Clinical Technologies: Concepts, Methodologies, Tools and Applications (3 Volume Set)

Clinical Technologies:

Concepts, Methodologies, Tools and Applications Information Resources Management Association USA

Volume I

Senior Editorial Director: Director of Book Publications: Editorial Director: Acquisitions Editor: Development Editor: Production Coordinator: Content Systems Analyst: Assistant Production Editor: Typesetters:

Cover Design:



Kristin Klinger Julia Mosemann Lindsay Johnston Erika Carter Chris Wozniak Jamie Snavely Devvin Earnest Deanna Zombro Michael Brehm, Keith Glazewski, Natalie Pronio, Jennifer Romanchak, Milan Vracarich, Jr., Deanna Zombro

Nick Newcomer

Published in the United States of America by Information Science Reference (an imprint of IGI Global) 701 E. Chocolate Avenue Hershey PA 17033 Tel: 717-533-8845 Fax: 717-533-8661 E-mail: [email protected] Web site: http://www.igi-global.com/reference and in the United Kingdom by Information Science Reference (an imprint of IGI Global) 3 Henrietta Street Covent Garden London WC2E 8LU Tel: 44 20 7240 0856 Fax: 44 20 7379 0609 Web site: http://www.eurospanbookstore.com

Copyright © 2011 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher. Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.

Library of Congress Cataloging-in-Publication Data Clinical technologies : concepts, methodologies, tools and applications / Information Resources Management Association, editor. p. cm. Summary: “This multi-volume book delves into the many applications of information technology ranging from digitizing patient records to highperformance computing, to medical imaging and diagnostic technologies, and much more”-- Provided by publisher. Includes bibliographical references and index. ISBN 978-1-60960-561-2 (hardcover) -- ISBN 978-1-60960-562-9 (ebook) 1. Medical technology. I. Information Resources Management Association. R855.3.C55 2011 610--dc22 2011011100 British Cataloguing in Publication Data A Cataloguing in Publication record for this book is available from the British Library. All work contributed to this book set is original material. The views expressed in this book are those of the authors, but not necessarily of the publisher.

Editor-in-Chief

Mehdi Khosrow-Pour, DBA Editor-in-Chief Contemporary Research in Information Science and Technology, Book Series

Associate Editors Steve Clarke University of Hull, UK Murray E. Jennex San Diego State University, USA Annie Becker Florida Institute of Technology USA Ari-Veikko Anttiroiko University of Tampere, Finland

Editorial Advisory Board Sherif Kamel American University in Cairo, Egypt In Lee Western Illinois University, USA Jerzy Kisielnicki Warsaw University, Poland Keng Siau University of Nebraska-Lincoln, USA Amar Gupta Arizona University, USA Craig van Slyke University of Central Florida, USA John Wang Montclair State University, USA Vishanth Weerakkody Brunel University, UK

Additional Research Collections found in the “Contemporary Research in Information Science and Technology” Book Series Data Mining and Warehousing: Concepts, Methodologies, Tools, and Applications John Wang, Montclair University, USA • 6-volume set • ISBN 978-1-60566-056-1 Electronic Business: Concepts, Methodologies, Tools, and Applications In Lee, Western Illinois University • 4-volume set • ISBN 978-1-59904-943-4 Electronic Commerce: Concepts, Methodologies, Tools, and Applications S. Ann Becker, Florida Institute of Technology, USA • 4-volume set • ISBN 978-1-59904-943-4 Electronic Government: Concepts, Methodologies, Tools, and Applications Ari-Veikko Anttiroiko, University of Tampere, Finland • 6-volume set • ISBN 978-1-59904-947-2 Knowledge Management: Concepts, Methodologies, Tools, and Applications Murray E. Jennex, San Diego State University, USA • 6-volume set • ISBN 978-1-59904-933-5 Information Communication Technologies: Concepts, Methodologies, Tools, and Applications Craig Van Slyke, University of Central Florida, USA • 6-volume set • ISBN 978-1-59904-949-6 Intelligent Information Technologies: Concepts, Methodologies, Tools, and Applications Vijayan Sugumaran, Oakland University, USA • 4-volume set • ISBN 978-1-59904-941-0 Information Security and Ethics: Concepts, Methodologies, Tools, and Applications Hamid Nemati, The University of North Carolina at Greensboro, USA • 6-volume set • ISBN 978-1-59904-937-3 Medical Informatics: Concepts, Methodologies, Tools, and Applications Joseph Tan, Wayne State University, USA • 4-volume set • ISBN 978-1-60566-050-9 Mobile Computing: Concepts, Methodologies, Tools, and Applications David Taniar, Monash University, Australia • 6-volume set • ISBN 978-1-60566-054-7 Multimedia Technologies: Concepts, Methodologies, Tools, and Applications Syed Mahbubur Rahman, Minnesota State University, Mankato, USA • 3-volume set • ISBN 978-1-60566-054-7 Virtual Technologies: Concepts, Methodologies, Tools, and Applications Jerzy Kisielnicki, Warsaw University, Poland • 3-volume set • ISBN 978-1-59904-955-7

Free institution-wide online access with the purchase of a print collection!

INFORMATION SCIENCE REFERENCE Hershey • New York

Order online at www.igi-global.com or call 717-533-8845 ext.100 Mon–Fri 8:30am–5:00 pm (est) or fax 24 hours a day 717-533-7115

List of Contributors

Aarts, Jos \ Erasmus University, Rotterdam, The Netherlands ........................................................... 25 Abeyratne, Udantha R. \ The University of Queensland, Australia ................................................. 295 Abraham, Chon \ College of William and Mary, USA ...................................................................... 108 Adang, Eddy M. M. \ Radboud University Nijmegen Medical Center, The Netherlands . ................... 1 Adler-Milstein, Julia \ Harvard University, USA . .......................................................................... 1075 Aggarwal, Vikram \ Johns Hopkins University, USA ....................................................................... 917 Ahmad, Yousuf J. \ Mercy Health Partners, USA ............................................................................. 132 Aittokallio, Tero \ University of Turku, Finland . .............................................................................. 422 Albert, Asunción \ University Hospital Dr. Peset/University of Valencia, Spain ........................... 1263 Alexander, Suraj M. \ University of Louisville, USA . .................................................................... 1393 Al-Jumaily, Adel A. \ University of Technology, Australia ............................................................... 965 Anderson, James G. \ Purdue University, USA . ............................................................................. 1491 Anderson, Kent \ University of California, Davis, USA ................................................................. 2094 Anselma, Luca \ Università di Torino, Torino, Italy ...................................................................... 1721 Antoniotti, Nina \ Marshfield Clinic Telehealth Network, USA ...................................................... 1843 Apweiler, Rolf \ European Bioinformatics Institute, UK . ............................................................... 1360 Arias, Pablo \ University of A Coruña, Spain . .................................................................................. 853 Armstrong, Alice J. \ The George Washington University, USA . ..................................................... 877 Astray, Loxo Lueiro \ University of Vigo, Spain . ............................................................................. 703 Atanasov, Asen \ Medical University Hospital “St. George”, Bulgaria ........................................... 172 Bali, R. K. \ Coventry University, UK . ............................................................................................ 1592 Barba, Pierluigi \ University of Pisa, Italy . ...................................................................................... 792 Battin, Malcolm \ National Women’s Health, Auckland City Hospital, New Zealand .................... 1215 Baxi, Hassan \ University of California, Davis, USA ...................................................................... 2094 Becker, Shirley Ann \ Florida Institute of Technology, USA .......................................................... 2047 Becker, Clemens \ Robert Bosch Gesellschaft für medizinische Forschung, Germany .................... 801 Beckers, Stefan \ University Hospital Aachen, Germany, . ............................................................... 974 Bergkvist, Sofi \ ACCESS Health Initiative, India .......................................................................... 1438 Bissett, Andy \ Sheffield Hallam University, UK ............................................................................. 1922 Biswas, Rakesh \ Manipal University, Malaysia & People’s College of Medical Sciences, India ....................................................................................................................... 1030, 1996, 2153 Borycki, Elizabeth M. \ University of Victoria, Canada .................................................................. 532 Botsivaly, M. \ Technological Education Institute of Athens, Greece . ............................................ 1674 Bottrighi, Alessio \ Università del Piemonte Orientale , Alessandria, Italy ................................... 1721 Bouchard, Stéphane \ University of Quebec in Outaouais, Canada .............................................. 2073

Bourantas, Christos V. \ Michailideion Cardiology Center, Greece & University of Hull, UK ..................................................................................................................................... 2114 Bowers, Clint \ University of Central Florida, Orlando, USA ........................................................ 2126 Boye, Niels \ Aalborg University, Denmark ..................................................................................... 1105 Bradford, John \ University of Ottawa, Canada . ........................................................................... 2073 Bratan, Tanja \ Brunel University, UK . .......................................................................................... 1047 Brimhall, Bradley B. \ University of New Mexico, USA . ............................................................... 1812 Brodziak, Tadeusz \ P3 Communications GmbH, Germany . ........................................................... 974 Bromage, Adrian \ Coventry University, UK ...................................................................................... 93 Burdick, Anne \ University of Miami Miller School of Medicine, USA .......................................... 1843 Byrd, Terry A. \ Auburn University, USA . ...................................................................................... 1461 Byrne, Elaine \ University of Pretoria, South Africa . ....................................................................... 444 Camon, Evelyn \ European Bioinformatics Institute, UK ............................................................... 1360 Camphausen, Kevin \ National Cancer Institute, USA .................................................................... 885 Cannon-Bowers, Jan \ University of Central Florida, Orlando, USA ........................................... 2126 Capilla, Rafael \ Universidad Rey Juan Carlos, Spain ..................................................................... 633 Carmona-Villada, Hans \ Instituto de Epilepsia y Parkinson del Eje Cafetero – Neurocentro, Colombia .................................................................................................................................... 1191 Cassim, M. \ Ritsumeikan Asia Pacific University, Japan ............................................................... 1770 Castellanos-Domínguez, Germán \ Universidad Nacional de Colombia, Colombia .................... 1191 Chartier, Sylvain \ University of Ottawa, Canada .......................................................................... 2073 Chatterjee, Aniruddha \ Johns Hopkins University, USA ................................................................ 917 Chatzizisis, Yiannis \ Aristotle University of Thessaloniki, Greece .................................................. 359 Chen, Dongqing \ University of Louisville, USA . ........................................................................... 1340 Chen, Hsiao-Hwa \ National Cheng Kung University, Taiwan ......................................................... 210 Cheung, Dickson \ Johns Hopkins Bayview Medical Center, USA ................................................... 917 Cho, Yoonju \ Johns Hopkins University, USA . ................................................................................ 917 Chorbev, Ivan \ Ss. Cyril and Methodius University, Republic of Macedonia ................................. 486 Citrin, Deborah \ National Cancer Institute, USA ............................................................................ 885 Clarke, Malcolm \ Brunel University, UK . ..................................................................................... 1047 Claster, William \ Ritsumeikan Asia Pacific University, Japan ...................................................... 1017 Colomo-Palacios, Ricardo \ Universidad Carlos III de Madrid, Spain ........................................... 995 Couto, Francisco M. \ Universidade de Lisboa, Portugal .............................................................. 1360 Covvey, H. Dominic \ University of Waterloo, Canada .................................................................. 1403 Cudeiro, Javier \ University of A Coruña, Spain .............................................................................. 853 Culhane, AedínC. \ Harvard School of Public Health, USA . ........................................................... 877 Curra, Alberto \ University of A Coruña, Spain ............................................................................... 368 Dadich, Ann \ University of Western Sydney, Australia .................................................................. 1759 Dahl, Yngve \ Telenor Research & Innovation, Norway ................................................................. 1171 Dale, Katherine \ Worcestershire Royal Hospital, UK . .................................................................. 1592 Daniel, Christel \ Université René Descartes, France .................................................................... 1235 Daskalaki, Andriani \ Max Planck Institute for Molecular Genetics, Germany .............................. 866 De Rossi, Danilo \ University of Pisa, Italy ....................................................................................... 792 del Río, Alfonso \ Universidad Rey Juan Carlos, Spain . .................................................................. 633 Delgado-Trejos, Edilson \ Instituto Tecnológico Metropolitano ITM, Colombia ........................... 1191 Dembowski, James \ Texas Tech University, Health Sciences Center USA .................................... 1554 Denzinger, Jörg \ University of Calgary, Canada ........................................................................... 1419

Di Giacomo, Paola \ University of Udine, Italy ................................................................................ 572 Díaz, Gloria \ National University of Colombia, Colombia .............................................................. 325 Dimmer, Emily \ European Bioinformatics Institute, UK ............................................................... 1360 Djebbari, Amira \ National Research Council Canada, Canada ..................................................... 877 Doyle, D. John \ Case Western Reserve University, USA & Cleveland Clinic Foundation, USA . ............................................................................................................................................. 190 Dryden, Gerald W. \ University of Louisville, USA ........................................................................ 1340 Duquenoy, Penny \ Middlesex University, London, UK . ................................................................ 1831 Efthimiadis, Efthimis N. \ University of Washington, USA ............................................................ 1121 Eldabi, Tillal \ Brunel University, UK ............................................................................................. 1738 Ellaway, Rachel \ Northern Ontario School of Medicine, Canada ................................................. 2153 Ergas, Henry \ Concept Economics, Australia . .................................................................................. 25 Espinosa, Nelson \ University of A Coruña, Spain ............................................................................ 853 Exarchos, Themis P. \ University of Ioannina, Greece ..................................................................... 412 Falk, Robert L. \ Jewish Hospital & St. Mary’s Healthcare, USA .................................................. 1340 Farag, Aly A. \ University of Louisville, USA .................................................................................. 1340 Farrell, Maureen \ University of Ballarat, Australia ...................................................................... 1504 Faxvaag, Arild \ Norwegian University of Science and Technology, Norway ................................ 1171 Fedoroff, Paul \ University of Ottawa, Canada .............................................................................. 2073 Feng, Dagan \ BMIT Research Group, The University of Sydney, Australia & Hong Kong Polytechnic University, Hong Kong ............................................................................................. 766 Fernández-Peruchena, Carlos \ University of Seville, Spain . ....................................................... 1284 Fielden, Kay \ UNITEC New Zealand, New Zealand ...................................................................... 1922 Fischermann, Harold \ University Hospital Aachen, Germany ....................................................... 974 Fitzgerald, Janna Anneke \ University of Western Sydney, Australia ............................................ 1759 Fotiadis, Dimitrios I. \ University of Ioannina, Greece, Michaelideion Cardiology Center, Greece & Biomedical Research Institute, Greece ........................................................................ 412 Fotiadis, Dimitrios I. \ University of Ioannina, Greece, Biomedical Research InstituteFORTH, Greece, & Michaelideion Cardiology Center, Greece ........................................ 305, 2114 Frenzel, Nadja \ University Hospital Aachen, Germany ................................................................... 974 Fursse, Joanna \ Brunel University, UK . ........................................................................................ 1047 Ganesh, A. U. Jai \ Sri Sathya Sai Information Technology Center, India . .................................... 1030 Gao, Bo \ Sichuan University, China ............................................................................................... 2191 García-Crespo, Ángel \ Universidad Carlos III de Madrid, Spain . ................................................. 995 Gasmelseid, Tagelsir Mohamed \ King Faisal University, Saudi Arabia ........................................ 250 Ghiassi, M. \ Santa Clara University, USA . ...................................................................................... 263 Ghotbi, Nader \ Ritsumeikan Asia Pacific University, Japan ......................................................... 1017 Giannakeas, Nikolaos \ University of Ioannina, Greece & National and Kapodistrian University of Athens, Greece .............................................................................................. 412, 2029 Giannoglou, George D. \ Aristotle University of Thessaloniki, Greece ............................................ 359 Gibson, Candace J. \ The University of Western Ontario, Canada ...................................... 1403, 1934 Gillies, Alan \ UCLAN, UK ................................................................................................................ 508 Godson, Nina \ Coventry University, UK ............................................................................................ 93 Goletsis, Yorgos \ University of Ioannina, Greece ............................................................................ 412 Gómez-Berbís, Juan M. \ Universidad Carlos III de Madrid, Spain ............................................... 995 González, Rubén Romero \ University of Vigo, Spain ..................................................................... 703 González Moreno, Juan Carlos \ University of Vigo, Spain ............................................................ 703

Gregory, Pete \ Coventry University, UK ........................................................................................ 1592 Gundlapalli, Adi V. \ University of Utah School of Medicine, USA . .............................................. 1470 Halazonetis, Demetrios J. \ National and Kapodistrian University of Athens, Greece . .................. 926 Hammer, Barbara \ Technical University of Clausthal, Germany ................................................... 478 Hammond, Kenric W. \ VA Puget Sound Health Care System, USA . ............................................ 1121 Hannan, Terry \ Australian College of Health Informatics, Australia ............................................... 25 Hartel, Pieter \ University of Twente, The Netherlands .................................................................... 391 Hayes, Richard L. \ University of South Alabama, USA . ............................................................... 1656 Hernández, Jesús Bernardino Alonso \ University of Las Palmas de Gran Canaria, Spain . ...... 1008 Hiner, Julie \ University of Calgary, Canada .................................................................................. 1419 Ho, Kendall \ University of British Columbia, Canada .................................................................... 147 Houliston, Bryan \ Auckland University of Technology, New Zealand . ........................................... 825 Ibraimi, Luan \ University of Twente, The Netherlands ................................................................... 391 Inoue, Yukiko \ University of Guam, Guam .................................................................................... 1530 Janß, A. \ RWTH Aachen University, Aachen, Germany ................................................................... 554 Jasemian, Yousef \ Engineering College of Aarhus, Denmark ......................................................... 717 Jean-Jules, Joachim \ Université de Sherbrooke, Canada ............................................................. 1962 Jiménez, Nicolás Victor \ University Hospital Dr. Peset/University of Valencia, Spain ................ 1263 Johnson, Roy D. \ University of Pretoria, South Africa .................................................................... 444 Joksimoski, Boban \ European University, Republic of Macedonia . ............................................... 486 Jones, Peter \ NHS Community Mental Health Nursing Older Adults, UK ...................................... 451 Jones, Erin \ University Health Network, Canada .............................................................................. 50 Jones, Russell \ Chorleywood Health Centre, UK ........................................................................... 1047 Jonker, Willem \ University of Twente, The Netherlands . ................................................................ 391 Kaan, Terry \ National University of Singapore, Singapore ........................................................... 1853 Kabene, Stefane M. \ University of Western Ontario, Canada & Ecole des Hautes Etudes en Sante Publique (EHESP), France ........................................................................................ 13, 1934 Kannry, Joseph L. \ Mount Sinai Medical Center, USA ................................................................. 1600 Karagiannis, G. E. \ Royal Brompton & Harefield NHS Trust, UK ................................................ 1093 Kastania, Anastasia N. \ Biomedical Research Foundation of the Academy of Athens, Greece & Athens University of Economics and Business, Greece ............................................... 118 Kasthuri, A. S. \ AFMC, India . ....................................................................................................... 1996 Kerstein, Robert B. \ Tufts University School of Dental Medicine, USA ......................................... 895 Khetrapal, Neha \ University of Bielefeld, Germany ........................................................................ 779 Khushaba, Rami N. \ University of Technology, Australia ............................................................... 965 Kim, Jeongeun \ Seoul National University, Korea ........................................................................ 2054 Kirsch, Harald \ European Bioinformatics Institute, UK ............................................................... 1360 Knight, David \ Mater Mother’s Hospital, Brisbane, Australia ...................................................... 1215 Kompatsiaris, Ioannis \ Informatics and Telematics Institue, Centre for Research and Technology Hellas, Greece . ......................................................................................................... 359 Kossida, Sophia \ Biomedical Research Foundation of the Academy of Athens, Greece ................. 118 Koul, Rajinder \ Texas Tech University, Health Sciences Center, USA .......................................... 1554 Koutsourakis, K. \ Technological Education Institute of Athens, Greece . ..................................... 1674 Krupinski, Elizabeth A. \ University of Arizona, USA ................................................................... 1843 Kuruvilla, Abey \ University of Wisconsin Parkside, USA ............................................................. 1393 Kuschel, Carl \ The Royal Women’s Hospital, Melbourne, Australia ............................................. 1215 Kushniruk, Andre W. \ University of Victoria, Canada ................................................................... 532

Kyriacou, Eftyvoulos \ Frederick University, Cyprus ...................................................................... 949 LaBrunda, Michelle \ Cabrini Medical Center, USA . .................................................................... 2183 Lagares-Lemos, Ángel \ Universidad Carlos III de Madrid, Spain ................................................. 995 Lagares-Lemos, Miguel \ Universidad Carlos III de Madrid, Spain ............................................... 995 Lauer, W. \ RWTH Aachen University, Aachen, Germany . ............................................................... 554 Leduc, Raymond W. \ University of Western Ontario, Canada . .................................................... 1934 Lee, Edwin Wen Huo \ Kuala Lumpur, Malaysia & Intel Innovation Center, Malaysia ...... 1030, 1996 Lee, Vivian \ European Bioinformatics Institute, UK ...................................................................... 1360 Lefkowitz, Jerry B. \ Weill Cornell College of Medicine, USA ...................................................... 1812 Li, Guang \ National Cacer Institute, USA ........................................................................................ 885 Liao, Guojie \ Sichuan University, China ........................................................................................ 2191 Ligomenides, Panos A. \ Academy of Athens, Greece . ..................................................................... 314 Linden, Ariel \ Linden Consulting Group & Oregon Health & Science University, USA . ............. 1075 Llobet, Holly \ Cabrini Medical Center, USA . ................................................................................ 2183 Llobet, Paul \ Cabrini Medical Center, USA ................................................................................... 2183 Logeswaran, Rajasvaran \ Multimedia University, Malaysia .......................................................... 744 López, M. Gloria \ University of A Coruña, Spain . .......................................................................... 368 Lum, Martin \ Department of Human Services, Australia . ............................................................ 1759 Macredie, Robert D. \ Brunel University, UK ................................................................................ 1738 Maliapen, Mahendran \ University of Sydney, Australia, National University of Singapore, Singapore and UCLAN, UK ......................................................................................................... 508 Manning, B. R. M. \ University of Westminster, UK ....................................................................... 1093 Manzanera, Antoine \ ENSTA-ParisTech, France ............................................................................ 325 Martin, Carmel M. \ Northern Ontario School of Medicine, Canada & Trinity College Dublin, Ireland ....................................................................................................... 1030, 1996, 2153 Martz, Jr., William Benjamin \ Northern Kentucky University, USA . ............................................ 132 Mato, Virginia \ University of A Coruña, Spain . .............................................................................. 368 May, Arnd T. \ University Hospital Aachen, Germany ..................................................................... 974 McDonald, Claudia L. \ Texas A&M University-Corpus Christi, USA .................................. 620, 2126 Melby, Line \ Norwegian University of Science and Technology, Norway ..................................... 1171 Mezaris, Vasileios \ Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece .............................................................................................................................. 359 Michalis, Lambros K. \ University of Ioannina, Greece & Michaelideion Cardiology Center, Greece ................................................................................................................................ 305, 2114 Miller, Thomas W. \ University of Connecticut, USA ..................................................................... 1566 Miller, Kenneth L. \ Youngstown State University, USA ................................................................. 1637 Miller, Robert W. \ National Cancer Institute, USA ......................................................................... 885 Miller, Susan M. \ Kent State University, USA . .............................................................................. 1637 Molino, Gianpaolo \ AOU San Giovanni Battista, Torino, Italy ..................................................... 1721 Montani, Stefania \ Università del Piemonte Orientale, Alessandria, Italy ................................... 1721 Moumtzoglou, Anastasius \ Hellenic Society for Quality & Safety in Healthcare, European Society for Quality in Healthcare Executive Board Member, Greece ............................................ 73 Mu, Ling \ West China Hospital, China . ......................................................................................... 2191 Mueller, Boris \ Memorial Sloan-Kettering Cancer Center, USA ..................................................... 885 Mukherji, Kamalika \ Hertfordshire Partnership NHS Foundation Trust, UK ............................. 1996 Mychalczak, Borys \ Memorial Sloan-Kettering Cancer Center, USA ............................................. 885 Na, In-Sik \ University Hospital Aachen, Germany .......................................................................... 974

Naka, Katerina K. \ Michailideion Cardiology Center, Greece ..................................................... 2114 Neofytou, Marios \ University of Cyprus, Cyprus . ........................................................................... 949 Noble, Robert \ The Robert Gordon University, Scotland . ............................................................. 1327 Nøhr, Christian \ Aalborg University, Denmark ............................................................................. 1105 Nóvoa, Francisco J. \ University of A Coruña, Spain ....................................................................... 368 Nytrø, Øystein \ Norwegian University of Science and Technology, Norway . ............................... 1171 Oladapo, Olufemi T. \ Olabisi Onabanjo University Teaching Hospital, Nigeria ......................... 1153 Orozco-Gutiérrez, Álvaro \ Universidad Tecnológica de Pereira, Colombia ................................ 1191 Ostermann, Herwig \ University for Health Sciences, Medical Informatics and Technology, Austria ........................................................................................................................................ 2035 Pain, Den \ Massey University, New Zealand .................................................................................. 1922 Paolucci, Francesco \ The Australian National University, Australia ................................................ 25 Papadogiorgaki, Maria \ Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece . ......................................................................................................... 359 Parise, Salvatore \ Babson College, USA . ...................................................................................... 1142 Parry, Gareth \ Horsmans Place Partnership, UK ......................................................................... 1518 Parry, Emma \ The University of Auckland, New Zealand ............................................................. 1581 Parsons, Thomas D. \ University of Southern California, USA ........................................................ 655 Pasupathy, Kalyan Sunder \ University of Missouri, USA ............................................................ 1684 Pattichis, Marios \ University of New Mexico, USA ......................................................................... 949 Pattichis, Constantinos \ University of Cyprus, Cyprus ................................................................... 949 Paul, Ray J. \ Brunel University, UK . ............................................................................................. 1738 Pekam, F. Chuembou \ RWTH Aachen University, Aachen, Germany . ........................................... 554 Pernefeldt, Hanna \ ACCESS Health Initiative, India .................................................................... 1438 Petroudi, Dimitra \ National and Kapodistrian University of Athens, Greece ............................... 2029 Pfeiffer, Klaus \ Robert Bosch Gesellschaft für medizinische Forschung, Germany ........................ 801 Piana, Michele \ Universita’ di Verona, Italy . ................................................................................... 353 Pongracz, Ferenc \ Albadent Inc., Hungary .................................................................................... 2143 Portas, Cesar Parguiñas \ University of Vigo, Spain . ...................................................................... 703 Prado-Velasco, Manuel \ University of Seville, Spain .................................................................... 1284 Price, Morgan \ University of Victoria, Canada, & University of British Columbia, Canada ....... 1874 Protogerakis, Michael \ RWTH Aachen University, Germany . ........................................................ 974 Proulx, Jean \ University of Montreal, Canada . ............................................................................. 2073 Quackenbush, John \ Harvard School of Public Health, USA ......................................................... 877 Quinn, Tom \ Coventry University, UK ........................................................................................... 1592 Radermacher, K. \ RWTH Aachen University, Aachen, Germany .................................................... 554 Raghavan, Vijay V. \ Northern Kentucky University, USA ............................................................... 132 Rapoport, Benjamin I. \ Massachusetts Institute of Technology, USA & Harvard Medical School, USA . ................................................................................................................................ 581 Rebholz-Schuhmann, Dietrich \ European Bioinformatics Institute, UK ..................................... 1360 Reid, Jonathan H. \ University of Utah School of Medicine, USA & Utah Department of Health, USA . .............................................................................................................................. 1470 Reis, Shmuel \ Technion- Israel Institute of Technology, Israel ........................................................ 160 Renaud, Patrice \ University of Quebec in Outaouais / Institut Philippe-Pinel de Montréal, Canada ....................................................................................................................................... 2073 Ribera, John \ Utah State University, USA ....................................................................................... 693 Ries, Nola M. \ University of Alberta, Canada, & University of Victoria, Canada ........................ 1948

Roberts, Jean M. \ University of Central Lancashire, UK . ............................................................ 1623 Rodríguez, Patricia Henríquez \ University of Las Palmas de Gran Canaria, Spain . ................. 1008 Rojo, Marcial García \ Hospital General de Ciudad Real, Spain .................................................. 1235 Root, Jan \ Utah Health Information Network, USA ....................................................................... 1470 Rossaint, Rolf \ University Hospital Aachen, Germany .................................................................... 974 Rouleau, Joanne L. \ University of Montreal, Canad ..................................................................... 2073 Ruiz-Fernández, Daniel \ University of Alicante, Spain ................................................................ 1782 Sahba, Farhang \ Medical Imaging Analyst, Canada ....................................................................... 377 Sánchez, José Antonio \ Universidad Politécnica de Madrid, Spain ............................................... 633 Sánchez Chao, Castor \ University of Vigo, Spain ........................................................................... 703 Sarpeshkar, Rahul \ Massachusetts Institute of Technology, USA . .................................................. 581 Schleif, Frank-M. \ University of Leipzig, Germany ........................................................................ 478 Schneiders, Marie-Thérèse \ RWTH Aachen University, Germany ................................................. 974 Scilingo, Enzo Pasquale \ University of Pisa, Italy . ......................................................................... 792 Seland, Gry \ Norwegian University of Science and Technology, Norway ..................................... 1171 Serrano, Antonio J. \ University of Valencia, Spain ....................................................................... 1263 Shachak, Aviv \ University of Toronto, Canada ................................................................................ 160 Shanmuganathan, Subana \ Auckland University of Technology, New Zealand ........................... 1017 Shibl, Rania \ University of the Sunshine Coast, Australia ............................................................. 1922 Silva, Mário J. \ Universidade de Lisboa, Portugal . ...................................................................... 1360 Skorning, Max \ University Hospital Aachen, Germany .................................................................. 974 Smith, Kevin \ National Digital Research Centre, Ireland ................................................... 1030, 2153 Sneed, Wanda \ Tarleton State University, USA .............................................................................. 2013 Soar, Jeffrey \ University of Southern Queensland, Australia . ....................................................... 1539 Song, Yulin \ Memorial Sloan-Kettering Cancer Center, USA .......................................................... 885 Sørby, Inger Dybdahl \ Norwegian University of Science and Technology, Norway ..................... 1171 Soria, Emilio \ University of Valencia, Spain .................................................................................. 1263 Soriano-Payá, Antonio \ University of Alicante, Spain .................................................................. 1782 Spera, C. \ Zipidy, Inc., USA .............................................................................................................. 263 Spyropoulos, B. \ Technological Education Institute of Athens, Greece ......................................... 1674 Spyrou, George M. \ Academy of Athens, Greece . ........................................................................... 314 Stamatopoulos, V. G. \ Biomedical Research Foundation of the Academy of Athens, Greece & Technological Educational Institute of Chalkida, Greece ......................................... 1093 Staudinger, Bettina \ University for Health Sciences, Medical Informatics and Technology, Austria ........................................................................................................................................ 2035 Staudinger, Roland \ University for Health Sciences, Medical Informatics and Technology, Austria ........................................................................................................................................ 2035 Stefurak, James “Tres” \ University of South Alabama, USA ........................................................ 1656 Sturmberg, Joachim \ Monash University and The University of Newcastle, Australia . ... 1030, 1996, 2153 Surry, Daniel W. \ University of South Alabama, USA ................................................................... 1656 Swennen, Maartje H. J. \ University Medical Centre Utrecht, The Netherlands . ......................... 1900 Tang, Qiang \ University of Twente, The Netherlands . ..................................................................... 391 Tanos, Vasilios \ Aretaeion Hospital, Nicosia, Cyprus ...................................................................... 949 Tantbirojn, Daranee \ Universtiy of Minnesota, USA .................................................................... 1374 Terenziani, Paolo \ Università del Piemonte Orientale, Alessandria, Italy .................................... 1721 Tesconi, Mario \ University of Pisa, Italy . ........................................................................................ 792

Thrasher, Evelyn H. \ University of Massachusetts Dartmouth, USA ............................................ 1461 Tong, Carrison K. S. \ Pamela Youde Nethersole Eastern Hospital, Hong Kong, China .............. 2173 Topps, David \ Northern Ontario School of Medicine, Canada ...................................................... 2153 Torchio, Mauro \ AOU San Giovanni Battista, Torino, Italy .......................................................... 1721 Toussaint, Pieter \ Norwegian University of Science and Technology, Norway ............................. 1171 Tsipouras, Markos G. \ University of Ioannina, Greece .......................................................... 305, 412 Tzavaras, A. \ Technological Education Institute of Athens, Greece .............................................. 1674 Übeyli, Elif Derya \ TOBB Ekonomi ve Teknoloji Üniversitesi, Turkey ............................................ 676 Umakanth, Shashikiran \ Manipal University, Malaysia .................................................... 1030, 1996 Urowitz, Sara \ University Health Network, Canada .......................................................................... 50 Valero, Miguel Ángel \ Universidad Politécnica de Madrid, Spain ................................................. 633 Versluis, Antheunis \ University of Minnesota, USA ...................................................................... 1374 Villablanca, Amparo C. \ University of California, Davis, USA .................................................... 2094 Villeneuve, Alain O. \ Université de Sherbrooke, Canada .............................................................. 1962 Villmann, Thomas \ University of Leipzig, Germany ....................................................................... 478 Vladzymyrskyy, Anton V. \ Association for Ukrainian Telemedicine and eHealth Development & Donetsk R&D Institute of Traumatology and Orthopedics, Ukraine . ....................................... 1062 Wadhwani, Arun Kumar \ MITS, India ........................................................................................... 467 Wadhwani, Sulochana \ MITS, India . .............................................................................................. 467 Walczak, Steven \ University of Colorado at Denver, USA ............................................................ 1812 Wang, Xiu Ying \ BMIT Research Group, The University of Sydney, Australia ............................... 766 Webbe, Frank \ Florida Institute of Technology, USA .................................................................... 2047 Wei, Wei \ West China Hospital, China ........................................................................................... 2191 Weir, Charlene R. \ George W. Allen VA Medical Center, USA ...................................................... 1121 White, Raymond \ Sunderland University, UK . ............................................................................. 1327 Whitehouse, Diane \ The Castlegate Consultancy, Malton, UK ..................................................... 1831 Wickramasinghe, Nilmini \ RMIT University, Melbourne, Australia ............................................ 1706 Wiljer, David \ University Health Network, Canada . ......................................................................... 50 Wilson, E. Vance \ The University of Toledo, USA . ........................................................................ 1800 Wolfe, Melody \ University of Western Ontario, Canada . .................................................................. 13 Wong, Eric T. T. \ The Hong Kong Polytechnic University, Hong Kong, China ............................ 2173 Wood, Jennifer A. \ VA Texas Valley Coastal Bend Health Care System, USA .............................. 1566 Xiao, Yang \ The University of Alabama, USA .................................................................................. 210 Xie, Shuyan \ The University of Alabama, USA ................................................................................ 210 Xu, Wu \ Utah Department of Health, USA . ................................................................................... 1470 Yergens, Dean \ University of Manitoba and University of Calgary, Canada ................................ 1419 Zhang, Zhiqiang \ Sichuan University, China ................................................................................ 2191 Zijlstra, Wiebren \ University Medical Center Groningen, The Netherlands .................................. 801

Contents

Volume I Section I. Fundamental Concepts and Theories This section serves as the groundwork for this comprehensive reference book by addressing central theories essential to the understanding of clinical technologies. Chapters found within these pages provide a tremendous framework in which to position clinical technologies within the field of information science and technology. Insight regarding the critical integration of global measures into clinical technologies is addressed, while crucial stumbling blocks of this field are explored. In the chapters comprising this introductory section, the reader can learn and choose from a compendium of expert research on the elemental theories underscoring the clinical technologies discipline. Chapter 1.1. The Implementation of Innovative Technologies in Healthcare: Barriers and Strategies............................................................................................................................. 1 Eddy M.M. Adang, Radboud University Nijmegen Medical Center, The Netherlands Chapter 1.2. Risks and Benefits of Technology in Health Care............................................................. 13 Stefane M. Kabene, University of Western Ontario, Canada Melody Wolfe, University of Western Ontario, Canada Chapter 1.3. The Effectiveness of Health Informatics........................................................................... 25 Francesco Paolucci, The Australian National University, Australia Henry Ergas, Concept Economics, Australia Terry Hannan, Australian College of Health Informatics, Australia Jos Aarts, Erasmus University, Rotterdam, The Netherlands Chapter 1.4. Personal Health Information in the Age of Ubiquitous Health......................................... 50 David Wiljer, University Health Network, Canada Sara Urowitz, University Health Network, Canada Erin Jones, University Health Network, Canada

Chapter 1.5. E-Health as the Realm of Healthcare Quality: A Mental Image of the Future.................. 73 Anastasius Moumtzoglou, Hellenic Society for Quality & Safety in Healthcare, European Society for Quality in Healthcare Executive Board Member, Greece Chapter 1.6. The Use of Personal Digital Assistants in Nursing Education.......................................... 93 Nina Godson, Coventry University, UK Adrian Bromage, Coventry University, UK Chapter 1.7. Reforming Nursing with Information Systems and Technology..................................... 108 Chon Abraham, College of William and Mary, USA Chapter 1.8. Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids................ 118 Anastasia N. Kastania, Biomedical Research Foundation of the Academy of Athens, Greece & Athens University of Economics and Business, Greece Sophia Kossida, Biomedical Research Foundation of the Academy of Athens, Greece Chapter 1.9. Adoption of Electronic Health Records: A Study of CIO Perceptions............................ 132 Yousuf J. Ahmad, Mercy Health Partners, USA Vijay V. Raghavan, Northern Kentucky University, USA William Benjamin Martz Jr., Northern Kentucky University, USA Chapter 1.10. Technology Enabled Knowledge Translation: Using Information and Communications Technologies to Accelerate Evidence Based Health Practices................................ 147 Kendall Ho, University of British Columbia, Canada Chapter 1.11. The Computer-Assisted Patient Consultation: Promises and Challenges..................... 160 Aviv Shachak, University of Toronto, Canada Shmuel Reis, Technion- Israel Institute of Technology, Israel Chapter 1.12. Quality and Reliability Aspects in Evidence Based E-Medicine.................................. 172 Asen Atanasov, Medical University Hospital “St. George”, Bulgaria Chapter 1.13. E-Medical Education: An Overview............................................................................. 190 D. John Doyle, Case Western Reserve University, USA & Cleveland Clinic Foundation, USA Chapter 1.14. Nursing Home............................................................................................................... 210 Shuyan Xie, The University of Alabama, USA Yang Xiao, The University of Alabama, USA Hsiao-Hwa Chen, National Cheng Kung University, Taiwan

Section II. Development and Design Methodologies This section provides exhaustive coverage of conceptual architecture frameworks to endow with the reader a broad understanding of the promising technological developments within the field of clinical technologies. Research fundamentals imperative to the understanding of developmental processes within clinical technologies are offered. From broad surveys to specific discussions and case studies on electronic tools, the research found within this section spans the discipline while offering detailed, specific discussions. From basic designs to abstract development, these chapters serve to expand the reaches of development and design technologies within the clinical technologies community. Chapter 2.1. Improving Clinical Practice through Mobile Medical Informatics................................. 250 Tagelsir Mohamed Gasmelseid, King Faisal University, Saudi Arabia Chapter 2.2. A Web-Enabled, Mobile Intelligent Information Technology Architecture for On-Demand and Mass Customized Markets....................................................................................... 263 M. Ghiassi, Santa Clara University, USA C. Spera, Zipidy, Inc., USA Chapter 2.3. A Framework for Information Processing in the Diagnosis of Sleep Apnea.................. 295 Udantha R. Abeyratne, The University of Queensland, Australia Chapter 2.4. Computer-Aided Diagnosis of Cardiac Arrhythmias...................................................... 305 Markos G. Tsipouras, University of Ioannina, Greece Dimitrios I. Fotiadis, University of Ioannina, Greece, Biomedical Research Institute FORTH, Greece, & Michaelideion Cardiology Center, Greece Lambros K. Michalis, University of Ioannina, Greece & Michaelideion Cardiology Center, Greece Chapter 2.5. Computer Aided Risk Estimation of Breast Cancer: The “Hipprocrates-mst” Project........................................................................................................... 314 George M. Spyrou, Academy of Athens, Greece Panos A. Ligomenides, Academy of Athens, Greece Chapter 2.6. Automatic Analysis of Microscopic Images in Hematological Cytology Applications......................................................................................................................... 325 Gloria Díaz, National University of Colombia, Colombia Antoine Manzanera, ENSTA-ParisTech, France Chapter 2.7. Computational Methods in Biomedical Imaging............................................................ 353 Michele Piana, Universita’ di Verona, Italy

Chapter 2.8. Visual Medical Information Analysis.............................................................................. 359 Maria Papadogiorgaki, Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece Vasileios Mezaris, Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece Yiannis Chatzizisis, Aristotle University of Thessaloniki, Greece George D. Giannoglou, Aristotle University of Thessaloniki, Greece Ioannis Kompatsiaris, Informatics and Telematics Institue, Centre for Research and Technology Hellas, Greece Chapter 2.9. Angiographic Images Segmentation Techniques............................................................ 368 Francisco J. Nóvoa, University of A Coruña, Spain Alberto Curra, University of A Coruña, Spain M. Gloria López, University of A Coruña, Spain Virginia Mato, University of A Coruña, Spain Chapter 2.10. Segmentation Methods in Ultrasound Images.............................................................. 377 Farhang Sahba, Medical Imaging Analyst, Canada Chapter 2.11. Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme to Securely Manage Personal Health Records......................................................................................... 391 Luan Ibraimi, University of Twente, The Netherlands Qiang Tang, University of Twente, The Netherlands Pieter Hartel, University of Twente, The Netherlands Willem Jonker, University of Twente, The Netherlands Chapter 2.12. Integration of Clinical and Genomic Data for Decision Support in Cancer................. 412 Yorgos Goletsis, University of Ioannina, Greece Themis P. Exarchos, University of Ioannina, Greece Nikolaos Giannakeas, University of Ioannina, Greece Markos G. Tsipouras, University of Ioannina, Greece Dimitrios I. Fotiadis, University of Ioannina, Greece, Michaelideion Cardiology Center, Greece & Biomedical Research Institute, Greece Chapter 2.13. Module Finding Approaches for Protein Interaction Networks.................................... 422 Tero Aittokallio, University of Turku, Finland Chapter 2.14. Networks of Action for Anti Retroviral Treatment Information Systems..................... 444 Elaine Byrne, University of Pretoria, South Africa Roy D. Johnson, University of Pretoria, South Africa Chapter 2.15. Socio-Technical Structures, 4Ps and Hodges’ Model................................................... 451 Peter Jones, NHS Community Mental Health Nursing Older Adults, UK

Chapter 2.16. Techniques for Decomposition of EMG Signals........................................................... 467 Arun Kumar Wadhwani, MITS, India Sulochana Wadhwani, MITS, India Chapter 2.17. Prototype Based Classification in Bioinformatics......................................................... 478 Frank-M. Schleif, University of Leipzig, Germany Thomas Villmann, University of Leipzig, Germany Barbara Hammer, Technical University of Clausthal, Germany Chapter 2.18. An Integrated System for E-Medicine: E-Health, Telemedicine and Medical Expert Systems..................................................................................................................................... 486 Ivan Chorbev, Ss. Cyril and Methodius University, Republic of Macedonia Boban Joksimoski, European University, Republic of Macedonia Chapter 2.19. The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy Simulation Analysis................................................................................................ 508 Mahendran Maliapen, University of Sydney, Australia, National University of Singapore, Singapore and UCLAN, UK Alan Gillies, UCLAN, UK Chapter 2.20. Use of Clinical Simulations to Evaluate the Impact of Health Information Systems and Ubiquitous Computing Devices Upon Health Professional Work.................................. 532 Elizabeth M. Borycki, University of Victoria, Canada Andre W. Kushniruk, University of Victoria, Canada Chapter 2.21. Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems............................................................................................................. 554 A. Janß, RWTH Aachen University, Aachen, Germany W. Lauer, RWTH Aachen University, Aachen, Germany F. Chuembou Pekam, RWTH Aachen University, Aachen, Germany K. Radermacher, RWTH Aachen University, Aachen, Germany Chapter 2.22. The European Perspective of E-Health and a Framework for its Economic Evaluation........................................................................................................................... 572 Paola Di Giacomo, University of Udine, Italy Chapter 2.23. A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders.............................................................................................................. 581 Benjamin I. Rapoport, Massachusetts Institute of Technology, USA & Harvard Medical School, USA Rahul Sarpeshkar, Massachusetts Institute of Technology, USA

Section III. Tools and Technologies This section presents an extensive treatment of various tools and technologies existing in the field of clinical technologies practitioners and academics alike must rely on to develop new techniques. These chapters enlighten readers about fundamental research on the many methods used to facilitate and enhance the integration of this worldwide phenomenon by exploring software and hardware developments and their applications—an increasingly pertinent research arena. It is through these rigorously researched chapters that the reader is provided with countless examples of the up-and-coming tools and technologies emerging from the field of clinical technologies. Chapter 3.1. Avatars and Diagnosis: Delivering Medical Curricula in Virtual Space......................... 620 Claudia L. McDonald, Texas A&M University-Corpus Christi, USA Chapter 3.2. Agile Patient Care with Distributed M-Health Applications........................................... 633 Rafael Capilla, Universidad Rey Juan Carlos, Spain Alfonso del Río, Universidad Rey Juan Carlos, Spain Miguel Ángel Valero, Universidad Politécnica de Madrid, Spain José Antonio Sánchez, Universidad Politécnica de Madrid, Spain Chapter 3.3. Affect-Sensitive Virtual Standardized Patient Interface System..................................... 655 Thomas D. Parsons, University of Southern California, USA Chapter 3.4. Telemedicine and Biotelemetry for E-Health Systems: Theory and Applications.......... 676 Elif Derya Übeyli, TOBB Ekonomi ve Teknoloji Üniversitesi, Turkey Chapter 3.5. Tele-Audiology in the United States: Past, Present, and Future..................................... 693 John Ribera, Utah State University, USA Chapter 3.6. SIe-Health, e-Health Information System....................................................................... 703 Juan Carlos González Moreno, University of Vigo, Spain Loxo Lueiro Astray, University of Vigo, Spain Rubén Romero González, University of Vigo, Spain Cesar Parguiñas Portas, University of Vigo, Spain Castor Sánchez Chao, University of Vigo, Spain Chapter 3.7. Sensing of Vital Signs and Transmission Using Wireless Networks.............................. 717 Yousef Jasemian, Engineering College of Aarhus, Denmark

Volume II Chapter 3.8. Neural Networks in Medicine: Improving Difficult Automated Detection of Cancer in the Bile Ducts...................................................................................................................... 744 Rajasvaran Logeswaran, Multimedia University, Malaysia

Chapter 3.9. Image Registration for Biomedical Information Integration........................................... 766 Xiu Ying Wang, BMIT Research Group, The University of Sydney, Australia Dagan Feng, BMIT Research Group, The University of Sydney, Australia & Hong Kong Polytechnic University, Hong Kong Chapter 3.10. Cognition Meets Assistive Technology: Insights from Load Theory of Selective Attention............................................................................................................................... 779 Neha Khetrapal, University of Bielefeld, Germany Chapter 3.11. Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis........................................................................................................................................ 792 Mario Tesconi, University of Pisa, Italy Enzo Pasquale Scilingo, University of Pisa, Italy Pierluigi Barba, University of Pisa, Italy Danilo De Rossi, University of Pisa, Italy Chapter 3.12. Wearable Systems for Monitoring Mobility Related Activities: From Technology to Application for Healthcare Services................................................................... 801 Wiebren Zijlstra, University Medical Center Groningen, The Netherlands Clemens Becker, Robert Bosch Gesellschaft für medizinische Forschung, Germany Klaus Pfeiffer, Robert Bosch Gesellschaft für medizinische Forschung, Germany Chapter 3.13. RFID in Hospitals and Factors Restricting Adoption.................................................... 825 Bryan Houliston, Auckland University of Technology, New Zealand Chapter 3.14. Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease............................................................................................................................. 853 Pablo Arias, University of A Coruña, Spain Nelson Espinosa, University of A Coruña, Spain Javier Cudeiro, University of A Coruña, Spain Chapter 3.15. Modeling of Porphyrin Metabolism with PyBioS......................................................... 866 Andriani Daskalaki, Max Planck Institute for Molecular Genetics, Germany Chapter 3.16. AI Methods for Analyzing Microarray Data................................................................. 877 Amira Djebbari, National Research Council Canada, Canada Aedín C. Culhane, Harvard School of Public Health, USA Alice J. Armstrong, The George Washington University, USA John Quackenbush, Harvard School of Public Health, USA

Chapter 3.17. 3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization........................................................................................................... 885 Guang Li, National Cacer Institute, USA Deborah Citrin, National Cancer Institute, USA Robert W. Miller, National Cancer Institute, USA Kevin Camphausen, National Cancer Institute, USA Boris Mueller, Memorial Sloan-Kettering Cancer Center, USA Borys Mychalczak, Memorial Sloan-Kettering Cancer Center, USA Yulin Song, Memorial Sloan-Kettering Cancer Center, USA Chapter 3.18. Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters.............................................................................................................. 895 Robert B. Kerstein, Tufts University School of Dental Medicine, USA Chapter 3.19. Ultrasound Guided Noninvasive Measurement of Central Venous Pressure................ 917 Vikram Aggarwal, Johns Hopkins University, USA Aniruddha Chatterjee, Johns Hopkins University, USA Yoonju Cho, Johns Hopkins University, USA Dickson Cheung, Johns Hopkins Bayview Medical Center, USA Chapter 3.20. Software Support for Advanced Cephalometric Analysis in Orthodontics................... 926 Demetrios J. Halazonetis, National and Kapodistrian University of Athens, Greece Chapter 3.21. Quantitative Analysis of Hysteroscopy Imaging in Gynecological Cancer.................. 949 Marios Neofytou, University of Cyprus, Cyprus Constantinos Pattichis, University of Cyprus, Cyprus Vasilios Tanos, Aretaeion Hospital, Nicosia, Cyprus Marios Pattichis, University of New Mexico, USA Eftyvoulos Kyriacou, Frederick University, Cyprus Chapter 3.22. Myoelectric Control of Prosthetic Devices for Rehabilitation...................................... 965 Rami N. Khushaba, University of Technology, Australia Adel A. Al-Jumaily, University of Technology, Australia Chapter 3.23. Med-on-@ix: Real-Time Tele-Consultation in Emergency Medical Services Promising or Unnecessary?.................................................................................................................. 974 In-Sik Na, University Hospital Aachen, Germany Max Skorning, University Hospital Aachen, Germany Arnd T. May, University Hospital Aachen, Germany Marie-Thérèse Schneiders, RWTH Aachen University, Germany Michael Protogerakis, RWTH Aachen University, Germany Stefan Beckers, University Hospital Aachen, Germany, Harold Fischermann, University Hospital Aachen, Germany Nadja Frenzel, University Hospital Aachen, Germany Tadeusz Brodziak, P3 Communications GmbH, Germany Rolf Rossaint, University Hospital Aachen, Germany

Chapter 3.24. DISMON: Using Social Web and Semantic Technologies to Monitor Diseases in Limited Environments......................................................................................................................... 995 Ángel Lagares-Lemos, Universidad Carlos III de Madrid, Spain Miguel Lagares-Lemos, Universidad Carlos III de Madrid, Spain Ricardo Colomo-Palacios, Universidad Carlos III de Madrid, Spain Ángel García-Crespo, Universidad Carlos III de Madrid, Spain Juan M. Gómez-Berbís, Universidad Carlos III de Madrid, Spain Section IV. Utilization and Application This section discusses a variety of applications and opportunities available that can be considered by practitioners in developing viable and effective clinical technologies programs and processes. This section includes over 20 chapters reviewing certain utilizations and applications of clinical technologies, such as artificial intelligence and social network analysis in healthcare. Further chapters show case studies from around the world, and the applications and utilities of clinical technologies. The wide ranging nature of subject matter in this section manages to be both intriguing and highly educational. Chapter 4.1. Speech-Based Clinical Diagnostic Systems.................................................................. 1008 Jesús Bernardino Alonso Hernández, University of Las Palmas de Gran Canaria, Spain Patricia Henríquez Rodríguez, University of Las Palmas de Gran Canaria, Spain Chapter 4.2. The Use of Artificial Intelligence Systems for Support of Medical Decision-Making................................................................................................................................ 1017 William Claster, Ritsumeikan Asia Pacific University, Japan Nader Ghotbi, Ritsumeikan Asia Pacific University, Japan Subana Shanmuganathan, Auckland University of Technology, New Zealand Chapter 4.3. Persistent Clinical Encounters in User Driven E-Health Care...................................... 1030 Rakesh Biswas, Manipal University, Malaysia Joachim Sturmberg, Monash University, Australia Carmel M. Martin, Northern Ontario School of Medicine, Canada A. U. Jai Ganesh, Sri Sathya Sai Information Technology Center, India Shashikiran Umakanth, Manipal University, Malaysia Edwin Wen Huo Lee, Kuala Lumpur, Malaysia Kevin Smith, National Digital Research Centre, Ireland Chapter 4.4. Remote Patient Monitoring in Residential Care Homes: Using Wireless and Broadband Networks......................................................................................................................... 1047 Tanja Bratan, Brunel University, UK Malcolm Clarke, Brunel University, UK Joanna Fursse, Brunel University, UK Russell Jones, Chorleywood Health Centre, UK Chapter 4.5. Telemedicine Consultations in Daily Clinical Practice: Systems, Organisation, Efficiency.................................................................................................................... 1062 Anton V. Vladzymyrskyy, Association for Ukrainian Telemedicine and eHealth Development & Donetsk R&D Institute of Traumatology and Orthopedics, Ukraine

Chapter 4.6. The Use and Evaluation of IT in Chronic Disease Management.................................. 1075 Julia Adler-Milstein, Harvard University, USA Ariel Linden, Linden Consulting Group & Oregon Health & Science University, USA Chapter 4.7. Safe Implementation of Research into Healthcare Practice through a Care Process Pathways Based Adaptation of Electronic Patient Records.................................................. 1093 V. G. Stamatopoulos, Biomedical Research Foundation of the Academy of Athens, Greece & Technological Educational Institute of Chalkida, Greece G. E. Karagiannis, Royal Brompton & Harefield NHS Trust, UK B. R. M. Manning, University of Westminster, UK Chapter 4.8. Towards Computer Supported Clinical Activity: A Roadmap Based on Empirical Knowledge and Some Theoretical Reflections................................................................. 1105 Christian Nøhr, Aalborg University, Denmark Niels Boye, Aalborg University, Denmark Chapter 4.9. Nursing Documentation in a Mature EHR System....................................................... 1121 Kenric W. Hammond, VA Puget Sound Health Care System, USA Charlene R. Weir, George W. Allen VA Medical Center, USA Efthimis N. Efthimiadis, University of Washington, USA Chapter 4.10. Applying Social Network Analysis in a Healthcare Setting........................................ 1142 Salvatore Parise, Babson College, USA Chapter 4.11. The Graphic Display of Labor Events......................................................................... 1153 Olufemi T. Oladapo, Olabisi Onabanjo University Teaching Hospital, Nigeria Chapter 4.12. The MOBEL Project: Experiences from Applying User-Centered Methods for Designing Mobile ICT for Hospitals................................................................................................. 1171 Inger Dybdahl Sørby, Norwegian University of Science and Technology, Norway Line Melby, Norwegian University of Science and Technology, Norway Yngve Dahl, Telenor Research & Innovation, Norway Gry Seland, Norwegian University of Science and Technology, Norway Pieter Toussaint, Norwegian University of Science and Technology, Norway Øystein Nytrø, Norwegian University of Science and Technology, Norway Arild Faxvaag, Norwegian University of Science and Technology, Norway Chapter 4.13. Processing and Communication Techniques for Applications in Parkinson Disease Treatment.............................................................................................................................. 1191 Álvaro Orozco-Gutiérrez, Universidad Tecnológica de Pereira, Colombia Edilson Delgado-Trejos, Instituto Tecnológico Metropolitano ITM, Colombia Hans Carmona-Villada, Instituto de Epilepsia y Parkinson del Eje Cafetero – Neurocentro, Colombia Germán Castellanos-Domínguez, Universidad Nacional de Colombia, Colombia

Chapter 4.14. Informatics Applications in Neonatology................................................................... 1215 Malcolm Battin, National Women’s Health, Auckland City Hospital, New Zealand David Knight, Mater Mother’s Hospital, Brisbane, Australia Carl Kuschel, The Royal Women’s Hospital, Melbourne, Australia Chapter 4.15. Digital Pathology and Virtual Microscopy Integration in E-Health Records............. 1235 Marcial García Rojo, Hospital General de Ciudad Real, Spain Christel Daniel, Université René Descartes, France Chapter 4.16. Clinical Decision Support System to Prevent Toxicity in Patients Treated with Digoxin......................................................................................................................... 1263 Asunción Albert, University Hospital Dr. Peset/University of Valencia, Spain Antonio J. Serrano, University of Valencia, Spain Emilio Soria, University of Valencia, Spain Nicolás Victor Jiménez, University Hospital Dr. Peset/University of Valencia, Spain Chapter 4.17. An Advanced Concept of Altered Auditory Feedback as a Prosthesis-Therapy for Stuttering Founded on a Non-Speech Etiologic Paradigm........................................................... 1284 Manuel Prado-Velasco, University of Seville, Spain Carlos Fernández-Peruchena, University of Seville, Spain Chapter 4.18. Reporting Clinical Gait Analysis Data........................................................................ 1327 Raymond White, Sunderland University, UK Robert Noble, The Robert Gordon University, Scotland Chapter 4.19. Variational Approach Based Image Pre- Processing Techniques for Virtual Colonoscopy.......................................................................................................................... 1340 Dongqing Chen, University of Louisville, USA Aly A. Farag, University of Louisville, USA Robert L. Falk, Jewish Hospital & St. Mary’s Healthcare, USA Gerald W. Dryden, University of Louisville, USA Chapter 4.20. Verification of Uncurated Protein Annotations........................................................... 1360 Francisco M. Couto, Universidade de Lisboa, Portugal Mário J. Silva, Universidade de Lisboa, Portugal Vivian Lee, European Bioinformatics Institute, UK Emily Dimmer, European Bioinformatics Institute, UK Evelyn Camon, European Bioinformatics Institute, UK Rolf Apweiler, European Bioinformatics Institute, UK Harald Kirsch, European Bioinformatics Institute, UK Dietrich Rebholz-Schuhmann, European Bioinformatics Institute, UK Chapter 4.21. Relationship Between Shrinkage and Stress............................................................... 1374 Antheunis Versluis, University of Minnesota, USA Daranee Tantbirojn, Universtiy of Minnesota, USA

Chapter 4.22. Predicting Ambulance Diversion................................................................................ 1393 Abey Kuruvilla, University of Wisconsin Parkside, USA Suraj M. Alexander, University of Louisville, USA Section V. Organizational and Social Implications This section includes a spacious range of inquiry and research pertaining to the behavioral, emotional, social and organizational impact of clinical technologies around the world. From demystifying human resources to privacy in mammography, this section compels the humanities, education, and IT scholar all. Section 5 also focuses on hesitance in some hospital members’ integration with clinical technologies, and methods therein. With more than 10 chapters, the discussions on hand in this section detail current and suggest future research into the integration of global clinical technologies as well as implementation of ethical considerations for all organizations. Overall, these chapters present a detailed investigation of the complex relationship between individuals, organizations and clinical technologies. Chapter 5.1. Demystifying E-Health Human Resources................................................................... 1403 Candace J. Gibson, The University of Western Ontario, Canada H. Dominic Covvey, University of Waterloo, Canada Chapter 5.2. Multi-Agent Systems in Developing Countries............................................................ 1419 Dean Yergens, University of Manitoba and University of Calgary, Canada Julie Hiner, University of Calgary, Canada Jörg Denzinger, University of Calgary, Canada Chapter 5.3. Primary Care through a Public-Private Partnership: Health Management and Research Institute............................................................................................................................... 1438 Sofi Bergkvist, ACCESS Health Initiative, India Hanna Pernefeldt, ACCESS Health Initiative, India Chapter 5.4. Strategic Fit in the Healtcare IDS................................................................................. 1461 Evelyn H. Thrasher, University of Massachusetts Dartmouth, USA Terry A. Byrd, Auburn University, USA Chapter 5.5. Regional and Community Health Information Exchange in the United States............. 1470 Adi V. Gundlapalli, University of Utah School of Medicine, USA Jonathan H. Reid, University of Utah School of Medicine, USA & Utah Department of Health, USA Jan Root, Utah Health Information Network, USA Wu Xu, Utah Department of Health, USA

Volume III Chapter 5.6. Regional Patient Safety Initiatives: The Missing Element of Organizational Change....................................................................................................................... 1491 James G. Anderson, Purdue University, USA Chapter 5.7. Use of Handheld Computers in Nursing Education...................................................... 1504 Maureen Farrell, University of Ballarat, Australia Chapter 5.8. Women’s Health Informatics in the Primary Care Setting............................................ 1518 Gareth Parry, Horsmans Place Partnership, UK Chapter 5.9. Assistive Technology for Individuals with Disabilities................................................. 1530 Yukiko Inoue, University of Guam, Guam Chapter 5.10. Ageing, Chronic Disease, Technology, and Smart Homes: An Australian Perspective.................................................................................................................. 1539 Jeffrey Soar, University of Southern Queensland, Australia Chapter 5.11. Synthetic Speech Perception in Individuals with Intellectual and Communicative Disabilities............................................................................................................... 1554 Rajinder Koul, Texas Tech University, Health Sciences Center, USA James Dembowski, Texas Tech University, Health Sciences Center USA Chapter 5.12. Telepractice: A 21st Century Model of Health Care Delivery.................................... 1566 Thomas W. Miller, University of Connecticut, USA Jennifer A. Wood, VA Texas Valley Coastal Bend Health Care System, USA Chapter 5.13. The Electronic Health Record to Support Women’s Health........................................ 1581 Emma Parry, The University of Auckland, New Zealand Chapter 5.14. Managing Paramedic Knowledge for Treatment of Acute Myocardial Infarction........................................................................................................................ 1592 Tom Quinn, Coventry University, UK R. K. Bali, Coventry University, UK Katherine Dale, Worcestershire Royal Hospital, UK Pete Gregory, Coventry University, UK

Section VI. Managerial Impact This section presents contemporary coverage of the social implications of clinical technologies, more specifically related to the corporate and managerial utilization of information sharing technologies and applications, and how these technologies can be facilitated within organizations. Section 6 is especially helpful as an addition to the organizational and behavioral studies of section 5, with diverse and novel developments in the managerial and human resources areas of clinical technologies. Typically, though the fields of industry and education are not always considered co-dependent, section 6 provides looks into how clinical technologies and the business workplace help each other. The interrelationship of such issues as operationalizing, supervision, and diagnosis management are discussed. In all, the chapters in this section offer specific perspectives on how managerial perspectives and developments in clinical technologies inform each other to create more meaningful user experiences. Chapter 6.1. Operationalizing the Science: Integrating Clinical Informatics into the Daily Operations of the Medical Center............................................................................................ 1600 Joseph L. Kannry, Mount Sinai Medical Center, USA Chapter 6.2. Current Challenges in Empowering Clinicians to Utilise Technology......................... 1623 Jean M. Roberts, University of Central Lancashire, UK Chapter 6.3. Challenges and Solutions in the Delivery of Clinical Cybersupervision...................... 1637 Kenneth L. Miller, Youngstown State University, USA Susan M. Miller, Kent State University, USA Chapter 6.4. Technology in the Supervision of Mental Health Professionals: Ethical, Interpersonal, and Epistemological Implications................................................................. 1656 James “Tres” Stefurak, University of South Alabama, USA Daniel W. Surry, University of South Alabama, USA Richard L. Hayes, University of South Alabama, USA Chapter 6.5. Optimization of Medical Supervision, Management, and Reimbursement of Contemporary Homecare................................................................................................................... 1674 B. Spyropoulos, Technological Education Institute of Athens, Greece M. Botsivaly, Technological Education Institute of Athens, Greece A. Tzavaras, Technological Education Institute of Athens, Greece K. Koutsourakis, Technological Education Institute of Athens, Greece Chapter 6.6. Systems Engineering and Health Informatics: Context, Content, and Implementation........................................................................................................................... 1684 Kalyan Sunder Pasupathy, University of Missouri, USA Chapter 6.7. Critical Factors for the Creation of Learning Healthcare Organizations...................... 1706 Nilmini Wickramasinghe, RMIT University, Melbourne, Australia

Chapter 6.8. Supporting Knowledge-Based Decision Making in the Medical Context: The GLARE Approach...................................................................................................................... 1721 Luca Anselma, Università di Torino, Torino, Italy Alessio Bottrighi, Università del Piemonte Orientale , Alessandria, Italy Gianpaolo Molino, AOU San Giovanni Battista, Torino, Italy Stefania Montani, Università del Piemonte Orientale, Alessandria, Italy Paolo Terenziani, Università del Piemonte Orientale, Alessandria, Italy Mauro Torchio, AOU San Giovanni Battista, Torino, Italy Chapter 6.9. Simulation Modeling as a Decision-Making Aid in Economic Evaluation for Randomized Clinical Trials................................................................................................................ 1738 Tillal Eldabi, Brunel University, UK Robert D. Macredie, Brunel University, UK Ray J. Paul, Brunel University, UK Chapter 6.10. How Can Human Technology Improve the Scheduling of Unplanned Surgical Cases?.................................................................................................................................. 1759 Janna Anneke Fitzgerald, University of Western Sydney, Australia Martin Lum, Department of Human Services, Australia Ann Dadich, University of Western Sydney, Australia Chapter 6.11. TACMIS: A Total Access Care and Medical Information System.............................. 1770 M. Cassim, Ritsumeikan Asia Pacific University, Japan Chapter 6.12. A Distributed Approach of a Clinical Decision Support System Based on Cooperation........................................................................................................................ 1782 Daniel Ruiz-Fernández, University of Alicante, Spain Antonio Soriano-Payá, University of Alicante, Spain Chapter 6.13. Applying Personal Health Informatics to Create Effective Patient-Centered E-Health................................................................................................................. 1800 E. Vance Wilson, The University of Toledo, USA Chapter 6.14. Diagnostic Cost Reduction Using Artificial Neural Networks: The Case of Pulmonary Embolism........................................................................................................................ 1812 Steven Walczak, University of Colorado at Denver, USA Bradley B. Brimhall, University of New Mexico, USA Jerry B. Lefkowitz, Weill Cornell College of Medicine, USA

Section VII. Critical Issues Section 7 details some of the most crucial developments in the critical issues surrounding clinical technologies. Importantly, this refers to critical thinking or critical theory surrounding the topic, rather than vital affairs or new trends that may be found in section 8. Instead, the section discusses some of the latest developments in ethics, law, and social implications in clinical technology development. Within the chapters, the reader is presented with an in-depth analysis of the most current and relevant issues within this growing field of study. Chapter 7.1. eHealth and Ethics: Theory, Teaching, and Practice..................................................... 1831 Diane Whitehouse, The Castlegate Consultancy, Malton, UK Penny Duquenoy, Middlesex University, London, UK Chapter 7.2. Standards and Guidelines Development in the American Telemedicine Association......................................................................................................................................... 1843 Elizabeth A. Krupinski, University of Arizona, USA Nina Antoniotti, Marshfield Clinic Telehealth Network, USA Anne Burdick, University of Miami Miller School of Medicine, USA Chapter 7.3. The Regulation of Genetic Testing and the Protection of Genetic and Medical Information in Singapore................................................................................................................... 1853 Terry Kaan, National University of Singapore, Singapore Chapter 7.4. A Bio-Psycho-Social Review of Usability Methods and their Applications in Healthcare................................................................................................................. 1874 Morgan Price, University of Victoria, Canada, & University of British Columbia, Canada Chapter 7.5. The Gap between What is Knowable and What We Do in Clinical Practice................ 1900 Maartje H.J. Swennen, University Medical Centre Utrecht, The Netherlands Chapter 7.6. Trust and Clinical Information Systems........................................................................ 1922 Rania Shibl, University of the Sunshine Coast, Australia Kay Fielden, UNITEC New Zealand, New Zealand Andy Bissett, Sheffield Hallam University, UK Den Pain, Massey University, New Zealand Chapter 7.7. Data Security in Electronic Health Records.................................................................. 1934 Stefane M. Kabene, Ecole des Hautes Etudes en Sante Publique (EHESP), France Raymond W. Leduc, University of Western Ontario, Canada Candace J Gibson, University of Western Ontario, Canada Chapter 7.8. Legal Issues in Health Information and Electronic Health Records............................. 1948 Nola M. Ries, University of Alberta, Canada, & University of Victoria, Canada

Chapter 7.9. Integrating Telehealth into the Organization’s Work System........................................ 1962 Joachim Jean-Jules, Université de Sherbrooke, Canada Alain O. Villeneuve, Université de Sherbrooke, Canada Chapter 7.10. Social Cognitive Ontology and User Driven Healthcare............................................ 1996 Rakesh Biswas, Manipal University, Malaysia Carmel M. Martin, Northern Ontario School of Medicine, Canada Joachim Sturmberg, Monash University, Australia Kamalika Mukherji, Hertfordshire Partnership NHS Foundation Trust, UK Edwin Wen Huo Lee, Intel Innovation Center, Malaysia Shashikiran Umakanth, Manipal University, Malaysia A. S. Kasthuri, AFMC, India Chapter 7.11. A Treatise on Rural Public Health Nursing................................................................. 2013 Wanda Sneed, Tarleton State University, USA Section VIII. Emerging Trends The final section explores the latest trends and developments, and suggests future research potential within the field of clinical technologies while exploring uncharted areas of study for the advancement of the discipline. The section advances through medical imaging techniques, diagnostics, virtual reality, and more new technologies by means of describing some of the latest trends in clinical research and development. These and several other emerging trends and suggestions for future research can be found within the final section of this exhaustive multi-volume set. Chapter 8.1. New Technologies in Hospital Information Systems.................................................... 2029 Dimitra Petroudi, National and Kapodistrian University of Athens, Greece Nikolaos Giannakakis, National and Kapodistrian University of Athens, Greece Chapter 8.2. IT-Based Virtual Medical Centres and Structures......................................................... 2035 Bettina Staudinger, University for Health Sciences, Medical Informatics and Technology, Austria Herwig Ostermann, University for Health Sciences, Medical Informatics and Technology, Austria Roland Staudinger, University for Health Sciences, Medical Informatics and Technology, Austria Chapter 8.3. Emerging Technologies for Aging in Place................................................................... 2047 Shirley Ann Becker, Florida Institute of Technology, USA Frank Webbe, Florida Institute of Technology, USA Chapter 8.4. The Development and Implementation of Patient Safety Information Systems (PSIS).................................................................................................................................. 2054 Jeongeun Kim, Seoul National University, Korea

Chapter 8.5. The Use of Virtual Reality in Clinical Psychology Research: Focusing on Approach and Avoidance Behaviors.................................................................................................. 2073 Patrice Renaud, University of Quebec in Outaouais / Institut Philippe-Pinel de Montréal, Canada Sylvain Chartier, University of Ottawa, Canada Paul Fedoroff, University of Ottawa, Canada John Bradford, University of Ottawa, Canada Joanne L. Rouleau, University of Montreal, Canad Jean Proulx, University of Montreal, Canada Stéphane Bouchard, University of Quebec in Outaouais, Canada Chapter 8.6. Novel Data Interface for Evaluating Cardiovascular Outcomes in Women................. 2094 Amparo C. Villablanca, University of California, Davis, USA Hassan Baxi, University of California, Davis, USA Kent Anderson, University of California, Davis, USA Chapter 8.7. New Developments in Intracoronary Ultrasound Processing....................................... 2114 Christos V. Bourantas, Michailideion Cardiology Center, Greece & University of Hull, UK Katerina K. Naka, Michailideion Cardiology Center, Greece Dimitrios I. Fotiadis, Michailideion Cardiology Center, Greece Lampros K. Michalis, Michailideion Cardiology Center, Greece Chapter 8.8. Pulse!!: Designing Medical Learning in Virtual Reality............................................... 2126 Claudia L. McDonald, Texas A&M University-Corpus Christi, USA Jan Cannon-Bowers, University of Central Florida, Orlando, USA Clint Bowers, University of Central Florida, Orlando, USA Chapter 8.9. Visualization and Modelling in Dental Implantology................................................... 2143 Ferenc Pongracz, Albadent Inc., Hungary Chapter 8.10. Patient Journey Record Systems (PaJR) for Preventing Ambulatory Care Sensitive Conditions: A Developmental Framework......................................................................... 2153 Carmel M. Martin, Trinity College Dublin, Ireland Rakesh Biswas, People’s College of Medical Sciences, India Joachim Sturmberg, Monash University and The University of Newcastle, Australia David Topps, Northern Ontario School of Medicine, Canada Rachel Ellaway, Northern Ontario School of Medicine, Canada Kevin Smith, National Digital Research Centre, Ireland Chapter 8.11. Picture Archiving and Communication System for Public Healthcare....................... 2173 Carrison K. S. Tong, Pamela Youde Nethersole Eastern Hospital, Hong Kong, China Eric T. T. Wong, The Hong Kong Polytechnic University, Hong Kong, China

Chapter 8.12. Imaging Advances of the Cardiopulmonary System................................................... 2183 Holly Llobet, Cabrini Medical Center, USA Paul Llobet, Cabrini Medical Center, USA Michelle LaBrunda, Cabrini Medical Center, USA Chapter 8.13. The Study of Transesophageal Oxygen Saturation Monitoring.................................. 2191 Zhiqiang Zhang, Sichuan University, China Bo Gao, Sichuan University, China Guojie Liao, Sichuan University, China Ling Mu, West China Hospital, China Wei Wei, West China Hospital, China

xxxii

Preface

Clinical technologies integrate the fields of Information Technology and Systems with healthcare, clinical design methodologies, and medicine. The constantly changing landscape of clinical technologies makes it challenging for experts and practitioners to stay informed of the field’s most up-to-date research. That is why Information Science Reference is pleased to offer this three-volume reference collection that will empower students, researchers, and academicians with a strong understanding of critical issues within clinical technologies by providing both extensive and detailed perspectives on cutting-edge theories and developments. This reference serves as a single, comprehensive reference source on conceptual, methodological, technical, and managerial issues, as well as providing insight into emerging trends and future opportunities within the discipline. Clinical Technologies: Concepts, Methodologies, Tools, and Applications is organized into eight distinct sections that provide wide-ranging coverage of important topics. The sections are: (1) Fundamental Concepts and Theories, (2) Development and Design Methodologies, (3) Tools and Technologies, (4) Utilization and Application, (5) Organizational and Social Implications, (6) Managerial Impact, (7) Critical Issues, and (8) Emerging Trends. Section 1, Fundamental Concepts and Theories, serves as a foundation for this extensive reference tool by addressing crucial theories essential to the understanding of clinical technologies. Chapters such as Risks and Benefits of Technology in Health Care, by Stefane Kabene and Melody Wolfe, introduce some of the fundamental issues involved with integrating technology into the field of healthcare. Other chapters introduce electronic health records, such as Adoption of Electronic Health Records, by Yousuf J. Ahmad, Vijay V. Raghavan, and William Benjamin Martz Jr., detailing the future of health records as they move forward in an increasingly electronic era. Other topics include medical (E-Medical Education by D. John Doyle) and nursing education (The use of Personal Digital Assistants in Nursing Education by Adrian Bromage and Nina Godson), and how e-learning has begun to take hold as a technological advancement allowing for off-site, online education. Overall, the first section is a fantastic introduction to some of the concepts and theories that will be addressed as the collection continues. Section 2, Development and Design Methodologies, presents in-depth coverage of the conceptual design and architecture of clinical technologies, focusing on aspects including the Balanced Scorecard framework, computer-aided diagnostics, biomedical imaging, and much more. Chapters in these areas include The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy Simulation Analysis by Mahendran Maliapen and Alan Gillies, detailing the potential for improving simulation analysis in healthcare policy, useful for tacticians and hospital designers interested in system dynamics and the Balanced Scorecard framework. Likewise, An Integrated System for E-Medicine (E-Health, Telemedicine and Medical Expert Systems) by Ivan Chorbev and Boban Joksimoski, is an included chapter, chosen for this reference book for its general breadth and articulation of the latest in design methodology within the clinical setting. Also included in this section are chapters on designing new medical vocabulary, and the alignment of clinical and general semantics. Together, section 2

xxxiii

comprises a comprehensive collection of the most recent publication in the design and development methodologies in clinical technologies. Section 3, Tools and Technologies, presents extensive coverage of the various tools and technologies used in the development and implementation of clinical technologies, including some of the newest applications to healthcare and bioinformatics. This comprehensive section includes such chapters that detail new methods for diagnosis and treatment of genetic, chronic, infectious, and other types of diseases and conditions. Such chapters include Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease by Javier Cudeiro, Nelson Espinosa, and Pablo Arias, and Quantitative Analysis of Hysteroscopy Imaging in Gynecological Cancer by Marios Neofytou, Constantinos Pattichis, Vasilios Tanos, Marios Pattichis, and Eftyvoulos Kyriacou. The section also specifically focuses on imaging technologies, with emphasis in chapters such as 3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization by Guang Li, Deborah Citrin, Robert W. Miller, Kevin Camphausen, Boris Mueller, Borys Mychalczak, and Yulin Song, as well as Image Registration for Biomedical Information Integration by Xiu Ying Wang and Dagan Feng. Section 3 concludes with a fascinating look at some recent developments in robot-assisted surgery. Section 4, Utilization and Applications, describes how clinical technology has been utilized, and offers insight on important lessons for its continued use and evolution. Due to the breadth of this section’s subject matter, section 4 contains the widest range of topics, including chapters such as The Use of Artificial Intelligence Systems for Support of Medical Decision-Making by William Claster, Nader Ghotbi, and Subana Shanmuganathan, and Applying Social Network Analysis in a Healthcare Setting by Salvatore Parise. As broad as the applications of clinical technology are, chapters within this section are pointed and precise, detailing case studies and lessons learned from integrating Information Technology with clinical systems. One intriguing example of this comes in the final chapter of the section, Predicting Ambulance Diversion by Abey Kuruvilla and Suraj M. Alexander, a detailed look at statistical and predictive analysis of one vital part of emergency care. Section 5, Organizational and Social Implications, includes chapters discussing the organizational and social impact of clinical technologies. Chapters are expository (Demystifying eHealth Human Resources by Candace J. Gibson and H. Dominic Covvey), pioneering, and based in research (Multi-Agent Systems in Developing Countries by Dean Yergens, Julie Hiner, and Joerg Denzinger). Overall, these chapters present a detailed investigation of the complex relationship between individuals, organizations and clinical technologies. Section 6, Managerial Impact, presents focused coverage of clinical technology as it relates to improvements and considerations in the workplace. In all, the chapters in this section offer specific perspectives on how managerial perspectives and developments in clinical technologies inform each other to create more meaningful user experiences. Typically, though the fields of industry and healthcare are not always considered co-dependent, section 6 provides looks into how clinical technologies and the business workplace help each other. Examples include Operationalizing the Science by Joseph L. Kannry and Technology in the Supervision of Mental Health Professionals by Daniel W. Surry, James R. Stefurak, and Richard L. Hayes. Section 7, Critical Issues, addresses some of the latest academic theory related to clinical technologies. Importantly, this refers to critical thinking or critical theory surrounding the topic, rather than vital affairs or new trends, which may be found in section 8. Instead, this section discusses some of the latest developments in ethics, law, and social implications in clinical technology development. Chapters include: eHealth and Ethics by Penny Duquenoy and Diane Whitehouse, Legal Issues in Health Information and Electronic Health Records by Nola M. Ries, and A Bio-Psycho-Social Review of Usability Methods and their Applications in Healthcare by Morgan Price.

xxxiv

Section 8, Emerging Trends, highlights areas for future research within the field of clinical technologies, while exploring new avenues for the advancement of the discipline. Beginning this section is New Technologies in Hospital Information Systems by Dimitra Petroudi and Nikolaos Giannakakis, which gives a summary view of some of the newest technological developments in hospital communications and health records. The section advances through medical imaging techniques, diagnostics, virtual reality, and more new technologies by means of describing some of the latest trends in clinical research and development. The book concludes with The Study of Transesophageal Oxygen Saturation Monitoring by Bo Gao, Guojie Liao, Ling Mu, Wei Wei, and Zhiqiang Zhang, a chapter from some of the most updated publication on digital clinical technology. Although the primary organization of the contents in this multi-volume work is based on its eight sections, offering a progression of coverage of the important concepts, methodologies, technologies, applications, social issues, and emerging trends, the reader can also identify specific contents by utilizing the extensive indexing system listed at the end of each volume. Furthermore, to ensure that the scholar, researcher, and educator have access to the entire contents of this multi volume set, as well as additional coverage that could not be included in the print version of this publication, the publisher will provide unlimited multi-user electronic access to the online aggregated database of this collection for the life of the edition, free of charge when a library purchases a print copy. This aggregated database provides far more contents than what can be included in the print version, in addition to continual updates. This unlimited access, coupled with the continuous updates to the database, ensures that the most current research is accessible to knowledge seekers. As a comprehensive collection of research on the latest findings related to using technology to providing various services, Clinical Technologies: Concepts, Methodologies, Tools and Applications, provides researchers, administrators and all audiences with a complete understanding of the development of applications and concepts in clinical technologies. Given the vast number of issues concerning usage, failure, success, policies, strategies, and applications of clinical technologies in healthcare organizations, Clinical Technologies: Concepts, Methodologies, Tools and Applications addresses the demand for a resource that encompasses the most pertinent research in clinical technologies development, deployment, and impact.

Section I

Fundamental Concepts and Theories This section serves as the groundwork for this comprehensive reference book by addressing central theories essential to the understanding of clinical technologies. Chapters found within these pages provide a tremendous framework in which to position clinical technologies within the field of information science and technology. Insight regarding the critical integration of global measures into clinical technologies is addressed, while crucial stumbling blocks of this field are explored. In the chapters comprising this introductory section, the reader can learn and choose from a compendium of expert research on the elemental theories underscoring the clinical technologies discipline.

1

Chapter 1.1

The Implementation of Innovative Technologies in Healthcare: Barriers and Strategies

Eddy M. M. Adang Radboud University Nijmegen Medical Center, The Netherlands

ABSTRACT Proven cost-effectiveness of innovative technologies is more and more a necessary condition for implementation in clinical practice. But proven cost-effectiveness itself does not guarantee successful implementation of an innovation. A reason for this could be the potential discrepancy between efficiency on the long run, on which cost-effectiveness is based, and efficiency on the DOI: 10.4018/978-1-60960-561-2.ch101

short run. In economics, long run and short run efficiency are discussed in the context of economies of scale. This chapter addresses the usefulness of cost-effectiveness for decision making considering the potential discrepancy between long run and short run efficiency of innovative technologies in healthcare, the potential consequences for implementation in daily clinical practice, explores diseconomies of scale in Dutch hospitals, and makes suggestions for what strategies might help to overcome hurdles to implement innovations due to that short run-long run efficiency discrepancy.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The Implementation of Innovative Technologies in Healthcare

COST-EFFECTIVENESS AND IMPLEMENTATION Cost-effectiveness analysis as part of the evaluation of medical innovations has become mainstream in several European countries as in Canada and Australia. For example in the UK, the National Institute of Clinical Excellence (NICE) uses costeffectiveness outcome, expressed as cost per quality adjusted life year gained, as a criterion for coverage recommendations to the National Health Service (Weinstein, 2008). In the Netherlands the Dutch Health Insurance Board (CVZ) uses the cost-effectiveness criterion in their advise to the minister about the inclusion of expensive intramural pharmaceuticals in the benefit package. Unlike the reimbursement authorities in Canada and Australia, and in many countries in Europe, in the US Medicare officials do not formally consider cost-effectiveness when determining the coverage of new medical interventions (Neumann, 2005). In general one would assume that if the evidence on therapeutic value and cost-effectiveness of innovations in health care is convincing, implementation in clinical practice is warranted. However, a characteristic of the health care sector is the somewhat fuzzy priority setting about implementation and the numerous potential conflicts between the stakeholders in the health system. Also, behavioral factors in individual health professionals, such as clinical inertia and persistent routine behaviors, may inhibit change. Therefore implementation of evidence-based innovations and guidelines does not follow automatically, as there might be barriers for change at different levels that need to be solved. Specific strategies targeting increasing speed and level of adoption of innovations can be launched, such as ‘tailored’ strategies directed at the medical profession and patients (for example, providing information, education, training, communication etc.). However, in a situation with convincing cost-effectiveness evidence and a high willingness to implement by the medical profession, implementation of

2

a technology or guideline into clinical practice might stagnate because there are negative consequences for specific stakeholders. The fact that economic evaluation might oversimplify complex health care decisions and disregards the multistakeholder issue has been noted before (Davies et al., 1994; Drummond et al., 2005; Eddama & Coast, 2008; Gold et al., 1996; Hoffmann & Graf von der Schulenburg, 2000; Johannesson, 1995; Lessard, 2007). This chapter argues that implementation of technologies, despite proven cost-effectiveness evidence, might be hampered due to specific economic barriers related to disincentives to implement by management of care providers. Key in the argumentation about successful implementation directed to technological change is the state of equilibrium in production: long run versus short run. Aletras (1999) concludes from his empirical work on estimating long-run and short-run cost functions in a sample of Greek NHS general hospitals that the use of long-run cost functions should be avoided since it might seriously mislead policymakers. Consequently, according to Aletras (1999), evidence on economies should presumably place lower validity weight on estimates derived from long-run as opposed to short-run cost functions. A necessary condition for achieving technical efficiency (an assumption underlying long-term cost-effectiveness) is that all inputs can be set at their cost-minimizing levels (Smet, 2002). However, inputs such as capital but also personnel with fixed labor contracts are difficult to adjust quickly to respond to changing output (levels) and will therefore not be set at their cost-minimizing level, given the output produced. This might have consequences for the management of health care providers who are often accountable for short run results like for example a balanced yearly budget or are restricted by tight financial frameworks. In health care systems where budgetary exceeds are sanctioned by a discount in the budget for next year (for example in the Netherlands) achieving

The Implementation of Innovative Technologies in Healthcare

budget becomes so important that those who perceive great pressure to meet budget, may be less inclined to partake of any activity that may cause temporarily inefficiencies in their environment (Adang, in press). Kallapur and Eldenburg (2005) found that if hospitals, when faced with increasing uncertainty, have a choice between technologies for a given activity, technologies with high variable and low fixed costs become more attractive relative to those with low variable and high fixed costs. The aim of this chapter is to discuss the usefulness of cost-effectiveness analysis in the context of implementation of innovations in health care. To come up with implementation strategies that are able to create a short run equilibrium that equals the long run equilibrium. This chapter is structured as follows: first the long run – short run discrepancy in efficiency is presented using an innovation in the treatment of psoriasis. Then the short-long run efficiency is discussed in an economic conceptual perspective focusing on returns to scale. Further, to get a quantitative impression whether diseconomies of scale are present in health care the prevalence of variable returns to scale is investigated in a sample of 33 Dutch hospitals (Adang & Wensing, 2008). Finally, using an example in diagnostics, implementation strategies that are able to overcome short run diseconomies are illustrated.

LONG RUN EFFICIENCY AND SHORT RUN LOSSES An earlier study about a new combined outpatient and home-treatment of psoriasis technology showed that 89% of the anticipated savings, based on the outcomes of a cost-effectiveness analysis, could not be achieved in the short run when implementing this technology due to inflexibility of production factors labor and infrastructure (Adang et al., 2005; Hartman et al., 2002). Consequently this meant that only €694 of the anticipated €6058

savings per patient could be freed and more than €5000 per patient could not be re-invested in the outpatient and home-treatment in the short run. This was considered a serious obstacle for health care management to implement the technology. On the other hand, the government and the medical profession strongly advocated the new technology as it was found convincingly cost-effective (Hartman et al., 2002). This resulted in the situation that both psoriasis treatment modalities existed next to each other. The “economic” reasoning behind this could be summarized as that the opportunity costs of using the inpatient treatment modality on the short run was very low. Hospital beds and personnel on the dermatology ward were fixed factors of production in the short run with little or no alternative use. At the end of the day both psoriasis treatment modalities co-existed in the same organization but were used at less than optimal capacity. Obviously information about an innovation’s cost-effectiveness provides not all economic information necessary for decision making about successful implementation. The evaluation of costs and benefits of new technologies and implementation of technologies is generally discussed in the context of Welfare Economics where welfare losses on the short run are considered ‘sunk’ (Adang, 2008; Drummond et al., 2005; Gold et al., 1996; Katz & Rosen, 1991). This might be true for stakeholders that decide on a long enough planning horizon where all production factors are flexible. In fact this convexity assumption is traditionally invoked in economics. In mathematical terms, in Euclidean space, an object is convex if for every pair of points within the object, every point on the straight line segment that joins them is also within the object. However, the convexity assumption is not without criticism. Farrel (1959) points to indivisibilities and economies of scale as sources of non-convexities. Allais (1977) confirms Farrel’s arguments where he rejects global convexity but favors local convexity. More recently for example Tone and Sahoo (2003) and Briec et al., (2004) point

3

The Implementation of Innovative Technologies in Healthcare

to non-convexity behavior of inputs. In a global convexity context, technologies implemented by health care providers are assumed to be infinitely divisible and production factors are supposed to be operating in a constant returns to scale region. Local convexity constraints this reasoning to a specific range. Global convexity is often not the case on the short run (Adang et al., 2005). Some factors of production are costly to adjust in the short run, and this induces short-run decreasing returns to scale and an upward sloping short run marginal cost curve. Short run inefficiencies might cause disincentives for health care management to implement innovations. According to Smith and Street (2005) in the short run many factors (like for example differences in efficiency and organizational priorities) are outside the control of the organization. On the long run this potentially can change. Implementation strategies directed to solve the short run – long run discrepancy can be exogenous and endogenous. Exogenous means the solution lays outside the care provider, for example changing the health system (and budgetary system) in a way that is consistent with decision making based on long run efficiency. This chapter focuses on endogenous implementation strategies meaning strategies that can be employed by the care provider itself. Therefore to optimally implement innovations that in the long run are found cost-effective, implementation strategies need to be directed at minimizing welfare losses, or inefficiencies, on the short run. To build effective implementation strategies information about potential short run diseconomies is necessary.

THE ECONOMICS BEHIND SHORT AND LONG RUN EFFICIENCY The discussion about long run and short run cost and efficiency is in economic literature discussed as part of “theory of the firm” (Katz & Rosen, 1991). The relationship between short run and

4

long run efficiency is associated with economies of scale and scope (Baumol, Panzar, Willig, 1988; Katz & Rosen, 1991). Returns to scale is a long run concept that reflects the degree to which a proportional increase in all inputs increases output. For example, one can consider scale economies to the degree that costs change in relation to changes in number of diagnostic performances. Cost-effectiveness analysis assumes constant returns to scale (Adang,2008; Drummond et al., 2005; Gold et al., 1996). This occurs when a proportional increase in all inputs results in the same proportional increase in output. This assumes that all production factors are functioning at optimal capacity. The case of the outpatient treatment of psoriasis innovation clearly shows that, at least on the short run, both treatment of psoriasis modalities do not function at optimal capacity and suffer from inefficiencies. Cremieux et al., (2005) argue that hospital unit costs may increase in response to increases in demand despite significant flexibility in inpatient care because other services such as laboratory exams may adjust poorly. Elbasha and Messonnier (2004) argue that technologies that are administered in health care settings often violate the assumption of constant returns to scale (costs increase linear with increasing production) that underlies cost-effectiveness analysis (CEA). These authors refer to various publications that illustrate the violation of the constant returns to scale assumption. For example, studies about nursing homes have revealed a mixture of findings, ranging from economies to diseconomies of scale (Dor, 1989; Gertler & Walman, 1992; Kass, 1987; Vitaliano & Toren, 1994). In fact, violation of the constant returns to scale assumption seems prevalent in health care. The constant returns to scale assumption is only appropriate when all health care programs or technologies are operating at an optimal scale. Imperfect competition, budgetary constraints, technology shifts, fixed contracts, etc. may cause a health care program to be not operating at optimal scale (Adang et al., 2005; Roberts, Frutos and Ciavarella, 1999).

The Implementation of Innovative Technologies in Healthcare

However, Kass (1987) presents empirical findings that show that economies of scale were not substantial in home health care. In this setting, according to Kass (1987), the ratio fixed to total costs was 5.6%, as labor (about 95%) could be considered variable costs and therefore moved linearly with output. On the other hand, Roberts, Frutos and Ciavarella (1999) found that about 84% of hospital costs were fixed. All this illustrates that diseconomies of scale are an important potential cause of inefficiency in the organization. Summarizing, diseconomies of scale refer to the relationship of average costs with volume of production. Diseconomies of scale arise when marginal costs of production get, with increasing volume of production, higher than average cost. This may be the result of a variety of factors: returns to scale, behavior of overheads, indivisibility of factors of production, nature of contracts between different stakeholders and organizational governance (Tone & Sahoo, 2003). Diseconomies of scope refer to the multipurpose use of capital investments. Diseconomies of scope are conceptually similar to diseconomies of scale. Where diseconomies of scale refer to changes in the output of a single technology, diseconomies of scope refer to changes in the number of different types of technologies (Adang, 2008). Being such a major source in explaining the potential long run – short run efficiency discrepancy raises the question whether (dis)economies of scale are prevalent in health care.

THE PREVALENCE OF DISECONOMIES OF SCALE IN DUTCH HOSPITALS To investigate the prevalence of returns to scale in health care a sample of 33 Dutch hospitals was researched (Adang & Wensing, 2008). The data came from annual reports of Dutch hospitals over the year 2003 (Adang, 2005). To create insight in whether these hospitals are working in the area of

constant returns to scale, increasing returns to scale or decreasing returns to scale, data envelopment analysis (DEA) was applied. In fact there are two principal approaches to analyzing the presence of returns to scale in organizations: parametric (corrected ordinary least squares and stochastic production frontiers) and non parametric (DEA). According to Coelli (1998) in the non-profit service sector, where random influences are less of an issue, multiple-output production is important, prices are difficult to define and behavioral assumptions, such as cost minimization or profit maximization, are difficult to justify, the DEA approach may often be the optimal choice. DEA is a non-parametric approach originally developed in the field of operations research for evaluating the performance of a set of peer entities using linear programming techniques (Charnes, Cooper and Rhodes, 1978). Farrel (1957) proposed a piece-wise-linear convex hull approach to efficient frontier estimation. Linear programming methods are used to construct a non-parametric piece-wise frontier over the data. Efficiency measures are then calculated relative to this frontier (Coelli, 2004). Charnes, Cooper and Rhodes (1978) developed a DEA model that assumed constant returns to scale (CRS). However, as is described earlier in this chapter the constant returns to scale assumption will be rejected if organizations are not operating at an optimal scale. Banker, Charnes and Cooper (1984) suggested an extension of the constant returns to scale DEA model to account for variable returns to scale (VRS). The analysis regarding the prevalence of (dis) economies of scale in hospitals begins with the determination of the hospital production function. The inputs of the hospital production function were: personnel costs, feeding and hotel costs, general costs, patient related costs and maintenance and energy costs. The outputs were: number of day-care, number of hospital days, and number of first outpatient visits. This production function is surely not complete, however the inputs and

5

The Implementation of Innovative Technologies in Healthcare

outputs are well acceptable and generally used, given a hospital environment. Inputs and outputs were on a yearly basis. The basic constant returns to scale DEA model is represented as follows: Min s.t .

Figure 1. Constant & Variable returns to scale

θ n

∑λ x

≤ θx i 0

i = 1, 2,..., m (inputs )

∑ λj yrj ≥ yr 0

r = 1, 2,..., s(outputs )

j =1 n

j

ij



j =1

λj ≥ 0

λj is the weight given to hospital j in its efforts to dominate hospital 0 and θ is the efficiency of hospital 0. Therefore, λ and θ are the variables to solve from the model. Figure 1 displays two production frontiers, one under constant returns to scale and one under variable returns to scale. Scale efficiencies are found by comparing efficiency on the variable returns to scale frontier to efficiency on the constant returns to scale frontier (Coelli, 2004). The variable returns to scale model adds the convexity constraint

n

∑λ j =1

j

= 1 to the constant returns

to scale DEA model. The convexity constraint ensures that an inefficient hospital is only benchmarked against hospitals of a similar size. This approach forms a convex hull of intersecting planes which envelope the data points more tightly than the constant returns to scale model. To explore returns to scale both DEA models were run using an input-orientation (Zhu, 2003). I, II, III, IV, V and VI are the regions associated with Table 1. Solving the VRS model (Table 1) shows that variable returns to scale are prevalent in Dutch hospitals. Results show that 55% of the hospitals were operating under decreasing returns to scale, 33% under constant returns to scale and 12% under increasing returns to scale. Based on these results it seems appropriate to assume that (dis)

6

economies of scale are prevalent in the Dutch hospital environment and consequently a potential concern for implementation of innovations.

SOLUTIONS FOR REDUCING SHORT RUN DISECONOMIES: CHOOSING THE RIGHT IMPLEMENTATION STRATEGIES Recently it was found that in patients with suspicion for prostate cancer diagnosis using a MRI device (with a specific contrast agent) is costeffective (the point estimate showed dominance: saving money and gaining quality adjusted life years) compared to diagnosis with a CT device (Hovels et al., 2008). From a long run perspective it is therefore clinically and economically sound to substitute CT for MRI in this particular patient group. Now let’s assume that prior to substitution both modalities CT and MRI were functioning at long run equilibrium i.e., optimal capacity (at constant returns to scale). To deal with the extra demand for MRI it is necessary to increase capacity by investing in an extra MRI device. On the short run output (number of MRIs) will increase less than proportionate with the input (production factors) (see Figure 2). This results in disecono-

The Implementation of Innovative Technologies in Healthcare

Table 1. Returns to scale in Dutch hospitals

Hospital location in the Netherlands

Input-oriented

Input-oriented

VRS

CRS

Input-oriented

RTS Region

Efficiency

Efficiency

1

Maastricht (UMC)

Region III

0,64782

0,48635

2,04154

Decreasing

2

Amsterdam (UMC)

Region III

0,56774

0,31385

3,02658

Decreasing

∑λ

RTS

3

Amsterdam (UMC)

Region III

0,48221

0,40303

1,94916

Decreasing

4

Groningen (UMC)

Region III

1,00000

0,50571

3,26753

Decreasing

5

Utrecht (UMC)

Region III

0,58401

0,33767

3,91163

Decreasing

6

Nijmegen (UMC)

Region III

0,58175

0,38371

2,37224

Decreasing

7

Leiden (UMC)

Region III

0,51136

0,44677

1,94743

Decreasing

8

Rotterdam (UMC)

Region III

1,00000

0,33021

3,31326

Decreasing

9

Deventer

Region III

0,85615

0,79930

1,32166

Decreasing

Tilburg

Region III

1,00000

0,77854

2,05430

Decreasing

10 11

Amsterdam

Region III

0,86786

0,69859

2,18329

Decreasing

12

Sittard

Region III

0,95313

0,85709

1,49226

Decreasing

13

Nieuwegein

Region III

0,89809

0,71196

1,66093

Decreasing

14

Ede

Region III

1,00000

0,99875

1,88144

Decreasing

15

Dordrecht

Region II

1,00000

1,00000

1,00000

Constant

16

Den Haag

Region III

1,00000

0,85115

1,65932

Decreasing

17

Delft

Region II

1,00000

1,00000

1,00000

Constant

18

Alkmaar

Region III

1,00000

0,78603

2,07834

Decreasing

19

Drachten

Region II

1,00000

1,00000

1,00000

Constant

20

Sneek

Region II

1,00000

1,00000

1,00000

Constant

21

Rotterdam

Region II

1,00000

1,00000

1,00000

Constant

22

Hilversum

Region VI

0,85072

0,85036

0,98041

Increasing

23

Winschoten

Region II

1,00000

1,00000

1,00000

Constant

24

Roosendaal

Region II

1,00000

1,00000

1,00000

Constant

25

Den Haag

Region I

0,97634

0,97023

0,92509

Increasing

26

Almere

Region II

1,00000

1,00000

1,00000

Constant

27

Lelystad

Region I

0,97577

0,95121

0,91390

Increasing

28

Utrecht

Region II

1,00000

1,00000

1,00000

Constant

29

Den Helder

Region I

0,97726

0,92168

0,74411

Increasing

30

Hoogeveen

Region II

1,00000

1,00000

1,00000

Constant

31

Rotterdam

Region III

1,00000

0,83530

2,55243

Decreasing

32

Enschede

Region III

1,00000

0,76230

2,45027

Decreasing

33

Zwolle

Region II

1,00000

1,00000

1,00000

Constant

RTS region means returns to scale region (see Figure 1). VRS means variable returns to scale CRS, IRS, DRS mean respectively constant, increasing and decreasing returns to scale:

7

The Implementation of Innovative Technologies in Healthcare

Figure 2. Returns to scale

mies of scale on the short run, assuming that on the long run constant returns to scale is restored. Implementation of the MRI modality has also consequences for the CT modality. To successfully implement MRI there is a need to establish substitution between both modalities. So an increase in demand for MRI comes with a decrease in demand for CT in the short run. This demand shift causes to reduce the multipurpose function of the CT modality and causes, on the short run, diseconomies of scope (see Figure 3). How do these phenomenon translate into barriers of implementation? This can be illustrated with the following numerical example. Table 2 shows the hypothetical production process on a yearly basis of the CT and MRI modality for a Figure 3. Returns to scope

8

particular organization. Let’s assume that fixed costs of production constitute of depreciation costs of the CT and MRI device respectively. Variable costs constitute of personnel and material cost like contrast agents. In the short run the health care provider has an incentive (marginal costs decrease) to meet the extra demand by investing in the variable input. However, in Table 2 the assumption is made that the health care provider is functioning at full capacity and in order to meet the extra demand for MRI needs to invest in an extra MRI device. Also assume that the health care provider negotiated earlier an internal transfer price or with the insurance companies a tariff of €75 for a CT image and €165 for an MRI image. Initially profitable performances need now to be compensated elsewhere in the health care organization to meet a balanced budget at the end of the year. The management which is accountable for a balanced budget at the end of the year has, under these circumstances, a disincentive to implement the MRI modality. This disincentive is due to inflexibility of the production factor MRI and CT capacity. In general innovations cause shifts in the demand for services. Implementation strategies consequently should be directed at increasing flexibility of production factors. The first step in the development of a framework for implementation strategies directed at increasing flexibility in production factors is the identification of fixed factors of production. Adang et al., 2005 developed a simple checklist specifically for the health care context to detect the proportion of fixed costs in the total production costs. Next, the fixed factors of production need to be disentangled into specific factors, like for example housing, capacity and personnel. Next a strategy directed at increasing flexibility of the specific production factor(s) should be developed. To illustrate this approach it will be applied to the diagnostic modality case. Employing the checklist to the diagnostic case shows that diagnostic capacity (both MRI and CT) is the major

The Implementation of Innovative Technologies in Healthcare

Table 2. Illustration of hurdle to implementation Production*

Fixed costs

Variable costs

CostPrice

New production*

New Fixed costs

Variable costs

CostPrice

CT

10000

200000

50

70

8000

200000

50

75

MRI

3500

350000

60

160

5500

700000

60

187

Cost in € *Production in number performances per year

inflexible factor. Now a tailored strategy directed at making diagnostic capacity more flexible needs to be developed. Making capacity more flexible urges for capacity planning. Depending on whether the production process is operating under decreasing returns to scale or increasing returns to scale a capacity planning strategy needs to be developed. Under increasing returns to scale capacity planning should be directed to increase production with the same inputs. Under decreasing returns to scale outsourcing production to meet extra demand can be a way of capacity planning without changing the size of operations in the organization. This means that the organization that is facing these inefficiencies needs to search for organizations that have excess MRI capacity available and are in need for extra capacity CT. Consequently this organization faces search costs that result in an increase in marginal cost. A regional market for diagnostic capacity could limit search costs and increase the adoption of innovations by limiting or even eliminating short run diseconomies. Such a market is not viable if all organizations in that particular market are operating at optimal capacity (i.e., constant returns to scale, meaning the most productive scale size). This means that some organizations need to operate under increasing returns to scale whereas others operate under decreasing returns to scale. These latter organizations produce above the optimal scale of operations and would improve their efficiency by downsizing or outsourcing, whereas for the organizations operating at increasing returns to scale it is efficient to increase production.

A minor inflexible factor of production (semifixed) was personnel. An MRI image takes more time than a CT image. An implementation strategy directed at personnel, being the inflexible production factor, could be for example teleradiology. According to Wachter the technical and logistic hurdles of remote teleradiology have been overcome (Wachter, 2006). This implementation strategy follows theoretically the same principles as in the capacity example. Other implementation strategies aiming at increasing flexibility of the labor force are ‘learning’ and ’making labor contracts more flexible’ (Alonso-Borrego, 1998).

RECOMMENDATIONS AND DISCUSSION In general firms’ ability to adjust to demand shifts due to innovations will determine fluctuations in the average cost of the firm. In a competitive market, a firms’ inability to cope with shifting demand will result in exit or major restructurings. In a non-market industry, like health care, similar rigidities may persist without exit or serious corrections, unless health care management identifies their sources and apply appropriate remedies (Cremieux et al., 2005). Implementation of cost-effective technologies (pharmaceuticals, technological devices, practice guidelines, etc.) in clinical practice is therefore a crucial process in healthcare, as in all industries. Health technology assessment has focused on long-term efficiency at societal level, ignoring short term inefficiencies for a specific hospital or

9

The Implementation of Innovative Technologies in Healthcare

other health care provider (Adang & Wensing, 2008). Cost-effectiveness analysis would be more informative if it extended the long run perspective with an additional short run perspective (Adang, 2008). Such an extended cost-effectiveness analysis provides not only insight in the cost-effectiveness of an innovation but also in the hurdles that need to be taken to implement the innovation. This leads to less co-existence of alternative treatment modalities, a higher adoption rate of the cost-effective modality and ultimately it leads to an increase in hospital value. Research on implementation processes in healthcare has focused mostly on perceived barriers for behavioral change and on educational interventions targeted at health professionals. This chapter aimed to contribute to both bodies of research. Key to solving this long run – short run efficiency discrepancies are implementation strategies directed towards making fixed factors of production more or entirely flexible. If all factors of production are entirely flexible the contrast between long run and short run efficiency will no longer exist and is then paradoxical. This chapter has been largely argumentative. There is little empirical evidence in support that the implementation strategies directed to increasing flexibility of factors of production lead to higher level and faster adoption compared to usual implementation strategies. Empirical research on this topic is important to quantify the impact of changes in production factors, and to examine potential side effects.

REFERENCES Adang, E. M. (2005). Efficiency of hospitals [in Dutch] [ESB]. Economic Statistical Bulletin, 4451, 40–41. Adang, E. M., & Voordijk, L., Wilt van der, G.J., & Ament, A. (2005). Cost-effectiveness analysis in relation to budgetary constraints and reallocative restrictions. Health Policy (Amsterdam), 74, 146–156. doi:10.1016/j.healthpol.2004.12.015 10

Adang, E. M. M. (2008). Economic evaluation in healthcare should include a short run perspective. The European Journal of Health Economics, 9, 381–384. doi:10.1007/s10198-007-0093-y Adang, E. M. M., & Wensing, M. (2008). Economic barriers to implementation of innovations in healthcare: Is the long run-short run efficiency discrepancy a paradox? Health Policy (Amsterdam), 2-3, 256–262. Aletras, A. (1999). Comparison of hospital scale effects in short run, long run cost functions. Health Economics, 8, 521–530. doi:10.1002/(SICI)10991050(199909)8:63.0.CO;2G Allais, M. (1977). Theories of general economic equilibrium and maximum efficiency. In G. Schwodiauer (Ed.), Equilibrium and Disequilibrium in Economic Theory (pp.129-201). Dordrecht: Reidel. Alonso-Borrego, C. (1998). Demand for labour inputs and adjustment costs: Evidence from Spanish manufacturing firms. Labour Economics, 5, 475–497. doi:10.1016/S0927-5371(98)00011-6 Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some models estimating technical and scale inefficiencies in data envelopment analysis. Management Science, 30, 1078–1092. doi:10.1287/mnsc.30.9.1078 Baumol, W., Panzar, J. C., & Willig, R. D. (1988). Contestable markets and the theory of industry structure. San Diego: Harcourt Brace Jovanovich. Briec, W., Kerstens, K., & VandenEeckhaut, P. (2004). Nonconvex technologies and cost functions: Definitions, duality, and nonparametric tests of convexity. Journal of Economics, 81, 155–192. doi:10.1007/s00712-003-0620-y Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2, 429–444. doi:10.1016/0377-2217(78)90138-8

The Implementation of Innovative Technologies in Healthcare

Coelli, T., Prasada Rao, D. S., & Battese, G. E. (2004). An introduction to efficiency and productivity analysis. Norwell, MA: Kluwer Academic Publishers.

Gertler, P., & Walman, D. (1992). Quality-adjusted cost functions and policy evaluation in nursing home industry. The Journal of Political Economy, 100, 1232–1256. doi:10.1086/261859

Cremieux, P. Y., Ouellette, P., & Rimbaud, F. (2005). Hospital cost flexibility in the presence of many outputs: A public-private comparison. Health Care Management Science, 8, 111–120. doi:10.1007/s10729-005-0394-6

Gold, M. R., Siegel, J. E., Russel, L. B., et al. (1996). Cost-effectiveness in health and medicine. New York: Oxford University Press.

Davies, L., Coyle, D., & Drummond, M., & The EC Network. (1994). Current status of economic appraisal of health technology in the European community: Report of the network. Social Science & Medicine, 38, 1601–1607. doi:10.1016/02779536(94)90060-4 Dor, A. (1989). The costs of medicare patients in nursing homes in the United States: A multiple output anapysis. Journal of Health Economics, 8, 253–270. doi:10.1016/0167-6296(89)90021-0 Drummond, M. F., Sculpher, M. J., Torrance, G. W., O’Brien, B., & Stoddart, G. L. (2005). Methods for the economic evaluation of health care programmes. Oxford, UK: Oxford University Press. Eddama, O., & Coast, J. (2008). A systematic review of the use of economic evaluation in local decision-making. Health Policy (Amsterdam), 86, 129–141. doi:10.1016/j.healthpol.2007.11.010 Elbasha, E. H., & Messonnier, M. L. (2004). Cost-effectiveness analysis and healthcare resource allocation: Decision rules under variable returns to scale. Health Economics, 13, 21–35. doi:10.1002/hec.793 Farrel, M. (1959). The convexity assumption in the theory of competitive markets. The Journal of Political Economy, 67, 377–391. doi:10.1086/258197 Farrel, M. J. (1975). The measurement of productive efficiency. Journal of the Royal Statistical Society. Series A (General), 3, 253–290.

Hartman, M., Prins, M., & Swinkels, O. Q. J. (2002). Cost-effectiveness analysis of a psoriasis care instruction programme with dithranol compared with UVB phototherapy and inpatient dithranol treatment. The British Journal of Dermatology, 147, 538–544. doi:10.1046/j.13652133.2002.04920.x Hoffmann, C., & Graf von der Schulenburg, J. M. (2000). The influence of economic evaluation studies on decision making: A European survey. The EUROMET group. Health Policy (Amsterdam), 52, 179–192. doi:10.1016/S01688510(00)00076-2 Hovels, A., Heesakkers, R. A. M., & Adang, E. M. M. (in press). Cost-effectiveness analysis of MRI with a lymph node specific contrast agent for the detection of lymph node metastases in patients with prostate cancer and intermediate or high risk of lymph node metastases. Radiology. Johannesson, M. (1995). Economic evaluation of health and policymaking. Health Policy (Amsterdam), 33, 179–190. doi:10.1016/01688510(95)00716-6 Kallapur, S., & Eldenburg, L. (2005). Uncertainty, real options, and cost behavior: Evidence from Washington state hospitals. Journal of Accounting Research, 43, 735–752. doi:10.1111/j.1475679X.2005.00188.x Kass, D. (1987). Economies of scale, scope in the provision of home health services. Journal of Health Economics, 6, 129–146. doi:10.1016/01676296(87)90003-8

11

The Implementation of Innovative Technologies in Healthcare

Katz, M. L., & Rosen, H. S. Microeconomics. Boston: Irwin. Lessard, C. (2007). Complexity and reflexivity: Two important issues for economic evaluation in healthcare. Social Science & Medicine, 64, 1754– 1765. doi:10.1016/j.socscimed.2006.12.006 Neumann, P. J., Rosen, A. B., & Weinstein, M. C. (2005). Medicare and cost-effectiveness analysis. The New England Journal of Medicine, 353(14), 1516–1522. doi:10.1056/NEJMsb050564 Roberts, R., Frutos, P., & Ciavarella, G. (1999). The distribution of variable vs. fixed costs of hospital care. Journal of the American Medical Association, 281, 644–649. doi:10.1001/ jama.281.7.644 Smet, M. (2002). Cost characteristics of hospitals. Social Science & Medicine, 55, 895–906. doi:10.1016/S0277-9536(01)00237-4 Smith, P. C., & Street, A. (2005). Measuring the efficiency of public services: The limits of analysis. Journal of the Royal Statistical Society, 168, 401–417. doi:10.1111/j.1467-985X.2005.00355.x Tone, K., & Sahoo, B. K. (2003). Scale, indivisibilities, and production function in data envelopment analysis. International Journal of Production Economics, 84, 165–192. doi:10.1016/ S0925-5273(02)00412-7 Vitaliano, D., & Toren, M. (1994). Cost and efficiency in nursing homes: A stochastic frontier approach. Journal of Health Economics, 13, 281–300. doi:10.1016/0167-6296(94)90028-0 Wachter, R. M. (2006) International teleradiology. N Engl J Med., 16, 354(7), 662-663.

Weinstein, M. C. (2008). How much are Americans willing to pay for a quality adjusted life year? Medical Care, 46, 343–345. doi:10.1097/ MLR.0b013e31816a7144 Zhu, J. (2003). Quantitative models for performance evaluation and benchmarking. New York: Springer Science +Business Media, Inc.

KEY TERMS AND DEFINITIONS Cost-Effectiveness Analysis: Tests the null hypothesis that the mean cost-effectiveness of one health care innovation is different from the cost-effectiveness of some competing alternative often common practice. Diseconomies of Scale: When average costs rise with the output level. Efficiency: How well an organization is performing with regard to optimizing its output for given inputs or its inputs for given output. Implementation: The way an objective is realized, here refers to an innovation adopted by the specific stakeholder(s). Long Run: A period where all factors of production are variable. Returns to Scale: An economic term that describes what happens if the scale of production changes. Short Run: A period where at least one of the factors of production is fixed. Technology in Health Care: A broad concept that deals with innovations in pharmaceuticals, diagnostics, disease management, guideline development and so on.

This work was previously published in Handbook of Research on Information Technology Management and Clinical Data Administration in Healthcare, edited by Ashish N. Dwivedi, pp. 321-332, copyright 2009 by Medical Information Science Publishing (an imprint of IGI Global).

12

13

Chapter 1.2

Risks and Benefits of Technology in Health Care Stefane M. Kabene University of Western Ontario, Canada Melody Wolfe University of Western Ontario, Canada

ABSTRACT The integration of technology into health care has created both advantages and disadvantages for patients, providers, and healthcare systems alike. This chapter examines the risks and benefits of technology in health care, with particular focus on electronic health records (EHRs), the availability of health information online, and how technology affects relationships within the healthcare setting. Overall, it seems the benefits of technology in health care outweigh the risks; however, it is imperative that proper measures are taken to ensure successful implementation and DOI: 10.4018/978-1-60960-561-2.ch102

integration. Accuracy, validity, confidentiality, and privacy of health data and health information are key issues that must be addressed for successful implementation of technology.

ELECTRONIC HEALTH RECORDS Technological advances in information and communication technologies (ICT) and computing have made way for the implementation of electronic health records (EHRs), the comprehensive compilation of health care provided to an individual over their lifetime —an exciting and impressive accomplishment. Despite the vast possibilities and efficiencies that EHRs can potentially

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Risks and Benefits of Technology in Health Care

offer, their implementation into existing healthcare systems poses some potentially deterring and serious risks, such as confidentiality breaches, identity theft, and technological breakdowns and incompatibilities. Therefore, electronic records should be not hastily integrated into healthcare systems without proper precautions.

Advantages Electronic records offer many advantages over conventional paper-based methods of recording patient data. The comprehensiveness of EHRs can help to bridge the geographic and temporal gaps that exist when several clinicians who are geographically dispersed treat the same patient. It is extremely important that all clinicians are aware of past and current medical histories when one patient is treated by several healthcare providers (Mandl, Szolovits, & Kohane, 2001). Since paperbased records are location specific, information contained in one record may differ substantially from records kept in another area or by another provider. When various specialists treat the same patient, patient communication is often hindered, as it can be extremely difficult and time consuming to share patient records between providers using conventional methods (for example, by phone, fax or mail, or physically transporting the record from location to location). Electronic health records, however, enable comprehensive databases of information to be viewed and used by authorized users when they need it and where they need it. Greater efficiency in accessibility of patient information is thus made possible by the use of electronic records. Accessibility allows for a faster transfer of medical history in a medical emergency or when visiting a new doctor, and also allows researchers and public health authorities—with the permission and consent of the patient—to efficiently collect and analyze updated patient data. Such access is imperative in emergency situations, and also allows public health officials to easily

14

conduct outbreak and incident investigations that may help control epidemics and pandemics, such as SARS, Listeriosis, or new strains of influenza. Accessibility also enables health care providers to reduce costs associated with duplicating tests, since providers have access to already performed test results (Myers, Frieden, Bherwani, & Henning, 2008). Additionally, clerical activities such as appointment reminders and notification of laboratory results can be handled electronically, resulting in greater efficiency and reduced human error. EHRs can also be equipped with authentication systems, a major guard against security breaches. Patients may be especially wary of having their personal health information part of a comprehensive database because they are unsure as to who will have access to their medical records. Authentication systems allow for the imposition of various security levels, providing greater control over access to personal information such as immunization records and diagnostic test results. Conversely, paper-based medical records allow healthcare staff to access any part of a patient’s medical records. By applying authentication and role-based access to EHRs, personnel such as secretaries and clerical staff will only have access to necessary information (such as that needed for scheduling appointments or providing reminders of scheduled visits) (Myers et al., 2008). In case of an emergency, however, it is possible to develop policies that allow medical professionals to override the protection barriers and gain immediate access to all medical information (Mandl et al., 2001). An additional security feature is accountability, which enables the system to track input sources and record changes. Accountability systems provide an audit trail that can help to eliminate security breaches and, at the very least, track user activities to ensure their appropriateness, authorization, and ethicality (Myers et al., 2008). Despite the impressive advantages EHRs offer, one must recognize the trade off that exists between accessibility and confidentiality. As noted by Rind et al. (1997, p. 138) “It is not always

Risks and Benefits of Technology in Health Care

possible to achieve both perfect confidentiality as well as perfect access to patient information, whether information is computerized or handwritten.” Confidentiality, among other issues, must be considered in order to utilize the EHR system to its fullest potential.

Disadvantages One potential deterrent to full implementation of the electronic health record is compatibility and interoperability across different health information systems. Electronic health records require a standardized system and technology to promote transfer, input and compilation of data from multiple sources; unfortunately, these standards are not easily achieved due to the complexity of linking disparate and often older legacy systems and the incumbent costs of doing so. Thus, despite the potential of EHRs to increase communication between practitioners, and practitioners and patients, the various already existing computer systems pose a major road block. Until a standard model for secure data transmission and linkage is efficiently in place, EHRs will remain fragmented and inconsistent. Of course, the most serious danger in widespread EHR implementation is the potential for security breaches. Electronically-bound information always comes with the risk of being exposed to inappropriate parties; “No matter how sophisticated security systems become, people will always manage to defeat them” (Mandl et al., 2001, p. 285). Due to the wealth of information contained in the EHR, they are an obvious target to for hacking and identity theft. Additionally, because electronic theft can occur in a variety of contexts—including the comfort of one’s home— electronic theft may be harder to track or detect and require less physical effort and planning. In contrast, physically stealing medical records from an office is difficult to accomplish successfully, since most medical offices implement effective security measures.

Aside from security breaches, there are core problems associated with using computers to house mass amounts of important and often sensitive information. Technological issues arise throughout the lifetime of a computer system. Not only are computer systems created and updated rapidly, there are few systems that last the entire lifespan of a person. Such technological issues raise questions regarding the safety of transferring medical information from system to system, as any breakdown may cause record loss, or pose a potential security breach (Mandl et al., 2001). The reaction of patients to confidentiality and privacy issues pose a major concern as well. As previously mentioned, patients have expressed concerns about who is seeing their medical information and for what purpose it is being used. Concerns also arise when patients are unable to see their own records, but secondary sources can view these records in an unregulated manner. Consequently, it has been noted that “they might fail to disclose important medical data or even avoid seeking medical care because of concern over denial of insurance, loss of employment or housing, or stigmatisation and embarrassment” (Mandl et al., 2001, p. 284). The failure to disclose important medical information can affect patient health and increase the risk of misdiagnosis. Since privacy is important for most patients when it comes to their medical records and health history, one option is to allow patients the right to decide who can examine and alter specific parts of their records. Although this seems like a plausible solution, its complexity can create problems that may lead to inferior care and uninformed practitioners (e.g., if certain portions of the history are restricted). Thus, granting patients the right to monitor their own privacy is not a solution in and of itself.

Discussion It is very likely that EHRs will be the key to linking disparate pieces of a patient’s medical record into a unified database. The comprehensiveness

15

Risks and Benefits of Technology in Health Care

and accessibility of EHRs will mend the gap in provider-provider communication that currently exists due to temporal and geographic differences in paper-based records. However, in order for these possibilities to actualize, proper precautions must be taken, and specific policies enforced. Authentication and audit trails of EHRs allow various health care professionals different levels of access to patient records, as well as a mechanism to track the use and changes of each patient’s file. Aside from these security enhancing features, various preventative actions should be taken in order to eliminate the risks associated with EHR use. When an electronic record system is put in place, staff and healthcare providers must use the system properly to obtain beneficial outcomes. Accordingly, education and training are also necessary components in enabling individuals to understand and maximize EHR system abilities. As well, a “change agent,” or expert on electronic record systems, could help successful implementation and adoption of digital medical records. This change manager would be in charge of training staff, monitoring security and confidentiality, and enforcing policy. In addition to educating employees, several preventative measures should be embedded into EHR systems; for example, passwords allowing for authentication and limiting the access of each person on an individual, role-based basis. Since different information is needed by various personnel, these passwords would allow doctors to access a wider range of information than, for instance, a secretary. Encryption, or the process of making the data unreadable unless the user possesses suitable authorization, is another preventative measure which hinders hackers from obtaining confidential information. Furthermore, it should be noted that already existing policies aid the privacy of an individual’s electronic health record. In Canada, federal privacy legislation under the Personal Information Protection and Electronic Documents Act (PIPEDA) ensures privacy via rules for the collection, use, and disclosure of

16

personal information (Taylor, 2003). In using EHR systems, health care providers must ensure that they comply with the PIPED Act in order to eliminate the risks of confidentiality and privacy breaches. With regards to the issue of authentication, access to a patients’ medical record for reasons incongruent with the individual’s health care should be a decision that only the patient themselves can make. Although public health and government research could significantly benefit from access to patient medical records, due to issues of privacy and confidentiality, it is advised that every patient should possess the right to deny access to their medical records for public use. Additionally, patients who choose to provide personal health information to government and public health associations should have the ability to limit what information is exposed. A technology such as “time keys” could be utilized in alignment with electronic records in allowing patients to give an organization access to their records only for a specified amount of time (Mandl et al., 2001). Since patients visit a number of healthcare service providers over the course of their life, each provider has a different computer system that encodes and stores patient data according to different levels and standards—which poses a great challenge when attempting to unify all patient information into a single database. It has not yet been fully understood how one system will capture all data from various physicians into a standard computer form or interface that shares the information with other computers (i.e. similar to the web browser and access to disparate web pages) or whether only a minimum data set will be captured in a central repository. In the future, universal coding and standardization of data must be implemented in order to manage fragmented pieces of patient data. Due to vast language barriers and numerous medical terminologies, universal coding is a challenge; however, it is hopeful that future advancements in technology can develop

Risks and Benefits of Technology in Health Care

systems able to appropriately incorporate EHRs and shared data repositories (Mandl et al., 2001).

ONLINE MEDICAL INFORMATION Demographics Fifty two million adults have used the Internet to access health or medical information through an estimated 20, 000 to 100,000 health related websites (Diaz et al., 2002). In general, 62% of Internet users go online for medical advice, and most of these Internet users are women between the ages of 50 and 64 (Rice, 2006). Searching for health information increases with education, as well as with years of Internet experience and access to a broadband connection (Rice, 2006, p 10). In a survey conducted by Diaz et al., (2002) 54% of respondents revealed that they used the Internet for health information, and a higher proportion of patients who used the Internet were college educated with an annual household income greater than $50 000.

Advantages Technological advancement—in particular, the use of Internet research for medical information—has a major influence on health care for both patients and providers alike. Because the Internet provides such accessible health information, individuals are often encouraged to seek out information regarding their own health, as well as about the health of others. Additionally, medical information on the Internet allows individuals to become exposed to a wide array of health information and become involved in their own personal health (Rice, 2006). In North America, the use of Internet research is an increasingly popular strategy for obtaining diagnostic information. In October 2002, 73 million American adults used the Internet to find health care information, and by November of 2004, this number had risen to 95 million (Rice, 2006).

There are many advantages associated with the use of the Internet for medical research, such as easy access to a wide array of information, personalized information that is uncontaminated by medical jargon, and anonymity while searching for health information (Rice, 2006). Additionally, online websites and support groups provide information, support, acceptance, and a sense of understanding to patients and their loved ones (Rice, 2006). Since the Internet allows for immediate access to an incredible amount of information directed at both health care providers and patients, it provides a sense of privacy, convenience, and perspective (Cotton & Gupta, 2004). The Internet also allows users to ask awkward, sensitive, or detailed questions without the risk of facing judgment, scrutiny, or stigma (Cotton & Gupta, 2004). Consequently, doctors are beginning to see a new type of patient: a patient with sharp intelligence and curiosity who knows how to utilize and benefit from information available online (Ferguson, 2000). Perhaps the most significant benefit online medical information offers is its ability to prevent long hospital wait-times for minor issues. Patients suffering from trivial matters who are simply seeking reassurance from doctors can go online and seek information without going to the hospital. Additionally, patients can gain medical knowledge and insight that may teach them to distinguish between symptoms that seem minor, but are actually serious. The Internet allows for patients to consult various perspectives on illness and treatments rather than visiting several doctors for multiple opinions. For example, if a patient is given a type of medicine to treat a thyroid problem, they can go online and research it, which enables them to gain perspective on their own medical issue. Since online medical information is free, it is fairly accessible to patients in tough financial positions. This is especially advantageous for countries such as the United States, in which private health care is extremely expensive. Online health information provides citizens of the United States with immediate information

17

Risks and Benefits of Technology in Health Care

about health problems; such knowledge may help citizens decide whether it is necessary to visit a doctor or hospital.

Disadvantages Although there are many advantages associated with the ability to research health-related information online, such as increased patient medical knowledge, it is imperative to consider the various disadvantages associated with hastily researching medical information on anonymous databases. For example, accurate information on credible websites may be hidden behind complicated medical language, and the quality and authenticity of available information is often questionable. At times, the vast amount of online information can be confusing, making it difficult for a patient to filter out what is important. Studies of postsurgery patients revealed that 83% had difficulties understanding online health information, and one third felt that the retrieved information was overwhelming (Rice, 2006). A major problem occurs when patients try to diagnose and treat a potentially serious medical condition without consulting a doctor. Research indicates that 11% of those using the Internet for medical information revealed that they did not discuss this information with their doctor (Diaz et al., 2002). According to Berland et al. (2001), less than one quarter of search engines’ first pages of links lead to relevant content; therefore, accessing health information using search engines and simple search terms is not sufficient. Additionally, Internet usage is often hindered by navigational challenges due to poor design features such as disorganization, technical language, and lack of permanent websites (Cline & Haynes, 2001). Support groups may also be problematic as they may distribute misleading information. Considering support information is often based on personal experience, it may lack the knowledgeable and experienced perspective developed by health care professionals who are trained to distinguish

18

among resources, determine information accuracy, and examine the quality of information (Cotton & Gupta, 2004). As most of the information available online lacks peer review, it may be misleading and inaccurate (Rice, 2006). Researchers consistently identify problems with the quality of online health information (Rice, 2006), and very few sites pass through traditional editorial processes or display sources of authorship or origin (Diaz et al., 2002). Much of this uncontrolled information commonly strays from recognized safety standards, is not updated regularly, and presents limited advice on avoiding dangerous drug interactions (Rice, 2006). Despite the existence of many reliable medical and health related websites, several reviews have demonstrated that patients may encounter potentially misleading or inaccurate information when navigating the Internet in search of them. Some health seekers even fear that confidentiality breaches will track their Internet searches and unfavorably reveal their private information to insurance companies and employers (Cotton & Gupta, 2004). Unfortunately, there exists a digital divide or gap between those who can access online resources and those who cannot. Internet health information remains inaccessible to large and specific parts of the population (lower income, those in rural communities with limited Internet access), and physically impaired and disabled persons are often at a disadvantage in a networked society (Rice, 2006). Additionally, those individuals who often cannot access the tools to seek health information online are usually those with preventable health problems or without health insurance (Rice, 2006). Sadly, it seems as though individuals who would benefit the most from online medical information are the least likely to acquire it.

Discussion As noted, there are a number of advantages and disadvantages associated with researching medi-

Risks and Benefits of Technology in Health Care

cal information online. The advantages include access to a wide array of information, personalized information that is easy to understand, and anonymous help. Help is provided by support groups, and allows for the maintenance of privacy and anonymity. The Internet also allows patients to remain current on information regarding their own health or the health of others. However, some disadvantages include medical language barriers on legitimate websites, unequal Internet access, self diagnosis, and inaccurate information (Rice, 2006). Support groups may rely too heavily on personal experience, and consequently distribute misleading information (Culver, Gerr, & Frumkin, 1997). The risk in researching medical knowledge online is that much available information lacks peer review and accuracy. Confidentiality of information searching online is also a risk to consider. With all issues in consideration, it is apparent that the benefits of researching medical information, if used with the proper discretion, can have an invaluable impact on the patient. As indicated in the demographics summarized above, it is evident that the majority of individuals currently researching medical information are well educated. Thus most individuals/patients who seek information online have the ability to understand the information, know the disadvantages of online data, and can utilize the online information to help address their own medical needs. The Internet also allows patients to seek medical advice immediately, easily, and at their convenience. Because Internet usage is on the rise, it is anticipated that the digital divide will narrow in scope, and more individuals will gain access to the information available on the Internet. Individuals who have not previously been able to take advantage of online information will have the opportunity to do so. However, not all individuals who gain access to online health information have the understanding to apply it; therefore, medical information on the Internet should be correct and user friendly. It is evident that there is already an increasing trend in search engines providing more valid, accurate

information. For example, “Google” has a more advanced search tool “Google Scholar,” which provides scholarly articles published by credible sources. Additionally, many medical websites inform patients of serious symptoms by stating that if certain symptoms are present, patients should seek medical attention immediately. If online medical information is credible and user-friendly, it can be extremely beneficial to patients and providers. Doctors are forced to stay up to date and well informed about all advances in medicine, and by using online medical information, can do so with ease. Patients become well versed in their medical issues, and can provide their doctors with insightful questions. Additionally, patients expect doctors to know about the information available online and answer their questions; they have the opportunity to print articles, bring information directly to their physicians, or email their doctor for a professional opinion. However, if the information online does not maintain a certain level of accuracy and validity, negative implications may occur. Overall, the Internet should be used as a supplementary tool. Professional opinions should always be used as the primary source.

TECHNOLOGY AND RELATIONSHIPS Technological progressions impact all relationships within healthcare, including those between patients and healthcare providers, patients and other patients, as well as those between various healthcare providers, including doctors, nurses, and specialists. In other words, the presence of information technology in health care has had a profound effect on relationship dynamics within the health care system. The relationships that exist between physicians and patients are important and central to the quality and efficiency of patient care—information technology (IT) affects both aspects of this relationship significantly. With the

19

Risks and Benefits of Technology in Health Care

emergence and growing prevalence of technology in health care, patients have been able take a more active role with regards to their health. Through increased access to information, the possibility of faster communication through e-mail beginning to emerge, and the existence of online patient portals, the presence of IT has had many positive effects on the patient-physician relationship (Wald, Dube, & Anthony, 2007).

Advantages The empowerment patients acquire by having access to their medical records along with online health researching enables them to intelligently discuss health issues with their clinician. In doing so, patients become a de facto part of the healthcare team. When properly counseled and mentored by healthcare professionals, individuals can potentially become true “clinical assistants” in their own health; they have more knowledge of their own problems and changes that enable clearer, more effective, and more efficient patientprovider communication. Technology can also aid patient decision making. For example, giving patients a visual “map” of the treatment options has been shown to help patients remember treatment options and their differential impacts with more accuracy, enabling patients to make faster, more appropriate decisions that increase the overall chances of a positive outcome (Panton et al. 2008). Along with the increase in online health resources, electronic communication via e-mail between patients and doctors is also increasing (Ferguson, 2002). For example, Weiner and Biondich (2006) found that 95% of surveyed patients who used e-mail communication (about 16.5% of the survey group) felt it was more efficient than the telephone, and that e-mail proved especially useful for patients with “sensitive issues” (Houston, Sands, Jenckes, & Ford, 2004; Weiner & Biondich, 2006). The patient-patient relationship is also facilitated in a positive manner by technology, and is

20

often accomplished through online patient helpers, who oversee medical communication between patients. Online patient helpers are individuals who provide help for other patients via the Internet (Ferguson, 2000). For example, blogs and online support groups provide an environment in which helpers can serve as a valuable resource by assisting patients with medical information, encouraging discussion, and facilitating support. It should be noted that these online resources are best used to supplement the patient-physician relationship by encouraging patient-patient relations, and should by no means be the sole source of health information for any patient (Ferguson, 2000).

Disadvantages While the provision of online health information can strengthen the patient-physician relationship, it can also hinder it. Since much of the information online is not credible, physicians may be required to spend unnecessary amounts of time separating fact from fiction with their patients (Dugdale, Epstein & Pantilat, 1999). Consequently, patients are putting pressure on physicians to help sort through and clarify their online findings—which is a timely and effortful endeavor for many physicians (Ferguson, 2002). Implementation of electronic health records (EHRs), and the increased reliance on technological devices in general, may also hinder the patient-physician relationship, as time-consuming, and potentially distracting, data entry can take valuable time away from patient-provider relations. The increasing use of IT in health care may be so time consuming that it actually diminishes the patient-physician relationship (Weiner & Biondich, 2006). Although patient helpers encourage and facilitate patient-patient relationships, many patients do not have the formal training to differentiate between credible and non-credible information. Without this knowledge, some patients put too much reliance on online helpers, which can jeopardize both their health and the patient-physician

Risks and Benefits of Technology in Health Care

relationship (Ferguson, 2000). Because some online sources are certainly not credible, and many patients are insufficiently equipped to differentiate between accurate and bogus sources, many doctors discourage the use of such information.

Discussion The increasing presence of IT in health has profound effects on patient-physician and patientpatient relationships. With the surplus of health information available online, patients are able to actively participate in the patient-physician relationship and provoke discussion that serves in the best case to help patients make more informed decisions when choosing medical options. The growing acceptance of e-mail communication between patients and health professionals is giving patients greater access to medical advice. Also, online patient helpers give patients a portal for communicating to others in cyberspace that increases knowledge, promotes coping, and forms patient-patient relationships. However, if online medical information is used irresponsibly, it can have detrimental effects on one’s health, and the patient-physician relationship. The presence of IT can waste time, and increase patient acceptance of non-credible information. If one substitutes online information for professional, clinical advice, the patient-physician relationship deteriorates. Negligence to seek out physician care can result in unfortunate health consequences. Overall, if these disadvantages are lessened or alleviated, than the presence of IT on the patient-physician relationship will be beneficial. The increasing prevalence of e-mail communication between patients and doctors allows for physicians to be more accessible to their patients and vice versa. If used correctly, e-mail communication allows the physician-patient relationship to be more efficient—an important factor given the time pressures placed on today’s physicians. Rapid communication also allows

patients who are experiencing extremely stressful medical issues to be in immediate contact with their physician. E-mail also helps to reduce the apprehension of a face-to-face interaction that is often experienced by many patients, and allows doctors to answer general inquiries that do not require face-to-face contact. This may make booking a doctor’s appointment easier for those who really need it (Wald et al., 2007). However, the inappropriate use of online patient-provider communication can be wasteful. It may lead to e-mail overload; patients may ask very similar questions that physicians would have to answer repeatedly. As well, patients may take advantage of such constant contact and abuse communication privileges (Weiner & Biondich, 2006). Technology has enabled the emergence of patient-patient relationships, and these relationships can offer patients much needed emotional and psychological support when dealing with various illnesses (Mandl et al., 2001). Although patient-patient relationships can be very beneficial, individuals should be cautious before entering a relationship for anything more than support. If a patient were taking specific medical advice from their patient-patient relationship, it should be discussed with their physician, as the credibility of the average Internet user to be giving medical advice is most likely questionable. It is also important to keep in mind that diseases can affect patients in many different ways, so a patient giving advice may not know what is best for someone else (Culver et al., 1997). Sometimes patients feel more comfortable conversing with patient helpers as opposed to doctors because there is a mutual understanding: each individual has experienced the same illness. Online patient-patient relationships can be especially helpful for patients who are dealing with sensitive issues, such as AIDS. Given the social stigma surrounding the disease, being able to do anonymous research and talk to others in the same situation can help to reduce the embarrassment that infected individuals may

21

Risks and Benefits of Technology in Health Care

feel. Patients can offer personal experiences about the disease and give advice (Wald et al., 2007). It should be mentioned that individuals seeking medical advice online may feel the need for faceto-face communication in person from a professional as opposed to a cyber friend—especially when seeking treatment.

CONCLUSION In order to combat issues threatening the successful integration of technology in health care, certain steps must be taken. One of the most important considerations when implementing electronic health records is the provision of proper training to all employees. For example, secretaries in medical offices must be extremely knowledgeable in all PIPEDA legislation, and must demonstrate proper system use to preserve privacy and confidentiality of patient information. Firewalls and other prevention measures must be developed and updated frequently to ensure that the records are private, confidential, and secure. A number of important implementations should be undertaken if the availability of online medical information is to be used to its fullest potential. First, governments and credible medical institutions should combine efforts in creating a large, easily understandable database of medical information accessible to laypersons (e.g. the National Library of Medicine’s Medline Plus web site, see http://www.nlm.nih.gov/medlineplus/). Such a database will provide credible information that is current and understandable from a reliable source. Second, doctors need to play a role in the information put forth on the Internet by researching it themselves and posting correct articles and information for their patients. Last, there should be a stamp or certification that ensures health information posted on various websites is legitimate (e.g. the HONcode certification now provided voluntarily by the Health on the Net

22

Foundation, see http://www.hon.ch/). As well, more stringent efforts to monitor non-credible information should be taken. A potential way of reducing harmful, non-credible online information is to implement legislation that restricts the release of medical information from those without formal training or licensure to certain groups (i.e. registered support groups). In addressing healthcare relationships affected by technology, it is important to consider the potential for patient e-mails to affect physician stress levels and time management; email overflow puts pressure on physicians. The creation of a “frequently asked questions” page would help solve the issue of redundant e-mails, whereas the establishment of guidelines for users may help reduce e-mail abuse. Additionally, doctors should inform patients about the potential harms online patient helpers may present. Patients should be aware that much information provided by online helpers is subjective, and may not be credible or valuable in their particular case. As well, specific patient-helper sites should make their purpose known to all visitors. Technology undoubtedly has the potential to benefit healthcare; the IT advances discussed above will improve the efficiency, accuracy, and effectiveness of health care systems. Although the benefits are shadowed by risks, EHRs allow for increased and more efficient patient care; researching medical information online, if used properly, can provide a wealth of invaluable information for patients; and technology has allowed patient-physician and patient-patient relationships to become more effective. If the negative effects on these relationships can be mitigated via proper precautions and responsibility, the benefits will be maximized. Although the medical system is able to function without advancements in technology, technology allows for innovation and efficiencies that can greatly improve and revamp the way health care systems work.

Risks and Benefits of Technology in Health Care

REFERENCES Berland, G. K., Elliott, M. N., Morales, L. S., Algazy, J. I., Kravitz, R. L., & Broder, M. S. (2001). Health information on the Internet: Accessibility, quality, and readability in English and Spanish. Journal of the American Medical Association, 285, 2612–2621. doi:10.1001/jama.285.20.2612 Cline, R. J. W., & Haynes, K. M. (2001). Consumer health information seeking on the Internet: The state of the art. Health Education Research, 16, 671–692. doi:10.1093/her/16.6.671

Houston, T. K., Sands, D. Z., Jenckes, M. W., & Ford, D. E. (2004). Experiences of patients who were early adopters of electronic communication with their physician: satisfaction, benefits, and concerns. The American Journal of Managed Care, 10, 601–608. Mandl, D. K., Szolovits, P., & Kohane, S. I. (2001). Public standards and patients’ control: How to keep electronic medical records accessible but private. British Medical Journal, 322, 283–287. doi:10.1136/bmj.322.7281.283

Cotton, S. R., & Gupta, S. S. (2004). Characteristics of online and offline health information seekers and factors that discriminate between them. Social Science & Medicine, 59, 1795–1806. doi:10.1016/j.socscimed.2004.02.020

Myers, J., Frieden, T. R., Bherwani, K. M., & Henning, K. J. (2008). Ethics in public health research: Privacy and public health at risk: Public health confidentiality in the digital age. American Journal of Public Health, 98, 793–800. doi:10.2105/ AJPH.2006.107706

Culver, J. D., Gerr, F., & Frumkin, H. (1997). Medical information on the internet: A study of an electronic bulletin board. Journal of General Internal Medicine, 12(8), 466–470. doi:10.1046/ j.1525-1497.1997.00084.x

Panton, R. L., Downie, R., Truong, T., Mackeen, L., Kabene, S., & Yi, Q. L. (2008). A visual approach to providing prognostic information to parents of children with retinoblastoma. PsychoOncology, 18(3), 300–304. doi:10.1002/pon.1397

Diaz, J. A., Griffith, R. A., Ng, J. J., Reinert, S. E., Friedmann, P. D., & Moulton, A. W. (2002). Patients’ use of the Internet for medical information. Journal of General Internal Medicine, 17, 180–185. doi:10.1046/j.1525-1497.2002.10603.x

Rice, R. E. (2006). Influences, usage and outcomes of Internet health information searching: Multivariate results from the pew surveys. International Journal of Medical Informatics, 75, 8–28. doi:10.1016/j.ijmedinf.2005.07.032

Dugdale, D. C., Epstein, R., & Pantilat, S. Z. (1999). Time and the patient-physician relationship. Journal of General Internal Medicine, 14(1), 34–40. doi:10.1046/j.1525-1497.1999.00263.x

Rind, D. M., Kohane, I. S., Szolovits, P., Safran, C., Chueh, H. C., & Barnett, G. O. (1997). Maintaining the confidentiality of medical records shared over the Internet and the World Wide Web. Annals of Internal Medicine, 127, 138–141.

Ferguson, T. (2000). Online patient-helpers and physicians working together: A new partnership for high quality health care. British Medical Journal, 321, 1129–1132. doi:10.1136/bmj.321.7269.1129 Ferguson, T. (2002). From patients to end users. British Medical Journal, 324, 555–556. doi:10.1136/bmj.324.7337.555

Smith, R. (2000). Getting closer to patients and their families. British Medical Journal, 321, Retrieved from http://www.bmj.com.proxy1.lib. uwo.ca:2048/cgi/reprint/321/7275/0. Taylor, S. (2003). Protecting privacy in Canada’s private sector. Information Management Journal, 37, 33–39.

23

Risks and Benefits of Technology in Health Care

Wald, H. S., Dube, C. E., & Anthony, D. C. (2007). Untangling the web – the impact of Internet use on health care and the physician-patient relationship. Patient Education and Counseling, 68(3), 218–224. doi:10.1016/j.pec.2007.05.016

Weiner, M., & Biondich, P. (2006). The influence of information technology on patient-physician relationships. Journal of General Internal Medicine, 21(Suppl 1), S35–S39. doi:10.1111/j.15251497.2006.00307.x

This work was previously published in Healthcare and the Effect of Technology: Developments, Challenges and Advancements, edited by Stéfane M. Kabene, pp. 60-71, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

24

25

Chapter 1.3

The Effectiveness of Health Informatics Francesco Paolucci The Australian National University, Australia Henry Ergas Concept Economics, Australia Terry Hannan Australian College of Health Informatics, Australia Jos Aarts Erasmus University, Rotterdam, The Netherlands

ABSTRACT Health care is complex and there are few sectors that can compare to it in complexity and in the need for almost instantaneous information management and access to knowledge resources during clinical decision-making. There is substantial evidence available of the actual, and potential, benefits of e-health tools that use computerized clinical decision support systems (CDSS) as a means for improving health care delivery. CDSS and associated DOI: 10.4018/978-1-60960-561-2.ch103

technologies will not only lead to an improvement in health care but will also change the nature of what we call electronic health records (EHR) The technologies that “define” the EHR will change the nature of how we deliver care in the future. Significant challenges relating to the evaluation of these health information management systems relate to demonstrating their ongoing cost-benefit, cost-effectiveness, and effects on the quality of care and patient outcomes. However, health information technology is still mainly about the effectiveness of processes and process outcomes, and the technology is still not mature, which may

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The Effectiveness of Health Informatics

lead to unintended consequences, but it remains promising and unavoidable in the long run.

INTRODUCTION The Institute of Medicine (IOM) report, To Err is Human: Building a Safer Health System provides a landmark review of the functionality of modern health care delivery in the information and technology revolutions (Kohn, Corrigan & Donaldson, 2000). It concludes that health care is error-prone and costly, as a result of factors that include persistent major errors and delays in diagnosis and diagnostic accuracy, under/ over-use of resources (e.g. excessive ordering or unnecessary laboratory tests), or inappropriate use of resources (e.g. use of outmoded tests or therapies) (Kohn, Corrigan & Donaldson, 2000). Health care is complex and there are few sectors that can compare to it in complexity as well as in the need for almost instantaneous information management and access to knowledge resources during clinical decision-making. An example of a comparable system of complex decision-making can be seen in air travel and is highlighted in the report on the factors contributing to the Tenerife air disaster on Sunday 17th, 1977. In the final summary on this disaster, Weick (1990) makes the following comments that could also be used to describe health care decision-making: The Tenerife air disaster, in which a KLM 747 and a PanAm 747 collided with a loss of 583 lives, is examined as a prototype of system vulnerability to crisis. It is concluded that the combination of interruption of important routines among interdependent systems, interdependencies that become tighter, a loss of cognitive efficiency due to autonomic arousal, and a loss of communication accuracy due to increased hierarchical distortion, created a configuration that encouraged the occurrence and rapid diffusion of multiple small errors (Weick, 1990, pp. 593).

26

This major air disaster led to significant changes in air travel and reforms in the regulatory framework that have resulted in a higher level of safety and quality in this industry. If we compare the changes that occurred in aviation to improvements in health care delivery following the To Err Is Human report (Kohn, Corrigan & Donaldson, 2000), which focused on documenting deaths due to medical errors in the U.S. health care system, then the changes have not been as significant. A number of studies have found evidence of a lack of improvement in health care delivery in the U.S. despite major public and private investments in technology. In 2005, Leape and Berwick (2005) reviewed the U.S. health delivery system five years after the IOM report was released. They found significant deficiencies and faults in nearly all aspects of health care delivery. For example, in patient diagnoses there remained significant errors in accuracy and delays. In attempts to evaluate a given diagnosis there were failures to employ appropriate tests (underuse), the continued use of outmoded tests or therapies (inappropriate use), and the failure to act on the results of tests or monitoring (ignoring medical alerts and reminders). In treatment protocols they reported significant errors in operations, surgical procedures, and tests. They also found evidence in the administration of medication, the continued administration of the wrong drugs, doses, and medications given to patients with a known allergy to the drug (also in Evans et al., 1998). Bates et al.(1994; 1997) and Rothschild et al., (2002) found the persistence of a high incidence of adverse drug events (ADE) and their associated costs during care delivery, and demonstrated a close relationship between the incidence of preventable ADE, costs and medical malpractice claims. Preventive care is also considered to be a significant area of health care where costs savings and better health outcomes can be delivered. Fries (1994) has estimated that healthcare cost savings of up to 70% can be achieved through the implementation of more effective preventive care measures. Other

The Effectiveness of Health Informatics

areas of healthcare information management that continue to impair healthcare delivery include persistent failures in communication, equipment failure, and other information systems failures. It is well known that human beings in all lines of work make errors, and available data show that the health care sector is complex and error-prone resulting in substantial harm (Leape & Berwick, 2005). We also know that current and emerging technologies have the potential to provide significant improvements in healthcare delivery systems. Similar trends have occurred in aviation, which has also provided a guide as to where the focus of change should lie (Coiera, 2003). In this chapter, the following questions are addressed: •

•

Can information technology and health information management tools improve the health care process, quality and outcomes, while containing costs? How do we assess the cost-effectiveness of health information technology?

This chapter will provide a historical perspective with examples of how information technologies, designed around computerised Clinical Decision Support Systems (CDSS), can facilitate the more accurate measurement of the healthcare delivery process, and have provided reproducible solutions for cost savings, improved patient outcomes, and better quality of care.

BACKGROUND: HISTORICAL PERSPECTIVE The importance of clinical decision making (CDM) and its effects on outcomes in care have been well documented since the 1970s (Dick, Steen & Detmer, 1997; Kohn, Corrigan & Donaldson, 2000; Osheroff et al., 2006). The care process is now understood to function across complex environments involving the patient,

primary care, prevention, in-hospital care, and sub-specialisation care. The information management interrelationships extend beyond the direct care process to research, epidemiology, planning and management, health insurance, and medical indemnity. In 1976, McDonald discussed the key limitations of CDM in complex health information-rich environments, in particular by showing the failure of CDM to meet pre-defined standards of care, and concluded that computerised (electronic) decision support (CDSS) was an essential “augmenting tool” for CDM (McDonald, 1976). McDonald’s research became a stimulus to the ongoing research in this domain of health care and revealed not only the benefits of CDSS in health care, but also how we can now begin to “measure the care processes” and evaluate what we do much more effectively. In 2008 the importance of CDSS in health care was reconfirmed in a policy document prepared by the American Medical Informatics Association (AMIA) entitled A Roadmap for National Action on Clinical Decision Support (Osheroff et al., 2006). In the Executive Summary, the functions of CDSS in a modern health care system are clearly defined: Clinical decision support provides clinicians, staff, patients or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care. It encompasses a variety of tools and interventions such as computerized alerts and reminders, clinical guidelines, order sets, patient data reports and dashboards, documentation templates, diagnostic support, and clinical workflow tools. Clinical decision support has been effective in improving outcomes at some health care institutions and practice sites by making needed medical knowledge readily available to knowledge users (Osheroff et al., 2006, p.4). and the summary concludes:

27

The Effectiveness of Health Informatics

Achieving desirable levels of patient safety, care quality, patient centeredness, and cost-effectiveness requires that the health system optimize its performance through consistent, systematic, and comprehensive application of available healthrelated knowledge – that is, through appropriate use of clinical decision support (Osheroff et al.,2006, p.4). The knowledge we have from more than 25 years of research in clinical information management systems allows us to conclude that “information is care” (Leao, 2007). In the words of Tierney et al. (2007, p. 373), “although health care is considered a service profession, most of what clinicians do is manage information.” A number of studies support the centrality of CDSS in improving CMD and the overall health care delivery system. From the 1950s to 1970s the technology supporting medical laboratory procedures was evolving. There was a dramatic escalation in the number of procedures performed with a minimal change in the number of technical personnel to support the care process. During this time the number of personnel in hospitals numbered in the 100s or 1000s, yet the number of procedures was rising to the millions each year (Speicher, 1983). Even though the average cost of many of these laboratory procedures at that time (e.g. chest X-rays, full blood analysis) was less than $20 US, the overall decision making process was already very costly. The results of the study by Speicher (1983) provide supportive evidence to the earlier study by Johns and Blum (1979) that linked CDM to resource utilisation and ongoing data generation in clinical environments and found that within a set of four nursing units where the annual expenditure was $44 million U.S., there were, on average, 2.2 million clinical decisions per year, 6,000 per day, and six per patient per day. A further example that linked CDM to compliance with care, clinical outcomes, and resource utilisation is that of immunization rates. In 1993 Gardner and Schaffner showed that vaccination

28

rates for common illnesses such as influenza, pneumococcus, hepatitis B, and tetanus-diphtheria ranged from 10 to 40%. These are diseases where the vaccines have a clinical effectiveness ranging from 60 to 99% (Gardner & Schaffner, 1993). There are several cost and quality implications of these vaccination rates. From a quality perspective Gardner and Schaffner were able to correlate low vaccination rates with preventable deaths. A similar study by Tang, LaRosa, Newcomb, and Gorden (1999) demonstrated similar effects of immunisation rates on clinical outcomes. Further historical evidence for the close relationship between CDM and outcomes of health care has been documented for a variety of parameters that measure healthcare processes. These include: the failure to comply with pre-defined standards of care, adverse drug event (ADE) detection, preventive healthcare procedures, health insurance claims management, and data acquisition for research (Bates et al., 1994; James, 1989; Tierney et al., 2007). To these factors can be added the significant cost inflation associated with the attempts to manage health care predominantly as a business or administrative organizational model. James (1989) analysed a range of common clinical scenarios for procedures such as cholecystectomy, prosthetic hip replacement, and transurethral resection of the prostate (TURP). He found a wide variation, not only in the care process amongst groups of clinicians within different health care institutions for standardised conditions, but also within each individual practitioner’s activities. He also found that low quality leads directly to higher costs; he defines these costs that arise from an initial process failure and the resulting low quality output, as “quality waste” (i.e., resources that are consumed, in the form of scrap or repairs, when a unit of output fails to meet quality expectations - in clinical care this can represent death or short and long term morbidity). James also emphasises the importance of documentation in determining the quality of care. He states that fundamental elements for quality improve-

The Effectiveness of Health Informatics

ment are to eliminate inappropriate variation and document continuous improvement (i.e., measure what we do). This is not possible using essentially paper-based record systems for decision support. It has also been shown in the Harvard Medical Practice Study looking at negligence in medical care that paper-based record systems actually hide decision-making errors that promote poor clinical outcomes (Brennan et al., 1991). ADE remains a significant reason for poor, and preventable, patient outcomes. In 1998, Cook, Woods, and Miller documented that 50% of ADEs are preventable and that they represent the highest incidence of medical deaths compared to motor vehicle accidents, breast cancer, and AIDS. These preventable events represent a cost of $17 to $29 billion U.S. per year. Bates et al. (1994) also demonstrated the costly nature of ADEs at the Brigham and Women’s Hospital. All ADEs at that hospital cost $5.6 million U.S. and of these, preventable ADEs represented $2.6 million U.S. These figures excluded costs of injuries to patients, malpractice costs, and the costs of admissions due to ADE (Bates et al., 1994). The close relationship between ADEs and malpractice claims (outcomes) was demonstrated by Rothschild et al. (2002) and Studdert et al. (2006). These studies show that many of the added costs of these events were related to litigation and administrative costs and approximately 50% of the events were preventable. Another factor contributing to the current status of health care delivery is the ability of clinicians to comply with pre-defined standards of care. For the three decades from 1979 to 1990, several studies demonstrated that the overall rate of what is done in routine medical practice that is based on published scientific research remained steady at between 10 to 20% (Ferguson, 1991; Williamson, Goldschmidt, & Jillson, 1979). In 2003 a RAND Corporation review revealed that, on average, patients received recommended care in only 54.9% of instances. While this reflects an improvement in compliance with care standards,

it also indicates that around 45% of patients do not receive standardised care (Farley, 2005). A final example that discusses the core principles of CDSS implementation is the Academic Model for Prevention and Treatment of HIV (AMPATH) in resource-poor Kenya (Tierney et al., 2007). This project saw the successful implementation of health information technologies based on electronic medical record (EMR) functionalities in a resource poor nation. The successful partnership now sees this EMR as the largest e-health system for developing nations and is implemented in more than 23 countries. To obtain an extensive review of successes, difficulties, and an understanding of the complexity of information management in health, as well as how solutions can be found, we refer to the 25 year review of electronic medical record systems that focuses on CDSS in North America and Europe in the full issue of the International Journal of Medical Informatics in 1999 (see Editorial by Safran, 1999, pp. 155-156).

CLINICAL INFORMATION SYSTEMS AND CLINICAL DECISION SUPPORT SYSTEMS (CDSS) IN THE 21ST CENTURY: THE IMPACT ON HEALTH CARE QUALITY, COSTS, AND OUTCOMES The statement “information (management) is care” (Tierney et al., 2007, p. 374) emphasises one of the core principals of health care, that is, everything a provider does with a patient involves the flow of information (e.g. clinical history, physical examination, orders for tests, instructions for care, follow-up). This process involves the collection, management, and reporting of data in readable formats to the provider, thereby facilitating the care process (Tierney et al., 2007). Coiera (2003) also provides a clear description of where “clinical informaticians” fit into the care process:

29

The Effectiveness of Health Informatics

Informaticians should understand that our first contribution is to see healthcare as a complex system, full of information flows and feedback loops, and we also should understand that our role is to help others ‘see’ the system, and re-conceive it in new ways (Coiera, 2003, p. xxii). Any clinical decision support system(s) (CDSS) provides clinicians, staff, patients, or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care (Coiera, 2003; Osheroff et al., 2006). CDSS encompass a variety of tools and interventions, such as computerized alerts and reminders, clinical guidelines, order sets, patient data reports and dashboards, documentation templates, diagnostic support, and clinical workflow tools. The following four key functions have been defined for CDSS (Perreault & Metzger, 1999): (1) Administrative: These systems provide support for clinical coding and documentation, authorization of procedures, and referrals. (2) Managing clinical complexity and details: Keeping patients on research and chemotherapy protocols; tracking orders, referrals follow-up and preventive care. That is, complying with pre-defined standards of care. (3) Cost control: This involves activities such as the monitoring of medication orders and avoiding duplicate or unnecessary tests. (4) Decision support: These are complex information management systems that support clinical diagnosis and treatment plan processes; and that promote use of best practices, condition-specific guidelines, and population-based disease management. Based on the existing evidence it is accepted that health information management systems centred on CDSS provide the most significant opportunity to improve health care delivery and management

30

(Brennan et al., 2007). One could wonder why their use is not universal (Ford, Menachemi, Peterson & Huerta, 2009). A major barrier to CDSS implementation is clinician involvement in the development of the information management tools. We know that age, gender, and computer literacy are not significant factors in the use of computers in health care (Sands, 1999; Slack, 2001). Uptake of health information technologies is related to the efficiency and useability of the information management tools that CDSS deliver through the total e-health system. The current focus to provide solutions to the problem of clinician involvement relates to Computerized Provider Order Entry (CPOE) (Brennan et al., 2007). It is believed that CPOE will facilitate safe, effective care for patients by insuring that clinical care directions are communicated in a timely, accurate, and complete manner. The integration of clinical decision support functions with CPOE systems provides functionality that incorporates contemporary knowledge and best practice recommendations into the clinical management process. Additionally, by ensuring the quality, accuracy, and relevance of the decision logic integrated within CPOE systems, a guaranteed method for creating safe and effective practice is ensured. Brennan emphasises that CPOE is not a technology, rather it is a design (or redesign) of clinical processes that integrates technology to optimize provider ordering of medications, laboratory tests, procedures, etc. CPOE is distinguished by the requirement that the provider is the primary user. It is not the “electronification” of the paper record system in existing formats. In summary the evidence to date indicates that the beneficial effects of health information technology on quality, efficiency, and costs of care can be found in three major areas (Chaudhry et al., 2006): • • •

increased adherence to guideline-based care, enhanced surveillance and monitoring, and decreased medication errors.

The Effectiveness of Health Informatics

A recent overview of health information technology studies by Orszag (2008) from the US Congressional Budget Office (CBO) suggests, albeit with some qualifications, that a more comprehensive list of potential benefits from the use of HIT would include the following: •

•

•

•

•

Eliminating paper medical records – however, the CBO notes that these savings might not apply in very small practices that have low but relatively fixed costs related to medical records; Avoiding duplicated or inappropriate diagnostic tests – according to the CBO, some studies (e.g. Bates et al., 1998; Bates et al., 1994) suggest that electronic health records with a notice of redundancy could reduce the number of laboratory tests by about 6%; Reducing the use of radiological services though the CBO notes that the evidence on this is weak. While studies (e.g. Harpole, Khorasani, Fiskio, Kuperman & Bates, 1997) show that HIT may ease the job of monitoring the use of radiological services, there is little evidence that it helps control costs; Promoting the cost-effective use of prescription drugs, particularly through decision support software and computerized provider order entry which prompts providers to use generic alternatives, lowercost therapies, and cost-effective drug management programs (Mullett, Evans, Christenson & Dean, 2001); Improving the productivity of nurses and physicians – This has to be qualified, as one study found that when HIT was in use, nurses in hospitals saw reductions in the time required to document the delivery of care, but physicians saw increases in documentation time (Poissant, Pereira, Tamblyn, & Kawasumi, 2005). However, the CBO notes that the latter effect may re-

•

•

flect a short-run learning phase for doctors. Few studies have measured effects on physicians’ efficiency in outpatient settings, and those that have show mixed results (Pizziferri et al., 2005). Reducing the length of hospital stays – HIT may reduce the average length of a hospital stay by speeding up certain hospital functions and avoiding costly errors (Mekhjian et al., 2002); General improvements in the quality of care through avoiding adverse drug events1; expanding exchanges of health care information thus reducing duplication of diagnostic procedures, preventing medical errors and reducing administrative costs; expanding the practice of evidencebased medicine2; and generating data for research on comparative effectiveness and cost-effectiveness of treatments. This benefit is also consistent with the results of an older literature review which found that HIT increased adherence to guideline- or protocol-based care (Chaudry et al., 2006). This increased quality of care would also be manifest in an associated decrease in malpractice claims, another prediction which is confirmed by a recent study (Virapongse et al., 2008).

Given the need for establishing the existing and ongoing benefits from investments in clinical information management technologies, it has been demonstrated that there are long-term financial benefits, in the form of an acceptable return on investment (ROI), in computerised provider order entry systems (Kaushal et al., 2006). Despite these established benefits there remain many barriers to the widespread successful implementation of CDSS. Most of these are not technical. They relate to the design of information management tools and their acceptance by clinicians who have a long history of autonomy in health care (Beresford, 2008).

31

The Effectiveness of Health Informatics

By contrast, the costs associated with implementation of HIT are: •

• • •

the initial fixed cost of the hardware, software, and technical assistance necessary to install the system; licensing fees; the expense of maintaining the system; and the “opportunity cost” of the time that health care providers devote to learning how to use the new system and how to adjust their work practices accordingly Orszag (2008).

On the costs of installation or implementation, the CBO notes that these may vary widely among physicians and among hospitals, depending on the size and complexity of those providers’ operations and the extent to which a system’s users wish to perform their work electronically. For instance, smaller practices will pay more per physician than larger practices to implement an HIT (Orszag, 2008). The estimation of these costs will also be complicated by the differences in the types and available features of the systems being sold and differences in the characteristics of the practices that are adopting them. The CBO notes that existing studies of costs have tended to make the mistake of not including estimates of indirect costs, such as the opportunity costs of time which providers dedicate to learn the new system and to adopt it in their work routines (Orszag, 2008). The initial opportunity costs in terms of learning time and adapting the operations of the practice around the implemented system can turn out to be quite significant, with one survey of health IT adoption finding that reported productivity in a practice may drop between 10 to 15 per cent for several months after implementation (Gans, Kralewski, Hammons & Dowd, 2005). One study of a sample of 14 small physicians’ offices implementing an HIT estimated the average drop in revenue from loss of productivity at about $7,500 per physi-

32

cian over a year (Miller, West, Brown, Sim & Ganchoff, 2005).

PUTTING HEALTH INFORMATION TECHNOLOGY (HIT) TO USE Health information technology (HIT) systems are basically a repository for patient data. The physician is able to retrieve information, often in a clinically meaningful way that may not necessarily have been entered by himself/herself in the electronic health record (EHR). The information might have been acquired and created, for instance, during the patient’s course in the healthcare organization. Increasingly, EHR systems are connected to regional health information networks enabling access to patient data in disparate systems, such as primary care.

Overcoming the Limitations of Paper-based Records The electronic patient record has been introduced to overcome perceived limitations encountered with the use of the paper-based medical record and to allow planning that goes beyond a static view of events. Some of the limitations of the paper-based patient record that can be overcome include: •

Accessibility. Often the record is not accessible when it is needed. It may be stored elsewhere, or another professional may be seeking to use it concurrently. Electronic records are accessible independent of place and time, and can be rapidly retrieved using a patient identifier. It is exactly this that is most valued by clinicians. However, access is usually constrained because of data protection and privacy. Authorizations and passwords are required to allow a clinician to review patient information. Making information available both from within the

The Effectiveness of Health Informatics

•

•

•

hospital and from ambulatory systems is a key goal of most national efforts to implement HITs. Completeness. Not all patient data are written in the record. This can pose problems when other professionals reading the record try to make sense of a patient problem or when a doctor tries to recall what she has done after seeing the information again. Forcing the user to enter data in all fields can improve the completeness of a patient record. Readability. Handwriting is often hard to read. On a medication list, numbers and units of dosages can be misinterpreted and require attention of a pharmacist checking the prescription or a nurse translating and transcribing the order or trying to prepare the medication to be administered. Entering the data digitally, and even structuring the fields, can enhance readability. Analysis. Information written in the record is generally not suited for quantitative analysis. Test results may be entered in pre-structured forms and even plotted on a graph, which may reveal a trend, but comparison with baseline data is painstakingly difficult. Auditing past records to identify and analyze patterns in a group of patients is very labor intensive and time consuming. Digitized data are exceptionally suited for computer analysis.

Some Reasons for Limited Diffusion of Health Information Technology & Electronic Health Records If the advantages are so obvious, why then is the electronic record not widely in use and why haven’t electronic records replaced paper records? It is often argued that physicians resist innovation and do not like to give up familiar tools. The wide adoption of advanced technology in health care, and certainly in emergency medicine, de-

fies this argument. Paper-based records have proved to be durable tools for medical practice and information technology specialists have only recently become aware of this (Clarke, Hartswood, Procter, Rouncefield & Slack, 2003). As a cognitive artefact, the physician can examine the paper record easily. The layout and structure can guide the physician to find the most relevant information and ignore other items. Often the use of tabs, coloured paper, tables, and flowcharts facilitates navigation through a paper-based record. Within a short period of time a complete mental picture of the patient can be created. In contrast, using a computer, a user can be forced to page through a large number of screens in order to find the needed piece of information. The paper-based record also allows the physician to judge the quality of the information. Handwriting or a signature can show who entered the information and inform the physician about the trustworthiness of the information. The absence of information does not necessarily imply that the record is incomplete; it often means that a particular item was considered not relevant for the case at hand (Berg, 1998). For example, if the patient has no known history of heart problems and is in good physical condition, blood pressure recordings or an electrocardiogram (ECG) may be missing from the record. This is not to say that there are no compelling arguments to adopt electronic records; there is ample evidence that the efficiency and quality of care benefits from their use (Dick, Steen & Detmer, 1997). However, one must look carefully at the role of paper-based records in medical practice and avoid simply translating it into an electronic system.

Particular Features of Electronic Health Records That Offer Advantages over Paper Records The two powerful and distinctive advantages that electronic records have over a paper record are

33

The Effectiveness of Health Informatics

summarized in the concepts of accumulating and coordinating (Berg, 1999). Accumulating refers to the fact that an EHR system allows for powerful analysis of data. For example, lab test outcome data can be accumulated to show time-dependent trends. When grouped for a large number of patients, the same data can be subjected to statistical analysis to reveal patterns. Combined with data from other sources, information can be used for planning, billing, and quality assessment. A most powerful application is the combination of patient data with decision support techniques that enable the physician to make decisions about patient care founded on accepted clinical rules stored in a database and patient data that is as complete as possible. The other concept of coordination provides the opportunity to plan and coordinate activities of healthcare professionals by allowing concurrent access to the electronic record. Access is no longer dependent on the physical location of the record, but possible from every computer that is connected to the system.

Standardization and Integration of Technologies Many conditions have to be met in order to successfully implement HIT and the EHR, most importantly standardization of the underlying technologies and of the content and meaning of healthcare data. This may seem obvious, but it has been shown that the lack of standardization is a major impediment to the introduction of electronic records in medical practice. Technological standards are required to link systems in networks. The wide diffusion of the Internet would have been impossible without underlying standards for communicating (e.g. messages, texts, and graphics) and establishing links and pointers to other sources of information. Yet, this achievement was not possible without negotiating and consultation about what and how to standardize. Often proprietary rights and the perceived need to protect markets stand in the way.

34

In health care, standardization is much harder since it is not only about the underpinning rationale and scientific evidence for data standards, but also about the diverse social and cultural values within and across geographic communities that influence choice of data standards. We will not elaborate on the standards relevant for health care; they are explained elsewhere. Typical examples include TCP/IP that determines how information is communicated on the Internet, HL7 that determines how health care information can be represented and communicated between diverse applications in health care, and SNOMED CT that describes how medical concepts are defined and represented. Standardization can be effected through national and international bodies with legislative power, such as ASTM (originally American Society for Testing and Materials), ISO (International Standards Organization) and CEN (Comité Européen de Normalisation). Also standardization can be achieved through market power, e.g., Microsoft’s Windows operating system accounts for about 90% of the personal computer market and can therefore be construed as a de facto standard for personal computer operating systems.

Implementing Health Information Technology The traditional approach to implementing health information technology has been top-down. An important factor in this respect was the perception that health information technology is an expensive resource that usually exceeded the financial capabilities of individual physicians and physician groups. However, the advent of personal computers to some degree altered this situation, particularly in primary care. In countries such as the Netherlands, the United Kingdom and Australia, up to 90% of GPs have adopted electronic health record systems (Jha, Doolan, Grandt, Scott & Bates, 2008). However, adoption, in many cases, was helped by financial incentives given by their governments.

The Effectiveness of Health Informatics

A key characteristic of implementing information systems is that organizational changes are an integral part of implementation. Unfortunately, however, the changes are not always for the better, and more often than not, the performance of organizations is worse after a system has been installed than before. The natural tendency is then to conclude that the system was somehow badly designed. In 1975, when the design and operation of information systems were considered primarily technical activities, Henry C. Lucas, Jr. wrote about failing systems. However, all our experience suggests that the primary cause for system failure has been organizational behavior problems (Lucas, 1975). Thirty years of research has increased our understanding of information systems in organizational contexts; yet, the record in terms of developing and implementing successful systems is still dismal (Ewusi-Mensah, 2003). HIT systems are particularly hard to implement because not only do they affect health care organizations as a whole, but also the work of health professionals who pride themselves on their professional autonomy. Implementing HIT is a social process (Aarts, Doorewaard & Berg, 2004). But the relevant organizational changes are not easily predictable. This creates a predicament for an implementer. S/he might like to design a system according to blueprints and to plan systematically its deployment. But Ciborra (2002) advises making organizational improvisation part of the implementation process, to allow prospective users to tinker with the system and let them find ways of working that fit them best, to plan for the unexpected and value emerging practices, and to give up strict control (Ciborra, 2002).

Adverse Effects of Health Information Technology Recent studies reveal that putting HIT to use, whatever its many advantages may be associated with unexpected outcomes. A study by Koppel et

al. (2005) of one CPOE system documented 22 different error-enhancing aspects of that system. Another study reported a doubling of infant mortality after the introduction of a CPOE system, probably resulting from increased time to enter orders, reduced communication among nurses and doctors, and the loss of advance information previously radioed in from the transfer team before patients arrived at the hospital (Han et al., 2005). Nebeker, Hoffman, Weir, Bennett, and Hurdle (2005), likewise, found high rates of ADEs in the highly computerized Veterans Administration system. Shulman, Singer, Goldstone, and Bellingan (2005) found that, compared to paperbased systems, CPOE was associated with fewer inconsequential errors, but also with more serious errors. Ash, Berg, and Coiera (2004); Campbell, Sittig, Ash, Guappone, and Dykstra (2006); and Aarts, Ash, and Berg (2007) have found unintended consequences from CPOE systems to be the rule, rather than the exception. Nemeth and Cook (2005, p. 262), noting these systems’ interactivity and complexity, add: “If [human error] exists, error is a consequence of interaction with IT systems. The core issue is to understand healthcare work and workers”. And although “healthcare work seems to flow smoothly,” the reality is “messy.”

METHODS OF ASSESSING COST-EFFECTIVENESS There have not been many rigorous studies of cost effectiveness of e-health measures in the literature. The most recent literature review of studies in this field concluded that there remained a “paucity of meaningful data on the cost-benefit calculation of actual IT implementation” (Goldzweig, Towfigh, Maglione & Shekelle, 2009, p. 292). The studies which this literature review collected, after a comprehensive selection process, can be broken into four different approaches. One approach looks at the experience of a few large organizations that had implemented

35

The Effectiveness of Health Informatics

multifunctional, interoperable electronic health records (EHRs), computerized physician order entry (CPOE), decision-support systems, and other functions. However, these studies cannot be described as appropriate cost effectiveness or cost benefit analyses since they only evaluated the impacts in terms of clinical performance/ quality improvement or potential benefit in terms of patient safety measures, but did not attempt to quantify the costs of these technologies or to then derive a net benefit calculation using some common measure (for instance, some imputed welfare gain in terms of dollars). Among the individual studies in this category: •

•

•

•

36

Three studies looked at a quality improvement project related to blood product administration that used automated alert technology associated with CPOE; electronic clinical reminders related to coronary artery disease and diabetes mellitus; and patient-specific e-mail to providers regarding cholesterol levels, and found small improvements in quality of care. (Lester, Grant, Barnett & Chueh, 2006; Rothschild et al., 2007; Sequist et al., 2005) Roumie et al. (2006) evaluated the impact of electronic provider alerts and found they provided a modest, non–statistically significant improvement over provider education alone as measured in terms of improvements in blood pressure control. Dexter, Perkins, Maharry, Jones, and McDonald (2004) compared the impact on rates of influenza and pneumococcal vaccinations of computer generated standing orders for nurses versus computerized physician reminders and found that immunization rates were significantly higher with the nurse standing order. Murray et al. (2004) and Tierney et al. (2005) evaluated computer-generated treatment suggestions for hypertension, and for asthma and chronic obstructive

•

pulmonary disease (COPD), and it was found that neither of these technologies resulted in improvements in care. Potts, Barr, Gregory, Wright, and Patel (2004), Butler et al., (2006) and Ozdas et al., (2006) evaluated the potential benefits of CPOE. These studies arrived at mixed results but generally found improvements in patient safety with the introduction of CPOE, as well as modest improvements in quality of care when CPOE was tailored to the management of patients with acute myocardial infarction.

Insofar as there was a basis for comparison implicit in these studies, this involved comparing the refinement in existing systems, addition of new applications or enhancement of existing functionalities, against the resulting improvements. The general finding was one of modest or even no benefits from the new applications or changed functionalities (Goldzweig et al., 2009, p. 285). A second approach found in the literature review was to look specifically at the experiences of commercial practices implementing commercially available or developed electronic health records. Among the individual studies in this category: •

•

Garrido, Jamieson, Zhou, Wiesenthal, and Liang (2005) compared outcomes before and after the implementation of a homegrown EHR at commercial hospitals and found that the number of ambulatory visits and radiology studies decreased after implementation, while telephone contacts nearly doubled. On the other hand, limited measures of quality (immunizations and cancer screening) did not change. O’Neill and Klepack, (2007) assessed the effect of implementing a commercial EHR in a rural family practice, looking specifically at financial impacts rather than quality of care measures. They found that average monthly revenue increased 11 per

The Effectiveness of Health Informatics

•

cent in the first year and 20 per cent in the second year, and the charge-capture ratio increased 65 to 70 per cent, because of better billing practices. Asaro, Sheldahl, and Char (2006); Del Beccaro, Jeffries, Eisenberg, and Harry (2006); Feldstein et al. (2006); Galanter, Polikaitis, and DiDomenico (2004); Han et al. (2005); Palen, Raebel, Lyons, and Magid (2006); Smith et al. (2006); Steele et al. (2005); and Toth-Pal, Nilsson, and Furhoff (2004) studied the effect of adding new functionalities to existing EHRs. Some of these studies found modest benefits, some found no benefits, and a few found marked benefits.

A third group of approaches involved evaluations of stand- alone applications such as health systems that link patients with their care providers and are intended to improve the management of chronic diseases; computer/video decision aids for use by patients and providers; text messaging systems for appointment reminders; electronic devices for use by patients to improve care; and patient-directed applications for use outside traditional settings. The main findings here were also mixed, with some studies showing no or only modest effects, and many more studies providing insufficient descriptions to reach strong conclusions. Among the individual studies in this category: •

•

McMahon et al. (2005) compared Webbased care management with usual care for patients with diabetes. Intervention patients had a statistically significant, modest improvement in their results for 2 out of 3 clinical measures. Cavanagh et al. (2006), Grime (2004), and Proudfoot et al. (2004) evaluated an interactive, multimedia, computerized cognitive behavioral therapy package. In two of the studies, statistically significant im-

•

provements of modest size were found for patients in the intervention groups compared to usual care, although the differences were no longer significant at three or six months. Cintron, Phillips, and Hamel (2006); Jacobi et al. (2007); and Wagner, Knaevelsrud, and Maercker (2006) looked at Internet applications that could be accessed directly by the patient with three involving randomized trials. Clinical improvements were found.

Another set of studies covered by the literature review that are not directly relevant to the issue of cost effectiveness or even general effectiveness of HIT, but nonetheless have implications for the likely costs of implementing HIT, looked at barriers to HIT adoption. One of these studies, which involved a survey of US paediatric practices found that the main barriers to HIT adoption were resistance from physicians (77 per cent of practices without an EHR reported this barrier), system downtime (72 per cent), increase in physicians’ time (64 per cent), providers having inadequate computer skills (60 per cent), cost (94 per cent), and an inability to find an EHR that met the practice’s requirements (81 per cent) (Kemper, Uren & Clark, 2006). A survey of the Connecticut State Medical Society Independent Practice Association found that the most commonly stated barrier was cost (72 per cent) and other barriers were time necessary to train staff (40 per cent), lack of proficiency among staff (26 per cent), and lack of an IT culture within the office (18 per cent) (Mattocks et al., 2007). One methodological problem with demonstrating that there are particular positive associations between clinical outcomes and use of HIT is that these associations are not necessarily causal hospitals that have more HIT tend to have greater resources and better performance. However, one study that at least attempted to control for these confounders still found a statistically significant

37

The Effectiveness of Health Informatics

relationship. Amarasingham, Platinga, DienerWest, Gaskin, and Powe (2009) looked at the relationship between HIT and both costs and clinical outcomes in hospitals in Texas. A particular focus in this study was whether increased automation of hospital information was associated with decreased mortality, complication rates and costs, and length of stay. They found strong relationships between the presence of several technologies and complication and mortality rates and lower costs. For instance, use of order entry was associated with decreases in mortality rate for patients with myocardial infarction and coronary artery bypass surgery. Use of decision support software was associated with a decrease in the risk of complications. Automated notes were associated with a decrease in the risk of fatal hospitalizations. The researchers controlled for the fact that hospitals that have more HIT tend to have more resources and still found that the relationships persisted, though there were also some instances in which relationships in the opposite direction were found. For example, electronic documentation was associated with a 35% increase in the risk of complications in patients with heart failure, though this may have been because it was easier to find these events due to better documentation (Bates, 2009). In short, there is a dearth of appropriate cost effectiveness studies and of useful data for conducting such studies. Although the review by Goldzweig et al. (2009, pp. 290-291) concluded on the basis of the individual studies surveyed that “there is some empirical evidence to support the positive economic value of an EHR;” they also found that the projections of large cost savings in previous literature assumed levels of health IT adoption and interoperability that had not been achieved anywhere. A less comprehensive survey by the US Congressional Budget Office (Orszag, 2008) focused on two prominent studies that attempted to quantify the benefits of HITs. One was a study by the RAND Institute (Girosi, Meili, & Scoville, 2005)

38

and the other a study by the Center for Information Technology Leadership (CITL) (Pan 2004). Both these studies had estimated annual net savings to the US health care sector of about $80 billion (in 2005 dollars) relative to total spending for health care of about $2 trillion per year, though they identified different sources of those savings. The RAND research had quantified savings that the use of health IT could generate by reducing costs in physicians’ practices and hospitals. In contrast, the CITL study narrowed the focus to savings from achieving full interoperability of health IT, while excluding potential improvements in efficiency within practices and hospitals. The approach adopted by both these studies did have in common the use of various extrapolations and these were a source of criticism by the CBO that found that they were inappropriate by the standards of a rigorous cost effectiveness analysis. In particular, according to the Orszag (2008), the RAND study had the following flaws: •

•

•

•

It assumed “appropriate changes in health care” from HIT rather than likely changes taking into account present-day payment incentives that would constrain the effective utilization of HIT. It drew solely on empirical studies from the literature that found positive effects for the implementation of health IT systems, thus creating a possible bias in the results. It ignored ways in which some cost reductions would be mitigated by cost shifting in other areas. Some of its assumptions about savings from eliminating or reducing the use of paper medical records were unrealistic for small practices.

Similarly, the CITL study came in for criticism by the CBO for, in particular, not fully considering the impact of financial incentives in the analysis, estimating savings against a baseline of little or

The Effectiveness of Health Informatics

no information technology use, and using overoptimistic assumptions. While the CBO identified the numerous ways in which two prominent studies of HIT may have overestimated the benefits of HIT, its list of benefits does suggest that there are some areas in which the long-term benefits of HIT may be underestimated insofar as the scale of use has not reached a critical mass. In particular, we would conjecture that the possible improvements in quality of care through the expansion of health care information and generation of data for research may presuppose a base of participating health care providers and institutions that have implemented HIT and are able to share data over their networks. It is possible that these benefits may not be significant until use of HIT is diffused over a greater percentage of health care providers and institutions. In other words, some of the benefits to be derived from health IT increase in value as the network of those using the technology expands, i.e. as other providers also purchase health IT systems. This phenomenon is known to economists as network effects and is not necessarily restricted to benefits arising from exchange of information for research purposes - providers who can perform functions electronically, such as sending and receiving medical records or ordering laboratory and imaging procedures, also gain when other providers develop similar electronic capabilities. For example, the cost to a general practitioner of sending medical data to a consulting specialist is potentially lower with an HIT system, but only so long as the consulting specialist has an interoperable system that can receive the data electronically. As a general matter, economists distinguish between direct and indirect network effects, where the former refer to “technological” externalities while the latter refer to “pecuniary” externalities. The former involve situations where use of a technology by agent A directly affects the value agent B derives from that technology (for instance, through increased inter-operability). In

the latter, the effects are mediated through the price system, so adoption by agent A reduces the cost of the technology (for instance, through the achievement of greater economies of scale) and hence yields a benefit to agent B. Generally, it is assumed that the price system will take account of pecuniary externalities (although this is not always correct), but the technological externalities can drive a wedge between private and social costs at the margin. When that occurs, it is crucial that cost-benefit studies appropriately distinguish between private and social costs and benefits; this is not generally the case with the studies of HIT deployment that we have reviewed. At the same time, when network effects are significant, there will typically be multiple equilibria; for example, private costs and benefits may be equalised at one, low level of adoption (with low net benefits), and at another, high level of adoption (with potentially higher net benefits). At the low level equilibrium, no individual non-adopter will face a private net gain from adopting. For example, in a telephone system with few customers, the marginal subscriber gains little by joining the network. A cost-benefit evaluation conducted at that low level of adoption will therefore find that the private benefits of adoption are less than the private costs. However, were adoption levels increased, joining the telephone system would allow the marginal user to communicate with a greater base of subscribers, so that the benefits of subscription (especially taking account of the gains made by those receiving the calls) exceed the private costs. These issues of differences between private and social costs and benefits and of multiple equilibria can create biases. For example, what may seem like low benefits relative to costs may reflect an initial low level equilibrium and an associated coordination failure. In other words, if instead of looking at a decision by say, an individual practitioner or medical practice to adopt HIT, one evaluated the costs and benefits of adding a large number of practitioners or medical practices to the

39

The Effectiveness of Health Informatics

installed base of HIT, the balance of the costs and benefits might differ. It is pertinent that according to a new analysis of HIT deployment in seven industrialized countries, US deployment lags well behind other countries (Davis, Doty, Shea, & Stramekis, 2009). Electronic medical records usage ranged from nearly all physicians in the Netherlands to 23 per cent in Canada and 28 per cent in the US. Incidentally, the same study also found that physicians with greater IT capacity were more likely to report feeling well-prepared to manage patients with chronic diseases. Insofar as the bulk of HIT studies have been from the US, the benefits documented from use of HIT in these studies may not be representative of benefits in countries with a higher deployment of HIT. However, this is not a hypothesis we are in a position to test. One barrier to the achievement of the full magnitude of network effects that would maximize the benefits of using HIT may be legal restrictions. A recent US study found that privacy regulations impose costs that deter the diffusion of EMR technology (Miller & Tucker, 2009). These regulations may inhibit adoption by restricting the ability of hospitals to exchange patient information with each other, which may be particularly important for patients with chronic conditions who wish to see a new specialist, or emergency room patients whose records are stored elsewhere. The study calculated that the inhibition of network benefits from privacy regulations reduced hospital adoption of EMR by 25 per cent. It is clear from the various literature reviews discussed so far, first, that there have been very few studies that have attempted to meet the rigorous standards of a cost effectiveness analysis, and second, that there are numerous pitfalls in conducting such analyses owing to the use of various assumptions and extrapolations in quantifying benefits or costs. Another issue which remains to be addressed with greater rigour in the literature is how to quantify the likely benefits of HIT taking

40

into account various projections of network effects associated with different uptake rates.

CONCLUSION The picture that emerges from our overview has several dimensions. First of all, it is evident that little is known about the overall impact of health IT on the outcomes of health care. In this chapter we have reported a number of studies that showed positive outcomes on, for example, the reduction of adverse drug events, better resource utilization, and improved adherence to clinical guidelines. These studies are well bounded in scope and size, and are mainly about the effectiveness of processes. However, the findings do not always unequivocally point to positive outcomes. Reminders and alerts are an essential feature of decision support in computerized physician order entry systems. They warn users of potentially dangerously interacting medications. A systematic review showed that they are suppressed frequently, and this may prevent detection of ADEs and thus compromise patient safety (van der Sijs, Aarts, Vulto, & Berg, 2006). Yet, the positive outcomes are often extrapolated to a larger population to make a case for the wide scale implementation of health IT. Equally, implementing health IT often entails significant organizational change, both at the level of practicing medicine and the structure of health care organizations. However, a study of organizational change in Australian hospitals by Braithwaite, Westbrook, Hindle, Iedema, and Black (2006) provides a sobering reminder that there can be a large gap between expectations and reality. The introduction of clinical directorates, in which clinical departments, wards, and units that best fitted conceptually were joined together, was seen to increase the effectiveness of health delivery. Using diachronic data the authors found that the introduction of clinical directorates had no effect. Unfortunately, we have no way of fully knowing the effectiveness of health IT unless it has

The Effectiveness of Health Informatics

been widely adopted and diachronic data become available for analysis. There is growing awareness that health IT is far from mature. In a report to the Office of the National Coordinator of Health Information Technology (ONCHIT), the American Medical Informatics Association writes that some current clinical decision support systems often disrupt clinical workflow in a manner that interferes with efficient delivery (Osheroff et al., 2006). In our overview we already mentioned the adverse effects of health IT, indicating how difficult implementation in practice is. In another study Koppel, Wetterneck, Telles, and Karsh (2008) also found that bar-coded medication administration systems did not reduce dispensing errors substantially because they induced workaround to mitigate unintended effects. The main causes seem to be the technology itself and the generally poor understanding of how technology affects work practices, let alone how it can improve them by introducing notions of patient-centered care and a working collaborative of different providers. A telling example is how computerized provider order entry systems are designed and implemented on the model of an individual physician prescribing medication, instead of a collaborative model involving physicians, pharmacists, and nurses who are all involved in providing medication to patients (Niazkhani, Pirnejad, Berg, & Aarts, 2009). Often expectations are overblown. In a Dutch hospital the implementers of a CPOE system expected that physicians would use the system, because the system that was being replaced was also about electronic order entry (Aarts et al., 2004). They did not realize that physicians were not at all accustomed to electronic order entry, and that the system requires a doctor to send electronic notes, but doctors don’t send notes as other people do that for them. Implementation should begin by asking the question ‘what organizational problem is going to be solved,’ and what can be done to engage problem owners. There is also a serious lack of organizational learning when a

system is designed and implemented (Edmondson, Winslow, Bohmer, & Pisano, 2003). This may be due to the fact that implementation teams are often dissolved after the project is considered finished, leading to a change in personnel who actually use the system. To conclude, we find ourselves in a double bind. The effectiveness of health IT is anecdotal, and increasingly unintended consequences are being reported (Ash et al., 2004). Health IT is far from mature. Hard work is needed to get safe and reliable systems to work in practice. Yet, it is a dictum that without IT, health care will grind to a halt. We have become dependent on health IT, and yet its overall contribution to health care is hard to quantify. It is comparable to the productivity paradox that some economists pointed to in the early 1990s. Society has become dependent and intertwined with information technology, and yet its contribution didn’t seem to show up in the productivity figures (Brynjolfsson, 1993; Landauer, 1995). While some of the positive macroeconomic impacts did become clearer over time, the situation with respect to health IT is still at the earlier stage. This leaves a need to identify proxies for health IT effectiveness and better capture longitudinal and diachronic data to assess its impact on process and outcome.

REFERENCES Aarts, J., Ash, J., & Berg, M. (2007). Extending the understanding of computerized physician order entry: Implications for professional collaboration, workflow and quality of care. International Journal of Medical Informatics, 76(Suppl 1), 4–13. doi:10.1016/j.ijmedinf.2006.05.009 Aarts, J., Doorewaard, H., & Berg, M. (2004). Understanding implementation: the case of a computerized physician order entry system in a large Dutch university medical center. Journal of the American Medical Informatics Association, 11(3), 207–216. doi:10.1197/jamia.M1372

41

The Effectiveness of Health Informatics

Amarasingham, R., Plantinga, L., Diener-West, M., Gaskin, D. J., & Powe, N. R. (2009). Clinical information technologies and inpatient outcomes: a multiple hospital study. Archives of Internal Medicine, 169(2), 108–114. doi:10.1001/archinternmed.2008.520 Asaro, P. V., Sheldahl, A. L., & Char, D. M. (2006). Embedded guideline information without patient specificity in a commercial emergency department computerized order-entry system. Academic Emergency Medicine, 13(4), 452–458. doi:10.1111/j.1553-2712.2006.tb00325.x Ash, J. S., Berg, M., & Coiera, E. (2004). Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. Journal of the American Medical Informatics Association, 11(2), 104–112. doi:10.1197/jamia.M1471 Bates, D. W. (2009). The effects of health information technology on inpatient care. Archives of Internal Medicine, 169(2), 105–107. doi:10.1001/ archinternmed.2008.542 Bates, D. W., Boyle, D. L., Rittenberg, E., & Kuperman, G. J., Ma’Luf, N., Menkin, V. et al. (1998). What proportion of common diagnostic tests appear redundant? The American Journal of Medicine, 104(4), 361–368. doi:10.1016/S00029343(98)00063-1 Bates, D. W., O’Neil, A. C., Boyle, D., Teich, J., Chertow, G. M., Komaroff, A. L., & Brennan, T. A. (1994). Potential identifiability and preventability of adverse events using information systems. Journal of the American Medical Informatics Association, 1(5), 404–411. Bates, D. W., Spell, N., Cullen, D. J., Burdick, E., Laird, N., & Petersen, L. A. (1997). The costs of adverse drug events in hospitalized patients. Adverse Drug Events Prevention Study Group. Journal of the American Medical Association, 277(4), 307–311. doi:10.1001/jama.277.4.307

42

Beresford, E. B. (1991, Jul. - Aug.). Uncertainty and the shaping of medical decisions. The Hastings Center Report, 21(4), 6–11. doi:10.2307/3562993 Berg, M. (1998). Medical work and the computerbased patient record: a sociological perspective. Methods of Information in Medicine, 37(3), 294–301. Berg, M. (1999). Accumulation and coordinating: occasions for information technologies in medical work. Computer Supported Cooperative Work, 8, 373–401. doi:10.1023/A:1008757115404 Braithwaite, J., Westbrook, M. T., Hindle, D., Iedema, R. A., & Black, D. A. (2006). Does restructuring hospitals result in greater efficiency?-An empirical test using diachronic data. Health Services Management Research, 19(1), 1–12. doi:10.1258/095148406775322016 Brennan, P. F. (2007). CPOE: sufficient, but not perfect, evidence for taking action. Journal of the American Medical Informatics Association, 14(1), 130–131. doi:10.1197/jamia.M2303 Brennan, T. A., Leape, L. L., Laird, N. M., Hebert, L., Localio, A. R., & Lawthers, A. G. (1991). Incidence of adverse events and negligence in hospitalized patients. Results of the Harvard Medical Practice Study I. The New England Journal of Medicine, 324, 370–376. Brynjolfsson, E. (1993). The productivity paradox of information technology. Communications of the ACM, 36(12), 67–77. doi:10.1145/163298.163309 Butler, J., Speroff, T., Arbogast, P. G., Newton, M., Waitman, L. R., & Stiles, R. (2006). Improved compliance with quality measures at hospital discharge with a computerized physician order entry system. American Heart Journal, 151(3), 643–653. doi:10.1016/j.ahj.2005.05.007

The Effectiveness of Health Informatics

Campbell, E. M., Sittig, D. F., Ash, J. S., Guappone, K. P., & Dykstra, R. H. (2006). Types of unintended consequences related to computerized provider order entry. Journal of the American Medical Informatics Association, 13(5), 547–556. doi:10.1197/jamia.M2042

Crosson, J. C., Ohman-Strickland, P. A., Hahn, K. A., DiCicco-Bloom, B., Shaw, E., Orzano, A. J., & Crabtree, B. F. (2007). Electronic medical records and diabetes quality of care: results from a sample of family medicine practices. Annals of Family Medicine, 5(3), 209–215. doi:10.1370/afm.696

Cavanagh, K., Shapiro, D. A., Van Den Berg, S., Swain, S., Barkham, M., & Proudfoot, J. (2006). The effectiveness of computerized cognitive behavioural therapy in routine care. The British Journal of Clinical Psychology, 45(Pt 4), 499–514. doi:10.1348/014466505X84782

Davis, K., Doty, M. M., Shea, K., & Stremikis, K. (2009). Health information technology and physician perceptions of quality of care and satisfaction. Health Policy (Amsterdam), 90(2-3), 239–246. doi:10.1016/j.healthpol.2008.10.002

Chaudhry, B., Wang, J., Wu, S., Maglione, M., Mojica, W., & Roth, E. (2006). Systematic review: impact of health information technology on quality, efficiency, and costs of medical care. Annals of Internal Medicine, 144(10), 742–752. Ciborra, C. (2002). The labyrinths of information, challenging the wisdom of systems. Oxford, UK: Oxford University Press. Cintron, A., Phillips, R., & Hamel, M. B. (2006). The effect of a web-based, patient-directed intervention on knowledge, discussion, and completion of a health care proxy. Journal of Palliative Medicine, 9(6), 1320–1328. doi:10.1089/ jpm.2006.9.1320 Clarke, K., Hartswood, M., Procter, R., Rouncefield, M., & Slack, R. (2003). Trusting the record. Methods of Information in Medicine, 42(4), 345–352. Coiera, E. (2003). Guide to health informatics (2nd ed.). New York: Arnold; Distributed in the USA by Oxford University Press. Cook, R. I., Woods, D. D., & Miller, C. (1998). A tale of two stories: contrasting views of patient safety. Chicago: National Patient Safety Foundation.

Del Beccaro, M. A., Jeffries, H. E., Eisenberg, M. A., & Harry, E. D. (2006). Computerized provider order entry implementation: no association with increased mortality rates in an intensive care unit. Pediatrics, 118(1), 290–295. doi:10.1542/ peds.2006-0367 Dexter, P. R., Perkins, S. M., Maharry, K. S., Jones, K., & McDonald, C. J. (2004). Inpatient computer-based standing orders vs physician reminders to increase influenza and pneumococcal vaccination rates: a randomized trial. Journal of the American Medical Association, 292(19), 2366–2371. doi:10.1001/jama.292.19.2366 Dick, R. S., Steen, E. B., & Detmer, D. E. (Eds.). (1997). The computer-based patient record (Rev. ed.). Washington, DC: National Academy Press. Edmondson, A. C., Winslow, A. B., Bohmer, R. M. J., & Pisano, G. P. (2003). Learning how and learning what: effects of tacit and codified knowledge on performance improvement following technology adoption. Decision Sciences, 34(2), 197–223. doi:10.1111/1540-5915.02316 Evans, R. S., Pestotnik, S. L., Classen, D. C., Clemmer, T. P., Weaver, L. K., & Orme, J. F. Jr,… Burke, J. P. (1998). A computer-assisted management program for antibiotics and other antiinfective agents. The New England Journal of Medicine, 338, 232–238. doi:10.1056/ NEJM199801223380406

43

The Effectiveness of Health Informatics

Ewusi-Mensah, K. (2003). Software development failures. Cambridge, MA: The MIT Press. Farley, D., Damberg, C., Berry, S., Sorbero, M., Teleki, S., Ricc, K., & Pollock, N. (2005). Assessment of the National patient Safety Initiative. Context and Baseline Evaluation Report 1. Rand Corporation, United States. Agency for Healthcare Research and Quality. Feldstein, A., Elmer, P. J., Smith, D. H., Herson, M., Orwoll, E., & Chen, C.,… Swain, M. C. (2006). Electronic medical record reminder improves osteoporosis management after a fracture: a randomized, controlled trial. Journal of the American Geriatrics Society, 54(3), 450–457. doi:10.1111/j.1532-5415.2005.00618.x Ford, E. W., Menachemi, N., Peterson, L. T., & Huerta, T. R. (2009). Resistance is futile: but it is slowing the pace of EHR adoption nonetheless. Journal of the American Medical Informatics Association, 16(3), 274–281. doi:10.1197/jamia. M3042 Fries, J. F. (1994). Can prevention lower health costs by reducing demand? Yes. Hospitals & Health Networks, 68(3), 10. Galanter, W. L., Polikaitis, A., & DiDomenico, R. J. (2004). A trial of automated safety alerts for inpatient digoxin use with computerized physician order entry. Journal of the American Medical Informatics Association, 11(4), 270–277. doi:10.1197/jamia.M1500 Gans, D., Kralewski, J., Hammons, T., & Dowd, B. (2005). Medical groups’ adoption of electronic health records and information systems. Health Affairs (Project Hope), 24(5), 1323–1333. doi:10.1377/hlthaff.24.5.1323 Gardner, P., & Schaffner, W. (1993). Immunization of adults. The New England Journal of Medicine, 328, 1252–1258. doi:10.1056/ NEJM199304293281708

44

Garg, A. X., Adhikari, N. K., McDonald, H., Rosas-Arellano, M. P., Devereaux, P. J., & Beyene, J. (2005). Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: a systematic review. Journal of the American Medical Association, 293(10), 1223–1238. doi:10.1001/jama.293.10.1223 Garrido, T., Jamieson, L., Zhou, Y., Wiesenthal, A., & Liang, L. (2005). Effect of electronic health records in ambulatory care: retrospective, serial, cross sectional study. British Medical Journal, 330, 581–585. doi:10.1136/bmj.330.7491.581 Girosi, F., Meili, R., & Scoville, R. (2005). Extrapolating evidence of health information technology and costs. Santa Monica, CA: RAND Corporation. Goldzweig, C. L., Towfigh, A., Maglione, M., & Shekelle, P. G. (2009). Costs and benefits of health information technology: new trends from the literature. Health Affairs (Project Hope), 28(2), w282–w293. doi:10.1377/hlthaff.28.2.w282 Grime, P. R. (2004). Computerized cognitive behavioural therapy at work: a randomized controlled trial in employees with recent stress-related absenteeism. Occupational Medicine (Oxford, England), 54(5), 353–359. doi:10.1093/occmed/ kqh077 Han, Y. Y., Carcillo, J. A., Venkataraman, S. T., Clark, R. S., Watson, R. S., & Nguyen, T. C. (2005). Unexpected increased mortality after implementation of a commercially sold computerized physician order entry system. Pediatrics, 116(6), 1506–1512. doi:10.1542/peds.2005-1287 Harpole, L. H., Khorasani, R., Fiskio, J., Kuperman, G. J., & Bates, D. W. (1997). Automated evidence-based critiquing of orders for abdominal radiographs: impact on utilization and appropriateness. Journal of the American Medical Informatics Association, 4(6), 511–521.

The Effectiveness of Health Informatics

Jacobi, C., Morris, L., Beckers, C., BronischHoltze, J., Winter, J., Winzelberg, A. J., & Taylor, C. B. (2007). Maintenance of internet-based prevention: a randomized controlled trial. The International Journal of Eating Disorders, 40(2), 114–119. doi:10.1002/eat.20344

Koppel, R., Wetterneck, T., Telles, J. L., & Karsh, B. T. (2008). Workarounds to barcode medication administration systems: their occurrences, causes, and threats to patient safety. Journal of the American Medical Informatics Association, 15(4), 408–423. doi:10.1197/jamia.M2616

James, B.C. (1989). Improving quality can reduce costs. QA Review: quality assurance and news, 1(1), 4.

Landauer, T. K. (1995). The trouble with computers: usefulness, usability, and productivity. Cambridge, MA: MIT Press.

Jha, A. K., Doolan, D., Grandt, D., Scott, T., & Bates, D. W. (2008). The use of health information technology in seven nations. International Journal of Medical Informatics, 77(12), 848–854. doi:10.1016/j.ijmedinf.2008.06.007

Leape, L. L., & Berwick, D. M. (2005). Five years after To Err Is Human: What have we learned? Journal of the American Medical Association, 293(19), 2384–2390. doi:10.1001/ jama.293.19.2384

Johns, R. J., & Blum, B. I. (1979). The use of clinical information systems to control cost as well as to improve care. Transactions of the American Clinical and Climatological Association, 90, 140–152.

Lester, W. T., Grant, R. W., Barnett, G. O., & Chueh, H. C. (2006). Randomized controlled trial of an informatics-based intervention to increase statin prescription for secondary prevention of coronary disease. Journal of General Internal Medicine, 21(1), 22–29. doi:10.1111/j.15251497.2005.00268.x

Kaushal, R., Jha, A. K., Franz, C., Glaser, J., Shetty, K. D., & Jaggi, T. (2006). Return on investment for a computerized physician order entry system. Journal of the American Medical Informatics Association, 13(3), 261–266. doi:10.1197/jamia. M1984 Kemper, A. R., Uren, R. L., & Clark, S. J. (2006). Adoption of electronic health records in primary care pediatric practices. Pediatrics, 118(1), e20– e24. doi:10.1542/peds.2005-3000 Kohn, L. T., Corrigan, J. M., & Donaldson, M. S. (Eds.). (2000). To err is human, building a safer health system. Washington, DC: National Academy Press. Koppel, R., Metlay, J. P., Cohen, A., Abaluck, B., Localio, A. R., Kimmel, S. E., & Strom, B. L. (2005). Role of computerized physician order entry systems in facilitating medication errors. Journal of the American Medical Association, 293(10), 1197–1203. doi:10.1001/jama.293.10.1197

Linder, J. A., Ma, J., Bates, D. W., Middleton, B., & Stafford, R. S. (2007). Electronic health record use and the quality of ambulatory care in the United States. Archives of Internal Medicine, 167(13), 1400–1405. doi:10.1001/archinte.167.13.1400 Lucas, H. C. Jr. (1975). Why information systems fail. New York: Columbia University Press. Mattocks, K., Lalime, K., Tate, J. P., Giannotti, T. E., Carr, K., & Carrabba, A. (2007). The state of physician office-based health information technology in Connecticut: current use, barriers and future plans. Connecticut Medicine, 71(1), 27–31. McDonald, C. J. (1976). Protocol-based computer reminders, the quality of care and the nonperfectability of man. The New England Journal of Medicine, 295(24), 1351–1355.

45

The Effectiveness of Health Informatics

McGlynn, E. A., Ash, S. M., Adams, J., Keesey, J., Hicks, J., DeCristoforo, A., & Kerr, E. A. (2003). The quality of health care delivered to adults in the United States. The New England Journal of Medicine, 348, 2635–2645. doi:10.1056/ NEJMsa022615 McMahon, G. T., Gomes, H. E., Hickson Hohne, S., Hu, T. M., Levine, B. A., & Conlin, P. R. (2005). Web-based care management in patients with poorly controlled diabetes. Diabetes Care, 28(7), 1624–1629. doi:10.2337/diacare.28.7.1624 Mekhjian, H. S., Kumar, R. R., Kuehn, L., Bentley, T. D., Teater, P., & Thomas, A. (2002). Immediate benefits realized following implementation of physician order entry at an academic medical center. Journal of the American Medical Informatics Association, 9(5), 529–539. doi:10.1197/ jamia.M1038 Miller, A. R., & Tucker, C. (2009). Privacy protection and technology diffusion: the case of electronic medical records. Management Science, 55(7), 1077–1093. doi:10.1287/mnsc.1090.1014 Miller, R. H., West, C., Brown, T. M., Sim, I., & Ganchoff, C. (2005). The value of electronic health records in solo or small group practices. Health Affairs (Project Hope), 24(5), 1127–1137. doi:10.1377/hlthaff.24.5.1127 Mullett, C. J., Evans, R. S., Christenson, J. C., & Dean, J. M. (2001). Development and impact of a computerized pediatric antiinfective decision support program. Pediatrics, 108(4), E75. doi:10.1542/peds.108.4.e75 Murray, M. D., Harris, L. E., Overhage, J. M., Zhou, X. H., Eckert, G. J., & Smith, F. E. (2004). Failure of computerized treatment suggestions to improve health outcomes of outpatients with uncomplicated hypertension: results of a randomized controlled trial. Pharmacotherapy, 24(3), 324–337. doi:10.1592/phco.24.4.324.33173

46

Nebeker, J. R., Hoffman, J. M., Weir, C. R., Bennett, C. L., & Hurdle, J. F. (2005). High rates of adverse drug events in a highly computerized hospital. Archives of Internal Medicine, 165(10), 1111–1116. doi:10.1001/archinte.165.10.1111 Nemeth, C., & Cook, R. (2005). Hiding in plain sight: What Koppel et al. tell us about healthcare IT. Journal of Biomedical Informatics,38(4), 262263. doi:10.1016/j.jbi.2005.05.010 Niazkhani, Z., Pirnejad, H., Berg, M., & Aarts, J. (2009). The impact of computerized provider order entry systems on inpatient clinical workflow: a literature review. Journal of the American Medical Informatics Association, 16(4), 539–549. doi:10.1197/jamia.M2419 O’Neill, L., & Klepack, W. (2007). Electronic medical records for a rural family practice: a case study in systems development. Journal of Medical Systems, 31(1), 25–33. doi:10.1007/ s10916-006-9040-1 Orszag, P. R. (2008). Evidence on the costs and benefits of health information technology. Washington, DC: The Office of the National Coordinator of Health Information Technology. Orszag, P. R., & Ellis, P. (2007). The challenge of rising health care costs--a view from the Congressional Budget Office. The New England Journal of Medicine, 357(18), 1793–1795. doi:10.1056/ NEJMp078190 Osheroff, J. A., Teich, J. M., Middleton, B. F., Steen, E. B., Wright, A., & Detmer, D. E. (2006). A roadmap for national action on clinical decision support (No. HHSP233200500877P). Bethesda, MD.

The Effectiveness of Health Informatics

Ozdas, A., Speroff, T., Waitman, L. R., Ozbolt, J., Butler, J., & Miller, R. A. (2006). Integrating “best of care” protocols into clinicians’ workflow via care provider order entry: impact on qualityof-care indicators for acute myocardial infarction. Journal of the American Medical Informatics Association, 13(2), 188–196. doi:10.1197/jamia. M1656 Palen, T. E., Raebel, M., Lyons, E., & Magid, D. M. (2006). Evaluation of laboratory monitoring alerts within a computerized physician order entry system for medication orders. The American Journal of Managed Care, 12(7), 389–395. Pan, W.T. (2004, November 18 - 19). Health information technology 2004: improving chronic disease care in California. California HealthCare Foundation. San Francisco, CA: SBC Park. Perreault, L., & Metzger, J. (1993). A pragmatic framework for understanding clinical decision support. Journal of Healthcare Information Management, 13(2), 5–21. Pizziferri, L., Kittler, A. F., Volk, L. A., Honour, M. M., Gupta, S., & Wang, S. (2005). Primary care physician time utilization before and after implementation of an electronic health record: a timemotion study. Journal of Biomedical Informatics, 38(3), 176–188. doi:10.1016/j.jbi.2004.11.009 Poissant, L., Pereira, J., Tamblyn, R., & Kawasumi, Y. (2005). The impact of electronic health records on time efficiency of physicians and nurses: a systematic review. Journal of the American Medical Informatics Association, 12(5), 505–516. doi:10.1197/jamia.M1700 Potts, A. L., Barr, F. E., Gregory, D. F., Wright, L., & Patel, N. R. (2004). Computerized physician order entry and medication errors in a pediatric critical care unit. Pediatrics, 113(Pt 1), 59–63. doi:10.1542/peds.113.1.59

Proudfoot, J., Ryden, C., Everitt, B., Shapiro, D. A., Goldberg, D., & Mann, A. (2004). Clinical efficacy of computerised cognitive-behavioural therapy for anxiety and depression in primary care: randomised controlled trial. The British Journal of Psychiatry, 185, 46–54. doi:10.1192/bjp.185.1.46 Rothschild, J. M., Federico, F. A., Gandhi, T. K., Kaushal, R., Williams, D. H., & Bates, D. W. (2002). Analysis of medication-related malpractice claims: causes, preventability, and costs. Archives of Internal Medicine, 162(21), 2414–2420. doi:10.1001/archinte.162.21.2414 Rothschild, J. M., McGurk, S., Honour, M., Lu, L., McClendon, A. A., & Srivastava, P. (2007). Assessment of education and computerized decision support interventions for improving transfusion practice. Transfusion, 47(2), 228–239. doi:10.1111/j.1537-2995.2007.01093.x Roumie, C. L., Elasy, T. A., Greevy, R., Griffin, M. R., Liu, X., & Stone, W. J. (2006). Improving blood pressure control through provider education, provider alerts, and patient education: a cluster randomized trial. Annals of Internal Medicine, 145(3), 165–175. Safran, C. (1999). Editorial. International Journal of Medical Informatics, 54(3), 155–156. doi:10.1016/S1386-5056(99)00003-9 Sands, D. Z. (1999). Electronic patient-centered communication: managing risks, managing opportunities, managing care. The American Journal of Managed Care, 5(12), 1569–1571. Sequist, T. D., Gandhi, T. K., Karson, A. S., Fiskio, J. M., Bugbee, D., & Sperling, M. (2005). A randomized trial of electronic clinical reminders to improve quality of care for diabetes and coronary artery disease. Journal of the American Medical Informatics Association, 12(4), 431–437. doi:10.1197/jamia.M1788

47

The Effectiveness of Health Informatics

Shulman, R., Singer, M., Goldstone, J., & Bellingan, G. (2005). Medication errors: a prospective cohort study of hand-written and computerised physician order entry in the intensive care unit. Critical Care (London, England), 9(5), R516– R521. doi:10.1186/cc3793 Slack, W. V. (2001). Cybermedicine: how computing empowers doctors and patients for better care. San Francisco, CA: Jossey-Bass. Smith, D. H., Perrin, N., Feldstein, A., Yang, X., Kuang, D., & Simon, S. R. (2006). The impact of prescribing safety alerts for elderly persons in an electronic medical record: an interrupted time series evaluation. Archives of Internal Medicine, 166(10), 1098–1104. doi:10.1001/ archinte.166.10.1098 Speicher, C. D. S.J.J. (1983). Choosing Effective Laboratory Tests, W.B. Saunders, (Ed.). Philadelphia, PA: W.B.Saunders. Steele, A. W., Eisert, S., Witter, J., Lyons, P., Jones, M. A., Gabow, P., & Ortiz, E. (2005). The effect of automated alerts on provider ordering behavior in an outpatient setting. PLoS Medicine, 2(9), e255. Retrieved from Doi: 10.1371/journal. pmed.0020255 Studdert, D. M., Mello, M. M., Gawande, A. A., Gandhi, T. K., Kachalia, A, & Yoon, C.,… Brennan, T.A. (2006). Claims, errors, and compensation payments in medical malpractice litigation. The New England Journal of Medicine, 354, 2024–2033. doi:10.1056/NEJMsa054479 Tang, P. C., LaRosa, M. P., Newcomb, C., & Gorden, S. M. (1999). Measuring the effects of reminders for outpatient influenza immunizations at the point of clinical opportunity. Journal of the American Medical Informatics Association, 6, 115–121.

48

Tierney, W. M., Overhage, J. M., Murray, M. D., Harris, L. E., Zhou, X. H., & Eckert, G. J. (2005). Can computer-generated evidencebased care suggestions enhance evidence-based management of asthma and chronic obstructive pulmonary disease? A randomized, controlled trial. Health Services Research, 40(2), 477–497. doi:10.1111/j.1475-6773.2005.0t369.x Tierney, W. M., Rotich, J. K., Hannan, T. J., Siika, A. M., Biondich, P. G., & Mamlin, B. W. (2007). The AMPATH medical record system: creating, implementing, and sustaining an electronic medical record system to support HIV/AIDS care in western Kenya. Studies in Health Technology and Informatics, 129(Pt 1), 372–376. Toth-Pal, E., Nilsson, G. H., & Furhoff, A. K. (2004). Clinical effect of computer generated physician reminders in health screening in primary health care--a controlled clinical trial of preventive services among the elderly. International Journal of Medical Informatics, 73(9-10), 695–703. doi:10.1016/j.ijmedinf.2004.05.007 van der Sijs, H., Aarts, J., Vulto, A., & Berg, M. (2006). Overriding of drug safety alerts in computerized physician order entry. Journal of the American Medical Informatics Association, 13(2), 138-147. Virapongse, A., Bates, D. W., Shi, P., Jenter, C. A., Volk, L. A., Kleinman, K. et al. (2008). Electronic health records and malpractice claims in office practice. Archives of Internal Medicine,168(21), 2362-7. Wagner, B., Knaevelsrud, C., & Maercker, A. (2006). Internet-based cognitive-behavioral therapy for complicated grief: a randomized controlled trial. Death Studies, 30(5), 429–453. doi:10.1080/07481180600614385 Walker, J., Pan, E., Johnston, D., Adler-Milstein, J., Bates, D. W., & Middleton, B. (2005). The value of health care information exchange and interoperability. Health Affairs (Project Hope), (Suppl Web Exclusives), W5-10–W15-18.

The Effectiveness of Health Informatics

Weick, K. E. (1990). The vulnerable system - an analysis of the Tenerife air disaster. Journal of Management, 16(3), 571–596. doi:10.1177/014920639001600304 Williamson, J. W., Goldschmidt, P. G., & Jillson, I. A. (1979). Medical Practice Information Demonstration Project: Final Report. Baltimore, MD: Policy Research Inc. Office of the Assistant Secretary of Health, Department of Health, Education, and Welfare, contract 28277-0068GS.

ENDNOTES 1.



2.



between 50 per cent and over 90 per cent (Evans et al., 1998; Potts, Barr, Gregory, Wright, & Patel, 2004). A review of studies on clinical decision support found that most such functions improved the performance of practitioners – see Garg et al., 2005. On the other hand, other research finds no evidence of an increase in physicians’ adherence to evidence-based standards of treatment for a wide variety of conditions – see for instance Crosson et al., 2007, and Linder, Ma, Bates, Middleton, & Stafford, 2007.

Some studies suggest potential reductions in error rates from the use of health IT of

This work was previously published in Healthcare and the Effect of Technology: Developments, Challenges and Advancements, edited by Stéfane M. Kabene, pp. 13-37, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

49

50

Chapter 1.4

Personal Health Information in the Age of Ubiquitous Health David Wiljer University Health Network, Canada Sara Urowitz University Health Network, Canada Erin Jones University Health Network, Canada

ABSTRACT We have long passed through the information age into an information perfusion in health care, and new strategies for managing it are emerging. The ubiquity of health information has transformed the clinician, the public, and the patient, forever changing the landscape of health care in the shift toward consumerism and the notion of the empowered patient. This chapter explores essential issues of ubiquitous health information (UHI), beginning with its origins in the explosion of health information and the advent of new technologies. Challenges of UHI include privacy issues, change management, and the lack of basic infrastructure. DOI: 10.4018/978-1-60960-561-2.ch104

However, benefits for patients include improvements in access to information, communication with providers, prescription renewals, medication tracking, and the ability to self-manage their conditions. Benefits at the organizational level include increased patient satisfaction, continuity of care, changes in costing models and improved standardization of care as organizations streamline processes to address this change in clinical practice.

INTRODUCTION In health care the “information age” has long passed and we have entered into the era of information perfusion. The ubiquity of health information

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Personal Health Information in the Age of Ubiquitous Health

has transformed the clinician, the public, and the patient. As technology progresses and we see exciting and innovative strategies for managing it emerge, ubiquitous health information (UHI) has brought on a tectonic shift that will forever change the landscape of health care. This chapter explores the essential issues of UHI: the debates and controversies, the risks and benefits, and efforts that must be made to manage them. To begin, we trace the origins of UHI – the rise and explosion of several genres of health information in conjunction with the evolution of new technologies. We look at the definition and components of UHI, the types of health information that are available, and the methods for their exchange. We also explore the rise of consumerism and the notion of empowered patients within the context of ubiquitous health information. The chapter also examines the role of UHI in the changing landscape of health care. We investigate the growing number of social, economic, cultural, ethical and legal issues, as technologies allow for the dissemination and exchange of personal health information. The impact of the perfusion of information on the public and patient is explored, as well as its impact on health professionals and the type of care delivered, and the new and alternative environments in which it is carried out. The benefits and the risks of UHI are discussed, including the educational, clinical and research opportunities. And finally, we offer a consideration of future research directions, and potential frameworks for the evaluation and assessment of UHI.

BACKGROUND Ubiquitous Heath Information Today: Towards a Definition The perfusion of health information today can be overwhelming. Health information is now everywhere, all the time. We see messages about

our health from television, radio, newspapers, the Internet, billboards and advertising. Health services and health products are an enormous industry, and controlling the messages is an ongoing battle between multiple interest groups that is being waged at the expense of those individuals who need it most – the public and the patients. Media and web sites are becoming battlegrounds over major health issues of the twenty-first century, ranging from circumcision and vaccinations to novel and alternative treatments. The social and financial stakes are high, and information flows at a rate that is almost incomprehensible. As new strategies for navigating this sea of information develop, so our ability to generate more information increases, contributing to this state of UHI. The emergence of new technologies, and the changing expectations of health care consumers have each influenced the explosion of UHI, both in their own right, and through an intricate relationship to each other. It is evident that healthcare has experienced a shift toward consumerism and the notion of the empowered patient. Patients and the public are no longer satisfied with the status quo and a growing wave of public and patient expectation is mounting (Ball, Costin, & Lehmann, 2008; Hassol et al., 2004; Leonard, Casselman, & Wiljer, 2008; Leonard & Wiljer, 2007; Pyper, Amery, Watson, & Crook, 2004; Wiljer et al., 2006). The traditional health care system that was characterized by the physician driven model, an insular system of practice that neither had the means nor the inclination to share expert knowledge, is now a thing of the past. Today, patients are acknowledged as expert consumers. Health care providers work in partnership with their patients to make shared decisions, and there is the growing global trend of adopting legislation to ensure that patients are able to access, review, and amend their medical record (Beardwood & Kerr, 2005; Blechner & Butera, 2002; Chasteen, Murphy, Forrey, & Heid, 2003; Dietzel, 2002; France & Gaunt, 1994; Harman, 2005; Ishikawa et al., 2004; Jones, 2003; Kluge,

51

Personal Health Information in the Age of Ubiquitous Health

1993; Mathews, 1998; Mitchell, 1998; Pyper, Amery, Watson, Crook, & Thomas, 2002). As a result, the information asymmetry that once existed between health care professionals and the health care consumer is now significantly diminished and UHI is pervasive. At the same time, technology is advancing. The role of information and communications technologies (ICTs) in health care is growing, and evidence of this trend is all around us. Research has shown that the number of MEDLINE citations for “Web-based therapies” showed a 12-fold increase between 1996 and 2003 (Wyatt & Sullivan, 2005). There are virtual health networks and electronic health records (EHRs) being used to coordinate the delivery of services, and knowledge management ICTs used to establish care protocols, scheduling, and information directories. ICTs are also increasingly seen as a key element of consumer-based health education, and the delivery of evidence-based clinical protocols (Branko, Lovell, & Basilakis, 2003). Health care teams are finding increases in efficiency and efficacy when new technologies help to manage patient data and provide a method of coordinating team members’ interactions (Wiecha & Pollard, 2004). There are many potential benefits to be realized by effective use of UHI. Patient benefits include better access to health information, increased ability to self-manage chronic health conditions, increased medication tracking, safer prescription renewals, and improved connections for patients and providers (Ball, Smith, & Bakalar, 2007; Tang, Ash, Bates, Overhage, & Sands, 2006). Potential benefits at the professional level include increased patient satisfaction, continuity of care, changes in costing models and improved standardization of care as organizations streamline processes and information to address this change in clinical practice (Ball et al., 2007; Tang et al., 2006). There are still a number of barriers to overcome, including privacy and security issues, change management issues, and the lack of basic infrastructure such as EHRs (Tang et al., 2006).

52

UHI has transformed the entire concept of health, and we are seeing rapid changes in the health delivery system. One prominent idea is the prospect of chronic disease management and self-managed care, facilitated through UHI. This ambition has inspired a growing interest in harnessing the power of EHRs beyond the point of care delivery. Health care organizations are also realizing the potential benefits of patient accessible health records (PAEHRs), including improving the patient experience, supporting patients with chronic conditions, improving transparency, increasing referral rates, and ensuring the continuity of care beyond the hospital walls.

THE HISTORY OF UHI: HOW DID WE GET HERE? Health Care Consumerism The role of the patient has been impacted by societal influences, starting with the Civil Rights movement and continuing through the 1980’s and beyond. Expectations are changing, and patients are no longer seen to be passive recipients of health care information, but rather are considered to be health care consumers who actively seek out, and more recently, create knowledge (Cresci, Morrell, & Echt, 2004; Urowitz & Deber, 2008).

Shared Decision Making As consumers of health care, people are more invested in actively participating in their care and making informed health care choices. This is commonly referred to as shared decision making (Charles, Gafni, & Whelan, 1997; Deber, Kraetschmer, Urowitz & Sharpe, 2005; Deber, Kraetschmer, Urowitz & Sharpe 2007). Effective shared decision-making requires that people are able to access and understand large amounts of complex medical and health related information. Once, this would have been a daunt-

Personal Health Information in the Age of Ubiquitous Health

ing task for most individuals. It would require trips to specialized libraries, an understanding of professional textbooks, and synthesizing large amounts of information from both traditional and non-traditional sources. Now, the pervasiveness of technology-supported health information has simplified shared decision-making. Video, CD-ROMs and the Internet have brought health information more into the public domain, making expert information available to the masses. Decision aids are one possibility for supporting health care consumers through this process. Decision aids provide health information to help people understand their options, and to consider the personal importance of possible benefits and harms (O’Connor et al., 2002). Like in prostate cancer, they are most helpful when there is more than one medically reasonable option. Information in decision aids was traditionally delivered through pamphlets and more recently through videos, CDs and Internet based programs. For instance, the Ottawa Health Research institute offers a decision aids page (http://decisionaid.ohri. ca) that allows the public to search their inventory of decision aids, or download a generic decision strategy sheet, which can be applied to the process of making any health decision. Shared decision making has been greatly facilitated by the introduction of the Internet, which has enabled health care consumers to independently seek out information that ranges from the layperson’s to the expert’s level. The National Library of Medicine in the United States has a section of their web page (www.nlm.nih.gov) designated “especially for the public” where users can find current and accurate health information for patients, families and friends.

The Chronic Care Model and Chronic Disease Management The welcoming of patients’ participation in their own care, combined with the abundance of health related information available through the Internet,

is continuing to drive a shift in the delivery of health care and increasing the demand for accessible, timely and relevant health information. We are moving away from the clinician driven model of care toward a patient centred model, in which empowered patients are encouraged to play an active role, not only in the decision making process, but in their own care management. This model is particularly relevant to the growing segment of the population living with long-term chronic illnesses that require ongoing monitoring and treatment. The aging population and advances in modern medicine have resulted in larger numbers of people living longer and experiencing more chronic illnesses, many dealing with multiple conditions at the same time (Orchard, Green, Sullivan, Greenberg, & Mai, 2008; Ralston et al., 2007). Whereas a disease such as cancer once had a high probability of mortality, people are now living longer, disease-free lives, even after a diagnosis of this once fatal condition (Hewitt & Ganz, 2006). The Chronic Care Model (CCM) and the paradigm of chronic disease management (CDM) are responses to the changing health care needs of the population. The Chronic Care Model suggests that optimal functional and clinical outcomes can be achieved in a system that supports productive interactions between informed clients, which includes patients, their families, and knowledgeable providers, who are working as part of a team with a client-centered focus (Wagner, 1998). CDM espouses the importance of people who are living with chronic illness taking responsibility for effectively managing their illness. People need to see themselves as active partners in their care with a keen understanding of what is best for them and their bodies, who can make behaviour changes through goal setting, but also know when to ask for help. Adopting this approach to disease management can result in overall improvements in long-term health (Lorig, Sobel, Ritter, Laurent, & Hobbs, 2001). There are 5 steps to effective CDM 1) Identification of a problem 2) Goal setting 3) Planning

53

Personal Health Information in the Age of Ubiquitous Health

ahead and considering obstacles to achieving goals 4) Determining confidence level for achieving the goal, and 5) Following up and staying connected with the health care team. Applying these 5 steps is an effective approach to changing behaviours. Since behaviour changes can be difficult to effect, access to UHI can facilitate knowledge sharing, education, and an understanding of the condition which are all factors in ensuring success (Wyatt & Sullivan, 2005). CDM is most effective when coupled with personal empowerment interventions. These interventions might consist of education in improving individual decision-making, disease complication management, efficient use of health services, improved health behaviors, or life coping skills (Paterson, 2001; Wallerstein, 2006). Empowerment interventions aim to increase the capacity of individuals to make choices and transform those choices into desired actions and outcomes (Wallerstein, 2006). The empowerment facets of CDM include the promotion of new knowledge and skills, within a system that reinforces active participation (Lorig et al., 2001; Wallerstein, 2006), resulting in better health outcomes and improvements in quality of life for the chronically ill (Wallerstein, 2006).

information becomes key to maintaining good continuity of care, and now UHI is enabling the provision of this sort of care in a less fragmented manner (Anderson & Knickman, 2001). In fact, the emerging goal for information technologies managing UHI is to be unobtrusive, seamlessly working in the background to provide users with value-added services (Pallapa & Das, 2007). As we approach this goal, technology is being more and more commonly applied to the home monitoring of vital signs for patients with chronic disease, and telemedicine is beginning to replace some home nursing visits (Branko et al., 2003). For example, remote patient monitoring for home haemodialysis utilized a system that that acquired, transmitted, stored and processed patient vital signs that were monitored according to algorithms that triggered alarms if intervention was required. In addition, the system included an IP-based pan-tilt-zoom video camera for clinical staff to observe remotely if the patient desired (Cafazzo, K.J, Easty, Rossos, & Chan, 2008). The hope is that UHI, coupled with the right technology, will be responsive to patient demand; ensuring that information can be personalized for the individual seeking it, and that services can be delivered at the time and the place they are needed (Wyatt & Sullivan, 2005).

UHI and the Management of Chronic Illness

Exchanging Health Information

The growth of technology and the appearance of UHI have had an impact on the rise of CDM, especially in terms of the increasing capacity to share information. Often, patients with chronic illnesses are seeing many physicians and specialists, who will all need access to their information. For example, a diabetes patient may be seeing his or her family doctor along with any number of specialists, including nephrologists, podiatrists, and ophthalmologists. All of these specialists may be prescribing various medications, ordering tests or making additional referrals to still other health care providers. Timely transmission of this

The notion of technology assisted exchange of health information is not a new idea. It has been over four decades since physicians first started using interactive video as a way of providing and supporting health care when distances separated participants. This was one of the earliest forms of telemedicine, or the use of electronic information and communication technologies for the provision of health care. Most commonly telemedicine was restricted to the sub-specialities (Field & Grigsby, 2002), but as Internet technology progresses, and the growth of health consumerism increases, our expectations and information practices in health

54

Personal Health Information in the Age of Ubiquitous Health

care are changing. Innovative means of communication that were once limited to sub-speciality health care providers are becoming more common place. Traditional ways of providing information are still in use in many health care institutions, but health care is slowly beginning to harness the power of the Internet to store, manage, exchange, and share information. More recently, the collaborative properties of Web 2.0 applications are making information sharing possible for a wide range of participants, including patients, providers, caregivers and researchers through the use of collaborative, adaptive, and interactive technologies (CAIT) that (1) facilitate collaboration among users in traditional or novel ways, (2) support adaptation of form, function, and content according to user needs or preferences, and (3) enable users to interact with the technology via mechanisms of explicit interaction (O’Grady et al., 2009). Currently, there are already some guidelines in place for providing patients with information. Many jurisdictions have sophisticated legal requirements that protect not only the privacy of patients, but also their right to access this information in a timely and accessible manner (Wiljer & Carter, In Press). There are several obstacles to meeting these requirements using the traditional methods of delivering patient information; that is, locating a patient’s file – either paper or electronic – and producing a hard copy of the information. There are often limited resources to dedicate to providing this information. Nurses and administrative staff already face a time crunch in dealing with the volume of their work. In addition, certain jurisdictions suggest that an explanation of medical terms be given along with the information (Wiljer & Carter, In Press). This is often not feasible for the same reasons: time constraints on overloaded staff, and a lack of infrastructure for providing the information. In some cases there may be pre-existing pamphlets or glossaries available; however, these resources contain very generic content which may or may

not address the specific information the patient is looking for. The traditional methods of delivering information have always faced these barriers, but as information continues to proliferate, new strategies and, correspondingly, new challenges are coming to the fore.

Electronic Health Records (EHRs) The pervasiveness of the Internet supports new methods for accessing patient information. Electronic Health Records (EHRs) are now increasingly being used to record and manage patient data. An EHR can be defined in its simplest form as a computerized version of a traditional health record. Detailed definitions can be more complex, however, as an EHR may contain patient’s full medical record, or may be used only for recording certain aspects of care, such as lab results (Urowitz et al., 2008; Tang et al., 2006). EHRs can facilitate the sharing of information between health care providers and with in a health care system. This sharing of information is dependent upon the existence of interoperable systems. One promising approach is through the use of health level 7 (HL7), a message-based communication system implemented by an asynchronous common communication infrastructure between facilities with different EHRs (Berler, Pavlopoulos, & Koutsouris, 2004). HL7 is now commonly used as the standard interface for the exchange of health information. The benefits of an EHR include that it can be made widely accessible, and that it can facilitate the integration of care by allowing all members of a particular patient’s care team to access his or her information at need (Urowitz et al., 2008; Tang et al., 2006). However, the use of EHRs is still in its early stages in many health systems. A Canadian study in 2008 determined that over half of the nation’s hospitals (54.2%) had some sort of EHR, but that only a very few had records that were predominantly electronic (Urowitz et al., 2008). In Japan, a similar study determined

55

Personal Health Information in the Age of Ubiquitous Health

that most hospitals (92%) used computerized administrative systems, but few (21%) used electronic “official documents”. Barriers to adoption became evident when professionals participating in the study reported that they expected efficiency and quality improvements from their EHR, but worried that the proposed system might be too complicated, threatening to slow down workflow or compromise patient privacy (Ishikawa et al., 2007). These types of barriers are of great interest, since having data in an electronic format is key to facilitating access to information.

Personal Health Records (PHRs) The development of storing data in electronic formats opens up further opportunities to integrate information. As a compliment to EHRs, which increase accessibility for professionals, there are many proponents for the implementation of Personal Health Records (PHRs). Markle Foundation’s Connecting For Health group, a public-private collaborative for interoperable health information structures, defined PHR as “An electronic application through which individuals can access, manage and share their health information, and that of others for whom they are authorized, in a private, secure and confidential environment” (Markle, 2003). There is a broad spectrum of approaches to implementing a PHR. At its simplest, a PHR may allow a patient to enter, store, and access his or her own information in a stand-alone application. In a more integrated model, the PHR could allow patients to access the information stored in their provider’s EHR, or even request appointments, renew prescriptions or communicate with their doctors online (Tang et al., 2006). PHRs can also provide a platform for communication between patients and health care providers. For example, Partners Group in Massachusetts supports the Patient Gateway. Patient Gateway allows patients to access an extract of their medical record and facilitates online communication through secure

56

messaging with medical practices (Kittler et al., 2004). The actual benefits of PHRs are being researched and documented. In a study by Wuerdeman et al (2005) it was found that medical records are often incomplete, missing data from certain test results, for example. Patient-reported data became very useful in these situations, as practitioners who need comprehensive information must often ask the patient if they have had a certain test, or what the results were. While patients’ ability to report specific numeric test results did prove less accurate than information found in the medical record, patients were more reliable in reporting the absence or presence of a test and whether or not it fell in the normal range. In addition, patients were able to report problems that may not have appeared in the symptom list of the official record, such as depression (Wuerdeman et al., 2005).

Sharing Personal Information and Web 2.0 Patients uploading their own information would not only enable the sharing of information with their doctors, but also with one another. Traditionally, patients who wished to share information about their illness joined support groups or found other ways of meeting face-to-face. This is changing with the spread of Web 2.0 culture. Web 2.0 is a somewhat nebulous concept that is coming to be defined as a general spirit of open sharing and collaboration (Giustini, 2006). Generally, the idea is that the web can act as more than a unidirectional method of publishing information. Information can travel back and forth, and users are able to both access what they need as well as contribute what they have to share. Wikipedia, an information resource that functions as an encyclopaedia that anyone can contribute to, is a well-known example of the Web 2.0 philosophy. The proliferation of these Web 2.0 technologies is adding a new exponential dimension to UHI.

Personal Health Information in the Age of Ubiquitous Health

We are seeing the use of Web 2.0 technologies in health care, especially wikis and blogs (a contraction of web logs). A wiki is a website that allows multiple users to enter information, acting as a repository of information from different resources, as well as a mode of discussion about the information added. For example, Healthocrates. com is a health care wiki that offers free membership and encourages anyone and everyone to join and become a collaborator. Healthocrates claims to be home to over 8,500 articles, adding 500 to 1,000 new items each month. “Articles” include all types of media, from video and images, to discussion forums and case reports. A blog is another type of website, containing regular entries, much like an online diary. Blogs allow their authors to write articles, or even post personal videos, as well as have discussions with readers in the form of posted comments in a miniature forum, which generally appears below the day’s entry. Patients and clinicians alike have been known to blog. One well known example is Ves Dimov’s Clinical Cases and Images, which offers a reliable “one stop shopping” location for health news and research findings taken from the web daily (Giustini, 2006). Specialized and reliable sources like these can be helpful in reducing the time patients and professionals spend wading through ubiquitous health information on the web. The introduction of UHI, and collaborative technologies such as blogs and wikis is contributing to a new phenomenon of the democratization of health information. Information that was once firmly in the strong hold of the hierarchical health system is much more freely exchanged.

Types of Health Information Even as our abilities to store, exchange, and seek out information change, the information itself is changing. As technology increases, our ability to manage large amounts of information, and as the value of different kinds of information is becoming recognized, we are seeing information that

is increasingly specific and personalized joining stores of information that were once predominantly generic. Leonard et al. have argued that there is an identifiable information paradigm for patients who are actively managing their health, especially those patients managing chronic conditions (Leonard, Wiljer, & Urowitz, 2008). The paradigm encompasses several distinct information types, including general health information, and personal health information and experiential health information.

General Health Information General health information is very pervasive. A visit to a family physician can result in large amounts of information about a wide variety of health issues, from screening programs for a wide array of diseases to lifestyle choices and concerns. General health information covers a wide spectrum, and is usually basic information about a health issue (i.e. obesity), a condition, (i.e. diabetes), or medications, (i.e. diuretics or beta blockers). This information is intended to provide a starting point to give a layperson a basic understanding of a particular aspect of health. The literature shows that many patients are still most likely to seek information directly from physicians (Rutten, Arorab, Bakosc, Azizb, & Rowland, 2005). However, information is not always clear during a clinical encounter, which is often short and overwhelming, with large quantities of information being exchanged. Traditionally, patients left their physician’s office with the essential health information that they have heard from the physician or nurse reinforced by a pamphlet or a brochure. However, new technologies now provide important alternative sources of information. The Internet has changed the dynamic of general health information. The notion of searching for information, or turning to “Dr. Google”, has been commonplace for many individuals. The pervasiveness of general health information is astounding; for example,

57

Personal Health Information in the Age of Ubiquitous Health

type “diabetes” into Google and in a tenth of a second, 96,800,000 hits are returned, “asthma” 30,900,000, “cancer” 232,000,000. At that rate, if you looked at each page for a second, it would take 7 years to view all of the information on cancer. The amount of health information on the Internet alone is beyond ubiquitous and will only continue to grow at an astounding rate. There are, of course, many different types of general information available on the Internet, and they vary greatly in quality and reliability. Many sites offer static content that is updated at regular or irregular intervals. Other sites have been developed to be more dynamic, offering tools and resources to help patients find information much more specific to their particular circumstances. For example, a palliative care site, www.CaringtotheEnd.ca, allows patients and family members to complete questionnaires based on a series of questions and then provides specific information to meet the needs of the user (Figure 1). Several technologies have also been developed to enhance the exchange of general health information between patients and clinicians. For example, a patient education prescription system, known as PEPTalk, (Figure 2,3) has been developed, that allows physicians to prescribe general health information to their patients (Atack, Luke, & Chien, 2008).

Personal Health Information It is evident that general health information has become ubiquitous, and certainly the advent of new technologies has improved access to it. Nonetheless, this of type information is of a general nature and does not provide patients with access to the information that is directly relevant to their particular situation. One of the promises of new technologies is to provide people with access to their own information. As an example, this type of access has clearly revolutionized the banking industry with the advent of online banking. Yet the

58

Figure 1. Caring to the end of life

Figure 2. PEPTalk - clinician view

Figure 3. PEPTalk – patient view

Personal Health Information in the Age of Ubiquitous Health

health care industry, overall, has been much more reluctant to embrace these types of technologies. Over the last decade, however, there has been a growing interest in utilizing new technologies to improve access to personal health information. This type of information generally includes an individual’s physical or mental health and family history, documentation of provided care, a care or service plan as well as administrative data (Cavoukian, 2004). For the moment, personal health information has the potential to be accessible anywhere the user is, but the promise has not yet been fully realized. In 2005, a collaboration began in Canada between Shared Information Management System (SIMS) Partnership and the Weekend to End Breast Cancer (WEBC) Survivorship Program began. The project was a an online portal called InfoWell, designed specifically to empower cancer patients to be active partners in their care, to participate in self-management activities and to find the support and resources they need in dealing with long-term consequences of their illness and treatment. Built using a commercially available patient portal as a foundation and customized by SIMS system engineers, InfoWell provides general health information as well as personal health information, which encompasses patientgenerated information including a patient profile, medication lists, and treatment history (Figure 4).

InfoWell also allows patients to view elements of their own EHR (Figure 5) (Leonard, Wiljer, & Urowitz, 2008).

Experiential Health Information From an early age, many of us glean valuable health information from our parents, siblings, teachers, and physicians. We learn important life lessons about things like diet, nutrition, exercise, and personal hygiene. Some of this “experiential” information is based on scientific evidence, and some of it based on family folklore, but it will all have a large impact on our lifestyle choices. It is amazing how a story or experience that a family member has when we are young can influence the day-to-day decisions that we make. The ubiquity of health experience and health information informs the very essence of who we are, whether we are always conscious of it or not. There is a specific type of experiential information that is of particular interest to most patients: the experiences and feelings of others with the same or similar conditions. Online communities are on the rise as a new way of finding and sharing this kind of information; 84% of Internet users participate in online groups, and 30% of these participants are members of medical or health related groups (Johnson & Ambrose, 2006). There are several Figure 5. InfoWell, my health – access to EHR

Figure 4. InfoWell – patient-generated PHR

59

Personal Health Information in the Age of Ubiquitous Health

reasons people go looking for experiential information. There are information needs that can be met, such as to better understand their diagnosis, or to become more informed about its treatments. However, there are also social needs that can be fulfilled by online communities, such as finding support, helping others in the same situation, or to feel less alone or afraid (Preece, 2000). To help fulfill these needs among cancer patients, a large cancer centre in Canada developed CaringVoices. ca (Figure 6), an online community that offers support for people living with cancer and their caregivers. CaringVoices.ca provides access to current educational resources, peer support, and advice and education from health care and community experts. There is also a people-matching function that allows members of the community to connect and share experiential information. A real-time chat function enables social networking between patients, as well as discussions hosted by health care professionals, volunteer cancer survivors, and staff from community cancer agencies. There are several benefits to patients’ participation in online communities. A major cause of non-compliance with a care plan is a lack of understanding of the treatment. Online communities offer patients a place to share information and foster a greater understanding of the process; they also offer a chance to connect for patients who

are separated by distance or time. This is especially helpful to those who live in rural areas or may be homebound due to illness. The holistic approach to most online communities might also help to ease the burden on the health care system of patients who might otherwise over utilize the health care system. There are potential research benefits, as well. Observing interactions between large groups of patients provides an unobtrusive method of gathering information. Patterns in discussions may emerge, producing useful information about topics like adverse drug reactions, or clustering of disease types within age groups or geographical locations (Johnson & Ambrose, 2006). The use of online communities can also have drawbacks. Especially viewed from a standpoint of ubiquitous health information, there is concern that patients may be unable to find reliable resources, and may even be mislead by inaccurate information. As a response to this, there are many communities that now offer question-and-answer periods with a real physician. However, it has been argued that these too are insufficient, as the doctors leading these sessions do not treat the patient, and have no information as to their particular medical background (Preece, 2000). As such, it is important that patients have reliable means of way-finding through such vast expanses of ubiquitous information.

Figure 6. Caring voices

BENEFITS AND RISKS: WHERE ARE WE NOW? We have traced the history of UHI and gained some insight into the different circumstance and events that have led to the vast amounts of information that surround us today. In our exploration, we have touched on the idea that one of the major benefits of growth in both technology and levels of information, is the potential to personalize. With the management of large amounts of personal information, however, also comes risk. It

60

Personal Health Information in the Age of Ubiquitous Health

is important to critically examine both the risks and benefits that face us today in order to ensure that in the future we can properly take advantage of the opportunities as well as manage our risks.

Benefits of Personal Health Information Once personal information has been captured, the benefits that can be realized from utilizing it correctly are substantial. Personal information is more and more commonly found in electronic format. This development opens up possibilities for technologies that can improve access and transportability of information, and in turn, help patients to engage as members of their own care team.

Access to Information As the role of the modern patient continues to shift toward the idea of the empowered consumer, there is an increasing demand for access to, and control over, personal information. Leonard, Wiljer and Urowitz (Leonard, Wiljer, & Urowitz, 2008), estimate that 40- 45% of the population in the United States and Canada are chronic illness patients with a strong desire for information on their condition. Members of this demographic group are knowledgeable about the health care system and eager to bypass the painstaking traditional methods of getting access to information. Wagner estimated that 40% of the population accounts for 70-80% of health care spending globally (Wagner, 1998). These numbers suggest that there are huge potential financial savings to be made by helping patients to manage their own health (Leonard et al., 2008). It has been shown, however, that improving access to information can introduce benefits beyond cost reduction. In 2003 an award-winning project entitled Getting Results focused on the potential benefits of providing electronic access to laboratory results for haematology patients who

spend long periods of time waiting in the hospital for these results. Patients and professionals alike saw benefits from the approach. Staff identified potential benefits such as improved workflow, decreased workload and a lower occurrence of unnecessary hospital visits (Wiljer et al., 2006).

Transportability Another benefit of the electronic management of personal health information is that it becomes transportable. Information that is more easily transported, is also more easily shared. Some independent or “untethered” ways of storing data are quite physically transportable: smart card technology, USB drives, or CDs, for example (Tang et al., 2006). Information that is “tethered” to a network, or stored online, can also be easy to share (Tang et al., 2006). The communication benefits of this can be seen in all types health care relationships; patient to provider, provider to provider, and patient to patient. Information that can be passed freely between patient and care provider changes care from episodic visits to a continuous process, and could dramatically reduce the time taken to address medical problems (Tang et al., 2006). In the relationship between separate providers, interoperable systems could virtually eliminate the need for paper and make important patient records instantly available in any facility involved in treatment, resulting in efficiencies and quality improvements in care (Gloth, Coleman, Philips, & Zorowitz, 2005). Additional benefits are seen when information can be passed easily from patient to patient. With the advent of Web 2.0 technology, large numbers of people can now unite and collaborate (Deshpande & Jadad, 2006). Patients are able to find one another and create communities that allow them to feel less alone or anxious, offer each other support, and increase their understanding of the information they collect at a physician visit.

61

Personal Health Information in the Age of Ubiquitous Health

Self-Managed Care Once information is personal (that is, specific to a given individual), and can be easily accessed by that individual as well as shared with care providers that need it, another avenue of patient empowerment opens up, through the possibility of self-managed care. Self-managed care involves the patient or non-professional caregiver (usually a close friend or family member) taking an active role in the patient’s care. The degree to which a patient is involved can vary substantially from patient to patient: one patient may feel very comfortable in taking an active role, navigating the system and ensuring that they receive the care that is right for them. Another patient may want to participate in their care, but have much more involvement from their health care team. The management of more clinical aspects of care, or self-care requires patients to have very specific knowledge and some training around a medical task. The type of self-care is, of course, often very specific to the type of disease that is being managed: for diabetes, it may be monitoring blood levels and taking insulin as required; for heart disease, it may be monitoring daily weight and controlling diet and intake of sodium; for breast cancer, it may monitoring activities to avoid infections and performing self-massage to control lymphedema. Advents in technology such as wearable monitoring technologies have made these tasks easier for people. As early as 1999, one set of researches have reported on a subminiature implantable device to provide remote monitoring of glucose levels and the ability to transmit the glucose concentration data to a corresponding receiver and computer (Beach, Kuster, & Moussy, 1999). Now monitoring devices connected to specialized computer modems are used to reliably measure and transmit physiological parameters including blood pressure, heart rate, blood glucose level, and pulse oximetry data (Field & Grigsby, 2002). Technologies such as portable monitoring devices, wearable sensors and even smart shirts

62

can all be used for electronically support self-care (Tan, 2005).

RISKS OF UBIQUITOUS HEALTH INFORMATION We have largely focused on the benefits and potential of UHI, but there are also risks to ubiquitous health information that can threaten an individual or their family, an organization or a community. UHI has the potential to lead to the loss of privacy for individuals, to security threats to organizations, and to the loss of social justice for communities and society. All of these risks have real implications and the use of electronic media pose risks at a much larger magnitude than paper. There have occasionally been errors in the safe disposal of paper records that lead to sensitive data appearing in inappropriate places, or the risk of losing a medical file, but the potential for wide scale breaches is much larger when dealing with electronic data. It is important therefore, to understand the risks and implications of the directions UHI is taking, in order to create protocols, policies and procedures that will ensure that as the amount of health information grows, the risks are mitigated as much as possible. In discussing the protection of information, it is useful to distinguish between the terms security, privacy, and confidentiality. Security, in all its contexts, refers to the notion of protection. In the case of sensitive information, it is the information itself that must be protected, by extension protecting the owner of that information. Security then, refers to the systems or means we use to ensure that access to protected information is appropriate. Privacy, in turn, deals with denying access that would not be appropriate, and determining policies for what information should be protected, and who should have access. Lastly, confidentiality refers to defining those situations where some access to personal data is appropriate (Wiljer & Cater, In press). Any person who is accessing data

Personal Health Information in the Age of Ubiquitous Health

that they do not own is being held in confidence, and it is understood that they will use their access only in the manner and purpose that is intended (Schoenburg, 2005).

Risks to the Individual: Privacy The greatest potential risks to the individual are around privacy issues. Awareness of these risks is increasing, and Internet users are beginning to protect their anonymity online. Members of the online community at PatientsLikeMe.com, who visit with the express purpose of sharing their information still tend to choose user names and avatars that represent them while keeping their identities anonymous. This may be especially desirable for patients living with illnesses such as HIV or mood disorders. With respect to receiving care, the notion of privacy relates to the fiduciary relationship between the patient and individual members of their health care team. The majority of privacy issues relate to data in health records such as family history, diagnostic tests, treatment and medication records and clinical notes. In most circumstances the patient and provider relationship is a private one, nonetheless, there are complex ethical issues that, at times, can impact on an individual’s right to privacy. Most jurisdictions have legislation that protect the rights of patients to privacy in general or specifically with respect to information pertaining to their health information. In the United States for example, the Health Insurance Portability and Accountability Act (HIPAA) protects personal health data.

Risks to Organizations: Security UHI also has the potential to cause security risks for organizations. The potential for wide scale breaches is much larger when dealing with electronic data, than the risks involved with paper files; errors in the safe disposal of files or the risk of losing a medical record. It is because of

this that is it so important to create protocols, policies and procedures that mitigate this risk as much as possible. Data must be protected in three major domains: the server side, where information is stored; the network, through which it travels; and the client side, the place where the information is received and used (Schoenburg, 2005). Each of these domains will require different actions to protect the data. It is important to allot equal attention to creation of protocols for each domain, since a vulnerability at any point could be the area where a potential breach would occur. Yet it is also important to find a balance between security and the need to construct usable systems. Applications that are not easily understood and used, or that are not effective in the delivery of care invite “workaround” solutions from frustrated staff who have found more efficient ways to operate. Oftentimes, workaround solutions are indeed more efficient, but they may circumvent the security protocols that are in place, creating new vulnerabilities for information to leak through.

Risks to the Community: Social Justice and Equity In addition to the issues affecting every day operations in health care, UHI has social implications on quite a broad scale. A survey in the United States (Brodie et al., 2000) shows that the Internet provides health information to large numbers of lower income, less-educated, and minority citizens. Nearly one-third of adults under age sixty are seeking such information online. And yet, there is still progress to be made. A digital divide still exists in terms of computer ownership. Lower income African Americans in the United States are far less likely than lower income whites to own a computer at home (Brodie et al., 2000) Once this barrier has been overcome, however, this study suggests that use of the internet to find health care information among computer-owners is similar across income, education, race, and age.

63

Personal Health Information in the Age of Ubiquitous Health

It is clear that online health information is key to closing these gaps and improving equity in access to information. When it comes to personal health information, however, there is a delicate balance between the desire for accessibility and the need for security. Breaches in security could result in the loss of social justice for entire communities. Delicate information that is not well protected can lead to financial threats, misinformation, fraudulent information, and even identify theft (Setness, 2003).

your health” is a very generic piece of information. It applies to you whether you are male or female, 6 or 60, African American or Caucasian. One could argue it should be sufficient to influence behaviours. Yet, the United Nations predicts over 7.1 million tonnes of tobacco will be consumed in 2010 (“FAO Newsroom,” 2004). Even with 6.7 billion people around the globe (“World Factbook,” 2009), the number is staggering. So, the question arises whether or not a simple generic piece of information is sufficient to influence the behaviour of individuals.

OPPORTUNITIES

Opportunities for the Clinic: Personal Health Programs

The ubiquity of information is coming to be synonymous with increased accessibility, transportability, and capacity for detail. All of these important elements of information translate to health care opportunities on several levels. In education, personal data is enabling information to be tailored to the individual who needs it, making it more meaningful, and thereby improving learning. On the clinical level, personalized health and wellness programs are on the rise, offering the public the chance to engage as managers of their own health matters. And finally, in research, the opportunities for data mining the wealth of surrounding information are beginning to be realized, not only by health care professionals, but by patients who are eager for in-depth learning about their condition.

Opportunities in Education: Tailored Education Health information is everywhere, but it can be delivered in a number of different ways, depending on who the target audience is and what the goals are for disseminating the information. The same basic kernel of information can evolve into several different key messages, when we are informed about who will be receiving it. For example, “don’t smoke because it is harmful to

64

As developments like EHRs and Web 2.0 technologies allow us to make information more transportable, shareable, and specific to a particular individual, health care can become more and more personalized. Personal information gathered from health care professionals, and from patients themselves, can be combined with standard care plans, transforming them into personal health programs. An example of this type of synergistic effort can be found in employee wellness programs in the United States, where a partnership between CapMed and Staywell Health Management offers opportunities for managing personal health. CapMed offers PHR solutions that allow patients to monitor and manage their health information, while Staywell is a health assessment and disease management organization, creating wellness programs for employers. By adding the power of the personal information PHRs can capture, wellness programs can become personalized to a particular individual. Members will be able to access individually relevant patient education resources and set up health alerts and reminders. Additional health management tools are also available, including a drug interaction checker, data importer to collect information from remote health care systems, and authorizations that allow family

Personal Health Information in the Age of Ubiquitous Health

members and caregivers to view all or part of the data a patient is storing (“CapMed and Staywell Health Managment Offer PHR/Wellness Program Presentations at AHIP 2008,” 2008).

Opportunities in Research: Data Mining It is difficult to consider the concept of ubiquitous health information in depth, without touching on the prospect of public health and surveillance. Public health is an issue that is currently booming, due to the events of the past decade. The anthrax threats that occurred after September 11th, and natural disasters such as hurricane Katrina and the Asian tsunami have raised awareness of the need for strengthened public health infrastructures (Kukafka, 2007). Spending has been increased for public health informatics in the United States, and research is showing that syndromes surveillance systems, which operate by collecting existing information by mining ubiquitous data are now capable of detecting some types of disease outbreaks quite rapidly. This trend of public health being increasingly tied to developments in informatics and the proliferation of health information and geographical information systems (GIS) (Tan, 2005) appears to be continuing as we move forward into the future. Prior to September 11th, 2001 a PubMed search for articles on “public health informatics” yielded only six results. In 2007, this same search resulted in over 600 articles (Kukafka, 2007). The public health benefits of data mining go beyond prevention. There is also the potential for identifying areas for improvement in quality of care. As an example, a New York state initiative in 1979 began collecting data on all hospitalized patients (Setness, 2003). Ten years later, a report card was published giving comparisons of cardiac bypass surgery deaths, by hospital and surgeon. The statistics were a shock for staff at hospitals that did not score well in the report. This discovery

led to the revamping of several cardiac surgery programs throughout the state. Improvements to the system are not the only type of public health benefit; there are benefits to be seen on the level of the individual as well. Patients being treated for a wide spectrum of conditions are finding benefits from this capacity for data mining UHI. PatientsLikeMe.com, is a website open to the public, that offers patients being treated for a variety of illnesses the chance to form communities. The home page offers the visitor three options that sum up the motivations for seeking health information online: Share your experiences, find patients like you, and learn from others. The site also hosts message boards with postings on information on treatments and symptoms, although it is not clearly documented where the information originates from. When they join, patients create their own PHR, by filling in information fields and posting their own data about their symptoms, treatments and outcomes. Of particular interest is the Research section of the site. The research team at PatientsLikeMe have harnessed the self-reported information from the PHR aspects of the site. The data in these personal profiles is compiled and displayed graphically to the community in the form of symptom and treatment reports. Members can view these reports to gain a broader understanding of what an entire population of patients like themselves are experiencing. The site allows members to discuss these issues in forums, contact each other through private messaging, and post comments on each other’s personal profiles (Frost, 2008). Having such robust information has proved very valuable to patients. Being able to locate others with very specific experiences allows members to target information searching and sharing. Users could find the right person to ask a specific question to, and know who was likely to benefit most from hearing their own experiences. It became easier for patients to form relationships that were stronger and more edifying; when they

65

Personal Health Information in the Age of Ubiquitous Health

were communicating with people they had much in common with (Frost, 2008).

FUTURE RESEARCH DIRECTIONS The field of UHI is burgeoning. It is difficult to know which areas will really blossom as integral parts of the health care system. Certainly, the domain of EHRs and PHRs seems to be firmly taking root with the emergence of industry giants such as Google and Microsoft staking a claim to this virtual space. In other areas of social networking and information exchange, organizations such as PatientsLikeMe are transforming the paradigm for health information sharing in ways that are still not evident. What is clear is that there are many new and intriguing opportunities in this area. There are opportunities to implement solutions and services that will contribute to the empowerment of patients and their families and provide them with new opportunities to be active participants in control of their health care. Leadership in the health system is required to affect the required changes. Finally, research is essential to produce new evidence and new knowledge so that effective change can be made. Research is required into many, aspects of ubiquitous health information and this research needs to build on the growing literature in health informatics and information and communication technologies.

Benefits Evaluation and Research on UHI Ubiquitous Heath Information poses several challenges to current models of the evaluation of health information. Many evaluation and research approaches of online resources and tools assume a relatively static, homogenous model. From this model, there are a series of assumptions made about the reliability and quality of the online resource. The very notion of ubiquitous information, however, suggests that this type of information is

66

not only everywhere, but it is, at the same time, highly dynamic in nature. The current evaluation techniques, for the most part, depend on measuring a constant or static condition and, therefore, in the realm of ubiquitous health information, many of the current evaluation endpoints for online information must themselves be challenged and examined. Evaluation and research models for online health information are based on the assumption that there is a “single” content provider. This may be one author, a group of authors, or an organization that can be identified as the source of content. Based on this assumption, a user of the information can employ a series of endpoints or parameters to assess the nature of the site. For example, the reliability of the site may be assessed based on the credibility of the source of the content. One may look at the credentials of the source, any conflicts of interest, publication record, and so on. However, if there are multiple authors who have no other association than being contributors to a collaborative site, such as a wiki or community of practice, it is much more difficult to assess the “global” reliability of the content on a site. Each “fragment” of authored content must be assessed individually. Making matters somewhat more complex, it is not always easy to immediately identify the “voice” of an individual. For example, wiki technologies are designed to integrate the work of many authors into a single message within a set framework. Although many wiki sites track individual contributions, these unique contributions or content fragments are not always readily apparent to the reader. In the case of a wiki, assessing the reliability of the information may be more about assessing the ability of a wiki community to self-correct and self-monitor the quality of the information on its site than assessing the credentials or credibility of individuals generating the content. Difficulties also arise when applying current assessment tools to the quality of an online resource. In assessing the usability of the site, there are often

Personal Health Information in the Age of Ubiquitous Health

several distinct technologies stitched together in what are often referred to as “mashups”. In assessing the usability of the platform or information communication technology, a new dimension of the evaluation may be required to assess not only the ease of use of the technology, but also the “integration” of information being aggregated from a number of sources. New models of evaluation are required that consider the exchange and flow of information. These new models need to expand the parameters for evaluation and assessment. For example, the strength of an online community can be assessed by looking at the number of users, frequency of posting and quantity of collaboration. A shift in focus is also required to investigate health systems improvements that are linked to UHI. These improvements may include continuity of care, the quality of information provided, fewer duplicated tests and better surveillance. In addition, psychosocial outcomes need to be included, such as reductions in distress, anxiety and depression. Levels of patient empowerment, activation and participation also must be considered to incorporate adherence to guidelines and care plans, as well as improved self-efficacy and self-management.

CONCLUSION Starting with general health information and the advent of the Internet, advances in technology and changes in the expectations of health care consumers have brought on a massive shift in the treatment of information in health care. We have moved from a focus on protecting and securing data to understanding how to provide equitable access to information in a safe and useful format. Health information is truly all around us, in many forms. The Age of the Internet has brought us dynamic social networking technologies and the spread of experiential health information, from

video stories to the real-time exchange of health experiences. Slowly, health care is beginning to enjoy the benefits of UHI, especially the potential to personalize care. Patients are seeing improvements in access to health information and communication with their providers, and beginning to take an active role in their care through easier prescription renewals, medication tracking, and the power of Internet technologies to empower those with chronic illness to self-manage their conditions. At the organizational level, we are starting to see increased patient satisfaction, better continuity of care, reductions in cost and improved standardization of care. However, it is also important to properly manage the risks and opportunities that come with ubiquitous health information. We must remember the challenges, so that we may create policies that effectively manage the risks of privacy and security issues, and strive to overcome the barriers of change management and the lack of basic infrastructure in many areas. If we are to take full advantage of our opportunities and make sense of the “everywhere, all the time” presence of health information, researchers must create a roadmap for utilizing emerging technologies. This new map must plot a pathway that will ensure that these new, disruptive and transformative technologies will result in more effective and efficient health systems, in which patient outcomes and quality of life are dramatically improved.

REFERENCES Anderson, G., & Knickman, J. (2001). Changing the chronic care system to meet people’s needs. Health Affairs, 20(6), 146–160. doi:10.1377/ hlthaff.20.6.146

67

Personal Health Information in the Age of Ubiquitous Health

Atack, L., Luke, R., & Chien, E. (2008). Evaluation of Patient Satisfaction With Tailored Online Patient Education Information. Computers, Informatics, Nursing, 26(5), 258–264. doi:10.1097/01. NCN.0000304838.52207.90

Branko, C., Lovell, N., & Basilakis, J. (2003). Using information technology to improve the management of chronic disease. The Medical Journal of Australia, 179(5), 242–246.

Bader, J., & Strickman-Stein, N. (2003). Evaluation of new multimedia formats for cancer communications. Journal of Medical Internet Research, 5(5).

Brodie, M., Flournoy, R., Altman, D., Blendon, R., Benson, J., & Rosenbaum, M. (2000). Health information, the Internet, and the digital divide. Health Affairs, 19(6), 255–265. doi:10.1377/ hlthaff.19.6.255

Ball, M. J., Costin, M. Y., & Lehmann, C. (2008). The personal health record: consumers banking on their health. Studies in Health Technology and Informatics, 134, 35–46.

Cafazzo, J. A. K.J, L., Easty, A., Rossos, P. G., & Chan, C. T. (2008). Bridging the Self-Care Deficit Gap: Remote Patient Monitoring and the Hospitalat-Home. Paper presented at the eHealth 2008.

Ball, M. J., Smith, C., & Bakalar, R. S. (2007). Personal health records: empowering consumers. Journal of Healthcare Information Management, 21(1), 76–86.

CapMed and Staywell Health Managment Offer PHR/Wellness Program Presentations at AHIP 2008. (2008). Retrieved March 10, 2009, from http://www.bioimaging.com/assets/pdf/ press_release/2008/2008-CapMed-StaywellCollaborate.pdf

Beach, R. D., Kuster, F. V., & Moussy, F. (1999). Subminiature implantable potentiostat and modified commercial telemetry device for remote glucose monitoring. Instrumentation and measurement. IEEE Transactions on, 48(6), 1239–1245. Beardwood, J. P., & Kerr, J. A. (2005). Coming soon to a health sector near you: an advance look at the new Ontario Personal Health Information Protection Act (PHIPA): part II. Healthcare Quarterly (Toronto, Ont.), 8(1), 76–83. Berler, A., Pavlopoulos, S., & Koutsouris, D. (2004). Design of an interoperability framework in a regional healthcare system. Conference Proceedings. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Conference, 4, 3093–3096. Blechner, B., & Butera, A. (2002). Health Insurance Portability and Accountability Act of 1996 (HIPAA): a provider’s overview of new privacy regulations. Connecticut Medicine, 66(2), 91–95.

68

Cavoukian, A. (2004). A Guide to the Personal Health Information Act. Charles, C., Gafni, A., & Whelan, T. (1997). Shared decision-making in the medical encounter: what does it mean? (or it takes at least two to tango). Social Science & Medicine, 44(5), 681–692. doi:10.1016/S0277-9536(96)00221-3 Chasteen, J. E., Murphy, G., Forrey, A., & Heid, D. (2003). The Health Insurance Portability and Accountability Act: practice of dentistry in the United States: privacy and confidentiality. [Electronic Resource]. The Journal of Contemporary Dental Practice, 4(1), 59–70. Cresci, M., Morrell, R., & Echt, K. (2004). The Convergence of Health Promotion and the Internet. In Nelson, R., & Ball, M. (Eds.), Consumer Informatics: Applications and Strategies in Cyber Health Care. New York: Springer.

Personal Health Information in the Age of Ubiquitous Health

Deber, R. B., Kraetschmer, N., Urowitz, S., & Sharpe, N. (2005). Patient, consumer, client, or customer: what do people want to be called? Health Expectations, 8(4), 345–351. doi:10.1111/j.13697625.2005.00352.x Deber, R. B., Kraetschmer, N., Urowitz, S., & Sharpe, N. (2007). Do people want to be autonomous patients? Preferred roles in treatment decision-making in several patient populations. Health Expectations, 10(3), 248–258. doi:10.1111/ j.1369-7625.2007.00441.x Deshpande, A., & Jadad, A. (2006). Web 2.0: Could it help move the health system into the 21st century? Journal of Men’s Health & Gender, 3(4), 332–336. doi:10.1016/j.jmhg.2006.09.004 Dietzel, G. T. (2002). Challenges of telematics and telemedicine for health care: from e-Europe 2002 to the electronic health passport. International Journal of Computerized Dentistry, 5(2-3), 107–114. Eysenbach, G. (2008). Medicine 2.0: Social Networking, Collaboration, Participation, Apomediation, and Openness. Journal of Medical Internet Research, 10(3). doi:10.2196/jmir.1030 Field, M. J., & Grigsby, J. (2002). Telemedicine and remote patient monitoring. Journal of the American Medical Association, 288(4), 423–425. doi:10.1001/jama.288.4.423 France, F. H., & Gaunt, P. N. (1994). The need for security--a clinical view. International Journal of Bio-Medical Computing, 35, 189–194. Frost, J., & Massagli, M. P. (2008). Social uses of personal health information within PatientsLikeMe, an online patient community: what can happen when patients have access to one another’s data. Journal of Medical Internet Research, 10(2), e15. doi:10.2196/jmir.1053

Giustini, D. (2006). Scrooge and intellectual property rights. British Medical Journal, 333, 179–1284. Gloth, M., Coleman, E., Philips, S., & Zorowitz, R. (2005). Using Electronic Health Records to Improved Care: Will “Hight Tech” Allow a Return to “High Touch” Medicine? Journal of the American Medical Directors Association, 270–275. doi:10.1016/j.jamda.2005.05.003 Harman, L. B. (2005). HIPAA: a few years later. Online Journal of Issues in Nursing, 10(2), 31. Hassol, A., Walker, J. M., Kidder, D., Rokita, K., Young, D., & Pierdon, S. (2004). Patient experiences and attitudes about access to a patient electronic health care record and linked web messaging. Journal of the American Medical Informatics Association, 11(6), 505–513. doi:10.1197/ jamia.M1593 Hewitt, M., & Ganz, P. (Eds.). (2006). From Cancer Patient to Cancer Survivor: Lost in Transition. Washington, DC: The National Academies Press. Ishikawa, K., Konishi, N., Tsukuma, H., Tsuru, S., Kawamura, A., & Iwata, N. (2004). A clinical management system for patient participatory health care support. Assuring the patients’ rights and confirming operation of clinical treatment and hospital administration. International Journal of Medical Informatics, 73(3), 243–249. doi:10.1016/j.ijmedinf.2003.11.021 Ishikawa, K., Ohmichi, H., Umeatso, Y., Terasaki, H., Tsukuma, H., & Iwata, N. (2007). The guideline of the personal haelth data structure to secure safety healthcare: The balance between use and protection to satisfy the patients’ needs. International Journal of Medical Informatics, 76, 412–418. doi:10.1016/j.ijmedinf.2006.09.005 Johnson, G., & Ambrose, G. (2006). Neo-Tribes: The Power and Potential of Online Communities in Health Care. Communications of the ACM, 49(1), 107–113. doi:10.1145/1107458.1107463

69

Personal Health Information in the Age of Ubiquitous Health

Jones, T. M. (2003). Patient participation in EHR benefits. Health Management Technology, 24(10). Kittler, A. F., Carlson, G. L., Harris, C., Lippincott, M., Pizziferri, L., & Volk, L. A. (2004). Primary care physician attitudes towards using a secure web-based portal designed to facilitate electronic communication with patients. Informatics in Primary Care, 12(3), 129–138. Kluge, E. H. (1993). Advanced patient records: some ethical and legal considerations touching medical information space [see comment]. Methods of Information in Medicine, 32(2), 95–103. Kukafka, R. (2007). Guest Editorial: Public Health Informatics. Journal of Biomedical Informatics, 40, 365–369. doi:10.1016/j.jbi.2007.07.005 Leonard, K., Wiljer, D., & Urowitz, S. (2008). Yes Virginia, There Are System Benefits to be Gained from Providing Patients Access to Their Own Health Information. Healthcare Quarterly (Toronto, Ont.), 11(4), 66–70. Leonard, K., Wiljer, D., & Urowitz, S. (2008). Yes, Virginia, There are System Benefits to Be Gained From Providing Patients Access to Their Own Health Information. Healthcare Quarterly (Toronto, Ont.), 11(4), 66–77.

Lorig, K. R., Sobel, D. S., Ritter, P. L., Laurent, D., & Hobbs, M. (2001). Effect of a self-management program on patients with chronic disease. Effective Clinical Practice, 4(6), 256–262. Markle. (2003). Connecting for Health: A PublicPrivate Collaborative. http://www.connectingforhealth.org/resources/final_phwg_report1.pdf. Accessed: 2008-09-22. (Archived by WebCite® at http://www.webcitation.org/5b24kHscL). Mathews, S. (1998). Protection of personal data--the European view. Journal of American Health Information Management Association, 69(3), 42–44. Mitchell, P. (1998). France gets smart with health a la carte. Lancet, 351(9104), 7. doi:10.1016/ S0140-6736(05)78518-4 Newsroom, F. A. O. (2004). Retrieved March 10, 2009, from http://www.fao.org/english/newsroom/news/2003/26919-en.html O’Connor, A., Stacey, D., Entwistle, V., LlewellynThomas, H., Rovner, D., Holmes-Rovner, M., et al. (2002). Decision aids for people facing health treatment or screening decisions. (Publication no. 10.1002/14651858.CD001431).

Leonard, K. J. W., D., & Casselman, M. (2008a). An Innovative Information Paradigm for Consumers with Chronic Conditions: The value proposition. The Journal on Information Technology in Healthcare.

O’Grady, L., Witteman, H., Bender, J. L., Urowitz, S., Wiljer, D., & Jadad, A. R. (2009). Measuring the Impact of a Moving Target: Towards a Dynamic Framework for Evaluating Collaborative Adaptive Interactive Technologies. Journal of Medical Internet Research, 11(2), e20. doi:10.2196/jmir.1058

Leonard, K. J., Casselman, M., & Wiljer, D. (2008). Who will demand access to their personal health record? A focus on the users of health services and what they want. Healthcare Quarterly (Toronto, Ont.), 11(1), 92–96.

Orchard, M., Green, E., Sullivan, T., Greenberg, A., & Mai, V. (2008). Chronic disease prevention and management: implications for health human resources in 2020. Healthcare Quarterly (Toronto, Ont.), 11(1), 38–43.

Leonard, K. J., & Wiljer, D. (2007). Patients Are Destined to Manage Their Care. Healthcare Quarterly (Toronto, Ont.), 10(3), 76–78.

Pallapa, G., & Das, S. (2007). Resource Discovery in Ubiquitous Health Care. Paper presented at the Advanced Information Networking and Applications Workshops

70

Personal Health Information in the Age of Ubiquitous Health

Paterson, B. L. (2001). The shifting perspectives model of chronic illness. Journal of Nursing Scholarship, 33(1), 21–26. doi:10.1111/j.15475069.2001.00021.x Preece, J. (2000). Online Communities: Desigining Usability, Supporting Sociability. Chichester, England: Wiley & Sons Ltd. Pyper, C., Amery, J., Watson, M., & Crook, C. (2004). Patients’ experiences when accessing their on-line electronic patient records in primary care. The British Journal of General Practice, 54(498), 38–43. Pyper, C., Amery, J., Watson, M., Crook, C., & Thomas, B. (2002). Patients’ access to their online electronic health records. Journal of Telemedicine and Telecare, 2, 103–105. doi:10.1258/135763302320302244 Ralston, J. D., Carrell, D., Reid, R., Anderson, M., Moran, M., & Hereford, J. (2007). Patient Web Services Integrated with a Shared Medical Record: Patient Use and Satisfaction. Journal of the American Medical Informatics Association, 14(6), 798–806. doi:10.1197/jamia.M2302 Rutten, L., Arorab, N., Bakosc, A., Azizb, N., & Rowland, J. (2005). Information needs and sources of information among cancer patients: a systematic review of research (1980-2003). Patient Education and Counseling, 57(3), 250–261. doi:10.1016/j. pec.2004.06.006 Schoenburg, R. (2005). Security of Healthcare Information Systems. In D. Lewis, G. Eysenbach, R. Kukafka, P. Z. Stavri & H. Jimison (Eds.), Consumer Health Informatics. New York: Springer Science+Business Media Inc. Setness, P. (2003). When privacy and the public good collide. Does the collection of health data for research harm individual patients? Postgraduate Medicine, 113(5), 15. doi:10.3810/ pgm.2003.05.1418

Tan, J. (2005). The Next Health Care Frontier. San Francisco: Jossey-Bass. Tang, P. C., Ash, J. S., Bates, D. W., Overhage, J. M., & Sands, D. Z. (2006). Personal health records: definitions, benefits, and strategies for overcoming barriers to adoption. Journal of the American Medical Informatics Association, 13(2), 121–126. doi:10.1197/jamia.M2025 Urowitz, S., & Deber, D. (2008). How Consumerist Do People Want to Be? Preferred Role in Decision-Making of Individuals with HIV/AIDS. Health Policy (Amsterdam), 3(3). Urowitz, S., Wiljer, D., Apatu, E., Eysenbach, G., DeLenardo, C., Harth, T., et al. (2008). Is Canada ready for patient accessible electronic health records? A national scan. Journal, 8. Retrieved from http://www.pubmedcentral.nih.gov/articlerender. fcgi?tool=pubmed&pubmedid=18652695 Wallerstein, N. (2006). What is the evidence on effectiveness of empowerment to improve health?Albuquerque: Europe’s Health Evidence Network. Wiecha, J., & Pollard, T. (2004). The Interdisciplinary eHealth Team: Chronic Care for the Future. Journal of Medical Internet Research, 6(3). doi:10.2196/jmir.6.3.e22 Wiljer, D., Bogomilsky, S., Catton, P., Murray, C., Stewart, J., & Minden, M. (2006). Getting results for hematology patients through access to the electronic health record. Canadian Oncology Nursing Journal, 16(3), 154–164. World Factbook. (2009, March 5, 2009). Retrieved March 10, 2009, from https://www.cia.gov/library/ publications/the-world-factbook/print/xx.html Wuerdeman, L., Volk, L., Pizziferri, L., Ruslana, T., Harris, C., Feygin, R., et al. (2005). How Accurate is Information that Patients Contribute to their Electronic Health Record? Paper presented at the AMIA 2005.

71

Personal Health Information in the Age of Ubiquitous Health

Wyatt, J., & Sullivan, F. (2005). eHealth and the future: promise or peril? Journal. Retrieved from http://www.bmj.com/cgi/content/ full/331/7529/1391

ADDITIONAL READING Andersen, J., & Aydin, C. (Eds.). (2005). Evaluating the Organizational Impact of Healthcare Information Systems. New York: Springer Science+Business Media Inc. Brender, J. (2006). Evaluation Methods for Health Informatics. San Diego: Elsevier. Bruslilovsky, P., Kobsa, A., & Nejdl, W. (Eds.). (2007). The Adaptive Web: Methods and Strategies of Web Personalization. Pittsburgh: Springer. Committee on Quality of Health Care in America. (2001). Crossing the Quality Chasm: A New Health System for the 21st Century. Washington, DC: National Academy Press. Europe, W. H. O. (2006). What is the evidence on effectiveness of empowerment to improve health? Retreived March 13, 2009 from http://www.euro. who.int/Document/E88086.pdf Gerteis, M., Edgman-Levitan, S., & Daley, J. Delblanco, T. (Eds.) (1993). Through the Patient’s Eyes. San Francisco: Jossey-Bass. Johnston Roberts, K. (2001). Patient empowerment in the United States: a critical commentary. Health Expectations, 2(2), 82–92. doi:10.1046/ j.1369-6513.1999.00048.x Kemper, D., & Mettler, M. (2002). Information Therapy: Prescribed Information as a Reimbursable Medical Service. Boise, ID: Healthwise.

Kreuter, M., Farrell, D., Olevitch, L., & Brennan, L. (Eds.). (2000). Tailoring Health Messages: Customizing Communication with Computer Technology. New Jersey: Lawrence Earlbaum Associates Inc. Leonard, K. J., Wiljer, D., & Casselman, M. (2008). An Innovative Information Paradigm for Consumers with Chronic Conditions: The value proposition. The Journal on Information Technology in Healthcare. Lewis, D., Eysenbach, G., Kukafka, R., Stavri, P. Z., & Jimison, H. (Eds.). (2005). Consumer Health Informatics. New York: Springer Science+Business Media Inc. Picker, N. C. R. Eight Dimensions of PatientCentred Care. Retreived March 13, 2009 from http://www.nrcpicker.com/Measurement/Understanding%20PCC/Pages/DimensionsofPatientCenteredCare.aspx Preece, J. (2000). Online Communities: Desigining Usability, Supporting Sociability. Chichester, England: Wiley & Sons Ltd. Sands, D. Z. (2008). Failed Connections: Why Connecting Humans Is as Important as Connecting Computers. Medscape Journal of Medicine, 10(11), 262. Sands, D. Z. (2008). ePatients: Engaging Patients in Their Own Care. Medscape Journal of Medicine, 10(1), 19. Tan, J. (2005). E-Health Care Information Systems: An Introduction for Students and Professionals. San Francisco: Jossey-Bass. Wagner, E. H. (¾¾¾). Chronic disease management: What will it take to improve care for chronic illness? Effective Clinical Practice, 1, 2–4.

This work was previously published in Ubiquitous Health and Medical Informatics: The Ubiquity 2.0 Trend and Beyond, edited by Sabah Mohammed and Jinan Fiaidhi, pp. 166-189, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

72

73

Chapter 1.5

E-Health as the Realm of Healthcare Quality: A Mental Image of the Future

Anastasius Moumtzoglou Hellenic Society for Quality & Safety in Healthcare, European Society for Quality in Healthcare Executive Board Member, Greece

ABSTRACT E-health has widely revolutionized medicine, creating subspecialties that include medical image technology, computer aided surgery, and minimal invasive interventions. New diagnostic approaches, treatment, prevention of diseases, and rehabilitation seem to speed up the continual pattern of innovation, clinical implementation and evaluation up to industrial commercializa-

tion. The advancement of e-health in healthcare derives large quality and patient safety benefits. Advances in genomics, proteomics, and pharmaceuticals introduce new methods for unraveling the complex biochemical processes inside cells. Data mining detects patterns in data samples, and molecular imaging unites molecular biology and in vivo imaging. At the same time, the field of microminiaturization enables biotechnologists to start packing their bulky sensing tools and medical simulation bridges the learning divide

DOI: 10.4018/978-1-60960-561-2.ch105

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

E-Health as the Realm of Healthcare Quality

by representing certain key characteristics of a physical system.

INTRODUCTION There is a worldwide increase in the use of health information technology (IT), which holds promise of health system breakthroughs. Emerging information and communication technologies promise groundbreaking solutions for healthcare problems. Moreover, a global e-health consensus framework is beginning to take shape in IT discourse, which includes stakeholders, policy, funding, coordination, standards and interoperability (Gerber, 2009). However, in an unprecedented technological innovation, many aspects of health care systems require careful consideration due to error and inefficiency although e-health bridges clinical and nonclinical sectors. The endorsement of information technology in the health sector is spreading slowly. Few companies focus on population-oriented e-health tools partly because of perceptions about the viability and extent of the market segment. Moreover, developers of e-health resources are a highly diverse group with differing skills and resources while a common problem for developers is finding the balance between investment and outcome. According to recent surveys, one of the most severe restraining factors for the proliferation of e-health is the lack of security measures. Therefore, a large number of individuals are not willing to engage in e-health (Katsikas et al., 2008). E-health presents risks to patient health data that involve not only technology and appropriate protocols but also laws, regulations and professional security cultures. Furthermore, breaches of network security and international viruses have elevated the public awareness of online data and computer security. Although the overwhelming majority of security breaches do not directly involve health-related data, the notion that online data are exposed to security threats is widespread. Moreover, as we

74

understand the clinical implications of the genetic components of disease, we expect a remarkable increase in the genetic information of clinical records. As a result, there is likely to be considerable pressure in favor of specific laws protecting genetic privacy (Magnusson, 2002). Therefore, secure e-health requires not only national standardization of professional training and protocols but also global interoperability of regulations and laws. Professional health information organizations must take the lead in professional certification, security protocols and applicable codes of ethics on a global basis ((Kluge, 2007; Moor & Claerhoutb, 2004). On the other hand, clinicians have moved toward the Internet within the last few years while purchasers seek higher quality and lower costs. The Internet offers an unprecedented opportunity to integrate various health-related sectors while some Internet-related trends and technologies will have a substantial impact on the design, content, functionality, dissemination, and use of future e-health tools. Moreover, quality assurance and improvement are key issues for the e-health sector while strategic planning could provide an insightful view of the impacts (Asoh et al., 2008). However, the accessibility and confidentiality of electronic resources does not guarantee quality access (West and E. A. Miller, 2006) and the current quality assurance strategies do not address the dynamic nature of e-health technologies. Furthermore, the contribution of various socioeconomic factors to ‘the digital divide’, which refers to the difference in computer and Internet access between population groups segmented by various parameters, is controversial. However, recent data suggests that the digital divide may be closing in some aspects, due to access to PCs, and the Internet. Access, however, is only one facet of the digital divide, as health literacy (Bass, 2005), and relevant content are also key elements. Moreover, there are overlapping and gaps in e-health content due to the uncoordinated and essentially independent efforts. Current market

E-Health as the Realm of Healthcare Quality

forces are driving e-health development in clinical care support, and health care transactions, but they do not provide population health-related functions. Therefore, increased information exchange and collaboration among developers may result in efficient use of resources. The challenge is to foster collaborative e-health development in the context of fair competition. Greater collaboration presents new communication challenges, which include a standardized communication and information flow (Lorence & R. Churchill, 2008), which will enhance: • • • •

social support cognitive functioning clinical decision-making cost-containment

Many observers believe that a picture of interoperable clinical, laboratory, and public health information systems, will provide unprecedented opportunities for improving individual and community health care (Godman, 2008; Lorence & Sivaramakrishnan, 2008; Toledo et al., 2006). Interoperable electronic medical records (EMRs) contain the potential to improve efficiency and reduce cost (James, 2005). Moreover, semantic interoperability is essential for sound clinical care (Goodenough, 2009) although electronic prescribing does not increase steadily (Friedman et al., 2009). The lack of integration in health care also carries over into the online world. Therefore, universal data exchange requires a translating software and the development of data exchange standards (Friedman et al., 2009). There is also need to integrate the various features and functions of e-health tools to provide a seamless continuum of care, which might include: • • • •

health data processing of transactions electronic health records clinical health information systems

• •

disease management programs behavior modification and promotion

health

Finally, as the complexity and amount of genetic information expand, new e-health tools, which support clinicians and consumers decisionmaking in genetics will be in considerable demand. Moreover, once nanotechnology applications become available, we have to establish nanoethics as a new sub-discipline of applied ethics (Godman, 2008). Overall, although we have to understand care in a renewed dimension (Nord, 2007), we might consider e-health as a method to improve the health status of the population. The promise of applying emerging e-health technologies to improve health care is substantial. However, it is crucial to build partnerships among healthcare providers, organizations, and associations to ensure its continued development (Harrison & Lee, 2006). Thus, the perspective of the chapter is to examine the prospects of e-health to become the realm of healthcare quality.

BACKGROUND The term e-health has evolved through prolonged periods of time, which include the era of discovery (1989-1999), acceptance (1999-2009), and deployment (2009-). Each period has distinctive features, and critical applications. For example, in the age of discovery, we denounced traditional approaches to realize during the era of acceptance that we need a collaboration model. Key application areas of this period of time include electronic medical records (including patient records, clinical management systems, digital imaging & archiving systems, e-prescribing, e-booking), telemedicine and telecare services, health information networks, decision support tools, Internet-based technologies and services. E-health also covers virtual reality, robotics, multimedia, digital imaging, computer

75

E-Health as the Realm of Healthcare Quality

assisted surgery, wearable and portable monitoring systems, health portals. Finally, the distinctive features of the era of deployment might involve community care, evidence based medicine, collaborative care, and self-management. Furthermore, centrifugal or centripetal mobility will contribute to a new service integrating e-health and quality. However, the term e-health is a common neologism, which lacks precise definition (Oh et al., 2005). The European Commission defines e-health as ‘the use of modern information and communication technologies to meet needs of citizens, patients, healthcare professionals, healthcare providers, as well as policy makers’ while the World Health Organization offers a more precise definition. Specifically, e-health is ‘the cost-effective and secure use of information and communications technologies in support of health and health-related fields, including health-care services, health surveillance, health literature, and health education, knowledge and research’. Eysenbach, in the most frequently cited definition, defines e-health as ‘an emerging field in the intersection of medical informatics, public health and business, referring to health services and information delivered or enhanced through the Internet and related technologies. In a broader sense, the term characterizes not only a technical development, but also a state-of-mind, a way of thinking, an attitude, and a commitment for networked, global thinking, to improve health care locally, regionally, and worldwide by using information and communication technology. Finally, Pagliari et al., (2005) in a highly detailed analysis provide a definition, which covers human and organizational factors. E-health is not the only sector, which lacks a clear definition. A literature review reveals a multitude of definitions of quality of care (Banta, 2001). The World Health Organization has defined the quality of health systems as ‘the level of attainment of a health system’s intrinsic goals for health improvement and responsiveness to the legitimate expectations of the population’

76

(WHO, 2000). W. Edward Deming argued that ‘a product or service possesses quality if it helps somebody and enjoys a good and sustainable market’, Juran (1974) stated that ‘quality is fitness for use’, and Crosby (1979) ‘quality means conformance to requirements’. Still, the most widely cited definition of healthcare quality was formulated by the Institute of Medicine (IOM) in 1990. According to the IOM, quality consists of the ‘degree to which health services for individuals and populations increase the likelihood of desired health outcomes and are consistent with current professional knowledge’. Overall, the meaning of quality is complex, elusive, and uncertain (Burhans, 2007), and its meaning is an optimal balance between possibilities and the framework of norms and values (Harteloh, 2004). Nevertheless, the concept of healthcare quality involves the standardization and national endorsement of performance measures, the evaluation of outcomes, reporting for accountability, and technological innovation (Smith, 2007). Furthermore, we might distinguish three levels of quality, which include conformance quality, requirements quality, and quality of kind. Even so, the present and future healthcare systems demonstrate a quality chasm, formed by a large number of different factors. Medical science and advanced technology have advanced at an unprecedented rate but have also contributed to a growing complexity. Faced with such rapid changes, the healthcare system has fallen considerably short in its ability to refine theoretical knowledge into professional practice and to use modern technology. The healthcare needs have changed quite dramatically as well. Individuals are living longer, due to fundamental advances in medical science. Subsequently, the aging population translates into an exponential growth in the prevalence of chronic conditions. In addition, today’s complex health system overly deals with acute, episodic care needs and cannot consistently deliver today’s science and technology. It is less prepared to respond to the tremendous advances

E-Health as the Realm of Healthcare Quality

that surely will appear in the near future. Healthcare delivery cannot face the various challenges and is overly complicated and uncoordinated. Therefore, complex processes waste resources and lead to excessive loss of useful information. Eventually, healthcare organizations, hospitals, and physicians operate without the added value of information. As a result, state-of-the-art care requires a fundamental and extensive redesign. The health system should avoid injuries, provide responsive services based on scientific knowledge, reduce waits and delays, avoid waste, and eliminate differences in quality. Overall, according to the Institute of Medicine ‘Crossing the Quality Chasm: A New Health System for the 21st Century’ report, the health system should be safe, effective, patientcentered, timely, efficient, and equitable (IOM, 2001). To achieve these improvement aims, we should adopt ten simple statements (IOM, 2001): • • • • • • • • • •

patients should receive care or access to care whenever they need it the entire system should meet the most frequent types of needs and preferences empower the patient allow information flow freely and share knowledge decision-making is evidence-based patient safety is a system property transparency is necessary anticipate needs decrease waste enhance cooperation

Furthermore, redesigning the healthcare delivery system involves changing the structure and administrative processes of healthcare organizations. Such changes need to appear in four key areas (IOM, 2001): • •

application of evidence to healthcare delivery information technology

• •

alignment of payment policies with quality improvement preparation of the future workforce

Finally, we should not underestimate the importance of adequately preparing the future workforce to undergo the transition into a new healthcare system. There are three contrasting approaches, which we might use (IOM, 2001): • • •

redesign the educational approach (Hasman et al., 2006) change the way we accredit or regulate health professionals define and ensure accountability among health professionals and professional organizations

Conclusively, the Institute of Medicine’s ‘Crossing the Quality Chasm: A New Health System for the 21st Century’ report, implies: •

•

•

•

• •

to improve physician-patient interaction beyond encounter-oriented care (Leong et al., 2005) to introduce a partnership approach to clinical decision-making (Coulter & Ellins, 2006) to improve health literacy and encourage full disclosure of adverse events (Gallagher et al., 2007) to revolutionize patient safety by providing clinicians with clinical decision support and by involving patients (Coulter & Ellins, 2006) to raise public accountability by measuring the quality of care at the provider level to improve patient care on the basis of ongoing evaluation, lifetime learning, systems-based practice and interpersonal skills of clinicians

Although groundbreaking, IOM’s approach raises several limitations:

77

E-Health as the Realm of Healthcare Quality

•

•

• •

the evidence on health literacy demonstrates that there are substantial gaps in what we know about how to raise its standards (Coulter & Ellins, 2006) the implementation of innovations, which improve clinical decision-making and promote greater patient involvement, has yet to occur (Coulter & Ellins, 2006; Billings 2004; Graham et al., 2003) safety improvement, through patient involvement, should be enhanced patient feedback, provider choice, complaints, and advocacy systems should be carefully assessed

E-health holds enormous potential for transforming the quality aspect of healthcare by supporting the delivery of care, improving transparency and accountability, aiding evidence-based practice and error reduction, improving diagnostic accuracy and treatment appropriateness, facilitating patient empowerment, and improving costefficiency.

THE QUALITY PERSPECTIVE OF E-HEALTH E-health has its roots in physics, engineering, informatics, mathematics and chemistry and links with biosciences, especially biomedical physics, biomedical engineering, biomedical computing and medicine. Therefore, the improvement in science and technology accelerate its tremendous growth the last few decades. Furthermore, e-health has widely revolutionized medicine, creating subspecialties that include medical image technology, computer aided surgery, and minimal invasive interventions. New diagnostic approaches, treatment, prevention of diseases, and rehabilitation seem to speed up the continual pattern of innovation, clinical implementation and evaluation up to industrial commercialization. Some of the current technologies are the electronic patient

78

record, a routine for a few countries, the patient data card, health professional card, e-prescription, e-reporting, networking and tele-consulting. BioMEMOS, imaging technology, minimally invasive surgery, and computer-assisted diagnosis, therapy and treatment monitoring, e-health/telemedicine/ networking, and medical engineering for regenerative medicine are the fields of e-health technologies, which have emerged in recent years. For example, tissue engineering combines specific medical disciplines with cell and molecular biology, material sciences, physics and engineering aiming at developing methods for regenerating, repairing or replacing human tissue. Specifically, it uses in situ stem cell control focusing specifically on stem cells of the central nervous system for neurogeneration (Nusslin, 2006). Finally, beyond the boundaries of the medical field, at least in the highly industrialized countries, e-health has a substantial impact on the economy, the whole society and the ethical system. The advancement of e-health in healthcare derives large quality and patient safety benefits. Health information technology holds enormous potential to improve outcomes, and efficiency within the health care system. Electronic health records, decision support systems, computerized physician order entry systems, computerized adverse events systems, automatic alerts, national or regional incident or event reporting systems improve administrative efficiency, adherence to guideline-based care, and reduce medication errors. Electronic health records, which store information electronically, provide the opportunity to find patients with certain conditions, and look for specific issues (Honigman et al., 2001a; Honigman et al., 2001b).The IOM (2001) argued that an electronic based patient record system would be the single action that would most improve patient safety. Coiera et al., (2006) argued that the use of decision support systems, comprehensive solutions often incorporated in a variety of e-health applications, can improve patient outcomes and make clinical services more effective.

E-Health as the Realm of Healthcare Quality

Tierney et al (2003) found that the intervention had no effect on physicians’ adherence to care suggestions while Kawamoto et al (2005), in a meta-analysis of seventy studies, concluded that decision support systems significantly improve clinical practice. Computerized physician order entry systems, a process whereby physicians file electronically instructions to individuals responsible for patient care, improve the quality of care by increasing clinician compliance. It is an outstanding application (Sittig & Stead, 1994), which reduces preventable adverse drug events (Kaushal & Bates, 2003). Computerized adverse event systems have shown a notable increase in the number of reported adverse drug events, and automatic alerts reduce the time until treatment of patients with critical laboratory tests (2001). Many studies have shown that prevention guidelines and the reminder computerization improve adherence (Balas et al., 2000) while reminders are noteworthy in the care of chronic conditions, which constitute a large quotient of expenditures (Lobach & Hammond, 1994). Furthermore, national or regional incident or event reporting systems, which collect data from local sources, lead to remarkable changes at the community and national level (Runciman, 2002). Finally, a particularly compelling reason for promoting the adoption of e-health is its potential to improve the processes of health care. E-Health improves the likelihood that the processes will be successful; strengthens the distribution of evidence-based decision support to providers, narrows the gaps between evidence and practice, and might lead to exceptional savings in administrative costs. On the other hand, electronic clinical knowledge support systems have decreased barriers to answering clinical questions (Bonis et al., 2008), and organizational learning at the system level fosters improvement (Rivard et al., 2006). However, e-health is not just a technology but a complex technological and relational process (Metaxiotis et al., 2004). Therefore, the Internet, a platform

available to remote areas, constitutes a radical change, which empowers patients. Patient-centered care is an emerging approach, which the Picker Institute defines as informing and involving patients, eliciting and respecting their preferences; responding quickly, effectively and safely to patients’ needs and wishes; ensuring that patients are treated in an ennobled and caring manner; delivering well coordinated and integrated care. Patient empowerment, which means that patients play a more active role in partnership with health professionals, and relates to strategies for education and e-health information (Pallesen et al., 2006), is increasingly being seen as a essential component of a modern patient-centered health system. Kilbridge (2002) suggested that technologies, which allow access to general and specific health care information, technologies capable to conduct data entry and tracking of personal self-management, empower patients. These technologies included Personal Health Records, Patient Access to Hospital Information Systems, Patient Access to General Health Information, Electronic Medical Records (EMRs), Pre-Visit Intake, Inter-Hospital Data Sharing, Information for Physicians to Manage Patient Populations, Patient-Physician Electronic Messaging, and Patient Access to Tailored Medical Information, Online Data Entry and Tracking, Online Scheduling, Computer-Assisted Telephone Triage and Assistance, and Online Access to Provider Performance Data. Moreover, handheld devices increasingly allow expansion of desktop systems and capture vital signs or administer medications while much of the quality improvement revolves around refinement of processes. It is worthwhile mentioning that telemedicine, a branch of e-health that uses communication networks and it is deployed to overcome uneven distribution and lack of infrastructural and human resources, provides a category by itself (Sood et al., 2007). Finally, the sciences of chaos and complexity are generating considerable interest within the domains of healthcare quality and patient safety

79

E-Health as the Realm of Healthcare Quality

(Benson, 2005), and clinical narratives has the potential to improve healthcare quality and safety (Brown et al., 2008; Baldwin, 2008). Still, there is a lack of interoperability, which makes the provision of clinical decision support and the extracting of pertinent information vastly more complicated (Schiff & Rucker, 1998; Schiff et al., 2008), and risks to patient health data (Kluge, 2007).

reliability and quality of health data and tort-based liability while recommendations for legal reform include (Hodge et al., 1999):

Issues, Controversies, Problems

Technological barriers to e-health stem from the evolutionary nature of these systems but also rapid obsolescence, and the lack of standards or criteria for interoperability. Without the technical specifications that enable interoperability, data exchange between providers who use different e-health systems is severely limited. Moreover, security and confidentiality in information technology represent a serious matter (Anderson, 2007). Finally, implementation and dissemination issues are decipherable. The effects of e-health tools on patient behavior, the patient-clinician relationship, the legal and ethical implications of using health information technologies and clinical decision support systems are unclear. Furthermore, potential health inequalities resulting from the digital divide have to restrain within bounds. Overall, key questions include clinical decision support, refinement of guidelines for local implementation, implementation and dissemination of clinical information systems, patient involvement and the role of the Internet. On the other hand, the variation in the quality strategies is the effect of different levels of political commitment and/or available financial resources (Legido-Quigley, 2008; Lombarts, et al., 2009; Spencer & Walshe, 2008; Spencer & Walshe, 2009; Sunol et al., 2009). The most eminent drivers of policy have been governments, professional organizations, scientific societies, and media. Governments are the key players in developing and implementing quality improvement policies, and setting quality standards and targets. However, the lack of quality incentives, lack of funding for quality improvement and absence of professional

Notwithstanding the e-health potential for contributing to improved processes of health care, greater efficiency and enhanced understanding of clinically-effective care, financial and other barriers persist that result in comparatively low penetration rates of e-health. These barriers include acquisition and implementation costs, the lack of interoperability standards, skepticism about the business case for investment, uncertainty about system longevity, and psychological barriers related to uncertainty and change. E-health systems are expensive and providers face problems making the investment. Even if they can handle the initial acquisition and implementation costs, still must remain confident that an e-health system will improve efficiency or cover its upfront and ongoing costs to make the investment. In addition, some providers’ reluctance to adopt e-health may be a rational response to skewed financial incentives since the health care system provides little financial incentive for quality improvement (Blumenthal & Glaser, 2007). Perhaps the most controversial barrier to adoption of quality-related information technology is the lack of incentives for change. Therefore, non-governmental groups might play an extremely pivotal role in developing incentives, and regulatory agencies can ask for valid, standardized, and more easily gathered quality measurement data. Additional barriers to the spread of e-health include legal and regulatory concerns and technological issues (Davenport, 2007). Legal challenges include privacy of identifiable health information,

80

• • • •

sensitivity of health information privacy safeguards patient empowerment data and security protection

E-Health as the Realm of Healthcare Quality

training and education in quality improvement are the key barriers to progress. We might also argue that funding and financing cause conflict of interest. Furthermore, in the current economic crisis, governments are less inclined to invest in the development of quality strategies as they will not provide adequate financial return in the short or medium terms. Additionally, the existence of a statutory legal requirement to implement quality improvement strategies for healthcare systems and organizations is an overriding motivation for supporting progress in the development of quality improvement initiatives. Nevertheless, patient and service user organizations have the least impact on the development of quality improvement strategy. The preponderance of governments and health professions makes difficult for patient and user groups to set the limits of quality improvement policies and strategies. This means that the quality improvement activities reflect a professional and provider perspective. Another issue concerns evaluation and implementation strategies. Little information is available about the effects of the strategies. As a consequence, investing in quality strategies appears expensive and inefficient, and education and training in quality improvement constitutes a barrier to progress. Therefore, there is a role for quality improvement training in education and institutional development of healthcare professionals. Finally, quality improvement requires strong, engaged and informed professional leadership, which develops if healthcare professionals have access to appropriate training in healthcare quality improvement. However, such training is usually not available, the development of clinical guidelines, accreditation schemes, auditing of standards, and quality management are optional, and evaluation of quality improvement is not available to patients.

SOLUTIONS AND RECOMMENDATIONS The integration of e-health into the existing work flow is increasingly difficult due to the complexity of the clinical setting. Consequently, strong leadership within the clinical setting is essential in the successful implementation. Leadership should work in conjunction with the staff in order to mitigate any apprehensions (Iakovidis, 1998; Burton et al., 2004). With this internal backing and commitment, healthcare professionals will become involved and integrate e-health into their daily practice. Moreover, there is a critical relationship between organizational and technological change (Berg, 2001), which providers should understand prior to implementation because they constitute the driving force behind changes within the clinical setting. Conclusively, the successful implementation of e-health requires a thoughtful integration of routine procedures and information technology elements. E-health integration is a process, which is determined by the uniqueness of each setting (Berg, 2001) and social and human variables. Immediate policy changes, which promote e-health adoption, might include standards for interoperability, privacy protection, clinical effectiveness research, and amelioration of parallel working by software developers and health services researchers (Pagliari, 2007). Quality, access and efficiency are the general key issues for the success of e-health (Vitacca et al., 2009). However, a realistic definition of the dimensions of quality care is insufficient to accomplish the goal of continuous improvement. Therefore, healthcare quality improvement is a work in progress. Nonetheless, we might argue that quality improvement relates with leadership, measurement, reliability, practitioner skills and the marketplace. Kotter (1996) provides an eight-stage process to cope with change, which is necessary for the effective leadership. The eight processes include:

81

E-Health as the Realm of Healthcare Quality

• • • • • • • •

establish a sense of urgency set up the guiding coalition develop a vision and strategy spread the change perspective empower broad-based events trigger short-term wins consolidate gains and provide more change harbor new approaches in the culture

Change acceleration is also the issue of the ADKAR psychological model, which rates on a scale of 1 to 5 the following elements (Hiatt & Creasey, 2003): • • • • •

awareness of the need to change desire to participate and continue change understanding of how to change means to implement change support to keep change in place

Quality measurement involves the outcome or the process (Auerbach, 2009). However, process measurement is the commonest because we can easily measure changes in processes (Davis & Barber, 2004) that in the patient health status. There is also resistance although change has spawned a number of trends in the assessment of doctors’ competence (Norcini & Talati, 2009), when we recommend practicing medicine in a predictable and reliable way. Reliability revolves around command and control, risk appreciation, quality, metrics, and reward (Rochlin et al., 1987). However, a highly reliable organization must include mechanisms, which support flexibility, constrained improvisation, and cognition management (Bigley & Roberts, 2001). The problem in creating reliable processes is reducing variability, which interprets into high-quality decision making and high-quality performance of the practitioners. Finally, taking into account that quality is a vital component of healthcare business model, we have to understand the role it plays in the market. So far, quality has a tough time demonstrating its business case because of the complexity of care

82

and the difficulty in capturing the actual fixed and variable cost of patient treatment (Leatherman et al., 2003; Sachdeva & Jain, 2009).

FUTURE RESEARCH DIRECTIONS To substantiate the argument that e-health is the future realm of healthcare quality, we identify emerging scientific fields, and assess their impact on quality and patient safety. •

Biobanks, formally known as biological resource centers (BRCs), are depositories of ‘biological standards’. Haga & Beskow (2008) estimated that the U.S. store more than 270 million tissue samples, with a growth rate of approximately 20 million samples annually. There are many types of BRC, which differ in their functional role and the kinds of material they hold. Public and private entities have developed them in order to provide researchers the opportunity to explore collections of human biospecimen. However, we believe that all BRCs have the potential to unravel the causes of complex diseases, as well as the interaction between biological and environmental factors.

Biobanking is an emerging discipline (Riegman et al., 2008), which follows the continuing expansion of new techniques and scientific goals. Overall, it constitutes a device, which facilitates the understanding of the genetic basis of disease (Ormond et al., 2009) and holds taxonomic strains (Day et al., 2008). However, its establishment and maintenance is a skill-rich activity, which requires careful attention to the implementation of preservation technologies. Moreover, biobanking represents a challenge to informed consent (Ormond et al., 2009) while only appropriate quality assurance ensures that recovered cultures and other bio-

E-Health as the Realm of Healthcare Quality

logical materials behave in the same way as the originally isolated culture. •

•

A biochip is a series of microarrays arranged on a solid substrate which permits to perform various tests at the same time (Fan et al., 2009; Cady, 2009). It replaces the typical reaction platform and produces a patient profile, which we use in disease screening, diagnosis and monitoring disease progression. The development of biochips is a considerable thrust of the rapidly growing biotechnology industry, which encompasses a diverse range of research efforts. Advances in genomics, proteomics, and pharmaceuticals introduce new methods for unraveling the complex biochemical processes inside cells. At the same time, the field of microminiaturization enables biotechnologists to start packing their bulky sensing tools. These biochips are essentially miniaturized laboratories, which allow researchers to produce hundreds or thousands of simultaneous biochemical reactions, and rapidly screen large numbers of biological analyses. Reproducibility and standardization of biochip processes is, therefore, essential to ensure quality of results and provide the best tool for the elucidation of complex relationships between different proteins in detrimental conditions (Molloy et al., 2005). As traditional clinical investigation evolves, different needs have emerged, which require data integration. Data mining is the process of extracting patterns from data. It supports workflow analysis (Lang et al., 2007) and saves time and energy leading to less hassle for clinicians. While data mining can be used to detect patterns in data samples, it is crucial to understand that non-representative samples may cause non-indicative results. Therefore, we sometimes use the term in

a negative sense. As a result, in order to avoid confusion, we recently mention the negative perception of data mining as data dredging and data snooping. Some people believe that data mining is ethically neutral. However, the way we might use data mining can raise questions regarding privacy, legality, and ethics. Specifically, data mining may compromise confidentiality and privacy obligations through data aggregation. Data mining is rarely applied to healthcare, although it can facilitate the understanding of patient healthcare preferences (Liu et al., 2009). However, as we gather more data, data mining is becoming an increasingly indispensable tool in mining a wide selection of health records (Norén et al., 2008). Moreover, we might complement it with semantics-based reasoning in the management of medicines. Finally, data mining can support quality assurance (Jones, 2009), simplify the automation of data retrieval, facilitate physician quality improvement (Johnstone et al., 2008), and accurately capture patient outcomes if combined with simulation (Harper, 2005). Recently, there is interest in switching to algorithms and database development for microarray data mining (Cordero et al., 2008). •

Disease modeling, the mathematical representation of a clinical condition, summarizes the knowledge of disease epidemiology, and requires computational modeling, which follows two different approaches. The bottom-up approach accentuates complex intracellular molecular models while the top-down modeling strategy identifies key features of the disease.

However, despite ongoing efforts, there are complex issues regarding the use of computational modeling. A key question concerns the handling of model uncertainty since the selection of a computational model has to take into account the additional

83

E-Health as the Realm of Healthcare Quality

interactions and components beyond those of a discursive model. Therefore, a successful strategy is the identification of minimal models or search for parameter dimensions that are indispensable for the model performance. However, we cannot select competing structures on the basis of their relative fitness. As a result, a high priority for future disease modeling is model selection and hierarchical modeling (Tegnér et al., 2009). •

Genomics, the study of the genomes of living entities, encompasses considerable efforts to determine the complete DNA sequence of organisms through fine-scale genetic mapping efforts. It also includes studies of intragenomic phenomena and interactions between loci and alleles within the genome. However, it does not deal with the functions of individual genes, a primary focus of molecular biology or genetics. Research of individual genes does not fall into the meaning of genomics unless the purpose of the research is to clarify its impact on the networks of the genome.

Technology development has played a vital role in structural genomics (Terwilliger et al., 2009). Nowadays, we can quantify the difficulties of determining a pattern of a single protein. Moreover, the systems approach, which the post-genomics follow, interprets into a greater responsibility for artificial intelligence and robotics. Overall, many disciplines turn on the issue of automating the different stages in post-genomic research with a view to developing high-dimensional data of high quality (Laghaee et al., 2005). •

84

Molecular Imaging unites molecular biology and in vivo imaging while enabling the visualization of the cellular function and the follow-up of the molecular process (Weissleder & Mahmood, 2001). It differs from conventional imaging in that we use biomarkers to image specific reference

points. Biomarkers and their surroundings interact chemically altering the image according to molecular changes, which occur within the point of interest. This method is markedly different from previous methods of imaging which typically image differences in qualities. The accomplishment to image sheer molecular changes opens up an impressive number of exciting possibilities for medical attention. These include the optimization of preclinical and clinical tests of new medication and early detection and treatment of disease. Molecular imaging imparts a greater degree of fairness to quantitative tests, and numerous potentialities to the diagnosis of cancer, neurological and cardiovascular diseases. Therefore, it has a substantial economic impact due to earlier and more accurate diagnosis. •

Nanotechnology is the study of the atomic and molecular matter. Although, we might think that it only develops materials of the size of 100 nanometers or smaller, nanotechnology is extremely diverse. It encompasses the extensions of conventional device physics, different approaches based upon molecular self-assembly, and new materials with nanoscale dimensions.

Nanomedicine, the medical practice of nanotechnology, encompasses the use of nanomaterials and nanoelectronic biosensors and seeks to provide a valuable collection of research and tools (Wagner et al., 2006; Freitas, 2005a). New applications in the pharmaceutical industry include advanced drug delivery systems, alternative therapies, and in vivo imaging. Moreover, molecular nanotechnology, a preliminary subfield of nanotechnology, deals with the engineering of molecular assemblers. So far, it is highly speculative seeking to predict what inventions nanotechnology might produce. However, we already know that we will need nanocomputers to lead molecular assemblers and

E-Health as the Realm of Healthcare Quality

expect nanorobots to join the medical armamentarium (Cavalcanti et al., 2008; Freitas, 2005b). Finally, neuro-electronic interfacing, a visionary project dealing with the creation of nanodevices, will connect computers to the nervous system while nanonephrology will play a role in the management of patients with kidney disease. However, advances in nanonephrology require nano-scale information on the cellular molecular mechanism involved in kidney processes. There has been much discussion in which reasons are advanced for and against nanotechnology. However, nanotechnology has the inherent capacity to provide many different materials in medicine, electronics, and energy production. In contrast, it raises concerns about its ecological effects of nanomaterials, and their potential effects on international economics (Allhoff & Lin, 2008). These concerns have led to a dialogue among advocacy groups and governments on whether particular regulation of nanotechnology is warranted. •

Ontology is the philosophical study of the nature of an entity, which also examines the main categories of reality. As a part of the metaphysics philosophy, ontology deals with questions concerning the entity existence and their grouping according to similarities and differences. However, in computer science, ontology is a formal representation of a set of concepts and their relationships.

Ontologies have become a mainstream issue in biomedical research (Pesquita et al., 2009) since we can explain biological entities by using annotations. This type of comparability, which we call semantic similarity, assesses the length of connectedness between two entities using annotations similarity. The implementation of semantic similarity to biomedical ontologies is new. Nevertheless, semantic similarity is a valuable tool for the validation of gene clustering

results, molecular interactions, and disease gene prioritization (Pesquita et al., 2009). However, the capacity to assure the quality of ontologies or evaluate their eligibility is limited (Rogers, 2006). Therefore, we need a combination of existing methodologies and tools to support a comprehensive quality assurance scheme. However, an ontology of superlative quality is not verifiable and might not be useful since a ‘perfect’ ontology, which complies with all current philosophical theories, might be too complex (Rogers, 2006). •

Proteomics, a term coined in 1997 to make an analogy with genomics, is the comprehensive study of proteins, their structures and functions (Anderson & Anderson, 1998; Blackstock & Weir, 1999). The term proteome, a combination of protein and genome (Wilkins et al., 1996) defines the full complement of proteins (Wilkins et al., 1996), and encompasses the modifications made to a single set of proteins. Therefore, proteome varies with time rendering proteomics, the next step in the study of biological systems, as more complicated than genomics.

One of the most promising developments from the study of human genes and proteins is the discovery of potential new drugs. This relies on the identification of proteins associated with a disease, and involves computer software which uses proteins as targets for new drugs. For example, virtual ligand screening is a computer technique which attempts to fit millions of small molecules to the three-dimensional structure of a protein. The computer assigns a rank of quality matching to various sites in the protein, enhancing or disabling the role of the protein in the cell. In the last ten years, the field of proteomics has expanded at a rapid rate. Nevertheless, it appears that we have underestimated the level of stringency required in proteomic data generation

85

E-Health as the Realm of Healthcare Quality

and analysis. As a result, several published findings require additional evidence (Wilkins et al., 2006). However, clinical proteomics certify that the majority of errors occur in the preanalytical phase. Therefore, before introducing complex proteomic analysis into the clinical setting, we need standardization of the preanalytical phase. •

Modern health systems have focused their attention on micro issues, which include the minimization of diagnostic, treatment, and medication errors. However, safety culture, a macro issue, requires our attention. Safety culture is a sub-set of the organizational culture (Olive et al., 2006). Although an overriding concern in recent years, we need additional work to identify and soundly measure the key dimensions of patient safety culture (Ginsburg et al., 2009; Singer et al., 2008) and understand its relationship with the leadership.

A key finding of research is that efforts to improve culture may not succeed if hospital managers perceive patient safety differently from frontline workers. Research shows the pivotal role of managers in the promotion of employees’ safe behavior, both through their attitudes, and by developing a safety management system (Fernandez-Muniz et al., 2007). Senior managers perceive patient safety climate more positively than non senior managers while it has proved difficult to engage frontline staff with the concept (Hellings et al., 2007). Therefore, the agenda should move from rhetoric to converting the concept into observable behavior. A prerequisite for the realization of this perspective is the collection, analysis, and dissemination of information deriving from incidents and near misses as well as the adoption of the reporting, just, flexible and learning cultures (Ruchlin et al., 2004).

86

•

As patients become increasingly concerned about safety, clinical medicine focuses more on quality (Simmons & Wagner, 2009) than on bedside teaching and education. Educators react to these challenges by restructuring curricula, developing smallgroup sessions, and increasing self-directed education and independent research (Okuda et al., 2009). Nevertheless, there is a disconnect between the classroom and the clinical setting, resulting in medical students feeling that they are inadequately trained.

Medical simulation bridges the learning divide by representing certain key characteristics of a physical system. Historically, the first medical simulators were unsophisticated models of human patients (Meller, 1997). Later on, active models seemed to imitate living anatomy and more recently interactive models respond to actions taken by a student or physician. These are two dimensional computer programs, which constitute a textbook, and demonstrate the edge of allowing a student to fix errors. Quality improvement, patient safety, and the actual assessment of clinical skills have impelled medical simulation into the clinical arena (Carroll & Messenger, 2008). Still, there is convincing evidence that simulation training improves provider self-efficacy and effectiveness (Nishisaki et al., 2007) and increases patient safety. Finally, the process of iterative learning creates a much stronger learning environment and computer simulators are an ideal tool for evaluation of students’ clinical skills (Murphy et al., 2007).. On the contrary, there is no evidence that simulation training improves patient outcome. Therefore, we need ongoing academic research in order to evaluate the teaching effectiveness of simulation, and determine its impact on quality of care, patient safety (Cherry & Ali, 2008), and retention of knowledge. Even so, there is a plau-

E-Health as the Realm of Healthcare Quality

sible belief that medical simulation is the training device of the future.

CONCLUSION E-health, which already supports outstanding advances in healthcare quality, has the potential to become its realm, if we make a representation of the cognitive and social factors related to its design and use and associate them with the clinical practice.

REFERENCES Allhof, F., & Lin, P. (Eds.). (2008). Nanotechnology & Society: Current and Emerging Ethical Issues. New York: Springer. Anderson, J. G. (2007). Social, ethical and legal barriers to e-health. International Journal of Medical Informatics, 76(5-6), 480–483. doi:10.1016/j. ijmedinf.2006.09.016 Anderson, N. L., & Anderson, N. G. (1998). Proteome and proteomics: new technologies, new concepts, and new words. Electrophoresis, 19(11), 1853–1861. doi:10.1002/elps.1150191103 Auerbach, A. (2009). Healthcare quality measurement in orthopaedic surgery: current state of the art. Clinical Orthopaedics and Related Research, 467(10), 2542–2547. doi:10.1007/s11999-0090840-8 Balas, E. A., Weingarten, S., Garb, C. T., Blumenthal, D., Boren, S. A., & Brown, G. D. (2000). Improving preventive care by prompting physicians. Archives of Internal Medicine, 160, 301–308. doi:10.1001/archinte.160.3.301 Baldwin, K. B. (2008). Evaluating healthcare quality using natural language processing. Journal for Healthcare Quality, 30(4), 24–29.

Bass, L. (2005). Health literacy: Implications for teaching the adult patient. Journal of Infusion Nursing, 28(1), 15–22. doi:10.1097/00129804200501000-00002 Benson, H. (2005). Chaos and complexity: applications for healthcare quality and patient safety. Journal for Healthcare Quality, 27(5), 4–10. Berg, M. (2001). Implementing Information Systems in Health Care Organizations: Myths and Challenges. International Journal of Medical Informatics, 64, 143–156. doi:10.1016/S13865056(01)00200-3 Bigley, G. A., & Roberts, K. H. (2001). Structuring Temporary Systems for High Reliability. Academy of Management Journal, 44, 1281–1300. doi:10.2307/3069401 Blackstock, W. P., & Weir, M. P. (1999). Proteomics: quantitative and physical mapping of cellular proteins. Trends in Biotechnology, 17(3), 121–127. doi:10.1016/S0167-7799(98)01245-1 Blumenthal, D., & Glaser, J. (2007). Information technology comes to medicine. The New England Journal of Medicine, 356(24), 2527–2534. doi:10.1056/NEJMhpr066212 Bonis, P. A., Pickens, G. T., Rind, D. M., & Foster, D. A. (2008). Association of a clinical knowledge support system with improved patient safety, reduced complications and shorter length of stay among Medicare beneficiaries in acute care hospitals in the United States. International Journal of Medical Informatics, 77(11), 745–753. doi:10.1016/j.ijmedinf.2008.04.002 Brown, S. H., Elkin, P. L., Rosenbloom, S. T., Fielstein, E., & Speroff, T. (2008). eQuality for all: Extending automated quality measurement of free text clinical narratives. AMIA Annual Symposium Proceedings, 71-75.

87

E-Health as the Realm of Healthcare Quality

Burhans, L. D. (2007). What is quality? Do we agree, and does it matter? Journal for Healthcare Quality, 29(1), 39–44, 54.

Day, J. G., & Stacey, G. N. (2008). Biobanking. Molecular Biotechnology, 40(2), 202–213. doi:10.1007/s12033-008-9099-7

Burton, L. C., Anderson, G. F., & Kues, I. W. (2004). Using Health Records to Help Coordinate Care. The Milbank Quarterly, 82(3), 457–481. doi:10.1111/j.0887-378X.2004.00318.x

Eysenbach, G. (2001). What is e-health?In Keith E. Herold, K., & Rasooly, A. (Eds.) Journal of Medical Internet Research, 3(2), e20.

Cady, N. C. (2009). Microchip-based PCR Amplification Systems. Lab-on-a-Chip Technology: Biomolecular Separation and Analysis. Norwich, UK: Caister Academic Press. Carroll, J. D., & Messenger, J. C. (2008). Medical simulation: the new tool for training and skill assessment. Perspectives in Biology and Medicine, 51(1), 47–60. doi:10.1353/pbm.2008.0003 Cavalcanti, A., Shirinzadeh, B., Freitas, RA Jr., Hogg, T. (2008). Nanorobot architecture for medical target identification. Nanotechnology, 19 (1), 015103(15pp). Cherry, R. A., & Ali, J. (2008). Current concepts in simulation-based trauma education. The Journal of Trauma, 65(5), 1186–1193. doi:10.1097/ TA.0b013e318170a75e Coiera, E., Westbrook, J. I., & Wyatt, J. C. (2006). The Safety and Quality of Decision Support Systems. In Haux, R., & Kulikowski, C. (Eds.), IMIA Yearbook of Medical Informatics. Cordero, F., Botta, M., & Calogero, R. A. (2007). Microarray data analysis and mining approaches. Briefings in Functional Genomics & Proteomics, 6(4), 265–281. doi:10.1093/bfgp/elm034 Davenport, K. (2007). Navigating American Health Care: How Information Technology Can Foster Health Care Improvement. Center for American Progress. Retrieved April 24, 2009, from www. americanprogress.com Davis, K., Otwell, R., & Barber, J. (2004). Managing costs through clinical quality improvement. Healthcare Financial Management, 58(4), 76–82.

88

Fan, Z. H., Das, C., & Chen, H. (2009). TwoDimensional Electrophoresis in a Chip. In Keith E. Herold, K., & Rasooly, A. (Ed.), Lab-ona-Chip Technology: Biomolecular Separation and Analysis. (pp. 122-138). Norfolk: Caister Academic Press. Fernandez-Muniz, B., Montes-Peon, J. M., & Vazquez-Ordas, C. J. (2007). Safety culture: analysis of the causal relationships between its key dimensions. Journal of Safety Research, 38(6), 627–641. doi:10.1016/j.jsr.2007.09.001 Freitas, R. A. Jr. (2005a). What is Nanomedicine? Nanomedicine; Nanotechnology, Biology, and Medicine, 1(1), 2–9. doi:10.1016/j. nano.2004.11.003 Freitas, R. A. Jr. (2005b). Current Status of Nanomedicine and Medical Nanorobotics. Journal of Computational and Theoretical Nanoscience, 2, 1–25. Gandhi, T. K., & Bates, D. W. (2001). Computer Adverse Drug Event (ADE) Detection and Alerts. Chapter 8 in: University of California at San Francisco – Stanford University Evidence based Practice Center Making Health Care Safer: A Critical Analysis of Patient Safety Practices (pp 81). Ginsburg, L., Gilin, D., Tregunno, D., Norton, P. G., Flemons, W., & Fleming, M. (2009). Advancing measurement of patient safety culture. Health Services Research, 44(1), 205–224. doi:10.1111/j.1475-6773.2008.00908.x

E-Health as the Realm of Healthcare Quality

Haga, S. B., & Beskow, L. M. (2008). Ethical, legal, and social implications of biobanks for genetics research. Advances in Genetics, 60, 505–544. doi:10.1016/S0065-2660(07)00418-X Harper, P. (2005). Combining data mining tools with health care models for improved understanding of health processes and resource utilisation. Clinical and Investigative Medicine. Medecine Clinique et Experimentale, 28(6), 338–341. Harteloh, P. (2004). The Meaning of Quality in Health Care: A Conceptual Analysis Health Care Analysis. Health Care Analysis, 11(3), 259–267. doi:10.1023/B:HCAN.0000005497.53458.ef Hellings, J., Schrooten, W., Klazinga, N., & Vleugels, A. (2007). Challenging patient safety culture: survey results. International Journal of Health Care Quality Assurance, 20(7), 620–632. doi:10.1108/09526860710822752 Hiatt, J. M., & Creasey, T. J. (2003). Change Management. Loveland, CO: Prosci Research. Hodge, J. G. Jr, Gostin, L. O., & Jacobson, P. D. (1999). Legal issues concerning electronic health information: privacy, quality, and liability. Journal of the American Medical Association, 282(15), 1466–1471. doi:10.1001/jama.282.15.1466 Honigman, B., Lee, J., Rothschild, J., Light, P., Pulling, R. M., Yu, T., & Bates, D. W. (2001). Using computerized data to identify adverse drug events in outpatients. Journal of the American Medical Informatics Association, 8, 254–266. Honigman, B., Light, P., Pulling, R. M., & Bates, D. W. (2001). A computerized method for identifying incidents associated with adverse drug events in outpatients. Journal of the American Medical Informatics Association, 61, 21–32.

Iakovidis, I. (1998). Towards Personal Health Record: Current Situation, Obstacles, and Trends in Implementation of Electronic Healthcare Record in Europe. International Journal of Medical Informatics, 52, 105–115. doi:10.1016/S13865056(98)00129-4 Johnstone, P. A., Crenshaw, T., Cassels, D. G., & Fox, T. H. (2008). Automated data mining of a proprietary database system for physician quality improvement. International Journal of Radiation Oncology, Biology, Physics, 70(5), 1537–1541. doi:10.1016/j.ijrobp.2007.08.056 Jones, A. R. (2009). Data mining can support quality assurance. Journal of the Royal Society of Medicine, 102(9), 358–359. doi:10.1258/ jrsm.2009.090216 Kluge, E. H. (2007). Secure e-Health: managing risks to patient health data. International Journal of Medical Informatics, 76(5-6), 402–406. doi:10.1016/j.ijmedinf.2006.09.003 Kuperman, G. J., Teich, J., Bates, D. W., Hilz, F. L., Hurley, J., Lee, R. Y., et al. (1996). Detecting alerts, notifying the physician, and offering action items: a comprehensive alerting system. Proceedings of AMIA Annual Fall Symposium, (pp. 704-708). Laghaee, A., Malcolm, C., Hallam, J., & Ghazal, P. (2005). Artificial intelligence and robotics in high throughput post-genomics. Drug Discovery Today, 10(18), 1253–1259. doi:10.1016/S13596446(05)03581-6 Lang, M., Kirpekar, N., Burkle, T., Laumann, S., & Prokosch, H. U. (2007). Results from data mining in a radiology department: the relevance of data quality. Studies in Health Technology and Informatics, 129(Pt 1), 576–580. Leatherman, S., Berwick, D., Iles, D., Lewin, L., Davidoff, F., Nolan, T., Bisognano, M. (2003). The Business Case for Quality: Case Studies and an Analysis. Health Affairs, 22 (2),17–30, 18.

89

E-Health as the Realm of Healthcare Quality

Legido-Quigley, H., McKee, M., Walshe, K., Suñol, R., Nolte, E., & Klazinga, N. (2008). How can quality of care be safeguarded across the European Union? British Medical Journal, 336, 920–923. doi:10.1136/bmj.39538.584190.47

Nishisaki, A., Keren, R., & Nadkarni, V. (2007). Does Simulation Improve Patient Safety?: SelfEfficacy, Competence, Operational Performance, and Patient Safety. Anesthesiology Clinics, 25(2), 225–236. doi:10.1016/j.anclin.2007.03.009

Liu, S. S., & Chen, J. (2009). Using data mining to segment healthcare markets from patients’ preference perspectives. International Journal of Health Care Quality Assurance, 22(2), 117–134. doi:10.1108/09526860910944610

Norcini, J., & Talati, J. (2009). Assessment, surgeon, and society. International Journal of Surgery, 7(4), 313–317. doi:10.1016/j.ijsu.2009.06.011

Lobach, D. F., & Hammond, W. E. (1994). Development and evaluation of a computer-assisted management protocol (CAMP): improved compliance with care guidelines for diabetes mellitus. Proceedings of the Annual Symposium on Computer Applications in Medical Care, 787-791 Lombarts, M. J. M. H., Rupp, I., Vallejo, P., Suñol, R., & Klazinga, N. (2009). Application of quality improvement strategies in 389 European hospitals: results of the MARQuIS project. Quality & Safety in Health Care, 18, 28–37. doi:10.1136/ qshc.2008.029363 Meller, G. (1997). A Typology of Simulators for Medical Education. Journal of Digital Imaging, 10, 194–196. doi:10.1007/BF03168699 Metaxiotis, K., Ptochos, D., & Psarras, J. (2004). E-health in the new millennium: a research and practice agenda. International Journal of Electronic Healthcare, 1(2), 165–175. doi:10.1504/ IJEH.2004.005865 Molloy, R. M., Mc Connell, R. I., Lamont, J. V., & FitzGerald, S. P. (2005). Automation of biochip array technology for quality results. Clinical Chemistry and Laboratory Medicine, 43(12), 1303–1313. doi:10.1515/CCLM.2005.224 Murphy, D., Challacombe, B., Nedas, T., Elhage, O., Althoefer, K., Seneviratne, L., & Dasgupta, P. (2007). Equipment and technology in robotics (in Spanish; Castilian). Archivos Espanoles de Urologia, 60(4), 349–355.

90

Norén, G. N., Bate, A., Hopstadius, J., Star, K., & Edwards, I. R. (2008). Temporal Pattern Discovery for Trends and Transient Effects: Its Application to Patient Records. In Proceedings of the Fourteenth International Conference on Knowledge Discovery and Data Mining SIGKDD 2008 (pp. 963-971). Las Vegas NV. Nusslin, F. (2006). Current status of medical technology. Acta Neurochirurgica, 98, 25–31. doi:10.1007/978-3-211-33303-7_5 Okuda, Y., Bryson, E. O., DeMaria, S Jr., Jacobson, L., Quinones, J., Shen, B., & Levine, A. I. (2009). The utility of simulation in medical education: what is the evidence? Mount Sinai Journal of Medicine: A Journal of Translational and Personalized Medicine, 76(4), 330-343. Olive, C., O’Connor, T. M., & Mannan, M. S. (2006). Relationship of safety culture and process safety. Journal of Hazardous Materials, 130(12), 133–140. doi:10.1016/j.jhazmat.2005.07.043 Ormond, K. E., Cirino, A. L., Helenowski, I. B., Chisholm, R. L., & Wolf, W. A. (2009). Assessing the understanding of biobank participants. American Journal of Medical Genetics, 149A(2), 188–198. doi:10.1002/ajmg.a.32635 Pagliari, C. (2007). Design and evaluation in eHealth: challenges and implications for an interdisciplinary field. Journal of Medical Internet Research, 9(2), e15. doi:10.2196/jmir.9.2.e15

E-Health as the Realm of Healthcare Quality

Pallesen, B., Engberg, A., & Barlach, A. (2006). Developing e-health information by empowerment strategy. Studies in Health Technology and Informatics, 122, 776. Pesquita, C., Faria, D., Falcao, A. O., Lord, P., & Couto, F. M. (2009). Semantic similarity in biomedical ontologies. PLoS Computational Biology, 5(7), e1000443. doi:10.1371/journal. pcbi.1000443 Riegman, P. H., Morente, M. M., Betsou, F., de Blasio, P., & Geary, P. (2008). Biobanking for better healthcare. Molecular Oncology, 2(3), 213–222. doi:10.1016/j.molonc.2008.07.004 Rivard, P. E., Rosen, A. K., & Carroll, J. S. (2006). Enhancing patient safety through organizational learning: Are patient safety indicators a step in the right direction? Health Services Research, 41(4 Pt 2), 1633–1653. doi:10.1111/j.14756773.2006.00569.x Rochlin, G. I., La Porte, T.R., & Roberts, K. H., K.H. (1987). The Self-Designing High-Reliability Organization: Aircraft Carrier Flight Operations at Sea. Naval War College Review, 40(4), 76–90. Rogers, J. E. (2006). Quality assurance of medical ontologies. Methods of Information in Medicine, 45(3), 267–274. Ruchlin, H. S., Dubbs, N. L., & Callahan, M. A. (2004). The role of leadership in instilling a culture of safety: lessons from the literature. Journal of Healthcare Management, 49(1), 47–58. Runciman, W. B. (2002). Lessons from the Australian Patient Safety Foundation: setting up a national patient safety surveillance system – is this the right model? Quality & Safety in Health Care, 11, 246–251. doi:10.1136/qhc.11.3.246 Sachdeva, R. C., & Jain, S. (2009). Making the case to improve quality and reduce costs in pediatric health care. Pediatric Clinics of North America, 56(4), 731–743. doi:10.1016/j.pcl.2009.05.013

Schiff, G. D., Aggarwal, H. C., Kumar, S., & McNutt, R. A. (2000). Prescribing potassium despite hyperkalemia: medication errors uncovered by linking laboratory and pharmacy information systems. The American Journal of Medicine, 109, 494–497. doi:10.1016/S0002-9343(00)00546-5 Schiff, G. D., & Rucker, T. D. (1998). Computerized prescribing: building the electronic infrastructure for better medication usage. Journal of the American Medical Association, 279, 1024–1029. doi:10.1001/jama.279.13.1024 Simmons, B., & Wagner, S. (2009). Assessment of continuing interprofessional education: lessons learned. The Journal of Continuing Education in the Health Professions, 29(3), 168–171. doi:10.1002/chp.20031 Singer, S. J., Falwell, A., Gaba, D. M., & Baker, L. C. (2008). Patient safety climate in US hospitals: variation by management level. Medical Care, 46(11), 1149–1156. doi:10.1097/ MLR.0b013e31817925c1 Sittig, D. F., & Stead, W. W. (1994). Computerbased physician order entry: the state of the art. Journal of the American Medical Informatics Association, 1, 108–123. Smith, Moore T., Francis, M. D., & Corrigan, J. M. (2007). Health care quality in the 21st century. Clinical and Experimental Rheumatology, 25, 3–5. Sood, S., Mbarika, V., Jugoo, S., Dookhy, R., Doarn, C. R., Prakash, N., & Merrell, R. C. (2007). What is telemedicine? A collection of 104 peer-reviewed perspectives and theoretical underpinnings. Telemedicine journal and e-health: the official journal of the American Telemedicine Association, 13(5), 573-590. Spencer, E., & Walshe, K. (2008). Quality and safety in healthcare in Europe. A growing challenge for policymakers. Harvard Health Policy Review, 1, 46–54.

91

E-Health as the Realm of Healthcare Quality

Spencer, E., & Walshe, K. (2009). National quality improvement policies and strategies in European Healthcare Systems. Quality & Safety in Health Care, 18, 22–27. doi:10.1136/qshc.2008.029355

Wilkins, M. R., Appel, R. D., Van Eyk, J. E., Chung, M. C., Gorg, A., & Hecker, M. (2006). Guidelines for the next 10 years of proteomics. Proteomics, 6(1), 4–8. doi:10.1002/pmic.200500856

Suñol, R., Vallejo, P., Thompson, A., Lombarts, M. J. M. H., Shaw, C. D., & Klazinga, N. (2009). Impact of quality strategies on hospital outputs. Quality & Safety in Health Care, 18, 57–63. doi:10.1136/qshc.2008.029439

ADDITIONAL READING

Tegnér, J. N., Compte, A., Auffray, C., An, G., Cedersund, G., & Clermont, G. (2009). Computational disease modeling - fact or fiction? BMC Systems Biology, 3, 56. doi:10.1186/1752-0509-3-56 Terwilliger, T. C., Stuart, D., & Yokoyama, S. (2009). Lessons from structural genomics. Annual Review of Biophysics, 38, 371–383. doi:10.1146/ annurev.biophys.050708.133740 Vitacca, M., Mazzu, M., & Scalvini, S. (2009). Socio-technical and organizational challenges to wider e-Health implementation. Chronic Respiratory Disease, 6(2), 91–97. doi:10.1177/1479972309102805 Wagner, V., Dullaart, A., Bock, A. K., & Zweck, A. (2006). The emerging nanomedicine landscape. Nature Biotechnology, 24(10), 1211–1217. doi:10.1038/nbt1006-1211 Weissleder, R., & Mahmood, U. (2001). Molecular imaging. Radiology, 219, 316–333. Wilkins, M., Pasquali, C., Appel, R., Ou, K., Golaz, O., & Sanchez, J. C. (1996). From Proteins to Proteomes: Large Scale Protein Identification by Two-Dimensional Electrophoresis and Arnino Acid Analysis. Nature Biotechnology, 14(1), 61–65. doi:10.1038/nbt0196-61

Elger, B., Biller-Andorno, N., Mauron, A., Capron, A., & Morgan Capron, A. (2008). Ethical Issues in Governing Biobanks. Hampshire, UK: Ashgate Publishing Limited. Gottweis, H. (2008). Biobanks: Governance in Comparative Perspective. New York: Roputledge. Kewal, J. (2009). The Handbook of Nanomedicine. Totwa, New Jersey: Humana Press. Nisbet, R., Elder, J., & Miner, G. (2009). Handbook of Statistical Analysis and Data Mining Applications. Cambridge, MA: Elsevier. Pevsner, J. (2009). Bioinformatics and Functional Genomics. New Jersey: Wiley & Sons. doi:10.1002/9780470451496 Pisanelli, D. M. (2004). Ontologies in Medicine. Amsterdam: IOS Press. Riley, R. (2008). A Manual of Simulation in Healthcare. Oxford, UK: Oxford University Press. Xing, W. L., & Cheng, J. (2006). Frontiers in Biochip Technology. New York: Springer. doi:10.1007/b135657

This work was previously published in E-Health Systems Quality and Reliability: Models and Standards, edited by Anastasius Moumtzoglou and Anastasia Kastania, pp. 291-310, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

92

93

Chapter 1.6

The Use of Personal Digital Assistants in Nursing Education Nina Godson Coventry University, UK Adrian Bromage Coventry University, UK

ABSTRACT The use of Personal Digital Assistants (PDAs) and smartphones (combined mobile telephone and PDA) in Nurse Education is a relatively new development, in its infancy. The use of mobile technologies by health care professionals is increasing, and it seems likely to accelerate, as mobile information and communication technologies become more ubiquitous in wider society. The chapter reports on a small-scale feasibility study to evaluate the practicalities of supporting student nurses on their first clinical placements with PDAs that have been pre-loaded with reusable e-learning DOI: 10.4018/978-1-60960-561-2.ch106

objects. The student nurses generally found the PDA easy to use and carry on their person, valued the availability of the reusable e-learning object on their clinical placements and called for more of them to be made available to learners.

INTRODUCTION Clinical Nurse Tutors face the challenge of managing large numbers of students in clinical skills sessions, when demonstrating and practising skills acquisition, including interprofessional working. This involves maintaining accurate and fair assessments of individual performance that may later serve as evidence of safe practice for their

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The Use of Personal Digital Assistants in Nursing Education

clinical placements. There needs to be a means of continuing this supportive learning environment, in the transition from simulation to reality, to ensure the bridge from theory to practice is strong. Ideally, to enhance learning and support development of decision-making skills, students should have access to information at the moment and the physical location where it is needed. In relation to this, a major aim of pre registration nursing programs is to assist students to attain competencies for practice, for example, in the use of the tools and technology associated with nursing. Healthcare professionals need to continually update knowledge and skills, in order to enhance their clinical practice and professional development. Similarly, student nurses in any speciality must get into the habit of continually becoming familiar with the latest practice recommendations, such as guidelines for preventing the spread of infection and standards from regulatory agencies. Using the world wide web can help meet these needs, given that anyone can publish web-pages that are then available instantly across the globe. Information published on line offers advantages compared to text books and journal articles, which are often outdated by the time they reach publication and distribution. However, up to 2003, connection to the internet was typically on either a desktop or laptop personal computer. Since that time, pocket-sized personal digital assistants (PDAs), also known as palmtop computers, with wireless access to the World Wide Web have become available. The first PDA, the Apple Newton MessagePad, was launched in 1993. In 1996, Nokia introduced the first mobile telephone with full PDA functionality (a genre known as ‘smartphones’). However, early PDAs relied on wired connection to networked desktop or laptop PCs, until the launch in 2003 of the Hewlett-Packard iPAQ H5450. This was the first PDA featuring built-in Wi-Fi for connecting to wireless computer networks and the World Wide Web. In 2009, the majority of PDAs sold are Wi-Fi enabled smartphones, for

94

example, the RIM Blackberry, the Apple iPhone and the Nokia N-Series. It is argued in this chapter that the Personal digital assistants (PDAs) and smartphones can enable nurse educators can incorporate into the curriculum opportunities for students to develop critical approaches, and also introduce information technology as a tool in clinical decision making. This will serve as a foundation for using mobile computing technologies later in their careers. This chapter describes a small-scale feasibility study in which the practicalities of using smartphones to support the undergraduate nursing curriculum programme at a UK University were evaluated. The students who participated were drawn from all branches of nursing, and were in their first year of nurse training.

BACKGROUND In the mid-1990’s the World Wide Web radically changed the possibilities of our information landscape, and simultaneously became available to users of pocket-sized hand held computing devices such as Personal Digital Assistants (PDAs) (Murphy 2005). Even so, the use of PDAs in nursing was rare until comparatively recently. The Cumulative Index to Nursing and Allied Health Literature (CINAHL) has had a subject heading for PDAs (“Computers, Hand-Held”) since 1997, while the National Library of Medicine first used the term “Computers, Handheld” in the Medical Subject Headings (MeSH) in 2003. Discussions in the early articles focused on theories of informatics and technology. More recent articles indicate that PDAs are in wider use (De Groote, 2004), for example in nursing practice and student education. Interestingly, the Medical professions have written most of the literature on PDA use in health care, and it is clear that as a group they are very interested in this. Members of other health care disciplines also demonstrate interest in using PDAs (De Groote, 2004). The

The Use of Personal Digital Assistants in Nursing Education

integration of personal digital assistants (PDAs) into health care delivery continues to grow; and health professionals are adopting PDAs faster than the general public (Stolworthy et al., 2000). However, there has been little evidence that PDAs improve patient care (Fischer, 2003). PDA use by healthcare professionals tends to be in the context of on everyday routine. Many use their PDAs as a diary or address book, rather than as a knowledge base (Criswell & Parchman, 2002). Considering these facts, it might be argued that the use of PDA s has not been fully understood or its full potential recognised by health care professionals. However, the PDA can be considered a new learning tool, offering the potential to keep healthcare professionals up to date with developments in evidence based practice, and having potential to help nurse educators meet students’ educational needs, as will be seen. There is emerging evidence of PDA use in graduate nurse education as well as undergraduate nursing programs (Thomas et al., 2001). There have been accounts of how various institutions have used PDAs, for example, Huffstutler (2002). While university libraries have been known to provide information and support to students who use PDAs, many health care programs making use of PDAs seek help from institutional information technology department instead (Young, 2002). The incorporation of information and communication technologies into nursing practice and education by student nurses and staff is potentially challenging, particularly on their clinical work placements. To meet the healthcare needs of today’s society, student nurses are expected to demonstrate compassion when providing patient care. However, reduced time for patient contact may have eroded their ability to do so. It has become clear that increasing staff workloads have increased while staffing levels have decreased in the clinical setting (Aston & Molassiotis, 2003). Simultaneously, nursing curricula are forever evolving, introducing new ways of learning and new learning technologies, partly in response to

the challenge of accommodating increasingly large student cohorts. When student nurses go on clinical placements, teaching in the ward area is guided by both time and space limits. In light of nursing shortages, teaching that takes place at the patient’s bedside is not always possible, and it is becoming more difficult to take students away from the ward area to train them. Thus the students’ educational needs are not always reliably met within the current context of patient care settings. Nursing educators can usefully integrate technology as a means of support to student’s learning and to prepare the new generation of nurses as independent learners. Pre-registration nursing programmes aim to assist undergraduate students to attain competencies for practice, including interprofessional working, and the earlier these skills are introduced the better. Nurse tutors need to constantly review areas where information and training gaps exist, to highlight where nurse mentors at clinical placements and university-based tutors can play a larger part in training and preparation, both before and during students’ clinical placements. In the case described here, a need was identified for first-year undergraduate students on their first clinical placements to have access to learning materials and resources at the moment they are needed, to enhance learning and support development of practical clinical skills (Huffstutler, 2002). In the feasibility study reported in this chapter, tutors responded to these pressures by developing a way to meet students’ individual learning needs and styles.

The use of PDAs by Student Nurses The evidence discussed in the preceding section of increasing adoption of PDAs by healthcare professionals suggests that PDAs can be considered as if any other piece of healthcare equipment. It is arguable that nurse educators should begin to introduce PDAs to student nurses, given the importance of providing opportunities for students to become familiar with PDAs as a tool for clinical skills

95

The Use of Personal Digital Assistants in Nursing Education

development. Both tutors and students are more likely to learn to use PDAs effectively through consistent use in both clinical and classroom settings. However, it was seen in the preceding section that there is currently limited literature addressing use of PDAs in the nursing curriculum, and it appears that their educational potential in clinical settings is relatively unexplored. The preparation of student nurses will always involve the development of clinical skills at the patient bedside, therefore quick and easy access to information is needed, not only in the skills laboratory but also in clinical practice. PDAs can operate as if electronic text books to help student nurses manage the information they need to learn. In the past, when students had queries about practical skills, they would turn to textbooks for answers, often having to find relevant volumes. This can be very time consuming, when there are so many clinical skills to learn, not to mention the problem of carrying heavy books to clinical placements. A PDA can enable student nurses to retrieve information in the clinical area quickly and easily, conserving precious learning time (Koeniger-Donohue, 2008). PDAs can provide nurses with portable access to extensive reference materials as well as other organizational resources and time saving benefits. This can be in the form of reusable elearning objects accessible from a central point by an entire cohort of students. The experience of learning from a PDA potentially offers students an opportunity to revisit at will any practical skill and reflect on their interpretation and understanding of it, so knowledge is accumulated and transferred into an individual’s own understanding (White et al., 2005). This way of learning alongside other supportive mechanisms promises an enriching clinical experience for the student. This is especially relevant when mentors are busy with heavy workloads and staffing has decreased due to cut backs in the economy (Yonge et al., 2002), and ward areas are as a result struggling to meet the

96

educational needs of student on clinical placements. The ultimate goal of nurse educators is to develop competent and passionate student learners, who feel supported in transition from academia to clinical practice. PDAs can be a powerful tool that ‘holds the hand’ of trainee nurses when reassurance is needed. This could help reduce students’ stress, which has been found to lead them to miss work shifts or abandon their studies (Timmins & Kaliszer, 2002). However, handheld devices must fit into the ward routine, and not be perceived as extra work, or as interfering with the nurse-patient interaction or nurse mentor relationships. The question of the practicalities of PDAs on clinical placement required investigation before tutors at the university in which the case study reported in this chapter took place could decide whether to invest time, effort and money in their adoption as learning tools.

Feasibility Study: Knowledge in the Palm of your Hands In 2008, a small-scale feasibility study was undertaken at Coventry University to examine the fundamental practicalities and issues surrounding the use of hand held devices (PDAs) by first year student nurses on their first clinical placement. It was thought that by creating reusable learning objects covering relevant skills and competencies and loading them onto a PDA, student nurses from any branch of nursing can practice the skills and competencies in the clinical room, were their mentor unavailable. The student nurses can gain confidence and requisite background knowledge prior to the hands-on portion of their learning, saving the mentor time to work toward mastering these skills. The student nurse can then move to the patient bedside to gain the crucial supervised practice required to successfully and safely perform clinical skills. While the learning object used in the study was not explicitly designed to promote

The Use of Personal Digital Assistants in Nursing Education

interprofessional learning, this was appropriate, give the aims of the feasibility study to explore the fundamental practicalities of smartphones on clinical placements. It is worth briefly outlining the nature of the learning object, however, and considerations that were taken into account to conduct a fair trial of the PDA. The reusable e-learning object was designed to make the learning experience personally relevant and meaningful, related to unambiguous skills that students had previously learned in the skills laboratories run in face-to-face university-based teaching. This was so the nature of its content did not distort the students’ smartphone use by being perceived as irrelevant or too challenging in its content matter. It comprised a “teachable moment” that was developed by nurse educators at the University into a reusable e-learning object that depicts a generic clinical skill for all branches of nursing, a hand-washing technique known as the ‘Ayliff procedure’ (Ayliffe, 1992). The resulting reusable e-learning object features a set of film clips that demonstrate and explain the Ayliff procedure (Gould, 2008), and was loaded on to each smartphone, enabling the participants to take it into their clinical practice placements. The issue of using smartphones in a clinical area was discussed with an infection control nurse. It was decided that students should not use smartphones at patients’ bedside and only in a study area. Thus it was explained to the participants that as an infection control precaution, smartphones were only to be used in the study room in the clinical area, and never at patients’ bedsides. Ethical approval for the study was sought from the University ethics committee, and was duly granted.

Methodology PDA Choice

considered carefully. Having a standard device was crucial, and one that is simple to use by all. Basically, there are two operating systems systems: the ‘Palm’ System and the ‘Pocket PC’ System. This study used a smartphone running the palm operating system. It was found by the course tutors that this operating system has the advantage of simplicity and ease of use. Smartphones are becoming as popular as ordinary mobile phones, which most students already possess, thus it was thought that the students would be more likely to adapt to a smartphone rather than a pure PDA. It is worth noting that there is also a learning curve for the tutors involved. It was important that the research team had access to the log in information, so they themselves could not be ‘locked out’of the smartphones. Preparing the smartphones for use is time consuming, in that reusable e-learning resources must be loaded onto each one. However once loaded the data are stored permanently.

Participants The sample comprised eighteen female students volunteers from a cohort of 215 students, as the researcher only had funding for twenty smartphones. The students were from the first year of a pre-registration nursing program that ran in January 2008, covering all four branches of nursing; adult, mental health, child and young persons, and learning disabilities. They were about to go on their first clinical placement, and they were to visit different types of placements, hospital based and community. It was considered important that the participants expressed an interest in the use of such technology, as should any technical hurdles become apparent while the students are on placement, they would have to solve them with only telephone or email support from the researchers.

To ensure that the PDA has a long service life, given the capital investment necessary for their purchase, the best choice of product needs to be

97

The Use of Personal Digital Assistants in Nursing Education

Ethical Briefing of Participants The project was introduced to the students during one of their initial lectures, and a participant ethical briefing sheet was distributed. The briefing explained that each participant would be asked to complete both a pre- and post-placement questionnaire to evaluate the use of the smartphone, that they had the right to withdraw from the study at any time, and that their responses would be anonymised and pooled, then statistically evaluated and published. Students read the information and asked any questions, and those who were interested signed up to participate. A further date was arranged for those who signed up to meet with the researchers and learn how to use the smartphone and discuss the project further, after which they signed the necessary ‘consent to participate’ forms.

Training for Smartphone Use The smartphone is a tool that a student must learn to use effectively, just as they would a medical tool such as a tympanic thermometer or sphygmomanometer. Clinical Skills are developed through consistent practice in the clinical skills laboratory, and similar techniques can be used to familiarise individuals with a smartphone. To these ends, the participants were invited to a workshop on how to use the smartphone. This session provided hands on experience, in a structured supportive environment. Each student was briefed about how to use the smartphone and provided with an instruction leaflet, and was then shown how to log into the device. They were briefed that after their twelve-week placement had finished, they were to return the smartphone to the researchers. It was also explained that as an infection control precaution, the devices were only to be used in the study room in the clinical area, and never at patients’ bedsides. Getting started with a PDA is like learning any other new skill. A positive attitude, practice and willingness to make mistakes will help to over-

98

come initial unfamiliarity. The students were thus asked to keep the smartphone in their coat pocket or work bag while in placement, so as to have the reusable learning objects available whenever they felt they needed to review the handwashing technique. The students were reminded that they were not alone. A contact number for the researchers was provided, in case the participants needed technical support. They were also made aware that they were effectively in a networking group with other students in the placement using the same device.

Pre-Placement Questionnaire The students next completed a pre clinical placement questionnaire. Its purpose was to gather information about their general familiarity with information and communication technologies, and any learning resources they had used so far in their training for the hand washing technique. It was found that the majority of participants owned a mobile phone. Furthermore, nine of the eighteen participants posessed PDAs of their own, and were familiar with the function and layout of these devices. All students seemed keen to participate in the project, and did not appear worried about using the smartphones. The main learning resources that the students had used prior to clinical placement were clinical skills laboratories, online learning, handouts/ leaflets, Objective Structured Clinical Examinations and Clinical skills tutors. Objective structured clinical examinations are used as a form of assessment of clinical skills at the beginning of the nursing curriculum. Students found clinical skills laboratories and open laboratories the most effective way of learning the hand washing technique. The majority of participants found pictorial presentation, continual practice of the skill and memory aides the most effective ways of remembering the eight-step hand washing technique. Some felt that they needed reminders

The Use of Personal Digital Assistants in Nursing Education

of these steps, not only in the skills laboratories but in clinical practice placements. The participants felt that, pre-placement, they were able to demonstrate the correct hand washing technique, and would practice further to become proficient at this clinical skill. The participants were then asked to comment upon the least effective method of learning. The majority stated that the least effective was learning from textbooks and from previous practical experience before entering the nursing course. It could be argued that textbooks offer abstract experience, not ‘hands on’ concrete experience, and that previous care experience may not have taught the most up-todate technique. Finally the students were given the option to state their Objective Structured Clinical Examination (OSCE) score on the pre- placement questionnaire. All OSCEs include the assessment of hand washing. All of the participants in the sample indicated that they had passed the year one OSCE, however, some chose not to record their actual score on the pre-placement questionnaire.

Post-Placement Questionnaire Each participant then took their smartphone into their first clinical placement, and twelve weeks later returned from their placements to the University for the theoretical component of their course. At this point, each participant completed a post- placement evaluation questionnaire, so that their thoughts and experiences on using the smartphones while on their placements could be evaluated. The questionnaire comprised eighteen likert scale questions and answer options in three ‘blocks’ of questions. The first block covered the usability of the smartphone. The second covered the educational effectiveness of the reusable e-learning object. The final block covering the participants’ feelings about the value of using the learning objects and smartphone on placement. The data of interest in this account are the

first and third question blocks, which dealt with the usability of the smartphone, rather than the block covering the reusable e-learning object itself. This is partly because the latter was not explicitly designed to promote interprofessional learning, but mainly because this chapter seeks to examine the practicalities of mobile learning on clinical placements. The scores on each question for each participant were entered onto a MS-Excel spreadsheet, and the number of participants choosing each response option for each question were summed and then entered onto bar charts that summarised the overall results for each question. The measure of central tendency is the modal response, the response option chosen by the greatest number of participants. The findings are presented below.

Q1: Overall Ease of Use Thirteen of the eighteen participants rated the overall usability of the smartphone ‘good’ ‘very good’ or ‘excellent’. The central tendency, the mode, was six participants who agreed with the latter (see Figure 1).

Q2: Smartphone Portability Thirteen of the eighteen participants rated the portability of the smartphone ‘good’ ‘very good’ or ‘excellent’. The central tendency, the mode, was six participants who agreed that the smartphone was ‘very good’ in this respect (see Figure 2)

Q3: Comprehensibility of the Smartphone ‘User Guide’ Thirteen of the eighteen participants rated the user guide to the smartphone that was provided by the researcher to the participants as ‘good’ ‘very good’ or ‘excellent’. The central tendency, the mode, was eight participants who agreed that the user guide was ‘very good’ in this respect (see Figure 3).

99

The Use of Personal Digital Assistants in Nursing Education

Figure 1. Overall ease of use

Figure 2. Smartphone portability

Figure 3. Comprehensibility of the smartphone ‘user guide’

Q4: Device Interface Usability

Q5: Smartphone Battery Life

Fourteen of the eighteen participants rated the ease-of-use of the smartphone user interface as ‘good’ ‘very good’ or ‘excellent’. The central tendency, the mode, was six participants who agreed that the smartphone was ‘good’ in this respect (see Figure 4).

Ten of the eighteen participants rated battery life (the time between recharges) of the smartphone ‘satisfactory’ or ‘good’. The central tendency, the mode, was seven participants who agreed with the latter (see Figure 5).

100

The Use of Personal Digital Assistants in Nursing Education

Figure 4. Device interface easy to manipulate

Figure 5. Smartphone recharging frequency

Q6: Smartphone Usefulness in the ‘Clinical Area’ Ten of the eighteen participants rated the usefulness of the smartphone in the ‘clinical area’ as ‘satisfactory’ or ‘very good’. The central tendency, the mode, was six participants who agreed with the latter (see Figure 6).

Questions eight to twelve focus on the learning object itself. As has already been discussed, it is felt more appropriate to focus on the delivery technology and the practicalities of its use. This is in part because the learning object was intended for cross-professional use, rather than one which explicitly promotes interprofessional learn-

Figure 6. Smartphone use in the ‘clinical area

101

The Use of Personal Digital Assistants in Nursing Education

ing. For these reasons, the next question to be examined is number 13.

Q13: Reflection on Learning All 18 participants tended to agree that using the smartphone on clinical placement enabled them to reflect on clinical skills, with no ratings of ‘unsatisfactory’ given. The central tendency, the mode, was six participants rating the smartphone as ‘excellent’ in this respect (see Figure 7).

Q14: Smartphone Enhancement of Learning Fourteen of the eighteen participants rated the way the smartphone enhanced independent learning as ‘satisfactory’, ‘very good’ or ‘excellent’. The central tendency, the mode, was eight participants agreeing with the latter (see Figure 8).

Q15: Smartphone Facilitates Flexibility of Learning Place and Rime Fourteen of the eighteen participants rated the ability of the smartphone to facilitate flexibility of learning place and time as ‘satisfactory’ to ‘excellent’. The central tendency, the mode, was nine participants agreeing with the latter (see Figure 9).

Q16: Mobile Learning Accommodates Individual Learning Atyles Thirteen of the eighteen participants rated the ability of mobile learning to accommodate individual learning styles as ‘satisfactory’ to ‘excellent’. The central tendency, the mode, was five participants who agreed that mobile learning was ‘very good’ in this respect (see Figure 10).

Figure 7. Reflection on learning materials

Figure 8. Smartphone enhancement of independent learning

102

The Use of Personal Digital Assistants in Nursing Education

Figure 9. Smartphone facilitates flexibility of learning place and time

Figure 10. Mobile learning accommodates individual learning styles

Q17: Mobile Learning Sustains Individual Learner Motivation

Q18: Willingness to Engage in Mobile Learning Again

Thirteen of the eighteen participants rated the ability of mobile learning to sustain their motivation to learn as ‘satisfactory’ to ‘excellent’. The central tendency, the mode, was seven participants who agreed that mobile learning was ‘excellent’ in this respect (see Figure 11).

Fourteen of the eighteen participants rated their willingness to engage in mbile learning again as ‘satisfactory’ to ‘excellent’. The central tendency, the mode, was nine participants who agreed that mobile learning was ‘excellent’ in this respect (see Figure 12).

Figure 11. Mobile learning sustains learner motivation

103

The Use of Personal Digital Assistants in Nursing Education

Figure 12. Willingness to engage in mobile learning again

It is interesting to note that many students made comments that indicated they wanted more smartphone-based learning and learning objects, as the selected quotes below clearly demonstrate: “happy to do more learning supported by a mobile device in the future” “It would have been good to have all the clinical skills sessions on a device, the Aseptic technique especially”

DISCUSSION The small-scale feasibility study reported here provides evidence that smartphones pre-loaded with reusable e-learning objects are a useful and practical means by which to support students nurses on their first clinical placements. Opportunities do not always arise in the clinical area to practice certain skills, this study strongly suggests that the smartphone can act as a back up mechanism. It thus seems reasonable to conclude that nurse academics and hospital trusts can use such mobile information and communication technologies to take a more proactive approach to bridge the theory-practice gap. The students in the study felt that a smartphone pre-loaded with reusable e-learning objects facilitated independent learning. It seems plausible to imagine that their use may facilitate the phenomenon whereby students retain a higher percentage of knowledge when they learn at their own pace 104

(Nelson 2003). In addition, Students who for whatever reason had no opportunity to attend the clinical skills sessions offered will not lose out. There are however organizational and technical barriers to the large-scale adoption of PDAs or smartphones, and they will now be discussed briefly.

Change Management Moving towards supporting collaborative learning with PDAs and smartphones requires consideration of not only the student’s learning needs, but also those of their tutors and mentors, as well as the circumstances of their clinical placement. It must be remembered that clinical Staff are not always technical experts, and they are primarily concerned with providing patient care under tight time constraints. Imposed changes in work habits may be met with resistance from staff and students. It is important that these barriers are considered before any introduction of handheld devices in the clinical area. One possible barrier to student nurses’ use of PDAs is lack of knowledge of how to use them. While the majority of the participants in the feasibility study reported in this chapter owned a mobile telephone, not all individuals have grown up with mobile technology, for example, students returning to study aged over forty years. It could also be argued that the nursing culture has not encouraged nurses toward information and communication technology use either. Furthermore, training that enables students to use new technology takes time; time that is taken away from patients.

The Use of Personal Digital Assistants in Nursing Education

For PDAs to be successfully integrated into their education, it should be explained to student nurses why they should be using them, what they can offer, which models to buy, and how to operate their PDA applications. The value of introducing mobile computing devices to students early in their education has been recognised for some time (Thomas et al., 2001). Nurses are skilled in information management, routinely collecting and organizing data. Tutors can explain that by using a PDA that has relevant electronic resources uploaded onto it, nurses will find answers in their hands, not at the nurses’ station. It can be explained that PDAs can bring benefits such as increased productivity, reduced errors, better patient care, and increased worker satisfaction (Huffstutler et al., 2002). All nurses need to network with other nurses to share information, to assist in research, provide and receive mentoring, emotional support, and friendship. PDAs can play a role in many of these areas, thus their adoption can be justified in these terms. PDAs can also be a means by which to help health care staff stay abreast of new information about clinical skills and support mentors in their teaching role. It follows that the information on the PDA will need continually updating, as evidence based research changes as fast as technology. However, it is possible that, nurses can support PDA development by creating their own learning resources. The creation of learning objects that meet the needs of nursing students on placement and reflects and supports previous learning in clinical skills laboratories can itself be incorporated into the curriculum as a reflective learning exercise.

FUTURE DEVELOPMENTS The feasibility study reported in this chapter involved a small sample of eighteen students, due to only twenty PDA s being available. While sufficient to demonstrate the feasibility of PDAs on clinical

placement, however, it is too few for a thorough evaluation to be made and firm conclusions to be reached. Furthermore, the participants put themselves forward on the basis that they had an interest in using information and communications technologies. It is quite possible that those who chose to participate were eager to have access to as much help as possible, whereas those who did not may have been too anxious about using new technology. Further action research can shed light on these questions, which has implications for the possibilities of scaling-up the use of PDAs in nurse education. The next developmental step would be to develop and evaluate a core online repository of PDA-ready reusable e-learning objects to satisfy the needs of students on placement. Of particular interest would be the value of this approach as a means to support the development of interprofessional working in students on placement. This would require a set of reusable e-learning objects that explicitly promote and explore interprofessionalism, rather than a genric skills object that is essentially cross-professional rather than interprofessional in nature. On the other hand, cross-professional e-learning objects can form the basis of interprofessional learning if used as discussion triggers between students from different professional groups, for example, nurses, medics, occupational therapists, dieticians and social workers. The inherent potential for communication through smartphones with Wi-Fi internet capability opens interesting possibilities in this regard. For example, online discussion boards are already available to the students of the university where the feasibility study was conducted through the interprofessional learning module that all healthcare students, regardless of their professional pathway, must undertake.

CONCLUSION Student and practicing nurses are likely become more familiar with handheld technology in their

105

The Use of Personal Digital Assistants in Nursing Education

daily lives as these technologies become more ubiquitous in society, and such devices are likely to become more prevalent in the workplace. In the near future, it is thus likely that nurses will be increasingly open to the use of handheld technology to enhance nursing research and nursing practice. PDAs open the possibilities for effectively delivering educational material that is always available at the point of need, regardless of an individual’s physical location, as long as its use does not compromise infection control. The clinical mentor is supported during busy times, as students can use this as an intermediate measure. Learning can take place at any time and place, with one or multiple users, providing great flexibility and ensuring work coverage. The results of the small feasibility study described in this chapter indicate that most of the students participating in the study were comfortable with this approach to mobile learning, finding it useful and calling for greater use of PDAs. The findings demonstrate that this approach is a feasible way to enable universities to meet students’ individual needs to be on-demand while they are on clinical placement.

ACKNOWLEDGMENT The feasibility study that is reported in this chapter was funded as a small research project by the Centre of InterProfessional e-learning (CIPeL).

REFERENCES Aston, L., & Molassiotis, A. (2003). Supervising and supporting student nurses in clinical placements: the peer support initiative. Nurse Education Today, 3(3), 202–210. doi:10.1016/ S0260-6917(02)00215-0

106

Ayliffe, C. A. J., Lowbury, E. J. L., Geddes, A. M., & Williams, J. D. (1992). Control of Hospital Infection: A practical handbook (3rd ed.). London: Chapman, and Hall Medical. Criswell, D. F., & Parchman, M. L. (2002). Handheld computer use in U.S. family practice residency programs. Journal of the American Medical Informatics Association, 9(1), 80–86. De Groote, S. (2004). The use of personal digital assistants in the health sciences: results of a survey. Medical Library association, 92(3), 341-349. Fischer, S., Stewart, T. E., Mehta, S., Wax, R., & Lapinsky, S. E. (2003). Handheld computing in medicine. Journal of the American Medical Informatics Association, 10(2), 139–149. doi:10.1197/ jamia.M1180 Gould, D., & Drey, N. (2008). Hand hygiene technique. Nursing Standard, 22(34), 42–46. Huffstutler, S., Wyait, T. H., & Wright, C. R. (2002). The use of handheld technology in nursing education. Nurse Educator, 27(6), 271–275. doi:10.1097/00006223-200211000-00008 Koeniger-Donohue, R. (2008). Hand held computers in nurse education: a PDA pilot project. The Journal of Nursing Education, 47(2). doi:10.3928/01484834-20080201-01 Murphy, E. A. (2005). Handheld computers software for school nurses. The Journal of School Nursing, 21(6), 56–360. doi:10.1177/10598405 050210061101 Nelson, E. (2003). A Practical Solution for Training and Tracking in Patient-care Settings. Nursing Administration Quarterly, 27(I), 29. Stolworthy, Y. & Suszka-Hildebrandt, S. (2000). Mobile information technology at the point-ofcare: the grass roots origin of mobile computing in nursing. PDA Cortex 2000 [serial online].

The Use of Personal Digital Assistants in Nursing Education

Thotnas, B. A., Coppola, J. F., & Feldman, H. (2001). Adopting handheld computers for community-based curriculum: Case study. The Journal of the New York State Nurses’ Association, 32(1), 3–6. Timmins, F., & Kaliszer, M. (2002). Aspects of nurse education programmes that frequently cause stress to nursing students -- fact-finding sample survey. Nurse Education Today, 22(3), 203–211. doi:10.1054/nedt.2001.0698

Yonge, O., Krahn, H., Trojan, L., Reid, D., & Haase, M. (2002). Being a preceptor is stressful! Journal for Nurses in Staff Development, 18(1), 22–27. doi:10.1097/00124645-200201000-00005 Young, P. M., Leung, R. M., Ho, L. M., & McGhee, S. M. (2001). An evaluation of the use of hand-held computers for bedside nursing care. International Journal of Medical Informatics, 62(2-3), 189–193. doi:10.1016/S1386-5056(01)00163-0

White, A., Allen, P., Goodwin, L., Breckinridge, D., Dowell, J., & Garvy, R. (2005). Infusing PDA technology into nursing education. Nurse Educator, 30(4), 150–154. doi:10.1097/00006223200507000-00006 This work was previously published in Interprofessional E-Learning and Collaborative Work: Practices and Technologies, edited by Adrian Bromage, Lynn Clouder, Jill Thistlethwaite and Frances Gordon, pp. 336-351, copyright 2010 by Information Science Reference (an imprint of IGI Global).

107

108

Chapter 1.7

Reforming Nursing with Information Systems and Technology Chon Abraham College of William and Mary, USA

INTRODUCTION Much of healthcare improvement via technology initiatives addresses gaining physician by-in (Reinertsen, Pugh & Bisognano, 2005) and does not adequately address engaging nurses, despite the fact that nurses serve as the front-line caregivers and are a primary user group (Wiley-Patton & Malloy, 2004). However, the tide is changing, and visibility of nurses as information gatherers and processors in the patient care process is increasing (Romano, 2006). Nurses perform the majority of

the data-oriented tasks involved in patient care and would benefit most from having access to information at the point of care (Bove, 2006). RADM Romano, Chief Professional Officer of Nursing and advisor to the U.S. Surgeon General concerning public health, recently addressed the American Informatics Nursing Informatics Association and stated, “This is the year of the nurse. The technologies that have the means to improve the efficiencies in patient care are in the hands of the nurses.” Nurses need to embrace technology in everyday work or continue suffering the consequences of antiquated methods of computing

DOI: 10.4018/978-1-60960-561-2.ch107

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Reforming Nursing with Information Systems and Technology

that take us away from where we work—at the point of care (Abbott, 2006). Healthcare can benefit greatly from use of a diverse set of technologies as facilitators of change to improve quality and safety of patient care by decreasing errors made because of the lack of faster, more comprehensive, and more accessible patient documentation at the point of care (IOM, 2000). Healthcare institutions abroad share similar sentiments according to international healthcare reports by the International Council of Nurses (Buchan & Calman, 2004). In light of these concerns and a severe U.S. nursing shortage (i.e., an estimated 400,000 shortage by 2020) (Bass, 2002), institutions are beginning to consider employing point-of-care IT as a means to promote patient safety, decrease medical errors, and improve working conditions for overtaxed nurses (TelecomWeb, 2005). This chapter provides an overview of nursing reform and novel ubiquitous computing integrating voice, hands-free, and mobile devices being researched or used to make nursing more efficient and promote the following: •

• •

•

•

A decrease in laborious documentation aspects (i.e., decrease problems with access to information at the point of care, written errors or legibility issues precluding comprehension of medical regimens). An increase in recruitment and retention rates. Improvement in communication in which nurses are required to consolidate and process information during care. Promotion of involvement in systems analysis and design of information systems to increase the likelihood of technology acceptance. Baseline and continuing education in both professional training and on-the-job training to allay technology aversion and build computer self-efficacy.

•

•

•

Development of standards for electronic documentation of patient interaction and processes for nursing that can be codified. Identification of role changes in the nursing community because of the use of information systems and technologies. More rigorous research concerning information research in the nursing community.

BACKGROUND This chapter provides an overview of tactics to reform nursing sponsored by leading nursing healthcare information systems research in professional organizations such as the American Nurses Informatics Association (ANIA), Association of Medical Informatics (AMIA), and the International Council on Nursing (ICN). The sociotechnical systems framework (STS) (Bostrom & Heinen, 1977) will be employed to categorize tasks, technologies, people involved, roles, and structure of aspects of the reform. STS emphasizes workplace interactions with various technologies and is espoused as a realistic view of organizations and a way to change them (Bostrom & Heinen, 1977). STS concerns the social system that is comprised of contributions of people (i.e., attitudes, skills, and values), the relationships among people and their roles in the organization, reward systems, and authority structures (Bostrom & Heinen, 1977). STS is also an intervention strategy in which technology implementations intervene within a work system to improve task accomplishment, productivity, and work quality of life, and to prompt supportive organizational structural changes. Innovative technologies may improve task efficiency and task effectiveness by automating or reengineering antiquated/manual processes, changing people’s roles, and making organizational structures (Bostrom & Heinen, 1977). A graphical depiction of STS is displayed in Figure 1.

109

Reforming Nursing with Information Systems and Technology

Figure 1. STS (Bostrom & Heinen, 1977)

DESCRIBING NURSING REFORM USING AN STS FRAMEWORK STS is used as a frame to describe elements addressed in the nursing reform initiative, such as tasks, technology, people involved, changes in roles, and structure in conjunction with their interplay, outputs, and goals.

Task The tasks in this reform are indicative of the pervasive problems plaguing nursing and healthcare in general. There is a critical nursing shortage crisis in the United States alone in which more than 90,000 nursing position vacancies were reported in 2004, and nearly 400,000 are expected by 2020, which has grave impacts on quality of care (JCAHO, 2004). Similar statistics reported by the International Council of Nursing (ICN) also highlights the great supply and demand disparities. Much of this is attributed to the problems recruiting new nurses and retaining seasoned ones. Nursing is an aging workforce in which the following is reported:

110

• • • •

One in five registered nurses plan to leave the profession. 81% say morale is extremely low. 64% do not have enough time to spend with the patient. 60% note the manual paperwork burden to document information about patient care tasks is “a necessary evil.”

The last percentage indicates the manual paperwork burden, which is one problem that can be directly addressed via application of novel information systems (IS) and information technology (IT). Information-oriented patient care tasks for which nurses are primarily responsible entail the following: •

•

Triage: The process of taking vital statistics, documenting impending problems, and making initial assessments to categorize criticality of the emergent condition. Charting: The process of monitoring the patient and recording vital statistics, effectiveness of medical intervention, and overall patient wellness, which is an ongoing process across nurse shifts.

Reforming Nursing with Information Systems and Technology

•

•

Throughput flow: The process of managing when, where, and how patients are moved about in a patient care system across units within the hospital for labs, procedures, and so forth. Medication administration: The process of validating medication regimens with prescriptions, delivering the medication, and documenting the effectiveness.

Virtually 98,000 people die each year due to medical errors in hospitals (Institute of Medicine, 2000), and many of these are likely due to antiquated methods of information management and delivery during these fundamental patient care tasks. These errors can also be attributed to the inability of personnel relying on the information documented by nurses during these tasks to actually access it in a timely fashion. In essence, not having access to pertinent information when and where needed cripples the abilities of healthcare professionals such as nurses from validating medical information and diagnosing properly. So the primary task for reform becomes automating manual charting and equipping nurses, who are highly mobile in their work environment and move from patient to patient, with devices that enable easy and unobtrusive means of accessing and documenting needed information during patient care.

Technology The reform in this industry recognizes that the paperwork burden is one where technology can be used as a facilitator of change or as part of the intervention strategy to promote change. Therefore, the goal is to promote productivity in which nurses are able to decrease errors associated with manual charting and reduce charting time via easy non-time-consuming data input and output capabilities. These improvements enable a nurse to spend more time with patients, which is a stated desire among nurses (Abraham, Watson, Boudreau

& Goodhue, 2004). AMIA and ANIA encourage partnerships among vendors, academic researchers, and nurses in practice to discern technologies that are vehicles for promoting their workflow efficiencies. Figure 2 provides an overview of these technology categories. One aspect of technology that is specific to the healthcare industry and not akin to the average mobile business environment is the need to protect patients and caregivers from the transfer of infectious diseases or bacteria introduced by mobile devices. The typical mobile device is assigned to one nurse who carries it throughout his or her shift. Therefore, the nurse uses the same device/ technologies across every medical intervention in which the likelihood of bacterial transfer is high. Thus, antimicrobial protective coatings or materials on healthcare mobile devices are of utmost importance in order to avoid transfer of bacteria from one patient environment to another, or from one nurse to another as the devices turn over during a shift change.

People Nursing in general is a highly caring and nurturing profession (Patistea, 1999). Nurses have chosen their profession based on their individual desires to care for others and promote the sanctity of human life. Nurses, for example, who are satisfied in their jobs are characterized by high levels of affective empathy (i.e., the ability to be moved by other people’s feelings) and service-mindedness (i.e., a strong tendency to let other people be central and dominant) (Sand, 2003). In addition, they display a high degree of sensitivity to nonverbal communication and a strong need for affirmation, support, and dominance (Sand, 2003). The traits prioritize patient care tasks for servicing nurses. For example if a patient is exhibiting physiological problems associated with anxiety or is in pain, the primary task becomes physical care of the patient. Additionally, since nurses do display a strong need for affirmation from their patients,

111

Reforming Nursing with Information Systems and Technology

Figure 2. Technology overview

112

Reforming Nursing with Information Systems and Technology

peers, and superiors, the technologies need not interfere with the desire to establish mutual trust with the patient and the need for nurses to exhibit confidence in use of the technology, especially in the presence of the patient. The clinical nursing population can be divided into two demographics: A and B. Demographic A is an average age of 40, has work experience of 15 or more years, lacks professional computer training that is a systemic problem recognized by servicing IS personnel and nursing managers (Bass, 2002), and has low computer selfefficacy (Abraham et al., 2004). Demographic B is characterized by an average age of 25, has less than seven years experience, was trained under a “new” regimen that introduced him or her to PC-based documentation, and exhibits medium to high computer self-efficacy (Abraham et al., 2004). Both demographics use traditional PCs for other tasks, such as monitoring vital statistics or maintaining an apparatus for intravenous drug administration. However, information systems for electronic charting or mobile devices that provide access to these information systems are revered initially as foreign to this population (Abraham, forthcoming). The majority of nurses in Demographic A do not type effectively but regard narrative oriented data entry as problematic (Abraham et al., forthcoming). Additionally, the personality traits of nurses influence how they accept and use technology (Abraham et al., 2004). In essence, intended utilization for mobile technologies is for the nurse to use at the point of care and, most likely, in the presence of the patient. Therefore, the technology should not impede nurses from being able to maintain eye contact nor interfere with the interaction with their patients. Additionally, healthcare workplaces are very tumultuous, hectic, and characterized by a great deal of uncertainty, which contributes to stressful work environments. Thus, the technologies need to be designed in a manner that are easy enough not to excite anxiety among the nurse user group that causes nurses to fumble with the

technology, display a lack of confidence, or inhibit nurse/patient interaction (e.g., having to pay more attention to the computer because of ill-designed interfaces) (Abraham, forthcoming). Nurses feel that their lack of confidence concerning operating the technology is interpreted by patients as incompetence not only with the technology but with performing their required patient care tasks (Abraham et al., 2004). Not only does the technology need to support the way nurses work, but it also needs to support the way nurses think about servicing their patients. Also, adequate computer training, including how to most appropriately use the mobile technologies as vehicles to support the patient/nurse relationship, is of utmost importance. Physical care of the patient will always overcome the need to document the intervention until the hands-on care has been performed. However, providing nurses with easy unobtrusive access to medical information when and where needed enables nurses to make better well-informed decisions.

Interplay Between Task, Technology, and People in Healthcare Environments Mobile technologies make new dimensions salient that are driving novel interplay between the task, technology, and people in healthcare environments. For example, traits of the nurse concerning the need to attain affirmation or appear competent in front of the patient while operating the technology is novel since prior to mobile computing the act of operating the computer most likely was done away from the point of care (i.e., at the nurses’ station). When computing using technologies that were tethered, nurses were less likely to exhibit as much anxiety about using the system in front of the patient (Abraham, forthcoming). The interplay among task, technology, and people in healthcare environments has become more pronounced because of changes in computing methods that

113

Reforming Nursing with Information Systems and Technology

enable access to information when and where needed via wireless technologies. Nursing reform addresses initiatives in the development of technologies not only to better support the way nurses actually work, but also to develop information systems based on a unified nursing language known as the International Classification for Nursing Practice (ICNP). It is an arduous task to implement compositional terminology for the nursing practice since nurses affectionately note that each individual “nurses” differently (i.e., they label common patient care functions differently) (ICN, 2006). The goal of the ICNP is to provide a global infrastructure informing healthcare practice and policy to improve patient care via codifying and retaining tacit, experiential knowledge to inform the practice and promote easier cross-training, especially in light of the aging nurse workforce and the problems with recruitment and retention (ICN, 2006). In light of these globally recognized nursing shortages, persons with either formalized nursing training or interest in the field from foreign countries have been recruited and trained to fill many vacant positions (Bass, 2002). ICNP hopes to streamline the learning curve in order for these foreign nurses to become acclimated to the nursing profession. Additionally, as people relocate, which is increasingly the case, their patient information should move or be automatically accessible despite the physical location of the patient (Council on Systemic Interoperability, 2006). However, the limiting factor is the lack of systemic interoperability within the United States and globally to make this idea a reality. ICNP focus is to ensure that at least care providers speak a common language, which is documented in the patient’s medical record that patients transport when they relocate or acquire services from a multitude of different healthcare providers. Therefore, ICNP and systemic interoperability projects are addressing the interplay between fulfilling the task with suitable information systems and technologies to support the people who are both task doers and

114

task recipients in this field in order to promote better quality of care—the ultimate goal.

Roles and Structure These novel technologies and initiatives in improving the communication among nurses and about what they actually provide as value-added effort helps to better define and highlight the crucial role of nurses. By virtue of these technologies being placed in the hands of nurses, it is indicative of the technology being a facilitator of change to realize the role of the nurse as instrumental to the patient care process since they do perform the bulk of documentation for any medical intervention. Contributing to realization of the criticality of their role is the nurse’s ability to recognize in an easier and more timely fashion the potential errors in the prescribed regimen, such as medication prescriptions that may have been prescribed erroneously or entered into the system mistakenly by a physician but validated by the pharmacist. In this sense, having access to IT and IS empowers nurses to at least bring to light problems they discern regarding the prescribed regimen for a patient. Typically, nurses that are assigned to inhospital patients interface with their patients more frequently than any other medical personnel; they are very familiar with their patient’s regimen and have the astuteness to recognize when there may be a potential error in the prescribed regimen. For example, the servicing nurse may discern allergic reactions to the prescribed medication or an ineffectiveness of the medication requiring a physician’s order to be clarified or modified. Having access to information coupled with efficient means of communicating with other servicing medical personnel such as a patient’s physician allows the nurse to discern the validity of the regimen or prescription in question. In essence, having access to information when and where needed and having the ability to communicate effectively because of the aforesaid novel technologies enables the nurse to appear more competent and thus gain

Reforming Nursing with Information Systems and Technology

more respect for his or her level of knowledge and critical role in the ability of the patient to be serviced properly. As far as structural impacts, nurses contend that they have been somewhat neglected and overlooked as principal components in the care process, which contributes to poor morale and self-efficacy issues (Abraham et al., 2004; Bass, 2002). This role of the nurse as being critical in patient throughput needs to be adequately articulated to the population, management, and associated physicians (Abraham et al., 2004). With the dire status of the industry and the grave projections of the impact on patient care, more and more healthcare organization administrators recognize the need to establish a vision of change for nurses, articulate how they fit into the care process, and explain why it is important for nurses to accept the technologies that enable efficiencies. Bringing physicians into the fold to internalize and articulate this message will also undoubtedly promote acceptance of the technology among the nursing population since they have a strong need for external affirmation. The vision needs to manifest into useful but nonobtrusive IT and IS training for nurses in order to improve computer self-efficacy and increase the likelihood that the technology will be used as intended to produce desired efficiencies. Nurses commented that the majority of the physicians they work with do not appreciate their services, which contribute to the nurses rejecting many technologies that management presents to them because they feel that the technologies are being thrust upon them in an effort to make the physicians’ jobs less demanding and not for improving nursing work conditions (Abraham et al., 2004). The technology can be the main facilitator to empower nurses as the principal information managers in patient care. One area of future research is to explore more thoroughly how technology is facilitating a change in the authoritative structure between physicians and nurses. The hope is for physicians to see nurses as more competent and capable because of the

ability to more readily address questions of them made possible with the use of IT and IS (Abraham et al., 2004).

FUTURE TRENDS A future trend is the exploration of nursing decision support systems comparable to those developed for physicians, but with a focus on nursing-related patient care tasks that have elements of diagnoses. Another area for future research is the incorporation of project management skills and facilitating IS and IT into nursing curricula in order to aid nurse managers and clinical nurses with a means for balancing patient loads with fewer nurse resources.

CONCLUSION The primary contribution of this chapter is to promote visibility of the needs of the nursing community regarding electronic information exchange, manipulation, and management that can be addressed in healthcare information systems research. Typically, IS research focus groups are geared toward physicians; however, nurses are responsible for the bulk of documentation for each medical intervention in each step of the patient care system. Physicians rely heavily on the information supplied by nurses concerning the historical documentation of charting patients’ ailments, effectiveness of prescribed regimens, and anomalies in their condition (Abraham et al., 2004). Focus has been placed on using technology to streamline processes and automate as much as possible to avoid human documentation error. However, healthcare IS research has frequently underexplored nurses or not recognized them as the principal personnel who supply the information to the system, whether the system is manual or automated. IS researchers can help to uncover nuances in the nature of nursing and design IT and IS to better support the way nurses work in

115

Reforming Nursing with Information Systems and Technology

hopes of helping our society achieve its goal of improving the quality of patient care.

REFERENCES Abbott, P. (2006). Innovations beyond the lab: Putting information into the hands of those who care. In Proceedings of the 2006 American Nursing Informatics Association Conference, Nashville, Tennessee. Abraham, C., Watson, R. T., & Boudreau, M. C. (forthcoming). Ubiquitous access: Patient care on the Frontlines. Communications of the ACM. Abraham, C., Watson, R. T., Boudreau, M. C., & Goodhue, D. (2004) Patient care and safety at the frontlines: Nurses’ experiences with wireless computing. Publication Series for IBM Center for Healthcare Management American Nurses Informatics Association. (2006). Description for Bayada Award for technological innovation in nursing education and practice. Retrieved from http://www.ania.org/Bayada.htm Bass, J. (2002). It’s not just nurses anymore: Survey of three other health occupations finds patient safety at risk due to understaffing. Retrieved from http://www.aft.org/press/2002/041102.html Bostrom, R., & Heinen, J. (1977). MIS problems and failures: A socio-technical perspective. Part I: The causes. MIS Quarterly, 1(3), 17–32. doi:10.2307/248710 Bove, L. (2006). Implementing a system with clinical transformation in mind. In Proceedings of the 2006 American Nursing Informatics Association Conference, Nashville, Tennessee. Buchan, J., & Calman, L. (2004). The global shortage of registered nurses: An overview of issues and actions. International Council of Nurses. Retrieved from http://www.icn.ch/global/shortage.pdf

116

Council on Systemic Interoperability. (2006). Interviews from practitioners speaking about ending the document game in U.S. healthcare. Retrieved from http://endingthedocumentgame. gov/video.html Institute of Medicine (IOM). (2000). To err is human: Building a safer health system. The National Academic Press. Retrieved from http://www.nap. edu/books/0309068371/html/ International Council of Nurses. (2006). International classification for nursing practice (ICNP) fact sheet. Retrieved from http://www.icn.ch/ icnp.htm Joint Commission on Accreditation of Healthcare Organizations (JCAHO). (2004). Healthcare at the crossroads: Strategies for addressing the evolving nursing crisis. Retrieved from http://www. jointcommission.org/NR/rdonlyres/5C138711ED76-4D6F-909F-B06E0309F36D/0/health_ care_at_the_crossroads.pdf Patistea, E. (1999). Nurses’ perceptions of caring as documented in theory and research. Journal of Clinical Nursing, 8(5), 487–495. doi:10.1046/ j.1365-2702.1999.00269.x Reinertsen, J., Pugh, M., & Bisognano, M. (2005). Seven leadership leverage points for organization-level improvement in healthcare. Retrieved from http://www.delmarvafoundation.org/html/ culture_club/docs/Seven%20Leadership%20 Leverage%20Points.pdf Romano, C. (2006). Improving the health of the nation: The promise of health information technology. In Proceedings of the 2006 American Nursing Informatics Association Conference, Nashville, Tennessee. Sand, A. (2003). Nurses’ personalities, nursesrelated qualities and work satisfaction: A ten year perspective. Journal of Clinical Nursing, 12, 177–187.

Reforming Nursing with Information Systems and Technology

TelecomWeb. (2005). U.S. healthcare providers seen growing IT budgets by over 10 percent. Retrieved from http://www.telecomweb.com/ news/1086273449.htm Wiley-Patton, S., & Malloy, A. (2004). Understanding healthcare professionals’ adoption and use of IT. In Proceedings of the Tenth Americas Conference on Information Systems, New York.

KEY TERMS AND DEFINITIONS Charting: The process of monitoring the patient and recording vital statistics, effectiveness of medical intervention, and overall patient wellness, an ongoing process across nursing shifts. Electronic Medical or Health Records: Electronic version of the manual chart that outlines historical medical information pertaining to one individual. Medication Administration: The process of validating medication regimens with prescriptions, delivering the medication, and documenting the effectiveness. Mobile or Wireless Computing: The use of mobile, untethered devices to access information

stored in information systems or communicate via wireless networks. Nursing: Encompasses autonomous and collaborative care of individuals of all ages, families, groups, and communities, sick or well and in all settings. Nursing includes the promotion of health, prevention of illness, and the care of ill, disabled, and dying people. Advocacy, promotion of a safe environment, research, participation in shaping health policies and patient and health systems management, and education are also key nursing roles. Systemic Interoperability: Comprehensive network of privacy-protected systems of electronic personal health information developed around the healthcare consumer to facilitate wellness and the safe delivery of optimal healthcare. Throughput Flow: The process of managing when, where, and how patients are moved about in the patient care system across units within the hospital for labs, procedures, and so forth. Triage: The process of taking vital statistics, documenting impending problems, and making initial assessments to categorize criticality of the emergent condition.

This work was previously published in Encyclopedia of Healthcare Information Systems, edited by Nilmini Wickramasinghe and Eliezer Geisler, pp. 1137-1145, copyright 2008 by Medical Information Science Publishing (an imprint of IGI Global).

117

118

Chapter 1.8

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids Anastasia N. Kastania Biomedical Research Foundation of the Academy of Athens, Greece & Athens University of Economics and Business, Greece Sophia Kossida Biomedical Research Foundation of the Academy of Athens, Greece

ABSTRACT The electronic healthcare in the modern society has the possibility of converting the practice of delivery of health care. Currently, chaos of information is characterizing the public health care, which leads to inferior decision-making, increasing expenses and even loss of lives. Technological progress in the sensors, integrated circuits, and the wireless communications have allowed designing low cost, microscopic, light, and smart sensors. These smart sensors are able to feel, transport one

or more vital signals, and they can be incorporated in wireless personal or body networks for remote health monitoring. Sensor networks promise to drive innovation in health care allowing cheap, continuous, mobile and personalized health management of electronic health records with the Internet. The e-health applications imply an exciting set of requirements for Grid middleware and provide a rigorous testing ground for Grid. In the chapter, the authors present an overview of the current technological achievements in the electronic healthcare world combined with an outline of the quality dimensions in healthcare.

DOI: 10.4018/978-1-60960-561-2.ch108

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

INTRODUCTION

BACKGROUND

E-health offers an optimistic vision of how computer and communication technology can help and improve healthcare provision at a distance. Health care provision is an environment that has presented remarkable improvement in the computing technology. For example, mobile e-health includes information and telecommunications technologies, which provide health care to patients that are at a distance from the supplier, and provide reinforcing tools for mobile health care. Recently, health care has embraced the mobile technology in electronic applications (Panteli et al., 2007). Various initiatives (public and private) have examined the different mobile applications of electronic health focusing on the mobility of doctors, the mobility of the patients, up to the Web based data access (Germanakos et al., 2005). Moreover, personalized electronic health care provision through autonomous and adaptive Web applications is noteworthy (American College of Physicians, 2008). The growth of high bandwidth wireless networks, such as GPRS and UMTS, combined with miniaturized sensors and the computers will create applications and new services that will change the daily life of citizens. The citizens will be able to get medical advice from a distance but will also be in a position to send, from any place, complete and accurate vital signs measurements (Van Halteren et al., 2004). However, grid technologies and standards, currently examined in health care, will be adopted if they prove that they face all the valid concerns of security and follow the ethical guidelines (Stell, et al., 2007). In this chapter, an overview of the existing personalized, mobile and Grid applications in electronic healthcare is presented. Further, quality aspects, which have successfully been applied in traditional health care quality assessment, are presented. A future challenge is to examine how these quality aspects can be successfully applied in electronic healthcare.

To realize the potential of electronic health, future electronic health should be able to support patient information management and medical decision support in an open and mobile medical environment. Such an environment will be strong in knowledge, sensitive and flexible in the needs of patients, and will allow collaboration of virtual teams that work in different geographic areas. Context-specific services to each individual are defined as personalized services. These are provided by agents usually with autonomy playing the role of personal assistant (Panayiotou & Samaras, 2004; Delicato et al., 2001). Personalization is a technique used to explore both interaction and information. It allows doctors to search personalized information about the health state of each patient (Nikolidakis, et al., 2008). Since the number and the use of wireless connection and portable devices are increasing, the complexity of designing and setting up adaptive e-health systems is also increasing. The current improvement in physiologic sensors allows realizing future mobile computing environments that can impressively strengthen health care that is provided to the community, and individuals with chronic problems. Finally, there is interest from the academic and industrial world to assess the challenge of e-health automation management of mobile Grid infrastructures (Lupu et al., 2008).

PERSONALIZED E-HEALTH The current tendency towards ubiquitous computing, the new sensor technologies, the powerful mobile devices and the wearable computers support different types of personalized electronic health applications. Telemonitoring applications have adopted this new technology to improve the quality of care and the quality of treatment for the sick and the elderly using questions and actions based on user preferences. The personalized application

119

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

collects user activities and begins interpreting them to act according to user’s wishes. The study of user activities also allows individualization of content. Reproduction of user experience, emotions, sentiments, thoughts, behavior and their actions is also performed (Lugmayr et al., 2009). The textile products and computers are combined to create a recent model in the individualized mobile data processing. To create a programmable device, as part of clothes, design and implementation of a ‘wearable’ motherboard architecture is needed to embody clothing items, hardware and software. This type of information processing should not only provide high bandwidth but should also have the capability of seeing, feeling, thinking, and acting (Park et al., 2002). A picture of remote health management supports the use of an individual cell phone, as a

mediator that transfers multiple data streams from multiple wearable sensors in a response infrastructure (Mohomed et al., 2008). Autonomous and adaptive Web applications encourage individual ‘participation’ in the medical decisionmaking and self-management (American College of Physicians, 2008). The personalization is based on adapting information content that is returned as an answer to a user request based on who is the end user (Figure 1). Internet-based applications allow individuals to shape the application based on their individual needs because patients require timely access to high quality personalized health care (Nijland et al., 2009). Using websites of health risk calculation, the users of the Internet can receive personalized predictions for their health (Harle et al., 2009). Life assistance protocols, determined by medical experts, are based on

Figure 1. Current aspects involved in personalized e-health

120

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

evidence-based medicine and have been proved acceptable practice that is applied in one population group. Developing a life assistance protocol in different user groups will provide, in the long run, enormous volumes of information on health risk, and documentation on different healthcare paths (Meneu et al., 2009). Moreover, it will allow creating diverse and personalized snapshots and adaptations of the life assistance protocol (Meneu, et al., 2009). The electronic health record (EHR) used in health administration reports and in clinical decision-making is an electronic record of patient history produced from personal health monitoring (Dong et al., 2009). Case-based reasoning has been used for the individualized production and delivery of health information to extract a personalized health record from the individual health profile (Abidi, 2002). Transport-independent personal-health data and protocol standards are defined by the ISO 11073/IEEE 1073 (known also as X73) for interoperability with the different sensors (Yusof et al., 2006; de Toledo et al., 2006). Finally, an emerging opportunity for artificial intelligence researchers is to develop technologies that would check continuously ‘healthy’ in their houses and constantly direct them to healthy behavior (Intille et al., 2002). The use of intelligent techniques to analyze user satisfaction ensures the medical accuracy of the dynamically created patient empowerment form (Abidi et al., 2001). Overall, the term context-awareness refers to the ability to obtain context information that describes the user environment and adjusts according to this context. Such an adaptation clearly leads to individualized services, but the sensitive information handling causes various privacy concerns (Chorianopoulos, 2008). Context-awareness is a promising model for flow control. The context is any information that characterizes the environment and the state of the user (for example setting and time) and any other component relative to the interaction between the user and the service (Wac et al., 2006). The personal user data and preferences

are the main ‘exploitation points’ for finding ways to add value to user experience. Therefore, essential areas of future e-health research should be personalization, the exchange of secure messages and the mobile services incorporating genetic information in electronic health applications (Carlsson & Walden, 2002). This will improve the medical practice, research and the convergence between the bioinformatics and the medical informatics (Oliveira et al., 2003). Integrating enormous sums of genetic information in the clinical setting will ‘inspire’ modern clinical practices where the clinical diagnosis and treatment will be supported by information on molecular level. Gene therapy provides the opportunity of hereditary illnesses minimization while molecular medicine requires an increasing exchange of knowledge between the clinicians and the biologists. The information that is available in biology and in medicine is so impressive that requires specific tools for information management. However, tools are needed to effectively connect the genetic and clinical information, electronic health applications and existing genetic databases. The usage of evolutionary algorithms opens many prospects for treatment optimization to combine biological and medical traits in the decision support.

Mobile Health In the mobile health world, we conduct clinical data collection, healthcare information delivery, direct provision of care and real time monitoring of vital signs using mobile devices and other wireless devices. Moreover, in order to ensure that we deliver care of high quality at any time and space, access to patient records is necessary as well as additional information from distant sources such as consultation with experts and direct contact with databases. The mobile, wireless technology can offer this support but there are few medical devices on the market, which can be adapted and personalized according to patient needs (Riva,

121

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

2003; Joshi, 2000). Finally, classifying medical devices is critical because their effects can fluctuate from no threat to life threat. On the other hand, we might integrate electronic health services in cell phones and connect biosensors with cell phones (Han et al., 2007). The necessary information is organized to establish an ontology for mobile electronic health services (Han, et al., 2007). Furthermore, mobile health applications introduce new mobile services in health care based on the technologies 2.5 (GPRS) and 3G (UMTS). This can be achieved by integrating sensors and activators to continuously measure and transmit vital signs, with sound and video, to suppliers of health care in a Wireless Body Area Network (BAN) (Konstantas et al., 2002). These sensors and the activators are improving the life of patients while introducing new services in disease prevention and diagnostics, remote support, recording of normal conditions and the clinical research. The Wireless Body Area Network will help the fast and reliable implementation of distant-aid services for sending reliable information from the place of accidents. Finally, the ubiquitous computing is a promising model for creating information systems. Databases improve data management for the patients, public health, drugstores and the workforce (Milenković et al., 2006). The field of mobile healthcare requires continuous planning, solutions, decision-making and intelligent support technology that will reduce both the number of faults and their range. In our wireless world, the mobile devices use various networking installations. The existing context aware applications are benefited from the user context (for example positioning information), nevertheless, few question the network quality of service (Wac, 2005; Broens et al., 2007; Hanak et al., 2007). However, the services of electronic health share several distinctive features concerning the structure of services, the component services and the data requirements. Therefore, the ideal is to

122

develop electronic health services in a common platform using standard features and services. Finally, various fundamental questions can affect the successful implementation of a mobile electronic health application. However, portable wireless applications at the point of care, based on information and connected with the clinical data, can reduce these problems. Interoperability of the existing applications of health care and/or databases is essential. The mobile technology of health care has the solution on not only supporting health care but also on easier management of mobile patients. The mobile applications can reduce faults and delays in creating accessible digital data (Archer, 2005).

GRID COMPUTING Grid computing is a term that involves a ‘hot’ model of distributed computing. This model embraces various architectures and forms but is restricted by detailed operational and behavior rules. Its precise definition is a difficult task because of the model nature and the large number of its different implementations. As a result, there are several misunderstandings of what the Grid is and what architectures can be called that way. Therefore, a clarification of the meaning of the Grid is needed. The computational Grid is practically a network of various types of heterogeneous computational engines and resources. Its size can vary from a small collection of stand-alone computers to a global network of thousands of interconnected units sharing distributed resources and working in parallel to establish a dominant global supercomputer. It involves computing, storage, application, coordination, and sharing. This architecture could be implemented at a local area network or the Internet to provide services to individuals or organizations all over the world. A Grid should include mechanisms for dividing a specific task into a subset of smaller ones,

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

scheduling and dynamic resource provisioning in order to perform parallel computation and merging the results. Overall, there are three basic principles that ought to be followed for a system to be called Computational Grid: •

•

•

The shared resources should not have centralized management, and there should be a successful coordination and service provision to users within different domains and geographical locations. A Grid should use standard, open and general-purpose protocols and interfaces for all its operations, such as authentication, resource sharing, and internode communication to avoid being an application-specific system. It should provide a competent service to the users about the collection of services, response time, computational performance and resource utilization to produce better resource utilization and performance rates than stand-alone units.

Grid computing is often confused with utility computing, cluster computing or P2P computing. Utility computing involves trade on-demand supply of various resources that are charged as services. Utility computing, although it is limited to the service providers’ network, resembles the Grid or can follow the Grids specifications. Cluster computing is a technology similar to the Grid computing but not quite the same. Clusters usually involve interconnected processors inside the restricted area networks. It is likely to be distributed and manage different resources, but they almost always require centralized control, which is the essential feature that distinguishes them from the Grid. Peer-to-peer applications, on the other hand, have remarkably similar infrastructure with the Grid since they use grid-like services for manipulating files. The main difference is based on the fact that P2P focus on services to be delivered to as many users as possible and these

services are detailed and restrained while the Grid model is set on other priorities such as resource integration, performance, reliability, quality of service and security. Grid computing is considered as the computing of the future. There is already an enormous amount of research around it, and new Grid applications are constantly emerging or take advantage of it. It already has a broad social value since it can be used by medical, biological and scientific research centers to handle resources and time-consuming calculations and tasks. However, there are issues under investigation for the Grid since it is a system of multiple objectives, a lot of which are difficult to be met. There is also a lot of research on the Quality of Service of the Grid, the performance, the effective coordination of users and resources and the serious issue of security at various levels and areas with attention to the remote computation and data integrity. Grid architecture makes extensive use of the Internet and the remote and distributed resources to create a global supercomputer that can perform complex tasks. It embraces the best features of many other distributed models, such as utility, cluster or P2P computing but its capacity and persistence to generalizing protocols, lack of centralized control, quality of service and failure tolerance make it unique and most probable ‘partner’ of the Semantic Web, which is the future of the Web. Its social and scientific contribution is incredible, and is expected to succeed in the future as its security, resource coordination and quality of service techniques will overcome the drawbacks of the present. Furthermore, the questions of data management in the Grids become more and more serious. From the side of data management, the Grid allows many replicated data objects, which allow a high-level of availability, reliability and performance to assure the user and application needs. A breakdown of communication happens when (Voicu et al., 2009): (1) a message is degraded during the communication between two regions (2) a message is lost because of dysfunction of

123

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

the network connection or (3) two regions cannot communicate because a network path is unavailable. In the first two cases, network protocols are used to maintain reliability (Voicu et al., 2009). Incorporating mobile devices in the Grid can lead to uncertainty and to under performance because of the inherent restrictions of mobile devices and the wireless communication connections. For the future generation of ubiquitous Grids, we have to find ways to rectify the restrictions that are innate in these devices and combine them in the Grid, so that the available resources are strengthened and the field of provided services is extended (Isaiadis & Getov, 2005). Tools and environments are available for simulation, data mining, knowledge discovery and collaborative work. There are various requirements for the Grid resource management strategies and its components (resource allocation, reservation management, job scheduling) since they should be in a position to react to failure of network connections (Volckaert et al., 2004). Health Grids are computing environments of distributed resources in which diverse and scattered biomedical data are accessed, and knowledge discovery and computations are performed. The growth of the information technology allows the biomedical researchers to capitalize in ubiquitous and transparent distributed systems and in a broad collection of tools for resource distribution (computation, storage, data, replication, security, semantic interoperability, and distribution of software as services) (Kratz et al., 2007). Electronic health records in collaboration with the Grid technologies and the increasing interdisciplinary collaborations offer the opportunity to participate in advanced scientific and prestigious medical discoveries. Medical Informatics based on a Grid can be extended to support the entire spectrum of health care: screening, diagnosis, treatment planning, epidemiology and public health. Health Grid was the first that determined the need for a particular middleware layer, between the global grid infrastructure, the middleware and the medical

124

or health applications (Breton et al., 2006). The Grid technology can support healthcare services integration but is usually unavailable at the point of care (for example in the house of the patient) (Koufi et al., 2008). The Share Project is a European Union financed research to establish a road map for future health grid research, giving priority to the opportunities, the obstacles and the likely difficulties (Olive et al., 2007a; Olive et al., 2007b). The primary technical road map has produced a series of technologies, models and essential growth points, development of a reference implementation for grid services, and has concluded with expanding a knowledge grid for medical research. In the Share Project, to create knowledge from data requires complex data integration, data mining and image processing applications and includes the use of artificial intelligence techniques to build relations between data from different sources and frameworks (Olive et al., 2007a; Olive et al., 2007b). Even if certain key pharmaceutical companies have established powerful Grid extensions for drug discovery, the specific skepticism remains for the value of Grid Computing in the sector. Therefore, in the Health Grid road map, particular attention should be given to the security and the choice of standards for the HealthGrid operating system and technologies. The security is an issue in all the technical levels: the networks should provide the protocols for secure data transfer, the Grid infrastructure should support secure mechanisms for Access, Authentication and Authorization, and the establishments should provide mechanisms for secure data storage (Breton et al., 2006). Computer systems should be autonomous (Sterritt & Hinchey, 2005a), where autonomicity implies self-management that often becomes visible in the frameworks of self-configuring, self-healing, self-optimizing, self-protecting and self-awareness. The autonomous computation is intended to improve the widespread utilization and computation possibilities of systems to assist the computer users. Since for most of the users access

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

to computation is by personal devices, research on autonomous computation focuses on this field (Sterritt & Bantz, 2004). The autonomous computing and other initiatives on the self-management systems are a powerful new vision for constructing complex computer systems. They provide the ability of complexity control by implementing self-governance (autonomy) and self-management (autonomicity) (Sterritt & Hinchey, 2005b). The principal emerging strategies to achieve this requirement are the ‘autonomic networks’ and the ‘autonomic communications’. The autonomicity vision is incorporated into the existing efforts that aim to the advanced automation including artificial intelligence research. This implies (Sterritt & Hinchey, 2005b) that all the processes should be designed effectively with autonomicity and self-managing capabilities. Adding semantics in the Grid using semantic services and the support of self-organization will allow creating more flexible and heterogeneous Grids (Beckstein et al., 2006). Proposals for the support of reliable electronic health business systems deal with adopting of the autonomic Grid computing and the Services-Oriented Architecture (SOA) (Omar & Taleb-Bendiab, 2006). This improvement provides the benefit of high quality of services and on-demand health monitoring (ubiquitously and on-line) (Omar & Taleb-Bendiab, 2006). Overall, the wireless Grid allows the user mobility to share underused available resources in a network that is required for e-health applications. The applications of wireless Grid contain (Benedict et al., 2008) patient monitoring, multimedia, wireless devices, etc. Incorporating mobile devices in Grid technologies and server applications can provide (Ozturk & Altilar, 2007) the potential to control the state of supercomputers with mobile devices and allow applications to access valuable data ubiquitously (for example, the production of positioning information from GPS mobile devices). Moreover, integrating mobile, wireless devices in the Grid has recently (Jan et al., 2009) drawn the attention of the research community since this

initiative has led to new challenges. Mobile ad hoc grid is a new approach of Grid computing (Jan et al., 2009) where appealing real scenarios can be applied. Finally, Knowledge Grid is an intelligent and realistic environment of Internet applications that allow operation from people or virtual roles to capture, publish and share explicit knowledge resources (Anya, et al., 2009). However, handling situations that result from emergencies requires semantic information about the environment and one flexible self-organizing information technology infrastructure to provide services that can be used (Beckstein et al., 2006). For example, intelligent services can achieve personalized and proactive management of individual patients (Omar et al., 2006).

Quality Dimensions in Healthcare During the last decades extensive efforts were focused on improvement of the provided products and services and quality improvement theories were adopted (Deming, 2000; Juran, 1988). The meaning of quality and the quality management strategies are much more complex in the health sector, compared to the industrial sector, where initially the theories for the quality were applied (Deming, 2000; Juran, 1988). Avedis Donabedian defined the following three dimensions of quality in healthcare: (1) structure, (2) process and (3) outcome (Donabedian, 1980a; Donabedian, 1980b). Given these quality dimensions, Donabedian formulated the following definition for quality in health care: this nature of health care that is expected to maximize the well-being of the patient, considering the balance of benefits but also losses that follow the health care process (Donabedian, 1980a; Donabedian, 1980b). In Figure 2, the essential quality dimensions both for products and services, as these have been defined by distinguished researchers are presented (Donabedian, 1982; Donabedian, 1983; Donabedian, 1985; Donabedian, 1991; Donabedian, 2001; Garvin, 1988; Maxwell, 1984; Maxwell,

125

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

Figure 2. Quality dimensions as defined by Donabedian, Maxwell and Garvin

1992). Quality assessment and measurement in traditional health care systems is feasible through these dimensions.

FUTURE RESEARCH DIRECTIONS The Semantic Grid applies Semantic Web technologies to raising the technical challenges we have to overcome. With the available data flow and handling, these issues are of utmost importance. Nowadays, when we think about amounts of data produced by simulations, experiments and sensors it becomes apparent that we have to automate the discovery. Therefore, we need automatic data management. This, in turn, calls for automatic annotation of data with metadata describing attractive features of both of data storage and management. Lastly, we have to try to move towards automatic knowledge management.

126

Producing data are one issue; preserving data is a separate issue. This process involves automated, semi-automated and manual annotation and data cleaning. Overall, there are many technical challenges to be resolved to ensure the information created today can undergo changes in storage media, devices and digital formats. Furthermore, we should focus on utilizing the previously presented quality dimensions in healthcare by developing mathematical models, which will be useful for the quality assessment of public e-health systems and services.

CONCLUSION E-health is an ambitious idea. It expects to set up a novel and powerful middleware that encourages a novel way of providing ubiquitous, personalized, intelligent and proactive health care services at

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

a distance. The three promising visions embrace the three corners of the ‘Semantic-PervasiveGrid Triangle’. Exploring all these three visions together requires working across at least three communities. The proponents of this ‘holistic’ approach argue that only through exploring the combined world (the entire triangle) we will make a comprehensive infrastructure for the vision of ambient intelligence and e-Medicine. If crowned with success, the e-health approach will unite different communities creating reliable, virtual organizations, which tackle new challenges, utilize a wide range of distributed resources for improving the quality of care and quality of life. Issues of quality assurance in e-health should consider the dimensions already used for quality assessment in the traditional health care.

REFERENCES Abidi, S. S. R. (2002). A Case Base Reasoning Framework to Author Personalized Health Maintenance Information. Paper presented at the 15th IEEE Symposium on Computer-Based Medical Systems. Abidi, S. S. R., Han, C. Y., & Abidi, S. R. (2001). An Intelligent Info-Structure for Composing and Pushing Personalised Healthcare Information Over the Internet. Paper presented at the 14th IEEE Symposium on Computer-Based Medical Systems. American College of Physicians (2008). E-health and its impact on Medical Practice. Position paper. Anya, O., Tawfik, H., & Nagar, A. Amin, S. (2009). E-workbench: A Case for Collaborative Decision Support in E-health. Paper presented at the 11th International Conference on Computer Modelling and Simulation.

Beckstein, C., Dittrich, P., Erfurth, Ch., Fey, D., Konig-Ries, B., Mundhenk, M., & Sack, H. (2006). SOGOS - A Distributed Meta Level Architecture for the Self-Organizing Grid of Services. Paper presented at the 7th International Conference on Mobile Data Management. Benedict, S., Rejitha, R. S., & Vasudevan, V. (2008). Threshold Accepting Scheduling algorithm for scientific workflows in Wireless Grids. Paper presented at the Fourth International Conference on Networked Computing and Advanced Information Management. Breton, V., Blanquer, I., Hernandez, V., Legré, Y., & Solomonides, T. (2006). Proposing a roadmap for HealthGrids. Studies in Health Technology and Informatics, 120, 319–329. Broens, T., van Halteren, A., van Sinderen, M., & Wac, K. (2007). Towards an application framework for context-aware m-health applications. International Journal of Internet Protocol Technology, 2(2), 109–116. doi:10.1504/IJIPT.2007.012374 Carlsson, C., & Walden, P. (2002). Further Quests for Value-Added Products and Services in Mobile Commerce. Paper presented at the The Xth European Conference on Information Systems. Chorianopoulos, K. (2008). Personalized and mobile digital TV applications. Multimedia Tools and Applications, 36(1-2), 1–10. doi:10.1007/ s11042-006-0081-8 de Toledo, P., Galarraga, M., Martínez, I., Serrano, L., Fernández, J., & Del Pozo, F. (2006). Towards e-Health Device Interoperability: The Spanish Experience in the Telemedicine Research Network. Paper presented at the Annual International Conference of the IEEE Engineering in Medicine and Biology Society.

Archer, N. (2005). Mobile eHealth: Making the Case. Euromgov2005. Mobile Government Consortium International.

127

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

Delicato, F., Pirmez, L., & Carmo, L. (2001). Fenix – personalized information filtering system for WWW pages. Internet Research: Electronic Networking Applications and Policy, 11(1), 42–48. doi:10.1108/10662240110365706

Germanakos, P., Mourlas, C., & Samaras, G. (2005). A Mobile Agent Approach for Ubiquitous and Personalized eHealth Information Systems. Paper presented at the 10th International Conference on User Modeling.

Deming, W. E. (2000). Out of the crisis. Cambridge, MA: The MIT Press.

Han, D., Park, S., & Kurkuri, S. (2007). An Evolving Mobile E-Health Service Platform. Paper presented at the International Conference on Consumer Electronics.

Donabedian, A. (1980a). The definition of quality and approaches to its assessment. Ann Arbor, MI: Health Administration Press. Donabedian, A. (1980b). Explorations in quality assessment and monitoring (Vol. 6). Ann Arbor, MI: Health Administration Press. Donabedian, A. (1982). The criteria and standards of quality. Ann Arbor, MI: Health Administration Press. Donabedian, A. (1983). Quality Assessment and Monitoring Retrospect and Prospect. Evaluation & the Health Professions, 6(3), 363–375. doi:10.1177/016327878300600309 Donabedian, A. (1985). The methods and findings of quality assessment and monitoring. Ann Arbor, MI: Health Administration Press. Donabedian, A. (1991). Striving for quality in health care. Ann Arbor, MI: Health Administration Press. Donabedian, A. (2001). An Introduction to Quality Assurance in Health Care. New York: Oxford University Press. Dong, J. C., Hu, X. P., Zhang, Z.-M., Zhou, D., & Jiang, K. (2009). Analysis and Design on Standard System of Electronic Health Records. Paper presented at the First International Workshop on Education Technology and Computer Science. Garvin, D. A. (1988). Managing Quality: The strategic and competitive edge.

128

Hanak, D., Szijarto, G., & Takacs, B. (2007). A mobile approach to ambient assisted living. Paper presented at the IADIS Wireless Applications and Computing. Harle, C. A., Padman, R., & Downs, J. S. (2009). Design, Implementation, and Preliminary Evaluation of a Web-Based Health Risk Calculator. Paper presented at the 42nd Hawaii International Conference on System Sciences. Intille, S. S., Larson, K., & Kukla, C. (2002). Just-In-Time Context-Sensitive Questioning for Preventative Health Care. Paper presented at the AAAI 2002 Workshop on Automation as Caregiver: The Role of Intelligent Technology in Elder Care. Isaiadis, S., & Getov, V. (2005). Integrating Mobile Devices into the Grid: Design Considerations and Evaluation. Euro-Par 2005 (Vol. 3648). Berlin: Springer Berlin / Heidelberg. Jan, S., Shah, I. A., & Al-Raweshidy, H. S. (2009). Performance Analysis of Proactive and Reactive Routing Protocols for Mobile Ad-hoc Grid in ehealth Applications. Paper presented at the International Conference on Communication Software and Networks. Joshi, A. (2000). On proxy agents, mobility, and web access. Mobile Networks and Applications, 5(4), 233–241. doi:10.1023/A:1019120915034 Juran, J. (1988). Juran on Planning for Quality. Washington, DC: Free Press.

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

Konstantas, D., Jones, V., & Herzog, R. (2002). MobiHealth –innovative 2.5 / 3G mobile services and applications for healthcare. Paper presented at the Eleventh Information Society Technologies (IST) Mobile and Wireless Telecommunications. Koufi, V., Malamateniou, F., & Vassilacopoulos, G. (2008). A Medical Diagnostic and Treatment Advice System for the Provision of Home Care. Paper presented at the 1st international conference on Pervasive Technologies Related to Assistive Environments. Kratz, M., Silverstein, J., & Dev, P. (2007). HealthGrid: Grid Technologies for Biomedicine. Fort Detrick, Maryland: Telemedicine & Advanced Technology Research Center, U.S. Army Medical Research and Materiel Command. Lugmayr, A., Risse, T., Stockleben, B., Kaario, J., & Laurila, K. (2009). Special issue on semantic ambient media experience. Multimedia Tools and Applications, 331–335. doi:10.1007/s11042-0090283-y Lupu, E., Dulay, N., Sloman, M., Sventek, J., Heeps, S., & Strowes, S. (2008). AMUSE: autonomic management of ubiquitous e-Health systems. Concurrency and Computation, 20(3), 277–295. doi:10.1002/cpe.1194 Maxwell, R. J. (1984). Quality assessment in health. British Medical Journal, 288(6428), 1470–1472. doi:10.1136/bmj.288.6428.1470 Maxwell, R. J. (1992). Dimensions of quality revisited: from thought to action. Quality in Health Care, 1, 171–177. doi:10.1136/qshc.1.3.171 Meneu, T., Traver, V., Fernández, C., Serafin, R., Domínguez, D., & Guillén, S. (2009). Life Assistance Protocols (LAP) – A Model for the Personalization of Lifestyle Support for Health and Wellbeing. Paper presented at the International Conference on eHealth, Telemedicine, and Social Medicine.

Milenković, A., Otto, C., Jovanov, E. (2006). Wireless Sensor Networks for Personal Health Monitoring: Issues and an Implementation. Computer Communications (Special issue: Wireless Sensor Networks: Performance, Reliability, Security, and Beyond), 29(13-14), 2521-2533. Mohomed, I., Misra, A., Ebling, M., & Jerome, W. (2008). Context-Aware and Personalized Event Filtering for Low-Overhead Continuous Remote Health Monitoring. Paper presented at the International Symposium on a World of Wireless, Mobile and Multimedia Networks. Nijland, N., Seydel, E. R., van Gemert-Pijnen, J. E. W. C., Brandenburg, B., Kelders, S. M., & Will, M. (2009). Evaluation of an Internet-Based Application for Supporting Self-Care of Patients with Diabetes Mellitus Type 2. Paper presented at the International Conference on eHealth, Telemedicine, and Social Medicine. Nikolidakis, S., Vergados, D. D., & Anagnostopoulos, I. (2008). Health Care Information Systems and Personalized Services for Assisting Living of Elderly People at Nursing Home. Paper presented at the Third International Workshop on Semantic Media Adaptation and Personalization. Olive, M., Rahmouni, H., & Solomonides, T. (2007a). From HealthGrid to SHARE: A Selective Review of Projects. Studies in Health Technology and Informatics, 126, 306–313. Olive, M., Rahmouni, H., Solomonides, T., Breton, V., Legre, Y., & Blanquer, I. (2007b). SHARE Roadmap 1: Towards a Debate. Studies in Health Technology and Informatics, 126, 164–173. Oliveira, I. C., Oliveira, J. L., Santos, M., MartinSanchez, F., & Sousa Pereira, A. (2003). On the requirements of biomedical information tools for health applications: the INFOGENMED case study. Paper presented at the 7th Protuguese Conference on Biomedical Engineering.

129

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

Omar, W. M., Ahmad, B. A., & Taleb-Bendiab, A. (2006). Grid Overlay for Remote e-Health Monitoring. Paper presented at the IEEE International Conference on Computer Systems and Applications.

Sterritt, R., & Bantz, D. F. (2004). PAC-MEN: Personal Autonomic Computing Monitoring Environment. Paper presented at the Database and Expert Systems Applications, 15th International Workshop.

Omar, W. M., & Taleb-Bendiab, A. (2006). Service Oriented Architecture for e-Health Support Services Based on Grid Computing Overlay. Paper presented at the Proceedings of the IEEE International Conference on Services Computing.

Sterritt, R., & Hinchey, M. (2005a). Autonomicity – An Antidote for Complexity? Paper presented at the Proceedings of the 2005 IEEE Computational Systems Bioinformatics Conference - Workshops.

Ozturk, E., & Altilar, D. T. (2007). IMOGA: An Architecture for Integrating Mobile Devices into Grid Applications. Paper presented at the Fourth Annual Conference on Mobile and Ubiquitous Systems: Networking & Services. Panayiotou, C., & Samaras, G. (2004). mPERSONA: Personalized Portals for the Wireless User: An Agent Approach. Mobile Networks and Applications, 9(6), 663–677. doi:10.1023/ B:MONE.0000042505.07003.e6 Panteli, N., Pitsillides, B., Pitsillides, A., & Samaras, G. (2007). An E-healthcare Mobile application: A Stakeholders’ analysis: Experience of Reading. In Al-Hakim, L. (Ed.), Web MobileBased Applications for Healthcare Management. Hershey, PA: IGI Global. Park, S., Mackenzie, K., & Jayaraman, S. (2002). The wearable motherboard: a framework for personalized mobile information processing (PMIP). Paper presented at the 39th Design Automation Conference.

Sterritt, R., & Hinchey, M. (2005b). Why Computer-Based Systems Should be Autonomic. Paper presented at the Proceedings of the 12th IEEE International Conference and Workshops on Engineering of Computer-Based Systems. Van Halteren, A., Konstantas, D., Bults, R., Wac, K., Dokovsky, N., & Koprinkov, G. (2004). MobiHealth: Ambulant Patient Monitoring Over Next Generation Public Wireless Networks. In Demiris, G. (Ed.), E-Health: Current Status and Future Trends (Vol. 106). Amsterdam: IOS Press. Voicu, L. C., Schuldt, H., Breitbart, Y., & Schek, H. J. (2009). Replicated Data Management in the Grid: The Re:GRIDiT Approach. Paper presented at the 1st ACM workshop on Data grids for eScience. Volckaert, B., Thysebaert, P., De Leenheer, M., De Turck, F., Dhoedt, B., & Demeester, P. (2004). Grid computing: the next network challenge! Journal of the Communications Network, 3, 159–165.

Riva, G. (2003). Ambient Intelligence in Health Care. Cyberpsychology & Behavior, 6(3), 295– 300. doi:10.1089/109493103322011597

Wac, K. (2005). Towards QoS-awareness of Context-aware Mobile Applications and Services. Paper presented at the On the Move to Meaningful Internet Systems 2005: OTM Workshops, Ph.D. Student Symposium.

Stell, A., Sinnott, R., Ajayi, O., & Jiang, J. (2007). Security Oriented e-Infrastructures Supporting Neurological Research and Clinical Trials. Paper presented at the Proceedings of the The Second International Conference on Availability, Reliability and Security.

Wac, K., van Halteren, A., & Konstantas, D. (2006). QoS-Predictions Service: Infrastructural Support for Proactive QoS- and Context-Aware Mobile Services (Position Paper) On the Move to Meaningful Internet Systems 2006: OTM 2006 Workshops. Berlin: Springer Berlin / Heidelberg.

130

Quality Issues in Personalized E-Health, Mobile Health and E-Health Grids

Yusof, M. M., Paul, R. J., & Stergioulas, L. K. (2006). Towards a Framework for Health Information Systems Evaluation. Paper presented at the 39th Annual Hawaii International Conference on System Sciences.

This work was previously published in E-Health Systems Quality and Reliability: Models and Standards, edited by Anastasius Moumtzoglou and Anastasia Kastania, pp. 278-290, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

131

132

Chapter 1.9

Adoption of Electronic Health Records: A Study of CIO Perceptions Yousuf J. Ahmad Mercy Health Partners, USA Vijay V. Raghavan Northern Kentucky University, USA William Benjamin Martz Jr. Northern Kentucky University, USA

ABSTRACT Adoption of Electronic Health Records (EHR) can provide an impetus to a greater degree of overall adoption of Information Technology (IT) in many healthcare organizations. In this study, using a Delphi technique, input from 40 CIO responses is analyzed to provide an insight into acceptance and adoption of EHRs at an enterprise level. Many useful findings emerged from the study. First, a majority of the participants believed DOI: 10.4018/978-1-60960-561-2.ch109

that about 40-49 percent of the providers will be using EHRs by the year 2014, thus highlighting the need for studying EHR diffusion in hospitals. As predictors of successful implementations, physicians’ leadership and attitude was ranked as the most important factor. Another significant determinant of success was the business model of the physicians—whether they are affiliated with hospitals or working independently. This factor highlights the need to employ different strategies to encourage adoption of EHRs among these distinct groups. All findings are discussed, including their implications for IT diffusion in healthcare.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Adoption of Electronic Health Records

INTRODUCTION In spite of many widespread uses of information technology, its potential to help in delivering quality healthcare has not yet been realized. A centralized database of patient data that can be easily accessed by healthcare providers is not yet a reality. Companies do offer proprietary solutions but often these systems cannot interact with each other. Adoption and diffusion rates for technologies such as computerized physician order entry (CPOE) have been very low (Poon, Blumenthal, & Jaggi, 2004). We seek to understand the rationale behind the rather slow adoption rate for electronic health records (EHRs). For the purposes of this study, an electronic health record refers to an individual patient’s medical record in a digital format; a real-time patient health record with access to evidence-based decision support tools that can be used to aid clinicians in decision-making. EHRs can also automate and streamline a clinician’s workflow ensuring that all clinical information is clearly communicated across individuals and groups that use the patient data. They can also prevent delays in responding to individuals needing urgent care (Chau & Hu, 2002). One of the complexities with EHRs is that its database activities (create, read, update or delete) can originate from many diverse sources: at a physician’s office during patient visits, pharmacy counters or at hospitals when patients receive care from the hospital. In physician practices, adoption of EHR is often an individual or a small-group decision. While the adoption of technology in general at small-group levels has been studied previously (Baxley & Campbell, 2008), the current study focuses specifically on the institutional level adoption of electronic health record systems. We know from prior studies that organizational adoption of software product lines hinges on the adopter making business, technical, organizational, procedural, financial, and personnel changes (Clements, Jones, McGregor, & Northrop, 2006). Studies evaluating healthcare information systems from an enterprise

perspective have identified potential conflict areas and have proposed conceptual models to facilitate implementation of healthcare Information systems (Connell & Young, 2007). Meetings have been called, retreats have been held, and policy recommendations ranging from financial incentives to promoting through educational, marketing, and supporting activities have been proposed (Middleton, Hammond, Brennan, & Cooper, 2005). And yet, implementing EHR systems remains the least understood process in the area of healthcare information systems. CIOs of healthcare institutions comprise a primary stakeholder group that is on the forefront of implementing organization-wide EHRs. It seems unlikely that the implementation of an organization-wide EHR would be undertaken without the participation and concurrence of the organization’s CIO. Hence we pursue a study of CIO perceptions of critical issues surrounding organization adoption of EHRs.

AREAS OF INVESTIGATION This section segments the areas investigated in our Delphi study under three broad categories: issues of adoption, barriers to adoption, and finally determinants of EHR adoption success. Under issues of adoption a variety of aspects of EHR adoption are examined. Barriers to Adoption and Determinants of success are of special significance as judged from many prior studies (Baxley & Campbell, 2008; Cooper, 2005; Gans, Kralewski, Hammons, & Dowd, 2005; Lowes & Spikol, 2008; Withrow, 2008) and we treat them separately.

Issues of EHR Adoption There is no disagreement about the powerful ways in which IT can help the creation and management of EHRs and consequently improve the quality of healthcare. Computerized order entry can ensure that medication and laboratory orders are accurate

133

Adoption of Electronic Health Records

and legible. A system as envisioned can also allow logical checks for duplication or redundancy, thereby eliminating both risk and waste from the healthcare delivery process. Managing the results from a laboratory order is another key area where digitization and increased availability can provide significant benefits. The ideal system enables instant communication of laboratory results as well as archival of past results for immediate comparisons. Management of chronic diseases, reducing variations in care delivery, and disease registries are other outcomes enabled by EHRs that can provide significant benefits to both organizations and to patients. In spite of these potential benefits, an organization’s successful diffusion and adoption of EHRs depends on how it manages its way through a variety of nontechnical issues. Diffusion of EHRs: Varying estimates of the extent of diffusion of EHRs are reported. While most agree that EHR diffusion is low at present, its anticipated diffusion rate into healthcare systems needs to be understood more clearly. A high level of diffusion is not guaranteed. For example, while 60% of primary practitioners in UK use some form of EHR, most private practitioners in the UK still manage their patient information manually. Even those physicians who have adopted EHRs are not actively using functionalities that are necessary to improve health care quality and patient safety (Simon et al., 2008). The US has more complexity and therefore may have more root causes associated with lack of diffusion than the UK; these include such issues as US not being a single payer system, lack of governmental standards and mandates, fragmented nature of US healthcare, and lack of government ownership. Scope: Organization-wide EHR implementation projects, in general, have a very wide scope as they are designed as a blanket to help cover a variety of private and public health issues. EHRs include not only clinical information systems but also scheduling, billing, and other functionalities, and these broader functionalities are critical for the fiscal viability of the New Model of Family

134

Medicine (Newton et al., 2006). Size not only adds complexity, but also problems related to manageability. An understanding of what is really expected of EHRs will help projects focus on those issues deemed to have highest priority. Governance: Governance issues have remained an obstacle as institutions that have implement EHRs try to obtain high quality data for quality initiatives (Jain & Hayden, 2008). Governance of a state-wide or nationwide EHR must be properly defined. Who is going to be responsible for the implementation, monitoring of compliance, and ownership of the EHR? Who will be the ultimate custodian of data and who will be able to police? Experts agree (see for example, Collins, 2004; Bakker, 2004) that the information governance issues of the EHR have been the hardest part of maintaining an EHR. Although all of these are important issues, we focus especially on the question of dealing with privacy and security aspects of implementing an EHR. Incentivizing the providers: It has been suggested that policies should be designed to provide incentives to help practices improve the quality of their care by using EHRs (Miller et al., 2005; Vijayan, 2005). Incentivizing the providers of healthcare, especially physicians, is going to be very difficult. The state or the federal government would have to reimburse providers who are participating in this EHR better. There can be no significant penalty, perceived or real, associated with adopting EHRs. This will be the most challenging limitation since physicians are inherently risk averse, and therefore any level of risk that they would have to undertake would need to be managed and defrayed by the state. The government, including the Center for Medicare and Medicaid Services (CMS), payers and the employers need to work together not only on the criteria for evaluating physician performance, but also on the rewards. The incentive would have to be motivating and probably not just financial in nature. The state or federal government must decide on the outcomes it wishes to achieve. From

Adoption of Electronic Health Records

these outcomes, a proper set of incentives can be created. For example, not only should there be incentives for physicians to practice medicine using an EHR, but a proportionate incentive to see patients with an EHR in rural areas. Public health: Public Health: Information Technology can be used to improve the quality of healthcare for the general population. Examples include, but are not limited to, incident reporting, disease surveillance, transfer of epidemiology maps, maintenance of disease registries, and delivery of alerts and other information to clinicians and health workers (Institute of Medicine, 2001). Aims of the Institute of Medicine: The relationship between electronic health records and the quality of healthcare warrants further exploration. The Institute of Medicine has laid out the following six aims for healthcare: safety, timeliness, effectiveness, efficiency, equity, and patient-centeredness. Healthcare consumption is a multi-criteria decision process and a rank ordering of these factors by content experts will help us understand which one of the factors is important in promoting EHR adoption. Measurement: Measuring the benefits of the EHR is difficult, especially since in certain areas we do not have any baseline metrics with which to compare. For this project, we elicited CIO opinions on important metrics for measuring the adoption of EHRs. Barriers to adoption: Although the use of the EHR today is still low, there is an increased interest in the application of this technology and a sense of urgency is felt even at the federal level. Our study explores the perspective of CIOs of health care organizations and more especially those who are directly involved in EHR implementations. We build on some of the barriers already identified to explore how CIOs feel about these factors as possible barriers to adoption. For example, cost, resistance by physicians (Berner, Detmer, & Simborg, 2005), and vendor and product immaturity have been identified as barriers to implementing CPOEs (Poon et al., 2004). EHRs are more com-

plex systems with a broader scope and we borrow some of these barriers for the study including cost, cultural barriers, lack of standards and interoperability, complexity of medicine, and workflow changes (Ash & Bates, 2005). Determinants of success: A study by Menachemi, Perkins, Durme, and Brooks (2006) presents some evidence to support the claim that physicians of younger age are more likely to adopt EHR systems and male physicians are more likely to adopt EHRs. The same study also suggests that physician specialty may have some influence on EHR adoption. Family physicians are more likely to use robust features such as allergy and medication lists, diagnosis, problem lists, patient scheduling and educational materials, preventive services reminders and access to reference material. There is also some evidence to suggest that hospital-based practices and practices that teach medical students or residents were more likely to have an EHR (Simon et al., 2007). We decided to label this aspect the business model of the physicians. Medical training and state leadership in promoting EHRs were other factors included in the current study as determinants of success. The impact of governmental activity on EHR adoption has been explored in prior studies (Ford et al., 2009). Summary: Healthcare information systems necessarily span a wide area of literature and research. Although several of these have been mentioned and drive this research study, we are left with two broad research questions: • •

What are the factors that facilitate adoption of EHRs? Will EHRs deliver the administrative and clinical benefits that they are being touted for?

RESEARCH METHOD CIOs of healthcare institutions comprise a primary group that is on the forefront of implementing

135

Adoption of Electronic Health Records

EHRs. Their participation and backing are necessary to study any issue surrounding EHR. We employed a Delphi method survey to elicit feedback on areas surrounding the research questions. The value of the Delphi method is in the ideas it generates, both those that build consensus and ones that do not. Also, it encourages respondents to consider group opinions especially on those aspects of the study about which the respondents may be unsure. Delphi is particularly appropriate for subjective information where measurable data is not available (Dalkey & Helmer, 1963). Delphi is a method of choice to help achieve convergence or consensus on various aspects of an emerging topic. It achieves this consensus by providing for repeated responses to the same questions after a controlled feedback (Nevo & Chan, 2007; Hayes, 2007). Our study was conducted with the support of the College of Healthcare Information Management Executives (CHIME) and a total of 40 health care CIOs participated, 20 in each round. A total of 4 of these survey had missing responses for some questions and were not fully usable. The profile of the subset of CHIME subjects that participated in the study was similar to the organization as a whole. Three demographic characteristics of age, gender and operating budget of the IT organization as reported in Table 1 shows that the respondent characteristics were representative of the organization as a whole. While all questions were objective in nature, they solicited free-text commenting by the respondents to better explain their position(s). Although we have labeled it as a 3-round Delphi survey, in the pure traditional definition it is a 2-round Delphi with an informal intermediate round wherein two respondents who used the free-form response to suggest additional items to be included in the survey were contacted to discuss these items as well as items that were not be clear to them. Clarifications were provided before the third round on all items that were deemed not clearly understood. One of the authors who is himself a CIO of a healthcare organization made subjective judgments on items that might not have

136

been understood clearly, based on this feedback. Such enhancements helped improve the quality of the responses in the final round and do not detract from the spirit of the Delphi technique. This means that complete data were available for the first and final round with an informal second round used to improve the questions surveyed for the final round.

ANALYSIS AND RESULTS We employed three different types of non-parametric tests: (1) the Wilcoxon sign-ranked test was used to compare each pair of scores to understand the differences between the initial and final round responses of respondents, (2) the Mann-Whitney test was used; where the scores are not treated pair wise, but ranked as though they were all from a

Table 1. Demographic characteristics Characteristic Gender

N(40)a 14 Females 22 Males

Age Group Less than or equal to 30 years

0

31-39

6

40-49

5

50-59

23

60 years or Older

2

IT Operating Budget Less than $1 million

4

$1.1 million to 10 million

7

$10.1 million to 20 million

11

$20.1 million to 30 million

4

$30.1 million to 40 million

2

$40.1 million to 50 million

4

More than 50 million

4

4 Missing data

a

Adoption of Electronic Health Records

single list, then sum of the ranks for each subgroup list is calculated, For the Mann-Whitney test, if there is no real difference between conditions, the ranks of the subgroups will remain roughly equal. The data from the initial and final rounds were compared for significance, and (3) the KruskalWallis, which is similar to Mann-Whitney, but used for statistical significance when there were three or more conditions used for questions that contained answers with three or more categories. Considering the small sample size that is typical of Delphi studies and the other assumptions necessary for proper use of parametric statistics, we chose to employ these non-parametric methods.

Issues of Adoption Diffusion of EHRs: Four different aspects of diffusion of EHRs were surveyed: current level of diffusion in healthcare organizations, current level of diffusion in physician offices, expected level of diffusion in about five years, and finally the reason for differences in the level of diffusion compared to UK where it is expected to be about 60%. Four commonly given reasons for this variation are: differences in single payer system, lack of governmental standards and mandates, fragmented nature of US healthcare, lack of government ownership. The current penetration of EHR in health organizations were generally estimated to be less than 20% with 17 responses estimating it to be less than 9%. For physician practices, the responses were similar. The variations between the two rounds of the Delphi study were insignificant. Expected future penetration by the year 2014 was estimated to be between 40% and 59%. The expected future penetration in both rounds was estimated to be between 40% and 59% with the average increasing significantly from 2.80 in the first round to 3.39 in the second round where the percentages were split in quintiles and coded from 1 through 5 (n=38, df=4, Pearson’s _2=.047). This is surprisingly low considering that some

expect that there might be a presidential mandate for EHR in the U.S. by the year 2014. Finally, the reason for EHR utilization in the USA being low compared to UK was explored. This was an open-ended question in the first round and using these responses it was made a forced-choice response in the final round. Four respondents believed it was due to the differences in single-payer system but a majority of them believed it was due to the lack of mandates and standards. Fragmented nature of healthcare received 5 votes and the lack of government ownership received 3 votes. Considering a fairly even distribution of the responses, we believe that all these remain competing reasons for facilitating EHR adoption. Scope. Given the breadth of promises for EHRs, what components are most important to its users? Majority of the opinion both in round 1 and round 3 centered around improving communication between care givers. Reducing adverse drug events (ADEs), reducing avoidable deaths, and improving physician/hospital integration were other choices. In the final round, physician hospital integration received five votes. Although this aspect of EHR adoption, physician-hospital integration, is increasingly mentioned by the practitioners, it is not clearly defined in the academic literature nor was it elaborated in our survey questionnaire. Considering the number of votes it received, this may be an area of EHR that might increase its adoption at least among the hospitals. There was another category available for the respondents and two respondents chose this in the final round. Convergence to the majority opinion was statistically significant (p=0.096). Governance: Although there are multiple facets of governance, we focused mainly on the privacy concerns since this is often cited as one of the reasons for lack of EHR adoption. Some believe that the privacy protections in Healthcare Insurance Portability and Accountability Act (HIPAA) may be insufficient with the advent of Health Information Exchanges (Walker, 2008).

137

Adoption of Electronic Health Records

The question was phrased as: Do you believe the privacy and confidentiality of patient records are compromised by going from paper to electronic? The responses were elicited based on a five point Likert Scale ranging from Strongly Agree (1.00) to Strongly Disagree (5.00). Of the 35 valid responses received for this question, the mean was 3.74 with a standard deviation of 0.561. The frequencies of responses were Neutral 11, Disagree 22, and Strongly Disagree 2, making up a total of 35. This means that although privacy and confidentiality of patient records is cited as a reason for non-adoption, the CIOs believe that digitization of patient records does not adversely affect privacy and confidentiality. The Wicoxon signed rank test of the difference between the first and final round responses is significant with a p value of .034. Interestingly, this finding is counter to the popular discussion. Incentivizing the providers: The question of funding was addressed indirectly through the choices provided for the question: What will make private practice physicians purchase EHR on their own? Among the choices were: demand from their patients, pressure (mandate) from the governments, improved reimbursement from payers, realization of the return on investment, someone else paying for it and improved quality. The changes in responses from the initial to the final round indicate that improved reimbursement from payers would be a significant driver (p=0.096). These are conditions for which healthcare providers had traditionally been paid for, but starting in 2008, they will not be paid for and for certain cases healthcare providers will not be allowed to bill the payer or the patient for these cases. A statistically significant number of respondents agreed with the majority opinion of very likely, when asked whether engaging patients in the treatment of their own care improves their health outcomes (p=0.043). Given the pressures to earn the same income and with stagnating volume, it might be natural to think that unless physicians can get financial assistance in the form of payer

138

reimbursement, there really is not much incentive for them to switch from paper-based systems to EHRs. Public health: A clarification on what exactly is public health was made during the final round as two of the respondents did not know how inclusively to define. A definition on public health as well as distinguishing the work that public health professionals do from privatized medicine or clinical professionals was provided. In the final round, public health was defined as the science of protecting and improving the health of communities through education, promotion of healthy lifestyle, and research for disease and injury prevention. Additionally, public health professionals try to prevent problems from happening or recurring through implementing educational programs, developing policies, administering services, and conducting research, in contrast to clinical professionals such as doctors and nurses, who focus primarily on treating individuals after they become sick or injured. These definitions were taken from the Association of Schools of Public Health. The clarification not withstanding we saw that respondents had a statistically significant change in their response, especially on immunizations (p=0.010). Responses were elicited on six areas of public health and a seventh choice for free form responses was provided. Biosurveillance/Bioterrorism, detection of an epidemic, drug recall, natural disaster relief, mandatory disease reporting, and immunizations were the identified areas of public health. Respondents were asked to identify whether EHR could help facilitate the public health efforts in each of these areas. Frequencies of responses in both the initial and final rounds are provided in Figure 1. Improving the quality of healthcare: The CIOs were asked to rate the six aims for healthcare as proposed by the institute of medicine (safety, timeliness, effectiveness, efficiency, equity, and patient-centeredness) along with compliance and cost. The rating was done on Likert scale ranging

Adoption of Electronic Health Records

Figure 1. Potential use of EHRs in public health

from not at all important (1) to extremely important (5). The rating for all the components were above 3 indicating that the CIOs believed that all aspects of quality of healthcare delivery will be improved by the adoption of EHRs. The difference is rating between the first and the final round was not significant on any of these components indicating the absence of any influence of Delphi study on the perception of CIOs. The mean ratings for each of these items are given in table 2. Measurement: CIO impressions on return on investment (ROI) metrics for EHRs were elicited in five dimensions: reduction in transcription costs, reduction in paper costs, faster patient throughput, and patient satisfaction. Response was positive in all these dimensions indicating that it is important to measure EHR benefits in Table 2. Components of healthcare quality Quality Component

Initial Round

Final Round

Safety

4.15

4.33

Timeliness

4.15

3.89

Quality

3.84

3.83

Compliance

3.75

3.56

Effectiveness

3.85

3.78

Efficiency

3.85

4.00

Patient Focused

3.40

3.50

Cost

3.75

3.86

all these areas. There was no significant change between rounds. Barriers to adoption: Barriers to adoption identified in the study were: cost, cultural barriers, lack of standards and interoperability, complexity of medicine, and workflow changes for caregivers. Another item was included to capture the rest of the barriers not mentioned here. While the aggregate response did not change significantly, we did see that categorically, there was a statistically significant change (p = 0.034) between rounds. Respondents picked workflow changes for caregivers based on the majority opinion, which is still not abundantly mentioned in the literature. The Delphi technique clearly had an influence on how the group responded to this question as workflow was relayed to them to be the majority opinion from the first round. The frequency of responses is noted in Figure 2. Determinants of success: Age, specialty, gender, and medical training (such as school attended) of the physicians along with the business model (whether employed by hospital or working independently), leadership of the state in which the physician practices, and physician leadership and attitude were studied as possible determinants of successful adoption of EHRs. Physician leadership and attitude was included only the final round after feedback from the second round. Once

139

Adoption of Electronic Health Records

Figure 2. Barriers to EHR adoption

included, physician leadership received the most votes followed by business model of the physicians. Interestingly, traditional significant factors such as the physician’s age fared much less favorably. The votes for various items in both the rounds of Delphi are provided in Figure 3.

SUMMARY OF FINDINGS, LIMITATIONS AND FUTURE DIRECTIONS In summary, we see that the Delphi method did have an impact on the experts’ opinion between the initial and final rounds, but there were certain areas where the impact was not as expected or Figure 3. Determinants of success

140

did not conform to group opinion. A group of experts conforming on their opinion(s) does not mean that they are correct. Analysis across the gender, age category, or budget size of the survey respondents did not reveal any changes of statistical significance. We see that CIOs perspectives did change over time in other areas. The sharing and reiteration characteristics of the Delphi technique are suspected to be the cause. We learned some of the factors that will make EHRs successful. As there was no difference between rounds and the emphasis remains high, one can conclude that Business model of the physician, whether they are employed by the hospital or independent, and physician leadership and attitude are key factors that determine

Adoption of Electronic Health Records

adoption of EHRs. The results affirmed findings from the prior studies that EHRs are expected to improve safety, timeliness, quality, compliance, effectiveness, efficiency, patient-focus, and cost of healthcare. The Delphi method helped the majority opinion influence the respondents when asked about the critical barriers, which included cost, culture, complexity of medicine, and workflow changes for caregivers. While not found explicitly in the literature, workflow changes for caregivers was the majority opinion in the end. A majority of the respondents agreed that EHRs help place the patient at the center of care, which is what the literature reflects. Additionally, EHRs are believed to help bridge the chasm between inpatient and outpatient care. Not only are patients more engaged when an EHRs is being used, the experts concurred with the literature that their health outcomes are going to be better as a result of being engaged in their own care. However, the experts believe that only 40-59% of the providers will be using an EHR by 2014. This is concerning given that there is a presidential mandate to implement EHRS by 2014. Most of the respondents did not see privacy and security of the patient being compromised by the utilization of an EHR. EHRs do support public health efforts of biosurveillance, detection of an epidemic, such as, Avian Flu or SARS, natural disaster relief, mandatory disease reporting and immunizations. The results were influenced by the majority opinion as well as by giving the respondents an official definition of public health and how public health professionals differed from clinical professionals in private medicine. In looking at tangible metrics around financial savings, all the responses agreed that EHR utilization does result in the reduction of transcription and paper cost, increased patient throughput, and higher patient satisfaction. Despite the unanimously agreeing on the potential savings that a physician might observe, the survey findings revealed that unless there is higher reimbursement

from payers or unless someone else is going to defray the costs, the private physician sees no reason to start using an EHR today. The experts mirrored the statistic mentioned in the literature on the percentage of U.S. adults that receive the recommended care and if EHRs will help improve this number. Their final response, after being influenced by the majority opinion as a product of the Delphi technique, was 50% as opposite to 55% percent as stated in the literature. This was indeed very encouraging to see, given the rising costs of healthcare. This study identified many determinants in the successful deployment of an EHR and it does affirm that EHRs can help deliver clinical and administrative savings as trumpeted by the literature. The study is particularly pertinent as the panelists were experts who are in the forefront of EHR implementation. One of the interesting findings is the identification of changes to the workflow of a physician as being the biggest barrier to EHR implementation. Cost of an EHR was a close second. It was disturbing to observe how little respondents knew about the role of EHRs in public health, especially in its most broad definition. It was not until an explicit definition of public health as the science of protecting and improving the health of communities through education, promotion of healthy lifestyles, and research for disease and injury prevention in the third round that respondents understood the distinction between the fields of medicine and public health. However, once defined, the experts did agree that there were several public health efforts that EHRs can help extend. Another new finding, not generally discussed in the literature, is that the business model of the physician is identified as a predictor of success in the dissemination of EHRs. It makes a significant difference if a physician is employed (salaried) or if he or she owns the practice because it is difficult to swallow the cost of an EHR by a physician owner, especially given all the incremental cuts in reimbursement. While still difficult, it is

141

Adoption of Electronic Health Records

Table 3. Summary of findings Aspect Measured

Determinants of EHR Success

Critical Barriers to Implementing EHR

Initial Round

Final Round

Improved Workflows

12

15

+

Cost affordable

12

10

-

Lack of Standards

11

4

-

5

16

+

19

14

-

Complexity of Medicine Business model of the physician

Physician Specific Factors

Factors Influencing Purchase of an EHR By Private Practice Physicians

Aims of the Institute of Medicine

Change Direction

Physician leadership/attitude

0

14

+

Physician age

8

1

-

Physician gender

3

0

-

Physician specialty

6

2

-

State government’s leadership

3

2

-

Medical school attended

4

3

-

Improve payer reimbursement

No

Yes

Yes

Improve Quality

Yes

No

No

Pressure from government

No

Yes

No

Demand from patients

No

Yes

Yes

Improve safety

4.15

4.33

+

Improve timeliness

4.15

3.89

-

Improve efficiency

3.84

3.83

=

Improve effectiveness

3.85

3.78

-

Improve patient-centeredness

3.4

3.5

+

3.75

3.86

+

Reduce adverse drug events

5

1

-

EHR as an Enabler to Achieving

Improve communication between caregivers

4

10

+

Higher Quality c

Reduce avoidable deaths

3

1

-

Improve regulatory compliance

Improve physician-hospital integration

Public Health

c

5

2

-

Biosurveillance/bioterrorism

11

5

-

Detection of an epidemic

12

7

-

Drug Recall

14

11

-

Natural disaster relief

11

2

-

Mandatory disease reporting

16

16

=

Immunizations

Return on Investment

b

0

12

+

Transcription cost

4.06

4.11

+

Paper cost

3.76

3.83

+

Storage cost

4.12

4

-

Patient throughput

4.12

3.78

-

Patient satisfaction

3.82

3.94

+

Likert scale ranging from not important to extremely important

a

Likert scale ranging from Strongly Disagree to Strongly Agree

b

Checkboxes - number of respondents (frequencies)

c

142

Adoption of Electronic Health Records

comparatively palatable when the cost of an EHR does not directly hit the pocket of the physician – as usually in the case of employed physicians. The majority of the opinion supported the literature that EHRs help improve communication between care givers; however, another new finding surfaced in this area - that EHRs also assist with the physician-hospital integration. This is intriguing to observe given that the Stark laws have most recently been relaxed to allow hospitals to defray the cost of an EHR for an affiliated physician. Age, specialty, and gender of the physician were not seen as factors of significance impacting the adoption of EHRs even though the literature suggests that older physicians find it hard to learn and adapt to using an EHR in their everyday practice. The expert respondents did agree that EHRs enabled healthcare organizations to achieve the six aims of the IOM - for care to be safe, timely, efficient, effective, equitable, and patient-centered. There are several areas within healthcare and public health that can leverage the findings from this study and build upon it. The results of the survey show that physician-hospital integration is an important area that can benefit from the deployment of an EHR. Physician-hospital integration has had a checkered history and one can further research and leverage the benefits of an EHR toward improving physician-hospital relationship(s). Future research can study how an EHR can help address the market forces that are converging around the need for more physicianhospital integration. Some of these forces that can be addressed are unsustainable growth in health care costs due to structural issues, such as aging population and a failure of the managed care paradigm, the move by the federal government and commercial payers to value-based reimbursement, significant shifts in provider demographics, including the nursing shortage and the noticeable shift toward quality of life requirements by younger physicians. Another area for potential research is what key features and functionality of an EHR contribute

the most toward delivering high quality and safe patient care. Does Computerized Physician Order Entry (CPOE) save more lives than a bar-coded medication administration system that checks the 5-rights of drug delivery (right patient, right drug, right time, right dose, and right route)? On the public health side exclusively, one can build upon this research to find out how an EHR can contribute toward meeting basic public health requirements. For example, can an EHR automatically report on all reportable diseases as mandated by the respective State Health Departments? How can an EHR bring public health and private medicine closer so that the common goals of the two somewhat disparate areas are met? A second significant factor in EHR success found in this study is physician leadership. Interestingly, physician leadership was a significant factor as a determinant of EHR success, yet where does a physician go to in order to learn these leadership skills? Which schools and their respective curricula didactically teach physicians leadership skills? Similarly, one may wonder, how important is physician leadership in other areas of health delivery including public health? Given today’s financial pressures on healthcare providers, it is very important to know how changes in the regulatory world will impact utilization of an EHR and how hospitals can assist private physicians in purchasing such technologies without ending up in violation of a Stark or Antikickback stature. The impact of the relaxation of the Stark Law, effective October of 2007, may have a positive impact on the deployment of EHRs into the physician practices. This is because hospitals can now discount EHR software, up to 85%, for the physician with the intent that patient care is going to improve. Given the overwhelmingly supported return-on-investment metrics, such as transcription and paper cost, faster patient throughput, and increased patient satisfaction, one can argue why physicians or hospitals are not investing in EHR technology already. Until there is an incentive to do so, such as higher reimbursement (or perhaps

143

Adoption of Electronic Health Records

diminished reimbursement for lack of not using an EHR), physicians do not want to proactively make such an investment. We presently see that government and private payers are exerting downward pressure on the reimbursements for nosocomial infections and for conditions that are present on admission. These are cases for which health care providers have traditionally billed for and have received payment. In the future, providers cannot even bill for certain cases, let alone, expect payment. In addition, with the wave of consumerism and transparency of price, quality, and cost, there is a doubt on how much of an impact these pressures will have on healthcare to have an EHR as a cost of doing business.

CONCLUSION The nature of this study is exploratory and studies EHR adoption at an enterprise level. Before building richer conceptual models on EHR adoption, this preliminary study confirms as well as adds to factors determining successful adoption of EHR. The findings of the study have authenticity since the panel of experts is the CIOs of healthcare organizations that academic researchers have not surveyed thus far. Although the physicians are the ultimate adopters of EHR, healthcare organizations and hospitals have an important role to play in EHR adoption. Hence this study using CIOs as an expert panel sheds light on the perspective of an important constituency on the adoption of EHRs.

Bakker, A. (2004). Access to EHR and access control at a moment in the past: a discussion of the need and an exploration of the consequences. International Journal of Medical Informatics, 73(3), 267. doi:10.1016/j.ijmedinf.2003.11.008 Baxley, E., & Campbell, T. (2008). Electronic health records in academic family medicine practices: a tale of progress and opportunity. Annals of Family Medicine, 6(1), 87–88. doi:10.1370/ afm.807 Berner, E. S., Detmer, D. E., & Simborg, D. (2005). Will the wave finally break? A brief view of the adoption of electronic medical records in the United States. Journal of the American Medical Informatics Association, 12(1), 3–7. doi:10.1197/ jamia.M1664 Chau, P. Y. K., & Hu, P. J.-H. (2002). Investigating healthcare professionals’ decisions to accept telemedicine technology: an empirical test of competing theories. Information & Management, 39(4), 297. doi:10.1016/S0378-7206(01)00098-2 Clements, P. C., Jones, L. G., McGregor, J. D., & Northrop, L. M. (2006). Getting there from here: a roadmap for software product line adoption. Communications of the ACM, 49(12), 33–36. doi:10.1145/1183236.1183261 Collins, T. (2004, October 5). Wirral lights the way to electronic records. Computer Weekly, 1416. Retrieved July 7, 2009 from Business Source Premier database.

REFERENCES

Connell, N. A. D., & Young, T. P. (2007). Evaluating healthcare information systems through an enterprise perspective. Information & Management, 44(4). doi:10.1016/j.im.2007.04.002

Ash, J. S., & Bates, D. W. (2005). Factors and forces affecting EHR system adoption: Report of a 2004 ACMI discussion. Journal of the American Medical Informatics Association, 12(1), 8–12. doi:10.1197/jamia.M1684

Cooper, G. F. (2005). Defining a workable strategy to stimulate widespread adoption of electronic health records in the United States. Journal of the American Medical Informatics Association, 12(1), 1–2. doi:10.1197/jamia.M1706

144

Adoption of Electronic Health Records

Dalkey, N., & Helmer, O. (1963). An experimental application of the Delphi method to the use of experts. Management Science, 9(3), 458–467. doi:10.1287/mnsc.9.3.458 Ford, E. W., Menachemi, N., Peterson, L. T., & Huerta, T. R. (2009). Resistance is futile: but it is slowing the pace of EHR adoption nonetheless. Journal of the American Medical Informatics Association, 16(3), 274–281. doi:10.1197/jamia. M3042 Gans, D., Kralewski, J., Hammons, T., & Dowd, B. (2005). Medical groups’ adoption of electronic health records and information systems. Health Affairs, 24(5), 1323–1333. doi:10.1377/ hlthaff.24.5.1323 Hayes, T. (2007). Delphi study of the future of marketing of higher education. Journal of Business Research, 60(9), 927–931. doi:10.1016/j. jbusres.2006.12.009 Institute of Medicine. (2001). Crossing the quality chasm: A new health system for the 21st century. Washington, DC: National Academy Press. Retrieved July 9, 2009 from http://www.nap.edu/ books/0309072808/html/index.html Jain, A., & Hayden, S. (2008). Using electronic health records to collect patient-specific performance measures and outcomes. In Proceedings of the AMIA Annual Symposium (pp. 1242-1243). Lowes, R., & Spikol, L. (2008). Fighting through the EHR. Medical Economics, 85(6), 58–64. Menachemi, N., Perkins, R. M., van Durme, D. J., & Brooks, R. G. (2006). Examining the adoption of electronic health records and personal digital assistants by family physicians in orida. Informatics in Primary Care, 14(1), 1–9.

Middleton, B., Hammond, W. E., Brennan, P. F., & Cooper, G. F. (2005). Accelerating U.S. EHR adoption: How to get there from here. Recommendations based on the 2004 ACMI retreat. Journal of the American Medical Informatics Association, 12(1), 13–19. doi:10.1197/jamia.M1669 Miller, R. H., West, C., Brown, T. M., Sim, I., & Ganchoff, C. (2005). The value of electronic health records in solo or small group practices. Health Affairs (Project Hope), 24(5), 1127–1137. doi:10.1377/hlthaff.24.5.1127 Nevo, D., & Chan, Y. E. (2007). A Delphi study of knowledge management systems: Scope and requirements. Information & Management, 44(6), 583–597. doi:10.1016/j.im.2007.06.001 Newton, W., Thompson, B., Campbell, T., Spencer, D., & Mast, C. (2006). Keeping Our Eye On The Ball: Managing the Evolution of Electronic Health Records. Annals of Family Medicine, 4(2), 184–185. doi:10.1370/afm.532 Poon, E. G., Blumenthal, D., & Jaggi, T. (2004). Overcoming barriers to adopting and implementing computerized physician order entry systems in U.S. hospitals. Health Affairs, 23(4), 184–190. doi:10.1377/hlthaff.23.4.184 Simon, S. R., Kaushal, R., Cleary, P. D., Jenter, C. A., Volk, L. A., & Poon, E. G. (2007). Correlates of electronic health record adoption in office practices: A statewide survey. Journal of the American Medical Informatics Association, 14(1), 110–117. doi:10.1197/jamia.M2187 Simon, S. R., McCarthy, M. L., Kaushal, R., Jenter, C. A., Volk, L. A., & Poon, E. G. (2008). Electronic health records: Which practices have them, and how are clinicians using them? Journal of Evaluation in Clinical Practice, 14(1), 43–47.

145

Adoption of Electronic Health Records

Vijayan, J. (2005, November 7). Federal Panel Pushes Plan for EHR Adoption. Computerworld, 39(45), 5–5.

Withrow, S. C. (2008). Why can’t physicians interoperate? barriers to adoption of EHRs. Healthcare Financial Management, 62(2), 90–96.

Walker, T. (2008). Report calls for more EHR privacy. Managed Healthcare Executive, 18(3), 9–9.

This work was previously published in International Journal of Healthcare Delivery Reform Initiatives (IJHDRI), Volume 2, Issue 1, edited by Matthew W. Guah, pp. 18-42, copyright 2010by IGI Publishing (an imprint of IGI Global).

146

147

Chapter 1.10

Technology Enabled Knowledge Translation:

Using Information and Communications Technologies to Accelerate Evidence Based Health Practices Kendall Ho University of British Columbia, Canada

ABSTRACT Because of the rapid growth of health evidence and knowledge generated through research, and growing complexity of the health system, clinical care gaps increasingly widen where best practices based on latest evidence are not routinely integrated into everyday health service delivery. Therefore, there is a strong need to inculcate knowledge translation strategies into our health system so as to promote seamless incorporation of

new knowledge into routine service delivery and education to promote positive change in individuals and the health system towards eliminating the clinical care gaps. E-health, the use of information and communication technologies (ICT) in health which encompasses telehealth, health informatics, and e-learning, can play a prominently supportive role. This chapter examines the opportunities and challenges of technology enabled knowledge translation (TEKT) using ICT to accelerate knowledge translation in today’s health system with two

DOI: 10.4018/978-1-60960-561-2.ch110

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Technology Enabled Knowledge Translation

case studies for illustration. Future TEKT research and evaluation directions are also articulated.

KNOWLEDGE TRANSLATION: INTRODUCTION The tenet of modern healthcare practice is evidence-based, established from knowledge generated through medical research and proven practice patterns. Evidence-based practice takes time to evolve. It is estimated that incorporating advances advocated by current research into routine, everyday medical practice takes one to two decades or more (Haynes, 1998; Sussman, Valente, Rohrbach et al., 2006). The causes of this apparent lag time of translating evidence into routine health practice are multifactorial, including but not restricted to: explosion of research and generation of resultant evidence, ineffective continuing education for health professionals to propagate the knowledge, lack of adoption of the knowledge by health professionals after exposure and education, the complexity of health management strategies that commonly demand more than simple changes in treatment approaches, reduction in healthcare resources, a lack of mutual understanding and dialogue between researchers that generated the research and health practitioners and health policy makers who need to translate the research into routine practices, and the practitioners’ and policy makers’ own beliefs and experiences that influence how knowledge will ultimately be utilized in clinical situations and quality assurance initiatives. As a result, a clinical care gap occurs when the best evidence is not routinely applied in clinical practice (Davis, 2006; Grol & Grimshaw, 2003).

Definition Canadian Institutes of Health Research (CIHR), one of the three members of the Canadian Research Tri-council and the guiding force in Canadian Health Research, defines knowledge translation

148

as “the exchange, synthesis, and ethically-sound application of knowledge, within a complex set of interactions among researchers and users, to accelerate the capture of the benefits of research for Canadians through improved health, more effective services and products, and a strengthened healthcare system” (CIHR, 2007). The Social Sciences and Humanities Research Council of Canada (SSHRC, 2006), another member of the Tri-council with focus on humanities research, defines knowledge mobilization as “moving knowledge into active service for the broadest possible common good.” SSHRC further contextually defines knowledge to be “…understood to mean any or all of (1) findings from specific social sciences and humanities research, (2) the accumulated knowledge and experience of social sciences and humanities researchers, and (3) the accumulated knowledge and experience of stakeholders concerned with social, cultural, economic and related issues” (SSHRC, 2006). Both definitions speak to the central principle of the need for not only discovering new knowledge through research, but also utilizing the resultant knowledge effectively and routinely in order to fully realize the benefits of the body of research. For the rest of this chapter, knowledge translation (KT) will be used to denote this core concept of effective knowledge application.

Strategic Considerations Strategically, effective and sustainable KT requires synchronized efforts at several levels towards a common vision of evidence based practice (Berwick, 2003; Katzenbach & Smith, 2005; Senge, 1994): the personal level where individuals influence their own behaviors towards change, the team level where groups of individuals work together collaboratively and cooperatively to drive towards group-based change, and the system level where health organizations and policy making bodies evolve and innovate on policies and establish

Technology Enabled Knowledge Translation

organizational patterns and cultures to motivate members towards change. Driving forces to motivate change at the individual level include: • •

•

•

Helping individuals to arrive at their own willingness and commitment to change Recognizing explicitly the contributions that individuals would make in carrying out the change Providing individuals with appropriate compensation, either monetary or otherwise, in making the change Sharing successful results with and giving feedback to the individuals after practice change has been instituted

Key factors to promote effective change at the team level include: • •

•

•

•

Jointly owning a shared vision towards an important goal Having effective overall leadership of the team, and also distributive leadership of the various individuals in the team and corresponding power and responsibility to drive change Sharing mutual trust with and accountability to each other in carrying out the necessary work Having an effective conflict resolution mechanism to bring differences; respectfully to the table for understanding, discussion, and resolution Achieving and celebrating success together

Important change management levers at the system’s level include: • •

Creating and adjusting fair and appropriate recognition and reward systems Bringing understanding to the impact of change in healthcare service delivery pat-

•

•

•

tern towards the social, economic, and population health context Cultivating the spirit of innovation to motivate individuals in the system to generate better evidence and pathways against current standards and practice patterns Promoting transfer of functions amongst individuals in the health system as effective division of labor and recognition of increasing competence through expansion of responsibilities Implementing routine system’s level reflection for continuous quality improvement and iterative modifications towards excellence

Success in sustainable KT requires not only transformation at the various levels, but also the harmonization of efforts in the totality of all these levels towards the common vision. For example, the April 2003 Institute of Medicine report advocates five core competencies that the health professionals of the 21st century need to possess (Institute of Medicine, 2003): delivering patient-centered care; working as part of interdisciplinary teams; practicing evidence based medicine; focusing on quality improvement; and using information technology. As an illustration, let us examine the core competency of interdisciplinary team establishment. In order to translate the concept of interdisciplinary team into routine health practice, having research that demonstrates examples of successful team based practice is not enough to cause lasting change in practice patterns. These successful examples, or documented knowledge, need to be vivified in health professionals’ own practice context so they and their own teams can visualize how they can model after these successful examples to replicate success (Ho et al., 2006). This type of education is necessary but not sufficient either; innovative policy translation by the policy makers to promote team based practice, patient and health consumer demand or preference for same, health

149

Technology Enabled Knowledge Translation

system redesign and implementation by health administrators, and appropriate accreditation for the educational system to promote the values and transform the curriculum for health professional trainees should all occur in synchrony in order to bring about lasting change, that is, sustainable knowledge translation. Finally, it is important to note that knowledge translation is not a straightforward and linear approach, but rather a complex and adaptive process based on common vision, solid principles, shared commitment at different levels, and human ingenuity in flexible adaptation.

Effective KT in Health It is desirable and achievable to accelerate knowledge translation in health to expeditiously reap the benefits health research to realize optimal evidence-based care for patients. Known and tested KT pathways, such as the Model of Improvement developed by Associates in Process Improvement (API, 2007) and endorsed by The Institute for Healthcare Improvement (IHI, 2007), can lead to successful KT in health system transformation through the following sequence of steps: • • • • • •

Setting aims to decide what accomplishments are to be achieved Establishing measure in order to assess positive change Selecting the key changes that will result in improvement Testing changes through the plan-dostudy-act cycle Implementing changes after testing Spreading changes to other organizations

Each of these steps require not only systems based mind-shift, but also individuals in the systems changing their personal behaviors. Therefore, in the health system context, health research can and ideally should be synchronized with individual practice, education, and policy setting environ-

150

ments so as to accelerate KT towards expeditious evidence based healthcare delivery.

ICT IN HEALTH Modern information and communication technologies (ICT), including computers, personal digital assistants, cellular phones, and an ever expanding list of imaginative electronic communication devices, are making unprecedented and innovative impact on healthcare service access, delivery, education and research (Ho, Chockalingam, Best, & Walsh, 2003). The rapid growth of affordable, interoperable ICTs that can facilitate seamless data communication and increase connectivity to the Internet are breaking down geographic and temporal barriers in accessing information, service, and communication. These advances are transforming the ways regional, national and global health services, surveillance, and education are being delivered. Some of the clear advantages of using ICT in health service delivery and education, or commonly referred to as e-Health, include but not limited to (Health Canada, 2006; Miller & Sim, 2004; Shortliffe, 1999): •

•

•

Anywhere, anytime access to accurate and searchable health information for knowledge and clinical case exchange, such as the use of ePocrates to help health professionals in rapidly accessing drug information to promote safe medication prescribing Large capacity for information storage and organization for health surveillance, such as the World Health Organization’s international health surveillance system to monitor infectious disease outbreaks in real time Ease of synchronous and asynchronous communication between health professionals in different geographic areas for health

Technology Enabled Knowledge Translation

service delivery, knowledge exchange or consultations

Defining E-Health When first introduced, the term “e-health” was used to signify health service delivery and activities on the Internet. Today, this term has been adopted to become an over-arching term to denote generally any use of ICT in healthcare (U Calgary Health Telematics Unit, 2005; Health Canada 2006), from health service delivery to health data storage and analysis to health education through ICT use. E-health can be conceptually visualized to be supported by three distinct but inter-related pillars: telehealth, health informatics, and e-learning (Figure 1). Telehealth commonly refers to “…the use of ICT to delivery health services, expertise, and information over distance” (U Calgary Health Telematics Unit, 2005). Whereas in the past, telehealth focused on the use of the videoconferencing medium as a distinguishing feature, today in many circles the emphasis is placed on the service delivery aspect of the definition. Therefore, telehealth can be either video-based through a closed network such as ISDN or fibreoptic videoconferencing, or Web-based through the use of the multimedia capabilities of the Internet. Telehealth can also be delivered either synchronous-

ly where communication between individuals occur in real time, or asynchronously in a “store and forward” fashion where one individual can send the information and expecting a response from others in a delayed fashion (Ho et al., 2004). Ample examples of telehealth can be found in the literature ranging from tele-psychiatry to teledermatology, tele-ophthalmology, emergency medicine, nursing, physiotherapy, and usage by other health disciplines. Health informatics (HI) have many definitions by different institutions, as documented on the University of Iowa Health Informatics Web site (U Iowa Health Informatics, 2005). One such typical definition of HI from Columbia University is “…the scientific field that deals with the storage, retrieval, sharing and optimal use of biomedical information, data, and knowledge for problem solving and decision making. It touches on all basic and applied fields in biomedical science and is closely tied to modern information technologies, notably in the areas of computing and communication.” The emphasis on HI, then, is on the storage and utilization of information captured through ICT. Excellent examples of the application of HI in practice are electronic health records in clinics and hospitals in a regional scale, public health on line disease surveillance systems nationally (Public Health Agency of Canada, 2006) or the World Health Organization epidemic and

Figure 1. E-health components

151

Technology Enabled Knowledge Translation

pandemic alert and response (EPR) system (World Health Organization, 2007). E-learning is commonly defined as “the use of electronic technology to deliver education and training applications, monitor learner performance, and report learner progress” (Sales, 2002). The distinguishing aspect of e-learning is the focus on the acquisition, utilization and evaluation of the knowledge captured by and synthesized through ICT. Examples of e-learning in health and their utility in changing health professionals’ practice is well documented in the literature. A recent example of a randomized control trial demonstrating the equivalence of e-learning compared to face to face workshops in helping learners in knowledge retention, with a statistical significance favoring e-learning in helping learners to actually change their behaviours (e.g., Fordis, 2005). While telehealth, health informatics, and elearning have their own distinguishing features and pillars of pursuits, they also synergistically interact to offer maximal benefits to the health system as a whole. On the contrary, each pillar on its own can only have limited impact on the health system. For example, trans-geographic telehealth without capturing outcomes through health informatics and e-learning to teach and propagate the service delivery module might only lead to temporary adoption of telehealth without sustaining effects to the health system as a whole. Similarly, health data capturing would not be complete without considering the contextual elements of health service delivery and the accumulated practice knowledge to date in the communities from which the data was generated. It is also obvious that e-learning will be dependent upon accurate data and best practices models in health service delivery. Therefore, the exciting challenge and opportunity of e-health is indeed in the seamless and comprehensive and complimentary utilization of data, service and knowledge to drive the prospective transformation of health practices that are based on evidence, knowledge,

152

Figure 2. Optimizing TEKT

and the needs of the health consumers and the communities (Figure 2).

TECHNOLOGY ENABLED KNOWLEDGE TRANSLATION E-health is rapidly gaining momentum worldwide as a vital part of the healthcare system in and amongst nations. For example, National Health Services in the United Kingdom, in Australia, and Infoway in Canada are actively facilitating the establishment of infrastructure and implementation strategies in e-health to promote its entrenchment in health practices. Collaboration amongst different agencies and organizations is also evident in establishing emerging e-health networks in other countries such as the Africa Health Infoway (World Health Organization, 2006). Recognizing the potential, the UN General assembly categorically stated in its Millennium declaration that “… the benefits of new technologies, especially information and communication technologies … (should be made) available to all” (United Nations, 2000). When considering e-health in practice, one needs to question not only “how do we augment or replace existing services with ICT,” but also “how can ICT be innovatively used to best serve the health system that may not have any

Technology Enabled Knowledge Translation

precedence?” This question can further be asked in two ways: “How can ICT be used to make current health service and education pathways more efficient, accessible, or higher quality?” and the companion question “How can ICT be used to provide unprecedented health service and educational models that were not possible in the past?” These questions guide us towards both adoption and innovation in e-health, and also properly consider ICT as serving the agenda of the health system, rather than using the latest and greatest technologies regardless of whether or not the usage actually improves health services and education.

TEKT Defined Technology enabled knowledge translation (TEKT) is defined as the use of ICT to accelerate the incorporation of evidence and knowledge into routine health practices (Ho et al., 2003). By definition, TEKT is not only about using ICT to achieve one type of purpose, such as telehealth for service delivery or health informatics for data storage and analysis. Rather, TEKT strategies synchronize and coordinate data, service, and knowledge capturing and utilization to synergistically cause a system’s level change so as to translate evidence and health knowledge into routine practice and policy establishment. ICT can play a pivotal role in the health system for knowledge synthesis, evidence-based decisionmaking, building shared capacity for knowledge exchange, and minimization of duplication of decision support systems. Various Web-based data and information repositories networked together can facilitate just-in-time clinical consultations, and support access to the latest management information on diseases, treatments, and medications. Instant sharing and exchange of knowledge by healthcare providers facilitated in these Internet portals as discussion groups or informal e-mail exchange can play an important role in team building. Remote access to centralized data

repositories, such as electronic medical records via the Internet as well as intelligent information retrieval functionality and data pattern recognition are just some examples of the ways in which technologies can save time, eliminate laborious tasks, and interconnect to capture, disseminate, and help translate knowledge into practice in ways previously not possible.

Case Studies in TEKT The following section highlights two case studies in TEKT to provide a qualitative illumination on best practices in TEKT. However, this discussion is not meant to be comprehensive or exhaustive, but rather illustrations to highlight and celebrate the innovation and ingenuity of the applications of ICT to facilitate TEKT and improved health outcome as they meet the challenges and needs of the healthcare system with existing and emerging technological solutions. SARS (Srinivasan, McDonald, Jernigan et al., 2004; Marshall, Rachlis, & Chen, 2005; Wenzel, Bearman, & Edmond, 2005) From late 2002 to 2003, the world was gripped by the emergence of a deadly and, up till then, unknown health threat sudden acute respiratory syndrome (SARS). This rapidly spreading infection with associated high mortality made its notorious impact worldwide into early 2003. In particular the South East Asian countries were most severely affected, with transmission to other continents due to mobility of infected populations. The eventual toll of SARS was recorded to be 8,098 cases worldwide, with 774 deaths. During that period of outbreak, the many unknown features of SARS needed to be rapidly disseminated to health professionals and administrators around the world. Also, tracking of the spread of the infection, and patterns of spread to understand the modes of transmission and contact persons involved were paramount to bring effective control of this outbreak. SARS related infection control policies, criteria of diagnosis,

153

Technology Enabled Knowledge Translation

education of health professionals and the general public, quality assurance activities were all vital information that were rapidly disseminated and exchanged through the use of ICT worldwide. Data repositories and electronic systems, together with Web-based information dissemination and consultations, were vital aspects of global SARS management and decision support for health professionals worldwide. The urgent and intensive efforts to disseminate information, sharing of best practices in infection control methods, and careful preparation of unaffected counties were key lessons learned, and ICT playing key roles in TEKT were pivotal in these activities. Also, as a result of SARS, great attention is paid in different countries and globally on disease monitoring and surveillance systems to prepare for future expected and unexpected outbreaks such as influenza, avian flu, or other diseases. Medication Safety Surveillance (Bell et al., 2004; Graham, Campen, Hui et al., 2005; Leape & Berwick, 2005; Topol, 2004) Non-steroidal anti-inflammatory drugs (NSAIDs) is a class of medications effective for pain management in patients with arthritis. However, NSAIDs are known to have substantial side effects in the gastroenteral system in causing ulcers and erosions. A new class of COX-II NSAIDs were introduced, the first one of which were rofecoxib (Vioxx®). Initial trials seemed to affirm that COX-II NSAIDs were effective for pain management and had lower gastrointestinal side effects compared to traditional NSAIDs. However, subsequent analysis of the data, with additional studies done by other groups, suggested the potential of increased cardiovascular side effects including myocardial infarctions and deaths. Kaiser Permanente, an American Health Management Organization, wanted to clarify this controversy. It had an electronic health record system (HER) where, amongst a variety of health and administrative records, every doctor would track patient visits over time, medication prescriptions, and side effects. Using the Cali-

154

fornia database of this EHR system, 2,302,029 person-years follow-up over a three year period where patients were exposed to rofecoxib were analyzed. The researchers found that there were 8,143 cases of serious coronary heart disease that occurred, with 2,210 cases (27.1 percent) where the patients died. This represented a more than three times risk (Odd ratio as high as 3.58) for the use of rofecoxib compared to another NSAID agent. This data, together with other studies, led to the company that made rofecoxib voluntarily withdrew the medication from the market in September 2004, and the Food and Drug Administration (FDA) officially recommending its withdraw in February 2005. A very significant development in this medication surveillance was that, once Kaiser Permanente detected the significant cardiovascular side effect risk of rofecoxib, this information was passed onto physicians practicing in Kaiser, leading to a dramatic drop of rofecoxib prescription rate of four percent compared to the national average in United States of 40 percent, well before the medication was withdrawn from the market place. This case demonstrates the power of EHR in not only being able to rapidly and prospectively track medication side effects so as to increase safety, but also disseminating the evidence to practitioners to influence practice outcome rapidly.

CHALLENGES AND OPPORTUNITIES IN TEKT Barriers to ICT Uptake Despite the potential and real benefits of ICT use in health, health professionals’ uptake of ICT has been slow. Factors that impede the adoption of ICT tools include cost, lack of informatics platform standards, physicians’lack of time or technological literacy, the need for a cultural shift to embrace these necessary changes in medicine, and a lack

Technology Enabled Knowledge Translation

proaches to evidence based knowledge translation or create unprecedented models Understanding the human-technology interface how ICT can be configured to optimize the utilization and practice of these tools in healthcare contexts Understanding and assisting in evidence based e-health policy making, as policy innovation is essential to guide the optimal use of ICT in the system’s level Integrating ICT into teams and communities where human to human interactions and collaborations can be enhanced through ICT facilitation Cost effectiveness and return on investment evaluation as to how ICT can lead to increased capacity of the health system, cost avoidance or savings in health service delivery, or improved access and quality Building capacity in TEKT research over time

of integration of various ICT methodologies into a cohesive deployment strategy. For example, in a recent survey supported by The Commonwealth Fund (Schoen, Osborn, Huynh et al., 2006), the authors found that there was a wide variation of ICT uptake by primary care physicians amongst these seven countries, from as low as 23 percent to as high as 98 percent uptake in electronic patient medical records. This variation in electronic patient record uptake directly correlated with and underpinned issues related to quality and efficiency broached in this survey, such as coordination of care of patients, multifunctional capacity including automated alerts and reminders, or information sharing amongst interprofessional team members. Of note, both United States and Canada were lagging significantly behind the other five countries surveyed in terms of ICT uptake in practice. These variations were in large part due to the underlying policy choices of the different countries. Therefore, in consideration as to how best to overcome barriers to technology uptake, it is important to not only focus on health professionals to increase their attitudes, knowledge and skills in ICT use, but also place emphasis on policy innovation to motivate the health systems towards quality and the adoption of ICT in support of this important vision of care.

In order to accelerate the discipline of TEKT, it is important that efforts in this research and innovation be harmonized to enable cross study comparisons and standardized documentation of best practices. In this line of thinking, establishing a research evaluation framework towards TEKT would be an ideal approach (Ho et al., 2004).

FUTURE RESEARCH DIRECTIONS

CONCLUSION

As TEKT is an emerging field with a rapidly evolving environment, there is ample opportunity for research, innovation, and evaluation. Examples of dimensions of TEKT research could include but not restricted to:

ICT has tremendous potential to improve healthcare service delivery, and TEKT can help accelerate ICT adoption and change management to reap the corresponding benefits. Excellent literature based and practice based examples of TEKT have shone some best practice examples, and more innovative and effective models in the future are sure to come with the continuing improvement of technologies and practices. As a result, the practice of and research in TEKT are both timely

• •

Documentation of best practices in TEKT to date Innovative demonstrations of ICT enabled models of knowledge translation, where ICT is used to either augment current ap-

•

•

•

•

•

155

Technology Enabled Knowledge Translation

and urgently needed to help achieve excellence in healthcare delivery.

REFERENCES Associates in Process Improvement–API. (2007). The model for improvement. Austin, TX, Detroit, MI, Sacramento, CA, Washington, DC. Retrieved June 7, 2007, from http://www.apiWeb.org/ API_home_page.htm Bell, D. S., Cretin, S., Marken, R. S., & Landman, A. (2004). A conceptual framework for evaluating outpatient electronic prescribing systems based on their functional capabilities. Journal of the American Informatics Association, 11, 60–70. doi:10.1197/jamia.M1374 Berwick, D. M. (2003). Disseminating innovations in health care. Journal of the American Medical Association, 289(15), 1969–1975. doi:10.1001/ jama.289.15.1969 Canadian Institutes of Health Research. (2007). Knowledge translation strategy 2004-2009: Innovation in action. Retrieved June 9, 2007, from http://www.cihr-irsc.gc.ca/e/26574.html#defining Davis, D. (2006). Continuing education, guideline implementation, and the emerging transdisciplinary field of knowledge translation. The Journal of Continuing Education in the Health Professions, 26(1), 5–12. doi:10.1002/chp.46 Fordis, M., King, J. E., & Ballantyne, C. M. (2005). Comparison of the instructional efficacy of Internet-based CME with live interactive CME workshops: A randomized controlled trial. Journal of the American Medical Association, 294(9), 1043–1051. doi:10.1001/jama.294.9.1043

156

Graham, D. J., Campen, D., Hui, R., Spence, M., Cheetham, C., & Levy, G. (2005). Risk of acute myocardial infarction and sudden cardiac death in patients treated with cyclo-oxygenzse 2 selective and non-selective non-steroidal antiinflammatory drugs: Nested case-control study. Lancet, 365(9458), 475–481. Grol, R., & Grimshaw, J. (2003). From best evidence to best practice: Effective implementation of change in patients’ care. Lancet, 362, 1225–1230. doi:10.1016/S0140-6736(03)14546-1 Haynes, R. (1998). Using informatics principles and tools to harness research evidence for patient care: Evidence-based informatics. Medinfo, 9(Pt 1Suppl), 33–36. Health Canada. (2006). eHealth. Retrieved on May 4, 2007, from http://www.hc-sc.gc.ca/hcssss/ehealth-esante/index_e.html Ho, K., Bloch, R., Gondocz, T., Laprise, R., Perrier, L., & Ryan, D. (2004). Technology-enabled knowledge translation: Frameworks to promote research and practice. The Journal of Continuing Education in the Health Professions, 24(2), 90–99. doi:10.1002/chp.1340240206 Ho, K., Borduas, F., Frank, B., Hall, P., HandsfieldJones, R., Hardwick, D., et al. (2006). Facilitating the integration of interprofessional education into quality healthcare: strategic roles of academic institutions. Report to Health Canada Interprofessional Education for Collaborative Patient Centred Practice. October 2006. Ho, K., Chockalingam, A., Best, A., & Walsh, G. (2003). Technology-enabled knowledge translation: Building a framework for collaboration. Canadian Medical Association Journal, 168(6), 710–711.

Technology Enabled Knowledge Translation

Ho, K., Karlinsky, H., Jarvis-Selinger, S., & May, J. (2004). Videoconferencing for Telehealth: unexpected challenges and unprecedented opportunities. British Columbia Medical Journal, 46(6), 285–289. Institute for Healthcare Improvement. (2007). How to improve: improvement methods. Retrieved on June 7, 2007 from http://www.ihi.org/ IHI/Topics/Improvement/ImprovementMethods/ HowToImprove/ Institute of Medicine of the National Academies. (2003). Health professions education: a bridge to quality. Institute of Medicine April 8, 2003 Report. Retrieved June 9, 2007, from http://www.iom.edu/ CMS/3809/4634/5914.aspx Katzenbach, J. R., & Smith, D. K. (2005). reprint). The discipline of teams. Harvard Business Review, 83(7), 162–171. Leape, L. L., & Berwick, D. M.(n.d.). Five years after ‘to err is human’: What have we learned? Journal of the American Medical Association, 293(19), 2384–2390. doi:10.1001/ jama.293.19.2384 Marshall, A. H., Rachlis, A., & Chen, J. (2005). Severe acute respiratory syndrome: responses of the healthcare system to a global epidemic. Current Opinion in Otolaryngology & Head & Neck Surgery, 13(3), 161–164. doi:10.1097/01. moo.0000162260.42115.b5 Miller, R. H., & Sim, I. (2004). Physicians’ use of electronic medical records: Barriers and solutions. Health Affairs, 23(2), 116–126. doi:10.1377/ hlthaff.23.2.116 Public Health Agency of Canada. (2006). Disease surveillance on-line. Retrieved June 7, 2007, from http://www.phac-aspc.gc.ca/dsol-smed/ Sales, G. C. (2002). A quick guide to e-learning. Andover, MN: Expert Publishing.

Schoen, C., Osborn, R., Huynh, P. T., Doty, M., Peugh, J., & Zapert, K. (2006). On the front line of care: primary care doctors’ office systems, experiences, and views in seven countries. Health Affairs 25(6), w555-w571. Retrieved June 9, 2007, from http://content.healthaffairs.org/cgi/content/ abstract/hlthaff.25.w555?ijkey=3YyH7yDwrJSo c&keytype=ref&siteid=healthaff Senge, P. M. (1994). The fifth discipline: The art & practice of the learning organization. New York: Doubleday. Shortliffe, E. H. (1999). The evolution of electronic medical records. Academic Medicine, 74(4), 441– 419. doi:10.1097/00001888-199904000-00038 Social Sciences and Humanities Research Council of Canada. (2006). Knowledge Impact on Society: A SSHRC Transformation Program. Retrieved June 9, 2007, from http:// http://www.sshrc.ca/ Web/apply/program_descriptions/knowledge_ impact_e.asp Srinivasan, A., McDonald, L. C., & Jernigan, D. (2004). Foundations of the severe acute respiratory syndrome preparedness and response plan for healthcare facilities. Infection Control and Hospital Epidemiology, 25(12), 1020–1025. doi:10.1086/502338 Sussman, S., Valente, T. W., Rohrbach, L. A., Skara, S., & Pentz, M. A. (2006). Translation in the health professions: Converting science into action. Evaluation & the Health Professions, 29(1), 7–32. doi:10.1177/0163278705284441 Topol, E. J. (2004). Failing the public healthRofecoxid, Merck and the FDA. The New England Journal of Medicine, 351(17), 1707–1709. doi:10.1056/NEJMp048286 United Nations Millennium Declaration. (2000). Section III. 19 of the General Assembly resolution 55/2 of 8 September 2000. Retrieved June 9, 2007, from http://www.ohchr.org/english/law/ millennium.htm

157

Technology Enabled Knowledge Translation

University of Calgary Health Telematics Unit. (2005). Glossary of telhealth related terms, acronyms and abbreviations. Retrieved June 7, 2007, from http://www.fp.ucalgary.ca/telehealth/ Glossary.htm University of Iowa Health Informatics. (2005). What is Health Informatics? Retrieved June 7, 2007 from http://www2.uiowa.edu/hinfo/academics/what_is_hi.html Wenzel, R. P., Bearman, G., & Edmond, M. B. (2005). Lessons from severe acute respiratory syndrome (SARS): Implications for infection control. Archives of Medical Research, 36(6), 610–616. doi:10.1016/j.arcmed.2005.03.040 World Health Organization. (2006). WHO Partnership—The Africa Health Infoway (AHI). Retrieved June 7, 2007 from http://www.research4development.info/projectsAndProgrammes. asp?ProjectID=60416 World Health Organization. (2007). Epidemic and pandemic alert and response (EPR). Retrieved June 7, 2007 from http://www.who.int/csr/en/

ADDITIONAL READING Booz, A. Hamilton (2005). Pan-Canadian electronic Health record: Quantitative and qualitative benefits. Canada Health Infoway’s 10-year investment strategy costing. March 2005 Report. Retrieved June 7, 2007 from http://www.infowayinforoute.ca/Admin/Upload/Dev/Document/ VOL1_CHI%20Quantitative%20&%20Qualitative%20Benefits.pdf Bower, A. G. (2005). The diffusion and value of healthcare information technology. A RAND Report. Accessible at http://www.rand.org/pubs/ monographs/MG272-1/

158

Canadian Institutes of Health Research. (2007). Knowledge Translation Strategy 2004-2009: Innovation in Action. Retrieved June 9, 2007, from http://www.cihr-irsc.gc.ca/e/26574.html#defining eHealth ERA Report. (2007). eHealth priorities and strategies in European countries. European Commission Information Society and Media. Retrieved June 9, 2007, from http://ec.europa. eu/information_society/activities/health/docs/ policy/ehealth-era-full-report.pdf Fonkych, K., & Taylor, R. (2005). The state and pattern of health information technology adoption. A RAND report. Accessible at http://www.rand. org/pubs/monographs/MG409/ Ho, K., Bloch, R., Gondocz, T., Laprise, R., Perrier, L., & Ryan, D. (2004). Technology-enabled knowledge translation: Frameworks to promote research and practice. The Journal of Continuing Education in the Health Professions, 24(2), 90–99. doi:10.1002/chp.1340240206 Ho, K., Chockalingam, A., Best, A., & Walsh, G. (2003). Technology-enabled knowledge translation: Building a framework for collaboration. Canadian Medical Association Journal, 168(6), 710–711. Institute for Healthcare Improvement. (2007). How to improve: Improvement methods. Cambridge, MA. Retrieved on June 7, 2007, from http://www.ihi.org/IHI/Topics/Improvement/ ImprovementMethods/HowToImprove/ Schoen, C., Osborn, R., Huynh, P. T., Doty, M., Peugh, J., & Zapert, K. (2006). On the front line of care: primary care doctors’ office systems, experiences, and views in seven countries. Health Affairs 25(6), w555-w571. Retrieved June 9, 2007, from http://content.healthaffairs.org/cgi/content/ abstract/hlthaff.25.w555?ijkey=3YyH7yDwrJSo c&keytype=ref&siteid=healthaff

Technology Enabled Knowledge Translation

Stroetmann, K. A., Jones, T., Dobrev, A., & Stroetmann, V. N. (2006). eHealth is worth it: The economic benefits of implemented eHealth solutions at ten European sites. Awww.ehealth-impact. orgreport. Retrieved June 7, 2007 from http:// ec.europa.eu/information_society/activities/ health/docs/publications/ehealthimpactsept2006. pdf

Sussman, S., Valente, T. W., Rohrbach, L. A., Skara, S., & Pentz, M. A. (2006). Translation in the health professions: Converting science into action. Evaluation & the Health Professions, 29(1), 7–32. doi:10.1177/0163278705284441

This work was previously published in Human, Social, and Organizational Aspects of Health Information Systems, edited by Andre W. Kushniruk and Elizabeth M. Borycki, pp. 301-313, copyright 2008 by Medical Information Science Publishing (an imprint of IGI Global).

159

160

Chapter 1.11

The Computer-Assisted Patient Consultation: Promises and Challenges Aviv Shachak University of Toronto, Canada Shmuel Reis Technion- Israel Institute of Technology, Israel

ABSTRACT The implementation of electronic health records (EHRs) holds the promise to improve patient safety and quality of care, as well as opening new ways to educate patients and engage them in their own care. On the other hand, EHR use also changes clinicians’ workflow, introduces new types of errors, and can distract the doctor’s attention from the patient. The purpose of this chapter is to explore these issues from a

micro-level perspective, focusing on the patient consultation. The chapter shows the fine balance between beneficial and unfavorable impacts of using the EHR during consultations on patient safety and patient-centered care. It demonstrates how the same features that contribute to greater efficiency may cause potential risk to the patient, and points to some of the strategies, best practices, and enabling factors that may be used to leverage the benefits of the EHR. In particular, the authors point to the role that medical education should play in preparing practitioners for the challenges

DOI: 10.4018/978-1-60960-561-2.ch111

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The Computer-Assisted Patient Consultation

of the new, computerized, environment of 21st century medicine.

INTRODUCTION Mrs. Jones is a new patient to the practice who has come in for a certificate that the local gym requires for enrollment. She is 50 years old, married and mother of two, who works as a high school teacher. Dr. Smith introduces himself, welcoming her to the practice. He asks for her USB memory stick, and her cumulative electronic health record (EHR) pops-up on his screen, indicating no significant health concerns. Her family tree (genogram) is also generated and is displayed. Dr. Smith asks “anything else?” and she answers that she would like to be informed about needed health promotion advice. Her family history contains the existence of heart disease (coronary artery disease) and breast cancer. Her life style is generally healthy. Dr. Smith performs a focused physical examination, and hooks her up to the multi-task physiological monitor that gives, within 90 seconds, a reading of her blood pressure and pulse, electrocardiogram, and lung function (spirometry) - all are normal. Since she had laboratory tests three months ago as part of a woman’s health program she follows, Dr Smith shares with her the results of her cholesterol test (hypercholesterolemia or a high cholesterol level). The EHR automatically generates the recommended screening and health promotion recommendations for her age and risk status, which he explains, and then goes over some patient education materials that appear on the screen that he shares with her. Finally, she asks his opinion about the future consequences of her ten year old infection of the abdomen’s inner lining (peritonitis) and a subsequent surgical procedure to explore it (laparotomy). The doctor does not recall any, but he sends a query through the Inforetriever (a program for updated sound medical information retrieval) on his desktop. When the evidence-based answer arrives 15 sec-

onds later, he is able to share it with her. All the generated materials are beamed to her cell phone and emailed to her, together with the certificate. Throughout the encounter Dr. Smith has been applying the patient-doctor–computer communication skills he had trained in six months ago at the national simulation center. The subsequent results of her age and risk appropriate screening arrive at his desktop (and hers) automatically within the next week. Dr. Smith interprets the results for her and sends them over the encrypted office email. Mrs. Jones also shares her exercise and diet program data electronically, and he monitors those. Three months later, a new cholesterol screen indicates that she has much improved. A week later Dr. Smith receives the annual report of his performance: the clinical quality indicators and patient safety monitoring show another 5 point improvement. The patient satisfaction survey is at its usual high. When Dr. Smith sits to plan his next year needs-based goals for his continuous learning plan he chooses tropical and poverty medicine. The bonus he will receive will enable him to finally choose the medical relief to Africa he has been hoping to accomplish for some years now. To some readers, the above scenario may seem imaginary. However, the technologies which enable it have already been, or are being, developed. As this scenario demonstrates, the application of information and communication technology (ICT) in health care holds the promise to improve quality of care, as well as opening new ways to educate patients and engage them in their own care. However, there are also many challenges involved. In this chapter, we will review the present literature on the benefits of the computerized consultation, as well as the challenges and problems associated with it. We will discuss the fine line between benefits and risks of using the EHR during consultations, and demonstrate how the same features that contribute to efficiency may pose a risk or interfere with patient centeredness. Finally, we will discuss some of the strategies, best

161

The Computer-Assisted Patient Consultation

practices, and enabling factors that could enhance realization of the vision. In particular, we will point to the role that medical education should play in preparing practitioners for the challenges of the new, computerized, environment of 21st century medicine.

BACKGROUND The potential and actual outcomes of ICT in health care are often discussed at the systems level. In particular, impacts on quality of care and patient safety of ICTs such as the EHR, computerized provider order entry (CPOE) and clinical decision support systems (CDSS) have been examined. Table 1 describes these and some other commonly used clinical information systems. Although the majority of studies have emerged from a small number of large health care organizations such as the Department of Veterans Affairs (DVA) or

Kaiser Permanente in the US, a recent systematic review has concluded that health information systems can indeed improve quality of care and patient safety. Preventive care was the primary domain of quality improvement reported. The main benefits were increased adherence to guidelines, enhanced surveillance, and monitoring, and decreased medication errors. The major efficiency improvement was decreased utilization of care (Chaudhry et al., 2006). However, a growing number of studies - especially of CPOE systems - have drawn attention to some of the unintended consequences of ICT in health care that may be beneficial or adverse. Such unintended consequences include more or new work for clinicians, unfavorable changes in workflow, high system demands, problems related to paper persistence, untoward changes in communication patterns, negative emotions, new kinds of errors, unexpected changes in the power structure, and overdependence on the technology

Table 1. Types of commonly used clinical information systems. Type of system

Description

Electronic Medical Record (EMR)

An electronic record of health-related information on an individual that can be created, gathered, managed, and consulted by authorized clinicians and staff within one health care organization. The National Alliance for Health Information Technology (NAHIT), 2008

Electronic Health Record (EHR)

An electronic record of health-related information on an individual that conforms to nationally recognized interoperability standards and that can be created, managed, and consulted by authorized clinicians and staff across more than one health care organization. The National Alliance for Health Information Technology (NAHIT), 2008). The terms EMR and EHR are sometimes used interchangeably.

Personal Health Record (PHR)

An electronic record of health-related information on an individual that conforms to nationally recognized interoperability standards and that can be drawn from multiple sources while being managed, shared, and controlled by the individual. The National Alliance for Health Information Technology (NAHIT), 2008.

Computerized Provider Order Entry (CPOE)

A system which allows providers to post and communicate medical orders and their application. CPOE can facilitate patient safety and quality of care by means of eliminating illegible handwriting and use of predefined, evidence-based, order sets for specific medical conditions.

Clinical Decision Support Systems (CDSS)

Computerized systems which are designed to assist health care professionals in making clinical decisions (e.g. about diagnosis, treatment or care management). CDSS may be characterized by their intended function, the mode by which they offer advice, style of consultation and the underlying decision making process (Musen, Shahar & Shortliffe, 2006)

Picture Archiving and Communication System (PACS)

Computerized information system which is used to acquire, store, retrieve and display digital images; in particular diagnostic images such as x-rays, Computerized Tomography (CT) and Magnetic Resonance Imaging (MRI) images.

162

The Computer-Assisted Patient Consultation

(Campbell, Guappone, Sittig, Dykstra, & Ash, 2009; Campbell, Sittig, Ash, Guappone, & Dykstra, 2006). As system-level outcomes of information systems are thought to emerge from the combination of multiple individual impacts (DeLone & McLean, 1992; 2003), through the rest of this chapter we will explore the impact of ICT in health care at a micro-level, focusing on the patient consultation. In particular we argue that there is a fine balance between benefits and risks of using EHR during consultation, which is greatly influenced by cognitive factors, and that the computer has become a third actor in the clinical encounter and thus influences patient-doctor relationships.

THE COMPUTER ENABLED (OR DISABLED) CONSULTATION: AN IN-DEPTH LOOK Efficiency vs. Risk One of the most serious reported unintended consequences of CPOE systems is the generation of new kind of errors. It has been termed “e-iatrogenesis” (Weiner, Kfuri, Chan, & Fowles, 2007). A deeper look at how these new types of errors arise, demonstrates that the line between greater efficiency and error is often very thin, and that sometimes the same features and user characteristics that make work more efficient, and presumably safer, are those which generate computer-related errors. For example, in the EHR there are often dropdown lists, e.g. for patients’ names, diagnoses, or medications. Structuring data this way is very useful in that it enables data analysis for health management, administration, and research purposes. “Point and click” also eliminates the need to type in data, which is time consuming and may divert the doctor’s attention from the patient. Finally, these lists may serve as a quick reminder and thus minimize memory load (Shachak, Hadas-Dayagi,

Ziv, & Reis, 2009). However, the structured format of entering data into the EMR is very different from the narrative structure of the traditional patient record. It has been demonstrated that moving to an EMR system affected physicians’ information gathering and reasoning processes, and could lead to potential loss of information (Patel, Kushniruk, Yang, & Yale, 2000). Furthermore, experienced users perform selection from lists very quickly, in a semi-automatic manner. This automaticity, the term used in cognitive science to describe skilled performance that requires little or no conscious attention (Wheatley & Wegner, 2001), makes it very easy to accidentally select the wrong item. Two of the most commonly reported errors with a primary care electronic medical record (EMR) system—selecting the wrong medication and adding to the wrong patient’s chart—were associated, fully or in part, with selection from lists (Shachak et al., 2009). Similar errors were also reported with hospital CPOE systems, suggesting this is a universal problem (Campbell et al., 2009). Other examples are the use of copy-paste and user- (or system-) generated templates. The computer allows quick copying and pasting of data, e.g. from a previous to the present visit, or from one patient’s chart to another. Undoubtedly, copy-paste and templates make an efficient use of computers; however, they raise ethical questions and concerns about data quality (Thielke, Hammond, & Helbig, 2007). Moreover, copying and pasting poses the risk of inducing errors by accidental failure to update data which should be modified. In a cross-sectional survey held at two medical centers that use computerized documentation systems physicians’ overall attitude toward copying and pasting was positive. However, they noted that inconsistencies and outdated information were more common in copied notes (O’Donnell et al., 2009). Thielke et al. (2007) have reported a steep rate of copy-paste examination data in EHR, including notes copied from another author or from a document older than six months, which they defined as high risk.

163

The Computer-Assisted Patient Consultation

Similarly, templates may be inserted into the patient chart using keyboard shortcuts or a mouse click. Templates make work more efficient as they eliminate the need for extensive typing. Furthermore, if comprehensive templates are prepared they may serve as checklists that help clinicians make sure they have collected all required information, performed the necessary tests, or provided recommended treatment. However, the use of templates, too, can become semi-automatic. As in copy-paste, it is very easy to approve a template without correcting wrong data - “clicking enter, enter, enter is a prescription for error” as one of the physician participants in our study described it (Shachak et al., 2009). Finally, alerts of drug contraindications can be beneficial or detrimental. Clearly, notifying clinicians that the patient is allergic to the medication they had just prescribed, or that it interacts with another drug the patient takes, is important. A number of studies have attributed significant improvements in medication safety to the embedding of alerts in CPOE systems; these improvements include reduced numbers of medication errors and adverse drug events (Bates et al., 1999; Eslami, Keizer, & Abu-Hanna, 2007; Tamblyn et al., 2003). However, others have reported alert override rates greater than 80% in various systems and settings (Hsieh et al., 2004; Weingart et al., 2003). Why are alerts being overridden? Prescribers provide various reasons such as poor quality and high volumes of alerts, patients already taking the drug in question, and the monitoring of patients (Grizzle et al., 2007; Hsieh et al., 2004; Lapane, Waring, Schneider, Dube, & Quilliam, 2008; van der Sijs, Aarts, Vulto, & Berg, 2006; Weingart et al., 2003). A recent study also suggests that some specified reasons for alert override typed into a CPOE system were, in fact, failed attempts to communicate important medical information which no one has read (Chused, Kuperman, & Stetson, 2008). In addition to causing interference or increasing cognitive load, we propose that poor quality and too high a volume of alerts may result

164

in formation of an automatic behavior to instantly shut down alerts, similar to the way users attempt to immediately close pop-up ads on websites (McCoy, Galleta, Everard, & Polak, 2004). This automatic behavior—or “alert fatigue” as van der Sijs et al. (2006) describe it—may accidentally result in overriding crucial alerts. How can the risk-benefit balance shift to the benefits side? A key component in systems improvement is usability and human factors engineering. User interfaces need to be designed and evaluated with an understanding of these cognitive elements, in an attempt to reduce effort on one hand and prevent automaticity-related errors on the other. Examples include drop-down lists with greater distance between items, or highlighting items on mouse-over just before selection. Alerts acceptance can be improved by tiering alerts on the basis of their severity, and designating only critical alerts as interruptive (Paterno et al., 2009; Shah et al., 2005). To some extent, the system itself can detect and prevent errors. For example, one of the systems we had checked accepted an input that a three year old patient had been smoking for 15 years. When a doctor makes this error, it is likely s/he is adding to the wrong chart. Such errors are preventable by better system design. Another key factor is education or training. Our findings suggest that experts are aware of these potential errors and, therefore, may be more careful when performing risky actions (Shachak et al., 2009). Education that goes beyond the technical aspects of using the system to a broader view of the benefits, risks, and principles of high-quality use is required. Simulation-based training, in particular, may be highly effective.

The Computer as a Third Player in the Patient-Doctor Encounter In our initial scenario, the computer is an active actor in the consultation. It triggers some of the discussion and information exchange, and acts as a colleague or advisor. Indeed, in computerized

The Computer-Assisted Patient Consultation

settings, the consultation cannot be viewed as dyadic patient-doctor interaction anymore. Rather, it is now triadic relationships of the patient, the doctor, and computer (Pearce, Dwan, Arnold, Phillips, & Trumble, 2009). The computer influences the consultation in various ways, beginning with the physical location of the monitor, keyboard, mouse, and other ancillary devices such as printers. The space these hardware occupy, their location, and orientation have significant influence on the consultation. For example, large fixed monitors may narrow the doctor’s or the patient’s personal space and interfere with eye contact. Even when it does not, the monitor in some settings is oriented towards the doctor only, which excludes the patient from the doctor’s interaction with the computer. For some patients, this may be disturbing (Frankel et al., 2005; Pearce, Walker, & O’Shea, 2008). However, the other option of a monitor that can be viewed by both doctor and patient is also not optimal for everyone. In our opening scenario, both Dr. Smith and Mrs. Jones seem to be comfortable with technology, so the computer becomes a useful tool to facilitate their discussion, share information, and help with decision making. However, this is not the case for all doctors and patients. As Pearce and others have demonstrated, whether the monitor’s position promotes information sharing or becomes distracting depends on both the doctor’s and patient’s styles (Pearce, Trumble, Arnold, Dwan, & Phillips, 2008). Perhaps an optimal solution is the use of flat LCD monitors on mobile arms, or using mobile hand-held devices, which allow greater flexibility. A positive influence of the EMR is greater information exchange (Shachak & Reis, 2009). EMR use was positively related to biomedical discussion, including questions about therapeutic regimen, exchange of information about medications, and patient disclosure of health information to the physician. Physicians who used an EMR were able to accomplish information-related tasks such as checking and verifying patient history,

encouraging patients to ask questions and ensuring completeness at the end of a visit to a greater extent than physicians who used paper records (Arar, Wen, McGrath, Steinbach, & Pugh, 2005; Hsu et al., 2005; Kuo, Mullen, McQueen, Swank, & Rogers, 2007; Makoul, Curry, & Tang, 2001; Margalit, Roter, Dunevant, Larson, & Reis, 2006). On the negative side, it is very hard for physicians to divide attention between the patient and the EMR. It has been demonstrated, that even within the first minute of the consultation, much of the interaction is driven by the computer, not by the patient’s agenda (Pearce et al., 2008). Physicians often walked straight to the monitor after only a short greeting. The average physician screen gaze was 24% to 55% of the visit time and it was inversely related to their engagement in psychosocial question asking and emotional responsiveness (Margalit et al., 2006; Shachak et al., 2009). The computer often caused physicians to lose rapport with patients; e.g. physicians typed in data or screen gazed while talking to patients or while the patient was talking. Our literature review also suggests that the computer’s potential to assist in patient education is underutilized (Shachak & Reis, 2009). How can we make the ideal future scenario come true, and overcome the negative influences of the computer on patient centeredness? Computer skills are important enablers. Blind typing, navigation skills, the use of templates and keyboard shortcuts, and the ability to organize and search for information were associated with physicians’ ability to effectively communicate with patients in computerized settings (Frankel et al., 2005; Shachak et al., 2009; Shachak & Reis, 2009). Similarly, physicians’ styles as well as their basic communication skills are highly important. Both Ventres et al. and Booth et al. identified relatively similar styles and suggested the computer’s impact on patient centeredness depended on them (Booth, Robinson, & Kohannejad, 2004; Ventres, Kooienga, Marlin, Vuckovic, & Stewart, 2005; Ventres et al., 2006). We have recently provided

165

The Computer-Assisted Patient Consultation

a unified classification of these styles (Shachak & Reis, 2009). The three styles in this unified classification are: 1. Informational-ignoring style. This style is characterized by extensive information gathering, that is often facilitated by the EMR, and focus on details of information. Physicians with this style often lost rapport with patients while engaged with the EMR, e.g. they frequently talked while gazing at the monitor, hardly faced the patients, and often left them idle while engaging with the computer. 2. Controlling-managerial style, which is characterized by separating computer use from communication. Physicians with this style alternated their attention between the patient and the computer in defined intervals or stages of the encounter. While with the patient, they turned away from the computer and vice versa. Switches of attention were often indicated by these physicians using non-verbal cues such as turning body or gaze. 3. Interpersonal style. This style is characterized by its focus on the patient. Physicians of this style oriented themselves to the patient even when using the EMR, did not usually talk while using the computer, and utilized the computer to share and review information together with the patient. They spent less time on data entry and usually refrained from using the computer in the beginning of the encounter. Along the same lines, Frankel et al. (2005) have suggested that the computer facilitates both negative and positive baseline communication skills. Some of the physicians we observed were able to compensate for their lack of computer mastery by excellent communication skills (Shachak et al., 2009). Additionally, strategies and best practices may be employed. The major strategy we have

166

identified is dividing the encounter into patientand computer-focused stages that are clearly distinctive and indicated by body language and focus of gaze. Another is keeping the patients engaged by sharing the screen with them or reading out loud while typing (Shachak et al., 2009).

FUTURE RESEARCH DIRECTIONS As pointed out, education is important for enhancing safer use of computers during consultation. Education at all levels, from medical school to residency to continuing medical education, is also essential for acquiring and continuously improving computer and communication skills and learning effective strategies for integrating the computer into the patient-doctor relationships. Dr. Smith in our scenario was trained in using the EHR at the national simulation center. The training he received went well beyond the technical aspects of using the software, which are usually the scope of clinical information systems implementations. Rather, he learned how to avoid EHR-induced errors, accommodate for his own and various patient styles, employ strategies and best practices for patient-doctor-computer communication, and make optimal use of the computer for patient education and shared decision making. The development and evaluation of such educational interventions is one area for future research. Ten tips for physicians on how to incorporate the EMR into consultation had been suggested by Ventres, Kooienga, and Marlin (2006), and have been recently modified by us (Shachak & Reis, 2009). These tips include suggestions regarding the computer skills required of the physician, encounter management practices, and ways to engage the patient. However, as far as we can detect from the literature, educational modules of patient-doctor communication have not yet included the computer. As of February, 2008, we were able to find only two examples of instructional modules supporting the introduction

The Computer-Assisted Patient Consultation

of computers to clinical care settings (Institute for Healthcare Communication, 2008; Simpson, Robinson, Fletcher, & Wilson, 2005). We are currently in the process of developing and evaluating a simulation-based training intervention aimed at qualifying Family Medicine residents in incorporating the use of EMRs into consultations. A number of scenarios have been developed for training participants to identify the potential pitfalls, learn various strategies and skills for using an EMR during consultations, and employ them flexibly depending on the situation and the patient and physician styles. Technology also opens new ways to incorporate the computer into a seamless clinical encounter. Present systems rely heavily on text, but we can envision a multimedia EHR which combines image, text, sound, and video with technologies such as voice and handwriting recognition, automatic transcription, natural language processing, and automated tagging. These technologies can minimize the interference of data entry with communication, maintain a significant narrative component of the record in rich media that may reduce ambiguity (Daft, Lengel, & Trevino, 1987) and, therefore, support reasoning and decision making, while still enabling fast retrieval and data structuring for management and analysis. Multiple challenges are involved in the development of such multimedia EHRs including spoken and medical natural language processing, assigning metadata to videotaped medical interviews, language standardization, and semantic information retrieval. Furthermore, even if these technological challenges are overcome, numerous factors such as legal issues, cost, usability, time required to review information, and potential users’ perceptions of the system—especially concerns about privacy and confidentiality—may still hinder adoption of multimedia-EHRs. All of these challenges open multiple directions for future research. In our scenario Dr. Smith is conscious of the service the data he generates render for the

benefit of the larger population, and welcomes feedback on his data quality. However, we have not discussed the relationships between micro and macro levels, and particularly how the way that physicians document influences data quality and the use of data for clinical management, administration, and research. More research is required to better understand the micro-macro relationships. Finally, Web 2.0 applications such as RSS feeds, social networking, blogs, wikis, and Twitter open new ways of communication and knowledge exchange among patients and doctors. These applications can add a component of virtual consultation into the clinical practice. They also facilitate many-to-many instead of, or in addition to, the traditional one-to-one communication (Hawn, 2009; Jain, 2009; Kamel Boulos & Wheeler, 2007). In this chapter we have focused on the impact of integrating ICT into the traditional patient-doctor consultation. The effect of Web 2.0 applications on medical practice and consultations in particular, remains to be seen. As far as we can detect, there is a dearth in rigorous research with this focus. Most of the literature to date consists of opinions and personal reports from the field.

LIMITATIONS As the scope of this chapter is the medical consultation, the majority of research discussed emerged from ambulatory or outpatient settings. Acute care presents different challenges and employs much different workflow processes, though some of the challenges discussed here also apply (e.g. new types of errors—Campbell et al., 2009).

CONCLUSION The micro level perspective provided in this chapter reveals some of the processes that underlie the impacts of ICT on the health care system. Throughout this chapter we have demonstrated

167

The Computer-Assisted Patient Consultation

the thin line between beneficial and adverse consequences of using ICT during consultation. Human-centered design, medical education, as well as developing the technologies toward the multimedia-EHR have the potential to improve the individual-level impact of the computerized consultation.

REFERENCES Arar, N. H., Wen, L., McGrath, J., Steinbach, R., & Pugh, J. A. (2005). Communicating about medications during primary care outpatient visits: the role of electronic medical records. Informatics in Primary Care, 13(1), 13–22. Bates, D. W., Teich, J. M., Lee, J., Seger, D., & Kuperman, G. J., Ma’Luf, N. et al. (1999). The impact of computerized physician order entry on medication error prevention. Journal of the American Medical Informatics Association, 6(4), 313–321. Booth, N., Robinson, P., & Kohannejad, J. (2004). Identification of high-quality consultation practice in primary care: the effects of computer use on doctor-patient rapport. Informatics in Primary Care, 12(2), 75–83. Campbell, E. M., Guappone, K. P., Sittig, D. F., Dykstra, R. H., & Ash, J. S. (2009). Computerized provider order entry adoption: implications for clinical workflow. Journal of General Internal Medicine, 24(1), 21–26. doi:10.1007/s11606008-0857-9 Campbell, E. M., Sittig, D. F., Ash, J. S., Guappone, K. P., & Dykstra, R. H. (2006). Types of unintended consequences related to computerized provider order entry. Journal of the American Medical Informatics Association, 13(5), 547–556. doi:10.1197/jamia.M2042

168

Chaudhry, B., Wang, J., Wu, S., Maglione, M., Mojica, W., & Roth, E. (2006). Systematic review: impact of health information technology on quality, efficiency, and costs of medical care. Annals of Internal Medicine, 144(10), 742–752. Chused, A., Kuperman, G. J., & Stetson, P. D. (2008). Alert override reasons: a failure to communicate. AMIA Annual Symposium Proceedings, 111-115. Daft, R. L., Lengel, R. H., & Trevino, L. K. (1987). Message Equivocality, Media Selection, and Manager Performance - Implications for Information-Systems. Management Information Systems Quarterly, 11(3), 355–366. doi:10.2307/248682 DeLone, W. H., & McLean, E. R. (1992). Information systems success: the quest for the dependent variable. Information Systems Research, 3(1), 60–95. doi:10.1287/isre.3.1.60 DeLone, W. H., & McLean, E. R. (2003). The DeLone and McLean model of information systems success: a ten-year update. Journal of Management Information Systems, 19(4), 9–30. Eslami, S., Keizer, N. F., & Abu-Hanna, A. (2007). The impact of computerized physician medication order entry in hospitalized patients-A systematic review. International Journal of Medical Informatics. doi:.doi:10.1016/j.ijmedinf.2007.10.001 Frankel, R., Altschuler, A., George, S., Kinsman, J., Jimison, H., Robertson, N. R., & Hsu, J. (2005). Effects of exam-room computing on clinician-patient communication: a longitudinal qualitative study. Journal of General Internal Medicine, 20(8), 677–682. doi:10.1111/j.15251497.2005.0163.x Grizzle, A. J., Mahmood, M. H., Ko, Y., Murphy, J. E., Armstrong, E. P., & Skrepnek, G. H. (2007). Reasons provided by prescribers when overriding drug-drug interaction alerts. The American Journal of Managed Care, 13(10), 573–578.

The Computer-Assisted Patient Consultation

Hawn, C. (2009). Take two aspirin and tweet me in the morning: How Twitter, Facebook and other social media are reshaping medicine. Health Affairs, 28(2), 361–368. doi:10.1377/hlthaff.28.2.361 Hsieh, T. C., Kuperman, G. J., Jaggi, T., Hojnowski-Diaz, P., Fiskio, J., & Williams, D. H. (2004). Characteristics and consequences of drug allergy alert overrides in a computerized physician order entry system. Journal of the American Medical Informatics Association, 11(6), 482–491. doi:10.1197/jamia.M1556 Hsu, J., Huang, J., Fung, V., Robertson, N., Jimison, H., & Frankel, R. (2005). Health information technology and physician-patient interactions: impact of computers on communication during outpatient primary care visits. Journal of the American Medical Informatics Association, 12(4), 474–480. doi:10.1197/jamia.M1741 Institute for Healthcare Communication. (2008). Retrieved August 11, 2009, from http://www. healthcarecomm.org/index.php?sec=courses&s ub=faculty&course=3 Jain, S. H. (2009). Practicing medicine in the age of Facebook. The New England Journal of Medicine, 361(7), 649–651. doi:10.1056/NEJMp0901277 Kamel Boulos, M. N., & Wheeler, S. (2007). The emerging Web 2.0 social software: an enabling suite of sociable technologies in health and health care education. Health Information and Libraries Journal, 24(1), 2–23. doi:10.1111/j.14711842.2007.00701.x Kuo, G. M., Mullen, P. D., McQueen, A., Swank, P. R., & Rogers, J. C. (2007). Cross-sectional comparison of electronic and paper medical records on medication counseling in primary care clinics: a Southern Primary-Care Urban Research Network (SPUR-Net) study. Journal of the American Board of Family Medicine, 20(2), 164–173. doi:10.3122/ jabfm.2007.02.060113

Lapane, K. L., Waring, M. E., Schneider, K. L., Dube, C., & Quilliam, B. J. (2008). A mixed method study of the merits of e-prescribing drug alerts in primary care. Journal of General Internal Medicine, 23(4), 442–446. doi:10.1007/s11606008-0505-4 Makoul, G., Curry, R. H., & Tang, P. C. (2001). The use of electronic medical records: communication patterns in outpatient encounters. Journal of the American Medical Informatics Association, 8(6), 610–615. Margalit, R. S., Roter, D., Dunevant, M. A., Larson, S., & Reis, S. (2006). Electronic medical record use and physician-patient communication: An observational study of Israeli primary care encounters. Patient Education and Counseling, 61(1), 134–141. doi:10.1016/j.pec.2005.03.004 McCoy, S., Galleta, D., Everard, A., & Polak, P. (2004, December 10-11). A study of the effects of online advertising: A focus on pop-up and in-line ads. Paper presented at the Third Annual Workshop on HCI Research in MIS, Washington, D.C. Musen, M. A., Shahar, Y., & Shortliffe, E. H. (2006). Clinical decision support systems. In Shortliffe, E. H., & Cimino, J. J. (Eds.), Biomedical informatics: computer applications in health care and biomedicine (3rd ed.). New York: Springer. National Alliance for Health Information Technology (NAHIT). (2008). Report to the Office of the National Coordinator for Health Information Technology on defining key health information technology terms. Retrieved October 15, 2009 from http://healthit.hhs.gov/portal/server.pt/gateway/PTARGS_0_10741_848133_0_0_18/10_2_ hit_terms.pdf O’Donnell, H. C., Kaushal, R., Barron, Y., Callahan, M. A., Adelman, R. D., & Siegler, E. L. (2009). Physicians’ attitudes towards copy and pasting in electronic note writing. Journal of General Internal Medicine, 24(1), 63–68. doi:10.1007/ s11606-008-0843-2

169

The Computer-Assisted Patient Consultation

Patel, V. L., Kushniruk, A. W., Yang, S., & Yale, J. F. (2000). Impact of a computer-based patient record system on data collection, knowledge organization, and reasoning. Journal of the American Medical Informatics Association, 7(6), 569–585. Paterno, M. D., Maviglia, S. M., Gorman, P. N., Seger, D. L., Yoshida, E., & Seger, A. C. (2009). Tiering drug-drug interaction alerts by severity increases compliance rates. Journal of the American Medical Informatics Association, 16(1), 40–46. doi:10.1197/jamia.M2808 Pearce, C., Dwan, K., Arnold, M., Phillips, C., & Trumble, S. (2009). Doctor, patient and computer--a framework for the new consultation. International Journal of Medical Informatics, 78(1), 32–38. doi:10.1016/j.ijmedinf.2008.07.002

Simpson, L., Robinson, P., Fletcher, M., & Wilson, R. (2005). e-Communication skills. Oxford, UK: Radcliff. Tamblyn, R., Huang, A., Perreault, R., Jacques, A., Roy, D., & Hanley, J. (2003). The medical office of the 21st century (MOXXI): effectiveness of computerized decision-making support in reducing inappropriate prescribing in primary care. Canadian Medical Association Journal, 169(6), 549–556. Thielke, S., Hammond, K., & Helbig, S. (2007). Copying and pasting of examinations within the electronic medical record. International Journal of Medical Informatics, 76(Suppl 1), S122–S128. doi:10.1016/j.ijmedinf.2006.06.004

Pearce, C., Trumble, S., Arnold, M., Dwan, K., & Phillips, C. (2008). Computers in the new consultation: within the first minute. Family Practice, 25(3), 202–208. doi:10.1093/fampra/cmn018

van der Sijs, H., Aarts, J., Vulto, A., & Berg, M. (2006). Overriding of drug safety alerts in computerized physician order entry. Journal of the American Medical Informatics Association, 13(2), 138–147. doi:10.1197/jamia.M1809

Pearce, C., Walker, H., & O’Shea, C. (2008). A visual study of computers on doctors’ desks. Informatics in Primary Care, 16(2), 111–117.

Ventres, W., Kooienga, S., & Marlin, R. (2006). EHRs in the exam room: tips on patient-centered care. Family Practice Management, 13(3), 45–47.

Shachak, A., Hadas-Dayagi, M., Ziv, A., & Reis, S. (2009). Primary care physicians’ use of an electronic medical record system: A cognitive task analysis. Journal of General Internal Medicine, 24(3), 341–348. doi:10.1007/s11606-008-0892-6

Ventres, W., Kooienga, S., Marlin, R., Vuckovic, N., & Stewart, V. (2005). Clinician style and examination room computers: a video ethnography. Family Medicine, 37(4), 276–281.

Shachak, A., & Reis, S. (2009). The impact of electronic medical records on patient-doctor communication during consultation: a narrative literature review. Journal of Evaluation in Clinical Practice, 15, 641–649. doi:10.1111/j.13652753.2008.01065.x Shah, N. R., Seger, A. C., Seger, D. L., Fiskio, J. M., Kuperman, G. J., Blumenfeld, B., et al. (2005). Improving override rates for computerized prescribing alerts in ambulatory care. AMIA Annual Symposium Proceedings, (pp. 1110).

170

Ventres, W., Kooienga, S., Vuckovic, N., Marlin, R., Nygren, P., & Stewart, V. (2006). Physicians, Patients, and the Electronic Health Record: An Ethnographic Analysis. Annals of Family Medicine, 4(2), 124–131. doi:10.1370/afm.425 Weiner, J. P., Kfuri, T., Chan, K., & Fowles, J. B. (2007). “e-Iatrogenesis”: the most critical unintended consequence of CPOE and other HIT. Journal of the American Medical Informatics Association, 14(3), 387–388, discussion 389. doi:10.1197/jamia.M2338

The Computer-Assisted Patient Consultation

Weingart, S. N., Toth, M., Sands, D. Z., Aronson, M. D., Davis, R. B., & Phillips, R. S. (2003). Physicians’ decisions to override computerized drug alerts in primary care. Archives of Internal Medicine, 163(21), 2625–2631. doi:10.1001/ archinte.163.21.2625

Wheatley, T., & Wegner, D. M. (2001). Automaticity of action, psychology of. In Smelser, N. J., & Baltes, P. B. (Eds.), International encyclopedia of the social & behavioral sciences (pp. 991–993). Oxford, UK: Pergamon.

This work was previously published in Healthcare and the Effect of Technology: Developments, Challenges and Advancements, edited by Stéfane M. Kabene, pp. 72-83, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

171

172

Chapter 1.12

Quality and Reliability Aspects in Evidence Based E-Medicine Asen Atanasov Medical University Hospital “St. George”, Bulgaria

ABSTRACT This chapter is a brief survey on some e-medicine resources and international definitions focused on the three main subjects of the healthcare quality – the patient, the costs and the evidence for quality. The patients can find in e-medicine everything that they need, but often without data on the supporting evidence. The medical professionals can learn where to find e-information on cost, quality and patient safety, and, more importantly, how to distinguish claims from evidence by applying the principles of evidence based medicine. The goal is to spread and popularize the knowledge in this field with an emphasis on how one can DOI: 10.4018/978-1-60960-561-2.ch112

find, assess and utilize the best present evidence for more effective healthcare. The sites discussed below could assist in the retrieval of information about methods for obtaining evidence along with the ways of measuring evidence strength and limitations. These sites also provide information on implementing the ultimate evidence-based product – clinical guidelines for better medical practice and health service.

INTRODUCTION The international consensus document (CLSI, HS1-A2, 2004) emphasized that the top in the hierarchy of healthcare quality is the undivided of complete customer satisfaction at minimal cost

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Quality and Reliability Aspects in Evidence Based E-Medicine

and highest quality. The customer, who might be unsatisfied by healthcare quality or other reasons, may prefer to search the Internet for some alternative sources of health information. Looking for a solution to her/his own problem amongst hundreds of thousands of Webb sites, she/he might find exactly what is needed in “minimal cost” but sometimes with unknown quality. The health professionals can also find scientific basis for continuous education along with answers of almost all practical problems of their patients. However, information about the evidence strength and quality is not always available. This chapter shortly considers the three main subjects of total quality in healthcare - customer, cost, and evidence for quality, in the context of e-medicine. The goal is more people to obtain knowledge in this field and use properly the best present evidence. The patient can find in emedicine everything that she/he needs, but often without data on the supporting evidence. The medical professionals can be assisted to decide what kind of e-resources they are interested in and how they can learn more about evidence in medicine. The information concerning the achievement of basic medical science, specific regulations, laws, accreditation and healthcare audits remains beyond the scope of the chapter.

BACKGROUND A lot of e-medical information can be easily obtained through Internet and Wikipedia. Its usefulness, however, depends on the users’ willingness, behavior, knowledge and skills to distinguish a claim from actual evidence, and to users’ ability to use properly e-medical information. Although for some regions the access to e-resources might be a problem, Bratislava Declaration clearly outlined as a priority the validity and quality of electronic health information, education and training (UEMS, 2007).

THE CUSTOMER Most of us have been, are, or will be patients. The patient, a suffering human being, becomes “customer”, one of the numerous external and internal customers of the healthcare system (other patients, suppliers, institutions, factories, agencies, hospitals, doctors, nurses, pharmacists, technicians, and all other staff engaged in healthcare). As a customer, the patient is told that his/her welfare is paramount for the healthcare system. Thus, the patient-customer expects the best quality of help or, in other words, service. Being a customer and receiving service, the patient obtains the opportunity to actively assist the medical staff regarding his/her personal health. The result, however, is that the customer evaluates the health service rather than his/her health behavior. However, customer’s “satisfaction” with the quality of health service might be far away from the “evidence for quality”. Satisfaction is very subjective and cannot be objectively measured hence it is not the best end point for healthcare evaluation. Most patients today are well informed, but some prefer illusions in place of reality. People are not always able to make a clear distinction between personal satisfaction and healthy life style. The best health strategy for the society is not be the best approach for a single person. Any healthcare system needs money and can be easily destroyed by growing expectations of uncertain nature in an environment of limited and often badly managed resources. An organized group of active, even aggressive, patients might politically impose disproportional distribution of funds that otherwise might be spent more effectively for the advantage of more patients. Attractive new technologies, diagnostic instruments, tools and devices, new curative approaches and therapeutic drugs are often subject of commercial rather than medical interest. In such a complicated situation, the healthcare customers are looking at the Internet for health information because they want to obtain dependable

173

Quality and Reliability Aspects in Evidence Based E-Medicine

service, communicate online with their physicians, talk for their health problems, receive interpretation for their laboratory or instrumental records and medications, or arrange an appointment. They are ready to browse pay-for-performance programs in which recognized physicians are committed to providing quality care and service. Such e-Medicine and e-Prescribing sources need support for delivering high quality and efficient healthcare focused on the physician as the core of the network. Improvement of the customer’s experience with a physician’s knowledge contains the providers’ medical costs. The providers of medical advice directly use the technology and should be accountable for patient care. Moreover,, the providers should guarantee service and system quality together with the quality of information for the ultimate consumer of e-healthcare – the patient (LeRouge, & al, 2004). Another cash service is online clinical conferencing. It offers consultations on many clinical cases. The patient tells his/her medical history and answers clinical questionnaires, including data of physical examination, laboratory findings, medication, and other relevant topics. The patient receives third party comments, and expert opinion about diagnosis, medication, or other questions in sites like [http://www.thirdspace.org], [Clinical Conferences Online.mht], and [e-medicine.com] which are continuously updated. Web-based health services, such as [WebMD], also provide health information, including symptom checklists, pharmacy information, blogs of physicians with specific topics, and a place to store personal medical information. [WebMD], the Magazine is a patient-directed publication distributed bimonthly to physician waiting rooms in the USA. Similar health-related sites include [MedicineNet] - an online media publishing company; [Medscape] - a professional portal with 30 medical specialty areas which offers up-to-date information for physicians and other healthcare professionals; [RxList] – providing detailed pharmaceutical information on generic and name-brand drugs.

174

The web site formerly known as [eMedicine.com], created for physicians and healthcare professionals, is now [eMedicineHealth] - a consumer site offering similar information to that of [WebMD]. Most are last modified February 2009. Another useful site is MedHelp [http://www. medhelp.org/]. It delivers the opportunity for online discussion on different healthcare topics by a partnership with medical professionals from hospitals and medical research institutions. The site offers several forums of contacts - “Ask an Expert” for users’ questions, “Health Communities” for questions, comments, responses and support from other users, and “International Forums” on topics including addiction, allergy, cancer, cosmetic surgery, dental, and various other topics. The site Med-e-Tel [http://www.medetel.lu/ index.php] comprises e-health, Telemedicine and Health ICT as tools for service to medical and nurse practitioners, patients, healthcare institutions and governments. The site offers broad information on markets, various research and experience, and appears to be a gathering place, for education, networking and business aimed at a worldwide audience with diverse professional backgrounds. Another site destined for professionals is eMedicine - a separate medical website that provides regular clinical challenges comprising of a short case history accompanied by a visual cue in the form of a radiograph, ECG or photo, along with the clinical resolution for each case. The customer can click on one of the cases listed in the index, after which will leave the Global Family Doctor website and go to the eMedicine website [http://www.globalfamilydoctor.com/education/ emedicine/eMedicine.asp].↜eMedicine offers also a series of clinical problems in an interactive format with CME credit available after completion. E-medicine related sites are continuously updated and offer enormous information useful for everybody. For example, if one is concerned by healthy life styles, he or she may find information about which are the benefits of physical exercise, how to avoid undesirable events, how to incorpo-

Quality and Reliability Aspects in Evidence Based E-Medicine

rate exercise in customer lifestyle, how to succeed with weight control, what to do for a healthy pregnancy and post-natal care. Another example is the Journal Bandolier Extra, [www.ebandolier. com]. The top ten eMedicine Case Studies can be seen in recently updated [http://www.who.is/ whois-net/ip-address/emedcine.net/]. One analysis of American Sociological Association [http://www.asanet.org] showed that the users should be capable of evaluating the quality of e-health information since the information might be inaccurate or out of date. The users also have to distinguish between commercial advertising and appropriate health information, and recognize potential conflicts of interest. The abundance of health information permits customers to assume more responsibility for their own health care, and, therefore, patients should be included in developing standardized quality assurance systems for online health information [http://www.emedicine. com/med/]. An important support of the patient needs is so cold Bratislava Declaration on e-Medicine. It was published by the Council of European Union of Medical Specialists (UEMS) on October 13th 2007. [www.uems.net]. The Declaration establishes that there is existing potential for improvement of e-medicine in the context of quality and the manner in which the patient’ care is provided. The priority of quality over the cost-efficiencies is advocated, and the electronic development, recording, transfer, and storage of medical data are accepted as useful and inevitable. The Declaration also recognizes the need of support for (1) further progress in the accessibility of medical information; (2) developing higher standards in medical qualification and specialisation; (3) promotion of improvements in the well-being and healthcare of persons. The Council of EUMS also emphasizes that the misuse of e-Medicine could damage persons, communities and countries, and can become a risk to the security of data, patient confidentiality, the medical ethics, and the law. “The principles of a patient’s privacy and confi-

dentiality must be respected, and only patients have the right voluntarily to decide to have their data held in storage” – the Declaration says. In the context of the electronic recording, transfer and storage of medical information, the Council will make an effort, through registration and validation procedures, to implement, promote, develop and control future development of the fields •

• • •

respect for the security and privacy of persons, and the rights and laws governing these; respect for medical ethical principles; the validity of electronic health information; the quality of electronic medical education and training (UEMS, 2007).

Patients’ values and preferences are important and might differ markedly from those of physicians.

COSTS It is always useful to keep cost in mind, especially in times of financial crisis. An analysis of the first generation e-health companies showed that some of them lost billions in value during the early 2000 and failed to build a profitable business. Four most important factors in predicting the success or failure of an Internet healthcare company are identified - compelling value, unambiguous revenue model, competitive barriers, and organizational structure for cost control. Companies that make certain that they meet all or most of these factors, will have a better chance for success with a unique and valuable product and disciplined spending. Three factors introduce more challenges (Itagaki, & al, 2002). Some companies have a more impressive array of e-health properties that allow them to offer a bulk of services, earn enough revenue, and continue operating successfully. However, today’s economic situation introduces a great

175

Quality and Reliability Aspects in Evidence Based E-Medicine

deal of uncertainty into their future. The success depends on the stability of users’ interest and how long their cash reserves will last. Few industries can sustain the rapid economic downturn. However, despite the people less income, some medical supply industries continue to grow. An example is Total e-Medical - a physician-supervised supplier of diabetic testing materials, durable medical equipment, respiratory care, and arthritis and pain management products. The company realizes unprecedented growth of revenue, hired new employees in the past months, and intends to continue its workforce expansion [www.totalemedical.com], last updated 13.02.2009. On February 16, 2009 it was announced that as much as $21 billion in the economic stimulus package is designated for IT for medical records in the USA. The funds will help health care providers to implement such IT systems, which are expensive and difficult to deploy. About $3 billion will be directed to help health care providers buy IT systems while an additional $18 billion would be covering for additional Medicare and Medicaid payments to health care providers who use technology to improve patient care. Hospitals could receive up to $1.5 million while physicians would qualify for about $40,000 over several years. The hope is that the potential for e-health to improve medical care remains excellent. This is important because the improvement can produce savings for healthcare and profits for enterprising start-ups. The prospect that any person might have his/her personal e-medical record would be a realistic goal (Kolbasuk, & McGree, 2009). The e-medicine needs governmental support to improve the delivery of healthcare services to the elderly and poor customers. It is estimated that e-medicine or telemedicine can reduce about 60% of the on-site care cost by using e-mail, faxes and telephone consultations to link patients with health-care providers. It also improves access to health-delivery services in rural areas with shortages of doctors and hospitals. Instead, some

176

insurance agencies reimburse only “face-to-face consultations” – an expensive type of communication between patient and the doctor. Telemedicine was proven especially effective in Norway where even the installation of costly equipment turned out to be cheaper than flying doctors into remote regions [http://www.emedicine.com/med/]. The described events concern costs of e-health and e-medical information. However, the decisions about costs of the real health care are much more complicated and concern large parts of the society. Health professionals, especially governmental experts, should be familiar with economical analyses which allow objective choice between different health interventions. Unfortunately, the different economic analyses are often not well understood and they rarely assist in a political discussion or in the development of national health strategies. The cost-effectiveness analysis compares different health interventions by their costs and obtained health effect. It helps to assess which health intervention is worth to be implemented from the economic point of view. The costs are expressed in monetary units, and the health effect in life-years gained. The cost-effectiveness analysis, therefore, followed by sensitivity analysis (which tests the impact of best case and worse case scenarios) becomes a part of the decisionmaking process. The cost-utility analysis (utility is a cardinal value that represents the strength of an individual’s preferences for specific outcomes under condition of uncertainty) is applied when the healthcare area that is likely to provide the greatest benefit has to be identified. The cost-profit or cost-benefit analysis is useful when information is needed to assess which intervention will result in resource savings (that both sides of the equation are in same monetary units). A clear and very useful explanation of these analyses and their terminology can be found in the series named “What is?” sponsored by an educational grant from Aventis Pharma [www. evidence-based-medicine.co.uk], and, in a more

Quality and Reliability Aspects in Evidence Based E-Medicine

sophisticated version, in [http://www.cche.net/ usersguides/main.asp]. The economical analyses, along with survival analysis (Cox model) and many other methods are powerful tools that produce evidence. Although the educational material in the shown sites is only instructive, the references exist that explain the methods, calculations, and limitations, of each approach. Costs in medicine remain an important issue and now are incorporated in any decision analysis and clinical guideline.

QUALITY The following short list of common definitions is useful for better understanding of quality in health care and medicine: •

•

•

•

•

Quality – “Degree to which a set of inherent characteristics fulfills requirements” (ISO 9000, 2000, 3.1.1) Quality assurance – “Part of quality management focused on providing confidence that quality requirements will be fulfilled” (ISO 9000, 2000, 3.2.11). “Quality assurance indicates the quality of process performance” (HS1-A2, 2004). Quality control – “Part of quality management focused on fulfilling quality requirements” (ISO 9000, 2000, 3.2.10). “Quality control indicates the quality of procedural performance” (CLSI, HS1-A2, 2004). Quality improvement – “Part of quality management focused on increasing the ability to fulfill quality requirements (ISO 9000, 2000, 3.2.12). “Continuous quality improvement indicates how to achieve sustained improvement” (CLSI, HS1-A2, 2004). Quality management – “Coordinated activities to direct and control an organization with regard to quality (ISO 9000, 2000, 3.2.8).

•

•

•

Quality management system – “Management system to direct and control an organization with regard to quality” (ISO 9000, 2000, 3.2.3). “Quality management system will support all operational paths of workflow” (CLSI, HS1-A2, 2004). Quality policy – “Overall intentions and direction of an organization related to quality as formally expressed by top management” (ISO 9000, 2000, 3.2.4.) Reliability - the consistency with which the same information is obtained by a test or set of tests in the absence of intervening variables (Tudiver, & al, 2008).

Good quality of healthcare service would not be obtained without serious administrative support. A lot of documents of the Clinical and Laboratory Standards Institute (CLSI, former National Committee for Clinical Laboratory Standards, NCCLS), USA, are developed through a considerable consensus process. After worldwide expert discussion, each document is disseminated for global application [www.nccls.org]. Several CLSI documents describe, in detail, how a health organization that wishes to implement a quality management system can succeed. “These specialized documents are designed for any healthcare service manager who wishes to improve the processes involved in creating customer satisfaction by implementing proven standardized quality system concept” /GP22-A2, 2004, p. 7/. Continuous quality improvement (CLSI, GP22-A2, 2004) is a model focused specifically on the implementation of clinical service quality system management. It represents any healthcare service as working within two forms of infrastructural quality system matrices. The external matrix is created by external guidance from local and governmental administrative-level policies, processes and procedures. The internal matrix is built up from intrinsic quality system managerial-level policies, processes, and procedures. The internal

177

Quality and Reliability Aspects in Evidence Based E-Medicine

matrix contains the quality system and underlines service operations rather than healthcare operation. Another document (CLSI, HS1-A2, 2004) represents organizational hierarchy of quality administration in healthcare establishments. Quality control is the starting level. It is considered as an operational process control technique for measuring the effectiveness of the procedural performance that is expected to fulfill requirements for quality and governmental compliance. It needs quality indicators like precision, accuracy and other statistical variables. Quality assurance is the next level that measures the effectiveness of process performance and provides confidence that the quality requirements are fulfilled. Implementation of a quality management system is the crucial level - the practice facilitating systematic process-oriented improvement. All these levels are included in quality cost management that adds the economic activities – “cost of quality”. The highest level is the total quality management. It identifies the cost of quality for obtaining total quality management and insures sustainable high quality “by focusing on long-term success through customer satisfaction” (CLSI, HS1-A2, 2004, p. vii). In the above motion directed “upstairs” five key activities are implemented in a continuous spiral - quality planning, quality teamwork, quality monitoring, quality improvement and quality review (Plan-Do-Check-Act). The results of the last quality review represent the basis of the next cycle towards the next quality level. All necessary activities have a detailed description in the original document (CLSI, HS1-A2, 2004). The administration of a healthcare establishment, if decided, should surmount step-by-step the hierarchical ladder, from the lowest level of quality to the highest– total quality management – “An organization-wide approach centered on ongoing quality improvement as evidenced by total customer satisfaction and minimal cost of maximized quality” (CLSI, GP22-A2, 2004, p. xii).

178

Recently, the MARQuIS project (2009) outlined the EU perspectives in five papers. They commented quality improvement strategies for European cross-border healthcare, national quality improvement policies and strategies, applications of quality improvement strategies in European hospitals, the results of the MARQuIS project, and the future direction of quality and safety in hospital care in EU. However, the lack of staff motivation is able to transform all administrative efforts for high quality in empty formalism and bureaucratic curtain. If medical professionals understand, joint and support the administrative efforts for implementation of quality management system, the team surely will succeed. In fact it is too early to discuss the effectiveness of this systematic approach towards continuous improvement of the healthcare service. More experience should be obtained for a frank assessment of the described administrative approach for obtaining high quality in healthcare.

Quality Indicators Measurement of the quality is essential for its improvement. The tools intended to measure the quality of healthcare in fact measure the evidence for quality and confirm that a desirable level is obtained. •

•

Quality indicators are „observations, statistics, or data defined by the organization or service that typify the performance of a given work process and provide evidence that the organization or service is meeting its quality intentions” (AABB, 2003). The thresholds are “statistical measure of compliance of the specific indicator for acceptable outcome” (CLSI, HS1-A2, 2004).

Quality indicators are objective measures for analysis of quality achievement in health service. Their implementation is an attainable, realistic

Quality and Reliability Aspects in Evidence Based E-Medicine

goal for assessing compliance and quantifying improvement. Some quality indicators concern patient preferences and are obtained by patient interviews. These indicators are important for identification of gaps in quality at the population level and can be useful for improvement of health service. These are mainly indicators related to pain, symptom management and access to care. With palliative care and emergency room visits, they are identified as acceptable for assessment of quality. The structure indicators are, for instance, the patient-to-nurse ratio, the existing strategies to prevent medication errors, the measurements of patient satisfaction. The process indicators give a quick view of the quality of care. When indicators such like the length of stay in the hospital, duration of mechanical ventilation, the proportion of days with all beds occupied, the proportion of glucose measurement higher than 8.0 or lower than 2.2 mmol/L, are applied to an intensive care unit, they can characterize the quality of total activity of the unit (Steel, & al, 2004). The outcome indicators are the standardized mortality, the incidence of decubitus, the number of unplanned extubations and others. In each case, however, it is important to assess the validity of data used because the quality ranking of the hospitals depends not only on the indicators used, but also on the nature of the data used. By means of the same indicator, hospitals classified as highquality using routine administrative data have been reclassified as intermediate or low-quality hospitals using the enhanced administrative data (Glance, & al, 2008). The authors emphasize the need to improve the quality of the administrative data if these data are to serve as the information infrastructure for quality reporting. A recent study (Grunfeld, & al, 2008) assesses stakeholder acceptability of quality indicators of end-of-life care that potentially are measurable from population-based administrative health databases. A multidisciplinary panel of cancer care health professionals, patients with metastatic

breast cancer and caregivers for women died of metastatic breast cancer, is used to assess acceptability among the indicators for end-of-life care. The authors conclude that patient preferences, variation in local resources, and benchmarking should be considered when developing quality monitoring systems. Those quality indicators that stakeholders perceive as measuring quality care will be most useful. The acceptance of quality indicators in EU is a subject of another study (Legido-Quigley, & al, 2008). It is noticed that a few EU countries have adopted quality indicators, possibly because some contradictions in EU healthcare policy. Health care is a responsibility of the member states, but the free movement of healthcare customers is regulated by the European law. Some initiatives on quality are coming from national governments but others are from health professionals and providers; some solutions are offered by EU laws but others are entirely nationally based. The following convincing example is shown as evidence for the large variations in the perception of the meaningfulness of quality indicators: in Denmark, the quality of care provided by hospitals for six common diseases (lung cancer, schizophrenia, heart failure, hip fracture, stroke, and surgery for acute gastrointestinal bleeding) is measured for hospital assessment; in UK the performance of general practitioners is assessed with the quality and outcomes framework of about 140 measures developed through evidence and professional consensus. Another very important side of the problem deserves attention: “Quality indicator systems have been criticised for focusing on what is easily measured rather than what is important, and for being used in ways that encourage opportunistic behaviour, either by manipulating data or changing behaviour to achieve targets while compromising care”. (Legido-Quigley & al 2008). The best currently available evidence for quality of any medical care might be the evaluation of patient outcomes. The general criteria for conducting studies on patient outcomes are described

179

Quality and Reliability Aspects in Evidence Based E-Medicine

in a CLSI document (CLSI HS6-A, 2004) and deserve more attention. The document is focused on the planning, conducting and reporting patient outcomes research. It provides a brief review on research methodology of primary patient outcome studies such as observational studies (surveys or cross-sectional studies, case-control studies and cohort studies) and interventional studies (randomized controlled trials and nonrandomized studies). It outlines the role of systematic overviews, meta-analyses, decision analyses, cost-effectiveness analyses and simulations along with a limited reference to some published methodological sources. A good example of the rules implementation on an outcome study is the PATH project (Veillard, & al 2005). It is currently implemented as a pilot project in eight EU countries in order to refine its framework before its further expansion. The project describes the outcomes achieved, specifically the definition of the concept and the identification of key dimensions of hospital performance. It also designs the architecture directed to enhance evidence-based management and quality improvement through performance assessment, selection of the core and tailored set of performance indicators with detailed operational definitions. It identifies the trade-offs between indicators and elaborates on the descriptive sheets for each indicator in order to support the hospitals in interpreting their results, designing a balanced dashboard, and developing strategies for implementation of the PATH framework. The future implementation of this project could be of significant interest for grading of hospitals by quality and effectiveness.

Patient Safety The high quality and reliability in health care are inaccessible without suitable activity to insure patient safety. Recently the General Secretariat of the Council of the EU sent a recommendation on patient safety, prevention and control of

180

healthcare associated infections to the Working Party on Public Health [Interinstitutional file 2009/0003 (CNS) 6947/09, 27 February, 2009]. The document offers the important definitions concerning patient safety: • •

•

•

Adverse event – “incident which results in harm to a patient”; Harm – “impairment of structure or function of the body and/or any deleterious effect which arises from that”; Healthcare associated infections – “diseases or pathologies (illness, inflammation) related to the presence of an infectious agent or its products as a result of exposure to healthcare facilities or healthcare procedures”; Patient safety – “freedom for a patient from unnecessary harm or potential harm associated with healthcare”.

The draft document notices an estimation showing that between 8% and 12% of patients in EU Member States admitted to hospitals suffer from adverse events whilst receiving healthcare. That is why the document is focused on the expected action and national policies directed towards improving public health, preventing human illnesses and diseases, and obviating sources of danger to human health: “Community action in the field of public health shall fully respect the responsibilities of the Member States for the organisation and delivery of health services and medical care” the document says. The aim is “to achieve result-oriented behaviour and organizational change, by defining responsibilities at all levels, organizing support facilities and local technical resources and setting up evaluation procedures” [Interinstitutional file 2009/0003 (CNS) 6947/09, 27 February, 2009]. It is also emphasized that the realization of these recommendations needs dissemination of the document content to healthcare organisations, professional bodies, and educational institutions

Quality and Reliability Aspects in Evidence Based E-Medicine

along with improvement of the patient information. The document suggests that each institutional level should follow the recommended approaches. This would ensure that the key elements are put into everyday practice and the patient safety is receiving proper attention.

EVIDENCE During the last decades several powerful methods for the distinction of evidence from claims in medicine were effectively disseminated. Evidence can be measured and assessed. For example, the accuracy of the diagnostic tests can be measured by their diagnostic sensitivity (the probability of a positive result amongst target persons) and diagnostic specificity (the probability of a negative result amongst control persons); the post-test probability of disease can be calculated by Bayesian approach from pre-test probability and the new information from the diagnostic test; the ratio of the probability that a given test result is met in a target (ill) person to the probability that the same result is met in a control (healthy) person, is given by the likelihood ratio (LR); the various kinds of effects (risks) for treated and control populations (groups) could be evaluated by the odds ratio; the treatment effect – by the number needed to treat (NNT), that is the inverse of the absolute risk reduction and is translated as “how many patients should be treated to prevent an event”. There are many others. Most of these useful methods can be found in the book-series Clinical Evidence [www.clinicalevidence.com], Wikipedia and others Internet or published resources related to clinical epidemiology. There are other methods that have become more sophisticated. For example, summarizing the results of a number of randomized trials and other independent studies can be done by the statistical technique called meta-analysis. It is used mainly to assess the clinical effectiveness of healthcare interventions as well as to estimate precisely the

treatment effect. Meta-analysis is usually applied in systematic reviews, but as any method has some limitations. The outcome of a meta-analysis will be influenced by the inclusion or exclusion of certain trials, and by the degree of adherence to the rigorous standards for the eligibility criteria explicitness and appropriateness that should characterize a meta-analysis (Fried, & al, 2008). Another example for gathering evidence is the assessment of the future effect of a medical action by decision analysis. It includes building a decision tree, obtaining probabilities and utilities for each outcome, evaluating the outcomes, and performing a sensitivity analysis to compare alternative health strategies (Detsky, & al. 1997, [www.cche.net/usersguides/main.asp], [www. evidence-based-medicine.co.uk] and others). How to obtain evidence for healthcare costs through clinical economic analysis and measurement of the healthcare quality by quality indicators was already discussed. However, any calculations have limitations and need valid primary data.

EVIDENCE BASED MEDICINE Evidence-based medicine (EBM) is the conscientious, explicit and judicious use of the best current evidence in making decisions about the care of individual patients (Sackett, & al 2000). In this context ”best” means that the quality of the evidence can be measured, and “current” means that the best evidence today might not be the best tomorrow. EBM instructs clinicians how to access, evaluate, and interpret the medical publications. The clinician can find the relevant answer of the patient problem by performing critical appraisal of the information resources,. After the decision is taken and best health service applied to the patient, the clinical performance can be evaluated and continually monitored. The goal of this process is to improve patient care by more effective and efficient use of clinical information, including

181

Quality and Reliability Aspects in Evidence Based E-Medicine

diagnostic and prognostic markers, and to ensure effective and safe treatment in compliance with individual patient preferences. EBM also offers an approach to teaching the practice of medicine. Centre for Health Evidence maintains (on behalf of the Evidence-Based Medicine Working Group) the full text of the series “Evidence-Based Medicine: A New Approach to Teaching the Practice of Medicine” [http://www. cche.net/usersguides/main.asp]. The series was originally published in the Journal of the American Medical Association (JAMA) between 1992 and 2000 years and was last updated on August 15, 2007. The educational materials have been enhanced and re-introduced in a new interactive website [http://www.usersguides.org]. Access to the site is available by subscription through JAMA. In the „Users’ Guides to EBM”, the interested medical professional can find a step-by-step explanation how to use electronic information resources, how to select publications that are likely to provide valid results about therapy and prevention, diagnostic tests (including Bayesian approach and likelihood ratios), harm and prognosis, how to use a clinical decision analysis, clinical practice guidelines and recommendations, and many other basic principles of EBM. A study assessing the efficiency of EBM teaching shows the growing popularity of EBM education, practice teaching and evaluation of learning methods, but also indicates the lack of data on the application of the obtained EBM skills in the clinical practice. It is noticed also that users’ perception for the main barriers to EBM practice is the finding of contradictory results in the literature, the insufficient knowledge of English language, the limited access to PCs, and the lack of time and institutional support (Gardois, & al, 2004). The Objective Structured Clinical Examination (OSCE) measures the family medicine resident lifelong learning skills in evidence-based medicine as an important part of the evaluation of physician skills. The authors describe an innovative assessment tool for evaluating the skills of EBM. It was

182

found that the competencies are overlapped with well known major skills such as translation of uncertainty for an answerable question, systematic retrieval of best evidence available, critical appraisal of evidence for validity, clinical relevance and applicability, and application of results in practice. These skills are tested in a simulated OSCE. The results show that the development of standardized tools for assessing EBM skills is an essential part of the evaluation of physician skills. They also emphasize the importance of the physician’s understanding of the different levels of evidence (Tudiver, & al 2008) An interesting article describes the development and implementation of an evidence-based practice model named CETEP (Clinical Excellence Through Evidence-Based Practice). The model provides the framework for a process that can be easily adapted for use as a tool to document the critical appraisal process. It is focused on nurse’s activities that lead to evidence-based practice changes and can be adapted for use in any healthcare organization. CETER emphasizes the process evaluating the applicability of the evidence for the clinical practice. It is shown that using research to incorporate essential components into clinical practice needs serious consideration before changing practice at the bedside. The article encourages nurses to conduct research and use evidence that will have meaning for their practice. It is expected that addressing practice problems or evaluating changes in practice with this model will be empowering. Translation of research to practice is more complex than simply writing a policy or procedure based on a research study and expecting the staff to comply. Successful integration of research into practice requires critical appraisal of study methods and results, and most importantly, consideration of the applicability of the evidence to the particular clinical setting (Collins, & al, 2008). Table 1 gives a list of sites that are useful for any reader interested in the principles and application of EBM.

Quality and Reliability Aspects in Evidence Based E-Medicine

Table 1. Some websites, with information related to EBM http://www.ahrq.gov/clinic http://www.jr2.ox.ac.uk/bandolier/ http://cebm.jr2.ox.ac.uk/ http://www.clinicalevidence.com http://www.cochrane.org/ http://www.infopoems.com http://www.ICSI.org http://www.guidelines.gov

http://www.acponline.org/journals/acpjc/jcmenu.htm http://www.ceres.uwcm.ac.uk/frameset.cfm?section=trip http://www.ebponline.nehttp://agatha.york.ac.uk/darehp.htm http://www.york.ac.uk/inst/crd/ehcb.htm http://www.evidence-basedmedicine.com http://medicine.ucsf.edu/resources/guidelines http://www.ahrq.gov/clinic/uspstfix.htm http://www.york.ac.uk/inst/crd/

EBM practice starts by converting the need for clinical information into answerable questions. The answers are researched through e-information and library resources. If the Internet is accessible, a lot of library resources become available. A simple enumeration includes MEDLINE, Cochrane Databases of Systematic Reviews and Clinical Trials, DARE, ACP Journal Club, InfoRetriever, InfoPoems, UpToDate, DynaMed and others. The readings found are critically appraised for validity, clinical relevance, and applicability of the evidence. The application of the obtained results in practice and monitoring the effects are focused on the patient. When researching the evidence one should keep in mind the difference between POEM (Patient-Oriented Evidence that Matters) and DOE (Disease-Oriented Evidence). POEM deals with outcomes of importance to patients, such as changes in morbidity, mortality, or quality of life. DOE deals with surrogate end points, such as changes in laboratory values or other measures of response (Slawson & Shaughnessy, 2000). The preferable readings in researching the evidence are the high quality systematic reviews that can be found mainly in The Cochrane Database of Systematic Review, The Cochrane Controlled Trails Register, The Cochrane Review Methodology Database [www.york.ac.uk/inst/crd/ welcome.htm], [www.cebm.jr2.ox.ac.uk], [www. hiru.macmaster/ca/cochrane/default.htm], [www. ich.ucl.ac.uk/srtu] and other sites. The systematic reviews are an important source of evidence because their authors take care for finding all relevant studies on the question, as-

sessing each study, synthesizing the findings in an unbiased way, very often after performing a meta-analysis. Expert critical appraisal about the validity and clinical applicability of systematic reviews can be found in the Database of Abstracts of Reviews of Effects (DARE), at the Website of the Centre for Review and Dissemination, University of York [www.york.ac.uk/inst/crd/welcome. htm], the sites shown in Table 1, as well as in [www.mrw.interscience.wiley.com/cochrane/ cochrane_cldare_articles_fs.html]. The next preferable readings as sources of evidence are randomized controlled trials (RCT) – the most appropriate research design for studying the effectiveness of a specific intervention or treatment. There is considerable information about how to plan, perform, and report a RCT as STARD checklist of items to include in diagnostic accuracy study (Bossuyt, & al, 2003), and in a randomized trial (Altman, & al, 2001). All other study designs, especially already mentioned outcomes studies (CLSI HS6-A, 2004) can also produce high quality evidence. The findings of systematic reviews, RCT and other studies provide the evidence for the development of clinical guidelines based on the principles of evidence-based medicine. The clinical guidelines are destined to bridge the gap between research and practice, to base the clinical decisions on a research evidence, and to make this evidence available globally. The clinical guidelines are systematically developed statements designed to help practitioners and patients to decide on appropriate healthcare regarding specific clinical conditions and/or circumstances. Guidelines de-

183

Quality and Reliability Aspects in Evidence Based E-Medicine

velopment, legal implications, implementations, and evaluation, are described in „What is?” series [www.evidence-based-medicine.co.uk], in [www. healthcentre.org.uk/hc/library/guidelines.htm], [www.nice.org.uk], and some others sites. The guideline development follows the principles of EBM. It starts with a clear definition of key questions about patients involved, alternative strategies, clinical outcomes, and methodological quality of the available evidence. A systematic review of each question should be available, and the evidence in the guideline should be arranged by their levels of power - from randomized controlled clinical trials and basic clinical research to well-conducted observational studies focused on patient-important outcomes and inferences. Nonsystematic clinical studies and expert opinion also are acceptable. A concise, continuously updated, and easy-touse collection of clinical guidelines for primary care, combined with the best available evidence, is

offered by [http://ebmg.wiley.com]. The collection covers a wide range of medical conditions with their diagnostics and treatments. The high-quality evidence is graded from A (that means “strong evidence exists and further research is unlikely to change the conclusion”) to D (that means “weak evidence and the estimate of effect is uncertain”). All reviews cited in the site are coming from The Cochrane Database of Systematic Reviews. It is likely that a complete source of clinical guidelines is the National Guideline Clearinghouse, United States [http//www.guideline.gov/ index.asp]. As recommended by EBM (Sackett, & al, 2000), each guideline contains a description of the methods used to collect and select the evidence and the rating scheme for the strength of this evidence. Such a rating scheme, used in the diabetes mellitus guideline (revised 2007), is given in Table 2 as an example [http//www. guideline.gov].

Table 3. Evidence hierarchy scheme [http//www.guideline.gov] from the guideline of Diabetes mellitus (with small modifications) LEC*

Study Design or Information Type

Comments

1.

RCT** Multicenter trials Large meta-analyses With quality rating

Well-conducted and controlled trials at 1 or more medical centers Data derived from a substantial number of trials with adequate power; substantial number of subjects and outcome data Consistent pattern of findings in the population for which the recommendation is made – generalizable results Compelling nonexperimental, clinically obvious evidence (e.g., use of insulin in diabetic ketoacidosis); “all or none” evidence

2.

RCT Prospective cohort studies Meta-analyses of cohort studies Case-control studies

Limited number of trials, small number of subjects Well-conducted studies Inconsistent findings or results not representative for the target population

3

Methodologically flawed RCT Nonrandomized controlled trials Observational studies Case series or case reports

Trials with 1 or more major or 3 or more minor methodological flaws Uncontrolled or poorly controlled trials Retrospective or observational data Conflicting data with a weight of evidence unable to support a final recommendation

4

Expert consensus Expert opinion based on experience Theory-driven conclusions, Unproven claims, Experience-based information

Inadequate data for inclusion in level-of-evidence categories 1, 2, or 3; data necessitates an expert panel’s synthesis of the literature and a consensus. *Level-of-Evidence Category, **Randomized Controlled Trials.

184

Quality and Reliability Aspects in Evidence Based E-Medicine

According to the principles of EBM, guidelines for clinical practice are also subject of critical appraisal and continuous improvement. For example, the recommendations made in the diabetes mellitus guideline (edition 2000) were found to be in agreement when concerning the general management and clinical care of type 2 diabetes, but some important differences in treatment details were noticed. The influence of professional bodies such as the American Diabetes Association was seen as an important factor in explaining the international consensus. It was also noticed that the globalisation of recommended management of diabetes is not a simple consequence of the globalisation of research evidence and deserves more attention (Burgers, & al, 2002). Sometimes, the principles of EBM might escape the authors of clinical guidelines. The core question of the guideline may not be clearly defined, the outcomes of interest may not be explicit, the systematic reviews may be missing, the quality of the primary studies may contain different kinds of bias or patient values and preferences might be neglected. As a result, the quality of the guidelines may be compromised. On the other side, data from clinical trials are often focused on specific patient populations and conditions. As a result, patient safety and trial efficiency are maximized, but may not reflect the majority of patients in clinical practice. The information obtained from such an artificial environment to clinical practice can be problematic if transferred directly to the clinical practice Nevertheless the usefulness of the clinical guidelines was appreciated from medical professionals soon after their dissemination. The existing considerable variations in resources, needs and traditions, stimulated many countries to establish their own processes for development or adaptation of clinical guidelines. In 2001, the Council of Europe developed a set of recommendations for producing clinical guidelines and for assessing their quality. The international collaboration created the European

research project PL96-3669 (AGREE, Appraisal of Guidelines, Research and Evaluation in Europe). The efforts of researchers from Health Care Evaluation Unit at St George’s Hospital Medical School in London, Guidelines International Network (an international network for guidelines developing organisations), policy makers, and others have contributed substantially for creating an international collaboration for implementation of the project. The project ultimate goal is to improve the quality and effectiveness of clinical practice guidelines by the introduction of an appraisal instrument, development of standard recommendations for guideline developers, programmes, content analysis, reporting, and appraisal of individual recommendations. The development programme concerns guidelines on asthma, diabetes and breast cancer. All of the guidelines are based on the principals of EBM and are presented in [www.agreecollaboration.org/instrument/] and [http://www.agreecollaboration.org]. In connection with the AGREE, several studies show that guidelines on the same topic may differ, possibly due to different medical practice, insufficient evidence, different interpretations of evidence, unsystematic guideline development methods, influence of professional bodies, cultural factors such as differing expectations of apparent risks and benefits, socio-economic factors, cultural attitudes to health or characteristics of health care systems resources, and patterns of disease (Cluzeau, & al, 2003; Burgers, & al, 2003; Berti, & al, 2003). In such countries, a guideline developed on the basis of the best current evidence and prepared for global dissemination, might appear irrelevant (Fried, & al, 2008). Several sessions of AHRQ (an USA Federal Agency for Healthcare Research and Quality) are also directed towards assessment of the guideline implementation. The AHRQ website [www.ahrq. gov] contains high quality information for Clinical practice guidelines, Evidence-based practice, Funding opportunity, Quality assessment, Health IT Home, and others. The critical role of guideline

185

Quality and Reliability Aspects in Evidence Based E-Medicine

implementation is explored. The common conclusions are that access to knowledge must be close to the point of decision and that the importance of guideline implementation into practice depends on using recommendations at the right time for the right patient. The same information may be delivered in different manners and may lead to different results (Slutsky, 2008). Obviously, the opportunity to obtain the relevant information on time at the point of decision-making might influence critically the patient outcomes. On the other side, national evidence-based guidelines are missing in some countries and optimal evidence is often not available for many clinical decisions. Another study considers an important and sometimes contradictory aspect of evidence-based medical practice – rational prescribing. It is shown that many factors other than evidence drive clinical decision-making. Amongst them are patient preferences, social circumstances, disease-drug and drug-drug interactions, clinical experience, competing demands from more urgent clinical conditions, marketing or promotional activity, and system-level drug policies. For example, despite the availability of trials and strong evidence for the positive effect of pravastatin and simvastatine, the new statines with less evident effects are often prescribed. It is likely that the lack of time and training for evidence evaluation allows some well-intentioned clinicians to be influenced by promotional information at the time of pharmaceutical product launch (Mamdani, & al, 2008). The possible interference of industry in clinical trials is well summarized in an expert article. It is shown that the priorities of the pharmaceutical industry, in terms of disease targets and the entities that are chosen to test might not reflect the needs of patients or the community. “The goals of industry can, therefore, become the hidden agenda behind the generation of much of the evidence that is used to construct guidelines. The profusion of clinical trials in lucrative diagnostic categories is testament to this phenomenon” (Fried, & al, 2008).

186

Whatever a guideline is, the clinician has to translate the available evidence to the management of an individual patient. However, the patient conditions are complex and often presented with chronic or co-morbid complications for which a suitable guideline may not exist. The real patient can be very different from the population included in clinical trials and the real clinician is, in fact, practicing an “opinion-based medicine” (Hampton, 2002). The evidence-based medicine has made a substantial impact on medical research and practice of medicine through the clinical guidelines. However, the need of regular recalls of EBM concepts and skills becomes necessary for the critical appraisal, validity, and applicability of the guidelines. Only evidence-based recommendations may consider the balance between benefits, risks and burdens, and weigh these considerations using patient and societal, rather than expert, values. With the aim to avoid discrepancies some organizations use different systems to grade evidence and recommendations. As a result, the current proliferation and duplication of guidelines is wasteful (Townend, 2007). A revision of the current practice for development and dissemination of guidelines seems to be adequate for better understanding and easier application. It is proposed that a uniform grading system (GRADE) achieve widespread endorsement. Then, specialty groups can produce a central repository of evidence that is regularly updated with the available systematic reviews. “From this central resource, regional guidelines could be developed, taking into account local resources and expertise and the values and preferences important in that population” (Guyatt, & al, 2006; Fried, & al, 2008),(www. gradeworkinggroup.org/publications/index.htm; http://www.gradeworkinggroup.org/intro.htm. The first step in this direction is already completed. The National Guidelines Clearing House, USA, already developed syntheses of guidelines covering similar topics [http://www.ngc.org/ compare/synthesis.aspx] (accessed 17 October

Quality and Reliability Aspects in Evidence Based E-Medicine

2007). BMJ as well requests that authors should preferably use the GRADE system for grading evidence when submitting a clinical guidelines article to the journal [www.bmj.com]. With GRADE the patients and clinicians have the opportunity to weigh up the benefits and downsides of alternative strategies when making healthcare management decisions. Much more about the balance and uncertainty between benefits and risks, decisions about an effective therapy, resource and cost can be learned by UpToDate (an electronic resource widely used in North America), [www.uptodate.com]. “When dealing with resource allocation issues, guideline panels face challenges of limited expertise, paucity of rigorous and unbiased cost-effectiveness analyses, and wide variability of costs across jurisdictions or health-care systems. Ignoring the issue of resource use (costs) is however becoming less and less tenable for guideline panels” (Guyatt, & al, 2006).

CONCLUSION The e-medical resources are useful and probably will continue to improve. With some crucial governmental support the delivery of healthcare services to patients in rural, lonely, or distant regions with shortages of medical staff will be facilitated. The e-medicine undoubtedly can reduce the cost of on-site communication between patients, doctors, and insurance agencies. In many cases, however, the information about the quality of the supporting evidence of the e-health service will remain limited. The e-sources will continue to provide excellent specific information for health professionals who have recognized the need to learn and understand how to discover, assess, and use the best current evidence in decision-making about the individual patient care. The principles of EBM deserve to become a way of thinking for governmental clerks, in particular when decisions for population health strategies are made, and for politicians when promises are disseminated.

The invasion of the market principles into the realm of medicine is a reality and the customer satisfaction tends to become a priority. However, common sense and governmental regulations are needed to avoid the transformation of the quality medical care into hedonistic “satisfaction-based medicine”. The results from outcome studies and assessments based on quality indicators seem to be a more realistic foundation for measuring the quality of healthcare for both the patients and the societies.

ACKNOWLEDGMENT Special thanks to Dr. Geroge Daskalov, St Francis Hospital - Hartford CT, USA for help and support.

REFERENCES Altman, D. G., Schltz, K. F., Mohler, D., & al. (2001). The revised CONSORT statement for reporting randomized trials: Explanation and elaboration. Annals of Internal Medicine, 134, 663-694. See also [http://www.annals.org/cgi/ content/abstract/134/8/663] and associated Web site [www.consort-statement.org]. AABB. (2003). American Association of Blood Banks, Standards for Blood Banks and Transfusion Services (22nd ed). Bethesda, MD. Retrieved from www.aabb.org Burgers, J. S., Bailey, J. V., & Klazinga, N. S., & al. (2002). Inside Guidelines: Comparative analysis of recommendations and evidence in diabetes guidelines from 13 countries. Diabetes Care, 25, 1933–1939. doi:10.2337/diacare.25.11.1933 Burgers, J. S., Cluzeau, F. A., & Hanna, S. E., & al, & the AGREE Collaboration. (2003). Characteristics of high quality guidelines: evaluation of 86 clinical guidelines developed in ten European countries and Canada. International Journal of Technology Assessment in Health Care, 19(1), 148–157. 187

Quality and Reliability Aspects in Evidence Based E-Medicine

Bossuyt, P. M., Reitsma, J. B., & Bruns, D. E., & al. (2003). Towards complete and accurate reporting of studies of diagnostic accuracy. Clinical Chemistry, 49, 1–6. doi:10.1373/49.1.1

Gardois, P., Grillo, G., Lingua C., & al. (2004). Assessing the efficacy of EBM teaching in a clinical setting Santander 9 EAHIL Conference,September 24,2004.

CLSI publication GP22-A2. (2004). Continuous Quality Improvement: Integrating Five Key Quality System Components; Approved Guideline (2nd ed.). Wayne, Pennsylvania, USA: CLSI.

Glance, L. G., Osler, T. M., Mukamel, D. B., & Dick, A. W. (2008). Impact of the present-onadmission indicators on hospital quality measurement experiance with the Agency for Healthcare Research and Quality. (AHRQ) Inpatient Quality Indicators. Medical Care, 46(2), 112–119. doi:10.1097/MLR.0b013e318158aed6

CLSI publication HS1-A2. (2004). Quality Management System Model for Health Care; Approved Guideline (2nd ed.). Wayne, Pennsylvania, USA: CLSI. CLSI publication HS6-A. (2004). Studies to Evaluate Patients Outcomes; Approved Guideline. Wayne, Pennsylvania, USA: CLSI. Cluzeau, F. A., Burgers, J. S., & Brouwers, M., & al. (2003). Development and validation of an international appraisal instrument for assessing the quality of clinical practice guidelines: the AGREE project. Quality & Safety in Health Care, 12(1), 18–23. doi:10.1136/qhc.12.1.18 Collins, P. M., Golembeski, S. M., Selgas, M., & al. (2007). Clinical Excellence Through EvidenceBased Practice - A Model to Guide Practice Changes. Topics in Advanced Practice Nursing eJournal. 7(4), Medscape, Posted 01/25/2008. Retrieved from [http://www.medscape.com/ viewarticle]. Detsky, A. S., Naglie, G., & Krahn, M. D., & al. (1997). Primer on Medical Decision Analysis. Medical Decision Making, 17, 123–233. doi:10.1177/0272989X9701700201 Fried, M., Quigley, E. M., Hunt, R. H., & al. (2008). Is an Evidence-Based Approach to Creating Guidelines Always the Right one? National Clinical Practice in Gastroenterology and Hepatology, 5(2), 60-61. Nature Publishing Group. Posted 03/10/2008.

188

Grunfeld, E., Urquhart, R., Mykhalovskiy, E., & al. (2008). Toward population-based indicators of quality end-of-life care: testing stakeholder agreement. Cancer, 112(10), 2301-8. (PreMedline Identifier: 18219238). Guyatt, G., Gutterman, D., Baumann, M. H., & al (2006). Grading Strength of Recommendations and Quality of Evidence in Clinical Guidelines Report from an American College of Chest Physicians. Task Force Chest, 129, 174-181. Retrieved March 16, 2009 from [http://www.chestjournal. org/content/129/1/174.full.pdf+html] Guyatt, G. H., Oxman, A. D., Vist, G. E., & al. (2008). Rating quality of evidence and strength of recommendations GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. British Medical Journal, 336, 924-926 (26 April), for the GRADE Working Group doi: 10.1136/bmj.39489.470347.AD. Retrieved from [http://www.gradeworkinggroup. org/intro.htm] Hampton, J. R. (2002). Evidence-based medicine, opinion-based medicine, and real-world medicine. Perspectives in Biology and Medicine, 45, 549–568. doi:10.1353/pbm.2002.0070 ISO 9000. (2000). Quality management systems - Fundamentals and vocabulary. Geneva: International Organization for Standardization.

Quality and Reliability Aspects in Evidence Based E-Medicine

Itagaki, M. W., Berlin, R. B., Bruce, R., & Schatz, B. R. (2002). The Rise and Fall of E-Health: Lessons From the First Generation of Internet Healthcare. Medscape General Medicine 4(2). Retrieved August, 6, 2009 from [http://www. medscape.com/viewarticle/431144] .

Slawson, D. C., & Shaughnessy, A. F. (2000). Becoming an information master: using POEMs to change practice with confidence. Patient-oriented evidence that matters. [Retrieved from]. The Journal of Family Practice, 49, 63–67.

Kolbasuk, M., & McGree, N. (2009). Retrieved June 17, 2009, from http://www.docmemory.com/ page/news/shownews.asp?num=11826

Slutsky, J. (2008). Session Current Care - G-I-N abstracts. Retrieved March, 27, 2009 from [http:// www.g-i-n.net/download/files/G_I_N_newsletter_May_2008.pdf]

Legido-Ouigley, H., McKee, M., & Walshe, K., & al. (2008). How can quality of healthcare be safeguarded across the European Union? [Retrieved from]. British Medical Journal, 336, 920–923. doi:10.1136/bmj.39538.584190.47

Steel, N., Melzer, D., & Shekelle, P. G., & al. (2004). Developing quality indicators for older adults: transfer from the USA to the UK is feasible. Quality & Safety in Health Care, 13(4), 260–264. doi:10.1136/qshc.2004.010280

LeRouge, C., Hevner, A., Collins, R., & al. (2004). Telemedicine Encounter Quality: Comparing Patient and Provider Perspectives of a SocioTechnical System. Proc. 37th Hawaii International Conference on System Sciences. Retrieved March 25, 2009 from [http://www2.computer. org/portal/web/csdl/doi?doc=abs/proceedings/ hicss/2004/2056/06/205660149a.abs.htm]

Townend, J. N. (2007). Guidelines on guidelines. Lancet, 370, 740. doi:10.1016/S01406736(07)61376-2

Mamdani, M., Ching, A., Golden, B., & al. (2008). Challenges to Evidence-Based Prescribing in Clinical Practice. Annals of Pharmacotherapy, 42(5), 704-707. Harvey Whitney Books Company. Posted 07/15/2008. Retrieved from [http://www. medscape.com/viewarticle/576145]. MARQuIS research project. Qual Saf Health Care, 18, i1-i74. (2009) National Guideline Clearinghouse (NGC), Guideline Syntheses (accessed 17 October 2007). Retrieved from [http://www.ngc. org/compare/synthesis.aspx]. Sackett, D. L., Straus, S. E., Richardson, W. S., Rosenberg, W., & Haynes, R. B. (2000). EvidenceBased Madicine. Now to Practice and Teach EBM. Second Edition, Churcill Livingstone, Edinburg, London, New York & al., as well as in [http:// www.library.utoronto.ca/medicine/ebm/].

Tudiver, F., Rose, D., Banks, B., & Pfortmiller, D. (2004). Reliability and Validity Testing of an Evidence-Based Medicine OSCE Station. Society of Teachers in Family Medicine meeting, Toronto, May 14, 2004. Date Submitted: May 22, 2008. Retrieved March 22, 2009 from http://www.stfm. org/fmhub/fm2009/February/Fred89.pdf UEMS. 2007.19 – Bratislava Declaration on eMedicine. (2009, February 25). Retrieved June 17, 2009,from http://admin.uems.net/uploadedfiles/893.pdf. Veillard, J., Champagne, F., & Klazinga, N., & al. (2005). A performance assessment framework for hospitals: the WHO regional office for Europe PATH Project. International Journal for Quality in Health Care, 17(6), 487–496. doi:10.1093/ intqhc/mzi072

This work was previously published in E-Health Systems Quality and Reliability: Models and Standards, edited by Anastasius Moumtzoglou and Anastasia Kastania, pp. 100-117, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global). 189

190

Chapter 1.13

E-Medical Education: An Overview

D. John Doyle Case Western Reserve University, USA & Cleveland Clinic Foundation, USA

ABSTRACT The World Wide Web has made available a large variety of medical information and education resources only dreamed of two decades ago. This review discusses a number of Web-based e-Medical education concepts and resources likely to be of interest to the medical education community as well as a number of other groups. The resources described focus especially on those that are free and those that have an interactive component. The importance of interactivity and its role in the “constructivist” approach to educaDOI: 10.4018/978-1-60960-561-2.ch113

tion is emphasized. Problem-based learning in medical education is also discussed. In addition, the importance of “Web 2.0” and related developments is discussed, along with an overview of Web-based medical simulation software that can complement medical education programs. The importance of podcasts and videocasts as an educational resource are also emphasized. Other concepts such as mashups and the semantic Web are briefly described. Intellectual property issues are also discussed, such as Creative Commons license arrangements, as well as the concept of “information philanthropy”. Finally, the importance of peer-review and technology evaluation for online educational materials is discussed.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

E-Medical Education

INTRODUCTION The Rise of Web-Based Educational Systems “The student begins with the patient, continues with the patient, and ends his studies with the patient, using books and lectures as tools, as means to an end.” -- Sir William Osler, Aequanimitas, 1905 Since the coming of age of the World Wide Web a little over a decade ago, a true revolution in access to information of all kinds has taken place. In particular, recent years have witnessed a quasi-revolution in conventional and distance education as a result. Web-based educational technologies now allow for new approaches to education not previously dreamed of. While this is especially true in the field of distance education generally, it has also had a tremendous impact on medical education in particular (Gallagher, Dobrosielski-Vergona, Wingard, and Williams, 2005; Glinkowski and Ciszek, 2007; Jang, Hwang, Park, Kim and Kim, 2005; McKimm, Jollie and Cantillon, 2003; Sajeva, 2006; Schilling, Wiecha, Polineni and Khalil, 2006; Smith, Roberts and Partridge, 2007; Zary, Johnson, Boberg and Fors, 2006). Clinicians and medical students now have available to them a countless array of online medical journals, CME educational sites, discussion forums, medical search engines, podcasts, wikis, and blogs that they can use to enhance their ongoing learning. A great number of these resources are entirely free.

WEB-SITE DEVELOPMENT SOFTWARE Web pages are a particularly useful means of providing information of almost any kind. They are easily accessed and are updated relatively easily, at least with some development systems. Links to related documents can be provided, while support

for tables, graphics and multimedia objects like audio and video clips can usually be managed with minimal difficulty. As a result, Web pages have proven to be very popular for medical education. Web pages are primarily built using the HTML language or one of its successors (Katzman, 2001; Lynch and Horton, 1998; Ryan, Louis and Yee, 2005; Wiggins, Davidson, Harnsberger, Lauman and Goede, 2001). There are many advantages to using HTML for Web site development, one of which is that one does not need to buy any special software in order to use it - one can write Web pages directly in HTML using almost any text editor. Still, most Web developers prefer to use a specialized Web page editor, and there are a number of inexpensive or free HTML editors available (as a simple Google search will quickly reveal). Professionals usually use packages such as Macromedia’s Dreamweaver or Microsoft‘s Expression Web. While these are excellent, comprehensive, general-purpose Web page editors, they are not especially friendly to beginners, and for a number of small projects I have used a less well-known but very easy-to-use system known as Homestead (www.homestead.com). Regardless of the system chosen, consideration must also be given to the use of appropriate design principles in Web page construction (Cook and Dupras, 2004; Gotthardt et al, 2006; Wong, Greenhalgh, Russell, Boynton, and Toon, 2003). In this respect, Where possible, consideration should also be given to adding some degree of interactivity to the process (Coiera, 2003; Despont-Gros, Mueller, and Lovis, 2005; Reynolds, Mason and Harper, 2008; Ridley, 2007; Stout, Villegas, and Kim 2001). Web-based interactivity, discussed in more detail later in this article, can be implemented in various ways. For instance, a section of a Web page may ask the student a question, and offer four possible answers the student may select from. Depending on the student’s response, the Web page can provide a different commentary. Other forms of Web-based interactivity may involve the use of discussion forums or online surveys and

191

E-Medical Education

polls. Providing e-mail addresses (and “mailto” links) for authors or instructors can also help make material more interactive. Advanced interactivity may also be achieved using JavaScript, a popular and relatively easy to learn programming language for Web pages, as well as via a number of other specialized languages. Another concern is with Web site maintenance. All good web sites need periodic updating – otherwise the content becomes dated, hyperlinks become broken and other problems arise. You should plan for quarterly maintenance updates to your site to keep it topical and respectable. A final point concerns Intellectual Property (IP) issues. It is not uncommon for individuals involved in the development of educational materials to utilize in their work useful material (especially images) found on the Web. However, if done without permission, this practice is potentially problematic, even with appropriate attribution, as the use of the Intellectual Property belonging to others – whether commercial or otherwise generally requires that permission be granted. It should be noted, however, that a number of excellent educational resources such as Wikipedia operate under a “Creative Commons” license arrangement. Creative Commons is “a nonprofit corporation dedicated to making it easier for people to share and build upon the work of others, consistent with the rules of copyright.” Creative Commons works “to increase the amount of creativity (cultural, educational, and scientific content) in “the commons” — the body of work that is available to the public for free and legal sharing, use, repurposing, and remixing.” This arrangement often greatly facilitates the legal reuse of valuable educational materials. The interested reader is directed to http://creativecommons.org for further details. As an example of such an arrangement, an individual seeking to freely use an illustration of a sugammadex molecule encapsulating a rocuronium molecule can get one that the creator has generously released into the public domain at http://

192

en.wikipedia.org/wiki/Image:Sugammadex_encaps_rocuronium.jpg).

WHAT IS WEB 2.0 AND WHY IS IT IMPORTANT? The original launch of the World Wide Web circa 1994 soon developed into a series of offerings that took the online world by storm and quickly created no small number of multi-millionaires. Such early offerings soon included corporate and personal Web sites, Web-based mail, Internet commerce, Internet radio, etc. For convenience, some authors describe this “original” version of the Web as Web 1.0. Since that time, however, the Web has advanced and developed so much that some authors now use the term “Web 2.0” to describe the many recent developments that stress openness, sharing, cooperation and collaboration among Web users (Atreja, Messinger-Rapport, Jain, and Mehta, 2006; Boulos, Maramba, and Wheeler 2006; Kamel Boulos, and Wheeler, 2007; McGee, and Begg, 2008; McLean, Richards, and Wardman 2007). Such developments include Wikis, blogs, podcasts, mashups, and RSS feeds (vide infra). While a formal definition of Web 2.0 that everyone agrees on would be hard to find, one way of understanding it would be to point out its emphasis on participation through open applications and services, rather than define it (as others might) in terms of newer languages such as AJAX or XML, or advanced online applications such as Goggle Maps or Wikipedia. The emphasis in Web 2.0 is thus on developing applications and systems so that they get smarter and more comprehensive the more people use them (leveraging collective efforts and intelligence), rather than emphasizing any technical properties or specifications. It is this special ‘collaborationware’ focus of many Web 2.0 applications as well as their simplicity of use and speed of deployment, that has lead their enthusiastic adoption in many medical educational

E-Medical Education

arenas. Another emphasis in Web 2.0 is on the automatic granting of permission to use existing content in new and exciting ways (“information philanthropy”) (Doyle, Ruskin and Engel, 1996), discussed later. Yet another focus is that of services over software. Finally, as Web 2.0 continues to develop, it is expected to embrace new modes of communication and interaction such as the use of communication via cell phones, the use of instant messaging, learning via virtual reality techniques, and even learning via online social gaming. In some cases the connections between these technologies and clinical practice may be surprising. For instance, in a study by Rosser, Lynch, Cuddihy, Gentile, Klonsky and Merrell (2007) the authors discovered that surgical trainees who had played video games more than 3 hours per week worked faster and made fewer errors when performing laparoscopic surgical tasks. Let us now look at some of these tools in more detail.

Wikis A Wiki is a collaborative informational Web site whose content is open (freely accessible) and can be edited by anyone who is duly authorized. The name comes from the Hawaiian word wiki, meaning to hurry or swift. The best known Wiki is Wikipedia (www.wikipedia.org), which as of this writing has over 2.6 million articles in English. It is sometimes held out as a one of the best examples of Web 2.0, with its openness in both access and in its participation model. However, Wikipedia’s openness in allowing virtually anyone to participate has lead to criticism (Taylor-Mendes, 2007), even though there is evidence that it may be just as accurate as commercial encyclopedias (Giles, 2005). Still, such concerns have lead to the launching of more rigorously edited free online encyclopedias such as Scholarpedia (http://www.scholarpedia.org) and Citizendium (http://citizendium.org). For instance, the approach of Scholarpedia is to avoid competing with Wikipedia, but rather to comple-

ment it, covering a few narrow fields (such as computational neuroscience) in an exhaustive manner. It should also be pointed out that some of the “alternatives” to Wikipedia that have been developed may be seriously flawed. As an example, Conservapedia (http://www.conservapedia.com) is a Web-based encyclopedia written from a Conservative Christian point of view. As a result, it contains articles that support the “Young Earth” creationist model (the earth being a mere 6000 years old) or that present evolution as a discredited scientific theory in conflict with the fossil record.

Medical Education Wikis One challenge with Wikipedia is that its entries dealing with medical topics tend to vary greatly in quality. While blatant technical inaccuracies are unusual, many of Wikipedia’s medical entries lack the perspective and balance usually found in well-established medical textbooks. This situation exists in part because no special training or clinical experience is required to contribute to any Wikipedia article. While it is true that clinicians unhappy with any Wikipedia medical article can take it upon themselves to edit it to their satisfaction, this seems not to be done frequently. In all likelihood, this is at least partly because academics generally get no credit towards tenure or promotion for contributions of this kind. Because Wikipedia is often weak and incomplete in their treatment of medical subjects, efforts to develop medical Wikis have been launched. Ganfyd (http://ganfyd.org) is an online collaborative medical reference that is edited by medical professionals and other experts. Sermo (http:// sermo.com) is a wiki-like medical resource that is only accessible by physicians practicing in the USA who can prove their medical credentials by providing their “DEA number”. A more specialized medical wiki example is Flu Wiki (http://www. fluwikie.com), which was launched to assist in the planning for possible avian influenza pandemics.

193

E-Medical Education

Wikisurgery (http://www.wikisurgery.com) is a free collaborative surgical encyclopaedia “with over 1,300 articles for surgeons and patients, including news, articles, operation scripts, biographies and images.” Yet another example is the Diabetes Wiki (http://diabetes.wikia.com/wiki/ Diabetes_Wiki). A fairly comprehensive list of medical Wikis is available at http://www.davidrothman.net/ list-of-medical-wikis. For more information on Wikis in a medical context see a review by Johnson, Freeman and Dellavalle (2007).

Blogs Blogs (short for “Web logs”) are Web sites intended for online audiences containing a diary-like account of events and personal insights. While a great many blogs are social, political or technical in nature, a number focus on medical issues. Often written by doctors, these blogs regularly discuss the latest developments in diagnosis, treatment or basic science. Some give the authors a chance to rant at “Big Pharma” and their expensive drugs, to complain about HMOs and their “unreasonable” policies, or to complain about malpractice lawyers and the ever rising cost of medical liability insurance. Others are more focused on providing up-todate clinical information or in telling interesting clinical anecdotes. For an excellent example of a medical blog, see http://www.kevinmd.com. For more general information in a medical context see a review by Sethi (2007). Finally, loosely related to blogging is Twitter (http://twitter.com), a microblogging technology where users can send and receive concise, textbased information. While it has not yet seen deep penetration in the medical field, this may be only a matter of time as this and other social media services, such as YouTube, Facebook, MySpace, and Second Life emerge as popular sources of health information, especially for young adults.

194

Critique of Wikis and Blogs One concern with both wikis and blogs is the quality of their content, an issue identified earlier. However, there is hope that, at least for some popular resources, the collaborative nature or the resource will lead to self-correction. For instance, the openness of many wikis has given rise to the concept of “Darwikinism” where “unfit” wiki material content is zealously edited by other users (McLean, Richards, and Wardman, 2007). One interesting question that arises in relation to medical blogging is: “What ethical standards should medical bloggers be held to in their postings?” I would argue that where these individuals are health professionals, ideally at least, they should be ethically bound to the same rules and standards of professionalism as professional journalists: any specific medical information posted should be accurate, should come from a trusted, authoritative source, and a citation to that source should be provided. This is not always the case, as exemplified by some blogs that promote the view that HIV does not cause AIDS (e.g., http:// www.aids-science.blogspot.com) When medical bloggers write about situations or cases encountered in their own clinical practice, specific attention should be given to protecting patient confidentiality. Ideally, medical bloggers may wish to have their blog or medical Web site accredited by the Health on the Net Foundation, which provides endorsement to websites which adhere to the foundation’s guidelines. (More information is available at the foundation’s Web site (http://www.hon.ch), which outlines the eight principles that the Health on the Net Foundation requires for endorsement.)

Podcasts / Videocasts A podcast is a downloadable audio or video presentation for replay on a personal media player such as Apple’s iPod. A rich variety of medical podcasts are available to interested individual

E-Medical Education

(e.g., itunes.com, podcast.com), although many require a subscription or payment While podcasts are sometimes used for undergraduate medical education, they are most commonly used in postgraduate medical education. A great many medical education podcasts consist of edited recordings of lectures at major medical conferences. In the case of lecture podcasts offered by Audio Digest and some other suppliers, it is possible to get formal CME credit upon completion of a quiz. This is important to many physicians who must obtain a particular number of CME credits to maintain their medical license. Users wishing free medical podcasts may wish to take advantage of those offered by the New England Journal of Medicine (http://content.nejm.org/misc/podcast.dtl). For more information on podcasts see a review by Abbasi (2006).

Mashups A mashup is a web application that presents information integrated from two or more data sources. An example of a health mashup is HealthMap (http://healthmap.org), which uses information from Google Maps (http://maps.google.com) or other sources to add cartographic context to epidemiological data so as to create a new service not originally provided by either source.

RSS (Really Simple Syndication) An RSS document is a Web resource known as a “feed” or “channel” that scans the Web to provide links to updated content (blog entries, news headlines, podcasts, etc.) in which one has previously indicated an interest. RSS documents also include “metadata” such as publication dates and authorship information in a standardized file format (“XML”) allows the information to be viewed by many different programs. For more information on RSS services in a medical context see a review by Wu and Li (2007).

Semantic Web A member of individuals such as Tim BernersLee, the inventor of Web 1.0, have expressed concerns about a number of aspects of Web 2.0 and have proposed as an alternative the concept of the “Semantic Web”, an ambitious undertaking where information would be automatically exchanged and acted upon on our behalf by the Web itself. While humans are capable of using the Web to carry out tasks such as ordering merchandise on eBay or searching for a good price on a cruise vacation, because web pages are designed to be read by people and not machines it is exceedingly hard for a computer to accomplish the same tasks without human direction. However, the vision offered by the Semantic Web is that the information it carried would be formatted so that it is “understandable” by computers, thereby reducing much of the tedious work involved in information-related web activities. Tim BernersLee originally expressed the vision of the semantic web as follows (Berners-Lee and Fischetti, 1999): “I have a dream for the Web [in which computers] become capable of analyzing all the data on the Web – the content, links, and transactions between people and computers. A ‘Semantic Web’, which should make this possible, has yet to emerge, but when it does, the day-to-day mechanisms of trade, bureaucracy and our daily lives will be handled by machines talking to machines. The ‘intelligent agents’ people have touted for ages will finally materialize.” In the domain of biomedicine, the Semantic Web has now begun to be explored in earnest (Mukherjea, 2005; Good and Wilkinson, 2006; Antezana, Kuiper and Mironov, 2009). Related to the Semantic Web is the notion of “Semantic Publishing”, where information and documents are published with explicit “semantic markups” intended to allow computers to comprehend the structure and meaning of published information. While the Semantic Web and Seman-

195

E-Medical Education

tic Publishing is expected to eventually change the face of web publishing in dramatic ways, there is much progress to be made before it becomes commonplace. For more information the reader is directed to discussions by Berners-Lee and Hendler (Berners-Lee and Hendler, 2001) and Shadbolt and -Lee (Shadbolt and Berners-Lee, 2008).

INFORMATION PHILANTHROPY Recent years have seen the growth of a movement I have called “Information Philanthropy” (Doyle, Ruskin and Engel, 1996), which has as its goal making quality intellectual property available for free, usually as an Internet download. This material might be textbooks, graphic illustrations, photographs, videos, musical compositions, blueprints, schematics, DNA and amino acid sequences, computer programs, mathematical algorithms, poetry, chemical synthesis procedures, detective novels, simulation procedures, and even raw scientific data. In all such cases, the creator/owner of the material makes this material available without charge subject to possible restrictions—such as not allowing the material to be used for commercial or military purposes, or perhaps not permitting the material to be reworked (edited, remixed, etc.) into a derivative product. In a great many cases such materials are governed by a Creative Commons license, as discussed earlier.

Interactivity and the “Constructivist” Approach to Education Constructivism is the view that knowledge is “constructed” by the learner by testing ideas, concepts and approaches based on existing knowledge and one’s actively acquired experiences and those of others. Constructivist theory holds that knowledge is not merely acquired passively and that students learn best when they actively (often collaboratively) participate in problem-solving

196

and critical thinking while involved in a wellformulated learning activity. Many experts in adult learning argue that traditional lecture-based didactic teaching all-toooften encourages passive learning instead of the development of the higher order cognitive skills needed for “true” education. Arguing that active involvement is essential for effective learning, they suggest that adults learn best when drawing on previous experiences, using techniques such as group discussion, simulation exercises, and problem solving. That is, going beyond mere looking and listening motivates people to learn on their own, gives students the motivation to try out new ideas, and encourages them to critically examine issues that were once simple accepted passively. One educational element that has been advocated as a means to achieve active learning is “interactivity” (Table 1) since it is well-matched to the constructivist model of learning (ConCeição and Taylor, 2007). The Oxford English Dictionary defines interactivity as reciprocally active, allowing a two-way flow of information between source and user, responding to the user’s input. In recent years interactivity has been viewed as a particularly important component to educational undertakings, although the nature of the “interaction” will naturally vary with factors such as the subject domain (e.g., philosophy versus physics), and format (e.g., conventional versus distance education). Well-designed interactivity in educational systems can, at least in principle, help capture the learner’s interest, has the potential to speed learning, and even allows for continuous assessment of the degree to which the material is mastered. These goals are met by providing frequent and relevant user feedback, by recognizing when students misunderstand a concept, and by providing learning aids such as animations or graphs that vary depending on user input. However, badly designed interactivity can also impede student progress (vide infra). The importance of interactivity in learning is perhaps best illustrated by the fact that a number

E-Medical Education

Table 1. Types of Interactivity in teaching and education, with examples focusing on medical Education Person-to-person (student / teacher) interaction with both persons in direct verbal communication (“same place interaction”) formal classroom or auditorium teaching (such as a question and answer period at the end of a lecture) informal hallway or bedside discussions Person-to-person interaction (student / teacher) using remote communication technology (“different place interaction”) telephone and audio conferencing writing (paper, e-mail) video conferencing telemedicine (e.g., clinical discussion centering around a radiographic image) Paper-based interactivity [can also be made into a computer model] interactive paper scenarios (If X is true then go to page Y otherwise go to page Z) Example 1: Patient with chest pain presents to the ER where X= “admit patient to hospital” Example 2: Patient with abdominal pain presents to the ER where X= “operate on patient [laparotomy]” Paper-based interactivity based on “developer pen” technology [can also be made into to a computer model] developer pen technology allows the student to see the result of a study or laboratory test once it has been “ordered”; based on the result of the selected investigation the student selects other clinical options (e.g., asks more questions, attempts to elicit a particular clinical finding, orders another test) Computer based interactivity not using the World Wide Web. This usually involves using programming languages such as C++ or various older authoring platforms. Web-based interactivity. This usually involves using programming languages such as JavaScript or authoring platforms with Web support. Note: further distinctions can to be made between immediate (“synchronous”, “real time”) and delayed (“asynchronous”) interaction.

of journals either directly or peripherally address this topic. For instance, the Journal of Interactive Learning Research (JILR), available online at http://www.aace.org/pubs/jilr/default.htm, publishes manuscripts having to do with “the underlying theory, design, implementation, effectiveness, and impact on education and training of the following interactive learning environments.” Stevan Harnard (1992) has pointed out that electronic publication (and especially electronic publication via the Internet) provides a dimension of interactivity in document publication that is radically new. He notes: No other medium can offer an author the possibility of virtually instantaneous and ubiquitous interaction with his intended audience. Electronic publication is not just a more efficient implementation of paper publishing, it offers the possibility of a phase transition in the form (and hence also the content) of scholarly inquiry. Not only are the

boundaries between “informal” and “formal” literature blurred, but scholarly inquiry has always been a continuum, from the inchoate birth of new ideas and findings to their (in principle) endless evolution as inquiry carries on. Interactivity is also important in distance education. One simple definition of “distance education” is that it is the delivery of instruction that does not require the learner to be present in the same physical location as the instructor. Students in such learning environments are thus said to interact variously with the instructor, the content, the technology, and with other students. In addition to these four forms of interaction, further distinctions must to be made between instantaneous (“real time”, “synchronous”) and delayed (“asynchronous”) interaction, as well as between “same-place” and “different-place” interaction. Generally speaking, delayed and different-place interaction offer the student more flexibility and

197

E-Medical Education

opportunities for thought and reflection, while immediate and same-place interaction, some would say, allow for a greater sense of spontaneity, impulsiveness and even exhilaration. By contrast, with conventional classroombased teaching, the focus is predominantly on student-teacher interaction, with student-content interaction also taking place, especially when outof-classroom reading takes place. In the distance education setting, with the heavy use of self-study materials, the focus is usually on student-content interaction although interaction with technology is also frequently important. Various learners and instructors will have preferences regarding interactivity, and some forms of education will rely on some forms of interactivity more than others. Regardless, the wise use of interactivity will help “engage” students and help ensure an enduring learning experience. One particularly engaging form of interactivity in educational materials is the use of multiple choice questions (MCQs) (Collins, 2006; Moss, 2001). The use of MCQs not only make the material more engaging but also help detect learning gaps. In recent years the most popular format for MCQ’s is the “pick the best single answer” format (e.g., Which one of the following drugs would be the most likely to increase heart rate in a normal adult?). If the question has been answered correctly, the student is given an encouraging note with more details. If answered incorrectly, a different note is shown to the student. Until recent years, educational technology was not especially capable of maintaining a high degree of interactivity with a learner, in contrast, say, to the very high level of possible interaction between a teacher and a student discussing a complex philosophical matter, where a series of verbal exchanges may occur that involves a cognitive effort much more advanced than mere memorization or rote repetition. This matter is expected to change favorably as software technology continues to become more sophisticated

198

and as computer hardware advances in speed and power. In particular, advances in Artificial Intelligence (AI), and educational authoring systems will likely have a direct impact on the evolution of computer-based interactivity. Still, computerbased learning is much more complex than working through a textbook, and many students who are uncomfortable with computers see no reason to deviate from a curriculum based on well-written textbooks, particularly well-polished classics that are current and are accompanied by study guides and solutions manuals. Many advocates of computer-mediated distance education draw attention to the helpful aspects offered by computer-based methods and understate the kinds of communicative and technical difficulties learners may experience. When students use complicated equipment in their learning environment, it is important to ensure the technologies used help rather than hinder the learning experience. Students using unfamiliar technologies may experience frustration, anxiety and confusion and students involved in distance education using computer technologies may face numerous stumbling blocks unrelated to the material to be learned. These include computer hardware problems, software problems at the level of the operating system or the various software applications, as well as difficulties with Internet access. Files may mysteriously “disappear”. Attachments sent by e-mail may be “lost” in transit. Random “disconnects” from the Internet may frustrate the user. Unless students are moderately familiar with computers, substantial effort may be expended on technology-related issues rather than learning the intended materials. Thus technologies that are awkward, unintuitive or present the learner with a steep learning curve may hinder interaction and even impede learning. Various means by which technology can impede interactivity include: haphazard, unintuitive or disorganized user interfaces; poor screen layout; poor use of fonts and icons; systems that

E-Medical Education

respond too slowly to user inputs; or systems that crash frequently. While a detailed discussion of this matter is beyond the scope of this review, books by design gurus Jakob Nielsen (1994) and Donald Norman (1999) are recommended as a good place to explore the nature of good and bad user-interface design. Studies show that technical problems can occasionally render the educational undertaking to be a substantially negative experience. For instance, Hara and Kling (1999) carried out a study of a web-based distance education course at an American university and commented that distance learners may experience problems and frustrations that are not well documented in the distance education literature. They noted that three interrelated sources accounted for most of the problems: “lack of prompt feedback, ambiguous instructions on the web, and technical problems”. They concluded that the frustrations the students experienced “inhibited their educational opportunity”. In a related paper (Hara and Kling, 2000) similar concerns were voiced. The importance of interactivity in medical education is supported by a landmark metaanalysis of the effectiveness of formal CME by Davis, O’Brien, Freemantle, Wolf, Mazmanian and Taylor-Vaisey (1999) who showed that traditional didactic methods do not generally lead to a change of clinical practice, or to an improvement in patients’ health outcomes, whereas interactive techniques are more likely to. As result, some regulatory bodies, such as the Royal College of Physicians and Surgeons of Canada (RCPSC) specifically mandate that at least 25 per cent of the time of an approved educational event should be allocated for interactive learning, such as in the form of a question and discussion period. This serves to further engage the listener as well as to clarify some of the issues that may remain unanswered at the end of the formal part of the presentation.

PROBLEM-BASED LEARNING Traditional education practices from kindergarten through professional school often produce students who are disenchanted and bored with their education. All too often students are faced with a vast amount of information to memorize, a good deal of which is of little apparent relevance to future career goals. In particular, in the field of medical education, medical students have been traditionally provided with a series of lectures, laboratories and clinical sessions for four years (six years in Europe), and then released into internship and clinical practice with little in the way of actual skills in real-world clinical problem solving, the prevailing view being that these skills will emerge quickly enough in the hustle-bustle of the busy clinical world (Barrows, 1983). Studies in classical medical education have often been critical of this approach. What students learn, despite intense efforts on the part of both students and teachers, often fades quickly (Levine & Forman, 1973) and there is even evidence that natural problem solving skills may be impaired by this process (Barrows & Bennett, 1982). It also seems that many physicians are not able to continue their education after completion of their training (Ramsey & Carline, 1991). Problembased learning (PBL) has been advocated as one means to get around these difficulties (Albanese & Mitchell, 1993). Traditional education involves the delivery of information to students by means of lectures and demonstrations, usually supported by tutorials and laboratory sessions. In this model the instructor provides students with the information that they need. In Problem-based Learning (PBL) the students work in small groups under the guidance of an instructor or tutor to find for themselves the knowledge they need to solve real world problems. The tutors are usually not experts in the subject matter of the tutorial, but rather are facilitators who help the students find the answers for themselves. At the conclusion

199

E-Medical Education

of a PBL module, the students should have encountered the information necessary to solve the problem and, in so doing, should have gained knowledge and skills that in a traditional medical curriculum would have been circulated by lecture. PBL emphasizes group problem-analysis and independent self-directed study over teacher- or examination-driven education. As a consequence, it is believed that PBL may encourage students to become thoughtful problem-solvers and life-long learners (Barrows, 1983). The introduction of a PBL-based medical curriculum began in 1969 at McMaster University in Hamilton, Ontario, Canada (Schmidt et al., 1991). From the school’s outset, McMaster medical school structured the curriculum around actual clinical cases instead of teaching based on the traditional subjects of anatomy, biochemistry, physiology, pharmacology and so on. The introduction of PBL was based on a concern that many medical schools put too much emphasis on memorization of facts and little weight on developing problem solving skills or the self-directed study skills necessary for the life-long practice of medicine. Following its introduction at McMaster University, the PBL teaching model was later adopted by a significant number of other medical schools, as well as by schools teaching other professional disciplines such as business. PBL is now the instructional method of choice in an increasing number of medical schools around the globe (Schmidt et al., 1991). Problem-based learning is a ground-breaking and demanding approach to medical education ground-breaking because it involves a new way of using clinical material to help students learn, and demanding because it requires the instructor to play a facilitating and supporting role rather than a didactic one. For the student, problem-based learning emphasizes the application of knowledge and skills to the solution of problems (clinical cases) rather than the mere recall of facts.

200

Since the quantity of direct teaching is substantially reduced in a PBL curriculum, students are expected to assume greater responsibility for their own education, and the instructor’s role becomes variously one of coach, group facilitator, subject matter expert, instructional resource guide, and group consultant, where the instructor is expected to promote and support student involvement, to provide guidance to help keep students on track, to help prevent anyone giving inappropriate negative criticism, and even to assume the role of fellow learner. This arrangement is intended to promote group acquisition of information rather than the mere imparting of information to passive students by faculty (Vernon & Blake, 1993). In a PBL class, students are given a problem to work on. They then proceed through a number of stages as they move towards their solution: exploring the nature of the problem; making a clear statement of the problem; identifying the information needed to understand the problem; identifying resources needed to gather this information; generating possible solutions to the problem; scrutinizing the various solutions for strengths and weaknesses; identifying the best solution among the various possible solutions; and presenting the solution. These stages usually take place over several class discussion sessions along with students working individually on assigned tasks related to the problem outside of class time. Problem-based learning has now entered the realm of eMedical education. For example, Table 2 lists some sample medical PBL problems that may be viewed from the Internet.

WEB-BASED MEDICAL SIMULATION SYSTEMS In many respects, simulators offer the ultimate in interactivity and appear to have an excellent future in medical education. Computer-based simulators used in medical education fall into four general categories: (1) Screen-based text simulators, (2)

E-Medical Education

Table 2. Some sample medical Problem-Based Learning (PBL) cases available on the internet Bad Reaction - A Case Study in Immunology http://ublib.buffalo.edu/libraries/projects/cases/immunology/ immunology1.html Pediatric Infection Case http://medicalpblukm.blogspot.com/2009/08/case-files-3-pedsim-hot-and-drooling.html Into Thin Air - A Case Study in Physiology http://ublib.buffalo.edu/libraries/projects/cases/Denali1.html A Series of Cases in the Field of Obstetrics and Gynecology http://www.healthsystem.virginia.edu/internet/obgyn-clerkship/ PBLCases.cfm A Series of Biology Cases http://capewest.ca/pbl.html

Screen-based graphical simulators, (3) Manikinbased simulators, and (4) Virtual reality trainers. We briefly discuss the first two of these here, since both can be Web-based. Screen-based text simulators create verbal scenarios in which the user picks one of several responses and, based on the chosen response, a new text scenario is produced. For instance, in a scenario involving a patient presenting with a severe headache, the user may be offered options such as prescribing an analgesic or getting a CTscan of the head. Based on the user’s choice, a new narrative is then generated and more management choices are offered. Being purely text-based, screen-based text simulators are relatively simple to construct and require little memory, making them popular in early medical simulation efforts. However, since they are lacking in graphical elements, they are rarely used today. Screen-based graphical simulators such as ‘Gasman’ (Philip, 1986) and ‘Body’ (Smith and Davidson, 1999) aim to recreate elements of reality in graphic form on a computer screen, often to elucidate the pharmacokinetic and pharmacodynamic processes associated with drug administration. Usually, only a mouse interface is involved. While these simulations help one understand the

conceptual theory underlying clinical practice, they usually do not confer actual practical skills. Their strength lies in an ability to help one understand abstract concepts while remaining portable and relatively inexpensive. For individuals interested in respiratory medicine, ThoracicAnesthesia.com offers a free “education, information, and reference service” which include a series of didactic pieces and teaching slide sets, a series of journal article synopses, a bronchial anatomy quiz, and a bronchoscopy simulator. The bronchoscopy simulator is unquestionably the best part of the site. Here, using real time video based on an Adobe Flash platform one can navigate through the tracheobronchial tree with controls for moving your virtual bronchoscope forward, backwards, and in diagonal directions. This process is aided by a “macro” view on the left panel (called a “bronchial tree navigational map view”) and a “micro” view on the right panel (“bronchoscope view”). If desired, zooming in and out and adding or removing labels on the displayed material can be controlled simply by clicking on the appropriate controls. Most importantly, clicking on various labels in the bronchoscope view will launch further information in the map view, making the experience intuitive and even enjoyable.

PEER-REVIEW OF ONLINE EDUCATIONAL MATERIALS When an instructor selects teaching resources - especially in the case of medical teaching resources - he or she wants some assurance that the material is timely, accurate, well-written, and developed using sound pedagogical principles. Such assurances, some would argue, can be best obtained via a process of peer-review. While the peer-review process in scientific publishing is as old as scientific publishing itself, the idea of peer-review procedures for educational resources in relatively new. However, it is clearly an idea whose time has come.

201

E-Medical Education

Typically, scientific materials submitted for publication are reviewed by two or more expert referees who are particularly knowledgeable in the subject matter being discussed. Their comments and opinions form the basis upon which an editor will decide whether or not to publish the submitted material, and with what changes, if any. By contrast, formal peer-review of educational materials remains relatively uncommon. Historically, this was because until recently, publishing educational resources like books was usually always done in conjunction with a publishing house with its own internal quality control system. For instance, some book companies employ a pre-publication internal and external reviews to establish whether a manuscript is worthy of publication and to determine how it might be enhanced. With the rise of the Internet, however, the landscape has changed, and publishing no longer requires enormous expense and the cooperation of a publishing house. The prevalence of electronic scholarly publishing has thus increased dramatically in recent years due, largely as a consequence of the reach and immediacy of the Internet and its ability to handle varied forms of electronic media. The idea of establishing a peer-review process for educational resources is not entirely new. For instance, Project Merlot, located at http://merlot. org, seeks to provide high-quality teaching materials in a number of disciplines of higher education. It conducts reviews along three dimensions: 1) quality of the content, 2) usefulness as a teaching tool, and 3) ease of use. For peer-reviewed medical education materials, MedEdPORTAL, an offering from the Association of American Medical Colleges (AAMC) maintains an online presence at http://www.aamc. org/mededportal. According to its Web site MedEdPORTAL “is a Web-based tool that promotes collaboration across disciplines and institutions by facilitating the exchange of peer-reviewed educational materials, knowledge, and solutions” and serves “as a central repository of high quality educational materials such as PowerPoint presen-

202

tations, assessment materials, virtual patient cases, and faculty development materials”. Any medical educator may submit materials for possible publication on MedEdPORTAL. The material is then screened by an AAMC staff member to ensure that it meets the minimal requirements for inclusion and does not violate patient privacy. Suitable material is then assigned to three reviewers, at least one who will have expertise in the content area, and at least one who will have expertise in the educational format of the material. Another peer-reviewed medical resource is The Health Education Assets Library (HEAL), which maintains an online presence at http://www.healcentral.org. Established in 2000, HEAL has grown over time to provide over 22,000 resources such as images, video clips, audio clips, PowerPoint slidesets, documents in PDF format, etc. HEAL, MERLOT and MedEdPORTAL all make use of the well-known Creative Commons copyright system with their works. In general, all resources in these collections are licensed for free use, reproduction, and modification by users according to the specific terms of the Creative Commons license associated with that resource. Despite these encouraging developments, formal peer-review of educational resources remains uncommon. However, with academic credit now starting to be granted to individuals who produce freely downloadable peer-reviewed educational material such as that offered by HEAL and MedEdPORTAL, the volume and quality of such resources is bound to increase over time.

EDUCATIONAL TECHNOLOGY EVALUATION As new educational technologies (ETs) become available there is a need to evaluate them in some systematic way. Ideally any new ET would be compared to an existing standard ET using a randomized trial design, such as in randomly assigning students to one of two teaching systems

E-Medical Education

and testing for differences in outcome. In reality these kinds of studies are relatively rare because of high resource requirements and for other reasons, and we may be thus forced to use less perfect instruments to gauge the value of a new ET, using measures such as user preference scores, as a surrogate for rigorous outcome data. Thus, for instance, when some educators declare that interactive teaching provides better educational outcomes than the mere non-interactive display of information, they generally do not argue on the basis of randomized clinical trials (as one would generally do, for instance, in evaluating a new drug) but instead support their position on the basis of general principles. Thus the notion that adding interactivity to a teaching event such as a lecture improves learning retention has not been proven to be true on the basis of randomized trials such as those done in domains like oncology, but is held to be true on the basis of a collective wisdom in the medical education community via a process driven primarily by thought leaders in education, where these leaders periodically explore, review and comment on this and other issues on a regular basis.

CONCLUSION In the past, medical education relied predominantly on classic methods of instruction such as formal lectures, laboratory demonstrations, and examining patients in clinics or at the bedside. Even today, these methods of instruction remain important, and in particular, contact with patients as a learning experience remains paramount. However, the use of electronic and Web-based teaching methods has now become an important additional modality in medical education. When properly designed so as to be engaging and interactive, such methods can have a particularly favorable impact on the way medical education is carried out. As these methods continue to be developed and refined

it is expected that they will continue to impact favorably on medical education in the future. As compared to medical education from two decades ago that relied heavily on textbooks and lectures, medical education today utilizes a large number of web-based resources that make learning more engaging, more pleasant, and easier. For example, no longer is a trip to the library required to obtain up-to-date information on an unusual condition that a medical student may encounter in a clinical encounter. Instead, the student can get reliable clinical information virtually instantly from an increasingly rich number of trustworthy online clinical sources. The future of medical education is indeed very promising.

REFERENCES Abbasi, K. (2006). Journals join the podcast revolution. Journal of the Royal Society of Medicine, 99, 329. doi:10.1258/jrsm.99.7.329 Albanese, M., & Mitchell, S. (1993). Problembased learning: A review of the literature on its outcomes and implementation issues. Academic Medicine, 68, 52–81. doi:10.1097/00001888199301000-00012 Antezana, E., Kuiper, M., & Mironov, V. (2009). Biological knowledge management: the emerging role of the Semantic Web technologies. Briefings in Bioinformatics, 10, 392–407. doi:10.1093/bib/ bbp024 Atreja, A., Messinger-Rapport, B., Jain, A., & Mehta, N. (2006). Using Web 2.0 technologies to develop a resource for evidence based medicine. In Proceedings of the AMIA Annu. Symp., (pg. 847). Barrows, H. S. (1983). Problem-based, selfdirected learning. Journal of the American Medical Association, 250, 3077–3080. doi:10.1001/ jama.250.22.3077

203

E-Medical Education

Barrows, H. S., & Bennett, K. (1982). Experimental studies on the diagnostic (problem-solving) skill of the neurologist, their implications for neurological training. Archives of Neurology, 26, 273–277. Berners-Lee, T., & Fischetti, M. (1999). Weaving the Web: the original design and ultimate destiny of the world wide web by its inventor. San Francisco: Harper San Francisco. Berners-Lee, T., & Hendler, J. (2001). Publishing on the semantic web. Nature, 26, 1023–1024. doi:10.1038/35074206 Boulos, M. N., Maramba, I., & Wheeler, S. (2006). Wikis, blogs and podcasts: a new generation of Web-based tools for virtual collaborative clinical practice and education. BMC Medical Education, 15, 41. doi:10.1186/1472-6920-6-41 Coiera, E. (2003). Interaction design theory. International Journal of Medical Informatics, 69, 205–222. doi:10.1016/S1386-5056(02)00106-5 Collins, J. (2006). Education techniques for lifelong learning: writing multiple-choice questions for continuing medical education activities and self-assessment modules. Radiographics, 26, 543–551. doi:10.1148/rg.262055145 Conceição, S. C., & Taylor, L. D. (2007). Using a constructivist approach with online concept maps: relationship between theory and nursing education. Nursing Education Perspectives, 28, 268–275. Cook, D. A., & Dupras, D. M. (2004). A practical guide to developing effective web-based learning. Journal of General Internal Medicine, 19, 698–707. doi:10.1111/j.1525-1497.2004.30029.x Davis, D., O’Brien, M. A., Freemantle, N., Wolf, F. M., Mazmanian, P., & Taylor-Vaisey, A. (1999). Impact of formal continuing medical education: do conferences, workshops, rounds, and other traditional continuing education activities change physician behaviour or health outcomes? Journal of the American Medical Association, 282, 867–874. doi:10.1001/jama.282.9.867 204

Despont-Gros, C., Mueller, H., & Lovis, C. (2005). Evaluating user interactions with clinical information systems: a model based on humancomputer interaction models. Journal of Biomedical Informatics, 38, 244–255. doi:10.1016/j. jbi.2004.12.004 Doyle, D. J., Ruskin, K. J., & Engel, T. P. (1996). The Internet and medicine: past, present, and future. The Yale Journal of Biology and Medicine, 69, 429–437. Gallagher, J. E., Dobrosielski-Vergona, K. A., Wingard, R. G., & Williams, T. M. (2005). Web-based vs. traditional classroom instruction in gerontology: a pilot study. Journal of Dental Hygiene, 79, 7. Giles, J. (2005). Internet encyclopaedias go head to head. Nature, 15, 900–901. doi:10.1038/438900a Glinkowski, W., & Ciszek, B. (2007). WWWbased e-teaching of normal anatomy as an introduction to telemedicine and e-health. Telemedicine Journal and e-Health, 13, 535–544. doi:10.1089/ tmj.2006.0077 Good, B. M., & Wilkinson, M. D. (2006). The Life Sciences Semantic Web is full of creeps! Briefings in Bioinformatics, 7, 275–286. doi:10.1093/ bib/bbl025 Gotthardt, M., Siegert, M. J., Schlieck, A., Schneider, S., Kohnert, A., & Gross, M. W. (2006). How to successfully implement E-learning for both students and teachers. Academic Radiology, 13, 379–390. doi:10.1016/j.acra.2005.12.006 Hara, N., & Kling, R. (1999). Students’frustrations with a web-based distance education course: A taboo topic in the discourse. Indiana University, Rob Kling Center for Social Informatics. Retrieved June 15, 2009, from http://rkcsi.indiana. edu/archive/CSI/WP/wp99_01-octoberB.html

E-Medical Education

Hara, N., & Kling, R. (2000). Students’ distress with a web-based distance education course: An ethnographic study of participants’ experiences. Indiana University, Rob Kling Center for Social Informatics. Retrieved June 15, 2009, from http:// rkcsi.indiana.edu/archive/CSI/WP/wp00-01B. html Harnad, S. (1992). Interactive Publication: Extending American Physical Society’s DisciplineSpecific Model for Electronic Publishing. Serials Review, Special Issue on Economics Models for Electronic Publishing, pp. 58–61. Retrieved from http://users.ecs.soton.ac.uk/harnad/ Papers/ Harnad/harnad92.interactivpub.html Jang, K. S., Hwang, S. Y., Park, S. J., Kim, Y. M., & Kim, M. J. (2005). Effects of a Web-based teaching method on undergraduate nursing students’ learning of electrocardiography. The Journal of Nursing Education, 44, 35–39. Johnson, K. R., Freeman, S. R., & Dellavalle, R. P. (2007). Wikis: the application of Web 2.0. Archives of Dermatology, 143, 1065–1066. doi:10.1001/ archderm.143.8.1065 Kamel Boulos, M. N., & Wheeler, S. (2007). The emerging Web 2.0 social software: an enabling suite of sociable technologies in health and health care education. Health Information and Libraries Journal, 24, 2–23. doi:10.1111/j.14711842.2007.00701.x Katzman, G. L. (2001). Adobe acrobat: an alternative electronic teaching file construction methodology independent of HTML restrictions. Journal of Digital Imaging, 14, 9–13. doi:10.1007/ s10278-001-0018-8 Levine, H. G., & Forman, P. B. (1973). A study of retention of knowledge of neurosciences information. Journal of Medical Education, 48, 867–869.

Lynch, P. J., & Horton, S. (1998). Hypertext markup language as an authoring tool for CDROM production. The Journal of Biocommunication, 25, 20–25. McGee, J. B., & Begg, M. (2008). What medical educators need to know about “Web 2.0”. Medical Teacher, 30, 164–169. doi:10.1080/01421590701881673 McKimm, J., Jollie, C., & Cantillon, P. (2003). ABC of learning and teaching: Web based learning. BMJ (Clinical Research Ed.), 19, 870–873. doi:10.1136/bmj.326.7394.870 McLean, R., Richards, B. H., & Wardman, J. I. (2007). The effect of Web 2.0 on the future of medical practice and education: Darwikinian evolution or folksonomic revolution? The Medical Journal of Australia, 6, 174–177. Moss, E. (2001). Multiple choice questions: their value as an assessment tool. Current Opinion in Anaesthesiology, 14, 661–666. doi:10.1097/00001503-200112000-00011 Mukherjea, S. (2005). Information retrieval and knowledge discovery utilising a biomedical Semantic Web. Briefings in Bioinformatics, 6, 252–262. doi:10.1093/bib/6.3.252 Nielsen, J. (1994). Usability Engineering. San Francisco: Morgan Kaufmann Publishers. Norman, D. A. (1999). The Invisible Computer: why good products can fail, the personal computer is so complex, and information appliances are the solution. Cambridge, MA: MIT Press. Philip, J. H.(No date.). ‘Gasman’ - Understanding Anesthesia Uptake & Distribution. [Computer software]. Chestnut Hill, MA: Med Man Simulations. International Journal of Clinical Monitoring and Computing, 3, 165–173. doi:10.1007/ BF01716358

205

E-Medical Education

Ramsey, P. G., & Carline, J. D. (1991). Changes over time in the knowledge base of practicing internists. Journal of the American Medical Association, 266, 1103–1107. doi:10.1001/ jama.266.8.1103 Reynolds, P. A., Mason, R., & Harper, J. (2008). The many faces of interaction. British Dental Journal, 24, 565–570. doi:10.1038/sj.bdj.2008.409 Ridley, R. T. (2007). Interactive teaching: a concept analysis. The Journal of Nursing Education, 46, 203–209. Rosser, J. C., Lynch, P. J., Cuddihy, L., Gentile, D. A., Klonsky, J., & Merrell, R. (2007). The impact of video games on training surgeons in the 21st century. Archives of Surgery (Chicago, Ill.), 142, 181–186. doi:10.1001/archsurg.142.2.181 Ryan, A. G., Louis, L. J., & Yee, W. C. (2005). Informatics in radiology (infoRAD): HTML and Web site design for the radiologist: a primer. Radiographics, 25, 1101–1118. doi:10.1148/ rg.254045119 Sajeva, M. (2006). E-learning: Web-based education. Current Opinion in Anaesthesiology, 19, 645–649. doi:10.1097/ACO.0b013e328010bec8 Schilling, K., Wiecha, J., Polineni, D., & Khalil, S. (2006). An interactive web-based curriculum on evidence-based medicine: design and effectiveness. Family Medicine, 38, 126–132. Schmidt, H. G. (1993). Foundations of problem-based learning: some explanatory notes. Medical Education, 27, 422–432. doi:10.1111/j.1365-2923.1993.tb00296.x Sethi, S. K. (2007). Blog/web log - a new easy and interactive website building tool for a non-net savvy radiologist. Journal of Thoracic Imaging, 22, 115–119. doi:10.1097/01.rti.0000213578.47398. aa

206

Shadbolt, N., & Berners-Lee, T. (2008). Web science emerges. Scientific American, 299, 76–81. doi:10.1038/scientificamerican1008-76 Smith, N. T., & Davidson, T. M. (1999). ‘Body’ simulation - Anesthesia, physiology and pharmacology simulation [Computer software]. Point Roberts, WA: Advanced Simulation Corporation. Smith, S. F., Roberts, N. J., & Partridge, M. R. (2007). Comparison of a web-based package with tutor-based methods of teaching respiratory medicine: subjective and objective evaluations. BMC Medical Education, 1, 41. doi:10.1186/14726920-7-41 Stout, P. A., Villegas, J., & Kim, H. (2001). Enhancing learning through use of interactive tools on health-related websites. Health Education Research, 16, 721–733. doi:10.1093/her/16.6.721 Taylor-Mendes, C. (2007). Proceed with caution: using Wikipedia as a reference. Neonatal Network, 26, 140–141. Vernon, D. T., & Blake, R. L. (1993). Does problem-based learning work? A meta-analysis of evaluative research. Academic Medicine, 68, 550–563. doi:10.1097/00001888-19930700000015 Wiggins, R. H. III, Davidson, H. C., Harnsberger, H. R., Lauman, J. R., & Goede, P. A. (2001). Image file formats: past, present, and future. Radiographics, 21, 789–798. Wong, G., Greenhalgh, T., Russell, J., Boynton, P., & Toon, P. (2003). Putting your course on the Web: lessons from a case study and systematic literature review. Medical Education, 37, 1020–1023. doi:10.1046/j.1365-2923.2003.01673.x Wu, W. G., & Li, J. (2007). RSS made easy: a basic guide for librarians. Medical Reference Services Quarterly, 26, 37–50. doi:10.1300/ J115v26n01_04

E-Medical Education

Zary, N., Johnson, G., Boberg, J., & Fors, U. G. (2006). Development, implementation and pilot evaluation of a Web-based Virtual Patient Case Simulation environment--Web-SP. BMC Medical Education, 21, 10. doi:10.1186/1472-6920-6-10

Costin, D. L., Mackinnon, A. J., Griffiths, K. M., Batterham, P. J., Bennett, A. J., Bennett, K., & Christensen, H. (2009). Health e-cards as a means of encouraging help seeking for depression among young adults: randomized controlled trial. Journal of Medical Internet Research, 11, e42. doi:10.2196/jmir.1294

ADDITIONAL READING

Donnelly, L. S., Shaw, R. L., & van den Akker, O. B. (2008). eHealth as a challenge to ‘expert’ power: a focus group study of internet use for health information and management. Journal of the Royal Society of Medicine, 101, 501–506. doi:10.1258/jrsm.2008.080156

Atienza, A. A., Hesse, B. W., Baker, T. B., Abrams, D. B., Rimer, B. K., Croyle, R. T., & Volckmann, L. N. (2007). Critical issues in eHealth research. American Journal of Preventive Medicine, 32, S71–S74. doi:10.1016/j.amepre.2007.02.013 Atreja, A., Messinger-Rapport, B., Jain, A., & Mehta, N. (2006). Using Web 2.0 technologies to develop a resource for evidence based medicine. In Proceedings of AMIA Annu. Symp., (pg. 847). Beard, L., Wilson, K., Morra, D., & Keelan, J. (2009). A survey of health-related activities on second life. Journal of Medical Internet Research, 11, e17. doi:10.2196/jmir.1192 Boulos, M. N., Maramba, I., & Wheeler, S. (2006). Wikis, blogs and podcasts: a new generation of Web-based tools for virtual collaborative clinical practice and education. BMC Medical Education, 6, 41. doi:10.1186/1472-6920-6-41 Catwell, L., & Sheikh, A. (2009). Evaluating eHealth interventions: the need for continuous systemic evaluation. PLoS Medicine, 6, e1000126. doi:10.1371/journal.pmed.1000126 Cook, D. A., Beckman, T. J., Thomas, K. G., & Thompson, W. G. (2008). Adapting web-based instruction to residents’ knowledge improves learning efficiency: a randomized controlled trial. Journal of General Internal Medicine, 23, 985–990. doi:10.1007/s11606-008-0541-0

Eysenbach, G. (2008). Medicine 2.0: social networking, collaboration, participation, apomediation, and openness. Journal of Medical Internet Research, 10, 22. doi:10.2196/jmir.1030 Frame, I., Austen, K. F., Calleja, M., Dove, M. T., White, T. O., & Wilson, D. J. (2009). New tools to support collaboration and virtual organizations. Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences, 367, 1051–1056. doi:10.1098/rsta.2008.0193 Gerard, J. M., Scalzo, A. J., Laffey, S. P., Sinks, G., Fendya, D., & Seratti, P. (2006). Evaluation of a novel Web-based pediatric advanced life support course. Archives of Pediatrics & Adolescent Medicine, 160, 649–655. doi:10.1001/ archpedi.160.6.649 Gorini, A., Gaggioli, A., Vigna, C., & Riva, G. (2008). A second life for eHealth: prospects for the use of 3-D virtual worlds in clinical psychology. Journal of Medical Internet Research, 10, e21. doi:10.2196/jmir.1029 Hall, W., De Roure, D., & Shadbolt, N. (2009). The evolution of the Web and implications for eResearch. Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences, 367, 991–1001. doi:10.1098/rsta.2008.0252

207

E-Medical Education

Hellström, L., Waern, K., Montelius, E., Astrand, B., Rydberg, T., & Petersson, G. (2009). Physicians’ attitudes towards ePrescribing-evaluation of a Swedish full-scale implementation. BMC Medical Informatics and Decision Making, 9, 37. doi:10.1186/1472-6947-9-37 Hughes, B., Joshi, I., & Wareham, J. (2008). Health 2.0 and Medicine 2.0: tensions and controversies in the field. Journal of Medical Internet Research, 10, e23. doi:10.2196/jmir.1056 Lemley, T., & Burnham, J. F. (2009). Web 2.0 tools in medical and nursing school curricula. Journal of the Medical Library Association, 97, 50–52. doi:10.3163/1536-5050.97.1.010 McLean, R., Richards, B. H., & Wardman, J. I. (2007). The effect of Web 2.0 on the future of medical practice and education: Darwikinian evolution or folksonomic revolution? The Medical Journal of Australia, 187, 174–177. Mohammed, S., Orabi, A., Fiaidhi, J., Orabi, M., & Benlamri, R. (2008). Developing a Web 2.0 telemedical education system: the AJAX-Cocoon portal. International Journal of Electronic Healthcare, 4, 24–42. doi:10.1504/IJEH.2008.018919 Nordqvist, C., Hanberger, L., Timpka, T., & Nordfeldt, S. (2009). Health professionals’ attitudes towards using a Web 2.0 portal for child and adolescent diabetes care: qualitative study. Journal of Medical Internet Research, 11, e12. doi:10.2196/jmir.1152 Norman, G. J., Zabinski, M. F., Adams, M. A., Rosenberg, D. E., Yaroch, A. L., & Atienza, A. A. (2007). A review of eHealth interventions for physical activity and dietary behavior change. American Journal of Preventive Medicine, 33, 336–345. doi:10.1016/j.amepre.2007.05.007

208

Rethlefsen, M. L., Piorun, M., & Prince, J. D. (2009). Teaching Web 2.0 technologies using Web 2.0 technologies. Journal of the Medical Library Association, 97, 253–259. doi:10.3163/15365050.97.4.008 Somal, K., Lam, W. C., & Tam, E. (2009). Computer and internet use by ophthalmologists and trainees in an academic centre. Canadian Journal of Ophthalmology, 44, 265–268. doi:10.3129/ i09-057 Timpka, T., Eriksson, H., Ludvigsson, J., Ekberg, J., Nordfeldt, S., & Hanberger, L. (2008). Web 2.0 systems supporting childhood chronic disease management: a pattern language representation of a general architecture. BMC Medical Informatics and Decision Making, 8, 54. doi:10.1186/14726947-8-54 van Straten, A., Cuijpers, P., & Smits, N. (2008). Effectiveness of a web-based self-help intervention for symptoms of depression, anxiety, and stress: randomized controlled trial. Journal of Medical Internet Research, 10, e7. doi:10.2196/ jmir.954 van Zutphen, M., Milder, I. E., & Bemelmans, W. J. (2009). Integrating an eHealth program for pregnant women in midwifery care: a feasibility study among midwives and program users. Journal of Medical Internet Research, 11, e7. doi:10.2196/jmir.988

KEY TERMS AND DEFINITIONS Blogs: Web sites intended for online audiences containing a diary-like account of events and personal insights. Information Philanthropy: A movement which has as its goal making quality intellectual property available for free, usually as an Internet download.

E-Medical Education

Problem-Based Learning (PBL): An educational modality where students work in small groups under the guidance of an instructor to find for themselves the knowledge they need to solve real world problems. Podcasts: A podcast is a downloadable audio or video presentation for replay on a personal media player such as Apple’s iPod. Web 2.0: A term used to describe the recent technical developments (e.g., Wikis, blogs, podcasts, mashups, and RSS feeds) that stress openness, sharing, cooperation and collaboration among Web users

Wikis: A collaborative informational Web site whose content is open (freely accessible) and can be edited by anyone who is duly authorized. Semantic Publishing: A technology where information and documents are published with explicit “semantic markups” intended to allow computers to comprehend the structure and meaning of published information. Semantic Web: A vision of the future where information is automatically exchanged and acted upon on our behalf by the Web itself by virtue of formatting to make it readily “understandable” by computers.

This work was previously published in Biomedical Engineering and Information Systems: Technologies, Tools and Applications, edited by Anupam Shukla and Ritu Tiwari, pp. 219-238, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

209

210

Chapter 1.14

Nursing Home Shuyan Xie The University of Alabama, USA Yang Xiao The University of Alabama, USA Hsiao-Hwa Chen National Cheng Kung University, Taiwan

ABSTRACT A nursing home is an entity that provides skilled nursing care and rehabilitation services to people with illnesses, injuries or functional disabilities, but most facilities serve the elderly. There are various services that nursing homes provide for different residents’ needs, including daily necessity care, mentally disabled, and drug rehabilitation. The levels of care and the care quality provided by nursing homes have increased significantly over the past decade. The trend nowadays is the continuous quality development towards to residents’ satisfaction; therefore healthcare technology plays a significant role in nursing home operations. This chapter points out the general information about current nursing home conditions and functioning DOI: 10.4018/978-1-60960-561-2.ch114

systems in the United States, which indicates the way that technology and e-health help improve the nursing home development based on the present needs and demanding trends. The authors’ also provide a visiting report about Thomasville Nursing Home with the depth of the consideration to how to catch the trends by implementing the technologies.

INTRODUCTION Nursing home is a significant part of long-term care in the health care system. Nursing homes provide a broad range of long-term care services – personal, social, and medical services designed to assist people who have functional or cognitive limitations in their ability to perform self-care and other activities necessary to live independently. This

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Nursing Home

survey release some information about nursing home, including general introduction of nursing homes in the United States, resident information, quality of life, services that are providing, governmental regulations, and developing trends. Objectives of this chapter are providing the broad information about current nursing homes in the United States to help understand the positive and negative situations that confront to nursing homes. Furthermore, our nursing home visiting report points out the real residents’ experience and opinions on their life qualities, needs and attitudes on technologies. Based on this information we introduce more technological innovations to improve nursing home development.

PART I: NURSING HOMES IN THE UNITED STATES General Information A nursing home, a facility for the care of individuals who do not require hospitalization and who cannot be cared for at home, is a type of care of residents. It is a place of residence of people who require constant nursing care and have significant deficiencies with activity of daily living (“Analysis of the National Nursing Home Survey (NNHS),” 2004). People enter nursing homes for a variety of reasons. Some may enter for a brief time when they leave the hospital because they need subacute care, such as skilled nursing care, medical services, and therapies (“Analysis of the National Nursing Home Survey (NNHS),” 2004). Others, however, need long-term care (LTC). LTC is generally defined as a broad range of personal, social, and medical services that assist people who have functional or cognitive limitations in their ability to perform self-care and other activities necessary to live independently (“Analysis of the National Nursing Home Survey (NNHS),” 2004).

In the United States, nursing homes are required to have a licensed nurse on duty 24 hours a day, and during at least one shift each day, one of those nurses must be a Register Nurse (“Analysis of the National Nursing Home Survey (NNHS),” 2004). A registered nurse (RN) is a health care professional responsible for implementing the practice of nursing in concert with other health care professionals. Registered nurses work as patient advocates for the care and recovery of the sick and maintenance of the health (“Nursing Facts: Today’s Registered Nurse - Numbers and Demographics,” 2006). In April, 2005 there were a total of 16,094 nursing homes in the United States. Some states having nursing homes that are called nursing facilities (NF), which do not have beds certified for Medicare patients, but can only treat patients whose payments sources is Private Payment, Private Insurance or Medicaid (“Medical & You handbook,” 2008). Medicare is a social insurance program administered by the United States government, providing health insurance coverage to people who are aged 65 and over, or who meet other special criteria (Castle, 2008). Medicaid is the United States health program for eligible individuals and families with low incomes and resources. It is a means-tested program that is jointly funded by the states and federal government, and is managed by the states. Among the groups of people served by Medicaid are eligible low-income parents, children, seniors, and people with disabilities. Being poor, or even very poor, does not necessarily qualify an individual for Medicaid (“Overview Medicaid Program: General Information,” 2006).

Services Baseline Services Those services included in the daily rate. The following basic services should be made available to all the residents:

211

Nursing Home

• • • • • •

•

•

•

•

•

212

Lodging-a clean, healthful, sheltered environment, proper outfitted; Dietary services; 24-hour-per-day nursing care; pharmacy services; diagnostic services; the use of all equipment, medical supplies and modalities used in the care of nursing home residents, including but not limited to catheters, hypodermic syringes and needles, irrigation outfits, dressings and pads, etc.(“Durable medical equipment: Scope and conditions,” 2006). general household medicine cabinet supplies, including but not limited nonprescription medications, materials for routine skin care, dental hygiene, care of hair, etc., except when specific items are medically indicated and prescribed for exceptional use for a specific resident (“Durable medical equipment: Scope and conditions,” 2006). assistance and/or supervision, when required, with activities of daily living, including but not limited to toileting, bathing, feeding and assistance with getting from place to place (“Home Health Service,” 2003). use of customarily stocked equipment, including but not limited to crutches, walkers, wheelchairs or other supportive equipments, including training in their use when necessary, unless such items are prescribed by a doctor for regular and sole use by a specific resident (“Durable medical equipment: Scope and conditions,” 2006). activities program, including but not limited to a planned schedule of recreational, motivational, social and other activities together with the necessary materials and supplies to make the resident’s life more meaningful (“Home Health Service,” 2003). social services as needed;

• •

provision of optician and optometrist services (““Home Health Service,” 2003,). physical therapy, occupational therapy, speech pathology services, audiology services and dental services, on either a staff or fee-for-services basis, as prescribed by a doctor, administered by or under the direct supervision of a licensed and currently registered physical therapist, occupational therapist, speech pathologist, qualified audiologist or registered dentist (“Home Health Service,” 2003).

Adult Day Health Care (ADHC) ADHC program provides the health care services and activities provided to a group of persons, who are not residents of a residential health care facility, but are functionally impaired and not home bounded (Castle, 2008). Require supervision, monitoring, preventive, diagnostic, therapeutic, rehabilitative or palliative care or services but not require continuous 24-hour-a-day inpatient care and services to maintain their health status and enable them to remain in the community (“Home Health Service, 2003,) (Castle, 2008).

Behavioral Intervention Services This program must include a discrete unit with a planned combination of services with staffing, equipment and physical facilities designed to serve individuals whose severe behavior cannot be managed in a less restrictive setting (Castle, 2008). The program shall provide goal-directed, comprehensive and interdisciplinary services directed at attaining or maintaining the individual at the highest practicable level of physical, affective, behavioral and cognitive functioning (Castle, 2008).

Clinical Laboratory Service Clinical laboratory means a facility for the microbiological, immunological, chemical, hema-

Nursing Home

tological, biophysical, cytological, pathological, genetic or other examination of materials derived from the human body, for the purpose of obtaining information for the diagnosis, prevention or treatment of disease, or the assessment of a health condition, or for identification purposes (“Home Health Service, 2003,) (Castle, 2008). Such examinations shall include procedures to determine, measure, or otherwise describe the presence or absence of various substances, components or organisms in the human body (“Home Health Service, 2003,) (Castle, 2008).

Coma Services A resident admitted for a coma management shall be a person who has suffered a traumatic brain injury, and is in a coma (Castle, 2008). This resident may be completely unresponsive to any stimuli or may exhibit a generalized response by reacting inconsistently and non-purposefully to stimuli in a nonspecific manner (Castle, 2008).

Government Regulations All nursing homes in the United States that receive Medicare and/or Medicaid funding are subject to federal regulations. People who inspect nursing homes are called surveyors or called as state surveyors (“NURSING HOMESHOMES: Federal Monitoring Surveys Demonstrate Continued Understatement of Serious Care Problems and CMS Oversight Weaknesses,” 2008). The Center for Medicare & Medicaid (CMS) is the component of the Federal Government’s Department of Health and Human Services that oversees the Medicare and Medicaid programs (“Overview Medicaid Program: General Information,” 2006). A large portion of Medicaid and Medicare dollars is used each year to cover nursing home care and services for elderly and disabled. State governments oversee the licensing of nursing homes (“Overview Medicaid Program: General Information,” 2006). In addition, State

have contract with CMS to monitor those nursing homes that want to be eligible to provide care to Medicare and Medicaid beneficiaries (“Overview Medicaid Program: General Information,” 2006). Congress established minimum requirements for nursing homes that want to provide services under Medicare and Medicaid (“NURSING HOMESHOMES: Federal Monitoring Surveys Demonstrate Continued Understatement of Serious Care Problems and CMS Oversight Weaknesses,” 2008). CMS also publishes a list of Special Focus Facilities- nursing homes with “a history of serious quality issues.” The US government Accountability Office (GAO), however, has found that state nursing home inspections understate the numbers of serious nursing home problems that present a danger to residents (“NURSING HOMES” 2008). CMS contracts with each state to conduct onsite inspections that determine whether its nursing homes meet the minimum Medicare and Medicaid quality and performance standards. Typically, the part of state government that takes care of this duty is the health department or department of human services (“NURSING HOMES” 2008) (Inspectors Often Overlook Serious Deficiencies at U.S. Nursing Homes, 2008). A report issued in September of 2008 found that over 90% of nursing homes were cited for federal health or safety violations in 2007, with about 17% of nursing homes having deficiencies causing “actual harm or immediate jeopardy” to patients (Pear, 2008). Nursing homes are subject to federal regulations and also strict state regulations (Inspectors Often Overlook Serious Deficiencies at U.S. Nursing Homes, 2008). The nursing home industry is considered one of the two most heavily regulated industries in the United States (the other being the nuclear power industry) (Wolf, 2003). As for the state and federal regulations that affect health care IT, CMS interpreted HIPAA’s security aspects to cover CIA--confidentiality, integrity, and availability. To date, most of the emphasis has been on confidentiality to reduce

213

Nursing Home

citizens’ fears that employers, insurers, or governmental agencies could use their personal health data against them (Sloane, 2009). The federal government left the implementation details surrounding HIPAA up to the states to oversee, however, because the states themselves manage the Medicare reimbursements, typically through third-party insurance companies. The result, unfortunately, really does look like a quilted patchwork of confusing and conflicting regulations (Sloane, 2009). Therefore, when information technology steps into the nursing homes, there are some concerns caused by the regulation complications.

Quality of Life Resident-Oriented Care Resident oriented care is designed as the place that nurses are assigned to particular patients and have the ability to develop relationships with individuals (Shields, 2005). Patients are treated more as family members. Using resident-oriented care, nurses are able to become familiar with each patient and cater more to their specific needs, both in emotional and medical aspects (Home Care and Nursing Home Services). According to various findings residents who receive resident-oriented are experience a higher quality of life, in respect to attention and time spent with patients and the number of fault reports (Boumans, 2005). Although resident-oriented nursing does not lengthen life, nursing home residents may dispel many feelings of loneliness and discontent (Boumans, 2005). “Resident assignment” refers to the extent to which residents are allocated to the same nurse. With this particular system one person is responsible for the entire admission period of the resident (Shields, 2005). However, this system can cause difficulties for the nurse or care-giver when one of the residents they are assigned to pass away or move to a different facility, since the nurse may become attached to the resident(s) they are car-

214

ing for (Shields, 2005;Home Care and Nursing Home Services). Therefore, three guidelines must be assessed: structure, process and outcome. Structure is the assessment of the instrumentalities of care and their organization (Pear, 2008); Process being the quality of the way in which care is given (Shields, 2005); Outcome is usually specified in terms of health, well-being, patient satisfaction, etc. (Shields, 2005); Using these three criteria, find that they are strengthened when residents experience resident oriented care (Shields, 2005; Home Care and Nursing Home Services) (Boumans, 2005). Communication is also heightened when residents feel comfortable discussing various issues with someone who is experienced with their particular case. In this particular situation nurses are also better able to do longitudinal follow up, and this insures the implementation of more lasting results (Shields, 2005; Boumans, 2005).

Task-Oriented Care Task oriented care is where nurses are assigned specific tasks to perform for numerous residents on a specific ward (Shields, 2005). Residents in this particular situation are exposed to multiple nurses at any given time. Because of the random disbursement of tasks, nurses are declined the ability to develop more in depth relations with any particular resident (Shields, 2005). Various findings suggest that task-oriented care produces less satisfied residents (Shields, 2005). In many cases, residents are disoriented and unsure of whom to disclose information to and as a result decide not to share information at all (Shields, 2005). Patients usually complain of loneliness and feelings of displacement. “Resident assignment” is allocated to numerous nurses as opposed to one person carrying the responsibility of one resident (Shields, 2005). Because the load on one nurse can become so great, various nurses are unable to identify with gradual emotional and physical changes experienced by

Nursing Home

one particular resident (CMS, 2008). Resident information has the ability to get misplaced or undocumented because of the numerous amounts of nurses that deal with one resident (Shields, 2005; CMS, 2008).

Options Current trends are to provide people with significant needs for long term supports and services with a variety of living arrangements (Home Care and Nursing Home Services). Indeed, research in the U.S as a result of the Real Choice Systems Change Grants, shows that many people are able to return to their own homes in the community. Private nursing agencies (Nursing Agency, also known as Nurses Agency or Nurses Registry) is a business that provides nurses and usually health care assistants (such as Certified Nursing Assistants) to people who need the services of healthcare professionals. Nurses are normally engaged by the agency on temporary contracts and make themselves available for hire by hospitals, care homes and other providers of care for help during busy periods or to cover for staff absences; “Assisted Living Facilities (ALF),” 2007) may be able to provide live-in nurses to stay and work with patients in their own homes. When considering living arrangements for those who are unable to live by themselves, potential customers consider it to be important to carefully look at many nursing homes and assisted living; Assisted living residences or assisted living facilities (ALFs) provide supervision or assistance with activities of daily living (ADLs), coordination of services by outside health care providers, and monitoring of resident activities to help to ensure their health, safety, and wellbeing (“Assisted Living Facilities (ALF),” 2007). Assistance may include the administration or supervision of medication, or personal care services provided by a trained staff person (“Assisted Living Facilities (ALF),” 2007), facilities as well as retirement homes, where a retirement home is a

multi-residence housing facility intended for the elderly (“Assisted Living Facilities (ALF),” 2007). The usual pattern is that each person or couple in the home has an apartment-style room or suite of rooms. Additional facilities are provided within the building. Often this includes facilities for meals, gathering, recreation, and some form of health or hospice care (Inspectors Often Overlook Serious Deficiencies at U.S. Nursing Homes, 2008), keeping in mind the person’s abilities to take care of themselves independently. While certainly not a residential option, many families choose to have their elderly loved one spend several hours per day at an adult daycare center, which is a non-residential facility specializing in providing activities for elderly and/or handicapped individuals (Boumans, 2005). Most centers operate 10 - 12 hours per day and provide meals, social/ recreational outings, and general supervision (Inspectors Often Overlook Serious Deficiencies at U.S. Nursing Homes, 2008). Beginning in 2002, Medicare began hosting an online comparison site intended to foster quality improving competition between nursing homes (Boumans, 2005).

Culture Change Nursing homes are leading to the way they are organized and direct to create a more residentcentered environment, making it more home-like and less hospital like (Home Care and Nursing Home Services). In these homes, nursing home units are replaced with a small set of rooms surrounding a common kitchen and living room. The design and decoration are more home style. One of the staff giving care is assigned to be the “household” (Home Care and Nursing Home Services). Residents have way more choices about when and what they want to eat, when they want to wake up, and what they want to do during the day. They also have access to more companionship such as pets, plants (Shields, 2005). Due to the residents’ diversity, the nurses and staff are learning more

215

Nursing Home

about different cultural difference, and meet more of the residents needs. Many of the facilities utilizing these models refer to such change as the “Culture Shift” or “Culture Change” occurring in the Long Term Care industry (Shields, 2005). Due to the nursing shortage problem, more and more nursing homes are heading the tendency to have more technology involve. Robot nurses are proposed to be part of the nursing home’s asset in the near future (“Robots may be next solution to nursing shortage,” 2003).

Resident Characteristics Nursing home residents are among the frailest Americans. In 2005, nearly half of all residents had dementia, and more than half were confined to a bed or wheelchair. In 2004, nearly 80 percent of residents needed help with 4 or 5 activities of daily living (bed mobility, transferring, dressing, eating, and toileting) (“Analysis of the National Nursing Home Survey (NNHS),” 2004). Most nursing home residents are female, especially at older ages, shown in Table 1 (Carrillo, 2006). Widowhood is a key predictor of nursing

Table 1. Nursing Home Residents by Age and Gender (Carrillo, 2006) Age Group

Men

Women

64 or Younger

175,000

54%

64%

65 to 84

643,000

34%

66%

85 or older

674,000

18%

82%

home use – at time of admission, over half of nursing home residents were widowed, and only 1 in 5 was married or living with a partner (Carrillo, 2006). The number of nursing home residents has remained approximately constant since 1985, but as a proportion of the population likely to need long-term care, it actually has declined (“The 65 and Over Population: 2000-Census 2000 Brief,”; “Nursing Home Data Compendium “, 2007). Over the past 20 years, the age 65 to 84 population increased by more than 20 percentage and the age 85+ population increased by more than 80 percentage, shown in Figure 1. As the percentage of older people living in nursing homes has declined, the number of stays has grown because of increasing use for short-term

Figure 1. AARP public policy institute analysis of 2004 NNHS

216

Total

Nursing Home

post-acute care (CMS, 2008). There were close to 3.2 million total nursing home stays in Medicare and Medicaid certified facilities during 2005, up from 3 million in 2000 (“Nursing Home Data Compendium “, 2007). Projecting future trends is difficult, since nursing home usage is driven by care preferences as well as life expectancy and disability trends (“Nursing Home Data Compendium “, 2007). Current estimates are that 35% of Americans age 65 in 2005 will receive some nursing home care in their lifetime, 18% will live in a nursing home for at least one year, and 5% for at least five years (“The 65 and Over Population: 2000-Census 2000 Brief,”; Kemper, Komisar, & Alecixh, 2005). Women, with longer life expectancy and higher rates of disability and widowhood, are more likely that men to need nursing home care, and especially likely to need lengthy stays (Kemper et al., 2005).

Concerns on Nursing Home Quality Nursing Home Abuse It seems that a lot of information that we have received are positive and promising, however, negative news would never too small to negligent. Here is a piece of news from New York Times: nursing home inspectors routinely overlook or minimize problems that pose a serious, immediate threat to patients, said by congressional investigators in a new report (Pear, 2008a). In the report, the investigators from the Government Accountability Office claim that they have found widespread ‘’understatement of deficiencies,’’ including malnutrition, severe bedsores, overuse of prescription medications, and abuse of nursing home residents (Pear, 2008a). Nursing homes are typically inspected once a year by state employees working under a contract with the federal government, which sets stringent standards (Pear, 2008b). Federal officials try to validate the work of state inspectors by accom-

panying them or doing follow-up surveys within a few weeks. It released that Alabama, Arizona, Missouri, New Mexico, Oklahoma, South Carolina, South Dakota, Tennessee and Wyoming were the nine states most likely to miss serious deficiencies (Pear, 2008a). More than 1.5 million people live in nursing homes. Nationwide, about one-fifth of the homes were cited for serious deficiencies last year (Pear, 2008a). ‘’Poor quality of care -- worsening pressure sores or untreated weight loss -- in a small but unacceptably high number of nursing homes continues to harm residents or place them in immediate jeopardy, that is, at risk of death or serious injury,’’ the report said (Pear, 2008b). There are several studies point out similar quality problems that occur in nursing homes. As to logical thoughts, nursing homes must meet federal standards as a condition of participating in Medicaid and Medicare, which cover more than two-thirds of their residents, at a cost of more than $75 billion a year (Pear, 2008b). Later section in this survey, there is an illustration on more details about the financial part and dilemmas that bother nursing homes.

Study of Residents Experience in Nursing Homes There has been increasing interest in the quality of care in nursing homes. Nursing homes should protect resident’s integrity and autonomy, and it should be the place that residents can thrive (Anderson, Issel, & McDaniel, 2003). The relationship between management practice (communication openness, decision making, relationship-oriented leadership and formalization) and resident outcome (aggressive behavior, fractures) should be paid attention on. Several studies (Anderson, 2003, 2003) show that lack of competent personnel and nursing home residents with more complex caring needs leads to insufficient care. There are correlations between satisfaction, commitment, stress

217

Nursing Home

and quality of care as perceived by staff. Reports from family members are often positive but have also led to implementation of emotion-oriented care in some homes. Despite this, while generally positive, the overall assessment of nursing homes by family members has not improved (Harris & Clauser, 2002; Redfern, Hannan, Norman & Martin,2002; Finnema, de Lange, Droes, Ribbe & Tilburg, 2001). The quality of care as assessed by competent nursing home residents has been studied by some authors (Polit & Beck, 2004). They found that the residents were mostly satisfied with the extent to which their wishes were met in regards to meals, shower routines, opportunities to listen to radio and TV, and their feeling of safety. The greatest difference between what residents wanted and what they experienced concerned the opportunities they had for close social relationships (Anderson, 2003). Understanding how residents in nursing homes regard living is important and has implications for nursing care and practice. In the study “Residents are Safe but Lonely”, it describe the experiences of a group of nursing home residents (Anderson, 2003). The two research questions were: How do the residents experience care? How can nurses help the residents to have a good life in the nursing home (Anderson, 2003)? The analyses were conducted by using a qualitative content analysis method [48]. The main finding was that the experience of being ‘safe but lonely’ characterizes residents’ experiences of living in a nursing home (Anderson, 2003). The residents at the nursing home emphasized that moving into the nursing home had made them feel safer than they felt when living in their private home (Anderson, 2003). Many residents told of feeling insecure and anxious when they lived at home. They were afraid that something adverse might happen to them and that they would not be able to get help when they needed it (Harris & Clauser, 2002). The informants stressed that having people in the vicinity at all times was one of the substantial benefits of living in a nursing

218

home. However, most of them have a feeling of a lack of trust in the foreign nurses (Anderson, 2003). These nurses had weak language skills which led to problems with communication and understanding both for the residents and the nurses and in turn led to feeling of insecurity for the residents (Anderson, 2003). Another thing was “haste”. The nurses were too busy to attend to the residents properly, so the informants did not feel that they were respected and regarded as unique individuals. This gave them a feeling of insecurity. All informants, however, stressed the importance of safety as the greatest benefit of living in a nursing home (Harris & Clauser, 2002). The informants gave their particular nursing home more credit than they appeared to think that nursing homes were generally given in the mass media (Harris & Clauser, 2002). They felt that they were cared for, but that at the same time the standard of their care would have been even better if there had been more nursing personnel. Feelings of loneliness and sadness are felt strongly by the residents. The days were long, boring, and empty, and informants felt uncared for (Harris & Clauser, 2002). Even though the activities are organized sometimes, it is far from what they have expected. The nurses do not see, nor do they meet the residents’ social needs. Some informants pointed out that a language barrier made communication with some of the nurses difficult (Harris & Clauser, 2002). According to those residents, the lack of personnel led to loneliness among the residents because the nurses had little time to chat with the residents (Harris & Clauser, 2002). Residents felt that this lack of respect was expressed as nonchalance (Harris & Clauser, 2002). It gave the residents a feeling of not being understood. Several informants stressed that they could not discuss their problems with the nurses because the nurses did not understand what they said and thus did not understand their needs. Other informants said that they used humor to solve language problems and that by the combined use of hands and words they could communicate what

Nursing Home

they wanted from the nurses (Harris & Clauser, 2002). As to the feeling a lack of reliability, it is reported that the nurses often did not follow up on agreements such as making training appointments and other agreements related to care (Redfern et al., 2002). The informants stressed that if a nurse had promised the resident something, he or she should attend to this immediately or at an agreed time. This was not always the case, and the residents felt frustrated at having to wait for ‘eternities’ for what they had been promised. This, the informants agreed on, was an aspect of their care that could be improved (Finnema et al., 2001). A central phenomenon in this study is the resident’s feelings of being met with respect. Several of the informants mentioned their feelings of being met with respect as a fundamental aspect of good nursing care (Finnema et al., 2001). To be met with respect means that the person is being met with appreciation for the unique human being he or she is. Residents in a nursing home have an unexpressed right to be taken care of in such a way that he or she feels comfortable and is regarded as a human being (Redfern et al., 2002). A resident is a human being with unique integral qualities and capabilities (Redfern et al., 2002). If residents are met in this way, hopefully they experience that more of their social needs are met so that they will not experience such a high degree of loneliness. The fact is that only competent residents were counted into this report, and 70-80% of the residents in nursing homes suffer from a dementia disease (Finnema et al., 2001). Residents suffering from dementia may have a different experience of living in a nursing home (Polit & Beck, 2004). Even if the informants mentioned positive factors in their experience of living in the nursing homes, they also mentioned deficiencies in the care they experienced (Polit & Beck, 2004). Nurses are therefore faced with the huge challenge of providing holistic care that addresses both the social as well as the physical needs of the residents

(Finnema et al., 2001). Therefore, some nursing homes start reform and try to design different daily activities for their residents.

Daily Activities There are more and more people receiving long term care in an institutional setting and that population is going to continue to grow with the aging of the “baby boomers.”So it is important to find out the what the residents ‘daily life, which is a way to spark more ideas to improve the nursing homes quality. It has been proven that when people move into nursing homes they lose a sense of control and feel a sense of isolation. Activities are one the best ways to fight these feelings and give a sense of empowerment back to the people. The variety of programs available in any nursing home depends on the health and interests of the residents. After doing research, basically, it comes up with a fairly wide range of activities in nursing homes that are designed to meet the needs of the residents. A broad range of programs is directed by the activities coordinator (Daily Activities, 2008). A home certified for Medicare and Medicaid must have someone designated as an activities coordinator (Daily Activities, 2008). The planning and implementation of activities comes from requests by residents, families, staff, and volunteers (Daily Activities, 2008). The activities are usually posted on a calendar of events that is available to each resident. Through the web reveal, here are some of the examples: •

•

Monthly birthday parties. Parties to which all residents are invited. Families and friends may be invited to participate (Fun Times, 2008). Celebrations of various holidays, both secular and religious. Holidays are particularly difficult times for those away from their own homes, families, and friends. Valentine’s Day, Halloween, Christmas,

219

Nursing Home

•

•

•

•

220

Hannukah, Easter, and Memorial Day are a few examples (Daily Activities, 2008; Fun Times, 2008; Activities, 2008). Musical events can be enjoyed actively or passively depending on the abilities of the residents. Many homes have sing-alongs in which the residents request their favorite songs and sing along with a leader. Again, the involvement of families, and friends is crucial to the success of such a program. Sometimes concerts are given by a church or school group or friend of the nursing home. Hopefully, the public is invited to attend, for this allows the residents to provide a source of pleasure to their community (Daily Activities, 2008; Fun Times, 2008). Games foster both one-to-one relationships and group activity. Bingo is a favorite for many, but bridge, chess, and other games for smaller groups usually are available. Volunteers and families often are the ones to stimulate resident interest in a game and they may be able to help arrange suitable opponents. Contests sometimes are run with work games, and tournaments are arranged for bridge or game players (Daily Activities, 2008; Fun Times, 2008; Activities, 2008). Outdoor activities include gardening, cookouts, or just enjoying time in the sun alone or with a friend. Often the staff does not have the time to take the immobile residents outside (Fun Times, 2008; Activities, 2008). Trips and tours to community events. Some homes have a special resident fund from the sale of arts and crafts made and sold by the residents to finance transportation rentals and ticket purchases. Friends or volunteers may donate to the fund or sometimes the nursing home sets aside money. Transportation can be a problem for those in wheelchairs, but the activities coordina-

•

•

•

•

tor usually can find volunteer drivers who are taught to cope with the special needs of disabled people. Some communities have special vans that transport residents in wheelchairs. Trips outside the home offer variety and mental stimulation (Activities, 2008). Nursing home newsletter, especially if published by residents. This is an especially valuable method of expression and uses resident talent that otherwise may lie idle. Poetry, history, birthdays, and resident and staff personality profiles are all topics that can be included (Fun Times, 2008; Activities, 2008). As we know, lots of elders are losing their ability to read as they are getting old, and therefore, it will be great if they can introduce the electronic reading device or arrange a “story time” when one of the caregivers read the news to the residents. Resident discussion groups. Sometimes a resident is an expert on a particular subject and will be the group leader. Other times a volunteer may offer to lead a discussion group. Topics may include current events, literature, and religion. The residents choose the topics and those interested attend. Exercise fun and physical fitness. Community leaders often volunteer to lead yoga or other exercise sessions. Even wheelchair-bound residents find satisfaction in exercising on a regular basis. Books. Volunteers may run a book service, taking a cart of books to the room of immobile residents. There may be a central library or small bookcases on each floor. Talking books for the blind may be part of the service. Families, friends, and volunteers can buy, bring, and hand out books. Many people help with reading to those unable to see well.

Nursing Home

•

•

•

Coffee or cocktail hours. Policies vary from home to home, but social hours provide a time of resident interaction. It is a particularly nice time for volunteers, family, and friends to join the residents (Daily Activities, 2008; Fun Times, 2008; Activities, 2008). Arts and crafts programs. They separate from occupational therapy, and are frequently offered by the activities’ coordinator. Religious services. Every Medicare- and Medicaid-certified nursing home must, by federal regulation, provide the opportunity for residents to attend religious services of their preference. Many nursing homes welcome denominational groups to provide religious services in the home for those who wish to attend. Again, this often provides an opportunity for families and friends to join the resident in worship. The organization of such services is usually handled by the activities’ coordinator (Activities, 2008).

Nursing Home Financial Regulations Billing Model Nursing home billing model is very like hospitals (Webb, 2001). Staff are housed in accessible nursing homes. Residents live in utilitarian, hospitallike rooms with little or no privacy and they sleep on hospital beds and are usually referred to as “patients” by the staff. Hospital pricing models are also used. (Webb, 2001) Residents are charged daily flat rates for semiprivate or private rooms just like a hospital. Extra services and supplies are added to the bill. This pricing model assumes that all residents require the same supervision and care. Of course this is not true (Webb, 2001). A lot could be done to improve the current system. For example, if family or friends were to help in the care of loved ones, these services

could be deducted from the bill (Webb, 2001). A number of residents are also capable of helping with the care of fellow residents or they might help with the facility services such as cleaning, food preparation, social needs, and laundry and so on. These cost savings could be passed on to all residents (Webb, 2001). State and Federal governments pay about 70% of nursing home costs and for about 85% of all residents the government pays part of or all of their costs (More Can be Done, 2002). Because the government pays such a large portion, nursing homes structure their care delivery system around the government payment system (More Can be Done, 2002). Government reimbursement is based on nursing hours and aide hours per patient, plant costs, wages, utilities, insurance, ancillary services, etc. and basically follows a hospital model (Webb, 2001). Because government programs typically are burdened with massive stationery inertia, the current pricing model will be around for a long time (Webb, 2001; More Can be Done, 2002).

Payment Sources Medicare Medicare is the government health insurance plan for all eligible individuals age 65 and older. Because of its universal availability almost everyone over age 65 in this country is covered by Medicare. There are about 40 million Medicare beneficiaries nationwide (ELJAY, 2006). Medicare will pay for 20 days of a skilled nursing care facility at full cost and the difference between $114 per day and the actual cost for another 80 days (ELJAY, 2006). Private Medicare supplement insurance usually pays the 80 day deductible of $114 per day, if a person carries this insurance and the right policy form. However, Medicare often stops paying before reaching the full 100 days. When Medicare stops, so does the supplement coverage (ELJAY, 2006).

221

Nursing Home

To qualify for Medicare nursing home coverage, the individual must spend at least 3 full days in a hospital and must have a skilled nursing need and have a doctor order it (ELJAY, 2006). The transfer from a hospital must occur within a certain time period (ELJAY, 2006). Medicaid Medicaid is a welfare program jointly funded by the federal government and the states and largely administered by the states (Caring for the Elderly, 1998). In 1998, Medicaid paid for 46.3% of the $88 billion received by all US nursing homes (Caring for the Elderly, 1998). Although the bias is only to cover eligible patients in semiprivate rooms in nursing homes, a recent court decision is forcing the States to consider more Medicaid funding for home care and assisted living (Caring for the Elderly, 1998). To receive a Medicaid waiver for alternative community services, the patient must first be evaluated for 90 days in a nursing home. There is a certain requirement for Medicaid to cover residents’ nursing home care. In order to qualify for nursing home Medicaid, an applicant must meet the “medical necessity” requirement, which generally requires a medical disorder or disease requiring attention by registered or licensed vocational nurses on a regular basis (“National Health Accounts,”). The Director of Nursing in the nursing home or an applicant’s medical doctor should be able to assess whether someone meets the medical necessity requirement (“National Health Accounts,”). Once it is determined that there is a medical necessity for nursing home care, the applicant must meet two financial tests: the income test and the resources test. The general rule is that in order to qualify for Medicaid nursing home care, an unmarried applicant cannot have more than $1,737.00 in monthly income (for the 2005 calendar year) and no more than $2,000.00 in resources or assets. If both spouses apply, the incomes are combined and the income cap is twice the cap for the individual. The resource limita-

222

tions for a married couple, both of whom apply, is $3,000.00. However, the limitations are not as stringent when the applicant is married and his or her spouse is not institutionalized (“National Health Accounts,”). Insurance Insurance is an alternative source of funding for long term nursing home care. From virtually nothing 10 years ago, insurance paid 5% of nursing home receipts in 1998 (Sloan & Shayne, 1993). This percentage is increasing every year. The government is also sending a clear message it wants private insurance to play a larger role. This began with the recommendation of the Pepper Commission in 1992 and continued with the HIPAA legislation in 1996 and on to the offering in 2003 of long-term care insurance for federal workers, military, retirees and their families (“National Health Accounts,”; Sloan & Shayne, 1993). There are several bills now pending in Congress allowing full deduction of premiums and the pass-through of premiums in cafeteria plans. These tax breaks are meant as an incentive for the purchase of insurance. Medicare covers about 12% of private nursing home costs while Medicaid covers about 50%. The Veteran’s Administration nursing home operations bring total government support of nursing home costs to about 70% of the total. Such a large reliance on government support has made nursing homes vulnerable to vagaries in state and Federal reimbursement policies towards nursing homes (Sloan & Shayne, 1993).

Current Problems For Nursing Homes Funding Shortfalls The majority of nursing home income comes from government reimbursement (Polit & Beck, 2004). The industry claims that many of its nursing homes are losing money on government payments caus-

Nursing Home

ing yearly net business losses. For those homes that make profit, there’s not margin enough for improving infrastructure or hiring more or better qualified staff to improve quality of care (Polit & Beck, 2004). Quality of care eventually suffers with inadequate income. On the other hand, critics contend the current revelations of poor care with many nursing homes across the country stem not from lack of income but from greedy owners not willing to apply profits to improvement in care. The national chains, in particular, are accused of retaining profits to bolster stock prices in an effort to fund acquisitions (Staff Turnover and Retention, 2007; Peter, 2001).

Shortage of Nurses

Staffing

Elder Abuse and Lawsuits

A recent report from the Government Accounting Office cites widespread understaffing by nursing homes both in levels of nurses and certified nurse’s assistants (Mukamel & Spector, 2000). Staffing in these cases is below government-recommended adequate standards (Mukamel & Spector, 2000). Recently states such as California have mandated higher staffing ratios for hospitals and skilled nursing homes. But in most cases there is not additional money to cover the cost of more staff. So mandated staffing ratios will probably have little effect on the problems facing nursing homes and may actually increase their problems (Mukamel & Spector, 2000).

It shouldn’t come as a surprise with the problems of funding and staffing that reported incidents of patient neglect and abuse are on the rise (Williams, 2003). Of particular concern is the more frequent occurrence of abuse. Abuse is not only just physical assault or threats but it can also be such things as improper use of restraints, failure to feed or give water, failure to bathe, improper care resulting in pressure sores or allowing a patient to lie too long in a soiled diaper or bed linen. Lawsuits are increasing in number. (Williams, 2003; Webb, 2001) There is a great concern at the Federal and state levels to control abuse (Webb, 2001). So far, aside from proposing tougher laws to penalize the industry, there appears to be little effort in finding a way to improve the nursing home system of care delivery (Webb, 2001).

Turnover of Aides There’s no question that tight labor markets over the past decade have made it difficult to recruit and retain workers. But turnover of qualified aides is so high. Nursing homes claim they can’t afford to pay for the higher level of wages and benefits necessary to retain aides who will stay around for a while (Staff Turnover and Retention, 2007).

Next is the problem of nurses. There is currently a nationwide shortage of nurses. Nursing homes are willing to pay the salary to attract nurses but in many areas there aren’t enough nurses to meet demand (Peter, 2001). Nursing homes, as well as hospitals are using innovative work schedules to meet staffing requirements but in many cases, nurses are overloaded with too many patients. In other cases, less qualified workers are substituting for nurses (Peter, 2001). These shortages and high turnover affect the quality of care that a nursing home can provide (Peter, 2001).

PART II: E-HEALTH AND NURSING HOMES E-health is a type of telemedicine that encompasses any “e-thing” in health field; that is, any database or encyclopedia containing information pertinent

223

Nursing Home

to health, medicine, or pharmacy that is accessible via internet (Breen & Zhang, 2008). It is a typical form commonly referenced and understood as that consists of internet-based services such as WebMD.com, Medlineplus.gov, and Mentalhelp.net. Telemedicine involves the use of telecommunications technologies to facilitate clinical healthcare delivery and exchange at a distance (Thompson, Dorsey, Miller, & Parrott, 2003). This delivery method was initially designed to specifically benefit patients (Breen, & Matusitz, 2006; Wootton, 2001). E-health is a subunit of the “telemedicine umbrella” (Breen & Zhang, 2008). Because of the distance barrier that telemedicine and e-health services are able to transcend, Matusitz and Breen (Breen, & Matusitz, 2006) were convinced, through their published research, that telemedicine would be particularly beneficial in secluded towns and communities deficient in adequate healthcare services, such as bucolic regions, in mountains, on islands, and in locations of arctic climate (Breen & Zhang, 2008). Given the efficacy of telemedicine services noted by these scholars in their work in both metropolitan and more rural or remote regions, it makes logical sense that such e-health services would prove valuable when actively used by nursing staff in a variety of nursing homes, regardless of the structure and organization of the nursing home (e.g., nursing home chains in urban and rural areas, for-profit or non-profit homes, etc.) (Matusitz & Breen, 2007). E-health provides a faster and easier alternative, as well as a more cost-effective means, to accessing healthcare information, as opposed to more traditional and costly physical interactions between doctors and patients. Besides the fact that the use of e-health services is rapidly rising with, according to Matusitz and Breen (Breen, & Matusitz, 2006) 88.5 million American adults on average seeking on-line health information daily, studies have also been published that identify the quantity of and frequency of use among specific medical practitioners in a variety of specialties and settings: psychiatry, pathology, dermatol-

224

ogy, cardiology, in intensive care units, and in emergency rooms where surgery is performed (Matusitz & Breen, 2007; Thompson, Dorsey, Miller, & Parrott, 2003). More relevantly, many of these relate to types of care needed and applied in nursing home settings. WebMD.com is preferred by most medical practitioners (Matusitz, 2007). One of the most important financial advantages found within the WebMD.com site (Our Products and Services) for medical practitioners in particular is that “its products and services streamline administrative and clinical processes, promote efficiency and reduce costs by facilitating information exchange, and enhance communication and electronic transactions between healthcare participants” (Our Products and Services). Further, WebMD.com (Our Products and Services) offers, free-of-charge, disease and treatment information, pharmacy and drug information, and an array of forums and interactive e-communities where practitioners and consumers can correspond using specialized e-messages. WebMD.com (Our Products and Services) doesn’t stop there in its free provisions and benefits to medical practitioners and its potential assets to nursing home settings in particular (Breen & Zhang, 2008). E-health should be readily accepted and generally understood by the professional medical population, especially in nursing home settings across metropolitan, suburban, and rural sectors. Mitka (2003) also reported that telemedicine services, particularly e-health applications, improved delivery of psychiatric care to residents of rural nursing homes (Breen & Zhang, 2008).

Health IT Acceptance in Long-Term Care Nursing homes are often depicted as laggards when it comes to embracing technology tools. But an analysis by the American Association of Homes and Services for the Aging shows that they are more than holding their own.

Nursing Home

Nearly all nursing homes at the time had electronic information systems for MDS data collection and billing, and 43% maintained electronic health record systems, according to analysts, who studied data from the 2004 National Nursing Home Survey. By comparison, 25% of doctor offices and 59% of hospitals handled EHRs (O’Connor, 2008).

Staff Survey There are various factors determining the acceptance of health IT applications by caregivers in long-term care facilities, including social influence factors such as subjective norm and image understandable level and demographic variables including age, job level, long-term care work experience and computer skills in regard to their impact on caregivers’ acceptance of health IT applications. A nationwide survey, being applied a modified version of the extended technology acceptance model (TAM) to examine the health IT acceptance in long-term care, reveals a positive results: Although it is suggested that the effect of the Technology Acceptance Model on professionals and general users (King & He, 2006) and on people from different cultures (Schepers & Wetzels, 2007) is different, the survey results clearly suggest that the caregivers’ acceptance of IT innovations in long-term care environment at the pre-implementation stage. In fact, the caregivers show high levels of acceptance of health IT applications. In order to ensure the successful introduction of a new health IT application into a long-term care facility, a positive environment should be established with support from facility managers and RNs for the introduction of the new innovation. As computer skills directly impact on the caregivers’ perceived ease of use and intention to use an application, effort should be made to understand the caregivers’ computer skills, and provide adequate training and support to improve their skills if necessary. “Our findings also suggest that improving the caregivers’ understanding of the

importance of the introduced system for nursing care will improve their acceptance of the new innovation” (Yu, Li, & Gagnon, 2009). In this survey, it appears that 66.4% of the participants would be able to adapt to a health IT application with reasonable training and support 34% of caregivers’ intention to use an introduced IT application before any hands-on experience with the system established (Yu et al., 2009).

Health Information Technology in Nursing Homes Information technology (IT) has significant potential to reduce error and improve the quality and efficiency of health care (Bates et al., 2001; Institute of Medicine [IOM], 2001). Some researchers also believe that computer systems can be used to reduce error and improve the reporting of adverse incidents in health care settings (Wald & Shojania, 2001). Since October 1998, all state licensed nursing homes have been required to electronically transmit data generated by the federally mandated Resident Assessment Instrument (RAI) via state public health agencies to the Centers for Medicare and Medicaid Services (CMS) (Mor et al., 2003). Advanced features in the software were available to most (87% to 98%) of the facilities; however, most features were not being used all the time. There are huge potentials addressing the technology applications with various aspects such as enhancing the administrator’s ability to manage the facility, tracking quality, and monitoring multiple performance indicators. HIT is used to collect, store, retrieve, and transfer clinical, administrative, and financial health information electronically (General Accounting Office, 2004). The IOM publication To Err is Human (Kohn, Corrigan, & Donaldson, 1999) drew attention to the potential for HIT to improve quality of care and prevent medical errors across the health care system. The focus of HIT development and implementation has been

225

Nursing Home

mainly on acute and ambulatory care settings (American Health Information Management Association [AHIMA], 2005; Hamilton, 2006; Kramer et al., 2004). Commonly found HIT systems provide computerized physician order entry (CPOE), electronic health records (EHR), and point of care (POC) to access data for entry and retrieval (e.g., reviewing records, entering orders at the bedside or in the examination room; Hamilton, 2006). In addition, electronic systems for management operations such as patient scheduling and reimbursement have been available for longer and are used frequently (AHIMA, 2005; Baron, Fabens, Schiffman, & Wolf, 2005; Poon et al., 2006). An EHR is an electronic formatting of medical records documenting the patient’s medical history, written orders, and progress notes. CPOE is a system making information available for physicians at the time an order is entered (Bates et al., 1998; Bates et al., 1999). POC is a technology automating the care provider’s procedure, visit notes, and educational materials at the place of care (Anderson & Wittwer, 2004). Although the implementation of HIT in the long-term care sector is recognized to be lagging behind the acute and ambulatory settings (AHIMA, 2005; Hamilton, 2006), there is a wide range potentials for nursing homes such as functions covering domains such as financial management, administration, ancillary care, support, and resident care (McKnight’s LTC News, 2005).

Technology Streamlines LongTerm Care Operations Technology’s role in nursing homes has been greater importance as these types of facilities serve more older adults. Most of them are expanding their focus on the best ways to deploy technology to improve residents’ quality of care. In this section, we explored some equipments/devices that favor nursing homes, skilled nursing facilities (SNFs) and assisted living facilities.

226

Fall Management This care issue includes technologies aimed at decreasing the occurrence of resident falls, as well as technologies that alert caregivers and reduce injury when falls occur (Burland, 2008). Products include grab bars, non slips mats and socks, and differnt types of alarms. What fall management options are currently available? According to Gillespie SM, Friedman SM (2007), the major two categories are as following: •

Technologies aimed at reducing the risk of a fall: ◦⊦ Anti-slip footwears are socks and slippers with anti-slip material incorporated on the bottom. ◦⊦ Anti-slip matting and materials provide a slip resistant surface to stand on in slippery areas such as tubs and bathroom floors. ◦⊦ Grab bars provide stability and support in bathrooms and other areas. ◦⊦ Wheelchair anti-rollback devices prevent a wheelchair from rolling away when residents stand or lower themselves into a chair. ◦⊦ Chair, bed, and toilet alarms signal a caregiver when a resident who is at risk for falling attempts to leave a bed, chair, wheelchair, or toilet unattended. ◦⊦ Rehabilitation equipment and programs geared toward the restoration and maintenance of strength, endurance, range of motion, bone density, balance, and gait. Some examples include unweighting systems that enable residents to perform gait training in a supported reduced weight environment, systems that can test balance, and treadmills.

Nursing Home

•

Technologies aimed at reducing the risk of injury when falls occur: ◦⊦ Hip protectors are designed to protect the hip from injury in the event of a fall. ◦⊦ Bedside cushions may help reduce the impact of a fall if a resident rolls out of bed. ◦⊦ Technologies that notify caregivers when a resident has fallen: Fall detection devices use technologies that sense a change in body position, body altitude, and the force of impact to determine when a fall has occurred.

Medication Management

•

•

Products geared toward enabling residents to manage and adhere to their medication regime with greater independence (Field, 2008). •

•

•

•

Medication Applicators enable the user to apply lotions and ointments on hard to reach areas such as the back and feet. They typically consist of a sponge or pad that is attached to a long handle (Lilley RC 2006). Medication Reminder Software is installed on Personal Digital Assistants (PDA’s), personal computers, or mobile phones to provide reminders to take or administer medication at predetermined time. Some of these applications have the ability to manage complex medication regimens and can store medication and medical histories(Wolfstadt, 2008). Pill Organizers keep dry medications and vitamins arranged in compartments to assist with medication compliance and protocol adherence. Compartments are labeled for weekly or daily dosage frequencies and may be marked in Braille for individuals with visual impairments (McGraw, 2004). Multi-Alarm Medication Reminders and Watches are programmed to remind the

•

•

user to take medications at predetermined time schedules. They typically come in the form of a specialized watch, pager or pocket device, or medication bottle cap (Lilley RC 2006). Multi-Alarm Pill Boxes serve two purposes; to store medication and provide reminder alerts to take medications at prescribed times. Most alerts come in the form of an audible tone at specific times of the day or predetermined hourly intervals (Lilley RC 2006). These pill boxes also offer compartments to help organize medications by day of the week and time of day. Personal Automatic Medication Dispensers are programmable, locked devices that will automatically dispense a dose of dry medications at predetermined times. These devices also act as multi-alarm medication reminders that alert the resident when it is time to take their medication with audible alarms, lights, text and voice messages(Lilley RC 2006). Talking Medication Bottles contain recording mechanisms that enable a caregiver or pharmacist to record a message that can be played back anytime by the user. The recorded message verbally identifies bottle contents and provides reminders concerning the medication protocol (Lilley RC 2006). Automated Medication Dispensing Cabinets provide computer controlled storage, dispensation, tracking, and documentation of medication distribution on the resident care unit. Monitoring of these cabinets is accomplished by interfacing them with the facility’s central pharmacy computer. These cabinets can also be interfaced with other external databases such as resident profiles, the facility’s admission/ discharge/transfer system, and billing systems (MacLaughlin, 2005).

227

Nursing Home

•

•

Automated Medication Dispensing Carts offer secure medication dispensation and tracking at the bedside or wherever a resident may be located during their medication time. These carts feature a wireless computer terminal with tracking software to audit access to the system and control electronically locked medication drawers (Lilley RC 2006). Barcode Medication Administration utilizes barcode scanning technology to match the resident with his/her medication order at the point of administration. If any variables of the “5 Rights of Medication Administration” do not match correctly, the administering nurse will be notified via a combination of warning tones and text messages. In many cases, this technology is used to complement automated medication distribution systems (Greenly, 2002).

Assistance Call This care issue, also known as nurse call or emergency call, includes technologies that enable a resident to summon assistance by means of a transmitter that they can carry with them or access from somewhere in their living area (Alexander, 2008). •

•

228

Wired Systems: These systems rely on wires for communication between the main components of the system. Some wired systems allow for the addition of wireless features, such as wireless call stations, wireless phones, pagers, and locator systems. However these systems are not fully wireless and are categorized here as wired systems (Michael, 2000). Wireless Systems: These systems require no wiring for installation. All components of the system communicate wirelessly through radio waves (Douglas Holmes, 2007).

•

Telephone Based Systems: These systems use telephone lines to alert caregivers to a resident need. When a resident is in need of assistance, they press a wireless transmitter that they wear (typically a pendant or wristband) or a wall mounted transmitter that sends a signal to dialing device. The dialing device automatically sends a signal to a central CPU that alerts staff to the resident need.

Bathing This care issue includes products that focus on bath safety, enable residents to be more independent with bathing tasks, and that make bathing tasks easier for caregivers (Barrick, 2008). According to Gill and Han (2006), products that enable residents to access showers and tubs more independently and safely: •

•

•

•

•

Barrier Free and ADA Compliant Showers offer accessible features that enable residents to enter and exit the shower, sit, and access controls with greater ease and safety. Tub and Shower Chairs can be placed inside of a tub or shower and provide a seating surface while showering. Transfer Benches are placed in a tub and provide a seating surface that extends over the side of the tub, thereby easing transfers, eliminating the need to step over the side of the tub, and providing a place to sit while showering. In-Tub Bath Lifts are powered devices that are placed in a tub, creating an adjustable height surface for transfers. Mobile Commode/Shower Chairs serve several roles. They can be pushed in to barrier free and roll-in showers, thereby providing a means of transporting the resident to the showering area (self propelled or pushed by a caregiver). They provide seat-

Nursing Home

• •

ing during showers, and act as a commode or raised seat for toileting. Grab Bars provide stability and support in bathrooms and other areas. Anti-slip Matting and materials provide a slip resistant surface to stand on in slippery areas such as tubs and bathroom floors.

Products that enable residents to perform bathing tasks more independently: •

•

•

Wash Mitts offer a washing solution for those with decreased fine motor skills and an inability to handle a washcloth. Long Handled Brushes and Sponges enable residents with limited range to wash hard to reach areas such as their back, feet, and head. Rinse-Free Bathing Products enable residents to wash their body and hair without the need for running water or transferring in to a tub or shower (Gill TM, 2006).

According to Moller, Julie etc. (2003), products that assist caregivers with the transfer of residents to showers and tubs: •

•

•

•

Height Adjustable Bathtubs can be raised and lowered to assist with transfers and place the tub at a comfortable working height for caregivers. Side-Opening Bathtubs provide a side opening that facilitates transfers in and out of the tub. Bath Lifts fully support the resident during bathing as well as transfers in and out of a tub. Depending on the resident’s trunk support and sitting balance, bath lifts are offered that enable the resident to be transferred and bathed in a seated or recumbent (reclined) position. Shower Trolleys and Gurneys enable caregivers to transport and shower fully depen-

•

•

dent residents that need full body support at a comfortable working height. Showering Cabinets allow caregivers to shower residents while standing outside of the cabinet and provide the resident with some privacy. Showering cabinets open in the front to provide access to mobile commode/shower chairs or assist with transfers on to a sliding seat. Mobile Commode/Shower Chairs serve several roles. They can be pushed in to barrier free and roll-in showers, thereby providing a means of transporting the resident to the showering area (self propelled or pushed by a caregiver). They also provide seating during showers and act as a commode or raised seat for toileting (Moller, 2003).

According to Wendt (2007), products that assist caregivers with bathing tasks: •

•

•

Shampooing Basins and Rinse Trays allow caregivers to wash residents’ hair in bed, a chair, or a wheelchair. Rinse-Free Bathing Products enable caregivers to wash and shampoo residents without the need for running water or transferring residents into to a tub or shower. Height Adjustable Bathtubs can be raised and lowered to assist with transfers and place the tub at a comfortable working height for caregivers.

Detailed Examples A Study of Remote Monitoring Test Understanding senior’s attitudes toward technology and their willingness to adopt technological solutions that help them remain independent longer, which presents a significant challenge to the aging industry. (Reilly & Bischoff, 2008; “Intelligent remote visual monitoring system for

229

Nursing Home

home health care service,” 1996) until recently, limited information regarding the practical impact if remote monitoring systems on the elderly has been available. Fortunately, as the technology development, a recent study has yield new data about this remote monitoring test (“Intelligent remote visual monitoring system for home health care service,” 1996). The April 2008 study consulting firm at four Philadelphia-based NewCourtland Eleder Services communities measured the effectiveness of senior technology and captured the perceptions of residents, family member, and staff employing it (Reilly & Bischoff, 2008). Results of the study indicate that users had a very positive attitude toward the remote monitoring technology. The two greatest advantages of the system, according to the residents’ responses, were the assistance it provided to get help quickly in the event of an emergency, such as a fall or sudden illness, and the added benefit of enabling them to live independently for a longer period of time (Reilly & Bischoff, 2008). They conducted a survey for all the residents, and 100 percent of the survey respondents either “strongly agreed” or “agreed” with the two statements. Among those surveyed, only one resident reported a concern about intrusiveness (Reilly & Bischoff, 2008). The staff study indicated similar sentiments about the system’s ability to provide better care to their residents (“Intelligent remote visual monitoring system for home health care service,” 1996). “We thought at first that adapting to the technology would be a major issue for our residents, but clearly it was not,” says Kim Brooks, NewCourtland’s vice president of housing and community-based services. “The results of the survey demonstrate that even seniors with little or no prior exposure to this technology can readily adapt to it (Reilly & Bischoff, 2008).” A remote monitoring system, typically consisting of small wireless electronic sensors, monitors daily living activities (“Intelligent remote visual monitoring system for home health care service,”). The sensors are placed in stratefic areas of the

230

residents, including walls, to detect movements within the rooms; on the kitchen cupboards and refrigerator doors, to monitor whether the resident is eating regularly; and tilt sensors on the medicine boxes to monitor medication usage. The sensor in the bed can detect when a resident gets in or out of bed, toilet sensors monitor toilet usage, and homeor away sensors tell when the resident leaves or return to her residence (Reilly & Bischoff, 2008; “Intelligent remote visual monitoring system for home health care service,” 1996). A call pendant can be used as an emergency call button, which may be worn or carried by the resident. Besides, if the sensor detects abnormal activity, the system calls will help and automatically alters a caregiver via phone. A cancel button clears in-home alerts or emergency calls (Reilly & Bischoff, 2008). There is a central computing component in the nursing home—that receives all information transmitted by the sensors (Reilly & Bischoff, 2008). Based on information that is received, the base station will determine if there is a need to call for help, for example. In such a case, the base station will first sound an “in-home alert.” If the resident is okay, a cancel button can be pressed to discontinue the alert. If it is an emergency and the in-home alert is not cancelled, the system will automatically proceed through a user-defined call list, via the telephone line, until a responder accepts responsibility to check on the user. If no contacts in the personal caregiver network can be reached, an automated call can be placed to facility security services (Reilly & Bischoff, 2008). Forty-three of 54 residents were interviewed, and the response of the seniors who participated indicated that the peace of mind they obtain from knowing that they will receive help by using the new technology (Reilly & Bischoff, 2008). Only one of the residents commented as a doubt attitude to the remote monitoring method at the beginning, since they ran the risk by leaving the nurse/caregiver always “invisible”. After realizing the success on for awhile, she became happy and

Nursing Home

satisfied with the new system (Reilly & Bischoff, 2008). Mitzi Boegly, a 97-year-old resident who was interviewed for the study, said that she welcomed having the sensor technology in her apartment— especially after she used it to summon help after a fall. “It gives you a feeling of security knowing that if something happened to you, you can get help right away. I go to bed at night with peace of mind,” she said. The NewCourtland staff responses indicate that they also feel the sensor technology has improved the residents’ basic security and safety (Reilly & Bischoff, 2008). “The study shows our staffs appreciate the system’s ability to improve the efficacy of care delivery by directing it quickly to where it is most needed,” says Brooks. At the same time, she adds, the staff’s responses suggest a need for additional training, better reporting tools, and extensions into prognostics for chronic conditions (Reilly & Bischoff, 2008). From such a successful testing, we feel that it is promising to adapt this technology system in nursing homes. As to earlier mentioned that the shortage of nurse in the Unite State, while the continuous increasing of demands for caregivers, the installment of remote monitoring system definitely will be helpful for regulate such a lack. One thing, we need concern is about the cultural and situational difference in different areas. Research indicates that in most of northern United States that people are generally more open minded, and easier to accept new methods (“Intelligent remote visual monitoring system for home health care service,” 1996). So that in the later sections will release the indication on the attitudes of nursing home residents here in Alabama, by conducting the personal interviews to different nursing homes.

Carpet Sensor in Bedroom Due to the frequent accident happening on the seniors during the night time, a new technology has been applied to the carpet to make sure their

safety in year 1997, making it as big as a rug. Result: floors that can sound an alert when a nursing-home patient falls or when a nighttime intruder enters (Otis, 2007). It proves that such carpet has been commonly used in health care organizations, especially in nursing homes. Developed by Messet Oy, a five-person company in Kuopio, Finland, with the Technical Research Center of Finland’s VTT Automation Institute in Tampere, the room-size sensors are being tested in nursing homes in Helsinki and Tampere, says Messet Chairman Keijo Korhonen (Otis, 2007). The sensor is a thin polypropylene lamination that goes under carpeting or floor tiles (Otis, 2007). Inside the 0.002-inch-thick structure are tiny pillows of foamed plastic (Otis, 2007). These function as “electrets,” a type of electromagnet used in some microphones. When a weak current is flowing through the top surface, the pillows respond to the slightest changes in pressure by generating an electrical signal (Otis, 2007). Messet says that the structure is so sensitive that it can detect the breathing of a person lying on the floor — through the carpet. This talking carpet is welcomed by the patients as well as the healthcare providers.

Scheduling Software Using specialized software to schedule the nursing staff at St. Joseph Convent retirement and nursing home is easy and affordable. It’s also a time-saver. Marilyn Fuller says she couldn’t do without it (“St. Joseph”, 2007). Fuller, a scheduler at St. Joseph Convent, a home for 170 retired nuns, uses the Visual Staff Scheduler Pro by Atlas Business Solutions to ensure shift coverage, track and reduce overtime, view staff availability and contact information, and define custom shifts. “The software has helped eliminate mistakes and questions in the schedule and provides an accurate report on cost effectiveness by showing me a complete picture of how many hours each staff member is scheduled each week,” Fuller said (“St. Joseph”, 2007).

231

Nursing Home

The scheduler software eliminates the opportunity for staff to make changes to the work schedule. Only Fuller and the director of nursing can edit the schedule, but each unit can access a copy (“St. Joseph”, 2007). The Visual Staff Scheduler Pro software by Atlas Business Solutions is a quick and flexible tool that makes staff scheduling easy and very affordable (“St. Joseph”, 2007).

Seating System Since 1994, recliner and wheelchair design, developing cushioning and posturing appliances to make the patient comfortable, has been introduced to health care industry (David, 1994). Chairs were designed to fit the largest size potential users and modified on-site with pillows, cushions, foam pads, gel seats and, most recently, active pneumatic technology (David, 1994). This has its obvious limits, especially to whose are small. Such a high-tech seating system encountered a roadblock in nursing homes: Their funding mechanisms haven’t advanced to keep up with developments in equipment (David, 1994). These new seating systems have been developed for use in more intensive rehabilitation environments and haven’t reached long-term care yet, largely because of high cost and reimbursement difficulties (David, 1994). A higher level of seating care becomes available and is widely used nowadays, because there is a change in how therapy services are being provided to nursing homes (David, 1994). As nursing homes contract for physical and occupational therapy, they are being exposed to newer technology. Sub acute care is educating staff and creating advocates for more comprehensive seating solutions (David, 1994). Intelligent surface technology, according to BCA, allows surfaces which come in contact with the body to automatically change shape to facilitate comfort, fit, and safety (David, 1994). It first measures load distribution on the body, then calculates the most comfortable surface shape,

232

and changes the shape of the surface to optimize comfort (David, 1994). “There is a dichotomy in seating,” explained by Sprigle, PhD, biomedical engineer at the Center for Assistive Technology, State University of New York at Buffalo, and an expert on therapeutic seating design. “involving two concepts that are always in opposition -- support and mobility.” (David, 1994). By optimizing support and then allowing that support to change as a patient moves in a chair, intelligent surface technology is bringing new comfort and safety to long-duration seating.

Telecommunication The use of telemedicine technology provides an opportunity to bridge the geographic distance between family and nursing home residents (Debra, 2006). Several studies have focused on using communication technologies in community settings (Demiris, 2001). Videophone technology has been used to enhance communication between home-bound patients and healthcare providers. Several studies confirm the feasibility of using videophone technology in a community setting with few technological challenges (Debra, 2006) Generally, home-based research has found that the technology is accepted by the general public and is easily managed in the home setting. “These interventions have been demonstrated to be cost-effective and satisfactory overall to users.” (Whitten, 2001; Parker, Demiris, Day, Courtney, & Porock, 2006) While most studies have focused on home care, a few have been tried in the nursing home setting (Debra, 2006). A video link between a resident and family may also allow staff to use the technology to communicate with the family, enhancing their relationship with distant caregivers. A facility may consider a shared phone for the facility to decrease resident cost and allow family members the opportunity to connect not only with the resident but also with the entire care team (Debra, 2006). Nursing home providers may want to be prepared

Nursing Home

to accommodate family members experienced with various forms of telecommunications and engage them in the life of the resident, regardless of their geographic distance. Facilities should consider participating in projects that evaluate this and other technology designed to enhance and improve resident and family care (Johnson, Wheeler, Dueser, & Sousa, 2000).

Heart Guard Heart disease is the frequent cause of death and early diagnosis is essential to save lives. Monitoring the heart’s rhythm and electrical activity using an electrocardiogram (ECG) provides vital information about abnormalities and gives clues to the nature of a problem (Hoban, 2003). Some cardiac conditions need long-term monitoring inconvenient for patients as it requires them to be away from their everyday environment for indeterminate periods of time (Hoban, 2003). Six years ago, Latvian company Integris Ltd, a specialist in the development of mobile wireless telemedicine ECG recording devices, came up with the concept of an inexpensive, real-time heart activity monitor for personal use. After years research and tests, 3489 HEART GUARD was born (Hoban, 2003). According to Integris, we know that “The HEART GUARD system comprises a lightweight, simple to use, matchbox-size device with five electrodes that are strategically placed on the wearer’s chest. The wireless device transmits data in real time directly to the patient’s pocket computer or desktop PC for instant interpretation by the system’s unique software” (Hoban, 2003). The low-cost device is discreet enough to be worn 24 hours a day, recording, analyzing and reporting not only the rhythm and electrical activity of a patient’s heart but also his physical activity and body positions, as they go about their daily life (Hoban, 2003). “Effectively, it is an early warning system,” explains Juris Lauznis, Director of Integris, the

project’s lead partner. “If HEART GUARD detects a problem, patients are alerted by means of vibration or a buzzer, prompting them to check their PC for further information and advice. At the very least, the device will help to monitor and manage a patient’s condition and it could even save a life.” (Hoban, 2003). Currently HEART GUARD is being developed for home use only, with patients monitoring their own condition and only contacting a doctor or hospital if the system identifies a cause for concern (Hoban, 2003). HEART GUARD also has applications in a number of other areas, including telemedicine, sports medicine, patient rehabilitation following cardiac surgery or a heart attack and as a low-cost ECG monitoring system in hospitals and nursing homes with limited budgets (Hoban, 2003), which will be extremely beneficial for the elders that want to live home with the monitoring system in the nursing home. With the 30-month project completed and clinical trials of the prototype successfully concluded by Kaunas Medical University’s Institute of Cardiology, the Lithuania Academy of Physical Education and the Research Institute of Cardiology at the University of Latvia, the next steps are to satisfy the EU’s strict compliance requirements for medical devices and then source a company to manufacture and distribute the system (Hoban, 2003). If successful, the first commercial HEART GUARD devices could be on the market and saving lives by the end of 2008 or early 2009 (Hoban, 2003).

The Latest Innovations Regarding to HIT According to long-term care (LTC) health information technology (HIT) Summit, touting the trends and innovations in HIT for long-term care, the latest innovations include handheld devices for care documentation by staff; easy-to-use touch screens with graphic icons to assess the documentations and care by direct care staff; hands-free,

233

Nursing Home

eyes-free voice documentation of assessment and care for certified nursing assistants; and software for electronic prescription processing from electronic physician order systems (O’Connor, 2005). Other technologies include managing medication passes to robotic medication dispensing, capturing quality data electronically for benchmarking and ongoing quality improvement, and operational and management information systems to assist in management functions (Harrison, 2002). According to COMTEX News Network (2009), two of the latest innovations include electronic patient monitoring and e-prescribing devices. Some of the latest technologies include products Vigil Dementia System, which features intelligent software and sensors to detect unexpected behavior, such as extended time out of bed, leaving a room, or incontinence. Others are the monitors for fall prevention, wireless call with location tracking, Wi-Fi coverage for staff communications using voice-over Internet protocol phones, and use of mobile devices necessary for electronic health record (EHR) data entry. In the past, it was a paper-compliant process done at the end of a shift and was not always the most accurate (Andrews, 2009). Facilities moving toward point-of-care systems (like touch pads in the hallways near care delivery) for nursing assistants are seeing more accurate documentation that is impacting their reimbursement rates because it is reflecting what is actually being delivered.

Challenges to Success Wilt admits that training presented one of the biggest challenges to HIT implementation. Employees needed to develop proficiency in computer use, a process that may last for several years. This is one of the more significant barriers to implementation (O’Connor, 2008). A lot of time are needed to spend in training and support to make the communities successful. Nursing homes’ adoption of technology has been slow. We have learned studies from the

234

AHCA and the California HealthCare Foundation. Responses to the December 2006 AHCA and National Center for Assisted Living study “A Snap-Shot of the Use of Health Information Technology in Long Term Care” indicated that 46% of long-term care facilities continued to do the majority of their work on paper or were just beginning to use computers. Only 1% reported being paperless, and only 2% considered themselves fully computerized and just beginning to communicate or communicating fully with outside healthcare providers (O’Connor, 2008).

Concern of Standards In terms of other challenges, the issue of standards creates a concern. According to Andrews (2009), the technology integration has been a challenge mostly because there are standards out there, but they give the implementers of them great flexibility in how to go about using them. It enables to improve the monthly nursing summary that shared between the nurses and doctors by providing an electronic exchange between our two clinical systems. This has improved access to information and hopefully influenced decision making properly. “Many nursing homes are adopting a waitand-see approach until software products are certified by the Certification Commission for Healthcare Information Technology (CCHIT) using the Health Level Seven (HL7)-approved LTC EHR-S functional profile”, says Eileen T. Doll, RN, BS, NHA, president of Efficiency Driven Healthcare Consulting, Inc. in Baltimore. HL7 is an American National Standards Institute standard for healthcare-specific data exchange between computer applications (Sloane, 2006). Proper integration of technology is essential. True streamlining occurs when the clinical and administrative sides are fully integrated in all aspects—no duplicate entry, no data inconsistency, and integrated processes. Duplications are reduced or eliminated when common information

Nursing Home

is shared among care settings and information systems (Sloane, 2006). Reduced duplication also improves accuracy, with less likelihood that the same piece of information is different in other systems. One thing that helps with standardization is the LTC HIT Collaborative. CAST, the American Association of Homes and Services for the Aging, the NCAL, and the AHCA are partners in ensuring that the LTC sector is represented and the national HIT standards cater to the unique requirements of LTC applications (Sloane, 2006).

Nursing Home Size How large must nursing homes be to make various technologies cost-effective? What can be done for smaller homes interested in technology but unable to invest as much as larger operations? According to Majd Alwan, PhD, director of the Center for Aging Services Technologies (CAST) in Washington, DC, and a member of the LTC HIT Summit, “It really depends on the size of the facility, where they are in the process, whether they have the basic infrastructure in place or not, whether they are part of a chain or not, the management’s position, etc. Generally speaking, if shown a return on investment potential, including return in terms of quality and competitive advantage, providers should be willing to invest,” Even so, smaller homes should not be discouraged. “The best approach is to have a plan and to take EHR implementation in stages,” he says. “Plan one application or module at a time based on the organization’s tolerance for change, leadership, and financial resources. If resources are limited, start with one or two smaller applications that will make an impact, such as the nursing assistant documentation/kiosk touchpad, and keep building.” For smaller homes interested in technology, Alwan explains that most companies price technology based on usage, so smaller providers pay less than larger providers. “Many smaller homes choose not to host and manage their software and

data,” he says. “With software as a service, even the smallest [nursing home] providers can benefit.”

The Future of Technology After thorough research work, we believe that resident access to the electronic documentation will continue to be an area providers will eventually offer for their residents. We think that technologies that connect the resident and family closer together will continue to be the most requested items. Innovative technology enables residents and their families to access basic information such as lists of problems, medications, allergies, contact information, and lab reports. The ability to make this easy to use for all will be the challenge but we see an opportunity with telehealth devices that are coming onto the market today for other uses that will eventually be extended to the SNF environment (“Telehealth can improve quality of life,” 2009). As mentioned earlier, a variety of exciting technologies are on the drawing board, including advanced total quality systems that integrate several basic components already in existence, such as nurse call, wandering management, fall prevention, resident tracking, resident assessment, electronic medication administration record, and electronic treatment administration record systems. Also in the works are advanced fall prevention systems, advanced beds with embedded sensors, and comprehensive interoperable EHRs that allow sharing health information securely across different settings. Resistance to change presents one of the biggest challenges to technology integration in the nursing home environment. The industry has been manual and paper-based for so long, it requires a cultural shift to a technology base. We believe that communication is the key to overcoming this challenge. We look forward to seeing technology beyond EHR systems. According to CosmoCom Showcases Telehealth at World Health Care Congress (2009), in the near future, we will see

235

Nursing Home

current high-tech devices integrated with EHRs. Telehealth is an emerging and important technology for the future. Technology is needed to upgrade all equipment used in SNFs as well as changing the homes to meet the cultural change required for a higher quality of life.

PART III: REPORT ON THOMASVILLE NURSING HOME VISITING In this part, we will report our visiting a nursing home in south Alabama.

Background Thomasville Nursing Home is a well known nursing home in south Alabama, which is equipped with standardized facilities. It is a religious nonmedical health care institution with SNF/NF (dually certificated). Five residents are interviewed and four filled out the questionnaire. Three nurses, two social workers, and the administrator are also interviewed by giving out some brief opinions on the technology involved in the nursing home daily bases. Figures 2 (a) and (b) shows the outside photos of the nursing home.

Figure 2. Thomasville nursing home

236

Research Purpose Know the actual conditions in Nursing Home in Alabama rural areas. Research are specific in the residents room decoration, overall quality on the residents’ living conditions, residents satisfaction on Nursing Home care, and staff, residents’ attitudes about technology applying.

Outcomes In Thomasville nursing home, what makes residents happy or unhappy? What are their special wishes? In this report, three themes emerged from resident responses: community, care, supportive relationships. In addition to residents’ daily happiness in nursing homes, quality life includes another great factor: technology involvement.

Community As in any community, nursing home residents focus on their living space, neighbors, and what is happening. For some residents, having others around affords conversation and companionship, especially when their roommate is or becomes a friend. The nursing home provides a sense of “fellowship, friendship, and care.” In contrast, other residents may be uncomfortable with the number of people and the amount of activity level within the environment. These residents are unhappy

Nursing Home

because they “always have someone around,” and wish for “a private room” or “a lot of space like I had at home.” So that in Thomasville Nursing Home, they offer different types of rooms for their residents, though it is more demanding on the single rooms, all of the interviewed residents are somewhat satisfied with their accommodations. Figures 3(a) and (b) shows the single- room pictures. To some residents, there is a sensor pad underneath the bed sheet to secure residents’ safety. Though pets are not allowed, some of the patients still feel like to have a kitty or puppy toy around with them. “TV is another important accompanier to stay away from lonely, which helps us kill time and catch up with the everyday ongoing” said couple of the residents. In the single room, some of the residents feel that they have their own privacy and their own time. Residents enjoy going to the dining room and visiting with their tablemates. They are happy when the food is good, there are snacks at night, and they celebrate birthdays together. They are unhappy when the coffee is not freshly brewed, the same foods are served too often, or the food is not cooked to their personal taste. Residents’ special wishes include “to have food and meat like I had on the farm,” “for someone to bring food from outside for my birthday,” and “to have a Residents enjoy going to the dining room and visiting with their tablemates. They are happy

when the food is good, there are snacks at night, and they celebrate birthdays together. They are unhappy when the coffee is not freshly brewed, the same foods are served too often, or the food is not cooked to their personal taste”. Residents’ special wishes include “to have food and meat like I had on the farm,” “for someone to bring food from outside for my birthday,” and “to have a martini once in a while.” The dining room (Figure 4) and activity room (Figure 5) play a significant role in their daily life in the nursing home. The planned activities provided by nursing homes are an important part of the community. Residents like to be busy and enjoy “the chance to play bingo and laugh.” Some are happy because church services are offered in their home. One of the residents says that many residents want to get into town and see new sights, and others simply would like to “get out more and see flowers and trees.” Others want to go shopping or go out to lunch. Most residents are elders, and the nursing home community may be less satisfying for younger residents. A younger resident says, “There’s a bunch of old people here instead of people my own age.” In Thomasville, the nursing home staffs also say that they understand the different demands on the their residents, however, it is hard to satisfy everyone, say, it’s a complicated process to bring the elders to the shopping in terms of their safety concern; because of the

Figure 3. Residents’ bed room with her toy kitty and TV

237

Nursing Home

Figure 4. Dining room for the residents

small portion of young residents in the nursing home, considering of the budget concern, it is not realistic to arrange activities for couple of them only. Fulfilling residents’ special wishes could increase their sense of community. According to the administrator, Dania, she says that one approach would be for the care planning coordinator or social worker to ask residents about special

Figure 5. Activity scene: Inauguration Day

238

wishes when completing the quarterly or annual Myelodysplastic syndromes (MDS), a group of diseases that affect the bone marrow and blood, assessment. Many of the wishes are easy to grant through staff or family members, and that special treat could be added to the respective resident’s care plan as an onetime event. Using the MDS assessment process ensures that residents with limiting physical or mental impairments would

Nursing Home

be included. Other actions to increase residents’ sense of community might include providing quiet places where they can go to be alone or arranging activities targeted for younger residents that offer interactions with age-mates. Access to the Internet, with assistance in learning to use that technology, could facilitate participation in chat rooms and, perhaps, cyber friendships with peers in other facilities. The “Miss. Nursing Home”, shown in Figure 6, who is 90 year-old, told me with a big smile and proud that she has rewarded as the Miss Nursing Home last year, which brings her a lot of confidence and self-esteem. Such rewarding games and programs are consistently held by Thomasville Nursing Home, which bring all the residents get into their fun daily life.

Care Residents come to nursing homes because they need help with basic activities of daily life. Many Figure 6. “Miss Nursing Home” wears her crown

residents are happy because someone else does the housework, prepares the meals, washes the dishes, and does the laundry. They like having someone who pays attention to them and helps them meet their needs. Having things go the resident’s way is important. Residents appreciate special attention to unique interests or needs, such as having staff communicate with sign language, being able to listen to music of their choice, and having coffee with the chaplain. They need flexibility; for instance, too many baths make one resident unhappy, whereas another’s special wish is to have more baths. Residents are happy when they are satisfied with staff and believe they receive good care. They describe good care as having staff listen to what they say, having someone respond quickly to the call light, being handled gently by staff during care, receiving proper medications and treatments, and having the doctor visit. Personal cleanliness is valued--being kept clean, having clean clothes, and having clean teeth. Good care depends on the facility having enough qualified staff to meet resident needs. Long waits--for meals to be served, for help to the bathroom, or for the nurse--make residents unhappy, which has been changed in Thomasville Nursing Home after getting feedbacks from their residents. One of the residents says that sometimes busy staff save time by doing things residents could do for themselves. It may be faster to change a resident’s briefs than to take him to the bathroom, or to push a resident’s wheelchair to the dining room rather than letting her use a walker. High staff turnover also creates problems. One resident says, “The people who help me change all the time; they don’t know what they’re supposed to do.” Some residents are more vulnerable than others because of cognitive impairment, hearing or vision loss, or limited mobility. A touching special wish is that everybody be treated the same. A resident

239

Nursing Home

observes, “Some people who don’t know where they are get left behind on answering call lights.” According to the nursing home staffs and Diane, It is important that staff seek input from residents and family members about care. Feedback about how their concerns are addressed reinforces a need for their participation. Although maintaining independence is desirable, a resident’s ability to do a task warrants careful evaluation. For example, vision and fine motor skills must be considered in deciding how a resident participates in an insulin injection. Employee recognition programs should emphasize specific staff behaviors that make a difference for residents. They saw a program in another facility where residents and family members nominate staff for a “STAR” award based on outstanding performance, and a star is attached to the employee’s name tag.

Supportive Relationships Establishing new relationships with staff and other residents is part of settling into a nursing home community. Support from family, friends, and staff is critical to new residents’ successful transition and adaptation. Family and friends, “Being with those you love”, make residents happy. Their hearts will be warmed may be just by receiving a small gift, and services from family and friends, such as a cassette player from a nephew, a fish fillet from a friend, and dresses laundered with a daughter’s special touch. Residents who are unhappy about relationships with family most often feel greater separation because of distance or a perceived lack of concern. Special wishes regarding family and friends usually relate to having more contact. However, one man feels that there is too much pressure on his wife and wishes for someone to help her take care of him. Another man wishes that staff would watch his wife, a resident in the same facility, more closely. Staff members make residents happy through interactions based on respect, kindness, and

240

concern, and by taking time to help, listen, and share their lives. Residents are happy when staff are attentive, but without fuss; when they invite them to events, but also allow them to be alone; and, in general, when they try to make things better for them. Residents appreciate a pat on the shoulder and staff coming to tell them goodbye. It is important for them to have something in common with staff; for example, boosting the same sports team. Shared laughter is a crucial happiness factor. For some residents, teasing staff and being teased, kidding, telling jokes, playing tricks on each other, and “picking on” staff add spice to the day. In addition to the resident council, residents need an opportunity to voice concerns privately to someone who can do something about them. Residents may feel powerless or threatened by abandonment in conflicts with staff, and may need support from a family member or friend in voicing their concerns. Family members and friends who do support residents and add quality to their daily lives deserve recognition. Examples of uncomplicated acknowledgments include feedback from staff about how much a resident enjoyed a special outing or a snapshot of a resident wearing a new sweater sent by a brother in another town.

Technology The overall quality is satisfying, and the Thomasville nursing home has already been switch to home-like style, which makes the residents more comfortable to live than that of traditional style. From the interviews on the residents, it reveals that most of the patients are willing to accept the new technology equipments. The overall attitudes on their nursing home quality is fairly satisfied, including the hygiene, food, daily activities, and nursing home services. Two of them feel it is too hard to communicate with others, and lonely is the major issue they do not want to stay in the nursing home. All of them think that it is safer to have the advanced facilities to assist their livings,

Nursing Home

and they feel secured to have the sitting pad on the wheel chair, alarm sleeping sheet on the bed, but they reject the clips connect the cloth and the chair, which makes them feel like being framed and controlled by a string. “Tech-Bed” is another favorite assistant at Thomasville Home. It looks and feels the same as other normal beds, so that the residents do not take it as a weird stuff. The nurses absolutely love those tech beds. “It truly is like another pair of eyes.” Sensors in the zip-on, washable mattress cover provide continuous heart and respiratory rate data. Nothing attaches to the patient. A monitor sits next to the bed and picks up the signals. Hoana presets defaults, but nurses can easily program the sensors to alarm at rates specific to their patients. “We have caught someone going into atrial fibrillation, that we had checked vital signs on an hour before,” Diane, the administrator, said. “It’s an early warning system.” Nurses at Thomasville have also picked up respiratory depression in patients receiving patient- controlled narcotics. The nursing home has monitored nurse and residents satisfaction, fall rates and cardiac arrests. Diane notes positive trends in the preliminary data. The nurses felt that the covers provided a valuable service and they found the device easy to use, said Valerie Martinez, RN, BSN. “It’s a great tool for the nurse, to help them monitor the patients when they can’t be in there,” Martinez said. “You can go back and trend things.” The nurse can assess what the alert was for and how many times it activated. “It allows the nurse to be more involved in the early recognition of patients that are starting to fail,” said Heather Long, RN, chief operating officer of Thomasville. “It provides data that brings the nurse to the bedside.” The Tech-Bed also has a bed exit alarm, with three push-button settings for patients at different fall risks. At high risk, the bed will alert if the patient lifts his or her head or shoulders off the bed. For moderate risk patients, it waits until the patient sits up and for low risk it delays issuing

an alarm until the patient actually exits the bed. Loved ones or the nurse can program the device to play a personalized, prerecorded message to get back into bed and wait for the nurse. “It’s very helpful for the elderly,” Diane said. “They hear a human voice, saying ‘I’m coming to get you,’ and they will stop and wait.” Heather said that patients do not feel or see the sensors. Once the patient understands the system, she said, “It makes them feel more secure and safe.” All of the being interviewed are holding a positive attitudes to have made the nursing home computerized, such as EMR, monitoring equipments, and alarm system. Technology makes their work a lot easier, though it took a while to get used to it. It is useful and powerful tool to make the working place more organized. Dianesays though it is getting harder to get funds from the state or federal government, the government is still supporting the ideas of keeping getting technology into long-term care, because it worth investing by thinking of saving money in a long run. She also points out that the facilities usages are not from patients’ expenses, but from the nursing homes. As for the e-prescribing which is another revolution for healthcare industry, Diane said lots of concerns. “Standards for e-prescribing have recently been developed, so this isn’t an area of widespread adoption yet,” says Diane, the administrator of Thomasville Nursing Home, “This is a complex undertaking for long-term care facilities because it requires standards to communicate between three parties: the facility, the pharmacy, and the physician.” To date, cost has been a primary inhibitor. “We as an industry have not compiled and communicated enough quantitative ROI [return on investment] data,” Diane adds. “I believe that technology can improve the quality of care, reduce risk, save time, and grow the business. These data will continue to be made available and will definitively show providers that technology is good for business and good for resident care.”

241

Nursing Home

Thomasville nursing home has been successfully implementing electronic records management: nurses’ use of clinical notes, clinical assessments, orders, and electronic medication administration records. For nursing aides, the company has implemented electronic activities of daily living documentation. “We are in the process of implementing electronic lab and integration with our Doc EMR [electronic medical record] to allow for even better communication between the teams,” says Diane. “The biggest improvements have been on the management side of the implementation,” she says. “We have been able to provide greater insight into the amount and frequency of the documentation that we have never had before. We are able to monitor more efficiently the documentation process by providing reporting capabilities that are not feasible with paper.” The concern for the Thomasville Home is financial support, “we hope that either the federal government or states will come up with more money to fund general hightech operation and HIT efforts” It is a great visiting to Thomasville Nursing Home, which allowed us to approach the residents’ real life in south United State. Those three themes are the major concern for the residents in this specific nursing home. Additionally, it is a promising future to have technology involve into the nursing homes, because the demanding is still growing, because a lot of facilities in the homes are not greatly developed, however, the residents numbers are still growing. According to the personal visiting and research, it reveals that the elders are willing to accept the new technologies for the sake of their life quality. The staffs as well as the government are taking a positive attitude in considering of long term benefits.

242

CONCLUSION This chapter provides broad information on the current and future views of nursing home in the United States, in terms of the condition and management level, residents’ satisfactions and demands, government regulation and influence. The ability of e-health technology enables nursing homes to improve their quality to meet residents’ needs, though there are challenges. The visiting report about Thomasville Nursing Home reaches the depth of the consideration to how to catch the trends by implementing the technologies.

ACKNOWLEDGMENT This work is supported in part by the US National Science Foundation (NSF) under the grant number CNS-0716211, as well as RGC (Research Grants Committee) award at The University of Alabama 2008.

REFERENCES Activities. (n.d.). Retrieved October 20, 2008, from http://www.hardingnh.com/ Alexander, G. L. (2008). A Descriptive Analysis of a Nursing Home Clinical Information System with Decision Support. Perspectives in Health Information Management / AHIMA, American Health Information Management Association, 5(12). Alexander, G. L. (2008). A Descriptive Analysis of a Nursing Home Clinical Information System with Decision Support. Perspectives in Health Information Management / AHIMA, American Health Information Management Association, 5(12).

Nursing Home

American Health Information Management Association. (2005). A road map for health IT in long-term care. Chicago: Author. Analysis of the National Nursing Home Survey (NNHS) (2004). AARP Public Policy Institute. Anderson, R., Issel, L. M., & McDaniel, R. R. (2003). Nursing homes as complex adaptive systems: relationship between management practice and resident outcomes. Nursing Research, 52(1), 12–21. doi:10.1097/00006199-200301000-00003 Anderson, S., & Wittwer, W. (2004). Using barcode point-of-care technology for patient safety. Journal for Healthcare Quality, 26(6), 5–11. Andrews, J. (2009). Keeping record: long-term care may not deserve its reputation as a tech laggard. But more could be done to prepare for the electronic health record.(Information technology). McKnight’s Long-Term Care News, 30(1), 42(43). Andrews, J. (2009). Keeping record: long-term care may not deserve its reputation as a tech laggard. But more could be done to prepare for the electronic health record.(Information technology). McKnight’s Long-Term Care News, 30(1), 42(43). Assisted Living Facilities (ALF). (2007). Virginia Department of Social Services: Virginia.gov. Baron, R. J., Fabens, E. L., Schiffman, M., & Wolf, E. (2005). Electronic health records: Just around the corner? Or over the cliff? Annals of Internal Medicine, 143(3), 222–226. Barrick, A., & Rader, J. (2008). Bathing without a battle: person-directed care of individuals with dementia. New York: Springer Pub. Barrick, A., & Rader, J. (2008). Bathing without a battle: person-directed care of individuals with dementia. New York: Springer Pub.

Bates, D. W., Cohen, M., Leape, L. L., Overhage, J. M., Shabot, M. M., & Sheridan, T. (2001). Reducing the frequency of errors in medicine using information technology. Journal of the American Medical Informatics Association, 8(4), 299–308. Bates, D. W., Leape, L. L., Cullen, D. J., Laird, N., Petersen, L. A., & Teich, J. M. (1998). Effect of computerized physician order entry and a team intervention on prevention of serious medication errors. Journal of the American Medical Association, 280(15), 1311–1316. doi:10.1001/ jama.280.15.1311 Bates, D. W., Teich, J. M., Lee, J., Seger, D., & Kuperman, G. J., Ma’Luf, N., et al. (1999). The impact of computerized physician order entry on medication error prevention. Journal of the American Medical Informatics Association, 6(4), 313–321. Berger, L. (2006, January 13). Information technology feature: The IT payback. McKnight’s Long-Term Care News, (pp. 34-37). Boumans, N., Berkhout, A., & Landeweerd, A. (2005). Efforts of resident-oriented care on quality of care, wellbeing and satisfaction with care. Scandinavian Journal of Caring Sciences, 19(3), 11–15. doi:10.1111/j.1471-6712.2005.00351.x Boumans, N., Berkhout, A., & Landeweerd, A. (2005). Efforts of resident-oriented care on quality of care, wellbeing and satisfaction with care. Scandinavian Journal of Caring Sciences, 19(3), 11. doi:10.1111/j.1471-6712.2005.00351.x Breen, G., & Zhang, N. (2008). Theoretical Analysis of Improving Resident Care. Journal of Medical Systems, 32(2), 18–23. doi:10.1007/ s10916-007-9121-9 Breen, G. M., & Matusitz, J. (2007). An interpersonal examination of telemedicine: Applying relevant communication theories. e-health. International Journal (Toronto, Ont.), 3(1), 187–192.

243

Nursing Home

Burland, E. M. J. (2008). An Evaluation of a Fall Management Program in a Personal Care Home Population. Healthcare Quaterly, 11(3), 137–140. Burland, E. M. J. (2008). An Evaluation of a Fall Management Program in a Personal Care Home Population. Healthcare Quaterly, 11(3), 137–140. Caring for the Elderly: Is Adequate Long-Term Care Available. (1998). Congressional Quarterly, Inc., CQ Researcher. Carrillo, H., & Harrington, C. (2006). Cross the States 2006: Profiles of Long-Term Care and Independent Living. In AARP (Ed.). Castle, N. G. (2008). Nursing Home Evacuation Plans. American Journal of Public Health, 98(7), 1235–1241. doi:10.2105/AJPH.2006.107532 CMS Adds Searchable Database of LowestQuality Nursing Homes Nationwide to Web Site. (2008). Kaiser Family Foundation. Corporation, W. (2005). Our products and services. About WebMD.com. CosmoCom Showcases Telehealth at World Health Care Congress (2009). Health & Beauty Close-Up, NA. Daily Activities. (n.d.). Retrieved October 20, 2008, from http://seniors-site.com/nursingm/ index.html David, P. (1994). A revolution in seating - Nursing Home Technology. Retrieved November 03, 2008. Debra, O., & Brain, H. (2006). A promising Technology to Reduce Social Isolation of Nursing Home Residents. Journal of Nursing Care Quality, 21(4), 302–306. Demiris, G., Speedie, S. M., & Finkelstein, S. (2001). Change of patients’ perceptions of TeleHomeCare. Telemedicine Journal and e-Health, 7(3), 241–249. doi:10.1089/153056201316970948

244

Durable medical equipment: Scope and conditions. (2006). CFR, 38, 42. Eljay,L.(2006). A Report on Shortfalls in Medicaid Funding for Nursing Home Care: American Health Care Association. Facts, N. (2006). Today’s Registered Nurse Numbers and Demographics. Washington, DC: A. N. Association. Field, T., Rochon, Paula. (2008). Costs associated with developing and implementing a computerized clinical decision support system for medication dosing for patients with renal insufficiency in the long-term care setting. Journal of the American Medical Informatics Association, 15(4), 466–472. doi:10.1197/jamia.M2589 Field, T., & Rochon, P. (2008). Costs associated with developing and implementing a computerized clinical decision support system for medication dosing for patients with renal insufficiency in the long-term care setting. Journal of the American Medical Informatics Association, 15(4), 466–472. doi:10.1197/jamia.M2589 Finnema, E., de Lange, J., Droes, R. M., Ribbe, M., & Tilburg, W. (2001). The quality of nursing home care: do the opinions of family members change after implementation of emotion-oriented care? Journal of Advanced Nursing, 35(5), 728–732. doi:10.1046/j.1365-2648.2001.01905.x Fun Times. (n.d.). Retrieved October 20, 2008, from http://www.goldenlivingcenters.com/ GGNSC General Accounting Office. (2004). HS’s efforts to promote health information technology and legal barriers to its adoption. Washington, DC: Author. Gill, TM, A. H., Han L. (2006). Bathing disability and the risk of long-term admission to a nursing home. Journal of Gerontology, 61(8), 821–825.

Nursing Home

Gill, TM, A. H., Han L. (2006). Bathing disability and the risk of long-term admission to a nursing home. Journal of Gerontology, 61(8), 821–825. Greenly, M., & Gugerty, B. (2002). How bar coding reduces medication errors. Nursing, 32(5), 70. Greenly, M., & Gugerty, B. (2002). How bar coding reduces medication errors. Nursing, 32(5), 70. Hamilton, B. (2006). Evaluation design of the business case of health information technology in long-term care (Final report). Baltimore: Centers for Medicare and Medicaid Services. Harris, Y., & Clauser, S. B. (2002). Achieving improvement through nursing home quality measurement. Health Care Financing Review, 23(4), 13. Harrison, S. (2002). Telehealth skills hailed as answer to discharge delays: costs of high-tech monitoring systems compare favourably with long-term care. (news). Nursing Standard, 17(12), 7(1). Harrison, S. (2002). Telehealth skills hailed as answer to discharge delays: costs of high-tech monitoring systems compare favourably with long-term care. (news). Nursing Standard, 17(12), 7(1). Hetzel, L., & Smith, A. (2001). The 65 and Over Population: 2000-Census 2000 Brief. Washington, DC: U.S. CENSUS BUREAU. Hoban, S. (2003). Activities plus at Montgomery Place: the “new” active senior wants to be entertained, enlightened, and engaged--and will let you know how. Nursing Homes, 52(6), 52–56. Holmes, D. (2007). An evaluation of a monitoring system intervention: falls, injuries, and affect in nursing homes. Clinical Nursing Research, 16(4), 317–335. doi:10.1177/1054773807307870

Holmes, D. (2007). An evaluation of a monitoring system intervention: falls, injuries, and affect in nursing homes. Clinical Nursing Research, 16(4), 317–335. doi:10.1177/1054773807307870 Home Care and Nursing Home Services. (n.d.). Retrieved December 12, 2008, from http://dhs. dc.gov/dhs/cwp/view,a,3,Q,613301.asp Home Health Service. (2003). CFR484.30, 1(42). Inspectors Often Overlook Serious Deficiencies at U.S. Nursing Homes (GAO Report). (2008). Medical New Today. Institute of Medicine. (2001). Crossing the quality chasm: A new health system for 21st century. Washington, DC: National Academy Press. (1996). Intelligent remote visual monitoring system for home health care service. InPatient, U. S. (Ed.), United States Patient. Johnson, B., Wheeler, L., Dueser, J., & Sousa, K. (2000). Outcomes of the Kaiser Permanente Tele-Home Health Research Project. Archives of Family Medicine, 9(4), 40–45. doi:10.1001/ archfami.9.1.40 St. Joseph Convent retirement and nursing home. (2007). McKnight’s Long-Term Care News, 1. Kemper, P., Komisar, H. L., & Alecixh, L. (2005). Long-Term Care Over What can Current Retirees Expect? Inquiry, 42, 15. King, W. R., & He, J. (2006). A meta-analysis of the technology acceptance model. Information & Management, 43(6), 740–755. doi:10.1016/j. im.2006.05.003 King, W. R., & He, J. (2006). A meta-analysis of the technology acceptance model. Information & Management, 43(6), 740–755. doi:10.1016/j. im.2006.05.003 Kohn, L. T., Corrigan, J. M., & Donaldson, M. S. (1999). To err is human: Building a safer human system. Washington, DC: National Academy Press.

245

Nursing Home

Kramer, A., Bennett, R., Fish, R., Lin, C. T., Floersch, N., & Conway, K. (2004). Case studies of electronic health records in post-acute and longterm care. Washington, DC: U.S. Department of Health and Human Services. Lilley, R. C. L. P., Lambden P. (2006). Medicines management for residential and nursing homes: a toolkit for best practice and accredited learning. Seattle: Radcliffe. Lilley RC, L. P., Lambden P. (2006). Medicines management for residential and nursing homes: a toolkit for best practice and accredited learning. Seatle: Radcliffe. MacLaughlin, E. J. (2005). Assessing medication adherence in the elderly: which tools to use in clinical practice? Drugs & Aging, 22(3), 55–231. MacLaughlin, E. J. (2005). Assessing medication adherence in the elderly: which tools to use in clinical practice? Drugs & Aging, 22(3), 55–231. Matusitz, J., & Breen, G. M. (2007). Telemedicine: Its effects on health communication. Health Communication, 21(1), 10–21. McGraw, C. (2004). Multi-compartment medication devices and patient compliance. British Journal of Community Nursing, 9(7), 90–285. McGraw, C. (2004). Multi-compartment medication devices and patient compliance. British Journal of Community Nursing, 9(7), 90–285.

Michael, S. J. C. (2000). Wireless clinical alerts for physiologic, laboratory and medication data. Proceedings / AMIA 2000 Annual Symposium, 789-793. Mitka, M. (2003). Approach could widen access for older patients. Telemedicine eyes for mental health services, 290(14), 22-25. Moller, J., Renegar, Carrie. (2003). Bathing as a wellness experience. Nursing Homes Long Term Care Management, 52(10), 108. Moller, J., Renegar, Carrie. (2003). Bathing as a wellness experience. Nursing Homes Long Term Care Management, 52(10), 108. Mor, V., Berg, K., Angelelli, J., Gifford, D., Morris, J., & Moore, T. (2003). The quality of quality measurement in U.S. nursing homes. The Gerontologist, 43(2), 37–46. More Can Be Done to Protect Residents from Abuse. (2002). U.S. General Accounting Office. Mukamel, B. D., & Spector, D. W. (2000). Nursing home Costs and Risk-Adjusted Outcome Measures of Quality. Medical Care, 38(1), 32–35. doi:10.1097/00005650-200001000-00009 National Health Accounts. (n.d.). In C. f. M. a. M. S. (CMS) (Ed.). Nursing Home Data Compendium. (2007). In CMS (Ed.): HS.gov.

(2008). Medical & You handbook. InServices, M. M. (Ed.), C. f.

NURSING HOMES. (2008). Federal Monitoring Surveys Demonstrate Continued Understatement of Serious Care Problems and CMS Oversight Weaknesses. In Office, S. G. A. (Ed.), U (Vol. 517). GAO.

Michael, S. J. C. (2000). Wireless clinical alerts for physiologic, laboratory and medication data. In Proceedings / AMIA 2000 Annual Symposium, (pp. 789-793).

O’Connor, J. (2005). Feds provide info on LTC tech products on market.(Technology)(Long Term Care)(Brief Article). McKnight’s Long-Term Care News, 14(11).

McKnight’s LTC News. (2005, July). 2005 software. McKnight’s Long-Term Care News, 43.

246

Nursing Home

O’Connor, J. (2008). Nursing homes up on technology, (NEWS). McKnight’s Long-Term Care News, 29(12), 3(1).

Reilly, J., & Bischoff, J. (2008). Remote Monitoring Tested. Technology and Health Care, 34(9), 12–14.

O’Connor, J. (2008). Nursing homes up on technology.(NEWS). McKnight’s Long-Term Care News, 29(12), 3(1).

Robots may be next solution to nursing shortage. (2003). Managed Care Weekly, 102.

Otis, P. (1997). Don’t worry, The Carpet is Keep Watching. BusinessWeek,1. Overview Medicaid Program. (2006). General Information (In, C. M. S., Ed.). Parker, O. D., Demiris, G., Day, M., Courtney, K. L., & Porock, D. (2006). Tele-hospice support for elder caregivers of hospice patients: two case studies. Journal of Palliative Medicine, 9(2), 54–59. Patient Placement Systems Unveils Referral Automation Vision Center. (2009). PRWeb, NA. Pear, R. (2008a). Serious Deficiencies in Nursing Homes Are Often Missed. New York Times. Pear, R. (2008b). Violations Reported at 94% of Nursing Homes. New York Times. Peter, D. (2001). The Nurse Shortage: Perspectives from Current Direct Care Nurses and Former Direct Care Nurses (Report). Polit, F. D., & Beck, T. C. (2004). Nursing Research: Principles and Methods. Wickford, RI: Lippincott-Raven Publishers. Poon, E. G., Jha, A. K., Christino, M., Honour, M. M., Fernandopulle, R., & Middleton, B. (2006). Assessing the level of healthcare information technology adoption in the United States: A snapshot. BMC Medical Informatics and Decision Making, 6, 1. doi:10.1186/1472-6947-6-1 Redfern, S., Hannan, S., Norman, I., & Martin, F. (2002). Work satisfaction, stress, quality of care and morale of older people in a nursing home. Health & Social Care in the Community, 10(6), 17–35. doi:10.1046/j.1365-2524.2002.00396.x

Robots may be next solution to nursing shortage. (2003). Managed Care Weekly, 102. Schepers, J., & Wetzels, M. (2007). A metaanalysis of the technology acceptance model: Investigating subjective norm and moderation effects. Information & Management, 44(1), 90–103. doi:10.1016/j.im.2006.10.007 Schepers, J., & Wetzels, M. (2007). A metaanalysis of the technology acceptance model: Investigating subjective norm and moderation effects. Information & Management, 44(1), 90–103. doi:10.1016/j.im.2006.10.007 Shields, S. (2005, March 23). Culture Change in Nursing homes. The Commonwealth Fund. Sloan, F., & Shayne, M. (1993). Long-Term Care, Medicaid, and the Impoverishment of the Elderly. The Milbank Quarterly, 70(4), 19. Sloane, E. B. (2006). The emerging health care IT infrastructure, (Tech Talk). 24x7, 11(12), 42(43). Sloane, E. B. (2006). The emerging health care IT infrastructure.(Tech Talk). 24x7, 11(12), 42(43). Sloane, E. B. (2009). Regulatory overview (Networking). 24x7, 14(4), 24(22). Sloane, E. B. (2009). Regulatory overview.(Networking). 24x7, 14(4), 24(22). Staff Turnover and Retention. (2007). AAHSA. Telehealth can improve quality of life. (2009). Nursing Standard, 23(24), 21(21). Telehealth can improve quality of life. (2009). Nursing Standard, 23(24), 21(21).

247

Nursing Home

Thompson, L. T., Dorsey, M. A., Miller, I. K., & Parrott, R. (Eds.). (2003). Telemedicine: Expanding healthcare into virtual environments. Mahwah, NJ: Handbook of health communication. Wald, H., & Shojania, K. G. (2001). Incident reporting. In Shojania, K. G., Ducan, B. W., McDonald, K. M., & Wachter, R. M. (Eds.), Making health care safer: A critical analysis of patient safety practices (pp. 41–47). Rockville, MD: Agency for Healthcare Research and Quality. Webb, A. (2001). The Impact of the Cost of LongTerm Care on the Saving of the Elderly. New York: International Longevity Center. Whitten, P., Doolittle, G., & Heilmich, S. (2001). Telehospice: using telecommunication technology for terminally ill patients. Comput-Mediat Commun, 6(4), 43–47. Williams, L. (2003). Medication-Monitoring Lawsuit: Case Study and Lessons Learned. Nursing Homes/Long Term Care Management, 3. Wolf, A. (2003). Behind Closed Doors. NFPA Journal.

Wolfstadt, J. (2008). The effect of computerized physician order entry with clinical decision support on the rates of adverse drug events: a systematic review. Journal of General Internal Medicine, 23(4), 451–458. doi:10.1007/s11606-008-0504-5 Wolfstadt, J. (2008). The effect of computerized physician order entry with clinical decision support on the rates of adverse drug events: a systematic review. Journal of General Internal Medicine, 23(4), 451–458. doi:10.1007/s11606-008-0504-5 Wootton, R. (2001). Telemedicine. British Medical Journal, 323, 4. Wunderlich, G. S., & Kohler, P. O. (2001). Improving the quality of long-term care: 4. Information systems for monitoring Quality. Washington, DC: National Academy Press. Yu, P., Li, H., & Gagnon, M.-P. (2009). Health IT acceptance factors in long-term care facilities: A cross-sectional survey. International Journal of Medical Informatics, 78(4), 219–229. doi:10.1016/j.ijmedinf.2008.07.006 Yu, P., Li, H., & Gagnon, M.-P. (2009). Health IT acceptance factors in long-term care facilities: A cross-sectional survey. International Journal of Medical Informatics, 78(4), 219–229. doi:10.1016/j.ijmedinf.2008.07.006

This work was previously published in Pervasive and Smart Technologies for Healthcare: Ubiquitous Methodologies and Tools, edited by Antonio Coronato, pp. 39-77, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

248

Section II

Development and Design Methodologies This section provides exhaustive coverage of conceptual architecture frameworks to endow with the reader a broad understanding of the promising technological developments within the field of clinical technologies. Research fundamentals imperative to the understanding of developmental processes within clinical technologies are offered. From broad surveys to specific discussions and case studies on electronic tools, the research found within this section spans the discipline while offering detailed, specific discussions. From basic designs to abstract development, these chapters serve to expand the reaches of development and design technologies within the clinical technologies community.

250

Chapter 2.1

Improving Clinical Practice through Mobile Medical Informatics Tagelsir Mohamed Gasmelseid King Faisal University, Saudi Arabia

ABSTRACT This chapter introduces the use of mobile medical informatics as a means for improving clinical practice in Sudan. It argues that mobile medical informatics, combined with new techniques for discovering patterns in complex clinical situations, offers a potentially more substantive approach to understanding the nature of information systems in a variety of contexts. Furthermore, the author hopes that understanding the underlying assumptions and theoretical constructs through the use of DOI: 10.4018/978-1-60960-561-2.ch201

the Chaos Theory will not only inform researchers of a better design for studying information systems, but also assist in the understanding of intricate relationships between different factors.

INTRODUCTION Healthcare organisations are undergoing major transformations to improve the quality of health services. While emphasis tends to be made (especially in developing countries) on the acquisition of improved medical technology, part of this transformation is directed towards the management

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Improving Clinical Practice through Mobile Medical Informatics

and use of rapidly growing repositories of digital health data. However, the heterogeneity associated with such types of data (variety of formats, lack of data standardisation, mismatch between data and proprietary software architectures and interfaces, etc) call for addressing the problems of system and information integration. Medical and healthcare applications and services are becoming knowledge intensive. Therefore, advanced information systems’ technology is essential to produce, coordinate, deliver, and share such information. The migration from static “medical” decision support systems technologies (a.k.a medical informatics technologies) towards a new breed of more malleable software tools allow the users of medical information systems to work with data and methods of analysis within the context of a “teleportal hospital”. These medical information systems can remain efficient and effective by continuously adopting and incorporating the emerging mobile technologies. Mobile computing is going a long way in reshaping the context of healthcare provision and its efficiency. The growing physical mobility of patients mandates the use of mobile devices to access and provide health services at any time and any place. Example of such healthcare provisions range from car or sports-accidents through to research and cure of long-lasting diseases such as allergies, asthma, diabetes and cancer. The basic aim of this chapter is to investigate the potential of improving clinical practice in public hospitals in Sudan through the use of mobile medical informatics.

BACKGROUND The health care sector in Sudan is being challenged by many organizational, institutional, technical and technological issues that endangered its ability to provide quality services (UNFPA website, UNCEF website, Ministry of Health website). Because public hospitals are competing with other government units for public funds, they

failed to acquire appropriate medical technology and improve clinical practice through improved diagnosis and staff training and retention. The lack of a sound managing capacity has also reduced their ability to integrate backward (with community and rural hospitals) and coordinate forward (with educational institutions, industry and research community). The recent economic liberalization has also increased both the “financial” and “managerial” overheating of public hospitals who fail to run as self-sufficient units rather than “cost centers’. While the quality of the services provided by private clinics and hospitals (both inside and outside Sudan) tends to be high their paramountly high costs make them out of the reach of many patients. The deterioration of the quality of health services and clinical practice due to the following: 1. The lack of financial resources on the side of public hospitals due to the fact that they compete for “limited” public funds with other institutions. Their failure to acquire funds has also been accompanied with a considerable difficulty in developing appropriate plans for the effective management of healthcare institutions at the primary and secondary health service provision spectrum. Such mis-management issue has resulted into a considerable failure to develop a matrix of priorities according to which tasks “especially at the two main entrants or gates of service provision for critically I and critically III patients”: Accidents & Causality and Intensive Care Units (ICUs). Especially in public clinics and hospitals the deterioration of service quality originates from the fact that there is a lack of medical supplies. Due to the privatization trends” patients are required to pay for basic inputs. Medical consultants, whose presence at this gate is paramountly important, are not prepared to be there because they are spending much time in their private clinics. Although the

251

Improving Clinical Practice through Mobile Medical Informatics

management of ICUs tends to scientific, there is a considerable difficulty regarding the development and adoption of the suitable management model. There are five ICU models: (a) open ICU model, (b) closed ICU model, (c) co-managed ICU model, (d) managed by an intensive model and (e) mixed model. Noticeably all ICUs across the country are managed using the open model and this reflects the fact that those clinics are not patient-centered to address patient service through innovative solutions. The acquisition of medical technology tend also to be affected by the inappropriate policy making where hospital decision makers tend to think and manage in medical orientations in a way that limits their ability to understand the determinants and processes of technology acquisition (including costs of technological infusion, diffusion, maintenance and training) that they fail to address the context of competition and market efficiency forms. Medical training of newly appointed medical staff (houseman-ship training programs) is negatively affected by economic trends that made specialists and consultants unavailable at attainable periods to train graduates as well as medical students. Moreover, clinical training of medical students has been negatively affected by the growing number of medical institutes and the mismatch between the number of patients, clinically-trainable beds, medical specialists and consultants (in different areas of expertise) and the number of students to be trained. 2. The lack of appropriate incident reporting systems, adverse drug events and order issuance and management protocols to address medical errors. It worth mentioning that this component depends upon intensive R&D and is fast moving. The result of such issues is the deterioration of medication quality, error-inclusive clinical prac-

252

tice and the growing costs of health and medical services. Because the overall health sectors fails to develop and implement a maintainable financial and technical sustainability, patients continued to travel to other countries for further investigation and management. In addition to medication associated costs, patients continued to face additional travel, accommodation and misconduct costs. Costs of “mediation” and misconduct associated with abuse and payment policies (reserving even bodies till payment is made”, un-necessary medication and operations and others are all examples.

MOBILITY IN HEALTHCARE Mobile and wireless systems are gaining paramount importance and deployment in They have been widely used in medical informatics, e-business, e-learning, supply chain management, virtual enterprises (Jain et al 1999), information retrieval (Cabri, et al 2000), Internet-based auctions (Sandholm & Huai 2000), distributed network management (Du et al 2003), resource management and broadband intelligent networks (Chatzipapadopoulos et al 2000), telecommunication services, and mobile and wireless computing (Keng Siau et al, 2001). Their use in the healthcare system enhances operational and contextual functionality, improving the availability, accessibility and management of decentralized repositories of concurrent data (Gasmelseid, 2007d). Mobility allows different agents to move across different networks and perform tasks on behalf of their users or other agents, by accessing databases and updating files in a way that respects the dynamics of the processing environment and intervention mechanisms (Gasmelseid, 2007c). The complexity exhibited in the healthcare industry is motivating the distribution of both the organisational structure and enabling information systems. The basic objective of such distributed environment is to build a network of complementary and integrated medical and health centres

Improving Clinical Practice through Mobile Medical Informatics

such as hospitals, laboratories, ambulatories, coordination centres, pharmacies, diagnosis centres and support units in accordance with diversity and imperativeness of services. Within this context, although each individual centre operates as an autonomous unit devoted to the delivery of a particular set of services, they are interacting to provide efficient prevention and care. According to Francesco Fedele (1995), the information systems supporting the individual centres must be structured as a federation of autonomous systems, individually optimised according to the specific characteristics of the involved units. In parallel, the individual systems must be consistent with information and procedural standards, in order to permit the mutual interoperability to provide an effective and efficient support to the co-operation individually provided by each single unit. Mobility is defined as the ability to “wirelessly” access all of the services that one would normally have in a fixed wired line environment such as a home or office, from anywhere. It includes terminal mobility (provided by wireless access), personal mobility (based on personal numbers), and service portability supported by the capabilities of intelligent networks. Especially in technology-intensive and information-rich distributed environments, mobile agents interact to gather information, route processing outcomes, update a database and read performance levels, among others. The efficiency of any mobile multi agent system is based on its capacity to achieve objectives and adapt to changing environments (Gasmelseid, 2007c). Mobility in healthcare has been associated with the concepts of ubiquitous healthcare which extends the dimensions of web-based healthcare computation. It allows individuals to access and benefit from healthcare services through mobile computing devices. While such fully or semi automatic access of healthcare information and services assists in reducing operational complexity, it also turns healthcare institutions into effective learning organizations. Knowledge provided

through mobile medical informatics includes clinical, personal, situational (environmental), and historical medical and health histories (relevant past diseases / operations / therapies mirrored upon current symptoms or already available diagnosis) indicators. Although under certain conditions, some of that knowledge may be captured from patients themselves through mobile devices, other knowledge may be accessed through distributed databases located any of the three at levels of the healthcare control process: federal, territorial and patient care centres such as hospitals, family doctors, medical specialists, pharmacy, and diagnostic centres. Because the maintenance of financial sustainability and competitive advantage (market efficiency forms, service differentiation, process efficiency, and improved patient relationship management) is looming very big in the healthcare system, mobile medical informatics offers an ITenabled platform for continuous improvement. The implementation of mobile medical informatics initiatives tend to be based on alternative technological, architectural, and infrastructurebased methodologies. This paper aims at using a mobile agent-enabled architecture to improve clinical practice through improved knowledge intensive cooperation among humans and intelligent software agents involved in producing, delivering, controlling, and consuming health services. The proposed technology proved to be useful also in improving operational efficiency through shared medical processes. The importance and feasibility of using mobile multiagent systems stems from two main reasons: 1. The diversity and multi-dimensionality of the healthcare structure: The healthcare provision system in Sudan is spread over three levels each with different level of complexity. The federal level is concerned with the promotion of national health strategies. At the Territorial (state) level Epidemiological and coordination centres take the responsibility of epidemiological and planning activities.

253

Improving Clinical Practice through Mobile Medical Informatics

At the medical centres level, hospitals operate as the main providers of patient caring. Despite the fact that each hospital is composed of a set of specialized service-based units, they also interact consistently within the context of the overall functionality of the global structure. Mobility helps in improving the quality of healthcare services and medical practice through improved information provision and location-specific organizational and managerial control. It also maintains effective institutional (within each healthcare institution) and structural (across the three levels of health control) coordination. 2. The characteristics of software agents that make them good candidates to implement distributed management. Within the context of a distributed multiagent mobile medical information system, the entire structure can be populated with a wide range of software agents possessing different qualities and attributes. Based on such attributes they can be classified. An intelligent agent is as an autonomous, computational software entity that has access to one or more, heterogeneous and geographically distributed information sources, and which proactively acquires, mediates, and maintains relevant information on behalf of users or other agents (Gasmelseid, 2007a). While autonomy remains a leading agency attribute other attributes like mobility, pro-activity, conviviality, among others are characterizing the use of multiple agents within the context of an integrated organizational structure. The topology of agents is based on their attributes and userspecific functions. Therefore, the classification includes internet agents, information agents, interface agents, shopping agents, search agents etc. However, the autonomous behaviour of these agents is determined by their proactiveness, reactive and deliberative actions, and social interactions. In a multiagent system, agents jointly use

254

knowledge and resources to solve problems in a context-dependent way (Gasmelseid, 2007b). Within the context of a distributed healthcare, software agents can provide assistance in different ways. For example, they can facilitate (vertical and horizontal) coordination, cooperation and interaction among different remote health and medical centres at the three levels of the healthcare system. While this will improve the effectiveness of communication it also enhances interaction among medical staff and patient mobility resulting from embedded orchestration of service and physical mobility and service accessibility enabled by the uniformity of service points. The use of mobile agents allows the incorporation of a wide range of object oriented features necessary for information sharing, encapsulation and inheritance. This is because their conceptualization is based on the “articulation” of roles, players and relationships on the one hand and “drawing” road maps for system-wide critical success “technical” factors including security and authentication on the other hand. In addition to their roles as information gatherers, filters and learners, multi agent systems support various phases of decision making and problem solving by serving as “problem analyzers and solvers” and are engaged in “implementation”, “monitoring” and “negotiation” (Gasmelseid, 2006). When multi agent systems are used, the problem-solving tasks of each functional unit becomes populated by a number of heterogeneous intelligent agents with diverse goals and capabilities (Lottaz, et al, 2000; Luo, et al, 2001; McMullen, 2001; Ulieru, et al, 2000; Wu, 2001). The system simplifies problem-solving by dividing the necessary knowledge into subunits, associating an intelligent independent agent to each subunit, and coordinating the agents’ activity. Multi-agent systems offer a new dimension for coordination and negotiation by incorporating autonomous

Improving Clinical Practice through Mobile Medical Informatics

agents into the problem-solving process and improving coordination of different functional unit defined tasks, independent of both the user and the functional units under control (Byung & Sadeh, 2004).

IMPROVING CLINICAL PRACTICE THROUGH MOBILE MEDICAL INFORMATICS Epidemiologic techniques are suited to complications and major transfusion errors that occur with reasonable frequency rather than rare ones. The conceptualization of medical errors is based on the taxonomy provided by Reason (1990; 1997; 2000) who classified errors into active error (occurs at the point of human interface with a complex system) and latent error (represents failures of system

design). Medical errors are also associated with the lack or inappropriateness of existing patient safety practices (mainly Quality improvement practices) that increases the probability of adverse events within the health care system across a range of conditions or procedures. The use of mobile agent-mediated medical informatics to improve clinical practice in Sudan can be conceptualized using the work centred analysis shown in Figure 1 below. Work centred analysis is usually based on the identification of “users”, “major steps”, “basic approach”, and “technology”. The basic aim beyond the use of such kind of analysis as a methodological tool is to ensure the incorporation of an integrated distributed hospital environment. Therefore, emphasis tend to be made maintaining coordination between medical centres and units at the interface between inpatient and outpatient

Figure 1. A work-centered analysis

255

Improving Clinical Practice through Mobile Medical Informatics

settings; promoting performance improvement and decreasing the number of diagnostic errors of omission. As shown in Figure 2, the distributed environment of the system is based on building effective interfaces among the different medical areas of expertise in accordance with their shared functionalities and rule based integration and reasoning of acquired information. The functionality of mobile information systems in clinical fields is contingent upon its operational capacity to provide clinical support as well as the “optimality” of the operating environment surrounding it as shown in Figure 3. The contribution of such agent mediated mobile medical information system in the Sudanese context can be outlined below:

Figure 2. System based interaction

Figure 3. System enabling environment

256

Universality of Patient Data and Records The question of managing and using patient’s medical data looms very big in Sudan. Especially for kidney failure and dialysis patients their mobility is growing over the last couple of years due to the lack of reliable medical services in remote trajectories and/or their inability to meet the growing medication costs. Their migration to the capital originates from their interest to benefit from the financial support provided by social funds and the concentration of qualified medical staff. Under many conditions, when they move from one city to the other they miss to bring with them their medical history. Also the same applies for referred cases where the information provided by different hospitals lacks standardization in a

Improving Clinical Practice through Mobile Medical Informatics

way that make it difficult for different medical specialists to use them as valid entities in their operational databases. The use of mobile multiagent systems provides an opportunity for reducing such complexity through the following: 1. Patients’ data can be maintained by the firstto-see medical specialist in his hospital’s data warehouse or knowledge base. Records of different patients are kept updated every time the patient visits his/her medical specialist. To facilitate information access patients’ data can be stored on smart cards and updated frequently. Upon referring patients to another medical specialist or center, patient’s data can be accessed from the smart card and updated accordingly by keeping original data unchanged. Because of the unprecedented technological developments witnessed in the medical field, mobile software agents can room and move across the internet or shared WANs to access the database of the originating medical center or specialist to retrieve necessary data. The advantage of this process are the following: a. Originating medical center or specialist will be informed about the recent developments in his patient’s health record and accordingly he/she can intervene at any point of time because of his access to the resulting real time information. b. Access can be given to a third party for a second opinion about the medical status of the patient as well as the suitability of medical management practices. This process is supported by the possibility of sharing diagnosis data such as MRI and other ultra sound images and the outstanding interface qualities of software agent systems. 2. Due to such integrated counseling medication errors as well as adverse drug events

will be minimized within a context of shared “responsibility”. Because the process enables the update of incident reporting systems in different medical canters it assists in developing reasonable medical and clinical synergies.

Physician Order Entry Validation and Endorsement A review of major studies of adverse medical events and medication errors led to estimations that in the USA between 44,000 to 98,000 people die in hospitals each year as a result of medical errors. The solutions available range from electronic patient records, order processing, clinical documentation and reporting for nursing and physicians, to computerized physician order entry in combination with rules-based decision support and electronic medication dispensing and administration solutions (Baldauf-Sobez, et al, 2003). Incorporating the Computerized Physician Order Entry (CPOE) into the backbone of the entire agent-mediated mobile information system allows both patients and healthcare institutions to gain some benefits (Mekhjian et al, 2002) such as: significant reduction in medication turn-around times, improvement in result reporting times, eliminated all physician and nursing transcription errors, and medication dispensing.

Patient Hospitalization and Recovery-Oriented Support Geriatric Hospitals address the risks of hospitalization and provide inpatient care for elders suffering from multiple co-morbidities by focusing on their special needs and geriatric syndromes. The main strategy to be adopted is to provide care by Geriatric Evaluation and Management (GEM) Units equipped with multidisciplinary teams in pursuit of improved mortality and other clinically relevant

257

Improving Clinical Practice through Mobile Medical Informatics

outcomes. Alternatively, geriatric hospitalization can be provided through comprehensive geriatric consultation service. On the other hand, Acute Care of Elders (ACE) Units incorporate the GEM unit design with additional enhancements and admit patients with acute illnesses. ACE units often provide improved flooring and lighting, reorienting devices, and other environmental improvements such as common rooms for patient use (Palmer, 1994; 1998). Within the context of the ACE unit, the GEM unit concept was enhanced by the use of more nurse-initiated protocols and a greater number of environmental and design modifications to emphasize the early assessment of risk factors for iatrogenic complications and the prevention of functional decline (Landefeld, et al 1995). The integrated agent mediated medical information system provide assistance in this regard by the development of a platform that helps in strengthening and supporting the incorporation the GEM unit into ACE units. Within this context, patients can use the components and symbols of the entire system to know directions, ask for help, get drug and nutrition information and call for assistance in case of falling etc.

can provide useful feedback about the recovery progress of paediatric inpatients.

Paediatric

a. Institutional knowledge-oriented direction: The use of mobile medical informatics not only affects medical and clinical practice but also dictates new axioms for resources sharing, medical accountability (before institutions and patients). Because the majority of healthcare institutions are public establishments their ability to benefit from the use of mobile medical informatics necessitate the conversion of such establishments into learning organizations capable of managing technology intensive acquisitions. There is also a growing need for some degree of organizational flexibility to allow for automating some of the core clinical processes, widening the hand-eye dexterity of physicians to a wider domain of consultation and

Parents used to co-patient heir children in hospitals sometimes for long periods of time. While emphasis tends to be made on clinical management, paediatric inpatients can be supported by a paediatric smart room in which patients and their co-patients (mainly parents) spend sometime entertainment time. The room can also benefit from the entire integrated system and include system-based smart toys, medication-based smart screens, as well as physical and digital tools. In addition to creating a home environment that assists in recovery, such rooms also provide some confidence for parents regarding the health status of their children. Since it operates through the backbone of the entire integrated system, it

258

Medical Training The integrated counseling environment constitutes an outstanding training environment for medical parishioners. Because of the growing number of medical and health-related students and graduates both on-study and houseman-ship training is challenged by the inability of public medical and health institutions to develop a conducive learning and training environment. Such an integrated system-based counseling environment provides an environment through which good practices can be shared.

FUTURE DIRECTION The incorporation of agent-mediated mobile medical informatics to improve clinical practice in Sudan (as well as in other developing countries) calls for paradigm shifts in the following directions:

Improving Clinical Practice through Mobile Medical Informatics

enriching all knowledge bases across the federated medical centers and applications. The use of mobile medical informatics also demands careful restructuring of hospitals to orchestrate processes throughout modified and innovative “value chains” rather than “chains of command”. b. Technology-oriented direction: The implementation of innovative agent-mediated mobile medical informatics requires more appreciation of the importance of changing technological orientations of users. While the installation of technological solutions is directly and inexorably linked to “functional” processes, it is also contingent upon the planning paradigms used to ensure system-function match. This direction emphasizes the role to be played by systems analysts in articulating roles, processes and relationships in a useful value-chain-related format that promotes informed utilization and minimize the impact of dysfunctional change agents that move the whole situation into disarray due to the “blind” jump to technological platforms. Based on the above mentioned directions, further research can be directed towards the following critical success factors: 1. Intelligent systems development and assessment with more emphasis on bridging methodological gaps associated with agent oriented software engineering when such systems are reduced to medical applications. While different agent oriented theories and methodologies are being developed and used, little has been done to address and re-engineer them to fit different domains of applications with varying degrees of functional complexity. 2. Orchestrating the qualities of agents with the core value adding medical processes and allowing rooms for both concurrent processing and flexibility

of technological platforms in order to enable the incorporation of technological (and medical) developments and inventions.

CONCLUSION The decreasing efficiency and lack of quality healthcare services in many developing countries motivates patients to consider alternative mediation and hospitalization scenarios. While public healthcare institutions are being challenged by the lack of funds and public service standard operating procedures, private ones are challenged by the growing operational costs associated with the acquisition of advanced medical equipments and specialists. The marathon of private healthcare institutions originates from the need to improve competitive edges within the healthcare system. However, the dysfunctional side effects of such race-for-competitiveness game ranged from high medication and hospitalization costs in Sudan (compared to other countries) to the institutional intension to find escape gates to avoid regulatory controls and quality assurance practices. If managed properly, the use of agent mediated mobile medical informatics provides a wide range of organizational, institutional and process-oriented functionalities that significantly contribute to improving clinical practice.

REFERENCES Baldauf-Sobez, W., Bergstrom, W. M., Meisner, M. K., Ahmad, A., & Häggström, M. (2003). How Siemens’ Computerized Physician Order Entry Helps Prevent the Human Error. Electromedica, 71(1), 1–9.

259

Improving Clinical Practice through Mobile Medical Informatics

Byung, K., & Sadeh, N. (2004). Applying casebased reasoning and multi-agent intelligent system to context-aware comparative shopping. Decision Support Systems, 37(2), 199–213.

Gasmelseid, T. (2007b). From Operational Dash Boards to EBusiness: Multiagent Formulation of Electronic Contracts. International Journal of E-Business Research, 3(3), 77–97.

Cabri, G., Leonardi, L., & Zambonelli, F. (2000). Mobile agent coordination models for internet applications. IEEE Computer, 33(2), 82–89.

Gasmelseid, T. (2007c). Engineering Multi Agent Systems. Encyclopedia of Information Security and Ethics. Information╯Science╯Reference, Idea Group Publishing, USA. ISBN-10:╯159140987X, EAN:╯9781591409878.

Chatzipapadopoulos, F., Perdikeas, M., & Venieris, L. (2000). Mobile agent and CORBA technologies in the broadband intelligent network. IEEE Communications Magazine, 38(6), 116–124. doi:10.1109/35.846081 Datta, A., & Thomas, H. (1993). The cube data model: A conceptual model and algebra for on-line analytical processing in data warehouses. Decision Support Systems, 27(3), 289–301. doi:10.1016/ S0167-9236(99)00052-4 Du, T., Li, E., & Chang, A. (2003). Mobile agents in distributed network management. Communications of the ACM, 46(7), 127–132. doi:10.1145/792704.792710 Fedele, F. (1995). Healthcare and Distributed Systems Technology. Retrieved December 23, from: www.ansa.co.uk/ANSATech/95/ansaworks-95/ hltcare.pdf Gasmelseid, T. (2006). “A Multi Agent Negotiation Framework in Resource Bounded Environments”. Proceedings of the international conference on information and communication technologies: from theory to practice. Damascus Syria, 24-28 April, 2006. Also in Information and Communication Technologies, 2006. ICTTA ‘06. 2(1), 465-470, ISBN: 0-7803-9521-2. Retrieved from http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?tp= &arnumber=1684414&isnumber =35470. Gasmelseid, T. (2007a). A Multi agent Service Oriented Modelling of e-government Initiatives. International Journal of Electronic Government Research, 3(3), 87–105.

260

Goodhue, D. L. (1995). Understanding user evaluations of information systems. Management Science, 41(12), 1827–1844. doi:10.1287/ mnsc.41.12.1827 Goodhue, D. L., & Thompso, R. L. (1995). Tasktechnology fit and individual-performance. MIS Quarterly, 19(2), 213–236. doi:10.2307/249689 Jain, A., Aparicio, M., & Singh, M. (1999). Agents for process coherence in virtual enterprises. Communications of the ACM, 42(3), 62–69. doi:10.1145/295685.295702 Keng, S., Ee-Peng, L., & Zixing, S. (2001). Mobile commerce: Promises, challenges, and research agenda. Journal of Database Management, 12(3), 4–10. Kirn, S. (n/d). Ubiquitous Healthcare: The OnkoNet Mobile Agents Architecture. Retrieved December 1, 2007 from: http://www.citeseer. Koutsoukis, N.-S., Mitra, G., & Lucas, C. (1999). Adapting on-line analytical processing for decision modeling: the interaction of information and decision technologies. Decision Support Systems, 26(1), 1–30. doi:10.1016/S0167-9236(99)000214 Landefeld, C. S., Palmer, R. M., Kresevic, D. M., Fortinsky, R. H., & Kowal, J. (1995). A randomized trial of care in a hospital medical unit especially designed to improve the functional outcomes of acutely ill older patients. The New England Journal of Medicine, (332): 1338–1344. doi:10.1056/ NEJM199505183322006

Improving Clinical Practice through Mobile Medical Informatics

Lottaz, C., Smith, C., Robert-Nicoud, Y., & Faltings, B. V. (2000). Constraint-based support for negotiation in collaborative design. Artificial Intelligence in Engineering, 14(3), 261–280. doi:10.1016/S0954-1810(00)00020-0 Luo, X., Zhang, C., & Leung, H. F. (2001). Information sharing between heterogeneous uncertain reasoning models in a multi-agent environment: a case study. International Journal of Approximate Reasoning, 27(1), 27–59. doi:10.1016/S0888613X(01)00032-9 McMullen, P. R. (2001). An ant colony optimization approach to addressing a JIT sequencing problem with multiple objectives. Artificial Intelligence in Engineering, 15(3), 309–317. doi:10.1016/S0954-1810(01)00004-8 Mekhjian, H. S., Kumar, R. R., Kuehn, L., Bentley, T. D., Teater, P., Thomas, A., Payne, B., Ahmad, A. (2002). Immediate Benefits Realized Following Implementation of Physician Order Entry at an Academic Medical Center. Journal of American Medical Information Association, (9), 529-532. Ministry of Health. Sudan. http://www.fmoh. gov.sd Monica, C.hiarini T., Remblay, A., Robert F., Uller, A., Donald, B., Erndt, B., & James, S. (2006). Doing more with more information: Changing healthcare planning with OLAP tools. Decision Support Systems. doi:.doi:10.1016/j. dss.2006.02.008 Münch, E. (2002). Teleportal-Klinik. Internal presentation, TU Ilmenau. Palmer, R. M., Counsell, S., & Landefeld, C. S. (1998). Clinical intervention trials: The ACE unit. Clinics in Geriatric Medicine, (14): 831–849. Palmer, R. M., Landefeld, C. S., Kresevic, D., & Kowal, J. (1994). A medical unit for the acute care of the elderly. Journal of the American Geriatrics Society, (42): 545–552.

Reason, J. (2000). Human error: models and management. British Medical Journal, (320): 768–770. doi:10.1136/bmj.320.7237.768 Reason, J. T. (1990). Human Error. New York: Cambridge University Press. Reason, J. T. (1997) Managing the Risks of Organizational Accidents. Ashgate Publishing. Sandholm, T., & Huai, Q. (2000). Nomad: mobile agent system for an internet-based auction house. IEEE Internet Computing, 4(2), 80–86. doi:10.1109/4236.832950 Sheth, A. P., & Larson, J. A. (1990). Federated Database Management Systems for Managing Distributed, Heterogeneous, and Autonomous Databases. ACM Computing Surveys, 22(3), 1183–1236. doi:10.1145/96602.96604 Ulieru, M. D., & Norrie, R., Kremer, & Shen, W. (2000). A multi-resolution collaborative architecture for web-centric global manufacturing. Information Sciences, 127(1–2), 3–21. doi:10.1016/ S0020-0255(00)00026-8 UNFPA website: http://www.unfpa.org Unicef website: http://www.unicef.org Wu, D. J. (2001). Software agents for knowledge management: Coordination in multi-agent supply chains and auctions. Expert Systems with Applications, 20(1), 51–64. doi:10.1016/S09574174(00)00048-8

KEY TERMS AND DEFINITIONS Clinical Practice: The activities undertaken by medical staff at the different medical areas of expertise and specialization (such as surgery, pediatrics, etc) within the organizational domains of health-related organizations and prevailing professional medical codes of ethics.

261

Improving Clinical Practice through Mobile Medical Informatics

Healthcare Control Levels: Reflect both the organizational and institutional dimensions of the healthcare system which varies from country to another. Especially in developing countries where public healthcare institutions play a significant role, three control levels tend to be used: federal, territorial and local. The efficiency of such levels is affected by the existing decision making context and technological platforms. ICU Models: The key organizational characterization used for the description, coordination, and re-engineering (if necessary) of medical processes undertaken by medical (and supporting) personnel in the intensive care unit. Medical Informatics: A term widely used to describe the use of information systems (mainly decision support systems) in medical processes and interactions. The basic aim is to improve operational efficiency of medical and health centers, enhance clinical practice and the quality of medical care and promote good practices.

Mobile Agents: Are software programs that use their “mobility” qualities to room across networks in order to access information and carry out tasks for their own processes, on behalf of their owners and/or other agents. Multiagent Systems: A cluster of (homogenous or heterogeneous) software agents possessing diverse agency qualities and attributes orchestrated in an organization structure-alike context to share resources and achieve objectives within a predefined (center-specific) or universal (internet-based) processing environment. Users: All stakeholders and partners who are in direct (main and subordinate) interaction with the provision of health care services (such as physicians, pharmacists, etc) (a.k.a affecters), third party partners (such as suppliers) (a.k.a facilitators) and patients who are directly affected by the quality of medical and clinical processes (a.k.a affected).

This work was previously published in Handbook of Research in Mobile Business, Second Edition: Technical, Methodological and Social Perspectives, edited by Bhuvan Unhelkar, pp. 604-614, copyright 2009 by Information Science Reference (an imprint of IGI Global).

262

263

Chapter 2.2

A Web-Enabled, Mobile Intelligent Information Technology Architecture for On-Demand and Mass Customized Markets M. Ghiassi Santa Clara University, USA C. Spera Zipidy, Inc., USA

ABSTRACT This chapter presents a web-enabled, intelligent agent-based information system model to support on-demand and mass customized markets. The authors present a distributed, real-time, Javabased, mobile information system that interfaces with firms’ existing IT infrastructures, follows a build-to-order production strategy, and integrates order-entry with supply chain, manufacturing, and product delivery systems. The model provides DOI: 10.4018/978-1-60960-561-2.ch202

end-to-end visibility across the entire operation and supply chain, allows for a collaborative and synchronized production system, and supports an event-based manufacturing environment. The system introduces four general purpose intelligent agents to support the entire on-demand and mass customization processes. The adoption of this approach by a semiconductor manufacturing firm resulted in reductions in product lead time (by half), buffer inventory (from five to two weeks), and manual transactions (by 80%). Application of this approach to a leading automotive manu-

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

A Web-Enabled, Mobile Intelligent Information Technology Architecture

facturer, using simulated data, resulted in a 51% total inventory reduction while increasing plant utilization by 30%. Adoption of this architecture by a pharmaceutical firm resulted in improving accuracy of trial completion estimates from 74% to 82% for clinical trials resulting in reduced trial cost overruns. These results verify that the successful adoption of this system can reduce inventory and logistics costs, improve delivery performance, increase manufacturing facilities utilization, and provide a higher overall profitability.

INTRODUCTION The globalization of businesses and the infusion of information technology (IT) into every aspect of operations have introduced a strong demand for product variety and transformed business environments from a production-centric model to one that is information and customer-centric (Arjmand & Roach, 1999). Although the Internet has strengthened business with its convenience and 24x7global accessibility, it has also dramatically shifted the traditional business model to a new, competitive market space. People can now purchase anything, anywhere, at any time, and both product customization and customer requirements are increasing exponentially, making sales and inventory prediction a challenge. Meeting the wants and needs of such a heterogeneous customer population, in a global market, inevitably calls for product variety, while every efficiency-seeking supply chain prefers to process as few “flavors” as possible. Mass customization seeks an economical resolution of this fundamental conflict. Taking mass production as a foil implies that a mass customized product should not cost end customers much more than a mass produced near-equivalent, and that the customization process should not create too much of a delay. We believe that this can be realized with consistency and at scale only with a customercentric production system. This is one that enables

264

an end-customer to configure (partially design) the product online and provides real-time visibility of the resulting order directly to the manufacturing floor and throughout the supply chain. In such a production system, businesses focus on their core competencies and outsource activities that are not essential to this core. Improvements in information technology infrastructures and worldwide acceptance of the internet have strengthened this transition. As a result, complex products in the market can be the result of collaborative efforts of many companies (Anderson & Lee, 1999). The auto industry is an excellent example of such a collaborative effort. A car can have over 10,000 parts, with multiple stages of production, many suppliers, high degree of product complexity, and high degree of customization. The manufacturing operation of such a business often requires a high production rate, time and space constraints and often long cycle time. High technology is another example. Fabrication-less firms that design new components are common. These firms now concentrate on their core business of designing a new component and outsource the manufacturing to specialized semiconductor and PC board manufacturing contractors. Transportation and logistics systems are additional examples in which the Internet and online commerce have facilitated rapid movements of products and material in a time-sensitive production environment. Pharmaceutical companies are yet another example in which trials for new drugs are often conducted globally and concurrently and can significantly benefit from an on-demand, real-time production control environment. The participants in these markets include suppliers, retailers and transportation services providers. The efficient operation of such markets requires extensive collaboration among its many members. There are several common themes that characterize these markets. The first theme is the timesensitive nature of the demand in such markets. The order stream for these markets can change

A Web-Enabled, Mobile Intelligent Information Technology Architecture

in a short period of time. For example, for the semiconductor manufacturing system described later in this chapter, the order stream can arrive multiple times per day creating a turbulent production environment requiring adjustments to production schedules. Similarly, transportation and delivery systems need to account for last minute changes in orders, cancellation of existing orders, addition of new orders, break downs in the actual transportation facilities and complications due to weather or traffic conditions, all within just a few hours (Dorer & Calisti, 2005). Most mass customized production environments are time-sensitive and therefore exhibit such behavior. Traditional production systems cannot efficiently address these needs. The second theme associated with such markets is the complexity of the supply chain system. The auto industry is an example of such a production environment. The supply chain is often multilayered with complex arrangements. Supporting mass customization of products in these markets can have a major impact on inventory levels for the suppliers that are located upstream from the final production vendor. If timely demand data reflecting the latest change in final product is not available to all suppliers, the inventory bullwhip effect may require upstream suppliers to stock up on the wrong components. The coordination requirements for an efficient operation of such a system are often lacking in traditional production systems. The third theme present in such markets is the adaptive requirements of such operations. Consider clinical trials for new drugs. The supply chain supporting such operations needs to adapt rapidly to reflect intermediate results from several ongoing clinical trials. As results become available, the supply chain must rapidly adapt itself to support and/or shift operations from one trial site to another and to offer complete visibility in real-time and on-demand. These themes, associated with on-demand and mass customized markets, clearly introduce addi-

tional complexity into the production environment. For these markets, optimal production solutions tend to be order/demand driven, are sensitive to each order quantity, and should account for more frequent changes in order streams. Obtaining optimality in such a production environment requires continuous analysis in very short time intervals. Effective analysis in this environment requires visibility of the changing demand data to all participants in the production system in real-time. The firms participating in such a system often follow a short-term optimization approach to production problems that require coordination and collaborations among local decision makers. The decision making for this environment is often: decentralized and distributed; geographically dispersed; more data driven with less human intervention; benefits from a rule based automated negotiation system; attempts to reach local optimal solutions for the participating firms; and performs final coordination and adjustment of these solutions in a short time interval. In contrast, the traditional production models rely on a centralized decision making process that attempts to optimize a global, central production function representing the primary vendor, to address a longer term production system. We note that agent-based systems allow timely actions and provide optimal response when the rule structure – under which agents act – does not change. However when externalities – i.e. elements not modeled in the system – become effective and change the structure of the decision process, then human interaction becomes crucial to re-establish the optimality of the decision process. For on-demand and mass customized production environments exhibiting these attributes, an agent-based information system solution can offer improved performance. We present three case studies that show how the implementation of such a system has improved the productivity and profitability for a semiconductor manufacturing company, an automotive firm, and a pharmaceutical company. Actual and simulated results show that for the semiconductor firm, the implementa-

265

A Web-Enabled, Mobile Intelligent Information Technology Architecture

tion of our model reduced product lead time by 50%, reduced buffer inventory from five to two weeks, and reduced manual transaction processing by 80%. For the automotive case, our results, using simulated data, show a total inventory reduction of 51%, and an increase in production facilities utilization rates by more than 30%. Results from adoption of this approach has improved accuracy of trial completion estimates from 74% to 82% for clinical trials while reducing trial cost overruns. Other researchers report similar improvements in cost reduction (by 11.7%), reduced traveled kilometers (by 4.2%), and fewer trucks employed (by 25.5%) in a transportation system utilizing an agent-based transportation optimization system (Dorer & Calisi, 2005), and a 15% increase in revenue with improved profit margin from 2% to 4% in a logistics provider after adoption of an agent-based adaptive transportation network (Capgemini, 2004).

and reorder replenishment inventory cycle under this model will be more frequent, involve smaller lot sizes, and require shorter delivery schedules. Such a synchronized production process will necessitate demand, manufacturing and production information transparency and greater cooperation among the participating members, from the manufacturer to the primary and secondary suppliers (Ghiassi, 2001). Truly successful members of such a manufacturing environment must have stronger alliances and be willing to significantly improve their inter-firm communications. In addition, there is a need for an infrastructure that can:

ON-DEMAND AND MASS CUSTOMIZATION BUSINESS ENVIRONMENT

•

The business environment for successful implementation of on-demand and mass customized markets, thus, requires the establishment of business alliances and partnerships with selected suppliers. Mass customized manufacturing systems need to support a “demand–driven” and “make-to-order” production system. Such a production system is often component-based and may involve many business partners, suppliers, and other companies that affect the delivery of products to customers (Gilmore & Pine, 2000). Outsourcing exacerbates the challenge of coordination and planning of any such production system. Supporting a mass customized production system, therefore, requires a supply chain paradigm capable of a great degree of synchronization throughout the entire supply chain, including the entire inventory system. In particular, the order

266

• • •

• •

Reduce time-to-market for product development, enhancement, and customization. Provide end-to-end visibility along the entire supply chain. Directly tie order-entry and manufacturing planning systems to speed the availability of demand requirements and to react in real-time. Intelligently and selectively communicate with a manufacturer’s strategic trading partners. Respond expediently to orders, changes in order configuration, and level of demand. Support an event-based supply chain management system in a collaborative environment.

Supporting such a production system requires a collaborative, adaptive, and synchronized supply chain management system that spans multiple firms, is event-driven, distributed, and can operate in real-time and across many computer platforms (Ghiassi & Spera, 2003a). Such a synchronized supply chain operation no longer follows a linear model. This system must be a network-based operation that requires timely availability of information throughout the system in order to allow cooperative and synchronized flow of material, products, and information among all participants. The manufacturing environment for this system

A Web-Enabled, Mobile Intelligent Information Technology Architecture

requires an operational structure and organization that can focus on building and maintaining the capacity to adapt quickly to a changing order stream while minimizing response time. Clearly, the decision making process in this environment needs to shift from a centralized to distributed mode. In this model, operational and strategic decisions may still be driven centrally; however, the responsibility to adapt to a volatile condition needs to be resolved locally. The effects of a response to such events, then, must be communicated to all relevant layers of the organization, including participating suppliers. Obviously, the relationship among the supply chain members of such a production system also must be altered to support this new business paradigm. Transitioning from mass production to mass customization is a complex, enterprise-wide process which affects external supply chain members, which are not typically controlled by the core company, and thus necessitating alliances and collaborations among all supply chain members.

ON-DEMAND AND MASS CUSTOMIZATION IT INFRASTRUCTURE Implementing such a system would require reliance on an interoperable open system IT structure, a high degree of automation in the decision process, and integration of customers, products and production information across many participants with possibly heterogeneous IT infrastructures. The technology infrastructure for an ondemand and mass customized production system requires a collaborative and adaptive supply chain management system that can account for the entire system - the customer, manufacturers, entire supply chain and the supporting market structure (Ghiassi & Spera, 2003b). An effective infrastructure for such a vast domain requires an interoperable open system capable of operating in a multi-firm, heterogeneous IT infrastructure.

In this chapter, we present an IT framework and architecture to support on-demand and masscustomized production system. We report on prototype solutions that have helped firms achieve their mass-customized production strategies. We present an architecture that can manage production resources, perform negotiations, achieve supply chain information transparency, and monitor and manage processes. We use an intelligent agent technology that can tie together resources, parts and process knowledge, in single or multiple locations, to create an agile production system. The system defines “brokers and infomediaries” and uses an intelligent, mobile, agent-based trading system that can be utilized in an interactive mode that can react to job streams with customized orders while bringing together buyers and suppliers in an efficient production environment. We have developed scheduling algorithms to balance the “time” vs. “price” trade-off by examining the relations between manufacturers and their suppliers and subcontractors. A multi-agent technology is used to schedule, control and monitor manufacturing activities and facilitates real-time collaboration among design teams. Agents in this environment can provide information about the product development process, availability of resources, commitments to deliver tasks, and knowledge of expected product completion in a distributed platform. These mobile agents are used to reach across organizational boundaries to extract knowledge from resources in order to reconfigure, coordinate, and collaborate with other agents in solving enterprise problems. The agents can interface with a customer, a manufacturing resource, a database, or other agents anywhere in the network in seconds, enabling the supply chain system to continuously adapt to demand fluctuations and changing events. Agents in such an environment can facilitate a decentralized and bottom-up planning and decision making process. Our IT architecture supports an easily accessible (web-based) order and production management system where customers can:

267

A Web-Enabled, Mobile Intelligent Information Technology Architecture

•

Configure a product according to their preferences. Monitor progress and obtain status reports. Manage changes and gain overall visibility into production.

• •

In this system the firms can: • •

•

Execute and support a demand driven synchronized supply chain system. Select suppliers for non-strategic commodity components using mobile intelligent agents. Employ intelligent agent technology to support the human stakeholders for strategic sourcing in an expedient manner.

We describe a distributed, Java-based, eventdriven system that uses mobile agent technologies over the Internet to monitor multiple production systems, sense exceptions as they occur and dispatch intelligent agents to analyze broken links and offer alternative solutions. The system is designed to improve the coordination and control of the total supply chain network, through the real-time optimization of material movement. This network encompasses all entities of the supply chain, and builds a demand recognition capability that is as close to the demand source as possible, and then uses this information to drive and synchronize the replenishment of all interim raw materials, components and products. The availability of this system enables all participants to monitor, detect, predict, and intelligently resolve supply chain problems in real-time. Participants can obtain up-to-date, customized visibility into other members of the marketplace and can measure historical supply chain performance activities. This information can lead to a manufacturing operation that supports on-demand or “make-to-order” strategy and avoids inventory bloat. In a mature implementation, the supply chain operation can be transformed into a “demand

268

chain” operation, in which inventory is reduced to its minimal level throughout the entire system.

OVERVIEW OF THE MODEL By providing a common communication infrastructure along with intelligent agents and optimization algorithms, we offer a system for transforming static supply chains into dynamic, web-centric trading markets that are closely integrated with the manufacturer’s systems. This system formulates the production problem as a set of distributed, decentralized, event-driven optimization sub-problems that can efficiently support an on-demand and mass customized production environment, which are often characterized by high demand variability, fixed capacity, time to market sensitivity, and/or complex supply chain network. This approach models the supply chain problem as a hierarchical framework and extends it to support a synchronized production system. The framework presented in this chapter partitions the mass customization production environment into hierarchical levels of process, plan, and execution. The framework offers facilities for process monitoring and analysis, short and long term planning capabilities, and efficient, optimized execution of production opportunities. It also provides visibility throughout the system by allowing changes in orders to be communicated to all impacted participants. Each participant in turn reacts to these changes and broadcasts its locally optimized solutions. In the next level of the hierarchy, the local optimum solutions are coordinated and synchronized and discrepancies are broadcasted again to all affected parties. In the final decision making level of the hierarchy, automated and human assisted negotiating agents resolve all production conflicts in support of the final committed production. The responsiveness required for such highly synchronized production systems is significantly higher than those of non-

A Web-Enabled, Mobile Intelligent Information Technology Architecture

synchronized systems. This requirement necessitates real-time collaboration among all supply chain members. The supporting IT infrastructure and applications presented in this model are able to cope with high demand variability, short product life cycles and tight collaboration and interactions with suppliers. In this model, information about the entire supply chain system is visible to all authorized members, especially through any device that is Internet enabled. Our system allows the up-to-date demand data to drive the production system following a “make-to-order” production policy. In solving the production allocation sub-problems, the up-to-date demand data is used to monitor production allocation at various plants. The fluidity of the demand undoubtedly can create many exceptions to the centrally designed production plan. Decentralized decision-making can allow local players to alter execution plans to satisfy their constraints. To manage such scenarios, a system must be able to detect an exception or unplanned event, identify the players impacted by the exception and its resolution, and develop the models necessary to solve the local sub-problem and to resolve the exception. To be fully effective, such a decentralized decision-making environment requires real-time data running across an intelligent peer-to-peer infrastructure and integrated with contextual understanding of the local production sub-problems. In the past years, many researchers have examined using “agent technology” to find solutions to problems arising in supporting masscustomized production. (Sundermeyer, 2001 and Wilke, 2002) present an agent-based, collaborative supply chain system for the automotive industry (DaimlerChrysler). These authors identify rapidly changing consumer demands, planning uncertainties, information flow delays, and sequential, batch oriented policies as elements of traditional production systems that contribute to an inefficient production environment. The authors offer an agent-based, web-centric, collaborative solution

that integrates the entire supply chain and provides a more efficient production system. They report results from a simulation study which show that using a collaborative, distributed, agent-based supply network management system can reduce production and inventory oscillations, safety stock levels, and working capital. Similarly, (Dorer & Calisti, 2005) report on the integration of an agent-based system into a real-world IT architecture in order to solve and optimize transportation problems facing a major logistics provider. They note that “conventional systems for transportation optimization are limited in their ability to cope with the complexity, and especially with the dynamics, of global transportation business where plans have to be adjusted to new, changing conditions within shortest time frames.” In the following, we describe a Java based software system that uses intelligent agents to constitute a theoretical and practical basis for addressing mass customization markets.

Agent Taxonomy We define a mobile intelligent agent to be an autonomous software program, capable of achieving goals defined by a set of rules, with learning and adaptive capabilities that are loosely or tightly coupled with other peers. Software agents are often designed to perform a task or activity on behalf of the user and without constant human intervention (Ma, 1999; Adam et al., 1999). Existing literature lists applications for intelligent agents in a wide array of electronic businesses, including retailing, purchasing, and automated negotiation (Jain et al., 1999; Ma, 1999; Maes et al., 1999; Sandholm, 1999). These software tools are used as mediators that act on behalf of their users. Specifically, these agents perform repetitive and predictable actions, can run continuously without human intervention and are often stationary. The authors in (Singh et al., 2005) present an agent-based model that uses E-marketplaces and

269

A Web-Enabled, Mobile Intelligent Information Technology Architecture

infomediaries to register suppliers and consumers of an E-supply chain. This system facilitates purchasing of material by collecting demands from the participants, then locating and matching this demand with suppliers. The proposed system develops supplier profiles based on a satisfaction survey of past performances. When extended to multiple E-marketplaces, a global balancing of supply and demand becomes viable. This system, however, does not include the actual manufacturing operations and relies heavily on the existence of E-marketplaces, infomediaries, and the willingness of all participants to register their demands, capacity and their services. Researchers have also applied intelligent agent technology to manufacturing and production systems (Kalakota et al., 1998), and mass customized production environments (Baker et al., 1999). Others have used agent technology to support what they have termed “adaptive virtual enterprises” (Pancerella & Berry, 1999). They have used mobile agents to extend across organizational boundaries. In their model, these agents extract knowledge from resources in order to reconfigure, coordinate, and collaborate with other agents in solving enterprise problems. There are many mobile agent systems that provide a platform for agent applications; most of these systems are Java-based. Some examples include “Aglets” from IBM, “Concordia” from Mitsubishi, and “Voyager” from ObjectSpace (Lang & Oshima, 1998). We have applied agent technology to ondemand and mass customized production problems. We use mobile software agents to support a distributed decision-making process. A mobile agent, unlike a stationary agent, is not bound to one execution environment. A mobile agent can transport itself and its supporting data to other execution environments. It can then begin execution or use data from both sites to perform its tasks. The mobility also allows the agent to interact with other agents or traditional software systems on the new site. Mobile agents addition-

270

ally offer the opportunity for execution support across heterogeneous computing environments. They can reconfigure themselves to interface with destination environment’s libraries, database systems, and other computing components to perform their tasks. Mobile agents introduce a new paradigm to distributed computing. In this paradigm, any host computer in the network can benefit from the “know-how”, resources, and processors throughout the network. We introduce an architecture that deploys mobile agents in the value chain to streamline the efficiency and performance of the entire network under various constraints of materials and time. The agents perform the following primary functions: 1. Distribute and collect data at the right time to process and analyze. 2. Coordinate inter-enterprise business processes and propose the best response to avoid disruptions throughout the supply chain. 3. Tactically and strategically adjust the supply network in an optimal and timely manner to changes in market conditions. 4. Assist the decision-makers in their learning processes. The four functionalities (monitoring, analyzing, acting, and coordinating) characterize the first dimension of our agent taxonomy, the second being given by the role that each agent assumes in the value chain. We have identified three types of roles: 1. Consumer Assistant. 2. Demand Management. 3. Supply and Resource Management. The agents performing these roles are mobile and ubiquitous. These agents address the needs of end users, manage product demands, and coordinate supply and resource management continuously. The time-sensitive nature of an

A Web-Enabled, Mobile Intelligent Information Technology Architecture

on-demand production environment requires the decisions to be made expediently. The mobility of the agents allows the communication between them to occur in real-time. For instance, consider the auto industry case discussed in this chapter. The consumer assistant agent is used by the end user to define product features, place an order, check order processing, and monitor production status of the product all the way through actual delivery. Similarly, in a transportation/package delivery system, the consumer assistant agent can be used to even alter the actual delivery location of a package as the recipient relocates before the final routing of the scheduled delivery of the product. The demand and supply management agents can also perform their roles in real-time and assist decision makers in resolving conflicts and coordinating operations. In particular, conflict resolutions caused by supply or capacity shortages can be handled more expediently. The mobility of the agents and their real-time connectivity within the production system and with the actual end user is utilized to alter the production schedule and inventory levels. These agents can be pre-programmed

to handle some decisions automatically to conduct some negotiations on behalf of their clients. For example, when a product feature is in short supply, the demand management agent can notify the consumer assistant agent of possible delays in delivery and to offer alternative resolutions. At the same time, this agent can negotiate with supply management agents to locate new sourcing of components that are hindering timely production of the final product. Additionally, the mobility of the agents allows managers and consumers to participate in decision making in real-time, using the Internet or even hand-held devices. The technology used to implement these agents and their communication protocols are presented in next section. The complete Functionality versus Role determining our agent taxonomy is reported below. Agent goals are reported in the corresponding cell. The entire system uses an event-based architecture (Figure 1). An event in this system is defined as actionable information (an object) that appears as a message in the system. The monitoring agents are registered to listen to certain mes-

Table 1. Agent taxonomy and functionality Functionality vs. Role

Consumer Assistant

Demand Management

Supply & Resource Management

Monitoring

Detect variations of any relevant (for the consumer) product features.

Detect key demand variable changes at product model and location level.

Detect key supply variable changes or resource constraints at product model and location level.

Coordinating

Bring order and semantic structure to the collection of demand and supply events gathered from multiple sources by the mobile assistant.

Bring order and semantic structure to the responses for the mix of demand events that are firing synchronously from multiple sources.

Bring order and semantic structure to the responses for the mix of supply events that are firing synchronously from multiple sources.

Analyzing

Rationalize the follow on actions for comparative analysis based on features and customers’ preferences.

Rationalize the follow on actions from excess inventory projections, sub-optimal minimum inventory levels, networked replenishment requirements and multiple shipping alternatives and constraints.

Rationalize the follow on actions from excess/shortage inventory projections, sub-optimal minimum inventory levels, networked replenishment requirements and multiple assembling shipping alternatives and production constraints.

Acting

Adjust customers’ decisions as their preference changes.

Adjust projected demand through mix of techniques: Optimize base parameters, modify elasticity, adjust for cannibalization and influence demand through dynamic pricing.

Adjust projected supply and production rates through mix of techniques: Optimize base parameters, modify elasticity, adjust price and optimize supply through tight collaboration.

271

A Web-Enabled, Mobile Intelligent Information Technology Architecture

sages. These messages are interpreted and processed either synchronously (i.e., as soon as they arrive), or asynchronously (pending the agent’s reaching of some pre-defined state). An event bus is introduced as a building block that receives, sends and processes messages as objects. The event bus offers an interface that is visible to any component of the system that wishes to send and/ or receive events (Figure 1). The agents presented in this architecture behave dynamically, are rule-based, mobile and respond to events. The behavior of the agents can be in reaction to both internal and external events. For instance, consider an inventory management scenario; the “analyze” and “act” agents can use parameters defined by other agents (monitoring and/or coordinating agents) to determine inventory replenishment policies that reflect up-to-date information about market conditions or even macroeconomic indicators, in addition to specific product parameters such as available sources, lead time information, the supplier’s recent behavior profile, and the latest speed of inventory deliveries. The supporting agents

(monitoring and coordinating agents) can monitor internal and external events and processes to acquire and coordinate the necessary data that allows the analyzing and acting agents to manage inventory levels dynamically rather than the traditional static models that use predetermined inventory replenishment parameters. Finally, in production environments for which E-marketplaces exist, this architecture allows the agents to seamlessly interface with these markets to execute procurement activities. When E-marketplaces do not exist, the agents can additionally perform market-making activities, such as identifying candidate suppliers, communicating requirements, and soliciting quotations.

AGENT SERVICE ARCHITECTURE FOR MASS CUSTOMIZATION We present an architectural design which is agent-based, supports mass customization, and uses a “cluster model” representation as depicted in Figure 1. This architecture supports the firm’s

Figure 1. An agent-based architecture for mass customization

272

A Web-Enabled, Mobile Intelligent Information Technology Architecture

IT infrastructure, and interfaces with existing DBMS, ERP, and other manufacturing support database systems. The architecture uses existing IT networking systems to establish connectivity across participating firms in the marketplace. Clusters of agents can support multiple product lines and/or even different features of a product. The main features of the above design are: •

Agents Technology and Architecture: ◦⊦ Agents are developed using a framework based on the ABLE (Agent Building and Learning Environment from IBM) (Bigus et al., 2002) and JADE-LEAP (Java Agent Development Framework and its extension Light Extensible Agent Platform) (Bellifemine et al., 2003, Bergenti & Poggi, 2001) platforms. The multi-agent system introduced in this architecture uses the ACL language (Agent Communication Language) (ACL, 2006) which complies with FIPA (Foundation for Intelligent Physical Agents) specifications. The architecture employed is distributed, and supports multiple hosts and processes, including mobile systems. Each running instance of the JADE runtime environment is called a “container.” A container can include multiple agents (Figure 2). A set of active containers is called a JADE platform which can be distributed across several hosts. The distribution can be from one platform to many hosts and/or many platforms to many hosts. The inter-agent and intra-agent communications occur via ACL messaging. Each agent is represented by one java thread. ◦⊦ Agents are registered in JNDI (Java Naming and Directory Interface) for clustering support.

◦⊦

•

•

•

Agents communicate asynchronously through the event bus amongst themselves, and can interface with external services (web services, application services and external agents) synchronously. ◦⊦ Agents learn and acquire intelligence through optimized algorithms per agent functionality. Control Panel: ◦⊦ The control panel allows users to create a new agent on demand and register it with the event bus and the JNDI. ◦⊦ The control panel allows users to modify agent parameters and thus modify the behavior of existing agents. Communication Interface: ◦⊦ The event bus structure is introduced and developed to facilitate fast internal communications among agents. ◦⊦ The JMS (Java Message Services) services are introduced to facilitate communications among agents and external applications including web services interface, WMA (Windows Media Audio) services, agents from other applications, external users and the web environment at large. Persistence Layer: ◦⊦ The persistence layer assists agents in interfacing with existing databases and data retrieval systems.

These features offer the necessary elements for supporting visibility across the entire supply chain and allowing the establishment of collaborative efforts in support of a demand driven, mass customized production system. To illustrate these concepts, consider an agent-based system implementation for a manufacturing environment. Figure 2 presents the implementation of this technology for a hypothetical manufacturing system. In this system, agents are implemented as Java

273

A Web-Enabled, Mobile Intelligent Information Technology Architecture

container classes that perform the Monitoring, Coordinating, Analyzing and Acting functionalities. The server side of Figure 2 lists five agent containers. The client side shows how users can use mobile phones, PDAs and similar devices to gain access to the system. In Figure 2, the “I/O Container” offers monitoring agents that communicate with data entry systems and provide interface capabilities with existing DBMS, ERPs, and manufacturing databases. Agents from this container communicate and exchange information with the central coordinating agents. Similarly, the “Inference Container,” an analyzing agent, stores information about the manufacturing operation for all possible product configurations. It continuously analyzes production schedules, capabilities, and equipment capacities. It also analyzes manufacturing requirements based on production status and the order stream information for new orders to form a batch and to hold them until notified by another agent (the optimizing agent) to release its information. The communications among these agents is in a synchronized mode. The “Optimizing Container” is an acting agent that is invoked upon receiving a new batch of orders from the agents of the “Inference Container.” These agents optimize the processing of the orders. The optimization process includes (a) scheduling optimization using tabu search and genetic algorithms, (b) resource planning using linear and integer programming, and (c) inventory optimization using stochastic optimization methods. The “Back-end Splits Container” serves as the interface channel to the client side. The agents in this container receive and record orders from external agents. The orders are filtered and organized according to their manufacturing requirements and are aggregated and presented to the main container for further processing and optimization. The customers and/ or external agents can use this interface to check the status of their orders. Finally, the “Main Container” serves as the coordinating agent. It interfaces with all other agents, provides supervisory

274

functions to all agents, performs synchronization among processes, and serves as conflict resolution authority among agents. In this architecture, a persistence layer is provided for interfacing with the external database and data retrieval systems. The entire system is Java based and uses the agent technology framework developed by IBM (the “ABLE” technology) (Bigus et al., 2002) and JADE-LEAP framework from the open source software (Bellifemine et al., 2003; Bergenti & Poggi, 2001) and the ACL messaging language. These frameworks are extended and modified to support the mobile technologies that will allow users access to the system via mobile phones, PDAs, and any web-enabled device or interface (Moreno et al., 2005). The system presented supports the decentralized, distributed decision making model discussed earlier. It is based on a distributed, web-centric, client server architecture in which the server resides at the manufacturing (primary) vendor and the client operates at each supplier site. The clients’ implementations are lightly integrated into the suppliers’ existing IT architectures. The webcentric nature of the system ensures accessibility across the participating members as depicted in Figure 2. The consistency of the system is ensured by the choice of the communications and the agent frameworks used in this implementation. The use of standard protocols such as XML and the collaborative nature of the environment require each participant to either directly comply with these protocols, or provide converters and translators in support of the communication systems. An implementation of such system may include many geographical servers at various locations. Client installations can also be located at geographically and/or organizationally dispersed sites. Real-time production and consumption data is gathered from the entire supply chain system and distributed throughout the system for decision making (Figure 2).

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Figure 2. An agent-based architecture for a manufacturing system

CASE STUDIES In this section we will discuss case studies that report on adoption and implementation of our approach by three companies that produce ondemand or mass customized products. The first case study reports on adoption of this system in a semiconductor manufacturing environment, the second case study presents our experience with implementation of the system in the automotive industry, and the third case study reports use of the system by a pharmaceutical company.

The ST Microelectronics System The first case presents a semiconductor manufacturing firm (STM) that has adopted and implemented our software system. STM is the fourth largest semiconductor company in the world with revenue in excess of eight billion dollars. A large part of this revenue comes from integrated circuits that are heavily customized to STM’s customer specifics.

STM is the primary supplier of assembled chip sets to a number of mobile telephone providers. The mobile telephone companies offer a mass customized product line with multiple features (GPS, Camera, and Phone capabilities). There are a number of valid product configurations for consumers to choose. Orders from the phone companies to STM for the various chip sets often arrive daily or even twice per day. STM’s production system needs to respond to demands on daily or even shift by shift basis. The order/ reorder processing system, defined by the demand at STM centers, requires agile manufacturing facilities capable of supporting a demand-driven supply chain system in real-time. The architecture described here, and its implementation at STM enables the company to operate and manage a manufacturing system far more efficiently than older, batch driven, build-to-forecast alternatives. In July of 2000, STM adopted our software for their mass customized production system. The server side of the software system was installed at STM manufacturing facilities and the clients’ side

275

A Web-Enabled, Mobile Intelligent Information Technology Architecture

at customers (the telephone manufacturers) and STM suppliers. This implementation has enabled STM to develop collaborative programs with its top customers and has seen notable progress in its delivery schedule, lead time, inventory management, and profit margins. An important requirement of the project was the ability of STM to connect its IT infrastructure to its customers’ existing installed IT systems using mobile intelligent agents. Our software system, sitting on top of STM’s existing planning platform, detects the RosettaNet (RosettaNet, 1998) and XML messages, and provides visibility into different inventory scenarios, such as what parts to build and where to hold inventory to meet the highly customized demand. The system offers a set of agent-based coordination tools for extracting and transmitting data across the entire IT infrastructure. This strategy has offered STM a proactive approach for supporting its mass customization strategy. It has allowed STM to connect its long-range plans with short term execution and to optimize the inventory levels and production schedules in order to quickly build the integrated circuits that meet its customers’ demands. STM supplies IC boards (assembled chip sets) to several mobile phone providers. The boards need to be customized according to each customer specification. The challenges facing STM in supporting their customers are to: • •

•

•

276

Develop a long-term planning framework for process and systems improvements. Implement new demand planning, E-commerce communications, capacity planning, and efficient supply chain execution. Roll out E-commerce with supply chain partners, perform end-to-end supply chain planning, and implement an effective available-to-production (ATP) capability. Continue to follow a build-to-forecast (BTF) policy in wafer fabrication and

•

•

probe to maintain a pre-defined level of inventory in the die bank. Implement a build-to-order (BTO) policy in the assembly and test phase of the IC manufacturing. Develop a financial model to trade-off inventory, manufacturing, and supply chain costs with revenue potentials.

STM adopted the agent-based adaptive supply chain software system described earlier to achieve these objectives. The system is presented in Figure 3 and offers a collaborative supply chain network, supports adaptive demand synchronization and adjustment, provides an adaptive inventory optimization and replenishment policy, and optimizes order, material and flow management systems. In Figure 3 the event management system models supply networks, monitors network input, and triggers events. The agents in the system provide demand synchronization, network inventory optimization, multi-echelon inventory replenishment, delivery and routing planning, and coordination of the final decision resolution. Each of these components includes multiple intelligent agents that perform the monitoring, coordinating, analyzing, and acting functions described earlier. The embedded decision support tools of the system offer performance visibility of suppliers through collection of rankings and metrics, and provides a library of optimization algorithms and learning models. In Table 2 we provide a comparison of the demand management activities of STM under both the traditional approach and the new agentbased adaptive approach. Similarly, Table 3 presents the traditional and the new approach for inventory optimization module. The adaptive replenishment collaboration process, its traditional implementation, and its new implementation are presented in Table 4. Finally, the routing process used to manage order and reordering of the IC boards, material

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Figure 3. STM adaptive supply chain system

Table 2. Demand management activities for STM Traditional Approach Forecast sharing is not traditional, but if occurs, follows a passive process like the following: 1. Semiconductor supplier receives forecasts and firm orders from Distributor and OEM on a weekly/daily basis by ship-to location and SKU. 2. Preprocess data to adjust for standard customer request dates, and account for uncertainty. 3. New forecast and firm order data are updated in demand planning system. 4. Supply chain planners use new demand data to reevaluate and adjust production plans.

Agent-Based Demand Management Approach Agent-based adaptive solution is driven by demand changes that impact as far upstream the supply network as possible. 1. Semiconductor supplier receives forecast or actual sales data from Distributors and OEMs on a daily/weekly basis. This will depend on cycle for forecast updates. If actual sales are given, then forecast from Distributor and OEM is also provided as it changes. 2. Semiconductor supplier receives changes to firm orders from OEMs and Distributors on a daily basis 3. Event management system triggers a “New Demand” event whenever a new forecast or firm order is received into the system. 4. MONITOR agents are listening for any demand event and will apply the appropriate business rules to adjust the input streams and trigger the required ANALYZE agents to check for any trending anomalies. 5. ANALYZE agents perform the following actions: €€€€€€€€€€a. ANALYZE “Order Synchronize” is designed to detect inconsistencies with transition from forecasted to firm orders. €€€€€€€€€€b. ANALYZE “Volatility” is designed to check for excessive volatility between successive forecasts. €€€€€€€€€€c. ANALYZE “Forecast Error” is designed to check for significant error between forecast and semiconductor shipments. €€€€€€€€€€d. ANALYZE “POS Sales” is designed to check for significant error between customer forecast and daily/weekly sales from source location in supply network. In this case that is the OEM’s and Distributor’s. 6. ANALYZE agents will coordinate with ACT agents to engage demand synchronization actions €€€€€€€€€€a. ACT “Rationalize Error” agents deduce root cause actions from analysis results. €€€€€€€€€€b. ACT “Optimize Parameters” agents adjust forecasting model parameters to minimize forecast error. €€€€€€€€€€c. ACT “Forecast” agents create and recommend new forecasts.

277

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Table 3. Inventory management activities for STM Traditional Approach 1. Inventory performance is reviewed on a periodic basis. 2. Viewed from a single enterprise perspective. 3. Review of safety stock levels based on expected demand and variability, general service level targets, lead-times, lead-time variability.

Agent-Based Inventory Management Approach 1. Inventory management is driven from a network perspective. 2. Event management monitors demand trending, demand error trend, lead time trending, supply error trending, inventory level trend and service level performance. If there is a threshold violation, an event is created. 3. MONITOR agents deduce the semantic value of the mix of events and will trigger the appropriate ANALYZE agents. 4. The ANALYZE agents will focus on the following: €€€€€€€€€€a. ANALYZE “Stock Up” agents deduce where and why average inventory is increasing. €€€€€€€€€€b. ANALYZE “Stock Out” agents deduce where and why stock-outs are excessive. €€€€€€€€€€c. ANALYZE “Min Buffer” agents compute the minimum for all key buffers in network. 5. ANALYZE agents will coordinate with ACT agents to engage inventory optimization actions. €€€€€€€€€€a. ACT “Balance Network” agents will be used to adjust inventory positions and levels via ACT “Replenishment” agents to adjust for both inventory shortage and excess scenarios. €€€€€€€€€€b. ACT “Set Level” agents make the appropriate safety stock parameter adjustments at all buffers.

Table 4. Replenishment management activities for STM Traditional Approach 1. Long-term forecasts are created by customers for up to 3 months lead-time. 2. Forecast is frozen and used to negotiate for capacity from Semiconductor company. 3. Semiconductor company incurs significant manual process costs to continually match capacity to expected customer demand. 4. Semiconductor company is required to keep high levels of inventory to protect against high variability of demand. 5. Customers replenish inventory by continually placing orders with Semiconductor company.

Agent-Based Replenishment Approach 1. Updated demand plan is provided for target replenishment customers on a daily basis. 2. Event management system monitors the demand plan changes and inventory buffer changes. If a threshold limit is violated an event is created. 3. MONITOR agents capture such events and engage the appropriate ANALYZE agent. 4. The ANALYZE agents are of the following two types: €€€€€€€€€€a. ANALYZE “Buffer” agents compute the threshold values for all of the demand buffers for a new demand plan. €€€€€€€€€€b. ANALYZE “Shortage” agents evaluate the shortage situation when inventory is low. 5. ACT agents optimize replenishment over a targeted horizon in order to minimize costs and realize targeted service levels established by Semiconductor supplier and all OEM/ Distributor customers. €€€€€€€€€€a. ACT “Replenishment” agents create an optimized (min cost flow) replenishment plan for all items over supply/demand map and within targeted plan horizon. €€€€€€€€€€b. ACT “Shipment” agents coordinate with “Replenishment” agents to create and trigger shipments, with the goal of minimizing shipment costs over target shipping horizon.

flow, WIP, and movement of supplies throughout the supply chain and manufacturing plants are optimized to ensure timely availability of supplies and fulfillment and delivery of customer orders. Table 5 presents the new routing processes employed and compares it with the traditional approach used by STM. Implementation of the agent-based adaptive supply chain management system by STM has resulted in supply chain simplification, logistic process and provider improvements, and better delivery performance. Table 6 summarizes these improvements. 278

The agent-based system presented offers an effective infrastructure that enables and encourages updated demand data to be distributed among members of the production environment on daily or even shift-by-shift bases. In the traditional system, demand data was often only updated weekly. Clearly, accurate information offered by our solution allows decision makers to optimize their operations throughout the system. The distributed nature of this system allows STM to validate orders for consistency by further monitoring and analysis of demand and manufacturing data down the stream, up to the OEMs and dis-

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Table 5. Order routing process management activities for STM Traditional “Commit” Approach Whereas replenishment is used to support the indirect sales process, the “Order Commit” process supports direct sales. 1. Orders are entered and booked with a temporary Scheduled Ship Date and are localized in multiple offices. 2. Within 24 hours the centralized Sales Planning team determines the initial commit date. 3. Sales orders are continuously revisited to bring the Schedule Date closer to the Customer Request Date. 4. Management of sales order rescheduling is usually very manual, with limited ability for managing changes and exceptions.

Agent-Based Routing Optimization Adaptive Order Routing dynamically adjusts the slotting of orders to fulfillment routes that best meet the sales, service level and supply chain cost goals across a network. 1. Event management system is designed to monitor, new orders, inventory, customer request date changes, goods in transit. Significant changes trigger events. 2. MONITOR agents recognize such events and trigger the appropriate ANALYZE agents. 3. ANALYZE agents interpret changes and recommend the best actions: €€€€€€€€€€a. ANALYZE “Order” agents use order prioritization and current supply/ demand state to correct order slotting. €€€€€€€€€€b. ANALYZE “Supply” agents use change prioritization and current supply/demand state to correct order slotting response. 4. ACT agents are engaged by ANALYZE agents to continuously drive order schedule dates to the customer request date: €€€€€€€€€€a. ACT “Simple Slotting” uses heuristics to slot or change an order to respond to a critical supply event. €€€€€€€€€€b. ACT “Allocation Optimization” optimizes order slotting, given the user selectable weightings for sales, service levels and cost or profit goals subject to available material and capacity constraints.

Table 6. Overall improvements for STM manufacturing processes Process

Results

Supply Chain Simplification

1. Non-value-added movement of wafers or packaged part is reduced. 2. Redefined the testing process to increase offshore testing. 3. Reduced end-to-end cycle time by 8 to 10 days.

Logistic Improvements

1. Identified root cause of long transit between Asia and the US. 2. Changed sourcing activities from multiple freight forwards to single 3PL (third party logistic providers). 3. Reduced inter-company door-to-door cycle times from 7 days to 2 days. 4. Reduced total logistic costs by $2M/year.

Delivery Performance

1. Identified the root causes of the poor delivery performance. 2. Used inventory monitoring and optimization for allocating supply in capacity-constrained environments. 3. Improved delivery performance to first commit by more than 35% within five months on pilot product line.

tributors of its customers’ (phone companies) products. Since adoption and implementation of this strategy, STM has cut its product lead time in half, reduced buffer inventory from five to two weeks, and eliminated 80% of its manual transactions (Shah, 2002).

The Automotive Example: The XYZ Pick-up Truck Today, American, European, Japanese and other automakers are fiercely competing among each

other by giving consumers a unique level of choices in configuring their model. This extreme level of mass customization for a market that until few years ago was producing vehicles with very few choices has introduced numerous challenges on how automakers have aligned and adjusted their supply chain. In the following we report on the strategy and the solution adopted by one of the leading car manufacturers in addressing the challenges related to customization of one of the components of the pickup truck: the engine block. This customization introduces a high degree of variability in the cylinder machining operation at 279

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Engine Plant A, the coating process at the Coating Plant B, and the block casting process at the Casting Plant C. The mass customization process can often result in excessive inventory, which is not tolerable financially. The objectives of the automaker were: 1. To increase the responsiveness to high customized demand. 2. To reduce the inventory level in the supply chain. 3. To improve control of managing the overall material flow to the target performance levels. We first summarize the inventory objectives and production features of each plant and report the “as is” supply chain from B to A and C to B in Table 7, and depict them in Figures 4 and 5 respectively. To show the effect of mass customization to the supply chain we consider the customization of the engine. There are two sets of choices for customers: engine size in Liters (4.6, 5.4, and 6.8) and the number of cylinders (8 and 10). The only engine with 10 cylinders is the 6.8L V10. Even this simple set of choices can have a great impact on the production lead time. In particular, each time production shifts from V8 to V10 the line

becomes idle for some period of time in order to set up the new configuration. Similarly, each time there is a change of configuration from 4.6L to 5.4L or 6.8L the set up needs to be reconfigured and the line becomes idle. The inability to proactively forecast and shape customers’ demand results in: • • • • • •

Unscheduled downtime at the block production line Inability of block line to consistently support engine assembly at Plant A Excessive inventory carrying costs Transportation inefficiencies Unplanned overtime at all the plants Scheduling instability

To support this degree of mass customization, we have applied the “agent-based adaptive supply chain model” presented in this chapter to the manufacturing processes at the three plants. Since the auto industry’s supply chain is complex and often includes many tiers, the agent-based solution provided in this model can make changes in the demand data to be simultaneously visible throughout the system. The propagation of this up-to-date information will allow production and inventory level adjustments for all participants, including the entire supply chain members. This

Table 7. Plant inventory objectives and production features Location

Inventory Objectives

Plant A

1. Supply material needs for engine assembly, which is driven by end demand needs. 2. Buffer against short supply from Plant B for the very near-term JIT windows.

1. Unstable machine capability. 2. Low schedule stability.

Plant B

1. Supply scheduled material needs (JIT schedule) at Plant A. 2. Supply unexpected material needs at Plant A, which may be caused by end demand shift, high scrap rate or material consumption, overtime production, etc. 3. Buffer against short supply from Plant C, which may be caused by high changeover rate, unexpected machine downtime at Plant C, etc.

1. Batch production. 2. High demand uncertainty from Plant A. 3. High supply uncertainty from Plant C.

Plant C

1. Supply scheduled material needs (JIT schedule) at Plant A. 2. Buffer against short material supply from internal or external suppliers.

1. Long changeover time. 2. Batch production. 3. Inflexible product switch.

280

Production Features

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Figure 4. Supply chain model from Plant B to Plant A

Figure 5. Supply chain model from Plant C to Plant B

model minimizes downtime, reduces inventory throughout the production system, and offers a more efficient transportation system while achieving a more stable scheduling and delivering option. This model is presented in Figure 6. There are seven components of this adaptive model:

1. The Event Management: The Event Management Framework is a foundational capability of the Adaptive Supply Chain solution. Its primary purpose is to monitor events and orchestrate the adaptive response to the significant event changes that occur in the supply chain. An event is any supply chain variable that represents some key dynamic in

281

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Figure 6. The adaptive supply chain model for engine assembly

the operation of the supply chain process, e.g. demand or inventory level. Throughout the supply chain a change in an event can trigger one of two possible paths. If the change is considered within the normal variability, it will trigger the Changeover Optimization Engine, Pull Replenishment Engine or both. If the change is outside a preset tolerance limit it will trigger an exception and notify the appropriate user. 2 – 3. Optimal Changeover Sequence and Target Inventory: The Optimal Changeover and Target Inventory (OCTI) is an intelligent engine that will calculate both the optimal sequence of product runs and target inventory buffer levels for any series of product changeover constrained operations in the supply chain. The engine uses a mixed integer programming model to minimize the total supply chain costs subject to demand, inventory, and transportation and production constraints. The OCTI engine is used to calculate both the optimal block sequence for each of the supply chain operations and

282

adjust the associated min/max buffer levels. The supply chain system uses a push/pull control strategy for coordinating the material flow, with a pull replenishment buffer established at the ready to ship buffer in Coating. The OCTI results are used as input to both Casting and Coating stages to drive the build schedules that are used to push the engine blocks through Coating to the pull buffer. The min and max levels of the interim buffers are adjusted as needed. On the Plant A OP 10 side, the OCTI results are used as input to the OP 10 pull scheduling process to develop production runs that best support the actual engine assembly usage and the calculated optimal block sequence. When the event of the OP 10 input buffers reaches the reorder levels, it will trigger the Pull Replenishment engine to pull in blocks from the Coating replenishment buffer 4. Pull Replenishment: The Pull Replenishment is an intelligent replenishment engine designed

A Web-Enabled, Mobile Intelligent Information Technology Architecture

to trigger and control the “Pull” replenishment from any two entities in the supply chain. It also has a look-ahead capability, driven by forecasted demand that will assess the synchronization viability of the source supply. Plant A supply chain uses the Pull Replenishment engine to coordinate the pull signals between the OP 10 input buffer and the Coating ready to ship buffer. 5. Demand Visibility: The demand forecasting module offers intelligent agents that are invoked by the event management component. These agents use the longer term demand values and parameters as guidelines and receive up-todate demand information from the actual customer base. The agents revise and smooth the order stream using the longer term demand values and parameters. They use the current production status, buffer values, and other production data to determine final manufacturing orders for both the Casting and Coating processes. 6. Resolve Exceptions: Exceptions are defined to be violations of production values from manufacturing thresholds. In this system, thresholds have dynamic values and are determined by the demand levels. Once an exception is reached, i.e., a threshold is violated, the conditions need to be restored to normality and full or near optimality. Reaching a restoration level is attempted while keeping the noise level in the entire manufacturing environment to a minimum and without

causing major interruption to the system. For example, a generated exception might be allowed to proceed, and thus missing fulfilling a portion of the demand portfolio in the short term, in order to honor longer term production goals and constraints. 7. Measure Performance: This module uses intelligent agents to monitor key performances of the system through the event management component. It can measure performance of any event, both internal and external, by collecting statistics dynamically. Examples of internal events include inventory buffer levels, number of changeovers, production line throughput, and total inventory in the pipeline. Similarly, customer satisfaction and supplier performance, external events, can also be monitored and measured by this module. Implementation of this system has used several features of the model, described in this chapter, in the engine manufacturing process. The data requirement of the system necessitated an information technology infrastructure based on the agent technology for decision making, JDBC for data storage, and XML for interfacing and data access. The concept of synchronization was used to ensure sequencing of the steps. The system allowed for interaction with auto dealers, partners, and suppliers. This visibility brought demand information and fluctuation to the manufacturing floor and broadcasted manufacturing status, alerts, and exceptions to the affected targets. Finally, a graphical user interface was produced to allow authorized users to view, input, or edit the entire system information.

283

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Prior to the implementation of the system, a simulation model was developed to assess the effectiveness of the proposed system and compare its performance to the existing one. The objectives of the simulation were to measure inventory levels, changeover time, and utilization of the manufacturing facilities for given demand values. The demand information represented a dynamic stream of customized orders and reorders recorded daily over a 90 day period. The objective function used represented total production costs while allowing for switching among the different configurations. The constraint set included the various buffer limitations for inventory, changeover capacity and its opportunity cost, maximum throughput capacity, and equipment availability and maintenance constraints. The results of the simulation showed improved performance as measured by the two key indices (inventory levels and utilization rates) for the three plants. Table 8 shows that total inventory level for the system was reduced from 27,000 units to 13,283 units (a 51% reduction). The utilization rate for plant A increased from 60% to 90% and the combined utilization rate for plants B and C was improved from 41% to 61%. These improvements have increased the automaker’s responsiveness and its competitive edge in meeting customers’ demand.

The Pharmaceutical Company Example: Clinical Trials Development of a new drug is a complex process that is difficult to manage, and includes a lengthy Table 8. Simulation results for the engine manufacturing plant Models

Total Inventory

Utilization (Plant A)

Utilization (Plants B & C)

Existing System

27,000

60%

41%

Proposed System

13,283

90%

61%

284

cycle lasting often years or even decades. These projects have several variables, many stakeholders with differing interests, and are closely scrutinized because of the urgent need to bring new treatments to market. Study managers and their teams try to control a process that is largely outside of their sphere of influence and scope of control. A typical drug development cycle begins with laboratory research and animal experimentation (pre-clinical phase) and it is followed by five testing phases. Phase zero trials are also known as the “human micro-dosing studies” and are designed to speed up the development of promising drugs. Phase one of the trials usually involves few volunteers (less than 50). Phase two of testing involves larger numbers of volunteer patients with advanced diseases that might benefit from such a new treatment. This is often affiliated with a hospital and a team of research medical doctors. Phase three, the most costly and time consuming phase, involves a much larger number of patients (between 300 -3000) and must follow the FDA (or similar regulatory bodies) guidelines. The results of this phase are submitted to the FDA for final approval. Once the approved drug reaches the market, development enters phase four of the cycle. This phase addresses any anomalies or unforeseen adverse effects that might become present. This research focuses on phase three. As mentioned earlier, this phase is the most elaborate, costly, and time consuming phase. In today’s global market, most international pharmaceutical companies prefer to conduct this phase over geographically dispersed locations often expanding many countries. Clearly, conducting such clinical trials over multiple sites, spreading over many locations can significantly benefit from existence of an advanced IT infrastructure that will enable coordination, monitoring, analyzing, and acting in a timely manner. Adoption of the architecture described in this paper by pharmaceutical firms meets the requirements of this phase of the clinical trials and is described below.

A Web-Enabled, Mobile Intelligent Information Technology Architecture

A typical phase three of a clinical trial consists of the following steps. •

•

•

•

•

•

Since phase three of the clinical trial requires meeting regulatory guidelines, the first step is the approval of the study protocols by the FDA. This approval can range from medical considerations to approval of the sample size for the clinical trials such as the number of patients to be recruited for the study. The next step involves selection of the sites for the trial including location of hospitals, clinics, and doctors’ offices. Once the total number of patients and sites are defined, allocation of patients to sites is determined. In the next step the actual patient recruitments by the local health sites is attempted. The recruitment follows two stages: in stage one, patients are screened for appropriateness of the study. The second stage is the randomization of patients who have successfully passed stage one. In this stage a patient is either assigned to a treatment at a site or is assigned to a placebo group. Once groups are designed, patient monitoring begins. For clinical trials patients must follow a pre-defined visitation and monitoring plan. During this step necessary data are collected and transmitted from the investigators’ sites to a central location in the pharmaceutical company for analysis. Today’s clinical trials follow an advanced approach called “multi-arms.” In this approach a drug may contain many ingredients or components. Trials are designed to test multiple combination of dosage for each version of the drug. This approach results in many configurations of the medicine, allowing the pharmaceutical company to simultaneously evaluate the most effective combinations. Therefore, at each site multiple independent dosages

•

•

(versions) are being evaluated. Patients are assigned to each drug version following a uniform distribution. The prescribed dosage of the drug for each version is administered to members of the group following the pre-defined procedures for the pre-defined period of the treatment. An added complexity of these trials is the globally distributed nature of the trial. Clearly, logistics, management, coordination and visibility of the entire system requires an information architecture that enables all participants to have real-time access to all the necessary information. Once trials have begun, partial results are analyzed to determine whether a particular dosage is a promising one or not. When there is enough evidence that a particular dosage is not effective, that specific configuration (version) is stopped and patients from this group are re-assigned to begin taking other dosages from the beginning cycle of the new dosage. The process continues until the pre-determined trial cycles for all dosages (versions) are complete.

Managing the Supply Chain for distributed Clinical Trials The supply chain management for distributed clinical trials is very complex and includes many steps. Recent advances in medical science have increased introduction of new drugs. However, developments cycles have become more complex, exclusive patent timelines have shrunk, and costs have skyrocketed. Companies look for ways to reduce these costs and to find an answer to the “costs more, takes longer” problem. Monitoring clinical trial cycles, managing the supply chains, and accurately predicting the end date for recruitment are of great interest to pharmaceutical companies. We first discuss a pharmaceutical company’s traditional supply chain system for distributed clinical trials and then present a

285

A Web-Enabled, Mobile Intelligent Information Technology Architecture

new approach based on the architecture outlined in this research and show how adoption of this new paradigm has reduced cost of the trials and shortened the completion time. The current landscape for distributed clinical trials for the pharmaceutical firm in this study is presented in Figure 7. The process begins with identification and acquisition of raw material, processing, packaging, delivery, administering, investigation, and drug accounting. Existing system has used forecasting techniques, inventory control, Manufacturing Resource Planning (MRP), Intelligent Voice Recognition Systems (IVRS), and Electronic Data Communications (EDC) systems. The traditional approach follows the “maketo-stock” production policy. In this approach once the dosages (versions) are defined, the entire demand for all versions and all sites are produced in batches and are shipped to the sites. Clearly, in this approach when the trials reveal a dosage not to be promising the associated dosage and its supporting material and clinical and support staff are no longer useful and must be discarded or disbanded. The costs associated with the non-

promising dosages are significant and the firm cannot reuse the dosage for other versions. Once “multi-arms” and globalization of the clinical trials are added, complexity of the system increases. Managing this complex system can significantly benefit from visibility, monitoring, coordinating, analyzing, and acting features of the architecture described earlier. One of the most important factors driving the cost and length of phase three of the clinical trial is recruitment time. Recruitment time is defined as the time needed to recruit a pre-determined number of patients for the entire study. In a multi-arms system, the recruitment problem is significantly more complex. The study managers must recruit sufficient number of patients for each dosage of the drug. The traditional approach uses a batch or “make-to-stock” policy, in which the numbers of sites, number of patients per site and per dosage (version) are first defined and then the supply of drug for each site is produced, packaged, and shipped to each site. As depicted in Figure 7, raw materials are converted into compounds or drug ingredients. A dosage or version uses a pre-defined portion of each ingredient per its recipe. In the traditional

Figure 7. The traditional supply chain for clinical trials

286

A Web-Enabled, Mobile Intelligent Information Technology Architecture

drug production approach, once the number of patients per site and per version is determined, a complete set of dosage for the recipients is produced, packaged, and distributed to each site. Additionally, clinical and support staff as well as instructions, guidelines, and administering procedures are defined and organized for each version proportionally. This approach is neither cost effective nor timely. The process begins with patient arrivals and administration of specific dosage and monitoring of the patients. The traditional approach does not provide a timely support structure for immediate analysis of the results and immediate termination of the less promising dosages. This in turn limits re-use of the ingredients and re-assignment of patients from failed dosages to the ongoing promising ones thus lengthening the entire process or even missing crucial deadlines. Cost overruns for this system are observed in all stages of the process, from wasted dosages (for the less promising versions) to costs associated with recruitment and administration of the trials. We next describe a system that follows the “on-demand” or “make-to-order” policy that with its supporting IT infrastructure can reduce the cost of the trials while simultaneously shortens the trail length.

the company to respond to changes in trial environment in the short term without compromising long term (strategic) goals. This architecture uses agent-based technology to support a distributed decision making environment. The system ensures optimal service to the patients, offers an integrated network, provides visibility into patient recruitment and treatment, and coordinates recruitment and trial program rate forecasts with drug and treatment availability at local trial sites. Figure 8 presents the overall optimal supply chain design based on this paradigm. In this paradigm, a plan consists of a set of activities and processes are event driven. We acknowledge recruitment strategies as a driving force for the entire clinical trial. The recruiting strategy has been a subject of many studies. Literature in this field offer simulation and stochastic modeling as the two most widely used approaches for managing recruitment policies. Patel, et al. (2009) offer simulation modeling and tools for managing drug supply chains. Anisimov & Fedorov (2007) recommend stochastic solutions for recruitment management of a clinical trial. We use a stochastic modeling approach to dynamically manage patient recruitment for the trials. The primary objective is to complete the recruitment of patients within the defined time of the study. Specifically, project objectives are:

An Adaptive, Agile, Agent-Based, Intelligent Supply Chain System

•

We present a new paradigm to manage supply chain for distributed clinical trials and show how our solution architecture can reduce cost of trials and shorten time to market. The paradigm applies the concepts of visibility, monitoring, coordinating, analyzing, and acting presented in this architecture to create an adaptive environment for trial management. The agent technology introduced earlier is combined with web services technology and the RFID technology to track and trace the elements of the trials (dosage, packages, team status, etc.). The system offers agility that allows

•

To enroll the required number of subjects within the specified timeframe, and To manage to the allocated budget for the study.

To achieve these goals, study managers must be able to quickly monitor changes, analyze the new information, decide on corrective course of actions and act expediently. We have introduced intelligent agents that use pre-defined algorithms as a means for providing the needed insights for making intelligent decisions. Essentially, study managers want the algorithms to define:

287

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Figure 8. Optimized supply chain system for clinical trials

•

•

The distribution of productive sites, showing when they become active and when they are forecast to screen their first patients, and The distribution of screened and randomized/enrolled patients.

These two interdependent time series are distributions of stochastic variables over a time horizon and therefore represent two interlinked “stochastic processes.” The ability to accurately predict these stochastic processes is the key for a study manager to define and maintain a predictable study plan. Many existing analytical tools for such systems have focused on the use of simulation to address this problem (Patel et al., 2009). The system described in this research, however, provides an intelligent user interface with predictive capabilities that eliminates the need for users to deeply understand the framework, yet still apply it when developing plans and managing recruitment. Our solution effectively visualizes these two interdependent time series across all phases of the enrollment process and provides proactive alerts for projected violations of the

288

trial’s business goals. We have examined both deterministic and stochastic model frameworks. A deterministic model framework assumes the patient arrival rate that follows a known pattern or estimated deterministic linear or non-linear function, whereas a stochastic model framework allows the definition of probability functions with expected mean values and variance. Our results show that the stochastic framework delivers superior results. Under the stochastic model framework the user is able to define instances of the model by specifying the recruitment problem parameters (expected initial recruitment estimates, by countries and across centers, as well as the rules for recruitment: competitive, balanced, or restricted recruitment). The advantage of this approach is that the model framework remains invariant and multiple instances can be generated and tested. The instanced model can now be solved using an “efficient” algorithm. Efficiency is characterized mainly by two factors: time and accuracy of the result. Because our stochastic model framework allows us to define algorithms that provide closed form solutions, accuracy and time are optimal and outperform existing simulation techniques utilized in traditional models.

A Web-Enabled, Mobile Intelligent Information Technology Architecture

In modelling the recruitment problem we consider the following: •

•

•

•

Uncertainties during the recruitment period. There are uncertainties (unpredictable events) in patient recruitment over the time horizon. Because of this, recruitment rates generally fluctuate over time and centers may delay screening patients due to unexpected events; Expected recruitment mean and confidence boundaries. To correctly model the uncertainties outlined above, we characterize the expected recruitment average for mean and its confidence boundary value. This will allow the algorithm to generate the expected number of randomized patients and its confidence bounds over the defined time horizon; The minimum number of centers required. The model determines the minimum number of centers needed to complete recruitment by a certain date at a given confidence. Estimated future performance of centers. Information about the future initiation of centers and estimated performance is factored in to improve the significance of the prediction.

The above characteristics are the driving assumptions of the model. Literature on patient recruitment (Senn, 1998, Anisimov & Fedorov, 2007, Cox, 1955) offer guidelines for patient arrivals and assignments to the centers. We follow the same two general assumptions: •

•

Given the stochastic fluctuations of recruitment over time, patients arrive at centers following the distribution of a Poisson process. Variations in recruitment rates between centers are estimated as samples from a gamma distributed population.

Additionally, we add some extensions to these assumptions. They are: •

• •

The minimum number of centers required to complete the recruitment by the target date. Knowledge of the information about future performance of centers. Capacity of the centers and their constraints (facility constraints, maximum number of patients).

We use these assumptions to extend the Cox stochastic forecasting model to develop a specific algorithm to determine the distribution of the patient arrival rate for each center. Figure 8 presents the proposed approach based on the outlined architecture to manage the supply chain for such clinical trials, including the multi-arms approached used in this study and uses the extended algorithms for its patient arrival forecasting. Results from adoption of this approach show significant improvements in cost and time to completion. The savings are the consequences of the “visibility, monitoring, coordinating, analyzing, and acting” features of the system. Specifically, the de-centralized framework of the system and the intelligent agents combine to realize the improvements. This architecture uses a de-centralized production policy. In this approach, instead of the finished drug dosage used in the traditional system, ingredients (drug components) and recipes are sent to each site and the actual dosages are produced on an on-demand or “make-to-order” bases at each site and upon patient arrivals. In this strategy once a dosage is deemed non-promising, the production of that dosage immediately stops and the remaining ingredients (drug components) are re-used for more promising dosages using their recipes. Clearly, in this de-centralized production system, expensive drug components are not wasted thus reducing the overall costs of the trial. Similarly, the predictive algorithms used in this study form the intelligence of the system allowing

289

A Web-Enabled, Mobile Intelligent Information Technology Architecture

the study managers to forecast events accurately, specially the patient recruitments events, and manage the entire process more efficiently. These features have resulted in faster decision making, allowing less promising trials to end sooner and re-assignment of patients and clinical and support staff to more promising cases, thus, completing the trials faster. The production, packaging, shipment, and administration of the dosages are based on on-demand or “make-to-order” policy, are done more expediently, and result in cost reductions and faster completion time of the trials.

Study Results The pharmaceutical company referenced in this study leverages a store of previously completed clinical studies—classified by therapeutic area and organized according to features (such as target number of subjects, number of sites and cycle time)—to test and refine the implementation of the model and its associated algorithms. Studies are sorted by category values from High (+), Medium (O), to Low (-) in order to better understand and analyze possible pattern behaviors, similarities, and differences. To arrive at the results presented in the examples outlined here, a “post-mortem” analysis was conducted to compare our results with the traditional approach. In the traditional approach, before the trials begin, the study manager offers his projection for the Last Subject Randomized (LSR) based on his past experiences. This input is a very rough estimate and is grounded on the manager’s intuition and earlier similar studies. His estimates are used for budgetary purposes, design of the trial framework, and to define system’s parameters. These parameters consists of (i) the number of countries in the study, (ii) the number of centers per country, and (iii) the total recruitment per center. and are used to create the initial recruitment plan. Clearly, these rough estimates need to be revised and corrected as soon as real data from the field becomes available. In

290

the traditional approach, once the process begins and some information from the field is collected, the LSR projections are revised. In this approach, the revision is deterministic and is triggered when 25% of the sites are operational and are actively recruiting. The study manager uses the field data to revise its LSR projections. Therefore, in a post-mortem analysis, the projection error in this approach can be computed by subtracting the LSR projections @ 25% from the actual completion date. We will use this error rate to compare with our corresponding error. We introduced a stochastic approach to compute the LSR projections. To be able to compare our results with the traditional approach, we also choose the 25% site activation metric as our trigger point. The projected LSR values based on this approach, on the other hand, uses the stochastic process described in this section. Therefore, in a post-mortem analysis, the projection error can be computed by subtracting the LSR projections @ 25% from the actual completion date based on the stochastic algorithm used in this approach. The improvement in the process can be computed by comparing these two error values. Table 9 describes the study features of three sample studies, followed by the results demonstrated by our methodology. The results presented below are illustrated through the use of two performance functions: “Absolute Time Error” and “Relative Time Error.” The “Absolute Time Error” measures the differTable 9. Study features of three trials Study Features

Study 1: Oncology

Study 2: Immunology

Study 3: Respiratory

Number of patients

O/-

+

+

Number of centers

O

-/O

+

Estimated recruitment rate

-

+

+

Expected # of active centers

-

O

O

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Table 10. Oncology study targeting 680 patients Study Start

Projected Last Subject Randomized

Planned First Site Initiated

Planned Last Subject Randomized

Deterministic Model – 25% Sites Active

Stochastic Model – 25% Sites Active

Actual – Study Completed

3-Jan2004

3-Jan2006

10-Jun2007

8-Jan2007

1-Feb2007

ence between the model estimates at the 25% trigger point and the actual end day the last patient was enrolled. The “Relative Time Error” measures the per cent error relative to model estimates at the 25% trigger point. The simplicity of these metrics provides for an immediate normalized comparison of our results across different studies as well as an immediately measurable understanding of the tangible economic impact of proper and accurate forecasting. The actual recruitment completion date for the oncology study was 1-Feb-2007, which was 129 days before the originally forecasted plan (based on the traditional approach). Using the stochastic model described in this research, at 25% centers active, the projections would have been within 23 (8-Jan-2007) days of the actual completion time. The improvement rate of the stochastic algorithm over the traditional deterministic approach is 82%. The actual recruitment completion date for the immunology study was 8-Jan-2007, which was Table 11. Immunology study targeting 2000 patients Study Start

Projected Last Subject Randomized

27 days later than the originally forecasted plan (based on the traditional approach). Using the stochastic model described in this research, at 25% centers active, the projections would have been within 6 days. The improvement rate of the stochastic algorithm over the traditional deterministic approach is 78%. The actual recruitment completion date for the respiratory study was 3-Aug-2007, which was 127 days later than the originally forecasted plan (based on the traditional approach). Using the stochastic model described in this research, at 25% centers active, the projections would have been within 33 days. The improvement rate of the stochastic algorithm over the traditional deterministic approach is 74%. The web-enabled, intelligent, agent-based system with specialized patient recruitment algorithms presented offers a more accurate trial progress forecast. Better accuracy improves and impacts the entire process, from dosage production, distribution, delivery, and administering, to drug version monitoring, information analysis, and site management. The cost savings in this system are spread throughout the supply chain and are the results of reduction in expensive dosage production (by immediate stoppage of non-promising versions) to patients’ assignment and drug administration. The stochastic model presented serves as the “intelligent component” of the agent technology. The stochastic model can determine the best and worst case scenarios for trial completion times. Table 12. Respiratory study targeting 2600 patients Study Start

Projected Last Subject Randomized

Planned First Site Initiated

Planned Last Subject Randomized

Deterministic Model – 25% Sites Active

Stochastic Model – 25% Sites Active

Actual – Study Completed

Planned First Site Initiated

Planned Last Subject Randomized

Deterministic Model – 25% Sites Active

Stochastic Model – 25% Sites Active

Actual – Study Completed

1-Jan2006

21-Nov2006

5-Feb2007

2-Jan2007

8-Jan2007

1-May2005

30-Dec2006

28-Mar2007

30-Jun2007

3-Aug2007

291

A Web-Enabled, Mobile Intelligent Information Technology Architecture

The agents can use these parameters to determine the feasibility of achieving projected goals. When these goals are deemed to be infeasible, a new trigger point is generated. A second agent can then use this trigger point and re-evaluate the LSR values using the latest available information. In the above examples, we used the 25% center activation metric as the trigger point only to allow us to compare our results with the traditional approach. In our model, the value of the trigger point is dynamically computed by the agent technology. Finally, the system presented here has been recently adopted by a pharmaceutical company and early actual results are very promising and are in line with the experimental values. Since such a system brings significant cost reduction, the savings might translate to further research for drugs targeted to specific customer needs, thus allowing the pharmaceutical company to move toward a “personalized medicine” model.

entire supply chain, allows for a collaborative and synchronized production system, and supports an event-based manufacturing environment. The system introduces an agent-based architecture with four general purpose intelligent agents to support the entire mass customization process. Experiences with implementations of this approach at a semiconductor manufacturer, an automotive company, and a pharmaceutical company show the effectiveness of the proposed architecture in enhancing the ‘velocity of execution’ of supply chain management activities, including order management, planning, manufacturing, operations, and distributions. Results verified that successful adoption of this system can reduce inventory and logistics costs, improve delivery performance, increase manufacturing facilities utilizations, and provide a higher overall profitability.

CONCLUSION

ACL. Agent Communications Language. (2006). Retrieved from http://www.fipa.org/repository/ aclspec.html.

The goal of an on-demand and mass customization strategy is to deliver customized products at costs that are near-equivalent to their mass produced versions without significant delays. To achieve this goal, we presented a web-enabled information system model that uses mobile intelligent agent technology and can be interfaced with the firms’ existing IT infrastructures. Successful implementation of any mass customized strategy requires a collaborative environment that follows a build-to-order production strategy. In such an environment, customers can configure their orders online and monitor the orders’ status at any time, suppliers can view demand stream dynamically in real-time, and the manufacturers can react to changing orders efficiently and expediently. We presented a distributed, Java-based, mobile intelligent information system model that is demand driven, supports a build-to-order policy, provides end-to-end visibility along the

292

REFERENCES

Adam, N. R., Dogramaci, O., Gangopadhyay, A., & Yesha, Y. (1999). Electronic Commerce. Upper Saddle River, NJ: Prentice Hall. Anderson, D. L., & Lee, H. (1999). The New Frontier. In Achieving Supply Chain Excellence Through Technology. Montgomery Research (pp. 12–21). Synchronized Supply Chains. Anisimov, V. V., & Fedorov, V. V. (2007). Modeling, Prediction and Adaptive Adjustment of Recruitment in Multicentre Trials. Statistics in Medicine, 26(27), 4958–4975. doi:10.1002/ sim.2956 Arjmand, M., & Roach, S. (1999). Creating Greater Customer Value by Synchronizing the Supply Chain. In Achieving Supply Chain Excellence Through Technology, Montgomery Research, 154-159, also in http://arjmand.ascet.com.

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Baker, A.D., Van Dyke Parunak, H., & Kutluhan, E. (1999). Agents and the Internet: Infrastructure for Mass Customization. IEEE Internet Computing, Sept.-Oct., 62-69. Bellifemine, F., Caire, G., Poggi, A., & Rimassa, G. (2003). JADE: A White Paper. Exp 3(3), 6-19. Bergenti, F. & Poggi. A. (2001). LEAP: A FIPA Platform for Handheld and Mobile Devices. In Proc. Eighth International Workshop on Agent Theories, Architectures, and Languages (ATAL2001), Seattle, WA, 303-313. Bigus, J. P., Schlosnagle, D. A., Pilgrim, J. R., Mills, W. N. III, & Diago, Y. (2002). ABLE: A Toolkit for Building Multi-agent Autonomic Systems – Agent Building and Learning Environment. IBM Systems Journal, (September): 1–19.

Ghiassi, M., & Spera, C. (2003b). A Collaborative and Adaptive Supply Chain Management System. Proceedings of the 31st International Conference on Computers and Industrial Engineering., San Francisco, Ca., 473-479. Gilmore, J. H., & Pine, B. J., II. (2000). Markets of One: Creating Customer-Unique Value Through Mass Customization. A Harvard Business Review Book. GPRS: General Packet Radio System. Retrieved from http://www.gsmworld. com/technology/gprs/index.html JADE. http://jade.cselt.it & http://jade.tilab.com Jain, A. K., Aparicio, M. IV, & Singh, M. P. (1999). Agents for Process Coherence in Virtual Enterprises. Communications of the ACM, 42(3), 62–69. doi:10.1145/295685.295702

Capgemini. (2004). A Collection of Agent Technology Pilots and Projects. Retrieved from http:// www.capgemini.com/resources/thought_leadership/putting_agents_towork/

Kalakota, R., Stallaert, J., & Whinston, A. B. (1998). Implementing Real-Time Supply Chain Optimization Systems. Global Supply Chain and Technology Management, POMS, 60-75.

Cox, D. (1955). Some Statistical Methods Connected with Series of Events (with Discussion). Journal of the Royal Statistical Society. Series B. Methodological, 17, 129–164.

Lange, D. B., & Oshima, M. (1998). Programming and Developing Java Mobile Agents with Aglets. Reading, MA: Addison-Wesley.

Dorer, K., & Calisti, M. (2005). An Adaptive Solution to Dynamic Transport Optimization. In Proc. Of the Fourth International Joint Conference on Autonomous Agents & Multi-agent Systems, AAMAS ’05, Utrecht, The Netherlands. Also at:http://www.whitestein.com/pages/downloads/ publications. Ghiassi, M. (2001). An E-Commerce Production Model for Mass Customized Market. Issues in Information Systems, 2, 106–112. Ghiassi, M., & Spera, C. (2003). a). Defining the Internet-based Supply Chain System for Mass Customized Markets. Computers & Industrial Engineering Journal, 45(1), 17–41. doi:10.1016/ S0360-8352(03)00017-2

Ma, M. (1999). Agents in E-Commerce. Communications of the ACM, 42(3), 79–80. doi:10.1145/295685.295708 Maes, P., Guttman, R. H., & Moukas, A. G. (1999). Agents that Buy and Sell. Communications of the ACM, 42(3), 81–91. doi:10.1145/295685.295716 MIDP. Mobile Information Device Profile. Retrieved from http://java.sun.com/products/midp/ index.jsp Moreno, A., Valls, A., & Viejo, A. (2005). Using JADE-LEAP to Implement Agents in Mobile Devices (Research Report 03-008, DEIM, URV). Retrieved fromhttp://www.etse.urv.es/recerca/ banzai/toni/MAS/papers.html Pancerella, A., & Berry, N. (1999). Adding Intelligent Agents to Existing EI Frameworks. IEEE Internet Computing, Sept.-Oct., 60-61. 293

A Web-Enabled, Mobile Intelligent Information Technology Architecture

Patel, N. R., & Tourtellotte, E. (2009). Drug Supply for Adaptive Trials. http://www.cytel.com/ Knowledge/stimulating_adaptive_1_08.pdf

Singh, R., Salam, A. F., & Iyer, L. (2005). Agents in E-Supply Chains. Communications of the ACM, 48(6), 109–115. doi:10.1145/1064830.1064835

RosettaNet. (1998). http://www.rosettanet.org

Sundermeyer, K. (2001). Collaborative Supply Net Management. In Baader, F., Brewka, G., & Eiter, T. (Eds.), KI:2001, Advances in Artificial Intelligence (pp. 467–470). doi:10.1007/3-54045422-5_35

Sandholm, T. (1999). Automated Negotiation. Communications of the ACM, 42(3), 84–85. doi:10.1145/295685.295866 Senn, S. (1998). Some Controversies in Planning and Analysing Multicentre Trials. Statistics in Medicine, 17(15-16), 1753–1756. doi:10.1002/(SICI)10970258(19980815/30)17:15/163.0.CO;2-X

Wilke, J. (2002). Using Agent-Based Simulation to Analyze Supply Chain Value and Performance. Supply Chain World Conference and Exhibition, New Orleans, La.

Shah, J. B. (2002). ST, HP VMI Program Hitting Its Stride. Electronics Business News (EBN), 42,http://www.ebnonline.com. This work was previously published in Interdisciplinary Advances in Adaptive and Intelligent Assistant Systems: Concepts, Techniques, Applications, and Use, edited by Gunther Kreuzberger, Aran Lunzer and Roland Kaschek, pp. 83-123, copyright 2011 by Information Science Reference (an imprint of IGI Global).

294

295

Chapter 2.3

A Framework for Information Processing in the Diagnosis of Sleep Apnea Udantha R. Abeyratne University of Queensland, Australia

INTRODUCTION Obstructive sleep apnea (OSA) is one of the most common sleep disorders. It is characterized by repetitive obstruction of the upper airways during sleep. The frequency of such events can range up to hundreds of events per sleep-hour. Full closure of the airways is termed apnea, and a partial closure is known as hypopnea. The number of apnea/hypopnea events per hour is known as the AHI-index, and is used by clinical community as a measure of the severity of OSA. DOI: 10.4018/978-1-60960-561-2.ch203

OSA, when untreated, presents as a major public health concern throughout the world. OSA patients use health facilities at twice the average rate (Delaive, Roos, Manfreda, & Kryger, 1998), causing huge pressures on national healthcare systems. OSA is associated with serious complications such as cardiovascular disease, stroke, (Barber & Quan, 2002; Kryger, 2000,), and sexual impotence. It also causes cognitive deficiencies, low IQ in children, fatigue, and accidents. Australian Sleep Association reported (ASA, 1999) that in the state of New South Wales alone 11,000–43,000 traffic accidents per year were attributable to untreated-OSA.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

A Framework for Information Processing in the Diagnosis of Sleep Apnea

OSA is a highly prevalent disease in the society. An estimated 9% of the women and 24% of the men in the U.S. population of 30 to 60 years was found to be having at least mild OSA (Young, Evans, Finn, & Palta, 1997). In Singapore, about 15% of the total population has been estimated to be at risk (Puvanendran & Goh, 1999). In a recent study in India (Udwadia, Doshi, Lonkar, & Singh, 2004), 19.5% of people coming for routine health checks were found to have at least mild OSA. The full clinical significance of OSA has only recently been understood. Partly as a result of this, the public awareness of the disease is severely lacking. Healthcare systems around the world are largely unprepared to cater to the massive number of OSA patients. This problem is especially severe in the developing world, where OSA diagnostic facilities are rare to find.

The average number of obstructive sleep apnea and hypopnea events per hour of sleep, as computed over the total sleep period, is defined as the Apnea Hypopnea Index (AHI).

BACKGROUND

Drawbacks of PSG and Possible Improvements

Definition of Sleep Apnea and Hypopnea Sleep apnea refers to a cessation of breathing at night, usually temporary in nature. The American Academy of Sleep Medicine Task Force formally defines apnea as: a. Cessation of airflow for a duration ≥10s, or b. Cessation of airflow for a duration < 10s (for at least one respiratory cycle) with an accompanying drop in blood oxygen saturation by at least 3%. Hypopnea is defined as a clear decrease (≥50%) in amplitude from base line of a valid measure of breathing (eg., airflow, air pressure) during sleep for a duration ≥10s, plus either: a. An oxygen desaturation of ≥3%, or b. An EEG-arousal (EEGA) (Flemons & Buysse, 1999).

296

The Current Standards in Apnea/Hypopnea Diagnosis The current standard of diagnosis of OSA is Polysomnography (PSG). Routine PSG requires that the patients sleep for a night in a hospital Sleep Laboratory, under video observation. In a typical PSG session, signals/parameters such as ECG, EEG, EMG, EOG, nasal/oral airflow, respiratory effort, body positions, body movements, and the blood oxygen saturation are carefully monitored. Altogether, a PSG test involves over 15 channels of measurements requiring physical contact with the patient.

At present, the hospital-based PSG test is the definitive method of diagnosis of the disease. However, it has the following drawbacks, particularly when employment as a community-screening tool is considered: 1. Poor data integrity is a common problem in routine PSG tests. Even when the test is done in the hospital, it is common to see cases of data loss (or quality deterioration), due to various reasons (eg., improper sensor contact due to electrodes/sensors coming loose, SpO2 sensor falling off, and measurement problems, such as movement artifacts). 2. PSG interpretation is a tedious task, due to the size of the data gathered, complexity of the signals, and measurement problems such as data loss. 3. PSG requires contact instrumentation; channels such as EEG/ECG/EOG/EMG

A Framework for Information Processing in the Diagnosis of Sleep Apnea

require Galvanic contact with the patient. It is especially unsuited for pediatric use. 4. PSG is not suitable for mass screening of the population. A trained medical technician is required to connect the patient to the PSG equipment, and the patient needs to be monitored overnight to avoid incurring data losses. 5. PSG is expensive; this is another factor working against mass screening uses. 6. AHI index (and a variant of it known as the respiratory disturbance index, RDI) is used as the golden clinical measure on the severity of apnea. However, AHI (or RDI) does not always correlate strongly with the symptoms of apnea as experienced by patients. There is an enormous clinical need for a simplified diagnostic instrument capable of convenient and reliable diagnosis/screening of OSA at a home setting (Flemons, 2003). Similarly, hospital-based, full PSG testing requires better measures to characterize the disease. This article explores possible solutions to both of these problems. There has been a flurry of recent activities at developing technology to address the issue of home screening of OSA. Four different classes of OSA monitors are under development (Flemons, 2003, and references therein). These devices varied from two-channel (eg., airflow and oximetry) systems (designated Type-IV), to miniaturized full-PSG (Type-I) units. Their major drawbacks are: •

Existing take-home devices have at least one sensor which requires physical contact. This makes them difficult to use by untrained persons, and cumbersome to use on pediatric populations. TYPE-IV systems, with the smallest number of sensors, suffers from the fact that oxymetry identifies oxygen saturation in blood only as a surrogate for OSA, and the absence of significant desaturation does not mean the absence of the disease (Flemons, 2003).

•

•

Presence of a medical technologist is still required, if acceptable sensitivity/specificity performance is required. High rates of data loss (up to 20% loss at home compared with 5% at sleep lab) (Flemons, 2003) (nine) have been reported when a medical technologist is not in attendance. Unattended systems have not led to high enough sensitivity/specificity levels to be used in a routine home monitoring exercise (Flemons, 2003; Portier, 2000). Type-I and II devices use channel counts from 7–18 and are difficult to use by an untrained person. The quality and the loss of data is a serious problem.

In this article, we present an instrumentation and signal processing framework addressing current problems in OSA diagnosis.

METHODS A Framework for the Noncontact Diagnosis of OSA Snoring almost always accompanies OSA, and is universally recognized as its earliest symptom (Hoffstein, 2000; Kryger, 2000; Puvanendran & Goh, 1999). Logic dictates that it should be providing us with the earliest opportunity to diagnose the disease. At present, however, quantitative analysis of snore sounds is not a practice in OSA detection. The vast potential of using snoring in the noninvasive diagnosis of OSA remains unused. In this article, we argue that the human speech and snore sounds share many similarities (see Figure 1), and biological “wetware” used for the generation processes. The upper airway acts as an acoustic filter during the production of snoring sounds, just as the vocal tract does in the case of speech sounds. Episodes of OSA, by definition, are associated with partial or full collapse of upper airways. As such, changes to the upper airways

297

A Framework for Information Processing in the Diagnosis of Sleep Apnea

Figure 1. Similarities and differences between speech and snoring: (a),(b),(c) snoring at different time scales; (d),(e),(f) speech at different time scales

brought about by this collapse should be embedded in snoring signals. Figure 2 shows a typical snore sound recorded from a patient undergoing routine PSG testing at the sleep diagnostic laboratory. The data acquisition system used a sampling rate of 44.1k samples/s, and a bit resolution of 16bits/sample. The system had a 3dB bandwidth of 20.8kHz. A typical hospital based routine PSG test runs for up to eight hours, and it is quite common to observe up to 8,000 events of snores within a recorded sound data.

Snore Sounds: A Working Definition One of the major problems towards automation is that there is no objective definition of what a “snore” is (Hoffstein, 2000). Recently, we proposed (Abeyratne, Wakwella, & Hukins, 2005) an objective definition for “snoring” independent of the sound intensity. It is based on the observation that sounds perceived as “snores” by humans are characterized by repetitive packets of energy that are responsible for creating the vibratory sound peculiar to snores (Figure 2, bottom). We call the distance between such packets as the “pitch” of

Figure 2. (Top) A snore episode (SE); (bottom) different segments in SE

298

A Framework for Information Processing in the Diagnosis of Sleep Apnea

snoring. A snoring episode (SE) will consist of a segment with pitch (“voiced-segment”), and possibly “silence” and segments without pitch (“unvoiced segment”). An inspiration-expiration cycle without any segment associated with a pitch is termed as a pure breathing episode (PB).

Mathematical Framework for Sound Analysis Snoring originates from acoustical energy released by the vibratory motions of upper airway site(s) during sleep. The airway cavity acts as an acoustical filter, and modifies the source sounds to produce snoring sounds we hear. Inspired by the source/vocal-tract system for human speech synthesis, we model a block of recorded sound s(n) as a convolution between: (i) “source signal x(n)” representing the source acoustical energy, and (ii) “Total Airway Response (TAR) h(n),” which captures the acoustical features of the upper airways as given by: s(n)=h(n)∗x(n)+b(n)

(1)

The symbol b(n) denotes background activities, x(n) is the source excitation, “*” denotes convolution, and h(n) is the TAR function. The term b(n) is due to a range of independent reasons and hence considered a Gaussian distribution independent of x(n). The nature of x(n) depends on the nature of the sound segment under consideration. For the case of a segment without any pitch (i.e, either a PB or unvoiced-snoring segment), we model the source excitation, x(n) as a white random process. The x(n) for voiced snoring segments is modeled by a pseudo-periodic pulse-train, drawing from techniques used in speech analysis. The TAR function h(n) is modeled in this article as a mixed-phase system considering that in OSA, multiple-source locations with temporal relations between eachother can be found, and thus, phaseinformation cannot be neglected in general.

It is our hypothesis that the state of the upper airways can be characterized by the pair ζ = {g{x(n)}, f{h(n)}}, where g and f represent the operation of extracting features out of the x(n) and h(n). We developed a snore segmentation algorithm (Abeyratne et al., 2005), which takes in full night sound data, and separates into snoring and pure-breathing episodes. It then further divides snore episodes into sub-categories “voiced,” “unvoiced,” and “silence” sections. The category “voiced” will then be further analyzed to estimate the pitch associated with each segment. Consider an arbitrary j-th Snoring Episode (SE) in the sound recordings. Divide the voiced segment ssv,j of the j-th Snoring Episode into Lj number of data blocks {Bjk},k =1… Lj, each of length N. Thus, at the output of the pitch-detector, each data block in the set {Bjk} is associated with a pitch period µjk. We term the series {µjk}, k=1,2, .., Lj as the Intra-snore Pitch Series for the j-th Snoring Episode. We consider the structure of the Intra-snore Pitch Series, and show that it is characterized by discontinuities, which can be used in the diagnosis of OSA. We propose a new measure called ISPJProbability to capture intra-snore jumps in pitch, via Definition-1 and Definition-2 given below: •

•

Definition-I: Suppose that in the j-th arbitrary snoring episode ssv,j there are at least q (< Nj +1) data blocks in the set {Bjk}, k =1… Nj with pitch periods µjk greater than a pitch threshold γ. Then the entire Snoring Episode j is labeled as having the feature ISPJ at level (q, γ). We call the quantity q as the “jump multiplicity.” Definition II: Define a quantity ISPJprobability Pqγ(rD) at level (q, γ.) for the signal s(n) of a length D as: Pqγ(rD) = 100 nqγ(rD)/rD%, where rD is the total number of snore episodes contained within the data of length D. The quantity nqγ(rD) is the number of episodes within D that were labeled

299

A Framework for Information Processing in the Diagnosis of Sleep Apnea

as having the feature ISPJ according to Definition-I. The snore sound based test statistic proposed in this article is Pqγ(rD), with the corresponding decision threshold symbolized by Pth. If Pqγ(rD)> Pth, the snore-based test is positive for OSA, and vice versa. For each subject in the database, a set S of ISPJ-Probabilities is defined by: S = {Pqγ(rD)| q = 1,2,3 and γ = 10,11, …25ms}

(3)

where Pqγ(rD) is calculated via (2). We use S to derive ROC curves of Pqγ(rD) at different ISPJ levels (q, γ). To draw Receiver Operating Characteristics (ROC curves, Figure 3), we need to know the “true clinical diagnosis” of the patients. The PSG based diagnosis is considered the absolute truth in this article. The diagnosis based on the test statistic Pqγ(rD)>Pth is compared to the “absolute truth,” and the nature of the decision is noted as one among (i) true positives (TP), (ii) true negatives (TN), (iii) false positives (FP), or (iv) false

negatives (FN). Sensitivity, a measure of success in detecting TPs is defined as: TP/(TP +FN)%. Specificity, a measure of success in rejecting nondiseased subjects, is defined as TN/(TN +FP)%. The ROC curve is a graph of sensitivity vs. 1-specificity. The ROC curve (Figure 3) allows us to conclude that snoring carries valuable information on the disease of OSA. The performance of the ISPJ-probability indicates the possibility of using the feature to diagnose OSA. Large-database trials are needed for further validation of results. The noncontact nature of the snore acquisition process should provide a significant advantage over competing techniques in developing a community-screening device for OSA. Results obtained in this article have been based on snorers referred to the sleep clinic for suspected OSA. For this reason, the screening utility of the proposed method needs further investigation with healthy snorers who do not display symptoms normally associated with OSA. The estimation of the TAR, the quantity h(n), can be carried out by using any established technique of blind signal identification (eg., Abeyratne, 1999). This article will not discuss the methodol-

Figure 3. ROC curves for AHIth =10, 15 and 30 at q=2 and γ = 14−17

300

A Framework for Information Processing in the Diagnosis of Sleep Apnea

ogy involved. Interested readers are referred to the article by Karunajeewa (2005).

A Framework for Arousal Analysis in Understanding and Characterizing Sleep Disorders Electrical activities measured on the scalp, as emanated from the brain, are called EEG signals. EEG is one of the most important signals in overnight PSG recordings. In assessing sleep disorders such as apnea, EEG plays an indispensable role. In the current clinical practice it is being used to study the sleep structure of a patient during sleep, mainly via determining the sleep stages. It is also used to estimate the total sleep time of a patient in computing the AHI and RDI indices. In the past, (Deegan & McNicholas, 1995; McNicholas, 1998) researchers have developed various hypotheses and experimental methods to investigate the changes in EEG signal during sleep apnea and hypopnea. Spectral analysis of the EEG has revealed that termination of apnea and hypopnea events are often associated with changes in EEG power in the δ- band of frequency (0.5 – 4Hz) (Dingli & Assimakopoulos, 2002; Svanborg, 1996). The issue of asymmetry of brain during sleep has been investigated by groups of researchers (Bolduc, 2002; Goldstein & Stoltzfus, 1972; Sekimoto, 2000). Asymmetry studies have been confined to undisturbed sleep in normal individuals. None of these studies considered the significance of apnea/hypopnea on the hemispheric correlations, or how the correlation behaved in the cases of peoples with symptoms of sleep apnea. Despite the importance of EEG in studying sleep disorders, the main index used to measure the severity of apnea, AHI (or RDI), does not use EEG, except for the total sleep time derived indirectly through EEG. EEG, however, is a high-temporal resolution window to the brain, and contains much more information than what is currently being used in clinical practice. We

hypothesize that EEG may provide one avenue to alleviate the important problem mentioned under the Background Section. Recently, the phenomenon known as the EEG Arousals (EEGA) has received attention as a possible pointer of sleep disorders. EEGA in sleep is defined as an abrupt shift in EEG frequency, lasting for 3s or more. According to the criteria developed by the American Sleep Disorder Association (ASDA), EEGA are markers of sleep disruption and should be treated as detrimental. Excessive Daytime Sleepiness (EDS) is one of the defining symptoms in OSA patients (Martin, & Wraith, 1997; McNicholas, 1998). The reason for EDS in SDB patients is the modification of sleep texture, due to reoccurring episodes of EEGA causing sleep fragmentation. In this article, we show that there exists a relationship between the EEGA and informationflow between the left and right hemispheres of the brain, as revealed via the left-right correlation of EEG. This finding may open up a new vista of research in understanding how information flows in the brain during phenomena associated with sleep (eg., EEGA, sleep spindles, and so on) and help with localizing sources of origin of some brain activities. Furthermore, considering the relationship between EEGA and dominant symptoms of OSA (ie., day time sleepiness), we believe a proper understanding of EEG will help in designing a better measure for OSA severity. Figure 4 shows (Vinayak, 2007) the spectral correlation coefficient computed between EEG signals measured from electrode pairs A1-C4 and A2-C3 (of the International 10/20 electrode system) during an overnight sleep study. In Figure 4, (f) and (g) arousal events and apnea event have been marked to help visualize their relationship with interhemispheric EEG correlation. In Figure 4(a), 4(b), 4(c), and 4(d) correlation coefficients have been computed separately in the δ,θ,α, and β spectral bands of the measured EEG. Our results showed that apnea/EEG arousal events generally lead to an increase in IHA. This

301

A Framework for Information Processing in the Diagnosis of Sleep Apnea

Figure 4. (a) Delta, (b) Theta, (c) Alpha, (d) Beta, correlation coefficient during the NREM and REM sleep, (e) dleep stages, base line indicates NREM sleep and 1st level indicates REM sleep, (f) arousal events, (g) apnea events

change is most prominent during NREM sleep affected by EEGA, particularly in θ, α, and β frequency bands. Our results strongly suggest that the information flow between the left-right hemispheres of the brain is affected by events of apnea/EEG-arousals (Vinayak, 2007).

expected to expand far beyond its current role of sleep-staging (using decades old criteria), making significant inroads into the subject of characterizing apnea. EEG will also help understand the brain mechanisms associated with the processes of sleep and sleep disturbances.

FUTURE TRENDS

CONCLUSION

With the rapidly increasing public awareness of the apnea syndrome, the demand for in-home noncontact methods for screening sleep apnea are expected to rise in the future. Noncontact measurement techniques based on snore (and breathing) sound analysis will, if successful, provide enabling technology for home-monitoring applications. Once the home-based screening test identifies individuals who should be properly diagnosed, hospital-based PSG testing should be considered. In a hospital-based PSG test, we expect EEG signals to play a pivotal role in the future, in diagnosing sleep apnea. The role of EEG is

In this article, we explored two techniques that addressed a range of difficult problems associated with diagnosing/ screening obstructive sleep apnea. The snore sound-based technique provided a basis for noncontact methods suitable community screening of the disease. Our results indicate that the performance of the feature Intra-Snore-Pitch-Jump Probability (ISPJ-probability) is on par with other competing technologies, in terms of the sensitivity/specificity characteristics. Snore-based diagnosis proposed in this article is superior, in that it does not involve contact instrumentation, thus solving a major problem in population screening of the disease.

302

A Framework for Information Processing in the Diagnosis of Sleep Apnea

We investigated the interhemispheric asymmetry of the brain during apnea and EEG-arousals (EEGA) events. The EEG asymmetry was determined via the Spectral Correlation coefficient of data, which computed the correlation between EEG measured from the two hemispheres of the brain. Results showed that apnea/EEG arousal events generally lead to a decrease in correlation. This change is most prominent during NREM sleep affected by EEGA, particularly in θ, α, and β frequency bands. Our results strongly suggest that the information flow between the left-right hemispheres of the brain is affected by events of apnea/EEG-arousals. The data used in this article came from: (i) patients medically diagnosed with sleep apnea, and (ii) subjects medically deemed to be without apnea, but nevertheless displaying some symptoms of the disease.

REFERENCES Abeyratne, U. R., Karunajeewa, A. S., & Hukins, C. (2007). Mixed-phase modelling in snore sound analysis. Medical & Biological Engineering & Computing, 45(8), 791–806. doi:10.1007/s11517007-0186-x

Bolduc, C. (2002). A study of EEG hemispheric asymmetry during REM sleep in young, healthy adults. International Journal of Bioelectromagnetism, 2(4), 145–146. Deegan, P. C., & McNicholas, W. T. (1995). Athophysiology of obstructive sleep apnoea. The European Respiratory Journal, 8(7), 1161–1178. doi:10.1183/09031936.95.08071161 Delaive, R. K., Roos, L., Manfreda, J., & Kryger, M. H. (1998). Obstructive sleep apnea patients use more health care resources ten years prior to diagnosis. Sleep Research Online, 1, 71–74. Dingli, K., & Assimakopoulos, T. (2002). Electroencephalographic spectral analysis: Detection of cortical activity changes in sleep apnoea patients. The European Respiratory Journal, 20(5), 1246–1253. doi:10.1183/09031936.02.00288202 Ehrlichman, H., & Antrobus, J. (1985). EEG asymmetry and sleep mentation during REM and NREM. Brain and Cognition, 4, 477–485. doi:10.1016/0278-2626(85)90035-1 Flemons, J. (2003). Home diagnosis of sleep apnea: A systematic review co-sponsored by the American Academy of sleep medicine. Chest, 124(4), 1543–1579. doi:10.1378/chest.124.4.1543

Abeyratne, U. R., Wakwella, A. S., & Hukins, C. (2005). A snore sound based method for the mass screening of sleep apnea. Physiological Measurement, 26, 779–798. doi:10.1088/09673334/26/5/016

Flemons, W. W., & Buysse, D. (1999). Sleeprelated breathing disorders in adults: Recommendations for syndrome definition and measurement techniques in clinical research. Sleep, 22(5), 667–689.

Australian Sleep Association (ASA), Fatigue and Transport Working Party (1999, June). House of Representatives Standing Committee on Communications, Transportation and the Arts. Australia.

Goldstein, L., & Stoltzfus, N. (1972). Changes in interhemispheric amplitude relationships in the EEG during sleep. Physiology & Behavior, 8, 811–815. doi:10.1016/0031-9384(72)90289-2

Babar, S. I., & Quan, S. F. (2002). Through the looking glass, the unrecognized tragedies of obstructive sleep apnea. Sleep Medicine, 3, 299–300. doi:10.1016/S1389-9457(02)00061-8

Hoffstein, V. (2000). Snoring, in the principles and practice of sleep medicine. In T. Kryger, & W. D. Roth (Eds.), (3rd Ed. pp. 813-826). Philadelphia: W. B. Saunders Co.

303

A Framework for Information Processing in the Diagnosis of Sleep Apnea

Karunajeewa, A. S. (2005). Abeyratne, U. R., & Hukins, C. Mixed-phase modeling of snore sounds within a nonlinear framework for component identification. IEEE/EURASIP International Workshop on Nonlinear Signal and Image Processing. Sapporo, Japan.

Swarnkar, V., Abeyartne, U., & Hukins, C. (2007). Inter-hemispheric asynchrony of the brain during events of apnea and EEG arousals. Physiological Measurement, 28, 869–880. doi:10.1088/09673334/28/8/010

Kryger, M. H. (2000). Management of obstructive sleep apnea-hyopnea syndrome: Overview. In The Principles and Practice of Sleep Medicine. W. B. Saunders Co.

Udwadia, Z. F., Doshi, A. V., Lonkar, S. G., & Singh, C. I. (2004). Prevalence of sleep-disordered breathing and sleep apnea in middle-aged urban Indian population. American Journal of Respiratory and Critical Care Medicine, 169, 2.

Martin, S., & Wraith, P. (1997). The effect of nonvisible sleep fragmentation on daytime function. American Journal of Respiratory and Critical Care Medicine, 155(5), 1596–1601.

Young, T., Evans, L., Finn, L., & Palta, M. (1997). Estimation of the clinically diagnosed proportion of sleep apnea syndrome in middle-aged men and women. Sleep, 20(9), 705–707.

McNicholas, W. (1998). Arousal in the sleep apnoea syndrome: A mixed blessing? The European Respiratory Journal, 12(6), 1239–1241. doi:10.1 183/09031936.98.12061239 Portier, F., Portmann, A., Czemichow, L., Vascaut, L., Devin, E., Cuvelier, A., & Muir, J. F. (2000)... American Journal of Respiratory and Critical Care Medicine, 162, 814–818. Puvanendran, K., & Goh, K. L. (1999)... Sleep Research Online, 2(1), 11–14. Sekimoto, M. (2000). Asymmetric interhemispheric delta waves during all-night sleep in humans. Clinical Neurophysiology: Official Journal of the International Federation of Clinical Neurophysiology, 111, 924–928. Svanborg, V., & Guilleminault, C. (1996). EEG frequency changes during sleep apneas. Sleep, 3(19), 248–254.

KEY TERMS AND DEFINITIONS β: α, θ and δ EEG Frequency Bands: Bandpass components of EEG signals respectively covering the frequency bands of 18-30 Hz, 8-13Hz, 4-7Hz, and 1-3.5Hz. EEG Arousals: An abrupt shift in EEG frequencies during sleep which may include α, θ, and/or frequencies 0 is a gain parameter, and ∆I = ∂2I / ∂x 2 + ∂2I / ∂y 2 is the Laplacian of the image. By adjusting the width of the spatial kernel used to compute the derivatives, the local contrast enhancement techniques can be adapted to the scale of the relevant features. Finally, the spectral domain methods refer to techniques operating in transformed domains, such as Fourier or wavelet bases. One useful operation in this category is the homomorphic filter, which allows correcting non-uniform background. Its principle is to model the background variation

by a low frequency multiplicative noise, and to remove it in the log-Fourier domain: I new = e

(F −1 (F (ln(I old )).H f ))



where F and F-1 are the direct and inverse Fourier transforms respectively, and Hf is the ideal highpass filter of minimal frequency f. Figure 3 illustrates two examples of image enhancement procedures. First row corresponds to the original blood smears and bottom row displays their corresponding enhanced images. Left image is the result of applying an unsharp mask on the V component of HSV color space, followed by a color median filter computed by minimization of the L1 distance in the RGB color space. Right image corresponds to the application of the color normalization process proposed by Tek et al. (2006). Note that object properties, such as color or contrast between objects, are visually highlighted.

OBJECT SEGMENTATION Generally, segmentation is the process of partitioning an image into disjoint regions based on a homogeneity criterion (Lucchese & Mitra, 2001). Two main levels of segmentation are commonly used in blood smear analysis: cell level segmentation, which aims at separating the whole cells from the background or plasma, and component level segmentation, which tries to separate the different components into the cell, such as nucleus from cytoplasm or intracell hemoparasites. Latter case is commonly used in applications in which the cell class depends on morphological features of its components. However, when finding the boundaries between components is not feasible, analyzing the components as a whole and trying to describe these morphological features as properties of the cell can be useful (Díaz et al. 2009).

329

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Figure 3. Image enhancement approaches. Fist row corresponds to the original images. Bottom row displays enhanced images obtained by application of an unsharp masking followed by a color median filter (left), and the color normalization process proposed by Tek et al. (right)

In this section the most popular cell segmentation approaches will be presented. Image segmentation approaches have been divided in two complementary perspectives: the region-based and boundary-based approaches. In the former the goal is to determine whether a pixel belongs to an object or not, whereas in the latter the objective is to locate the frontier curves between the objects and the background.

Region Based Segmentation Thresholding This is the simplest technique used to segment an image. This method allows separating the objects from the background using a pixel feature value that is compared with a threshold value in order to determine the class of the pixel. The feature generally used is the illumination value, although other feature descriptors have been used, such as gradient information (Bacus et al., 1976), entropy histogram (Bikhet et al., 2000) and intensity computed as mean of red, green and blue components 330

from the RGB color space (Hengen et al., 2002). Other authors have found that specific color components in different color spaces stress the differences between blood smear components. Cseke et al.(1992) found that nuclei of white cells are most contrasted on the green component of RGB color space, while differences between cytoplasm and erythrocytes are most perceptible in the blue component. This has been used by other authors for segmenting leukocytes (Katz, 2000) and hemoparasites (Le et al., 2008). The saturation component from the HSV color space also allows distinguishing the nucleated component (Di Ruberto et al. 2002; Ross et al., 2006; Wu & Zeng. 2006). Moreover, thresholds may be either global or local. In the latter case, some information relative to the pixel’s neighborhood is taken into account to perform adaptive threshold (Cseke et al., 1992; Hengen et al., 2002). The critical issue for the performance of the threshold algorithm is a good selection value, which can be determined heuristically from the grey level histogram of the image to be segmented (Micheli-Tzanakou et al., 1997) or computed

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

automatically. Harms et al. (1986) found that it is possible to estimate automatically a threshold for segmenting leukocyte nucleus, using the largest color difference value from a grey level image obtained as the combination of the red, green and blue color components (R + B – 2G). A similar approach was used by Poon et al. (1992), using a subtraction of the red and blue components (R – B +128). Analysis of the histogram shape for detecting the separating value between modal peaks was used by Le for segmenting nucleated components (Le et al., 2008). A straight line is drawn from the first prominent peak to the first empty bin of the histogram; the threshold is then selected as the abscissa of the point of the histogram with maximal perpendicular distance to the line. Clustering algorithms are used as another way of performing automatic thresholding. Originally, these algorithms assume that the image contains two classes of pixels i.e. foreground and background, and calculate the optimum threshold separating those two classes. Otsu (1979) proposes an optimal threshold by maximizing the between-class variance through an exhaustive search. This approach was adapted to multiple thresholds for separating background, cytoplasm and nucleus (Cseke, 1992; Scotti, 2006). In Wu et al. (2006), Otsu’s algorithm was applied on a circular histogram from the HSI color space; this representation prevents to lose the periodicity of the hue attribute (Wu & Zeng, 2006). Figures 4a and 4b shows segmentation results of blood and bone marrow smears shown in Figures 2 and 3. In this case, automatic thresholds on the green value histograms were computed using the Otsu algorithm. Segmentation based on threshold can work well in blood and bone marrow smears because background and foreground objects maintain constant visual feature values which constitute multimodal histograms. In particular, staining procedures highlight cell components containing DNA, such as white cell nuclei and parasites, in such a way that finding appropriate thresholds for

segmenting them becomes easy. However, boundaries between cytoplasm, plasma and earlier parasite stage are harder to find (see Figure 4b).

Pixel Based Classification Given an input image I, the classification of a target pixel I(x,y) is performed according to a feature vector (μ1,…,μm) describing it. Strategies in this category are composed of two steps: feature extraction and pixel class identification. Intensity or chromatic features, along with supervised classification, explained in section 3.5, are commonly used approaches. Tek et al. (2006) classify the pixels in a blood smear as stained or not stained, according to a their R,G,B values using a Bayesian classifier which is trained using a set of pixels classified by an expert. Classification of whole color space was proposed by Díaz et al. (2007) for reducing the image segmentation time. This method is based on a classification process that labels all components of a color space in one of the three classes: background, erythrocytes and parasites. Then, the labeled color space is used as a look-up-table for defining the class of each pixel in the images. The combined choice of a color space representation and a particular classifier was evaluated, showing that a normalized RGB color space together with a K-nearest neighbor (K-nn) classifier obtained the best performance. Images in Figures 4c and 4d show results of pixel classification for images of Figure 2. A K-nn classifier (k set to 15) was trained for separating the normalized RGB color space in three classes. In Figure 4c, parasites, erythrocytes and background were segmented. In Figure 4d, classifier was trained for distinguishing nuclei cells from cytoplasm (erythrocytes and leukocytes) and background.

Region Growing These are classical algorithms for which a set of regions are defined, each one initially constituted by a single pixel, and then an iterative growing

331

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Figure 4. Segmentation result samples. (a,b) Segmentation obtained by applying the Otsu algorithm on the green value histograms. (c,d) segmentation using a pixel classification approach. (e,f) Segmentation by k-means clustering

procedure is performed according to a homogeneity criterion. Main issues of region growing are: to establish a criterion that decides whether a neighbor pixel is similar to the seed, and to find the suitable seed. Region growing algorithms have been used in segmentation of leukocytes. Pixels inside nuclei, which are easily identified, are used as seeds, and cytoplasm are segmented by growing region; once nucleus is located, cytoplasm is detected by iterative growing of a ring surrounding the nucleus, and a gradient threshold is used as stopping condition (Kovalev et al., 1996). Lezoray et al. (1999) pro-

332

posed a leukocyte segmentation approach based on region growing in which region seeds are found by prior knowledge on color information of nucleus pixels. Then a growing procedure is applied based on color homogeneity and gradient information.

Region Clustering Region clustering approaches are similar to region growing, but the region clustering focuses on the region directly without any seed. In contrast with pixel based classification strategies these approaches identify coherent regions instead

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

of classifying single pixels independently. The k-means algorithm is a well-known clustering method. In (Sinha & Ramakrishnan, 2003), pixels represented by a vector of 3 components from the HSV color space were automatically clustered, using k-means algorithm, for identifying leukocyte nucleus, cytoplasm and background. Then results were further refined by an ExpectationMaximization algorithm. In (Thera-Umpon, 2005), fuzzy k-means algorithm was proposed as an over segmentation algorithm applied on grey level image for detecting regions corresponding to nucleus in leukocytes. Comaniciu & Meer (2001) proposed a method, which detects clusters in the L*u*v* color space and delineates the border cells using the gradient ascent mean shift procedure. Segmentations obtained by application of the k-mean algorithm on green value component of previous images are presented in Figures 4e and 4f.

region for determining whether one point is inside the region. Likewise, markers were extracted from bone marrow images by a mean shift procedure, and used as the seed in a watershed segmentation applied on the color gradient image (Pan et al., 2003). Multiple markers within the same object can be obtained when they are extracted by thresholding. Then, a unique marker is associated to one object by performing a morphological closing on the markers so that they touch each other. The parameters used for these operations are related to the maximum object size which can be previously determined from a granulometric analysis (Mahona Rao & Dempster, 2002). In addition, watershed region merging was also used for enhancing the watershed segmentation by Wang et al. (2006). For doing this, a series of rules on surface, depth and volume of adjacent watershed regions were used as merge criteria.

Morphological Segmentation: Watershed Transform

Boundary Based Segmentation

The watershed transform is the fundamental tool of the mathematical morphology for segmentation of images. For a complete review of this topic we refer the reader to (Dougherty, 2003). The watershed concept comes from considering the image as a topographic surface, in which grey levels correspond to terrain altitudes. Segmentation is performed by flooding or rain falling simulation, and the resulting catchment basins (which correspond initially to the regional minima of the image) form the final segmentation. Classical watershed transform is applied to a gradient image, generally presenting many regional minima, which leads to oversegmentation. In order to overcome this problem many strategies have been proposed. An extended watershed approach for color images was proposed by Lezoray et al. (1999, 2002), in which color information is used for extracting the markers and for defining a function that combines the color gradient module with the mean color value of the

Edge Detectors Classical edge detection approaches are based on detection of abrupt neighborhood changes in the pixel values. In many cases, the boundaries between cells and their components are not clearly defined, and then edge detection performs poorly on these images (Wu & Zeng, 2006). Even so, edge operator performance can be improved when it is combined with other techniques. The Teager energy filter and morphological operators were proposed for segmenting leukocytes (Kumar, 2002). Similar scheme was used by Piuri et al. (2004), but the Canny edge operator was used.

Deformable Contour Models or Active Contours A deformable model is a curve that evolves toward boundary contours using local information of the image. Two main models of active contours

333

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

have been proposed: explicit models (snakes) and implicit models (level sets). Explicit models aim at minimizing the energy attached to a curve describing the contour of the object. The optimization function is defined as a linear combination of external forces (computed from the image properties) and internal forces (defined by elasticity and stiffness of the curve). Ongun et al. (2001b) proposed a snake for segmenting leukocytes, for which external forces are computed from the intensity and gradient magnitude of the image and a corner force attraction. Likewise, a gradient flow vector was used to define these external forces in (Theerapattanakul et al., 2004; Zamani & Safabakhsh, 2006). This approach was also used by Yang et al. (2005) but the gradient flow vector was computed from the L*u*v* color space (this was also used by Tuzel et al., 2007). Implicit models represent the evolving object contour through a function of higher dimension defined over the image area, which is positive inside the region, negative outside and zero on the boundaries. The deformation of the contour is generally described by a partial differential equation (PDE). Level sets can be viewed as region or boundary based approaches. In boundary based methods, the function determining the evolution of the contour (stopping function) is based on the gradient of the image. Nilsson & Heyden (2001; 2005) used this scheme to segment leukocyte components. For nuclei segmentation, the stopping function is defined by a threshold value on the RGB color space, while the image gradient is used for cytoplasm segmentation. On the other hand, region based level sets define the contour derivative as a function that depends on the stiffness of the curve and an energy term, which is minimum if inside and outside regions are homogeneous. Dejmal et al. (2005) proposed a linear combination of region and boundary criteria for segmenting erythrocytes. Active contours are useful in segmentation of clusters of cells, however they require a relatively

334

high computational cost, and resulting contours do not correspond with the exact borders of the cells. Then, they are not appropriate in differentiation process based on the contour shape, like in erythrocyte classification.

Segmentation Improvement After segmentation, an image can be thought of as a binary map of background and foreground objects. In some cases, this initial segmentation is not satisfactory, as holes and artifacts can appear. Improving the segmentation results can be done through a set of operations based on a priori knowledge. Morphological operators are commonly used for this purpose. Basic morphological operators are often used after segmentation process for reducing artifacts, filling holes or removing border objects (Anoraganingrum, 1999; Sabino et al., 2004; Jian et al., 2006, among others). Binary erosion shrinks image regions and eliminates small objects, whereas binary dilatation enlarges image regions, allows connecting separated blobs and fills small holes. Segmented objects that are not interesting for the analysis can be removed using connected operators, such as the opening by reconstruction or the area opening. These operators are filtering techniques that eliminate some regions without modifying the boundaries of the remaining regions. Connected operators implemented as pruning strategies of a tree representation was presented by Díaz et al. (2009) for removing staining artifacts and border touching cells.

SPLITTING OF CLUMPED CELLS An important problem in analysis of blood and bone marrow smears is the clumped or touching cells. Many approaches have been proposed for separating them, some of them included as part of the segmentation and other specifically dedicated to separate superposed cells. For instance, some

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

leukocytes segmentation approaches working on a sub-image that are extracted from the original image, cutting a square around the segmented nucleus (Kovalev et al., 1996; Jiang et al, 2003, Sinha & Ramakrishnan, 2003; Theerapattanakul et al., 2004; Zamani & Safabakhsh, 2006). Under the assumption that each sub-image has only one white cell, other a priori are included for segmenting it. Kovalev et al. (1996) and Katz (2000) introduce circle-shaped restrictions, whereas color information is used for performing clustering around the detected nucleus by Jiang et al. (2006) and by Sinha & Ramakrishnan (2003). On the other hand, restrictions included into the fitness function of deformable models allow to deal with overlapped cells (Liu & Sclaroff, 2001; Theerapattanakul et al., 2004; Zamani & Safabakhsh, 2006). Single and complex morphological operators have proved useful for the segmentation of touching blood cells. Opening operator was applied for separating leukocytes (Kumar, 2002; Dorini et al., 2007; Pan et al., 2003). Hence, a series of morphological operations was applied for splitting composite erythrocytes by Di Ruberto et al. (2002). This procedure was composed of a morphological opening by a flat disk-shaped structuring element, followed by a morphological gradient, an area closing and a thinning operator. Distance transform (i.e. a function associating to every pixel of a binary image its distance to the border) of the segmented clumped shape is also used in some clump splitting approaches. The maxima in the distance image can be used as markers for subsequent segmentation of the original image by the watershed algorithm (Malpica et al. 1997; Angulo & Flandrin, 2003a, Nilsson

& Heyden, 2005) or another region growing approach (Hengen et al., 2002). But the watershed segmentation can also be applied directly on the distance image for separating circular shapes (Lindbland, 2002). These approaches present good results for splitting clumped cells with small overlaps, but they fail when the overlapping area is too important. An improved approach which introduces geometrical restrictions on the distance function was presented by Pan et al. (2006a). The purpose of these restrictions is to improve the segmentation accuracy and to reduce the computational cost of the watershed computation. Figure 5 shows an example of splitting clumped cells using the watershed transform applied on distance transform. On the segmented image (Figure 5b), the Euclidean distance transform is computed (Figure 5c) and the regional maxima of the distance transform (Figure 5d), are used as markers of the watershed transform (Figure 5e). Other approaches for separating clumped cells are based on a concavity analysis, which assume that superimposed objects can be separated by one line joining two specific cut points where the boundary curvature abruptly changes. Concavity analysis is commonly composed of three sequential steps: detection of concavity points, detection of candidate split lines and selection of best split lines. Poon et al. (1992) presented a method for splitting clumped cells that uses the difference of the tangent angles between neighborhood boundary pixels for finding two cut points on the boundary which are then joined by a line for separating the shape. This approach separates pairs of cells but cannot handle clusters of multiple cells.

Figure 5. Clump splitting using distance transform and watershed segmentation

335

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

An approach that allows separating complex clumped cells proposed by Kumar (2002) was used in (Sio et al., 2007). In this approach, cut points are detected by measuring the concavity depth, defined as the distance from the boundary point to the convex hull border. Then candidate split lines are selected from those obtained by joining all possible pairs of concavity pixels, based on the distance and alignment of points. Finally a “measure of split” is computed for selecting the best split lines (Kumar, 2002). The drawback of these methods is that they depend on implicit conditions about cell shape or size. Morphological operations are limited by the overlapping degree between cells since two shapes can be separated if their centers are separated (Serra, 1982). Whereas, concavity analysis demands very accurate segmentation in order to detect the cut points. Template matching strategies attempt to find parts in the image which correspond to predefined shapes named templates, without any other a priori information. These approaches have two important components: the template definition and the matching strategy, which is commonly formulated as an optimization problem. Halim et al. (2006) proposed a template matching strategy for detection of erythrocytes. A healthy erythrocyte represented by a gray scale template was constructed from the original image, based on cross correlations between the predefined binary templates and the image. Then, cross correlation was also used as optimization criterion in order to detect the actual erythrocytes. On the other hand, Díaz et al. (2007a) proposed a matching strategy based on superposition of the chain code representation of the clumped shape contour and an ideal erythrocyte, estimated from the original image by an Expectation-Maximization algorithm. Although these strategies can overcome the drawbacks previously described, they are computationally expensive and are applicable only for separating shapes that do not present large shape and size variations.

336

FEATURE EXTRACTION AND SELECTION Differential analysis of blood and bone marrow smears is feasible thanks to the capacity of visual system to distinguish image patterns, which are associated to specific cell classes. Thereby, the challenge of automatic image analysis systems is to carry out computational methods to calculate objective cell measures related to those visual patterns. The aim of feature extraction step is to obtain a set of descriptors, which will be further separated in different classes by a classification procedure. The feature extraction can be formally defined as a function f which maps the original image I onto a feature vector x, i.e. f:I→x=(x1, x2,…, xd), where d is the number of features used for characterizing the image.

Visual Descriptors Computational visual descriptors can be classified as chromatic, textural and shape measures. Chromatic features are related to the color image information, textural features provide statistical information on the local structures in the image, and shape measures describe geometrical information of objects like perimeter, area, circularity, etc. In the rest of this section we discuss some common measurements used for describing the blood and bone marrow cells.

Chromatic Features These features describe the grey-level or color distribution of the images, which are the most discriminative features of blood and bone marrow cells. Let R={p1, p2,…, pn} be a set of pixels that belong to a segmented object (nucleus, cytoplasm, parasite or complete cell). Intensity and chromatic features are computed from histograms of R. Color histograms can be represented by one single multi

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

dimensional histogram or multiple separate 1-D histograms. Considering an image as a stochastic process where every pixel value is modeled as a random variable, the histogram can be regarded as the probability density function of grey level or color distribution. The most used measures from a histogram h are provided by its moments: mean, variance, standard deviation, kurtosis and skewness, which describe the histogram shape and are computed as1:

∑ kh(k ) Variance : v = ∑ (k − µ) h(k ) Mean : µ =

k

2

k

Standard Deviation : σ = v Skewness : µ3 = σ −3 ∑ (k − µ)3 h(k )



k

Kurtosis : µ4 = σ −4 ∑ (k − µ)4 h(k ) − 3 k

Entropy : e = −∑ h(k ) ln(h(k )) k

The standard deviation is a measure of dispersion of the histogram, whereas the skewness and kurtosis respectively measure the dissymmetry and flatness (or peakedness) of the distribution. An important issue in chromatic descriptors is the color space used for representing the color in the images. Traditionally, images are acquired using the RGB color model, in which each color pixel is represented by its three components: R(Red), G(Green) and B(Blue). But the statistical distributions of those three components are highly correlated, and then, many decorrelated models have been proposed for digital image processing (Plataniotis & Venetsanopoulos, 2001) and some of them have been used for blood and bone marrow cells description. Discrimination of platelets, red and white cells in peripheral blood smears has been correctly accomplished using only the average intensities of R, G and B color components and cell size

(Lin et al. 1998). Nevertheless, discrimination of subclass cells as leukocyte or erythrocyte types usually requires other features, such as textural or geometrical measures. Leukocyte classification systems have used chromatic statistical measures as part of a feature vector. Kovalev et al. (1996) used standard deviation of red and green intensities for describing chromatic features of leukocytes. R, G and B components and color ratios green/red and green/ blue were used by Katz (2000), together with the nucleus area and perimeter, for the same purpose. Similar features were used by Song et al. (1998) and by Sinha & Ramakrishnan (2003). Likewise, RGB color histograms computed from original and gradient images, were used along with many other features (Siroic et al., 2007). On the other hand, CieLab and HSV transformations have also been used in leukocyte color characterization (Ongun et al, 2001; Comaniciu & Meer, 2001; Angulo & Serra, 2002). Chromatic features have also been used for distinguishing live stages of hemoparasites, particularly malaria. HSV color model histograms were used for describing segmented parasites which were classified as mature and immature throphozoites (Di Ruberto et al., 2002). For the authors, parasites in these live stages are differentiated by their nucleus and spot of chromatins, which are evident in the hue and saturation components. Statistical features derived from the red, green, blue, hue and saturation components were part of a set of features used by Ross for classifying malaria parasites into different live stages (Ross et al., 2006). Same measures computed from the normalized RGB color model, were used for classifying erythrocytes as healthy or infected at any infection stage by Díaz et al. (2009).

Shape Descriptors Another important feature to the human visual system is the shape. Many shape measurements have been proposed in the literature. A comprehensive

337

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

review of general purpose shape descriptors was published in (Zhang & Lu, 2004). In hematological applications, the geometrical or “simple” shape descriptors are the most popular because they are easy to compute, but indiscriminative for small dissimilarities. Other used descriptors include geometrical moments, structural representations and spectral features.

Geometrical Features Geometrical features describe different aspects of the cell or component structure, such as size or shape. In hematological applications they are widely used because various cells differ greatly by their size or nucleus shapes. Geometrical features are computed on a region of interest R, which has a well defined closed boundary composed of a set of consecutive pixels S={s1, s2,…,sn}. Simplest geometrical features are area, perimeter, centroid, tortuosity (area/perimeter) and radius. The area feature is computed as the number of pixels enclosed by the cell boundaries (Bacus, 1976), A={cord(R)}. The perimeter, n −1

P = | sns1 | +∑ i =1 | sisi +1 | , corresponding to the

total distance between consecutive points of the boundary. The centroid is the average of coordi1 nates of all points of R: x = ∑ x , A (xi ,yi )∈R i y =

1 ∑ y . The radius is measured by A (xi ,yi )∈R i

averaging the length of the radial line segments defined by the centroid and border points n 1 R = ∑ i =1 | si (x , y ) | . Other features for den scribing shape structure include measures of major and minor axis, which are provided by the eigenvalues of the second order moments matrix; the circularity or elongation (Thera-Umpon & Gader, 2002), computed as the ratio of the square of the perimeter to the area (P2/A); the eccentricity, defined as the ratio between the major and

338

minor axes (Sinha & Ramakrishnan, 2003); the symmetry, calculated from the distance between lines perpendicular to the major axis and the cell boundary (Stanislaw et al., 2004; Markiewicz & Osowski, 2005); and the spicularity, proposed by Bacus et al. (1976), spicules are defined as the points where chain code derivate change of sign. In addition, other specific features have been proposed, which are more strongly linked with blood cell analysis. For example, the nucleus and cytoplasm areas ratio and the number and structure of nucleus lobes are prominent features used for identifying the class of the leukocytes (Beksak et al., 1997, Sinha & Ramakrishnan, 2003, Hengen et al., 2002, Ballaro et al., 2008). Likewise, number of chromatin dots, and the ratio between parasite area and cell area have been used for identifying the infection stage and class in malarial image analysis (Ross et al, 2006). The main drawback of the geometrical features is that their application demands accurate segmentation of the region of interest, and then they are commonly used together with other features more robust to segmentation errors, such as texture or chromatic descriptors.

Structural Based Representations Structural methods decompose the shape or contour region into parts that are geometrically meaningful for the image. A structural representation widely used is the convex hull (de Berg et al., 2000), corresponding to the smallest convex polygon H containing the region R (R⊂H). Once computed the convex hull, shape can be represented by a string of concavities as chain code representations or by measures based on convex hull such as its area or region compactness computed as cell area/convex hull area ratio (Mircic & Jorgovanovic, 2006; Ballaro et al, 2008). Region skeleton it is another shape representation that allows to encode both boundary and region information for every object (connected component). Many algorithms for generating

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

skeletons have been proposed. Di Ruberto et al. (2002) applied the sequential thinning process for representing segmented malarial parasites as connected thin lines. The end points of the obtained skeletons are used as shape descriptors for distinguishing either immature disk-shaped parasites or mature irregularly-shaped parasites.

Spectral Based Features In order to provide descriptors less sensitive to noise and boundary variations, several authors have proposed to use a representation of pattern shapes in frequency domain. Those spectral features include Fourier descriptors and wavelet descriptors, which are usually derived from spectral transform on shape signatures (functions derived from the shape boundary points). Features obtained from the direct application of the discrete Fourier transformation are not invariant to geometrical transformation, and then different extensions have been proposed. Elliptic Fourier descriptor (EFD) proposed by Kuhl & Giardina (1982) was used for characterizing nuclei shapes of leukocytes (Comaniciu et al., 1999) and megakaryocytes (Ballaro, et al., 2008). EFD is a Fourier expansion of the chain coding contour, which is represented as a composition of ellipses defined as contour harmonics that result from expanding separately the components x(s) and y(s) in the complex function of coordinates u(s)=x(s)+jy(s). The EFD corresponding to any closed curve S with Euclidean length l and composed of k points is described by the nth harmonics given by: ∆x  2πns 2πns  ∑ ∆s i cos l i − cos l i−1  i =1 i k 2πnsi −1  ∆x i  2πnsi l  sin − sin bn = 2 2 ∑  l l 2π n i =1 ∆si  k ∆yi  2πnsi 2πnsi −1  l  cos − cos cn = 2 2 ∑  l l 2π n i =1 ∆si  k 2πnsi −1  ∆yi  2πnsi l  sin dn = 2 2 ∑ − sin  l l 2π n i =1 ∆si  l an = 2 2 2π n

k

i

Where si = ∑ ∆s j , is the length of the first j =1

i vectors, ∆si =

2

(∆x ) i

2

+ (∆yi ) , ∆xi=(x1-xi-1)

and ∆yi=(y1-yi-1). The coefficients of the EFDs are normalized to be invariant with respect to the size, rotation, and starting point, using the ellipse of the first harmonic. Another extension of the Fourier descriptor used as cell shape feature is the UNLF (Universidade Nova de Lisboa, Portugal) descriptor. The UNLF descriptor is computed by applying 2-D Fourier transform on the image curves transformed to the normalized polar coordinates. The main advantage of this descriptor is that it is able to handle open curves, lines and patterns composed of parametric curves as well as cells with interior components. Kyungsu et al. (2001) used it as shape descriptor of erythrocytes in order to include information about its concavity, observed as holes in a binarization process. The main drawback of UNLF descriptor is that it produces feature vectors of high dimensionality, and then feature selection approaches should be applied. The use of wavelet transformation has been proposed as shape descriptor in many applications in order to achieve description of shape features in a joint spatial and frequency space. However, they have not been much used for cell description. In Sheik et al. (1996) only, largest wavelets coefficients were used for classifying cells as platelets, white or red cells.

Textural Features Traditional machine vision and image processing approaches assume uniformity of intensities in local image regions, but some objects present a repeated pattern as principal visual feature, which is called texture. In hematological applications the texture property has proved valuable for distinguishing some abnormal cells and parasite presence and evolution.

339

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Texture analysis is generally applied on a grey scale representation of the image. However, as in segmentation approaches, different color models have been used for textural analysis. Hengen et al. (2002) perform the texture analysis in H component of HSI model (Hengen et al. 2002). Ongun et al. (2001) use an image distance as input for computing a grey level co-occurrence matrix which is computed as the distance between the pixel color and coordinate origin in the CIELab model. Finally, Angulo and Serra (2002) extend the texture analysis to all RGB channels that are then integrated to compose one single texture feature (2002). Methods for extracting textural measures can be classified as: statistical, model based and geometric.

Statistical Approaches These are the most popular texture analysis methods, in which the texture can be defined by its statistical features, related to the pixel gray level conditional probability given its neighborhood. The goal of these methods is to deduce statistical properties of this function from observed data (gray levels value of pixels). These methods are easy to implement, which is their main advantage. The gray-level co-occurrence matrix (GLCM) is a second order statistics, which is defined as follows: GLCM d (i, j ) = #{(x , y ) ∈ {0,W − 1} × {0, H − 1}, I (x , y ) = i ∧ I (x + dx , y + dy ) = j )} / (W * H )

where d=(dx,dy) is a displacement vector, W and H are the width and height of the image I, and #S is the cardinality of set S. Hence, GLCMd(i,j) is the probability of finding two pixels separated by vector d, which have respective grey values i and j. Many feature vectors can be computed from the GLCM (Haralick, 1979, Tuceryan & Jain, 1998). For example, Sinha and Ramakrishnan used the entropy, energy and correlation of the GLCM for

340

characterizing the cytoplasm in leukocytes (Sinha & Ramakrishnan, 2003). Additionally, Sabino et al. (2004a) used the contrast feature for recognition of leukocyte. The main drawback of the GLCM approach is that their calculation is computationally expensive even if the grey level values are quantized. Sum and difference histograms are a modification of GLCM in which two histograms are calculated counting grey level differences and sums for all the pixel couples separated by vector d. In hematological applications this method has been used in the differentiation of blood cells type (Siroic et al., 2007) and the classification of leukemia blast cells (Stanislaw et al., 2004; Markiewicz & Osowski, 2006). Another statistical approach used to characterize blood cells is the autocorrelation coefficients (Sinha & Ramakrishnan, 2003); this second order statistic may be calculated as: W −p

ACC (p, q ) =

∑ i=1 W *H (W − p)(H − q )

p = 1… P, q = 1…Q

∑

H −q j =1 W

I (i, j )I (i + p, j + q ) H

2

∑ i =1 ∑ j =1 I (i, j )

,

where P,Q are neighborhood size parameters.

Model Based Approaches These approaches are based on the assumption that the texture corresponds to instances of mathematical models, in which parameter models should be found. Feature extraction based on the Markov random fields (MRF) is a representative model based approach. It is a generalization of Markov chains in 2-dimensional space, where the time index is replaced by spatial indexes and the Markov restriction is applied into the neighborhood of each pixel. Hence, texture is characterized by the parameters that define the MRF for observed pixel values and a certain topology, defining the local dependencies. According to the Hammersley-Clifford theorem, solving this problem is equivalent to estimating the parameters

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

of a Gibbs distribution if the MRF has translation invariance and isotropic properties, which is assumed in image processing applications. MRFs have been used as feature descriptor of leukemia blast cells (Stanislaw et al., 2004; Markiewicz & Osowski, 2006). Other model based approaches used in blood cells characterization are the autoregressive models, which were used for characterizing bone marrow cells for differential counting analysis (Ongun et al., 2001) and malignant lymphomas and chronic lymphocytic leukemia in peripheral blood smears (Comaniciu et al., 1999; Comaniciu & Meer, 2001). The autoregressive model of order n+2 is a time series of order n, in which the time variable is defined in a 2D spatial domain. The other 2 parameters correspond to the mean and variance of a white noise added to the model. Maximum likelihood estimation is commonly used for inferring the model parameters.

Geometric Approaches Texture features may also be derived from geometric models. In these approaches, texture is defined as a repetitive pattern composed of texture element named textons. Granulometry analysis was proposed as a texture feature for describing nucleus granules, in order to characterize different kinds of leukocytes. Angulo & Serra (2002) applied the granulometry analysis for calculating texture features used in cell retrieval and leukocyte classification systems. A similar work was presented by Theera-Umpon and Dhompongsa in classification of leukocytes (Theera-Umpon, 2007). A texture descriptor based on resulting regions of a watershed segmentation was proposed in (Hengen et al., 2002). A cell nucleus is segmented applying standard watershed and then the ratio between the interior area (i.e. without boundary) of watershed regions and the area of the whole n

nucleus is computed. ( ∑ i =1 Ai / Anucleus with n= number of watershed regions).

Recently, image analysis approaches based on analogy with syntactical document analysis have been proposed. In texture description, the idea of a texton element associated to a visual word was used by Tuzel et al. (2007) for characterizing cell nucleus and cytoplasm. A texton dictionary is generated from a set of training images of each class. For this, a filter bank was designed, composed of two rotationally symmetric filters (Gaussian and Laplacian of Gaussian) and 36 edge and bar filters with different scales and orientations. Then filter responses inside the segmented images were clustered using the k-means algorithm and the cluster centers were the selected textons. Using the resulting texton dictionary, the new cells are represented with their texton histograms.

FEATURE SELECTION Feature selection is a process commonly used in pattern recognition, which allows determining the most relevant, discriminating and uncorrelated features of a classification task, while reducing the dimensionality of vectors associated to objects. Although original feature vectors extracted from blood and bone marrow cells are usually large, feature selection processes are not often applied in published studies. There are many techniques of feature selection, which can be classified into three categories: filter methods, wrapper methods and embedded methods (Molina et al., 2002; Saeys et al., 2007). In the first category, feature selection is based on quantitative measures of how well the features discriminate the classes, independently of a specific classification algorithm. In the wrapper approaches, the contribution performance of every component is taken into account by the classifier algorithm. Embedded methods refer to classification procedures that include the feature selection as a step in the whole classification procedure, for example decision trees. Wrapper models outperform other models because they are

341

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

optimized for the specific classification approach, but they are computationally feasible only for small feature vectors. Another relevant issue in the feature selection processes is the search strategy which can be exhaustive (brute-force or complete search), heuristic (sequential) or non-deterministic (randomized) (Molina et al., 2002). Exhaustive search generates the optimal subset of n features in a search space of possible subsets. Heuristic approaches select features that maximize a criterion function. Sequential forward selection starts selecting the best single feature and successively adds one feature at the time, whilst sequential backward selection starts with the whole set of features and then deletes one feature at the time. In hematological applications, comparison of several feature selection approaches has been performed. Markiewicz et al. (2006) evaluated two filter approaches (correlation analysis, mean and variance measures) and one wrapper of linear support vector machine (SVM) for selecting features from a vector of 164 dimensions, in order to distinguish 10 classes of leukocytes. The best approach was the correlation between the feature and the class as well as the linear SVM ranking. Likewise, a wrapper method with sequential forward search strategy applied on a naive Bayes classifier was used in (Sabino et al., 2004a). 12 relevant features were selected exploring morphometry, texture and color feature sets separately, i.e. 4 features for each one. Genetic algorithms have been proposed as a feature selection process in (Siroic et al., 2007). In this approach, each set of possible features is represented as chromosomes and genetic operators such as mutation and crossover are applied in order to find the best solution(s) according to a fitness function, defined as the classification error on the validation data set when a SVM is used as classifier algorithm. From reported results, features selected with this approach proved better classification performance than the ones selected by wrapper of linear SVM.

342

On the other hand, some approaches called by many authors “feature selection”, do not properly perform a selection of features, but carry out a reduction of dimensionality by projection or combination of them. Principal component analysis (PCA) is the most popular technique for dimensionality reduction. The aim of the PCA is to find a set of orthogonal vectors in feature space corresponding to directions along which the data present the maximal variance. Dimensionality reduction is performed projecting the data from their original space onto the orthogonal subspace. Kyungsu et al. (2001) used PCA for reducing a 76d vector in order to classify erythrocytes and leukocytes, achieving dimension reductions between 12% and 50%. A non-linear extension of PCA, called kernel-PCA, which maps input data into high dimension feature space using the kernel trick was employed by Pan et al. (2006)

SINGLE CELL CLASSIFICATION Once the cell features are extracted and selected, they should be input into a process that classifies the cells according to hematological concepts. An automatic classification system should be able to identify these concepts within the context of real images, i.e. noisy images with visual differences between concepts, which are variable. This problem may be addressed from two different perspectives: analytic and inductive. The analytic standpoint requires a deep understanding of the way the low-level features are combined to structure concepts. The inductive standpoint automatically builds a model of concepts based on a number of training samples, a framework commonly used through several machine learning approaches. Many learning approaches have been used to classify blood cells. The simplest model is the Bayesian classification (Aksoy, 2002; Sabino, 2004; Theera-Umpon, 2004), which is based on applying the Bayes theorem with independence

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

assumptions between features. In the training step the a priori probabilities of each class are computed, and then used for estimating the class of new instances by the maximum a posteriori rule. Despite its simplicity, Bayesian classification can perform accurate classifications if features are discriminative. Conversely, k-NN classifies unlabeled samples based on their similarity with samples in the training set without any knowledge of a priori class probabilities. Given the knowledge of N prototype features and their correct classification into M classes, the k-NN rule assigns an unclassified pattern to the class that is most heavily represented among its k nearest neighbors in the pattern space, under some appropriate metric. k-NN was used by Tek et al. (2006) to classify stained regions as parasites or not. Likewise, Comaniciu et al. (1999) use this technique for distinguishing lymphoproliferative disorders. Artificial neural networks (ANN) are the most used classifiers in hematological applications. ANNs are networks of simple processors that emulate the behavior of human brain. Many network architectures and training algorithms have been proposed and are well described in literature (Bishop, 1995). The most popular learning algorithm is the multilayer back propagation (ML-BP), which was used in many studies (Lin et al, 1998; Kim et al., 2001; Mircic & Jorgovanović, 2006; Ross, 2006; Theera-Umpon & Gader, 2002). Other algorithms used in the literature include ALOPEX (Lin et al., 1998), Radial basis function – (RBF, Djemal et al., 2005) and Linear Vector Quantization (Ongun et al., 2001). Recently Support Vector Machines (SVM) approaches have received increasing interest because they have outperformed other methods in several pattern recognition problems. The SVM is based on the mapping of the input vectors into a high dimensional feature space, induced by a kernel function chosen a priori. In the feature space the learning algorithm produces an optimal separating hyperplane between two classes. In principle,

a SVM is a linear discriminator; however it can perform non-linear discrimination thanks to the fact that it is a kernel method. The multi-class classification problem is solved constructing several binary classifiers and combining them. The most used strategies are one-against-one and one-against-all combinations. In the former, a binary classifier is trained for all combinations of classes and final decision is made by majority rule among the classifiers (Stanislaw et al. 2004; Ramoser et al., 2005; Markiewicz et al, 2005; Guo et al., 2006; Siroic, 2007; Tuzel, 2007). In the latter, for each class a binary classifier is trained by labeling the samples from the class as positive examples and samples from the other classes as negative examples, the final decision is assigned to the class having maximum decision function among all classes (Pan, 2006). An important issue found in hematological applications is the fact that cells can be classified using hierarchical classifier structures, which take advantage of morphological taxonomy of the blood cells for extracting the most discriminative features in each classification level. Kim et al. (2001) proposed a two stage structure of ANN for classifying erythrocytes according to their shape features. Firstly, the most varied shapes were discriminated, and then circular shapes were classified as normal, target, spherocyte or stomatocyte. The same strategy was used by Ross (2006) in the classification of infected erythrocytes; a first stage decides whether erythrocyte is infected or not and a second stage defines the infection stage. Similar strategy was used by Díaz et al. (2009), but based on a SVM classifier.

CASE OF APPLICATION To conclude this chapter with an application case, we briefly present in this section the application of the general framework to the analysis of malarial infected blood smears. More details can be found in Díaz et al. (2009). The main goal of analysis of

343

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

malarial infected smears is to estimate the infection level of each smear. For this, we need to count the number of infected and non-infected erythrocytes but also to estimate the life stage of each parasite. As shown in Figure 6, discrimination of these cells is very difficult due to the slight differences between consecutive life stages.

Image Acquisition Images were digitized using a Sony high resolution digital video camera, which was coupled to a Carl Zeiss Axiostar Plus microscope. Use of intermediate lens and a ×100 power objective yielded a total magnification of ×1006. Optical

image corresponded to 102×76μm2 for a 640×480 image size, resulting in a resolution of 0.0252μm2.

Image Preprocessing Original images presented luminance variations mainly due to film inhomogeneities and varying illumination condition of acquisition devices. In order to correct the background inhomogeneities, a local low-pass filter was applied on the luminance channel from the YCbCr color space i.e. a low pass filter applied on sub-windows of approximately the larger size of erythrocytes in the image, and smoothed out using a moving window whose size was adjusted in order to eliminate the tiling

Figure 6. Single erythrocyte samples. From left to right: two healthy erithrocytes, two erythrocytes infected with ring stage parasites, two erythrocytes infected with throphozoite stage parasites and two erythrocytes infected with schizont stage parasites

Figure 7. Processing results of malarial infected blood smear images. (a) Original Image. (b) Pre-processed Image. (c) Pixel-based classification result. (d) Result after filtering by inclusion-tree representation

344

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

effect of the filter. Figure 7b shows an example of application of this procedure.

Erythrocyte Segmentation Automatic quantification of malarial infected blood smears has been designed on the base of detecting parasites and erythrocytes independently, and to locate the former into the latter to quantify the infection level. However, in early infection stages, the boundaries are not clearly defined, and segmentation is very difficult. So, the analysis of the erythrocyte-parasite as a whole was proposed as an alternative procedure. This stage tries to extract single erythrocytes from the background as follows: first, a pixel classification allows labeling each image pixel as either background or foreground, based on its color features. Then, segmentation is improved using an Inclusion-Tree structure, which is simplified to satisfy the restrictions imposed by the erythrocyte morphological structure and its spatial organization. This simplification allows removing artifacts generated at the staining or digitization processes. Figure 7c shows an example of an image segmented with pixel classification procedure and its corresponding result after filtering process is shown in Figure 7d. As the presence of clumped cells affects the automation of erythrocyte quantification, splitting was achieved by a template matching strategy that

searched for the best match between a chain code representation of the clumped shape contour and an ideal erythrocyte, estimated from the original image by an Expectation-Maximization algorithm (see Figure 8). This approach attempts to find cells in clumped shapes similar to erythrocytes found in the same image.

Feature Extraction Visual features of erythrocytes were described by the moments (mean, variance, standard deviation, kurtosis and skewness) of a set of histograms that represent the probability density function of the color, saturation, intensities, texture and edges inside the whole cell. So, an image was represented by 25 features that characterize the five histograms.

Cell Classification Erythrocyte classification was carried out by learning from samples approach composed of a hierarchical (two levels) ensemble of classifiers. At first level, a classifier was trained for deciding whether or not an erythrocyte was infected. Once an erythrocyte classified as infected, a second level determined the stage of infection using a set of classifiers, composed of four learning models: one model for each class and one more for detecting possible artifacts. The erythrocytes

Figure 8. Splitting of clumped cells. Top: Original overlapped cells. Down: splitting results. In white image edges, in green cells found by the proposed approach (Díaz et al. 2009)

345

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

that were misclassified into none, two or more classes were ruled out and left to the user for a later visual decision. Training algorithm for any learning model was selected by evaluation of their performance using a one-against-all strategy, resulting in an array of: one polynomial SVM, two radial basis function SVM and a neural network training models. The performance of classification task was evaluated using a Fβ measure, which is related to effectiveness. Parameter β allows assigning relative weights to the true positive and precision rates for dealing with class imbalance problem. A set of 12,557 erythrocytes composed of 11,844 healthy ones, 521 in ring stage infection, 109 in trophozoite stage infection and 83 in schizont stage infection, was used, which were automatically extracted from malarial infected blood smears. The whole set was randomly divided into training (40%), validation (30%) and test (30%) sets. Training and validation sets were used for parameter tuning of the classifiers while the test set was used for the final evaluation of the method performance. Using the strategy of classification presented above, this approach achieves good performance in classification of the different stages of infection on single erythrocytes extracted from malarial blood smears. Table 1 presents result obtained for the two levels of classification. Each Fβ value corresponds to the average of 10 experiments from a 10-fold cross validation process.

Table 1. Fβ measures for training, validation and test sets of the automatic infection stage Infection Detection

Ring stage

Throphozoite

Schizont

Training Set

0.937

0.973

0.771

0.780

Validation Set

0.954

0.923

0.519

0.739

Test Set

0.961

0.947

0.677

0.882

346

FUTURE RESEARCH DIRECTIONS An open issue in automatic hematological diagnosis is the development of algorithms that perform accurate segmentation of single cells and their components, and splitting of the cell clusters. Likewise, an emerging trend is the development of feature extraction techniques that allow a suitable object representation, which is less dependent on the quality of the segmentation. Generalization of learning models is another open research question. Approaches for classifying many types of cells have been developed, however most of them are very specialized. The framework for analyzing these images is very similar and the main difference lies in the features used for describing the objects, and the classification models varying by the training data. Modern machine learning approaches include on-line learning classifiers, which can dynamically extract object features and perform classification without retraining the learning model. These technique combined with robust descriptors can be used for constructing a reliable hematological analysis system.

SUMMARY AND CONCLUSION Automatic differential diagnosis of blood and bone marrow smears is a typical problem of pattern recognition, which is usually realized in four stages: preprocessing, segmentation, feature extraction/ selection, and classification. This chapter presents a comprehensive review of the methods proposed in the literature for performing each stage. The accuracy of computer-assisted hematology image analysis depends on the slide preparation, staining procedures and digitization settings. Consequently, algorithm efficiency may be enhanced by developing optimized acquisition protocols. Unfortunately there exist few system studies on this issue for this kind of images.

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

REFERENCES Aksoy, S. (2004). Interactive classification and content-based retrieval of tissue images. In Proceedings from Conference on Applications of digital image processing, Seattle, WA. Angulo, J., & Flandrin, G. (2003). Microscopic image analysis using mathematical morphology: Application to haematological cytology. In A. Mendez-Vilas (Ed.), Science, Technology and Education of Microscopy: An overview (pp. 304312). Padajoz, Spain: FORMATEX. Angulo, J., & Flandrin, G. (2003). Automated detection of working area of peripheral blood smears using mathematical morphology. Analytical Cellular Pathology, 25(1), 37–49. Angulo, J., & Serra, J. (2002). Morphological Color Size Distributions for Image Classification And Retrieval. In Proceedings of Advanced Concepts for Intelligent Vision Systems Conference, Ghent, Belgium. Anoraganingrum, D. (1999). Cell Segmentation With Median Filter and Mathematical Morphology Operation. In Proceedings of the International Conference on Image Analysis and Processing. Bacus, J. W., Belanger, M. G., Aggarwal, R. K., & Trobaugh, F. E. (1976). Image processing for automated erythrocyte classification. The Journal of Histochemistry and Cytochemistry, 24, 195–201. Ballaró, B., Florena, A. M., Franco, V., & Tegolo, D. (2008). An automated image analysis methodology for classifying megakaryocytes in chronic myeloproliferative disorders. In Medical Image Analysis. Beksac, M., Beksac, M., Tipi, V., Duru, H., Karaks, M., & Cakar, A. (1997). An artificial intelligent diagnostic system on differential recognition of hematopoietic cells from microscopic images. Cytometry, 30, 145–150. doi:10.1002/ (SICI)1097-0320(19970615)30:33.0.CO;2-K

Bentley, S., & Lewis, S. (1975). The use of an image analyzing computer for the quantification of red cell morphological characteristics. British Journal of Haematology, 29, 81–88. doi:10.1111/j.1365-2141.1975.tb01801.x Bikhet, S. F., Darwish, A. M., Tolba, H. A., & Shaheen, S. I. (2000). Segmentation and classification of white blood cells. In Proceedings of the International Conference on Acoustics, Speech, and Signal Processing. Bishop, C. M. (1995). Neural Networks for Pattern Recognition. Oxford, UK: Oxford University Press. Ceelie, H., Dinkelaar, R. D., & Gelder, W. V. (2006). Examination of peripheral blood films using automated microscopy; evaluation of Diffmaster Octavia and Cellavision DM96. Journal of Clinical Pathology, 60, 72–79. doi:10.1136/ jcp.2005.035402 Comaniciu, D., & Meer, P. (2001). Cell image segmentation for diagnostic pathology. In J. S. Sigh & K. Setarehdan (Eds.), Advanced Algorithmic Approaches to Medical Image Segmentation: State-Of-The-Art Applications in Cardiology, Neurology, Mammography and Pathology (pp. 541-558). Berlin, Germany: Springer. Comaniciu, D., Meer, P., Foran, D., & Medl, A. (1998). Bimodal system for interactive indexing and retrieval of pathologyimages. In Applications of Computer Vision. Comaniciu, D., Meer, P., & Foran, D. J. (1999). Image-guided decision support system for pathology. Machine Vision and Applications, 11, 213–224. doi:10.1007/s001380050104 Cseke, I. (1992). A fast segmentation scheme for white blood cell images. In Proceedings of the IAPR International Conference on Image, Speech and Signal Analysis.

347

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

de Berg, M., van Kreveld, M., Overmans, M., & Schwarzkopf, O. (2000). Convex Hulls: Mixing Things. In Computational Geometry: Algorithms and Applications (2nd rev. ed, pp. 235-250). Berlin, Germany: Springer-Verlag. Di Ruberto, C., Dempster, A., Khan, S., & Jarra, B. (2000). Automatic thresholding of infected blood images using granulometryand regional extrema. In Proceedings of the International Conference on Pattern Recognition. Di Ruberto, C., Dempster, A., Khan, S., & Jarra, B. (2002). Analysis of infected blood cell images using morphological operators. Image and Vision Computing, 20(2), 133–146. doi:10.1016/S02628856(01)00092-0 Díaz, G., Gonzalez, F. A., & Romero, E. (2007). Infected Cell Identification in Thin Blood Images Based on Color Pixel Classification: Comparison and Analysis. In Progress in Pattern Recognition, Image Analysis and Applications (LNCS 4756). Díaz, G., Gonzalez, F. A., & Romero, E. R. (2007a). Automatic Clump Splitting for Cell Quantification in Microscopical Images. In Progress in Pattern Recognition, Image Analysis and Applications (LNCS 4756). Díaz, G., González, F. A., & Romero, E. (2009). A Semi-Automatic Method for Quantification and Classification of Erythrocytes Infected with Malaria Parasites in Microscopic Images. Journal of Biomedical Informatics, 42, 296–307. doi:10.1016/j.jbi.2008.11.005 Djemal, K., Chettaoui, C., & Maaref, H. (2005). Shapes description for cells sickle illness Recognition. In Proceedings of the IEEE International Conference on Systems, Signals & Devices, Communication and Signal Processing. Dorini, L. B., Minetto, R., & Leite, N. J. (2007). White blood cell segmentation using morphological operators and scale-space analysis. In Proceedings of the Brazilian Symposium on Computer Graphics and Image Processing. 348

Dougherty, E. R., & Lotufo, R. A. (2003). Handson Morphological Image Processing. Bellingham, WA: SPIE Press. Halim, S., Bretschneider, T. R., Li, Y., Preiser, P. R., & Kuss, C. (2006). Estimating Malaria Parasitaemia from Blood Smear Images. In Proceedings of the International Conference on Control, Automation, Robotics and Vision. Haralick, R. M. (1979). Statistical and Structural Approaches to Texture. Proceedings of the IEEE, 67, 786–808. doi:10.1109/PROC.1979.11328 Harms, H., Aus, H. M., Haucke, M., & Gunzer, U. (1986). Segmentation of stained blood cell images measured at high scanning density with high magnification and high numerical aperture optics. Cytometry, 7, 522–531. doi:10.1002/ cyto.990070605 Heijden, F. V. D. (1995). Image Based Measurement Systems, Object Recognition and Parameter Estimation. New York: John Wiley & Sons. Heijmans, H. (1995). Composing morphological filters. Amsterdam: Centrum voor Wiskunde en Informatica. Hengen, H., Spoor, S., & Pandit, M. (2002). Analysis of blood and bone marrow smears using digital image processing techniques. Progress in biomedical optics and imaging, 3(2), 624-635. Hoffbrand, V., Moss, P., & Pettit, J. (2006). Essential haematology. Malden, MA: Blackwell. Houwen, B. (2000). Blood Film Preparation and Staining Procedures. Clinics in Laboratory Medicine, 22, 1–14. doi:10.1016/S02722712(03)00064-7 Jenkins, C., & Hewamana, S. (2008). The blood film as a diagnostic tool. British Journal of Hospital Medicine, 69, M144–M147.

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Jiang, K., Liao, Q., & Dai, S. (2003). A novel white blood cell segmentation scheme using scale-space filtering and watershed clustering. In Proceedings of the International Conference on Machine Learning and Cybernetics. Jiang, K., Liao, Q., & Xiong, Y. (2006). A novel white blood cell segmentation scheme based on feature space clustering. Soft Computing, 10, 12–19. doi:10.1007/s00500-005-0458-z Katz, A. R. J. (2000). Image Analysis and Supervised Learning in the Automated Differentiation of White Blood Cells from Microscopic Imag. Melbourne, Australia: RMIT University. Kovalev, V. A., Grigoriev, A. Y., & Ahn, H. (1996). Robust Recognition of White Blood Cell Images. In Proceedings of the 13th International Conference on Pattern Recognition (pp. 371-375). Kuhl, F. P., & Giardina, C. R. (1982). Elliptic Fourier features of a closed contour. Comp Graphics Image Processing, 18, 236–258. doi:10.1016/0146-664X(82)90034-X Kumar, B. R. J. (2002). Teager energy based blood cell segmentation. In Proceedings of the International Conference on Digital Signal Processing. Kyungsu, K., Jeonghee, J., Wankyoo, C., Pankoo, K., & Yo-Sung, H. (2001). Automatic cell classification in human’s peripheral blood images based on morphological image processing. In Proceedings of the Australian joint conference on artificial intelligence.

Lezoray, O., Elmoataz, A., Cardot, H., & Revenu, M. (1999). Segmentation of cytological images using color and mathematical morphology. Acta Stereologica, 18, 1–14. Lin, W., Xiao, J., & Micheli-Tzanakou, E. (1998). A computational intelligence system for cell classification. In Proceedings of the IEEE International Conference on Information Technology Applications in Biomedicine. Lindblad, J. (2002). Development of algorithms for digital image cytometry. Uppsala University, Faculty of Science and Technology. Liu, L., & Sclaroff, S. (2001). Medical image segmentation and retrieval via deformable models. In Proceedings of the International Conference on Image Processing. Lucchese, L., & Mitra, S. (2001). Color image segmentation: A state-of-the-art survey. In Proceedings of the Indian National Science Academy. Malpica, N., Solorzano, C. O. D., Vaquero, J. J., Santos, A., Vallcorba, I., & GarcíaSagredo, J. M. (1997). Applying Watershed Algorithms to the Segmentation of Clustered Nuclei. Cytometry, 28, 289–297. doi:10.1002/ (SICI)1097-0320(19970801)28:43.0.CO;2-7 Markiewicz, T., & Osowski, S. (2006). Data mining techniques for feature selection in blood cell recognition. In Proceedings of the European Symposium on Artificial Neural Networks.

Le, M., Bretschneider, T. R., Kuss, C., & Preiser, P. R. (2008). A novel semi-automatic image processing approach to determine Plasmodium falciparum parasitemia in Giemsa-stained thin blood smears. BMC Cell Biology, January, 9.

Markiewicz, T., Osowski, S., Marianska, B., & Moszczynski, L. (2005). Automatic recognition of the blood cells of myelogenous leukemia using SVM. In Proceedings of the IEEE International Joint Conference on Neural Networks.

Lezoray, O., & Cardot, H. (2002). Cooperation of color pixel classification schemes and color watershed: a study for microscopic images. IEEE Transactions on Image Processing, 11, 783–789. doi:10.1109/TIP.2002.800889

Micheli-Tzanakou, E., Sheikh, H., & Zhu, B. (1997). Neural Networks and Blood Cell Identification. Journal of Medical Systems, 21, 201–210. doi:10.1023/A:1022899519704

349

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Miller, M. N. (1972). Leukocyte classification by morphological criteria. In Proceedings of the Engineering Foundation Conference on Automatic Cytology. Mircic, S., & Jorgovanovic, N. (2006). Automatic Classification of Leukocytes. Journal of Automatic Control, 16, 29–32. doi:10.2298/JAC0601029M Mohana Rao, K. N., & Dempster, A. G. (2002). Modification on distance transform to avoid over-segmentation and under-segmentation. In Proceedings of the International Symposium on VIPromCom. Mohana Rao, K. N., & Dempster, A. G. (2002). Use of area-closing to improve granulometry performance. In Proceedings of the International Symposium on VIPromCom. Molina, L. C., Belanche, L., & Nebot, A. (2002). Feature selection algorithms: a survey and experimental evaluation. In Proceedings of the IEEE International Conference on Data Mining, 2002. Nilsson, B., & Heyden, A. (2001). Segmentation of dense leukocyte clusters. In Proceedings of the IEEE Workshop on Mathematical Methods in Biomedical Image Analysis. Nilsson, B., & Heyden, A. (2005). Segmentation of complex cell clusters in microscopic images: application to bone marrow samples. Cytometry, 66, 24–31. doi:10.1002/cyto.a.20153 Ongun, G., Halici, U., Leblebicioglu, K., Atalay, V., & Beksac, M. (2001). An Automated Differential Blood Count System. In Proceedings of the EMBS International Conference. Ongun, G., Halici, U., Leblebicioglu, K., Atalay, V., Beksac, M., & Beksac, S. (2001). Feature extraction and classification of blood cells for anautomated differential blood count system. Proceedings of the International Joint Conference on Neural Networks.

350

Ongun, G., Halici, U., Leblebicioglu, K., Atalay, V., Beksac, S., & Beksac, M. (2001). Automated Contour Detection in Blood Cell Images by An Efficient Snake Algorithm. Nonlinear Analysis: Series B –Applications, 47, 5839–3847. doi:10.1016/ S0362-546X(01)00707-6 Otsu, N. (1979). A Tlreshold Selection Method from Gray-Level Histograms. IEEE Transactions On Systrems. Man And Cybernetics, 9, 62–66. doi:10.1109/TSMC.1979.4310076 Pan, C., Fang, Y., Yan, X., & Zheng, C. (2006). Robust Segmentation for Low Quality Cell Images from Blood and Bone Marrow. International Journal of Control, Automation, and Systems, 4, 637–644. Pan, C., Yan, X., & Zheng, C. (2006). Recognition of Blood and Bone Marrow Cells using Kernelbased Image Retrieval. International Journal of Computer Science and Network Security, 6, 29–35. Pan, C., Zheng, C., & Wang, H. (2003). Robust color image segmentation based on mean shift and marker-controlled watershed algorithm. In Proceedings of the International Conference on Machine Learning and Cybernetics. Park, J., & Keller, J. (1997). Fuzzy Patch Label Relaxation in Bone Marrow Cell Segmentation. In Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics. Piuri, V., & Scotti, F. (2004). Morphological classification of blood leucocytes by microscope images. In Proceedings of the IEEE International Conference on Computational Intelligence for Measurement Systems and Applications. Plataniotis, K. N., & Venetsanopoulos, A. N. (2001). Color Image Processing and Applications. New York: Springer - Verlag. Poon, S. S., Ward, R. K., & Palcic, B. (1992). Automated image detection and segmentation in blood smears. Cytometry, 13, 766–774. doi:10.1002/ cyto.990130713

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Ross, N. E., Pritchard, C. J., Rubin, D. M., & Dusé, A. G. (2006). Automated image processing method for the diagnosis and classification of malaria on thin blood smears. Medical & Biological Engineering & Computing, 44, 427–436. doi:10.1007/ s11517-006-0044-2

Sio, S. W. S., Sun, W., Kumar, S., Bin, W. Z., Tan, S. S., & Ong, S. H. (2007). MalariaCount: An image analysis-based program for the accurate determination of parasitemia. Journal of Microbiological Methods, 68, 11–18. doi:10.1016/j. mimet.2006.05.017

Sabino, D. M., Costa, L. F., Rizzatti, E. G., & Zago, M. A. (2004a). Toward leukocyte recognition using morphometry, texture and color. In Proceedings of the International Symposium on Biomedical Imaging: Nano to Macro.

Siroic, R., Osowski, S., Markiewicz, T., & Siwek, K. (2007). Support Vector Machine and Genetic Algorithm for Efricient Blood Cell Recognition. In Proceedings of the Instrumentation and Measurement Technology Conference.

Sabino, D. M. U., Costa, L. D. F., Rizzatti, E. G., & Zago, M. A. (2004). A texture approach to leukocyte recognition. Real-Time Imaging, 10, 205–216. doi:10.1016/j.rti.2004.02.007

Song, X., Abu-Mostafa, Y., Sill, J., & Kasdan, H. (1998). Incorporating contextual information in white blood cell identification. In Proceedings of the Conference on advances in neural information processing systems.

Saeys, Y., Inza, I., & Larrañaga, P. (2007). A review of feature selection techniques in bioinformatics. Bioinformatics (Oxford, England), 23, 2507–2517. doi:10.1093/bioinformatics/btm344 Scotti, F. (2006). Robust Segmentation and Measurements Techniques of White Cells in Blood Microscope Images. In Proceedings of the Instrumentation and Measurement Technology Conference. Serra, J. (1982). Image Analysis and Mathematical Morphology (Vol. 1). Orlando, FL: Academic Press. Sheikh, H., Zhu, B., & Micheli-Tzanakou, E. (1996). Blood cell identification using neural networks. In Proceedings of the 1996 IEEE Twenty-Second Annual Northeast Bioengineering Conference. Sinha, N., & Ramakrishnan, A. G. (2003). Automation of differential blood count. In Proceedings of the Conference on Convergent Technologies for Asia-Pacific Region.

Stanislaw, O., Tomasz, M., Bozena, M., & Leszek, M. (2004). Feature generation for the cell image recognition of myelogenous leukemia. In Proceedings of the European Signal Processing Conference. Tek, F., Dempster, A., & Kale, I. (2006). Malaria Parasite Detection in Peripheral Blood Images. In Proceeding of the British Machine Vision Conference. Tek, F. B., Dempster, A. G., & Kale, I. (2006). A Colour Normalization Method for Giemsa-Stained Blood Cell Images. In Proceedings of the IEEE 14th Signal Processing and Communications Applications. Theera-Umpon, N. (2005). Patch-Based White Blood Cell Nucleus Segmentation Using Fuzzy Clustering. Ecti Transactions On Electrical Eng., Electronics. Communications, 3, 15–19. Theera-Umpon, N. (2007). Automatic White Blood Cell Classification in Bone Marrow Images using Morphological Features. In Proceedings of the 25th Electrical Engineering Conference.

351

Automatic Analysis of Microscopic Images in Hematological Cytology Applications

Theera-Umpon, N., & Dhompongsa, S. (2007). Morphological Granulometric Features of Nucleus in Automatic Bone Marrow White Blood Cell Classification. IEEE Transactions on Information Technology in Biomedicine, 11, 353–359. doi:10.1109/TITB.2007.892694 Theera-Umpon, N., & Gader, P. D. (2002). System-Level Training of Neural Networks for Counting White Blood Cells. IEEE Transactions on Systems, Man and Cybernetics. Part C, Applications and Reviews, 32, 48–53. doi:10.1109/ TSMCC.2002.1009139 Theerapattanakul, J., Plodpai, J., & Pintavirooj, C. (2004). An efficient method for segmentation step of automated white blood cell classifications. In Proceedings of the IEEE Region 10 Conference TENCON. Tuzel, O., Yang, L., Meer, P., & Foran, D. J. (2007). Classification of Hematologic Malignancies using Texton Signatures. Pattern Analysis & Applications, 10, 277–290. doi:10.1007/s10044007-0066-x Wang, W., Song, H., & Zhao, Q. (2006). A modified Watersheds Image Segmentation Algorithm for Blood Cell. In Proceedings of the International Conference on Communications, Circuits and Systems.

Wu, J., & Zeng, P. Z. (2006). A novel color image segmentation method and its application to white blood cell image analysis. In Proceedings of the International Conference on Signal Processing. Yang, L., Meer, P., & Foran, D. J. (2005). Unsupervised Segmentation Based on Robust Estimation and Color Active Contour Models. IEEE Transactions on Information Technology in Biomedicine, 9, 475–486. doi:10.1109/TITB.2005.847515 Zamani, F., & Safabakhsh, R. (2006). An unsupervised GVF snake approach for white blood cell segmentation based on nucleus. In Proceedings of the International Conference on Signal Processing. Zhang, D., & Lu, G. (2004). Review of shape representation and description techniques. Pattern Recognition, 37, 1–19. doi:10.1016/j.patcog.2003.07.008

ENDNOTE 1



In some cases only mean and standard deviation are utilized together with features from other categories such as texture or geometrical measures (Beksak, 1997; Bikhet, 2000; Hengen, 2002.)

This work was previously published in Biomedical Image Analysis and Machine Learning Technologies: Applications and Techniques, edited by Fabio A. Gonzalez and Eduardo Romero, pp. 167-196, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

352

353

Chapter 2.7

Computational Methods in Biomedical Imaging Michele Piana Universita’ di Verona, Italy

INTRODUCTION Biomedical imaging represents a practical and conceptual revolution in the applied sciences of the last thirty years. Two basic ingredients permitted such a breakthrough: the technological development of hardware for the collection of detailed information on the organ under investigation in a less and less invasive fashion; the formulation and application of sophisticated mathematical tools for signal processing within a methodological setting of truly interdisciplinary flavor. A typical acquisition procedure in biomedical imaging requires the probing of the biological tissue by means of some emitted, reflected

or transmitted radiation. Then a mathematical model describing the image formation process is introduced and computational methods for the numerical solution of the model equations are formulated. Finally, methods based on or inspired by Artificial Intelligence (AI) frameworks like machine learning are applied to the reconstructed images in order to extract clinically helpful information. Important issues in this research activity are the intrinsic numerical instability of the reconstruction problem, the convergence properties and the computational complexity of the image processing algorithms. Such issues will be discussed in the following with the help of several examples of notable significance in the biomedical practice.

DOI: 10.4018/978-1-60960-561-2.ch207

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Computational Methods in Biomedical Imaging

BACKGROUND The first breakthrough in the theory and practice of recent biomedical imaging is represented by X-ray Computerized Tomography (CT) (Hounsfield, 1973). On October 11 1979 Allan Cormack and Godfrey Hounsfield gained the Nobel Prize in medicine for the development of computer assisted tomography. In the press release motivating the award, the Nobel Assembly of the Karolinska Institut wrote that in this revolutionary diagnostic tool “the signals[...]are stored and mathematically analyzed in a computer. The computer is programmed to reconstruct an image of the examined cross-section by solving a large number of equations including a corresponding number of unknowns”. Starting from this crucial milestone, biomedical imaging has represented a lively melting pot of clinical practice, experimental physics, computer science and applied mathematics, providing mankind of numerous non-invasive and effective instruments for early detection of diseases, and scientist of a prolific and exciting area for research activity. The main imaging modalities in biomedicine can be grouped into two families according to the kind of information content they provide. •

354

Structural imaging: the image provides information on the anatomical features of the tissue without investigating the organic metabolism. Structural modalities are typically characterized by a notable spatial resolution but are ineffective in reconstructing the dynamical evolution of the imaging parameters. Further to X-ray CT, other examples of such approach are Fluorescence Microscopy (Rost & Oldfield, 2000), Ultrasound Tomography (Greenleaf, Gisvold & Bahn, 1982), structural Magnetic Resonance Imaging (MRI) (Haacke, Brown, Venkatesan & Thompson, 1999) and some kinds of

•

prototypal non-linear tomographies like Microwave Tomography (Boulyshev, Souvorov, Semenov, Posukh & Sizov, 2004), Diffraction Tomography (Guo & Devaney, 2005), Electrical Impedance Tomography (Cheney, Isaacson & Newell, 1999) and Optical Tomography (Arridge, 1999). Functional imaging: during the acquisition many different sets of signals are recorded according to a precisely established temporal paradigm. The resulting images can provide information on metabolic deficiencies and functional diseases but are typically characterized by a spatial resolution which is lower (sometimes much lower) than the one of anatomical imaging. Emission tomographies like Single Photon Emission Computerized Tomography (SPECT) (Duncan, 1997) or Positron Emission Tomography (PET) (Valk, Bailey, Townsend & Maisey, 2004) and Magnetic Resonance Imaging in its functional setup (fMRI) (Huettel, Song & McCarthy, 2004) are examples of these dynamical techniques together with Electroand Magnetoencephalography (EEG and MEG) (Zschocke & Speckmann, 1993; Hamalainen, Hari, Ilmoniemi, Knuutila & Lounasmaa, 1993), which reproduce the neural activity at a millisecond time scale and in a completely non-invasive fashion.

In all these imaging modalities the correct mathematical modeling of the imaging problem, the formulation of computational algorithms for the solution of the model equations and the application of image processing algorithms for data interpretation are the crucial steps which allow the exploitness of the visual information from the measured raw data.

Computational Methods in Biomedical Imaging

MAIN FOCUS From a mathematical viewpoint the inverse problem of synthesizing the biological information in a visual form from the collected radiation is characterized by a peculiar pathology. The concept of ill-posedness has been introduced by Jules Hadamard (Hadamard, 1923) to indicate mathematical problems whose solution does not exist for all data, or is not unique or does not depend uniquely on the data. In biomedical imaging this last feature has particularly deleterious consequences: indeed, the presence of measurement noise in the raw data may produce notable numerical instabilities in the reconstruction when naive approaches are applied. Most (if not all) biomedical imaging problems are ill-posed inverse problems (Bertero & Boccacci, 1998) whose solution is a difficult mathematical task and often requires a notable computational effort. The first step toward the solution is represented by an accurate modeling of the mathematical relation between the biological organ to be imaged and the data provided by the imaging device. Under the most general assumptions the model equation is a non-linear integral equation, although, for several devices, the non-linear imaging equation can be reliably approximated by a linear model where the integral kernel encodes the impulse response of the instrument. Such linearization can be either performed through a precise technological realization, like in MRI, where acquisition is designed in such a way that the data are just the Fourier Transform of the object to be imaged; or obtained mathematically, by applying a sort of perturbation theory to the non-linear equation, like in diffraction tomography whose model comes from the linearization of the scattering equation. The second step toward image reconstruction is given by the formulation of computational methods for the reduction of the model equation. In the case of linear ill-posed inverse problems, a wellestablished regularization theory exists which

attenuates the numerical instability related to illposedness maintaining the biological reliability of the reconstructed image. Regularization theory is at the basis of most linear imaging modalities and regularization methods can be formulated in both a probabilistic and a deterministic setting. Unfortunately an analogously well- established theory does not exist in the case of non-linear imaging problems which therefore are often addressed by means of ‘ad hoc’ techniques. Once an image has been reconstructed from the data, a third step has to be considered, i.e. the processing of the reconstructed images for the extraction and interpretation of their information content. Three different problems are typically addressed at this stage: •

•

•

Edge detection (Trucco & Verri, 1998). Computer vision techniques are applied in order to enhance the regions of the image where the luminous intensity changes sharply. Image integration (Maintz & Viergever, 1998). In the clinical workflow several images of a patient are taken with different modalities and geometries. These images can be fused in an integrated model by recovering changes in their geometry. Image segmentation (Acton & Ray, 2007). Partial volume effects make the interfaces between the different tissues extremely fuzzy, thus complicating the clinical interpretation of the restored images. An automatic procedure for the partitioning of the image in homogeneous pixel sets and for the classification of the segmented regions is at the basis of any Computed Aided Diagnosis and therapy (CAD) software.

AI algorithms and, above all, machine learning play a crucial role in addressing these image processing issues. In particular, as a subfield of machine learning, pattern recognition provides a sophisticated description of the data which, in

355

Computational Methods in Biomedical Imaging

medical imaging, allows to locate tumors and other pathologies, measure tissue dimensions, favor computer-aided surgery and study anatomical structures. For example, supervised approaches like backpropagation (Freeman & Skapura, 1991) or boosting (Shapire, 2003) accomplish classification tasks of the different tissues from the knowledge of previously interpreted images; while unsupervised methods like Self-Organizing Maps (SOM) (Kohonen, 2001), fuzzy clustering (De Oliveira & Pedrycz, 2007) and ExpectationMaximization (EM) (McLachlan & Krishnan, 1996) infer probabilistic information or identify clustering structures in sets of unlabeled images. From a mathematical viewpoint, several of these methods correspond more to heuristic recipes than to rigorously formulated and motivated procedures. However, since the last decade the theory of statistical learning (Vapnik, 1998) has appeared as the best candidate for a rigorous description of machine learning within a functional analysis framework.

FUTURE TRENDS Among the main goals of recent biomedical imaging we point out the realization of •

•

•

microimaging techniques which allow the investigation of biological tissues of micrometric size for both diagnostic and research purposes; hybrid systems combining information from different modalities, possibly anatomical and functional; highly non-invasive diagnostic tools, where even a modest discomfort is avoided.

These goals can be accomplished only by means of an effective interplaying of hardware development and application of innovative image processing algorithms. For example, microtomography for biological samples requires the introduc-

356

tion of both new X-ray tubes for data acquisition and computational methods for the reduction of beam hardening effects; electrophysiological and structural information on the brain can be collected by performing an EEG recording inside an MRI scanning but also using the structural information from MRI as a prior information in the analysis of the EEG signal accomplished in a Bayesian setting; finally, non- invasivity in colonoscopy can be obtained by utilizing the most recent acquisition design in X-ray tomography together with sophisticated softwares which allow virtual navigation within the bowel, electronic cleansing and automatic classification of cancerous and healthy tissues. From a purely computational viewpoint, two important goals in machine learning applied to medical imaging are the development of algorithms for semi-supervised learning and for the automatic integration of genetic data with information coming from the acquired imagery.

CONCLUSION Some aspects of recent biomedical imaging have been described from a computational science perspective. The biomedical image reconstruction problem has been discussed as an ill-posed inverse problem where the intrinsic numerical instability producing image artifacts can be reduced by applying sophisticated regularization methods. The role of image processing based on machine learning techniques has been described together with the main goals of recent biomedical imaging applications.

REFERENCES Acton, S. T., & Ray, N. (2007). Biomedical Image Analysis: Segmentation. Princeton: Morgan and Claypool.

Computational Methods in Biomedical Imaging

Arridge, S. R. (1999). Optical Tomography in Medical Imaging. [Evolutionary Programming, Genetic Algorithms. Oxford University Press.]. Inverse Problems, 15, R41–R93. doi:10.1088/02665611/15/2/022 Bertero, M., & Boccacci, P. (1998). Introduction to Inverse Problems in Imaging. Bristol: IOP. Boulyshev, A. E., Souvorov, A. E., Semenov, S. Y., Posukh, V. G., & Sizov, Y. E. (2004). Threedimensional Vector Microwave Tomography: Theory and Computational Experiments. Inverse Problems, 20, 1239–1259. doi:10.1088/02665611/20/4/013 Cheney, M., Isaacson, D., & Newell, J. C. (1999). Electrical Impedance Tomography. SIAM Review, 41, 85–101. doi:10.1137/S0036144598333613 De Oliveira, J. V., & Pedrycz, W. (2007). Advances in Fuzzy Clustering and its Applications. San Francisco: Wiley. Duncan, R. (1997). SPECT Imaging of the Brain. Amsterdam: Kluwer. Freeman, J. A., & Skapura, D. M. (1991). Neural Network Algorithms: Applications and Programming Techniques. Redwood City: AddisonWesley. Greenleaf, J., Gisvold, J. J., & Bahn, R. (1982). Computed Transmission Ultrasound Tomography. Medical Progress Through Technology, 9, 165–170.

Hadamard, J. (1923). Lectures on Cauchy’s Problem in Partial Differential Equations. Yale: Yale University Press. Hamalainen, M., Hari, R., Ilmoniemi, R. J., Knuutila, J., & Lounasmaa, O. V. (1993). Magnetoencephalography: theory, instrumentation and applications to non-invasive studies of the working human brain. Reviews of Modern Physics, 65, 413–497. doi:10.1103/RevModPhys.65.413 Hounsfield, G. N. (1973). Computerised Transverse Axial Scanning (Tomography). I: Description of System. The British Journal of Radiology, 46, 1016–1022. doi:10.1259/0007-1285-46-5521016 Huettel, S. A., Song, A. W., & McCarthy, G. (2004). Functional Magnetic Resonance Imaging. Sunderland: Sinauer Associates. Kohonen, T. (2001). Self-Organizing Maps. Berlin: Springer. Maintz, J., & Viergever, M. (1998). A Survey of Medical Imaging Registration. Medical Imaging Analysis, 2, 1–36. doi:10.1016/S13618415(01)80026-8 McLachlan, G., & Krishnan, T. (1996). The EM Algorithm and Extensions. San Francisco: John Wiley. Rost, F., & Oldfield, R. (2000). Fluorescence Microscopy: Photography with a Microscope. Cambridge: Cambridge University Press.

Guo, P., & Devaney, A. J. (2005). Comparison of Reconstruction Algorithms for Optical Diffraction Tomography. Journal of the Optical Society of America. A, Optics, Image Science, and Vision, 22, 2338–2347. doi:10.1364/JOSAA.22.002338

Shapire, R. E. (2003). The Boosting Approach to Machine Learning: an Overview. Nonlinear Estimation and Classification, Denison, D. D., Hansen, M. H., Holmes, C., Mallik, B. & Yu, B. editors. Berlin: Springer.

Haacke, E. M., Brown, R. W., Venkatesan, R., & Thompson, M. R. (1999). Magnetic Resonance Imaging: Physical Principles and Sequence Design. San Francisco: John Wiley.

Trucco, E., & Verri, A. (1998). Introductory Techniques for 3D Computer Vision. Englewood Cliffs: Prentice Hall.

357

Computational Methods in Biomedical Imaging

Valk, P. E., Bailey, D. L., Townsend, D. W., & Maisey, M. N. (2004). Positron Emission Tomography. Basic Science and Clinical Practice. Berlin: Springer. Vapnik, V. (1998). Statistical Learning Theory. San Francisco: John Wiley. Zschocke, S., & Speckmann, E. J. (1993). Basic Mechanisms of the EEG. Boston: Birkhaeuser.

KEY TERMS AND DEFINITIONS Computer Aided Diagnosis (CAD): The use of computers for the interpretation of medical images. Automatic segmentation is one of the crucial task of any CAD product. Edge Detection: Image processing technique for enhancing the points of an image at which the luminous intensity changes sharply. Electroencephalography (EEG): Noninvasive diagnostic tool which records the cerebral electrical activity by means of surface electrodes placed on the skull. Ill-Posedness: Mathematical pathology of differential or integral problems, whereby the solution of the problem does not exist for all data, or is not unique or does not depend continuously on the data. In computation, the numerical effects of ill-posedness are reduced by means of regularization methods.

Image Integration: In medical imaging, combination of different images of the same patient acquired with different modalities and/or according to different geometries. Magnetic Resonance Imaging (MRI): Imaging modality based on the principles of nuclear magnetic resonance (NMR), a spectroscopic technique used to obtain microscopic chemical and physical information about molecules. MRI can be applied in both functional and anatomical settings. Magnetoencephalography (MEG): Noninvasive diagnostic tool which records the cerebral magnetic activity by means of superconducting sensors placed on a helmet surrounding the brain. Segmentation: Image processing technique for distinguishing the different homogeneous regions in an image. Statistical Learning: Mathematical framework which utilizes functional analysis and optimazion tools for studying the problem of inference. Tomography: Imaging technique providing two-dimensional views of an object. The method is used in many disciplines and may utilize input radiation of different nature and wavelength. There exist X-ray, optical, microwave, diffraction and electrical impedance tomographies.

This work was previously published in Encyclopedia of Artificial Intelligence, edited by Juan Ramón Rabuñal Dopico, Julian Dorado and Alejandro Pazos, pp. 372-376, copyright 2009 by Information Science Reference (an imprint of IGI Global).

358

359

Chapter 2.8

Visual Medical Information Analysis Maria Papadogiorgaki Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece Vasileios Mezaris Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece Yiannis Chatzizisis Aristotle University of Thessaloniki, Greece George D. Giannoglou Aristotle University of Thessaloniki, Greece Ioannis Kompatsiaris Informatics and Telematics Institute, Centre for Research and Technology Hellas, Greece

INTRODUCTION Images have constituted an essential data source in medicine in the last decades. Medical images derived from diagnostic technologies (e.g., Xray, ultrasound, computed tomography, magnetic resonance, nuclear imaging) are used to improve the existing diagnostic systems for clinical purposes, but also to facilitate medical research. Hence, medical image processing techniques are constantly investigated and evolved. DOI: 10.4018/978-1-60960-561-2.ch208

Medical image segmentation is the primary stage to the visualization and clinical analysis of human tissues. It refers to the segmentation of known anatomic structures from medical images. Structures of interest include organs or parts thereof, such as cardiac ventricles or kidneys, abnormalities such as tumors and cysts, as well as other structures such as bones, vessels, brain structures and so forth. The overall objective of such methods is referred to as computer-aided diagnosis; in other words, they are used for assisting doctors in evaluating medical imagery or in recognizing abnormal findings in a medical image.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Visual Medical Information Analysis

In contrast to generic segmentation methods, techniques used for medical image segmentation are often application-specific; as such, they can make use of prior knowledge for the particular objects of interest and other expected or possible structures in the image. This has led to the development of a wide range of segmentation methods addressing specific problems in medical applications. In the sequel of this article, the analysis of medical visual information generated by three different medical imaging processes will be discussed in detail: Magnetic Resonance Imaging (MRI), Mammography, and Intravascular Ultrasound (IVUS). Clearly, in addition to the aforementioned imaging processes and the techniques for their analysis that are discussed in the sequel, numerous other algorithms for applications of segmentation to specialized medical imagery interpretation exist.

BACKGROUND Magnetic Resonance Imaging Magnetic Resonance Imaging (MRI) is an important diagnostic imaging technique attending to the early detection of the abnormal conditions in tissues and organs because it is able to reliably identify anatomical areas of interest. In particular for brain imaging, several techniques which perform segmentation of the brain structures from MRIs are applied to the study of many disorders, such as multiple sclerosis, schizophrenia, epilepsy, Parkinson’s disease, Alzheimer’s disease, and so forth. MRI is particularly suitable for brain studies because it is virtually noninvasive, and it achieves a high spatial resolution and high contrast of soft tissues. To achieve the 3D reconstruction of the brain morphology, several of the existing approaches perform segmentation on sequential MR images. The overall process usually includes noise filtering of the images and edge detection for

360

the identification of the brain contour. Following, perceptual grouping of the edge points is applied in order to recover the noncontinuous edges. In many cases, the next step is the recognition of the various connective components among the set of edge points, rejection of the components that consist of the smallest number of points, and use of the finally acquired points for reconstructing the 3D silhouette of the brain, as will be discussed in more detail in the sequel.

Mammography Mammography is considered to be the most effective diagnostic technique for detecting abnormal tissue conditions on women’s breast. Being used both for prevention and for diagnostic purposes, it is a very commonly used technique that produces mammographic images by administering a low-dose of x-ray radiation to the tissue under examination. The analysis of the resulting images aims at the detection of any abnormal structures and the quantification of their characteristics, such as size and shape, often after detecting the pectoral muscle and excluding it from the further processing. Methods for the analysis of mammographic images are presented in the sequel.

Intravascular Ultrasound IVUS is a catheter-based technique that renders two-dimensional images of coronary arteries and therefore provides valuable information concerning luminal and wall area, plaque morphology and wall composition. An example IVUS image, with tags explaining the most important parts of the vessel structure depicted on it, is shown in Figure 1. However, due to their tomographic nature, isolated IVUS images provide limited information regarding the burden of atherosclerosis. This limitation can be overcome through 3D reconstruction techniques in order to stack the sequential 2D images in space, using single-plane

Visual Medical Information Analysis

Figure 1. Example IVUS image with tags explaining the most important parts of the vessel structure depicted on it

VISUAL MEDICAL INFORMATION ANALYSIS TECHNIQUES Magnetic Resonance Imaging Analysis

or biplane angiography for recovering the vessel curvature (Giannoglou et al., 2006; Sherknies, Meunier, Mongrain, & Tardif, 2005; Wahle, Prause, DeJong, & Sonka, 1999). The analysis of IVUS images constitutes an essential step toward the accurate morphometric analysis of coronary plaque. To this end, the processing of IVUS images is necessary so that the regions of interest can be detected. The coronary artery wall mainly consists of three layers: intima, media and adventitia, while three regions are supposed to be visualized as distinguished fields in an IVUS image, namely the lumen, the vessel wall (made of the intima and the media layers) and the adventitia plus surroundings. The above regions are separated by two closed contours: the inner border, which corresponds to the lumen-wall interface, and the outer border representing the boundary between media and adventitia. A reliable and quick detection of these two borders in sequential IVUS images constitutes the basic step towards plaque morphometric analysis and 3D reconstruction of the coronary arteries.

Several techniques have been proposed for the analysis of MR images. In Grau, Mewes, Alcaniz, Kikinis, and Warfield (2004), a modification of a generic segmentation technique, the watershed transform, is proposed for knee cartilage and gray matter/white matter segmentation in MR images. This introduces prior information in the watershed method via the use of a previous probability calculation for the classes present in the image and via the combination of the watershed transform with atlas registration for the automatic generation of markers. As opposed to Grau et al. (2004), other methods are more application specific; in Woolrich, Behrens, Beckmann and Smith (2005), for example, segmentation tools are developed for the study of the function of the brain, that is, for the classification of brain areas as activating, deactivating, or not activating, using functional magnetic resonance imaging (FMRI) data. This method performs segmentation based on intensity histogram information, augmented with adaptive spatial regularization using Markov random fields. The latter contributes to improved segmentation as compared to nonspatial mixture models, while not requiring the heuristic fine-tuning that is necessary for nonadaptive spatial regularization previously proposed. Because MR images contain a significant amount of noise caused by operator performance, equipment, or even the environment, the segmentation on them can lead to several inaccuracies. In order to overcome the effects of noise, Shen, Sandham, Granat and Sterr (2005) propose a segmentation technique based on an extension to the traditional fuzzy c-means (FCM) clustering algorithm. The segmentation performance is improved using neighborhood attraction, which

361

Visual Medical Information Analysis

depends on the relative location and features of neighboring pixels. The degree of attraction is optimized by applying a neural network model. Greenspan, Ruf and Goldberger (2006) have also developed an algorithm for the automated brain tissue segmentation on noisy, low-contrast (MR) images. Under their approach, the brain image is represented by a model that is composed of a large number of Gaussians. For the algorithm’s initialization an atlas or parameter learning are not required. Finally, segmentation of the brain image is achieved by affiliating each voxel to the component of the model that maximizes an a posteriori probability. In Valdes-Cristerna, Medina-Banuelos and Yanez-Suarez (2004) a hybrid model for the segmentation of brain MRI has been investigated. The model includes a radial basis network and an active contour model. The radial basis network algorithm generates an initial contour, which is following used by the active contour model to achieve the final segmentation of the brain.

Mammography Image Analysis Several applications have been proposed, which process the mammographic images in order to assist the clinicians in their diagnostic procedure. In Székely, Toth and Pataki (2006) the mammographic images are analyzed using segmentation in order to identify regions of interest. The applied segmentation technique includes texture features, decision trees, and a Markov random field model. The extracted features which refer to the object’s shape and texture parameters are linearly combined to lead to the final decision. Because the pectoral muscle should be excluded from processing on a mammogram intended for the breast tissue, its identification is important. Kwok, Chandrasekhar, Attikiouzel and Rickard (2004) have developed an adaptive segmentation technique for the extraction of the pectoral muscle on digitized mammograms. The method uses knowledge about the position and shape of the pectoral muscle. Other approaches, such as

362

Cascio, Fauci, Magro, Raso, Bellotti, De Carlo et al. (2006) use supervised neural networks for detecting pathological masses in mammograms. A segmentation process provides features of geometrical information, or shape parameters which constitute the input to the neural network that computes the probability of the lesion existence.

IVUS Image and Image Sequence Analysis Traditionally, the segmentation of IVUS images is performed manually, which is a time consuming procedure with results affected by the high inter- and intra-user’s variability. To overcome these limitations, several approaches for semiautomated segmentation have been proposed in the literature. In Herrington, Johnson, Santago and Snyder (1992) after the manual indication of the general location of the boundary of interest by the user, an edge detection filter is applied to find potential edge points within the pointed neighborhood. The extracted image data are used for the estimation of the closed smooth final contour. Sonka, Zhang, Siebes, Bissing, DeJong, Collins et al. (1995) implemented a knowledge-based graph searching method incorporating a priori knowledge on coronary artery anatomy and a selected region of interest prior to the automatic border detection. Quite a few variations of active contour model have been investigated, including the approach of Parissi et al. (2006), where a user interaction is required, by drawing an initial contour as close as possible to its final position. Thus, the active contour is initialized and tends to approximate the final desired border. The active contour or deformable models principles have been used to allow the extraction of the luminal and medial-adventitial borders in three dimensions after setting an initial contour in Kovalski, Beyar, Shofti and Azhari (2000). However, the contour detection fails for low contrast interface regions such as the luminal border where the blood-wall interface in most images

Visual Medical Information Analysis

corresponds to weak pixel intensity variation. In order to improve the included active surface segmentation algorithm for plaque characterization, Klingensmith, Nair, Kuban and Vince (2004) use the frequency information after acquiring the radiofrequency (RF) IVUS data through an electrocardiogram scheme. Radio frequency data are also used in Perrey et al. (2004), after in vivo acquisition for the segmentation of the luminal boundary in IVUS images. According to this approach, tissue describing parameters are directly estimated from RF data. Subsequently, a neuro-fuzzy inference system trained to several parameters is used to distinguish blood from tissue regions. For clinical practice the most attractive approaches are the fully automatic ones. A limited number of them has been developed so far, such as the segmentation based on edge contrast (Zhu, Liang, & Friedman, 2002); the latter is shown to be an efficient feature for IVUS image analysis, in combination with the gray level distribution. Specific automated approaches which utilize the deformable models principles in combination with other various techniques and features reported in the related literature have been investigated. Brusseau, de Korte, Mastik, Schaar and van der Steen (2004) exploited an automatic method for detecting the endoluminal border based on an active contour that evolves until it optimally separates regions with different statistical properties without using a preselected region of interest or initialization of the contour close to its final position. Another automated approach based on deformable models has been reported by Plissiti, Fotiadis, Michalis and Bozios (2004), who employed a Hopfield neural network for the modification and minimization of an energy function as well as a priori vessel geometry knowledge. An automated approach for segmentation of IVUS images based on a variation of an active contour model is presented in Giannoglou et al. (2007). This technique is in vivo evaluated in images originated from human coronary arteries. The initialization of the contours in each IVUS frame

is automatically performed using an algorithm, which is based on the intensity features of the image. The initially extracted boundaries constitute the input to the active contour model, which then deforms the contours appropriately, identifying their correct location on the IVUS frame. A fuzzy clustering algorithm for adaptive segmentation in IVUS images is investigated by Filho, Yoshizawa, Tanaka, Saijo and Iwamoto (2005). Cardinal et al. (2006) present a 3D IVUS segmentation applying Rayleigh probability density functions (PDFs) for modeling the pixel gray value distribution of the vessel wall structures. Other approaches are based on the calculation of the image’s energy. In Luo, Wang and Wang (2003) the lumen area of the coronary artery is estimated using the internal energy, which describes the smoothness of the arterial wall and the external energy which represents the grayscale variation of images that constitute an IVUS video. The minimal energy which defines the contour is obtained using circular dynamic programming. Other methods include statistical analysis, such as Gil, Hernandez, Rodriguez, Mauri and Radeva’s (2006), where the presented approach uses statistical classification techniques for the IVUS border detection. In Papadogiorgaki, Mezaris, Chatzizisis, Kompatsiaris and Giannoglou (2006), a fully automated method for the segmentation of IVUS images and specifically for the detection of luminal and medial-adventitial boundaries is presented. This technique is based on the use of the results of texture analysis, performed by means of a multilevel Discrete Wavelet Frames decomposition. Following image preprocessing, to remove catheter-induced artifacts, a two-step process is employed for the detection of the boundaries of interest. Objective of the first step, termed contour initialization, is the detection of pixels that are likely to belong to the lumen and media-adventitia boundaries, taking into consideration the previously extracted texture features. As a result of this step, initial contours are generated; however, these are not smooth and are characterized by

363

Visual Medical Information Analysis

Figure 2. Sample IVUS image (left) and the corresponding analysis result of a texture-based approach to the detection of luminal and medial-adventitial boundaries (right)

discontinuities, as opposed to the true lumen and media-adventitia boundaries. Thus, at the second step, a filtering or approximation procedure is applied to the initial contour functions, so as to result in the generation of smooth contours that are in good agreement with the true lumen and media-adventitia boundaries. This approach does not require manual initialization of the contours and demonstrates the importance of texture features in IVUS image analysis. A sample IVUS image and the corresponding analysis result of this approach are illustrated in Figure 2.

FUTURE TRENDS With the number of medical imaging techniques used in everyday practice constantly rising and the quality of the results of such imaging techniques constantly increasing, to the benefit of the patients, it is clear that the need for accurate and automated to the widest possible extent analysis of medical images is a key element in supporting the diagnosis and treatment process for a wide range of medical conditions. To this end, future research will continue to concentrate on the development of analysis methods that are automated, robust and reliable. In addition to that, particular emphasis is expected to be put on the coupling of

364

the results of automated analysis with techniques for the formal representation of them. Examples of early systems for the formal representation of knowledge extracted from medical images, though not in an automated manner, include those discussed in Dasmahapatra et al. (2006), Hu, Dasmahapatra, Lewis and Shadbolt (2003). These are concerned with the annotation of medical images used for the diagnosis and management of breast cancer, such as those generated by mammography and MRI, by expressing all the extracted features and regions of interest using domain knowledge and assigning them to specific concepts of a knowledge structure. In an analogous approach, in Gedzelman, Simonet, Bernhard, Diallo and Palmer (2005) a knowledge structure of cardiovascular diseases is constructed in order to be used for the representation of the findings of the relevant imaging techniques, so as to support concept-based information retrieval. Combining automated analysis results with techniques such as those briefly discussed above for the formal representation of them will empower new possibilities in the areas of retrieval in extensive medical databases and reasoning over the results of analysis, consequently providing the physicians not only with the analysis results themselves but also with hints on their meaning, minimizing the risk of misinterpretation.

Visual Medical Information Analysis

CONCLUSION In this article, visual medical information analysis was discussed, starting with an introduction on the current use of medical imaging and the needs for its analysis. The current article then focused on three important medical imaging techniques, namely Magnetic Resonance Imaging (MRI), Mammography, and Intravascular Ultrasound (IVUS), for which a detailed presentation of the goals of analysis and the methods presented in the literature for reaching these goals was given. The future trends identified in the relevant section provide insights on how the algorithms outlined in this article can be further evolved, so as to more efficiently address the problem of medical image analysis and consequently pave the way for the development of innovative doctor decision support applications that will make the most out of the available image data.

REFERENCES Brusseau, E., de Korte, C. L., Mastik, F., Schaar, J., & van der Steen, A. F. W. (2004). Fully automatic luminal contour segmentation in intracoronary ultrasound imaging—a statistical approach. IEEE Transactions on Medical Imaging, 23(5), 554–566. doi:10.1109/TMI.2004.825602 Cardinal, M.-H. R., Meunier, J., Soulez, G., Maurice, R. L., Therasse, E., & Cloutier, G. (2006). Intravascular ultrasound image segmentation: A three-dimensional fast-marching method based on gray level distributions. IEEE Transactions on Medical Imaging, 25(5), 590–601. doi:10.1109/ TMI.2006.872142 Cascio, D., Fauci, F., Magro, R., Raso, G., Bellotti, R., & De Carlo, F. (2006). Mammogram segmentation by contour searching and mass lesions classification with neural network. IEEE Transactions on Nuclear Science, 53(5), 2827–2833. doi:10.1109/TNS.2006.878003

Dasmahapatra, S., Dupplaw, D., Hu, B., Lewis, H., Lewis, P., & Shadbolt, N. (2006). Facilitating multi-disciplinary knowledge-based support for breast cancer screening. International Journal of Healthcare Technology and Management, 7(5), 403–420. Dos, S., Filho, E., Yoshizawa, M., Tanaka, A., Saijo, Y., & Iwamoto, T. (2005, September). Detection of luminal contour using fuzzy clustering and mathematical morphology in intravascular ultrasound images. In Proceedings of the 27th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (IEEE-EMBS), China. Gedzelman, S., Simonet, M., Bernhard, D., Diallo, G., & Palmer, P. (2005, September). Building an ontology of cardiovascular diseases for concept-based information retrieval. Computers in Cardiology, Lyon, France. Giannoglou, G. D., Chatzizisis, Y. S., Koutkias, V., Kompatsiaris, I., Papadogiorgaki, M., & Mezaris, V. (2007). (accepted for publication). A novel active contour model for fully automated segmentation of intravascular ultrasound images: In-vivo validation in human coronary arteries. Computers in Biology and Medicine. Giannoglou, G. D., Chatzizisis, Y. S., Sianos, G., Tsikaderis, D., Matakos, A., & Koutkias, V. (2006). In-vivo validation of spatially correct threedimensional reconstruction of human coronary arteries by integrating intravascular ultrasound and biplane angiography. Coronary Artery Disease, 17(6), 533–543. doi:10.1097/00019501200609000-00007 Gil, D., Hernandez, A., Rodriguez, O., Mauri, J., & Radeva, P. (2006). Statistical strategy for anisotropic adventitia Modelling in IVUS. IEEE Transactions on Medical Imaging, 25(6), 768–778. doi:10.1109/TMI.2006.874962

365

Visual Medical Information Analysis

Grau, V., Mewes, A. U. J., Alcaniz, M., Kikinis, R., & Warfield, S. K. (2004). Improved watershed transform for medical image segmentation using prior information. IEEE Transactions on Medical Imaging, 23(4), 447–458. doi:10.1109/ TMI.2004.824224 Greenspan, H., Ruf, A., & Goldberger, J. (2006). Constrained gaussian mixture model framework for automatic segmentation of MR brain images. IEEE Transactions on Medical Imaging, 25(9), 1233–1245. doi:10.1109/TMI.2006.880668 Herrington, D. M., Johnson, T., Santago, P., & Snyder, W. E. (1992, October). Semi-automated boundary detection for intravascular ultrasound. In Proceedings of Computers in cardiology (pp. 103-106). Durham, NC, USA. Hu, B., Dasmahapatra, S., Lewis, P., & Shadbolt, N. (2003, November). Ontology-based medical image annotation with description logics. In Proceedings of the 15th IEEE International Conference on Tools with Artificial Intelligence, Sacramento, CA, USA. Klingensmith, J. D., Nair, A., Kuban, B. D., & Vince, D. G. (2004, August). Segmentation of three-dimensional intravascular ultrasound images using spectral analysis and a dual active surface model. In Proceedings of the IEEE Ultrasonics Symposium, Montreal, Canada. Kovalski, G., Beyar, R., Shofti, R., & Azhari, H. (2000). Three-dimensional automatic quantitative analysis of intravascular ultrasound images. Ultrasound in Medicine & Biology, 26(4), 527–537. doi:10.1016/S0301-5629(99)00167-2 Kwok, S. M., Chandrasekhar, R., Attikiouzel, Y., & Rickard, M. T. (2004). Automatic pectoral muscle segmentation on mediolateral oblique view mammograms. IEEE Transactions on Medical Imaging, 23(9), 1129–1140. doi:10.1109/TMI.2004.830529

366

Luo, Z., Wang, Y., & Wang, W. (2003). Estimating coronary artery lumen area with optimizationbased contour detection. IEEE Transactions on Medical Imaging, 22(4), 564–566. doi:10.1109/ TMI.2003.811048 Papadogiorgaki, M., Mezaris, V., Chatzizisis, Y. S., Kompatsiaris, I., & Giannoglou, G. D. (2006, September). A fully automated texture-based approach for the segmentation of sequential IVUS images. In Proceedings of the 13th International Conference on Systems, Signals & Image Processing (IWSSIP), Budapest, Hungary. Parissi, E., Kompatsiaris, Y., Chatzizisis, Y. S., Koutkias, V., Maglaveras, N., Strintzis, M. G., et al. (2006, December). An automated model for rapid and reliable segmentation of intravascular ultrasound images. In Proceedings of the 7th International Symposium on Biological and Medical Data Analysis (ISBMDA), Thessaloniki, Greece. Perrey, C., Scheipers, U., Bojara, W., Lindstaedt, M., Holt, S., & Ermert, H. (2004, August). Computerized segmentation of blood and luminal borders in intravascular ultrasound. In Proceedings of the IEEE Ultrasonics Symposium, Montreal, Canada. Plissiti, M. E., Fotiadis, D. I., Michalis, L. K., & Bozios, G. E. (2004). An automated method for lumen and media–adventitia border detection in a sequence of IVUS frames. IEEE Transactions on Information Technology in Biomedicine, 8(2), 131–141. doi:10.1109/TITB.2004.828889 Shen, S., Sandham, W., Granat, M., & Sterr, A. (2005). MRI fuzzy segmentation of brain tissue using neighborhood attraction with neural network optimization. IEEE Transactions on Information Technology in Biomedicine, 9(3), 459–467. doi:10.1109/TITB.2005.847500 Sherknies, D., Meunier, J., Mongrain, R., & Tardif, J.-C. (2005). Three-dimensional trajectory assessment of an IVUS transducer from singleplane cineangiograms: A phantom study. IEEE Transactions on Bio-Medical Engineering, 52(3), 543–549. doi:10.1109/TBME.2004.843295

Visual Medical Information Analysis

Sonka, M., Zhang, X., Siebes, M., Bissing, M. S., DeJong, S. C., & Collins, S. M. (1995). Segmentation of intravascular ultrasound images: A knowledge-based approach. IEEE Transactions on Medical Imaging, 14(4), 719–732. doi:10.1109/42.476113

Zhu, H., Liang, Y., & Friedman, M. H. (2002). IVUS image segmentation based on contrast. In Proceedings of SPIE, Durham, NC, USA, (Vol. 4684, pp. 1727-1733).

Székely, N., Toth, N., & Pataki, B. (2006). A hybrid system for detecting masses in mammographic images. IEEE Transactions on Instrumentation and Measurement, 55(3), 944–952. doi:10.1109/ TIM.2006.870104

KEY TERMS AND DEFINITIONS

Valdes-Cristerna, R., Medina-Banuelos, V., & Yanez-Suarez, O. (2004). Coupling of radial-basis network and active contour model for multispectral brain MRI segmentation. IEEE Transactions on Bio-Medical Engineering, 51(3), 459–470. doi:10.1109/TBME.2003.820377 Wahle, A., Prause, G. P. M., DeJong, S. C., & Sonka, M. (1999). Geometrically correct 3-D reconstruction of intravascular ultrasound images by fusion with biplane angiography—methods and validation. IEEE Transactions on Medical Imaging, 18(8), 686–699. doi:10.1109/42.796282 Woolrich, M. W., Behrens, T. E. J., Beckmann, C. F., & Smith, S. M. (2005). Mixture models with adaptive spatial regularization for segmentation with an application to FMRI data. IEEE Transactions on Medical Imaging, 24(1), 1–11. doi:10.1109/TMI.2004.836545

Active Contour Model: Energy-minimizing parametric curve that is the basis of several medical image analysis techniques. Computer-Aided Diagnosis: The process of using computer-generated analysis results for assisting doctors in evaluating medical data. Coronary Angiography: X-ray diagnostic process for obtaining an image of the coronary arteries. Intravascular Ultrasound (IVUS): Diagnostic catheter-based technique that renders twodimensional images of coronary arteries. Magnetic Resonance Imaging (MRI): Imaging technique that uses a magnetic field to provide two-dimensional images of internal body structures. Mammography: Diagnostic X-ray technique which produces breast images and is used to detect breast tissue abnormalities. Medical Image Segmentation: The localization of known anatomic structures in medical images.

This work was previously published in Encyclopedia of Information Science and Technology, Second Edition, edited by Mehdi Khosrow-Pour, pp. 4034-4040, copyright 2009 by Information Science Reference (an imprint of IGI Global).

367

368

Chapter 2.9

Angiographic Images Segmentation Techniques Francisco J. Nóvoa University of A Coruña, Spain Alberto Curra University of A Coruña, Spain M. Gloria López University of A Coruña, Spain Virginia Mato University of A Coruña, Spain

INTRODUCTION Heart-related pathologies are among the most frequent health problems in western society. Symptoms that point towards cardiovascular diseases are usually diagnosed with angiographies, which allow the medical expert to observe the bloodflow in the coronary arteries and detect severe narrowing (stenosis). According to the severity, extension, and location of these narrowings, the expert pronounces a diagnosis, defines a treatment, and establishes a prognosis. DOI: 10.4018/978-1-60960-561-2.ch209

The current modus operandi is for clinical experts to observe the image sequences and take decisions on the basis of their empirical knowledge. Various techniques and segmentation strategies now aim at objectivizing this process by extracting quantitative and qualitative information from the angiographies.

BACKGROUND Segmentation is the process that divides an image in its constituting parts or objects. In the present context, it consists in separating the pixels that

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Angiographic Images Segmentation Techniques

compose the coronary tree from the remaining “background” pixels. None of the currently applied segmentation methods is able to completely and perfectly extract the vasculature of the heart, because the images present complex morphologies and their background is inhomogeneous due to the presence of other anatomic elements and artifacts such as catheters. The literature presents a wide array of coronary tree extraction methods: some apply pattern recognition techniques based on pure intensity, such as thresholding followed by an analysis of connected components, whereas others apply explicit vessel models to extract the vessel contours. Depending on the quality and noise of the image, some segmentation methods may require image preprocessing prior to the segmentation algorithm; others may need postprocessing operations to eliminate the effects of a possible oversegmentation. The techniques and algorithms for vascular segmentation could be categorized as follows (Kirbas, Quek, 2004): 1. Techniques for “pattern-matching” or pattern recognition 2. Techniques based on models 3. Techniques based on tracking 4. Techniques based on artificial intelligence 5. Main Focus This section describes the main features of the most commonly accepted coronary tree segmentation techniques. These techniques automatically detect objects and their characteristics, which is an easy and immediate task for humans, but an extremely complex process for artificial computational systems.

Techniques Based on Pattern Recognition The pattern recognition approaches can be classified into four major categories:

Multiscale Methods The multiscale method extracts the vessel method by means of images of varying resolutions. The main advantage of this technique resides in its high speed. Larger structures such as main arteries are extracted by segmenting low resolution images, whereas smaller structures are obtained through high resolution images.

Methods Based on Skeletons The purpose of these methods is to obtain a skeleton of the coronary tree: a structure of smaller dimensions than the original that preserves the topological properties and the general shape of the detected object. Skeletons based on curves are generally used to reconstruct vascular structures (Nyström, Sanniti di Baja & Svensson, 2001). Skeletonizing algorithms are also called “thinning algorithms”. The first step of the process is to detect the central axis of the vessels or “centerline”. This axis is an imaginary line that follows each vessel in its central axis, i.e. two normal segments that cross the axis in opposite sense should present the same distance from the vessel’s edges. The total of these lines constitutes the skeleton of the coronary tree. The methods that are used to detect the central axes can be classified into three categories:

Methods Based on Crests One of the first methods to segment angiographic images on the basis of crests was proposed by Guo and Richardson (Guo & Ritchardson, 1998). This method treats angiographies as topographic maps in which the detected crests constitute the central axes of the vessels. The image is preprocessed by means of a median filter and smoothened with non-linear diffusion. The region of interest is then selected through thresholding, a process that eliminates

369

Angiographic Images Segmentation Techniques

the crests that do not correspond with the central axes. Finally, the candidate central axes are joined with curve relaxation techniques.

Methods Based on Regions Growth Taking a known point as seed point, these techniques segment images through the incremental inclusion of pixels in a region on the basis of an a priori established criterion. There are two especially important criteria: similitude in the value, and spatial proximity (Jain, Kasturi & Schunck, 1995). It is established that pixels that are sufficiently near others with similar grey levels belong to the same object. The main disadvantage of this method is that it requires the intervention of the user to determine the seed points. (Figure 1) O’Brien and Ezquerra (O’Brien & Ezquerra, 1994) propose the automatic extraction of the coronary vessels in angiograms on the basis of temporary, spatial, and structural restrictions. The algorithm starts with a low-pass filter and the user’s definition of a seed point. The system then starts to extract the central axes by means of the “globe test” mechanism, after which the detected regions are entangled through the graph theory. The applied test also allows us to discard the regions that are detected incorrectly and do not belong to the vascular tree.

Figure 1. Regions growth applied to an angiography

370

Methods Based on Differential Geometry The methods that are based on differential geometry treat images as hypersurfaces and extract their features using curvature and surface crests. The points of hypersurface’s crest correspond to the central axis of the structure of a vessel. This method can be applied to bidimensional as well as tridimensional images; angiograms are bidimensional images and are therefore modelled as tridimensional hypersurfaces. Examples of reconstructions can be found in Prinet et al (Prinet, Mona & Rocchisani, 1995), who treat the images as parametric surfaces and extract their features by means of surfaces and crests.

Correspondence Filters Methods The correspondence filter approach convolutes the image with multiple correspondence filters so as to extract the regions of interest. The filters are designed to detect different sizes and orientations. Poli and Valli (Poli, R & Valli, 1997) apply this technique with an algorithm that details a series of multiorientation linear filters that are obtained as linear combinations of Gaussian “kernels”. These filters are sensitive to different vessel widths and orientations.

Angiographic Images Segmentation Techniques

Mao et al (Mao, Ruan, Bruno, Toumoulin, Collorec & Haigron, 1992) also use this type of filters in an algorithm based on visual perception models that affirm that the relevant parts of the objects in images with noise appear normally grouped.

Morphological Mathematical Methods Mathematical morphology defines a series of operators that apply structural elements to the images so that their morphological features can be preserved and irrelevant elements eliminated (Figure 2). The main morphological operations are the following: • • • • • •

Dilatation: Expands objects, fills up empty spaces, and connects disjunct regions. Erosion: Contracts objects, separates regions. Closure: Dilatation + Erosion. Opening: Erosion + Dilatation. “Top hat” transformation: Extracts the structures with a linear shape “Watershed” transformation: “Inundates” the image that is taken as a topographic map, and extracts the parts that are not “flooded”.

Eiho and Qian (Eiho & Qian, 1997) use a purely morphological approach to define an algorithm that consists of the following steps:

1. Application of the “top hat” operator to emphasize the vessels 2. Erosion to eliminate the areas that do not correspond to vessels 3. Extraction of the tree from a point provided by the user and on the basis of grey levels. 4. Slimming down of the tree 5. Extraction of edges through “watershed” transformation

MODEL-BASED TECHNIQUES These approaches use explicit vessel models to extract the vascular tree. They can be divided into four categories: deformable models, parametric models, template correspondence models, and generalized cylinders.

Deformable Models Strategies based on deformable models can be classified in terms of the work by McInerney and Terzopoulos (McInerney & Terzopoulos, 1997). Algorithms that use deformable models (Merle, Finet, Lienard, & Magnin, 1997) are based on the progressive refining of an initial skeleton built with curves from a series of reference points: • •

Root points: Starting points for the coronary tree. Bifurcation points: Points where a main branch divides into a secundary branch.

Figure 2. Morphological operators applied to an angiography

371

Angiographic Images Segmentation Techniques

End points: Points where a tree branch ends.

•

These points have to be marked manually.

Deformable Parametric Models: Active Contours These models use a set of parametric curves that adjust to the object’s edges and are modified by both external forces, that foment deformation, and internal forces that resist change. The active contour models or “snakes” in particular are a special case of a more general technique that pretends to adjust deformable models by minimizing energy. (Figure 3) Klein et al. (Klein, Lee & Amini, 1997) propose an algorithm that uses “snakes” for 4D reconstruction: they trace the position of each point of the central axis of a skeleton in a sequence of angiograms.

Figure 3. “Snakes” applied to a blood vessel. http://vislab.cs.vt.edu/review/extraction.html

372

Deformable Geometric Models These models are based on topographic models that are adapted for shape recognition. Malladi et al. (Malladi, Sethian & Vemuri, 1995) for instance adapt the “Level Set Method” (LSM) by representing an edge as a level zero set of a hypersurface of a superior order; the model evolves to reduce a metric defined by the restrictions of edges and curvature, but less rigidly than in the case of the “snakes”. This edge, which constitutes the zero level of the hypersurface, evolves by adjusting to the edges of the vessels, which is what we want to detect.

Propagation Methods Quek and Kirbas (Quek & Kirbas, 2001) developed a system of wave propagation combined with a backtracking mechanism to extract the vessels from angiographic images. This method basically labels each pixel according to its likeliness to belong to a vessel and then propagates a wave through the pixels that are labeled as belonging to the vessel; it is this wave that definitively extracts the vessels according to the local features it encounters. Approaches based on the correspondence of deformable templates: This approach tries to recognize structural models (templates) in an image by using a template as context, i.e. as a priori model. This template is generally represented as a set of nodes connected by a segment. The initial structure is deformed until it adjusts optimally to the structures that were observed in the image. Petrocelli et al. (Petrocelli, Manbeck, & Elion, 1993) describe a method based on deformable templates that also incorporates additional previous knowledge into the deformation process.

Angiographic Images Segmentation Techniques

Parametric Models These models are based on the a priori knowledge of the artery’s shape and are used to build models whose parameters depend on the profiles of the entire vessel; as such, they consider the global information of the artery instead of merely the local information. The value of these parameters is established after a learning process. The literature shows the use of models with circular sections (Shmueli, Brody, & Macovski, 1983) and spiral sections (Pappas, & Lim, 1984), because various studies by Brown, B. G., (Bolson, Frimer, & Dodge, 1977) (Brown, Bolson, Frimer & Dodge, 1982) show that sections of healthy arteries tend to be circular and sections with stenosis are usually elliptical. However, both circular and elliptical shapes fail to approach irregular shapes caused by pathologies or bifurcations. This model has been applied to the reconstruction of vascular structures with two angiograms (Pellot, Herment, Sigelle, Horain, Maitre & Peronneau, 1994), which is why both healthy and stenotic sections are modeled by means of ellipses. This model is subsequently deformed until it corresponds to the shape associated to the birth of a new branch or pathology.

Generalized Cylinder Models A generalized cylinder (GC) is a solid whose central axis is a 3D curve. Each point of that axis has a limited and closed section that is perpendicular to it. A CG is therefore defined in space by a spatial curve or axis and a function that defines the section in that axis. The section is usually an ellipse. Tecnically, GCs should be included in the parametric methods section, but the work that has been done in this field is so extense that it deserves its own category. The construction of the coronary tree model requires one single view to build the 2D tree and estimate the sections. However, there is no infor-

mation on the depth or the area of the sections, so a second projection will be required.

ARTERIAL TRACKING Contrary to the approaches based on pattern recognition, where local operators are applied to the entire image, techniques based on arterial follow-up are based on the application of local operators in an area that presumibly belongs to a vessel and that cover its length. From a given point of departure the operators detect the central axis and, by analyzing the pixels that are orthogonal to the tracking direction, the vessel’s edges. There are various methods to determine the central axis and the edges: some methods carry out a sequential tracking and incorporate connectivity information after a simple edge detection operation, other methods use this information to sequentially track the contours. There are also approaches based on the intensity of the crests, on fuzzy sets, or on the representation of graphs, where the purpose lies in finding the optimal road in the graph that represents the image. (Figure 4) Lu and Eiho (Lu, Eiho, 1993) have described a follow-up algorithm for the vascular edges in angiographies that considers the inclusion of branches and consists of three steps: 1. Edge detection 2. Branch search 3. Tracking of sequential contours The user must provide the point of departure, the direction, and the search range. The edge points are evaluated with a differential smoothening operator in a line that is perpendicular to the direction of the vessel. This operator also serves to detect the branches.

373

Angiographic Images Segmentation Techniques

Figure 4. Tracking applied to an angiography

TECHNIQUES BASED ON ARTIFICIAL INTELLIGENCE Approaches based on Artificial Intelligence use high-level knowledge to guide the segmentation and delineation of vascular structures and sometimes use different types of knowledge from various sources. One possibility (Smets, Verbeeck, Suetens, & Oosterlinck, 1988) is to use rules that codify knowledge on the morphology of blood vessels; these rules are then used to formulate a hierarchy with which to create the model. This type of system does not offer any good results in arterial bifurcations or in arteries with occlusions. Another approach (Stansfield, 1986) consists in formulating a rules-based Expert System to identify the arteries. During the first phase, the image is processed without making use of domain knowledge to extract segments of the vessels. It is only in the second phase that domain knowledge on cardiac anatomy and physiology is applied. The latter approach is more robust than the former; but it presents the inconvencience of not combining all the segments into one vascular structure.

FUTURE TRENDS It cannot be said that one technique has a more promising future than another, but the current tendency is to move away from the abovementioned

374

classical segmentation algorithms towards 3D and even 4D reconstructions of the coronary tree. Other lines of research focus on obtaining angiograph images by means of new acquisition technologies such as Magnetic Resonance, Computarized High Speed Tomography, or two-armed angiograph devices that achieve two simultaneous projections in combination with the use of ultrasound intravascular devices. This type of acquisition simplifies the creation of tridimensional structures, either directly from the acquisition or after a simple processing of the bidimensional images.

REFERENCES Brown, B. G., Bolson, E., Frimer, M., & Dodge, H. T. (1977). Quantitative coronary arteriography. Circulation, 55, 329–337. Brown, B. G., Bolson, E., Frimer, M., & Dodge, H. T. (1982). Arteriographic assessment of coronary atherosclerosis. Arteriosclerosis (Dallas, Tex.), 2, 2–15. Eiho, S., & Qian, Y. (1997). Detection of coronary artery tree using morphological operator. Computers in Cardiology, 1997, 525–528. Gonzalez, R. C., & Woods, R. E. (1996). Digital Image Proccessing. Addison-Wesley Publishing Company, Inc. Reading, Massachusets, USA.

Angiographic Images Segmentation Techniques

Greenes, R. A., & Brinkley, K. F. (2001). Imaging Systems. De Medical informatics: computer applications in health care and biomedicine. Pp. 485 – 538. Second Edition. 2001. Ed. SpringerVerlag. New York. USA. Guo, D., & Richardson, P. (1998). Automatic vessel extraction from angiogram images. Computers in Cardiology, 1998, 441–444. Jain, R. C., Kasturi, R., & Schunck, B. G. (1995). Machine Vision.McGraw-Hill. Kirbas, C., & Quek, F. (2004). A review of vessel extraction techniques and algorithms. ACM Computing Surveys, 36(2), 81–121. doi:10.1145/1031120.1031121 Klein, A. K., Lee, F., & Amini, A. A. (1997). Quantitative coronary angiography with deformable spline models. IEEE Transactions on Medical Imaging, 16(5), 468–482. doi:10.1109/42.640737 Lu, S., & Eiho, S. (1993). Automatic detection of the coronary arterial contours with sub-branches from an x-ray angiogram.In Computers in Cardiology 1993. Proceedings., 575-578. Malladi, R., Sethian, J. A., & Vemuri, B. C. (1995). Shape modeling with front propagation: a level set approach. Pattern Analysis and Machine Intelligence. IEEE Transactions on, 17, 158–175. Mao, F., Ruan, S., Bruno, A., Toumoulin, C., Collorec, R., & Haigron, P. (1992). Extraction of structural features in digital subtraction angiography. Biomedical Engineering Days, 1992.,Proceedings of the 1992 International, 166-169. McInerney, T., & Terzopoulos, D. (1997). Medical image segmentation using topologically adaptable surfaces. In CVRMedMRCAS ‘97: Proceedings of the First Joint Conference on Computer Vision, Virtual Reality and Robotics in Medicine and Medial Robotics and Computer-Assisted Surgery, 23-32, London, UK, Springer-Verlag.

Nyström, I., Sanniti di Baja, G., & Svensson, S. (2001). Representing volumetric vascular structures using curve skeletons. In Edoardo Ardizzone and Vito Di Gesµu, editors, Proceedings of 11th International Conference on Image Analysis and Processing (ICIAP 2001), 495-500, Palermo, Italy, IEEE Computer Society. O’Brien, J. F., & Ezquerra, N. F. (1994). Automated segmentation of coronary vessels in angiographic image sequences utilizing temporal, spatial and structural constraints. (Technical report), Georgia Institute of Technology. Pappas, T. N., & Lim, J. S. (1984). Estimation of coronary artery boundaries in angiograms. Appl. Digital Image Processing VII, 504, 312–321. Pellot, C., Herment, A., Sigelle, M., Horain, P., Maitre, H., & Peronneau, P. (1994). A 3d reconstruction of vascular structures from two x-ray angiograms using an adapted simulated annealing algorithm. Medical Imaging. IEEE Transactions on, 13, 48–60. Petrocelli, R. R., Manbeck, K. M., & Elion, J. L. (1993). Three dimensional structure recognition in digital angiograms using gauss-markov methods. In Computers in Cardiology 1993. Proceedings., 101-104. Poli, R., & Valli, G. (1997). An algorithm for realtime vessel enhancement and detection. Computer Methods and Programs in Biomedicine, 52, 1–22. doi:10.1016/S0169-2607(96)01773-7 Prinet, V., Mona, O., & Rocchisani, J. M. (1995). Multi-dimensional vessels extraction using crest lines. In Engineering in Medicine and Biology Society, 1995. IEEE 17th Annual Conference, 1:393-394. Quek, F. H. K., & Kirbas, C. (2001). Simulated wave propagation and traceback in vascular extraction. In Medical Imaging and Augmented Reality, 2001. Proceedings. International Worksho, 229-234.

375

Angiographic Images Segmentation Techniques

Shmueli, K., Brody, W. R., & Macovski, A. (1983). Estimation of blood vessel boundaries in x-ray images. Optical Engineering (Redondo Beach, Calif.), 22, 110–116. Smets, C., Verbeeck, G., Suetens, P., & Oosterlinck, A. (1988). A knowledge-based system for the delineation of blood vessels on subtraction angiograms. Pattern Recognition Letters, 8(2), 113–121. doi:10.1016/0167-8655(88)90052-9 Stansfield, S. A. (1986). Angy: A rule-based expert system for automatic segmentation of coronary vessels from digital subtracted angiograms. PAMI, 8(3), 188–199.

KEY TERMS AND DEFINITIONS Angiography: Image of blood vessels obtained by any possible procedure. Artery: Each of the vessels that take the blood from the heart to the other bodyparts. Computerized Tomography: Exploration of X-rays that produces detailed images of axial cuts of the body. A CT obtains many images by rotating around the body. A computer combines all these images into a final image that represents the bodycut like a slice.

Expert System: Computer or computer program that can give responses that are similar to those of an expert. Segmentation: In computer vision, segmentation refers to the process of partitioning a digital image into multiple regions. The goal of segmentation is to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze. Image segmentation is typically used to locate objects and boundaries (structures) in images, in this case, the coronary tree in digital angiography frames. Stenosis: A stenosis is an abnormal narrowing in a blood vessel or other tubular organ or structure. A coronary artery that’s constricted or narrowed is called stenosed. Buildup of fat, cholesterol and other substances over time may clog the artery. Many heart attacks are caused by a complete blockage of a vessel in the heart, called a coronary artery. Thresholding: A technique for the processing of digital images that consists in applying a certain property or operation to those pixels whose intensity value exceeds a defined threshold.

This work was previously published in Encyclopedia of Artificial Intelligence, edited by Juan Ramón Rabuñal Dopico, Julian Dorado and Alejandro Pazos, pp. 110-117, copyright 2009 by Information Science Reference (an imprint of IGI Global).

376

377

Chapter 2.10

Segmentation Methods in Ultrasound Images Farhang Sahba Medical Imaging Analyst, Canada

ABSTRACT Ultrasound imaging now has widespread clinical use. It involves exposing a part of the body to high-frequency sound waves in order to generate images of the inside of the body. Because it is a real-time procedure, the ultrasound images show the movement of the body’s internal structure as well. It is usually a painless medical test and its procedures seem to be safe. Despite recent improvement in the quality of information from an ultrasound device, these images are still a challenging case for segmentation. Thus, there is much interest in understanding how to apply an image segmentation task to ultrasound data and any improvements in this regard are desirable. DOI: 10.4018/978-1-60960-561-2.ch210

Many methods have been introduced in existing literature to facilitate more accurate automatic or semi-automatic segmentation of ultrasound images. This chapter is a basic review of the works on ultrasound image segmentation classified by application areas, including segmentation of prostate transrectal ultrasound (TRUS), breast ultrasound, and intravascular ultrasound (IVUS) images.

INTRODUCTION Among different image modalities, ultrasound imaging is one of the most widely used technologies for the diagnosis and treatment of diseases such as breast and prostate cancer. Ultrasound

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Segmentation Methods in Ultrasound Images

equipment is less expensive to purchase and maintain than many other imaging systems such as X-ray, computed tomography (CT), or magnetic resonance imaging (MRI). These images are the result of reflection, refraction, and deflection of ultrasound beams from different types of tissue with different acoustic impedances. The detection of the object boundaries in such images is crucial for diagnostic and classification purposes. However, attenuation, speckle, shadows, and signal dropout can result in missing or diffused boundaries. Also the contrast between areas of interest is often low. These obstacles make segmentation of these images a challenge. Further complications arise when the quality of the image is influenced by the type and particular settings of the machine. Despite these factors, ultrasound imaging still remains an important tool for clinical applications and any effort to improve segmentation of these images is highly desirable. Thus, there is currently an interest in understanding how to apply image segmentation to ultrasound data. Figure 1 demonstrates the basic principle of an ultrasound imaging transducer. Using an ultrasound transducer, a pulse of energy is transmitted into the body along the path shown by line 1. After this beam encounters any surface, including tissue or structures within an organ, a part of the transmitted energy is backscattered along the original trajectory and received by the transducer which now acts as a receiver. These returning waves are converted to electrical signals, amplified, and finally shown. After that, the direction of the transmitted beam changes to attain the data from the next line close to the first one. The ultrasound transducer repeats the same procedure to cover 64-256 lines and makes the entire image (Webb, 2003). This chapter contains an overview of the ideas representing the ultrasound segmentation problem in particular clinical applications.

378

Figure 1. The basic principle of an ultrasound imaging transducer (© 2003 IEEE, Reprinted, with permission from IEEE Press Series in Biomedical Engineering 2003. “Introduction to Biomedical Imaging”, by A. Webb).

BACKGROUND Many methods have been introduced to facilitate more accurate segmentation of ultrasound images. The performance of these methods is generally improved through the use of expertise or prior knowledge. All segmentation methods usually require at least some user interaction to adjust critical parameters. The type of user interaction varies, depending on the amount of time and effort required from the user. This chapter is a review of ultrasound image segmentation methods and focuses on clinical applications that have been investigated in different clinical domains. It centers on reviewing the ideas behind the incorporated knowledge of ultrasound physics such as speckle structure, as well as prior information about the intensity or shape model. We review some principal works in this area according to their applications where the majority of efforts have been focused. These include segmentation of prostate transrectal ultrasound (TRUS), breast ultrasound, and intravascular ultrasound (IVUS) images.

Segmentation Methods in Ultrasound Images

ULTRASOUND IMAGE SEGMENTATION ACCORDING TO CLINICAL APPLICATIONS Based on clinical application, ultrasound image segmentation can be categorized in various groups. In this section, we mention some important methods in each group.

Prostate Segmentation Prostate cancer is one of the most frequently diagnosed malignancies in the adult and aging male population (Mettlin,1995). Ultrasound imaging is a widely used technology for the detection and intervention of this cancer and may help to reduce death rate if used in early stages. As the prostate boundaries play an important role in the diagnosis and treatment of prostate cancer, it is crucial for many clinical applications to accurately detect them. These applications include the accurate placement of needles during the biopsy, accurate prostate volume measurement from multiple frames, constructing anatomical models used in treatment planning, and estimation of tumor border. These images are the result of reflection, refraction, and deflection of ultrasound beams from different types of tissues with different acoustic impedances (Insana et al., 1993). Some factors, such as poor contrast, speckle, and weak edges, however, make the ultrasound images inherently difficult to segment. Furthermore, the quality of the image may be influenced by the type and particular settings of the machine. Currently, the prostate boundaries are generally extracted from TRUS images (Insana et al., 1993). This kind of imaging has been a fundamental tool for prostate cancer diagnosis and treatment. Prostate boundaries must generally be outlined in 2D TRUS image slices along the length of the prostate. But as previously mentioned, the signal-to-noise ratio in these images is very low. Therefore, traditional edge detectors fail to extract the correct boundaries.

Consequently, many methods have been developed to facilitate automatic or semi-automatic segmentation of the prostate boundaries from the ultrasound images. Figure 2 shows segmentation of some sample ultrasound images of the prostate using the method presented by Nanayakkara et al. (2006); it contains the ground truth boundaries drawn by an expert as well as contours generated by the computerized method. Knoll et al. (1999) proposed a technique for elastic deformation of closed planar curves restricted to particular object shapes. Their method is based on a one-dimensional dyadic wavelet transform as a multi-scale contour parameterization technique to constrain the shape of the prostate model. Important edges at multiple resolutions are extracted as the first step. Then a template matching procedure is used to obtain an initial shape of the contour. The shape of the contour is constrained to predefined models during deformation. While they reported that the method provides an accurate and fully automatic segmentation of 2D objects, the dependence of the statistically derived prior model has limited its capability for segmentation of aberrant shapes. Richard et al. (1996) presented a texturebased algorithm for prostate segmentation. This method segments a set of parallel 2D images of the prostate into prostate and non-prostate regions to form a 3D image. This algorithm is a pixel classifier which classifies each pixel of an ultrasound image using four associated texture energy measures. One of the drawbacks of this approach is that the number of clusters cannot be predicted for an image; therefore, the resulting image may be represented by a set of disconnected regions and no post-processing technique is proposed for preventing or overcoming the problem of such discontinuities. Ladak et al. (2000) proposed a cubic spline interpolation technique for semi-automatic segmentation of the prostate. The algorithm uses an initial contour based on four points given by the user. The user selects four points around the pros-

379

Segmentation Methods in Ultrasound Images

Figure 2. Four samples for prostate boundary detection. The solid line shows the contour generated by the computerized method and the dotted line shows the corresponding manual outline (Reprinted (partially), from Physics in Medicine and Biology, 51, N.D Nanayakkara, J. Samarabandu, and A. Fenster, “Prostate segmentation by feature enhancement using domain knowledge and adaptive region based operations”, pp. 1831–1848, 2006, with permission from IOP publishing Ltd).

tate and then uses the discrete dynamic contour. A model is used to refine the boundary. Although this semi-automatic algorithm can segment a wide range of prostate images, at least four initial points must be defined accurately by the user (radiologist). In addition, it is less satisfactory when the prostate has an irregular shape and cannot be perfectly approximated by the initial points. For such cases, further human intervention is required to achieve satisfactory results. Wang et al. (2003) presented two methods for semi-automatic three-dimensional (3D) prostate boundary segmentation using 2D ultrasound images. The segmentation process is initiated by manually placing four points on the boundary of a selected slice. Then an initial prostate boundary is determined. It is refined using the discrete dynamic contour until it fits the actual prostate boundary. The remaining slices are then segmented by iteratively propagating the results to other slices and implementing the refinement.

380

Hu et al. (2003) proposed an algorithm for semi-automatic segmentation of the prostate from 3D ultrasound images. In this method, the authors use model-based initialization and mesh refinement using deformable models. Six points are required to initialize the outline of the prostate using shape information. The initial outline is then automatically deformed to better fit the prostate boundary. Chiu et al. (2004) introduced a semi-automatic segmentation algorithm based on the dyadic wavelet transform and the discrete dynamic contours. In this method, a spline interpolation is first used to determine the initial contour based on four userdefined initial points. Then the discrete dynamic contour refines the initial contour based on the approximate coefficients as well as the wavelet coefficients generated using the dyadic wavelet transform. A selection rule is also used to choose the best contour.

Segmentation Methods in Ultrasound Images

Abolmaesumi et al. (2004) used an interactive multi-model probabilistic data association filter to extract prostate contours from transrectal ultrasound images. As the first step, a Sticks filter is used to enhance the image. The problem is then addressed by considering several equally spaced radii from a seed point towards the boundary of the prostate. In this method, the border of the prostate is represented as the trajectory of a moving object. This motion is modeled using a set of dynamical models where their measurement points are presented as candidate edge points along each radius. As the method does not use any numerical technique, their results show that the convergence is also fast. Pathak et al. (2000) proposed an edge-guided boundary delineation algorithm for prostate segmentation provided as a visual guide to the observer. This step is followed by manual editing. For automatic edge detection, the algorithm first uses a Sticks filter to enhance contrast and reduce speckle. Then, an anisotropic diffusion filter is applied to smooth the result of the previous stage. Finally, some basic prior knowledge such as shape and echo pattern is used to extract the most probable edges. After these stages, by using a manual linking procedure on the detected edges, the final boundary is indicated. Shen et al. (2003) presented a statistical shape model for segmentation of the prostate in ultrasound images. A Gabor filter bank is employed in both multiple scales and multiple orientations to represent the image characteristics around the prostate boundaries. As these features give both edge directions and edge strengths, they can provide the information for deformation of the prostate model. This strategy can generate both coarse and fine image features that further allow the model to focus on particular features at different deformation steps. To make the proposed method more robust against local minima, several hierarchical deformation strategies are proposed as well. In another work, the authors have also introduced an adaptive focus deformable model,

which uses the concept of an attribute vector (Shen et al., 2001). Figure 3 shows segmentation of an ultrasound image of the prostate using the method presented in (Shen et al., 2003). This figure demonstrates the results of the algorithm in the different steps (iterations). As can be seen, the method is robust with respect to a bad initialization. Betrounia et al. (2005) proposed a method for the automatic segmentation of trans-abdominal ultrasound images. Adaptive morphological and median filtering are employed together to detect the noisy areas and smooth them. Using this method, the contours of the prostate can be enhanced without changing the critical information in the image. An optimization algorithm is then employed to search for the contour initialized from a prostate model. The algorithm has been shown to have accurate results in terms of average distance and average coverage index in comparison to those obtained by manual segmentation.

Breast Cancer Breast cancer is one of the leading causes of death in women. Along with digital mammography, ultrasound has been one of the most commonly used methods for early detection and diagnosis of breast cancer in the last decade. Ultrasound can help to distinguish whether a mass is benign or malignant (cancerous). Automatic segmentation of breast tumors using ultrasound imaging can assist physicians in making faster and more accurate diagnoses (Noble et al., 2006). The problem with segmenting ultrasonic breast images is mostly the variance of the lesion’s shape and the fact that often the borders of the lesion are not well distinguished. Figure 4 demonstrates three different manually segmented ultrasonic breast lesions with great differences in their shapes and sizes as represented in (Madabhushi et al., 2003). Huang et al. (2004) proposed an approach that combines a neural network classifier with a morphological watershed segmentation method to

381

Segmentation Methods in Ultrasound Images

Figure 3. Demonstration of the algorithm represented in (Shen et al., 2003). The dashed contour indicates the manually segmented prostate. The solid contours show the resulting the automatic algorithm in the different steps (iterations). The final segmentation results are shown in the right-bottom corner; a case of the bad initialization. (© 2003 IEEE, Reprinted with permission from IEEE Transaction on Medical Imaging, 2003, 22(4), pp. 539-551, “Segmentation of Prostate Boundaries from Ultrasound Images Using Statistical Shape Model”, by D. Shen, Y. Zhan and C. Davatzikos).

Figure 4. Three manually segmented tumors with different shapes and sizes (© 2003 IEEE, Reprinted with permission from ╯IEEE Transaction on Medical Imaging, 2003, 22(2), pp. 155–169, “Combining low-, high-level and empirical domain knowledge for automated segmentation of ultrasonic breast lesions”, by A. Madabhushi and D. N. Metaxas).

extract precise contours of a breast tumor from ultrasound images. Textural analysis is employed to generate the inputs of the neural network to classify US images. The features of the texture contain auto covariance coefficients to classify US images using a self-organizing map. After

382

these features have been classified, an adaptive preprocessing procedure is selected by neural network output and then watershed transformation automatically determines the contours of the tumor. This method contains a preprocessing step which helps the watershed algorithm through a

Segmentation Methods in Ultrasound Images

good selection of markers. The self-organizing map is used in order to select the appropriate preprocessing filter locally from a set of nine predefined filters. Figure 5 demonstrates two malignant and benign cases of breast ultrasound images. In this figure, the first, second, and third rows show the original magnified monochrome breast image, the contours manually sketched, and the contours determined by the proposed system in (Huang et al., 2004), respectively. Chen et al. (2002) presented a method based on a neural network. The aim of this method is to make the classification based on a set of input features. These features are variance contrast, autocorrelation contrast, and the distribution

distortion in the wavelet coefficients. These are inputs of a multilayer perceptron neural network with one hidden layer which is trained by error backpropagation. Image texture is an important component in their method. Xiao et al. (2002) discussed a method for simultaneous estimation of video-intensity inhomogeities and segmentation of US image tissue regions. The number of regions (classes) needs to be specified. It employs a combination of the maximum a posteriori (MAP) and Markov random field (MRF) methods to estimate the US image distortion field. The approach shows consistent segmentations under different time gain compensation (TGC) settings on the tested data.

Figure 5. Contour segmentation. The first row: Original magnified monochrome breast ultrasound images (left malignant and right benign case); the second row, manual sketch contour; and the third row, automatic sketch contour. (Reprinted from Ultrasound in Medicine & Biology, 30(5), Y. L Huang and D. R. Chen, “Watershed segmentation for breast tumor in 2-D sonography”, pp. 625–632, 2004, with permission from Elsevier).

383

Segmentation Methods in Ultrasound Images

Horsch et al. (2001) introduced a segmentation technique that is based on maximizing a utility function over partition margins which are defined through gray-value thresholding of a preprocessed image that has enhanced mass structures. It requires the manually defined lesion center. The problem of shadowing is not discussed. Shape, echogeneity, margin, and posterior acoustic behavior are computed as four features to test the effectiveness when directed at distinguished malignant and benign masses. The authors have further evaluated their method in (Horsch et al., 2004) to assess the advantages of the different mentioned features using linear discriminant analysis. It is shown that the two best features are depth-to-width ratio for shape and normalized radial gradient for margin. Sahiner et al. (2004) developed 2D and 3D intensity-gradient active contour models for automated segmentation of the mass volumes. For initialization of the active contour, texture and morphological features were automatically extracted from the segmented masses and their margins. Algorithm parameters were determined empirically. To find the segmentation result, depth to width ratio, a posterior shadowing feature measure, and texture features were computed around the boundary of each 2D slice and then linear discriminant analysis was used to classify volumes. Madabhushi et al. (2003) proposed a method that uses the combination of intensity and texture with empirical domain-specific knowledge, along with directional gradient and a deformable shape-based model. A second-order Butterworth filter is used to remove speckle noise, and then the contrast of the tumor regions is enhanced. The image pixels are probabilistically classified based on intensity and texture information and then a region growing technique is used to obtain an initial segmentation of the lesion. The boundary points are found to supply an initial estimate to a deformable model. It attempts to limit the effects of shadowing and false positives by incorporat-

384

ing empirical domain-specific knowledge. The method requires a small database for training. There are also some other methods that take into account the domain knowledge and consider the specification of ultrasound images such as speckle or artifacts.

Intravascular Ultrasound Intravascular Ultrasound (IVUS) is a non-invasive technique which provides real-time high-resolution images with valuable anatomical information about the coronary arterial wall and plaque. IVUS consequently provides new insights into the diagnosis and therapy of coronary disease. IVUS imaging can be used as a complementary imaging tool to contrast X-ray angiography (Radeva et al., 2003), (Noble et al., 2006). Due to the ultrasound speckle, catheter artifacts, or calcification shadows, processing of IVUS image sets is a challenge. To perform an accurate quantitative analysis, a good segmentation of the lumen, the plaque, and the wall borders is required. Methods used for IVUS segmentation usually apply contour modeling (Kovalski et al., 2000; Sonka et al., 1995; Cardinal et al., 2003; Olszewski et al., 2004; Noble et al., 2006). Figure 6 shows a schematic form of a cross-sectional anatomy of a diseased coronary artery (Sonka et al., 1995). Cardinal et al. (2003) presented a three-dimensional IVUS segmentation model based on the fast-marching method and gray level probability density functions of the vessel wall structures. A mixture of Rayleigh PDFs model the gray level distribution of the whole IVUS pullback. Using this method, the lumen, intima plus plaque structure, and media layers of the vessel wall were computed simultaneously. The results were obtained with average point to point distances between segmented vessel wall borders and ground truth. The authors have also shown the potential of gray level PDF and fastmarching methods in 3D IVUS image processing.

Segmentation Methods in Ultrasound Images

Figure 6. A schematic cross-sectional anatomy of a diseased coronary vessel. (©1995 IEEE, Reprinted with permission from IEEE Transaction on Medical Imaging, 1995, 14(4), pp. 719–732, “Segmentation of intravascular ultrasound images: A knowledge-based approach”, by M. Sonka, X. M. Zhang, M. Siebes, M. S. Bissing, S. C. DeJong, S. M. Collins and C. R. McKay).

Sonka et al. (1995) introduced a semi-automatic knowledge-based method for segmentation of intravascular ultrasound images which identifies the internal and external elastic laminae and the plaque-lumen interface. This approach attempts to incorporate a priori knowledge about crosssectional arterial anatomy, such as objectshape, edge direction, double echo pattern, and wall thickness. The method uses a cost function based on edge strength to find the border. However, the method does not take into account speckle statistics. To assess the performance of the method, they compared five quantitative measures of arterial anatomy derived from borders extracted by the algorithm, with measures derived from borders manually indicated by an expert. Shekhar et al. (1999) developed a threedimensional segmentation technique, called active surface segmentation, for semi-automatic segmentation of the lumen and adventitial borders in serial IVUS images in examinations of coronary arteries. The authors also presented a faster method, based on a fast active contours technique

(a neighborhood-search method) (Klingensmith et al., 2000). Both of their works used only intensity gradient information for snakes. The technique was assessed by computing correlation coefficients and by comparing the results to the expert tracings. Takagi et al. (2000) presented an automated contour analysis assisted by a blood noise reduction algorithm used as preprocessing step. Subtraction of two consecutive IVUS images acquired at the same position in time can increase the signal-to-noise ratio of the lumen area, i.e., obtain good contrast between the lumen and other parts in the image. As the blood echo speckles have higher temporal and spatial variations than the arterial wall, an adaptive filtering of speckle was applied based on pre-segmentation of blood and tissue areas. Their preliminary results showed that automated contour detection facilitated with a blood noise reduction algorithm appeared to be a reliable technique for area measurements in 40-MHz IVUS imaging. Kovalski et al. (2000) developed an algorithm to identify the lumen border and the media-adventitia border (the external elastic membrane). To represent the contours in this method, a 3D balloon model is recruited by using polar coordinates. In this representation, the control points can only move along the radial direction. The reported inter-observer area variability is comparable to the result in (Klingensmith et al., 2003) except that the method in (Kovalski et al., 2000) is automatic. Figure 7 illustrates three sets of images with the lumen and media-adventitia tracings with shadowing artifacts from (Kovalski et al., 2000). The left, middle, and right columns demonstrate the original image, the manual contours, and the result of automatic procedures, respectively. The features are extracted in three dimensions. The results for border contours obtained by this method were compared to the manually extracted results. The method suffers from some tuning of the algorithm’s parameters when using different scanners.

385

Segmentation Methods in Ultrasound Images

Figure 7. The comparison between the manual and the automatic tracings of the lumen and mediaadventitia for three different slices. Left: Original image; middle: Manual tracing and right: automatic detection. (Reprinted from Ultrasound in Medicine & Biology, 26(4), G. Kovalski, R. Beyar, R. Shofti, and H. Azhari, “Three-dimensional automatic quantitative analysis of intravascular ultrasound images”, pp. 527–537, 2000, with permission from Elsevier).

Haas et al. (2000) used an algorithm based on the optimization of a maximum a posteriori estimator. It implements an image generation model based on the Rayleigh distribution of the image pixels and a priori information about the contours. Using the property of the closed contours, they are modeled by first-order Markov random fields, expressing a strong correlation between a contour point and its two next neighbors. During the first stage, the algorithm detected the reliable points and then chose either the global or the first maximum along the radial direction depending on the contour prior energy. Dynamic programming is used to accelerate the estimation process. Additional information from the blood flow is used to initialize the segmentation in 3D image sets. Pardo et al. (2003) proposed a statistical deformable model for 3D segmentation of anatomical organs applied to IVUS images. A bank of Gaussian derivative filters is used to generate a feature space at different orientations and scales, to locally describe each part of the object

386

boundary. The feature space was reduced by linear discriminant analysis and a statistic discriminant snake is used in a supervised learning scheme to guide the model deformation. It is performed by minimizing the dissimilarity between the learned and found image features. The proposed approach is of particular interest for tracking temporal image sequences. Anatomical organs including IVUS are segmented and the results compared to expert tracings for validation. As one of the approaches incorporating highlevel knowledge in IVUS image segmentation, Olszewski et al. (2004) proposed a learning technique based on the human visual system. This method mimics the procedure performed by human experts for automatic detection of the luminal and the medial-adventitial borders. The approach requires no manual initialization or interaction. To verify the accuracy of the method, it is applied on a reasonable data set. A key drawback for Intravascular Ultrasound imaging is its inability to consider the vessel cur-

Segmentation Methods in Ultrasound Images

vature and the orientation of the imaging catheter (Noble et al., 2006). Therefore, the information extracted from this data is distorted, since the vessel curvature remains unconsidered. An answer for correct 3D reconstruction of the IVUS can be derived from the fusion between intravascular ultrasound images and biplane angiography (Noble et al., 2006).

FUTURE TRENDS Ultrasonic imaging stands as one of the most important medical imaging applications of physics and engineering. There have been recent advances in transducer technology and image formation procedure which significantly improve the quality of information obtained from ultrasound devices. There is a need for more effort in the area of segmentation validation to better evaluate the strengths and limitations of the existing methods on larger and more varied databases of images. This would be especially useful for the adoption of methods which are more appropriate in clinical applications. Future efforts for ultrasound segmentation methods should also more effectively consider imaging physics. Basically, a 3D ultrasound image can be constructed from 2D slices which are put together. For such an application, the size and shape of the object of interest is to be extracted in each slice. This means that a fast and reliable segmentation method for individual slices is needed. Considering the correlation between the sequential frames, this can be subjected to further research. 4D ultrasound system provides a next generation for medical imaging applications which allow faster diagnosis and improve treatment success rates. A 4D ultrasound device takes multiple images in rapid succession and creates a 3D motion video, which is very valuable for diagnosis purposes. Any effort towards applying current segmentation methods in these successive frames would be desirable.

CONCLUSION Ultrasound imaging is a non-invasive, easily portable, and relatively inexpensive image modality that has been used for clinical applications for a long time. Ultrasound imaging can be used to visualize the anatomy of the body organs. It also has excellent temporal resolution. In this technique, by exposing a part of the body to high-frequency sound waves, we can generate images of the inside of the body. On the other hand, segmentation is an important image processing task that partitions the image into meaningful regions. These regions are homogeneous with respect to specific characteristics or features such as gray level, texture, etc. Segmentation is an important tool in medical imaging and it is useful for many applications such as feature extraction and classification. Although ultrasound imaging is one of the most widely used technologies for diagnosis and treatment, it is still a challenging case for segmentation tasks due to attenuation, speckle, shadows, and signal dropout. Improvements in this area of research are thus highly desirable. Many methods have been introduced in existing literature to facilitate more accurate automatic or semi-automatic segmentation of ultrasound images. This chapter reviews the ultrasound image segmentation methods by focusing on clinical applications that contain important ideas demonstrating significant clinical usefulness. The focus of this chapter is on reviewing the works which have incorporated prostate transrectal ultrasound (TRUS), breast ultrasound, and intravascular ultrasound (IVUS) images. The future trend for ultrasound image segmentation can focus on improvement of the current methods such that they can be adopted for real clinical applications. Development of new and effective methods for construction of 3D images from 2D slices is also another interesting area of research.

387

Segmentation Methods in Ultrasound Images

As the next generation of ultrasound imaging devices, the 4D ultrasound system should be positioned as a top research area for image segmentation.

REFERENCES Abolmaesumi, P., & Sirouspour, M.R. (2004). Segmentation of prostate contours from ultrasound images. ICASSP04, 3, 517-520. Betrounia, N., Vermandela, M., Pasquierc, D., Maoucheb, S., & Rousseaua, J. (2005). Segmentation of abdominal ultrasound images of the prostate using a priori information and an adapted noise filter. Computerized Medical Imaging and Graphics, 29, 43–51. doi:10.1016/j.compmedimag.2004.07.007 Cardinal, M. H. R., Meunier, J., Soulez, G., Thrasse, E., & Cloutier, G. (2003). Intravascular ultrasound image segmentation: A fastmarching method. Medical Image Computing and Computer Assisted Intervention, Lecture Note Computer Science. Berlin: Springer-Verlag, (pp. 432–439). Chen, D. R., Chang, R. F., Kuo, W. J., Chen, M. C., & Huang, Y. L. (2002). Diagnosis of breast tumors with sonographic texture analysis using wavelet transform and neural networks. Ultrasound in Medicine & Biology, 28(10), 1301–1310. doi:10.1016/S0301-5629(02)00620-8 Chiu, B., Freeman, G. H., Salama, M. M. A., & Fenster, A. (2004). Prostate segmentation algorithm using dyadic wavelet transform and discrete dynamic contour. Physics in Medicine and Biology, 49, 4943–4960. doi:10.1088/00319155/49/21/007 Haas, C., Ermert, H., Holt, S., Grewe, P., Machraoui, A., & Barmeyer, J. (2000). Segmentation of 3-D intravascular ultrasonic images based on a random field model. Ultrasound in Medicine & Biology, 26(2), 297–306. doi:10.1016/S03015629(99)00139-8 388

Horsch, K., Giger, M. L., Venta, L. A., & Vyborny, C. J. (2001). Automatic segmentation of breast lesions on ultrasound. Medical Physics, 28(8), 1652–1659. doi:10.1118/1.1386426 Horsch, K., Giger, M. L., Vyborny, C. J., & Venta, L. A. (2004). Performance of computer-aided diagnosis in the interpretation of lesions on breast sonography. Academic Radiology, 11(3), 272–280. doi:10.1016/S1076-6332(03)00719-0 Hu, N., Downey, D. B., Fenster, A., & Ladak, H. M. (2003). Prostate boundary segmentation from 3D ultrasound images. Medical Physics, 30(7), 1648–1659. doi:10.1118/1.1586267 Huang, Y. L., & Chen, D. R. (2004). Watershed segmentation for breast tumor in 2-D sonography. Ultrasound in Medicine & Biology, 30(5), 625– 632. doi:10.1016/j.ultrasmedbio.2003.12.001 Insana, M. F., & Brown, D. G. (1993). Acoustic scattering theory applied to soft biological tissues. In Ultrasonic Scattering in biological tissues, Boca Raton, CRC Press. Klingensmith, J. D., Shekhar, R., & Vince, D. G. (2000). Evaluation of three dimensional segmentation algorithms for the identification of luminal and medial-adventitial borders in intravascular ultrasound images. IEEE Transactions on Medical Imaging, 19(10), 996–1011. doi:10.1109/42.887615 Klingensmith, J. D., Tuzcu, E. M., Nissen, S. E., & Vince, D. G. (2003). Validation of an automated system for luminal and medial-adventitial border detection in three-dimensional intravascular ultrasound. The International Journal of Cardiovascular Imaging, 19(1), 93–104. Knoll, C., Alcaniz, M., Grau, V., Monserrat, C., & Juan, M. C. (1999). Outlining of the prostate using snakes with shape restrictions based on the wavelet transform. Pattern Recognition, 32, 1767–1781. doi:10.1016/S0031-3203(98)00177-0

Segmentation Methods in Ultrasound Images

Kovalski, G., Beyar, R., Shofti, R., & Azhari, H. (2000). Three-dimensional automatic quantitative analysis of intravascular ultrasound images. Ultrasound in Medicine & Biology, 26(4), 527–537. doi:10.1016/S0301-5629(99)00167-2

Pathak, S. D., Chalana, V., Haynor, D. R., & Kim, Y. (2000). Edge-guided boundary delineation in prostate ultrasound images. IEEE Transactions on Medical Imaging, 19, 1211–1219. doi:10.1109/42.897813

Ladak, H. M., Mao, F., Wang, Y., Downey, D. B., Steinman, D. A., & Fenster, A. (2000). Prostate boundary segmentation from 2D ultrasound images. Medical Physics, 27, 1777–1788. doi:10.1118/1.1286722

Radeva, P., Suri, J. S., & Laxminarayan, S. (2003). Angiography and plaque imaging: Advanced segmentation techniques. Boca Raton, FL: CRC Press.

Madabhushi, A., & Metaxas, D. N. (2003). Combining low-, high-level and empirical domain knowledge for automated segmentation of ultrasonic breast lesions. IEEE Transactions on Medical Imaging, 22(2), 155–169. doi:10.1109/ TMI.2002.808364 Mettlin, C. (1995). American society national cancer detection project. Cancer, 75, 1790–1794. doi:10.1002/10970142(19950401)75:7+3.0.CO;2-Z Nanayakkara, N. D., Samarabandu, J., & Fenster, A. (2006). Prostate segmentation by feature enhancement using domain knowledge and adaptive region based operations. Physics in Medicine and Biology, 51, 1831–1848. doi:10.1088/00319155/51/7/014 Noble, J. A., & Boukerroui, D. (2006). Ultrasound Image Segmentation: A Survey. IEEE Transactions on Medical Imaging, 25, 987–1010. doi:10.1109/TMI.2006.877092 Olszewski, M. E., Wahle, A., Mitchell, S. C., & Sonka, M. (2004). Segmentation of intravascular ultrasound images: A machine learning approach mimicking human vision. International Congress Series, 1268, 1045–1049. doi:10.1016/j. ics.2004.03.252

Richard, W. D., & Keen, C. G. (1996). Automated texture-based segmentation of ultrasound images of the prostate. Computerized Medical Imaging and Graphics, 20(3), 131–140. doi:10.1016/08956111(96)00048-1 Sahiner, B., Chan, H. P., Roubidoux, M. A., Helvie, M. A., Hadjiiski, L. M., & Ramachandran, A. (2004). Computerized characterization of breast masses on threedimensional ultrasound volumes. Medical Physics, 31(4), 744–754. doi:10.1118/1.1649531 Shekhar, R., Cothren, R. M., Vince, D. G., Chandra, S., Thomas, J. D., & Cornhill, J. F. (1999). Three-dimensional segmentation of luminal and adventitial borders in serial intravascular ultrasound images. Computerized Medical Imaging and Graphics, 23, 299–309. doi:10.1016/S08956111(99)00029-4 Shen, D., Herskovits, E. H., & Davatzikos, C. (2001). An adaptive-focus statistical shape model for segmentation and shape modeling of 3D brain structures. IEEE Transactions on Medical Imaging, 20, 257–270. doi:10.1109/42.921475 Shen, D., Zhan, Y., & Davatzikos, C. (2003). Segmentation of Prostate Boundaries From Ultrasound Images Using Statistical Shape Model. IEEE Transactions on Medical Imaging, 22(4), 539–551. doi:10.1109/TMI.2003.809057

Pardo, X. M., Radeva, P., & Cabello, D. (2003). Discriminant snakes for 3-D reconstruction of anatomical organs. Medical Image Analysis, 7(3), 293–310. doi:10.1016/S1361-8415(03)00014-8 389

Segmentation Methods in Ultrasound Images

Sonka, M., Zhang, X. M., Siebes, M., Bissing, M. S., DeJong, S. C., Collins, S. M., & McKay, C. R. (1995). Segmentation of intravascular ultrasound images: A knowledge-based approach. IEEE Transactions on Medical Imaging, 14(4), 719–732. doi:10.1109/42.476113 Takagi, A., Hibi, K., Zhang, X., Teo, T. J., Bonneau, H. N., Yock, P. G., & Fitzgerald, P. J. (2000). Automated contour detection for highfrequency intravascular ultrasound imaging: A technique with blood noise reduction for edge enhancement. Ultrasound in Medicine & Biology, 26, 1033– 1041. doi:10.1016/S0301-5629(00)00251-9 Wang, Y., Cardinal, H. N., Downey, D. B., & Fenster, A. (2003). Semiautomatic three-dimensional segmentation of the prostate using two dimensional ultrasound images. Medical Physics, 30(5), 887–897. doi:10.1118/1.1568975 Webb, A. (2003). Introduction to Biomedical Imaging. IEEE Press Series in Biomedical Engineering.

Xiao, G. F., Brady, M., Noble, J. A., & Zhang, Y. Y. (2002). Segmentation of ultrasound B-mode images with intensity inhomogeneity correction. IEEE Transactions on Medical Imaging, 21(1), 48–57. doi:10.1109/42.981233

KEY TERMS AND DEFINITIONS Breast Ultrasound: Generation of ultrasound images from the breast for diagnostic and treatment purposes. Image Segmentation: This is an important image processing task that partitions the image into meaningful regions. Intravascular Ultrasound: This is an ultrasound medical imaging methodology that uses a catheter with a miniaturized ultrasound probe to visualize the inside walls of blood vessels. Prostate Ultrasound: Generation of ultrasound images from the prostate for diagnostic and treatment purposes. Ultrasound Imaging: This is an ultrasoundbased diagnostic imaging technique used for visualizing internal organs and their structures and possible lesions.

This work was previously published in Handbook of Research on Advanced Techniques in Diagnostic Imaging and Biomedical Applications, edited by Themis P. Exarchos, Athanasios Papadopoulos and Dimitrios I. Fotiadis, pp. 473-487, copyright 2009 by Medical Information Science Publishing (an imprint of IGI Global).

390

391

Chapter 2.11

Exploring Type-and-IdentityBased Proxy Re-Encryption Scheme to Securely Manage Personal Health Records Luan Ibraimi University of Twente, The Netherlands Qiang Tang University of Twente, The Netherlands Pieter Hartel University of Twente, The Netherlands Willem Jonker University of Twente, The Netherlands

ABSTRACT Commercial Web-based Personal-Health Record (PHR) systems can help patients to share their personal health records (PHRs) anytime from anywhere. PHRs are very sensitive data and an inappropriate disclosure may cause serious problems to an individual. Therefore commercial Web-based PHR systems have to ensure that the patient health data is secured using state-of-theart mechanisms. In current commercial PHR DOI: 10.4018/978-1-60960-561-2.ch211

systems, even though patients have the power to define the access control policy on who can access their data, patients have to trust entirely the access-control manager of the commercial PHR system to properly enforce these policies. Therefore patients hesitate to upload their health data to these systems as the data is processed unencrypted on untrusted platforms. Recent proposals on enforcing access control policies exploit the use of encryption techniques to enforce access control policies. In such systems, information is stored in an encrypted form by the third party and there

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

is no need for an access control manager. This implies that data remains confidential even if the database maintained by the third party is compromised. In this paper we propose a new encryption technique called a type-and-identity-based proxy re-encryption scheme which is suitable to be used in the healthcare setting. The proposed scheme allows users (patients) to securely store their PHRs on commercial Web-based PHRs, and securely share their PHRs with other users (doctors).

INTRODUCTION Recently, healthcare providers have started to use electronic health record systems which have significant benefits such as reducing healthcare costs, increasing the patient safety, improving the quality of care and empowering patients to more actively manage their health. There are a number of initiatives for adoption of electronic health records (EHRs) from different governments around the world, such as the directive on privacy and electronic communications in the U.S. known as the Health Insurance Portability and Accountability Act (HIPAA) (The US Department of Health and Human Services, 2003), which specify rules and standards to achieve security and privacy of health data. While EHR systems capture health data entered by health care professionals and access to health data is tightly controlled by existing legislations, personal health record (PHR) systems capture health data entered by individuals and stay outside the scope of this legislation. Before going into details on how to address the confidentiality issues, let us introduce the definition of PHR system (The personal health working group final report, 2004): “An electronic application through which individuals can access, manage and share their health information, and that of others for whom they are authorized, in a private, secure, and confidential environment.”

392

PHR systems are unique in their design since they try to solve the problem that comes from scattering of medical information among many healthcare providers which leads to unnecessary paper work and medical mistakes (The personal health working group final report, 2004). The PHR contains all kinds of health-related information about an individual (say, Alice) (Tang, Ash, Bates, Overhage & Sands, 2006). Firstly, the PHR may contain medical data that Alice has from various medical service providers, for example about surgery, illness, family history, vaccinations, laboratory test results, allergies, drug reactions, etc. Secondly, the PHR may also contain information collected by Alice herself, for example weight change, food statistics, and any other information connected with her health. Controlling access to PHRs is one of the central themes in deploying a secure PHR system. Inappropriate disclosure of the PHRs may cause an individual serious problems. For example, if Alice has some disease and a prospective employer obtains this, then she might be discriminated in finding a job. Commercial efforts to build Web-based PHR systems, such as Microsoft HealthVault (Microsoft, 2007) and Google Health (Google, 2007), allow patients to store and share their PHRs with different healthcare providers. In these systems the patient has full control over her PHRs and plays the role of the security administrator - a patient decides who has the right to access which data. However, the access control model of these applications does not give a patient the flexibility to specify a fine-grained access-control policy. For example, today’s Google Health access control is all-or-nothing - so if a patient authorizes her doctor to see only one PHR, the doctor will be able to see all other PHRs. Another problem is that the data has to be stored on a central server locked by the access control mechanism provided by Microsoft HealthVault or Google Health, and the patient loses control once the data is sent to the server. PHRs may contain sensitive information such as details of a patients disease, drug usage,

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

sexual preferences, therefore, many patients are worried whether their PHRs will be treated as confidential by companies running data centers. Inappropriate disclosure of a PHR can change patients life, and there may be no way to repair such harm financially or technically. Therefore, it is crucial to protect PHRs when they are uploaded and stored in commercial Web-based systems.

PROBLEM STATEMENT The problem addressed in this paper is the confidentiality of patient PHRs stored in commercial Web-based PHR systems. A solution to the problem is a system which would have the following security requirements: •

• • •

Protect patient PHRs from third parties (from the commercial Web-based PHR systems) Provide end-to-end security Allow only authorized users to have access to the patient PHRs Allow the patient to change the access policy dynamically

Contributions To solve the identified problem we propose a new public key encryption scheme called typeand-identity-based proxy re-encryption which helps patients to store their PHRs on commercial Web-based PHR systems, and to share their

PHRs securely with doctors, family and friends. In public key encryption, each user has a key pair (private/public key) and everyone can encrypt a message using a public key, also referred to as the encryption key, but only users who have the associated private key, also referred to as the decryption key, can decrypt the encrypted message. Proxy re-encryption is a cryptographic method developed to delegate the decryption right from one party, the delegator, to another, the delegatee. In a proxy re-encryption scheme, the delegator assigns a key to a proxy to re-encrypt all messages encrypted with the public key of the delegator such that the re-encrypted ciphertexts can be decrypted with the private key of the delegatee. The Proxy is a semi-trusted entity i.e. it is trusted to perform only the ciphertext re-encryption, without knowing the private keys of the delegator and the delegatee, and without having access to the plain text. In the context of public key encryption, proxy reencryption is used to forward encrypted messages without revealing the plaintext. For example, in Figure 1, Alice (delegator) is president of a company who wants to allow her secretary, Bob (delegatee), to read encrypted emails from Charlie when Alice is on vacation (Alice cannot give to Bob her private key). Using proxy re-encryption Alice can compute a re-encryption key which would allow the Proxy to transform a ciphertext for Alice generated by Charlie into a ciphertext for Bob, thus, Bob can decrypt the re-encrypted data using his private key. In practice the role of the Proxy can be played by a commercial enterprise which has enough computation power to perform

Figure 1. Proxy re-encryption

393

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

re-encryption services for a large number of users. Our proposed scheme is suitable for the healthcare setting and has the following benefits: •

•

•

Allow the patient to store her PHRs in a protected form on semitrusted commercial Web-based PHR system. The commercial Web-based PHR system cannot access the data since the data is stored in an encrypted form. Allow the patient to define a fine-grained access control policy. The patient only has to compute the re-encryption key and forward it the Proxy which will re-encrypt the data without decrypting them such that the intended user (e.g., doctor) can decrypt the re-encrypted data using his private key. In addition to that, the scheme allows the patient to change dynamically the access policy without necessarily decrypting the data. Allows the patient to categorize her messages into different types, and delegate the decryption right of each type to the doctor through a Proxy. Data categorization is needed since different data may have different levels of privacy requirements.

WEB-BASED PHRS In this section we discuss current Web-based PHRs systems such as Microsoft HealthVault and Google Health and their access control mechanisms. Moreover, we discuss existing techniques to enforce access policies using cryptography and discuss their limitations. In Microsoft HealthVault each patient has an account and an identifier. Each account has a special role called a custodian who has the right to view and modify all records, and grant or revoke access to users (e.g., doctors) and applications. Both users and applications have to be registered with HealthVault in order to access other accounts. An application is a third party service which can

394

read data stored in HealthVault records, store new data, or transfer data from one account to another account. An application can access patient accounts in two ways: a) online access - access the account when the patient is logged on to her account, and b) offline access – access the account at any time. If the patient uses an application for the first time, and the application requires access to a health record, the application sends an access request to the patient. After the patient approves the request, the application can access the health information. HealthVault uses discretionary access control (DAC) (Graham and Denning, 1971) where access to a patient PHR is based on user identity (e.g. email). For example, if the patient wants to allow the doctor to see her record, the patient has to send a sharing invitation to the doctors’ email and within 3 days the doctor has to accept or reject the invitation. The doctor may have one of the following permissions: a) view patient information, b) view and modify patient information or, c) act as a custodian. Microsoft HealthVault defines 74 types of information and allows granular access control for non-custodians. For example, the patient can allow the doctor to have access only to records of type Allergy, and block access to records of type Family History. However, the patient does not have the flexibility to grant access to the doctor to an individual record. When the patient grants access to the doctor to her health information, the patient can also specify an expiration date. After the expiration date, the doctor would not be able to access the patient information. However, the patient has the option to remove sharing access at any time (even before the expiration date). Similar to Microsoft HealthVault, Google Health allows a patient to import her PHRs, add test results and add information about allergies, among others. In Google Health each patient has an account and an identifier and a user (e.g., doctor) has to be registered to Google Health to access others health information. If the patient wants to allow the doctor to see her PHRs, the patient has

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

to send an invitation to doctors’ email, and within 30 days the doctor has to accept or reject the invitation. A patient can share her PHRs with other users or Google partners including Walgreens, CVS Long Drugs pharmacies, Cleveland Clinic and etc. However, data sharing in Google Health is all-or-nothing. The patients does not have the flexibility to chose a fine-grained access policy to share her data. For example, once the patient allows the doctor to see her blood result test, the doctor can access all health information of the patient. In the access-control mechanisms of Microsoft HealthVault and Google Health the receiving end of the information must provide a set of credentials to the access control manager (ACM) who is responsible for enforcing the access control policies. The ACM checks whether user credentials satisfy the access control policy. If so, the user can read the resource, otherwise not. However, the main limitation of this is that the patient still needs to trust the Microsoft HealthVault and Google Health to enforce access control policies when disclosing data, and specially the access control decisions and privacy enforcement has to be enforced when the data is moving from one enterprise to another. Another limitation is that despite the privacy policy the leakage of confidential information can happen due to compromise of the database maintained by the enterprise. Therefore, to solve the aforementioned problem, recent proposals on enforcing access control policies exploit the use of encryption techniques. In such systems, information is stored in an encrypted form by an enterprise and there is no need for an access control manager to check user credentials. Thus, each user can get the encrypted data, but only users who have the right credentials (the right key) can decrypt the encrypted data. This implies that data confidentiality is preserved even if the database maintained by the enterprise is compromised.

Current Solutions (And Their Drawbacks) Which Enforce Access Control Policies Using Encryption To prevent commercial Web-based PHR systems to access the content of patients PHRs, the patient can encrypt her data using traditional public key encryption algorithms and store the encrypted data (ciphertext) in a database, and then decrypt the ciphertext on demand. In this case, the patient only needs to assume that Microsoft HealthVault and Google Health will properly store her encrypted data, and even if Microsoft HealthVault and Google Health get corrupted, patients PHR will not be disclosed since the data is stored in an encrypted form. The problem with this solution is that the patient needs to be involved in every request (e.g., from her doctor, hospital) and perform the decryption. This is because only the patient knows the decryption key. One possible solution is to use more advanced public key encryption schemes such as CiphertextPolicy Attribute-Based Encryption (CP-ABE) scheme (Bethencourt, Sahai, & Waters, 2007), (Cheung & Newport, 2007), (Ibraimi, Tang, Hartel, & Jonker, 2009). The CP-ABE scheme is a type of attribute-based encryption scheme in which the data owner encrypts the data according to an access control policy τ defined over a set of attributes, and where the receiving end can decrypt the encrypted data only if his private key associated with a set of attributes satisfies the access control policy τ. For example, suppose Alice encrypts her data according to an access policy τ=(a1 AND a2) OR a3. Bob can decrypt the encrypted data only if his private key is associated with a set of attributes that satisfy the access policy. To satisfy the access control policy τ, Bob must have a private key associated with at least one from the following attribute sets: (a1, a2), (a3) or (a1, a2, a3). In CP-ABE the mapping user-attribute is many-to-many, which means that one user may possess many attributes and one attribute may be possessed by many users. The main drawbacks of

395

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

using CP-ABE to securely manage PHRs are: a) the patient has to know the list of users associated with an attribute, and b) an attribute is possesed by many users. Therefore, using CP-ABE, the patient cannot encrypt a PHR such that only one healthcare provider, with whom the patient has contract, can access it at a time. This is because one attribute can be possesed by many healthcare providers. This is one of the main reasons why both Microsoft HealthVault and Google Health use DAC where access to patients records is based on doctors identity, instead of using Attribute-Based Access Control (ABAC) where access to patients records is based on doctors attributes. Another solution is to use existing identitybased proxy re-encryption schemes (Green and Ateniese, 2007), in which the patient assigns a re-encryption key to the Proxy which re-encrypts the encrypted PHR under patients’ public key into encrypted PHR under doctors’ public key. However this approach has the following drawbacks: •

•

The Proxy is able to re-encrypt all ciphertexts of the patient such that the doctor can decrypt all ciphertexts using his private key. Thus, the patient does not have the flexibility to define fine-grained access control. If the Proxy and the delegatee get corrupted, then all PHRs may be disclosed to an illegitimate entity based on the fact that the re-encryption key can re-encrypt all ciphertexts.

THE CONCEPT OF TYPEAND-IDENTITY-BASED PROXY RE-ENCRYPTION In order to solve the aforementioned problems a new encryption scheme which would cryptographically enforce access policies is needed. We propose a type-and-identity-based proxy re-encryption scheme which consists of six algo-

396

rithms: Setup, Extract, Encrypt, Pextract, Preenc and Decrypt. The basic building block of the type-andidentity-based proxy re-encryption scheme is the Identity-Based Encryption (IBE) scheme (Figure 2). The concept of IBE was proposed by Shamir (Shamir, 1985), however IBE has become practical only after Boneh and Franklin (Boneh and Franklin, 2001) propose the first IBE scheme based on bilinear pairings on elliptic curve (In appendix A we review in more detail the concept of bilinear pairing). The IBE scheme consist of four algorithms: Setup, Extract, Encrypt and Decrypt. Unlike a traditional public key encryption scheme, an IBE does not require a digital certificate to certify the encryption key (public key) because the public key of any user can be an arbitrary string such as an email address, IP address, etc. Key escrow is an inherent property in IBE systems, i.e., the trusted authority (TA), also referred to as Key Generation Center (KGC), can generate each users’ private key, because the TA owns the master key mk used to generate users’ private keys. IBE is a very suitable technique to be used in healthcare to exchange emails more securely. For example, in Figure 2, when Alice (the patient) wants to send an encrypted email to Bob (the doctor), Alice can encrypt an email using the encryption key derived from the doctors identity and send the email via an insecure channel. The doctor can authenticate himself to the TA to get the decryption key (private key). After the private key is generated the doctor can decrypt the encrypted email. Unlike in traditional public-key encryption schemes where the private key and the public key has to be created simultaneously, in IBE the private key can be generated long time after the corresponding public key is generated. A type-and-identity-based proxy re-encryption scheme extends the IBE scheme by adding the Proxy entity to the existing two entities: the TA and users. Another type of extension has been made to the number of algorithms. In addition to the four algorithms of IBE scheme, the type- and-

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

Figure 2. Identity-based encryption

identity-based proxy re-encryption scheme uses Pextract algorithm to generate a re-encryption key, and Preenc algorithm to re-encrypt the ciphertext. These algorithms are needed to enable the patient (delegator) to specify a fine-grained access control policy for her PHRs. In the type-and-identity-based proxy reencryption scheme (Figure 3), Alice (delegator) using one key-pair can categorize messages (data) into different types and delegate the decryption right of each type to Bob (delegatee) through a Proxy. Grouping the data into different categories is needed since different PHRs may have different levels of privacy requirements. For example, Alice may not be seriously concerned about disclosing her food statistics to other persons, but she might wish to keep her illness history as a top secret and only disclose it to the appropriate person. In addition to categorizing her PHRs according to the sensitivity level, Alice may categorize her PHRs according to the type of information or according to the device which generated the PHR. There are a number of measurement devices in the market which can be used by the patient and can be connected via home hubs to remote back-end servers. Such examples are disease management services (such as Philips Motiva and PTS) or emergency response services (Philips Lifeline) in which the healthcare provider can remotely access the measurement data and help the patients. As illustrated in Figure 3, Alice uses only one public key to

encryp all messages, and delegates her decryption right (computes a re-encryption key) only for one type (type 1), and Bob can use his private key to access only messages which belong to type 1. If the Proxy and Bob get corrupted, then only health records belonging to type 1 may be disclosed to an illegitimate entity, while the other types of information remains secure. A full construction of the type-and-identity based re-encryption scheme is given in Appendix B. The six algorithms are defined as follows: •

•

•

•

Setup(k): run by the TA, the algorithm takes as input a security parameter k and outputs the master public key pk and the master private key mk. pk is used in the encryption phase by each user, and mk is used by the TA to generate users private keys. Extract(mk,id): run by the TA when a user request a private key. The algorithm takes as input the master key mk and an user identity id, the algorithm outputs the user private key skid. Encrypt(m,t,pkid): run by the encryptor, the algorithm takes as input a message to be encrypted m, a type t, and an the public key pkid associated with the identity id, and outputs the ciphertext c. Pextract(idi , id j , t, skid ) run by the delei

gator, this algorithm takes the delegator’s identifier idi, the delegatee’s identifier idj,

397

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

Figure 3. A type-and-identity-based proxy re-encryption

the type t, and the delegator’s private key skid as input and outputs the re-encryption

A formal security model and a formal security proof is given in appendix B.1. and B.2.

key rkid →id .

Applying the Scheme in Practice

i

i

•

j

Preenc(ci , rkid →id ) run by the Proxy, this i

j

algorithm, takes as input the ciphertext ci and the re-encryption key rkid →id , and i

•

j

outputs a new ciphertext cj Decrypt(c j , skid ) run by the decryptor, j

the algorithm takes as input the ciphertext cj and the private key skid , and output a message m.

j

Figure 4. Secure management of PHR

398

Figure 4 illustrates a general architecture of a PHR system that uses our type-and-identity-based proxy re-encryption scheme. The architecture consists of Trusted Authorities (TAs), a patient, an application hosting device, a Web PHR, a Proxy and a doctor. The TAs are used to generate key pairs for the patient, respectively the doctor. We assume that the patient and the doctor are from different security domains. The application hosting device can be implemented on a home PC of the data source (a patient) or as a trusted service. Its

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

role is to encrypt PHRs and forward them to the Web PHR. The Web PHR stores encrypted PHRs, and Proxy is used to re-encrypt encrypted data and forward them to the doctor. There are five basic processes in the management of PHRs: •

•

•

Setup: In this phase, TAs run the Setup and Extract algorithm and distribute the public parameters needed to run the algorithms of the type-and-identity-based proxy re-encryption scheme, and distributes the private keys, which are needed to decrypt encrypted messages, to the patient and doctor (1). We assume that there is a secure channel between the TA and the user, respectively the doctor. Note that the doctor can get his key pair during the decryption phase, while the Proxy can perform re-encryption of encrypted data under doctors public key, even if the doctor does not have a private key. This is possible since the doctors public key can be computed by everyone who knows doctors’ identity. The TA does not have to be online as long as each user gets his/her key pair. Data creation: The patient uses a number of healthcare devices and creates measurement data and forwards them to the application hosting device (2). In addition to that, the patient can forward to the application hosting device all kinds of information that the patient has from various medical service providers. Data protection: The patient categorizes her PHR according to her privacy concerns. For instance, she can set her illness history as type t1, her food statistics as type t2, and the necessary PHR data in case of emergency as type t3. Then the patient generates the encryption key (public key) derived from her identity (identity can be any type of public string) and run the Encrypt algorithm using the generated public key. After the encryption is performed, the pa-

•

•

tients uploads the encrypted data to Web PHR (3). As in the previous step, all this (data categorization, data encryption and data uploading) can be done by a hosting device on behalf of the patient. Data Sharing (allow the doctor to see patient data): In this phase, the patient runs the Pextract algorithm and generates the re-encryption key which will be used by the Proxy to re-encrypt encrypted data under patients public key to encrypted data under doctors public key, such that the doctor can decrypt the encrypted data using his private key. The generated key is forwarded to the Proxy (4). As in the above steps, all this can be done by a hosting device on behalf of the patient who specifies the access control policy. Data consumption (doctors’ request-response): When a doctor wants to use patient data, he contacts the Web PHR and specifies the ciphertext that he wants to decrypt (5). We assume that each ciphertext is associated with appropriate metadata - descriptive information about the patient. The encrypted data is forwarded to the Proxy (6). The Proxy checks if the doctor is allowed to see patient data (checks if it has a re-encryption key rkPatient→Doctor), and, if so, the Proxy runs the Preenc algorithm to re-encrypt the encrypted data. The re-encrypted data is sent to the doctor (7). After receiving the re-encrypted data the doctor runs the Decrypt algorithm using his private key.

Trust Assumptions User trust is very important aspect when deploying a Web-based PHR system. In practice, users have greater trust on systems where they can control access to their information, and lower trust on systems where they have to trust someone else to control access to their information (e.g., the user

399

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

has to trust the access control manager of the Web PHR). Therefore in this paper we provide a solution which compared to existing solutions reduces the trust that patients need to have on commercial Web-based PHR systems. As mentioned above, in our proposal the role of the Web PHR is twofold: a) to provide storage for PHRs, and b) to maintain the Proxy. Next to that, the patient has to put the following trust on Web PHR: •

•

400

The Web-based PHR system is trusted to store PHRs in publicly accessible storage only in an encrypted form, therefore the patient does not have to trust the Web-based PHR system to provide data confidentiality service. The data confidentiality service is provided by the patient at the moment when the data is encrypted. Encryption prevents sniffing software to access the data when the data travels from the user to the storage, and the storage cannot decrypt the data without having the private key. The Proxy is trusted to maintain a list of re-encryption keys, and to enforce the access policy by properly re-encrypting the encrypted data when the identity of the doctor (requester) is part of the re-encryption key. Note that, the list of re-encryption keys should be secret (if the list of reencryption keys is public then the patient cannot prevent an authorized doctor to see her data after the access decision is made), therefore the patient has to trust the Proxy to store all re-encryption keys securely. The difference between our approach and the access control mechanisms in existing Web PHR is that in our approach the Proxy who plays the role of the ACM cannot access the content of PHRs, therefore, the patient does not have to fully trust the Proxy, while in existing commercial Web-based PHR systems the patient has to fully trust the ACM because the ACM can access the content of PHRs.

•

The user should trust the Proxy to securely delete re-encryption keys when the user wants to prevent an authorized doctor to access users data further. For example, a patient might change her healthcare provider, and after some time she wants to prevent the doctor from an old healthcare provider to access her data.

Policy Updating To allow someone to access her data, the patient has to compute a new re-encryption key. For example, if Alice wants to allow Bob and Charlie to see her PHRs belonging to category tAlergy, Alice has to create two re-encryption keys: rkAlice→Bob and rkAlice→Charlie. Transmission of the reencryption keys to the Proxy should be secured using cryptographical protocols such as Transport Layer Security (TLS) which allows two entities to securely communicate over the internet. In practice the patient might want to update her access policy. Using our approach the patient might do this task without entirely decrypting the ciphertext. To update an access policy means to create new, and delete old, re-encryption keys. For example, if Alice wants to update her access policy from τ=Bob OR Charlie (this access policy implies that Bob and Charlie can access the data) to a different access policy τ ' = Bob OR Dave (this access policy implies that Bob and Dave can access the data), she has to follow the following two procedures: •

•

Notify the Proxy to delete (revoke) the re-encryption key rkAlice→Charlie. This would prevent the Proxy to re-encrypt encrypted data under Alice’s public key to encrypted data under Charlie’s public key. Create a new re-encryption key rkAlice→Dave and send it to the Proxy. This would allow the Proxy to re-encrypt encrypted data

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

under Alice’s public key to encrypted data under Dave’s public key.

RELATED WORK ON PROXY RE-ENCRYPTION Since (Mambo and Okamoto, 1997) first proposed the concept, a number of proxy re-encryption schemes have been proposed. (Blaze, Bleumer, and Strauss, 1998) introduce the concept of atomic Proxy cryptography which is the current concept of proxy re-encryption. In a proxy re-encryption scheme, the Proxy can transform ciphertexts encrypted with the delegator’s public key into ciphertexts that can be decrypted with the delegatee’s private key. (Blaze, Bleumer, & Strauss, 1998) propose a proxy re-encryption scheme based on the (ElGamal, 1985) encryption scheme. One property of this scheme is that, with the same re-encryption key, the Proxy can transform the ciphertexts not only form the delegator to the delegatee but also from the delegatee to the delegator. This is called the “bi-directional” property in the literature. Bi-directionality might be a problem in some applications, but it might also be a desirable property in some other applications. (Jakobsson, 1999) addresses this “problem” using a quorum controlled asymmetric proxy re-encryption where the Proxy is implemented with multiple servers and each of them performs partial re-encryption. (Ivan & Dodis, 2003) propose a generic construction method for proxy re-encryption schemes and also provide a number of example schemes. Their constructions are based on the concept of secret splitting, which means that the delegator splits his private key into two parts and sends them to the Proxy and the delegatee separately. During the re-encryption process the Proxy performs partial decryption of the encrypted message using the first part of the delegator’s private key, and the delegatee can recover the message by performing partial decryption using the second part of the delegator’s private key. One disadvantage of this method is that it is not collusion-safe, i.e. the

Proxy and the delegatee together can recover the delegator’s private key. Another disadvantage of this scheme is that the delegatee’s public/private key pair can only be used for dealing with the delegator’s messages. If this key pair is used by the delegatee for other encryption services, then the delegator can always decrypt the ciphertexts. (Ateniese, Fu, Green, & Hohenberger, 2006) propose several proxy re-encryption schemes based on the (ElGamal, 1985) scheme. In their schemes, the delegator does not have to interact and share his private key with the delegatee. The delegator stores two private keys, a master private key and a “weak” private key. The ciphertext can be fully decrypted using either of the two distinct keys. Their scheme is collusion safe, since only the “weak” private key is exposed if the delegatee and the Proxy collude but the master key remains safe. In addition, (Ateniese, Fu, Green, & Hohenberger, 2006) also discuss a number of properties for proxy re-encryption schemes. Recently, two IBE proxy re-encryption schemes were proposed by (Matsuo, 2007) and (Green & Ateniese, 2007), respectively. The Matsuo scheme assumes that the delegator and the delegatee belong to the same KGC and use the (Boneh & Boyen, 2004) encryption scheme. The (Green & Ateniese, 2007) scheme assumes that the delegator and the delegatee can belong to different KGCs but the delegatee possess the public parameter of the delegator’s KGC. (Sahai and Waters, 2005) introduce the concept of Attribute-Based Encryption (ABE) which is a generalized form of IBE. In ABE the ciphertext and user private keys are associated with a set of attributes. A user can decrypt the ciphertext if the user private key has the list of attributes specified in the ciphertext. In Ciphertext-Policy Attribute-Based Encryption (CP-ABE) the user private key is associated with a set of attributes and a ciphertext is associated with an access control over some attribute. The decryptor can decrypt the ciphertext if the list of attributes associated with the private key satisfies the access policy.

401

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

In Key-Policy Attribute-Based Encryption (KPABE) (Goyal, Pandey, Sahai, & Waters, 2006) the idea is reversed and the private key is associated with an access control over some attributes and the ciphertext is associated with a list of attributes. The decryptor can decrypt the ciphertext if the list of attributes associated with the ciphertext satisfy the access policy associated with the private key. (Liang, Cao, Lin, & Shao, 2009) proposed an attribute-based proxy re-encryption scheme. The scheme is based on the (Cheung & Newport, 2007) CP-ABE scheme and inherits the same limitations that (Cheung & Newport, 2007) has: it supports only access policies with AND boolean operator, and the size of the ciphertext increases linearly with the number of attributes in the system. Proxy re-encryption has many promising applications including access control in file storage (Ateniese, Fu, Green, & Hohenberger, 2006), email forwarding (Wang, Cao, T. Okamoto, Miao & E. Okamoto, 2006), and law enforcement (Mambo & Okamoto, 1997). With the increasing privacy concerns over personal data, proxy re-encryption, in particular IBE proxy re-encryption schemes, will find more and more applications. As we show in this paper, proxy re-encryption is a powerful tool for patients to enforce their PHR disclosure policies.

CONCLUSION This paper presents a new approach for secure management of PHRs which are stored and shared from a semitrusted web server (the server is trusted to perform only the ciphertext re-encryption, without having access to the plain text). We gave an overview of access control mechanisms employed in current commercial Web-based PHR systems and show that traditional access control mechanisms as well as traditional encryption techniques which enforce access control policies are not suitable to be used in scenarios where the data is outsourced to a third party data center. In

402

this paper we propose a type-and-identity-based proxy re-encryption scheme which allow patients to reduce the trust on commercial Web-base PHR systems and enable patients to provide different re-encryption capabilities to the Proxy while using the same key pair. This property has been shown to be useful in our PHR disclosure system, where an individual can easily implement fine-grained access control policies to her PHRs.

REFERENCES Ateniese, G., Fu, K., Green, M., & Hohenberger, S. (2006). Improved proxy re-encryption schemes with applications to secure distributed storage. [TISSEC]. ACM Transactions on Information and System Security, 9(1), 1–30. doi:10.1145/1127345.1127346 Bethencourt, J., Sahai, A., & Waters, B. (2007). Ciphertext-policy attribute-based encryption. In D. Shands (Ed.), Proceedings of the 28th IEEE Symposium on Security and Privacy (pp. 321-334). Oakland, CA: Citeseer. Blaze, M., Bleumer, G., & Strauss, M. (1998). Divertible protocols and atomic proxy cryptography. In K. Nyberg (Ed.), Proceeding of Eurocrypt 1998 (Vol. 1403 of LNCS) (pp. 127-144). New York: Springer Verlag. Boneh, D., & Boyen, X. (2004). Efficient selectiveid secure identity-based encryption without random oracles. In C. Cachin & J. Camenisch (Eds.), Proceedings of Eurocrypt 2004 (Vol. 3027 of LNCS) (pp. 223-238). New York: Springer Verlag. Boneh, D., & Franklin, M. K. (2001). Identitybased encryption from the weil pairing. In J. Kilian (Ed.), Proceedings of Crypto 2001 (Vol. 2139 of LNCS) (pp. 213-229). New York: Springer Verlag.

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

Chen, L. (2007). An interpretation of identitybased cryptography. In A. Aldini & R. Gorrieri (Eds.), Foundations of Security Analysis and Design, IV FOSAD 2006/2007 Tutorial Lectures (Vol. 4677 of LNCS) (pp. 183-208). New York: Springer Verlag. Cheung, L., & Newport, C. (2007). Provably secure ciphertext policy ABE. In P. Ning (Ed.), Proceedings of the 14th ACM Conference on Computer and Communications Security (pp. 456–465). New York: ACM. ElGamal, T. (1985). A public key cryptosystem and a signature scheme based on discrete logarithms. In G. R. Blakley & D. Chaum (Eds.), Proceedings of Crypto 1984(Vol. 196 of LNCS) (pp. 10-18). New York: Springer Verlag. Google. (2008). Google Health. Retrieved September 15, 2009 from http://www.google.com/ health Goyal, V., Pandey, O., Sahai, A., & Waters, B. (2006). Attribute-based encryption for finegrained access control of encrypted data. In A. Juels (Ed.), Proceedings of the 14th ACM Conference on Computer and Communications Security (pp. 89-98). New York: ACM. Graham, G. S., & Denning, P. J. (1971). Protection: principles and practice. In Proceedings of the November 16-18, 1971, Fall Joint Computer Conference (pp. 417–429). Green, M., & Ateniese, G. (2007). Identity-based proxy re-encryption. In J. Katz & M. Yung (Eds.), Proceedings of Applied Cryptography and Network Security (Vol. 4521 of LNCS) (pp. 288-306). New York: Springer Verlag. Ibraimi, L., Tang, Q., Hartel, P., & Jonker, W. (2009). Efficient and provable secure ciphertextpolicy attribute-based encryption schemes. In F. Bao, H. Li, & G. Wang (Eds.), Proceedings of Information Security Practice and Experience (Vol. 5451 of LNCS) (pp. 1-12). New York: Springer Verlag.

Ivan, A., & Dodis, Y. (2003). Proxy cryptography revisited. In C. Neuman (Ed.), Proceedings of the Network and Distributed System Security Symposium. Citeseer. Jakobsson, M. (1999). On quorum controlled asymmetric proxy re-encryption. In H. Imai & Y. Zheng (Eds.), Proceedings of Public Key Cryptography (Vol. 1560 of LNCS) (pp. 112–121). New York: Springer Verlag. Liang, X., Cao, Z., Lin, H., & Shao, J. (2009). Attribute based proxy re-encryption with delegating capabilities. In W. Li, W. Susilo, & U. Tupakula (Eds.), Proceedings of the 4th International Symposium on Information, Computer, and Communications Security (pp. 276-286). New York: ACM. Mambo, M., & Okamoto, E. (1997). Proxy Cryptosystems: Delegation of the power to decrypt ciphertexts. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Science, 80(1), 54–63. Matsuo, T. (2007). Proxy re-encryption systems for identity-based encryption. In T. Takagi, T. Okamoto, E. Okamoto, & T. Okamoto (Eds.), Proceedings of Pairing-Based Cryptography Pairing 2007 (Vol. 4575 of LNCS) (pp. 247-267). New York: Springer Verlag. Microsoft. (2007). HealthVault Connection Center. Retrieved September 15, 2009 from http:// www.healthvault.com/ Sahai, A., & Waters, B. (2005). Fuzzy identitybased encryption. In R. Cramer (Ed.), Proceedings of Eurocrypt 2005 (Vol. 3494) (pp. 457-473). New York: Springer Verlag. Shamir, A. (1985). Identity-based cryptosystems and signature schemes. In G. R. Blakely & D. Chaum (Eds.) Proceedings of Crypto1984 (Vol. 196 of LNCS) (pp. 47-53). New York: Springer Verlag.

403

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

Shoup, V. (2006). Sequences of games: a tool for taming complexity in security proofs. Retrieved October 15, 2009 from http://shoup.net/papers/ games.pdf Tang, P. C., Ash, J. S., Bates, D. W., Overhage, J. M., & Sands, D. Z. (2006). Personal health records: definitions, benefits, and strategies for overcoming barriers to adoption. Journal of the American Medical Informatics Association, 13(2), 121–126. doi:10.1197/jamia.M2025 The personal health working group final report. (2004). Connecting for health. Retrieved October 2, 2009 from http://www.connectingforhealth.org/ resources/wg_eis_final_report_0704.pdf

404

The US Department of Health and Human Services. (2003). Summary of the HIPAA privacy rule. Retrieved October 2, 2009 from http://www.hhs. gov/ocr/privacy/hipaa/understanding/summary/ privacysummary.pdf Wang, L., Cao, Z., Okamoto, T., Miao, Y., & Okamoto, E. (2006). Authorization-limited transformation-free proxy cryptosystems and their security analyses. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Science, (1): 106–114. doi:10.1093/ ietfec/e89-a.1.106

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

APPENDIX A. Review of Pairing We briefly review the basis of pairing and the related assumptions. More detailed information can be found in the seminal paper (Boneh and Franklin, 2001). A pairing (or, bilinear map) satisfies the following properties: 1. G and G1 are two multiplicative groups of prime order p; 2. g is a generator of G; 3. eˆ : G × G → G1 is an efficiently-computable bilinear map with the following properties: ◦⊦ Bilinear: for all u.v∈G and a, b ∈ Z*p , we have eˆ(u a , vb ) = eˆ(u, v )ab . ◦⊦ Non-degenerate: eˆ(g, g ) ≠ 1 . As defined in (Boneh and Franklin, 2001), G is said to be a bilinear group if the group action in G can be computed efficiently and if there exists a group G1 and an efficiently-computable bilinear map eˆ as defined above. The Bilinear Diffie-Hellman (BDH) problem in G is as follows: given g, ga, gb, gc ∈ G as input, output eˆ(g, g )abc ∈ G1 . An algorithm A has advantage ε in solving BDH in G if: Pr[ A(g, g a , g b , g c ) = eˆ(g, g )abc ] ≥ ε. Similarly, we say that an algorithm A has advantage ε in solving the decision BDH problem in G if: |Pr[A(g, ga, gb, gc, gabc)=0] – Pr[A(g, ga, gb, gc, T)=0]|≥ε. Here the probability is over the random choice of a, b, c ∈ Z*p , the random choice of T∈G1, and the random bits of A (the adversary is a nondeterministic algorithm). Definition 1.We say that the (decision) (t,ε) -BDH assumption holds in G if no t-time algorithm has advantage at least ε in solving the (decision) BDH problem in G. As in the general group, the Computational Diffie-Hellman (CDH) problem in G is as follows: given g, ga, gb ∈ G as input, output gab ∈ G. An algorithm A has advantage ε in solving CDH in G if:

Pr[A (g, ga, gb) = gab]≥ε.

Definition 2.We say that the (t,ε) -CDH assumption holds in G if no t-time algorithm has advantage at least ε in solving the CDH problem in G.

405

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

Given a security parameter k, a problem (say, BDH) is believed to be intractable if any adversary has only negligible advantage in reasonable time. We usually define a scheme to be secure if any adversary has only a negligible advantage in the underlying security model. The time parameter is usually be ignored. Definition 3.The function P(k):Z→R is said to be negligible if, for every polynomial f(k), there exists 1 an integer Nf such that P (k ) ≤ for all k≥Nf. f (k )

B. Our Construction In this section we give the construction of the type-and-identity-based proxy re-encryption scheme. In our scheme, the delegator and the delegatee are allowed to be from different domains, which nonetheless share some public parameters. •

Suppose that the delegator is registered at KGC1 in Boneh-Franklin IBE scheme (Setup1, Extract1, Encrypt1, Decrypt1) and the delegatee is registered at KGC2 in Boneh-Franklin IBE scheme (Setup2, Extract2, Encrypt2, Decrypt2). The algorithms are defined as follows. ◦⊦ Setup1 and Extract1 are the same as in the Boneh-Franklin scheme, except that Setup1 outputs an additional hash function H2 : {0,1}* → Z*p . The public parameter is params1 = (G, G1, p, g, H1, H2 , eˆ, pk1 ) , and the master key is mk1=α1. Encrypt1(m,t,id): Given a message m, a type t, and an identifier id, the algorithm outputs the ciphertext c= (c1, c2, c3) where r ∈R Z*p ,

◦⊦

r ⋅H (sk ||t ) 2 id

c1 = g r , c2 = m ⋅ eˆ(pkid , pk )

Decrypt1(c,skid): Given a ciphertext c= (c1, c2, c3), the algorithm outputs the message

◦⊦ m=

, c3 = t .

c2 H (sk ||c ) 2 id 3

eˆ(skid , c1 )



Without loss of generality, suppose the delegator holds the identity idi and the corresponding private key skid . Apart from the delegator, another party cannot run the Encrypt1 algorithm under the delegai

tor’s identity idi since he does not know skid . i

Suppose that the delegatee (with identity idj) possesses private key skid registered at KGC2 in the j

Boneh-Franklin IBE scheme, where the public parameter is params2 = (G, G1, p, g, H1, eˆ, pk2 ) , the α

master key is mk2=α2, and skid = H1(id j ) 2 . For the ease of comparison, we denote the IBE scheme as j

(Setup2, Extract2, Encrypt2, Decrypt2) although these algorithms are identical to those described in Section B.

406

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

The Delegation Process If the delegator wants to delegate his decryption right for messages with type t to the delegatee, the algorithms of the proxy re-encryption scheme are as follows. Pextract(idi , id j , t, skid ) : Run by the delegator, this algorithm outputs the re-encryption key

•

i

rkid →id , where X∈RG1 and i

j

−H (sk ||t ) 2 id i

rkid →id = (t, skid i

•

j

i

⋅ H1(X ), Encrypt2 (X , id j )).

Preenc(ci , rkid →id ) : Run by the Proxy, this algorithm, takes a ciphertext ci=(ci1, ci2, ci3) and the i

j

re-encryption key rkid →id as input where t=ci3, and outputs a new ciphertext cj=(cj1, cj2, cj3), i

where cj1=ci1 and

−H (sk ||c ) 2 id i 3 i

c j 2 = ci 2 ⋅ eˆ(ci 1, skid α

= m ⋅ eˆ(g 1 , pk

i rH (sk ||t ) 2 id i id i

j

⋅ H1 (X )) −H (sk ||t ) 2 id i

) ⋅ eˆ(g r , skid

i

= m ⋅ eˆ(g r , H1(X )),

⋅ H1(X ))

and cj3=Encrypt2(X,idj). •

Decrypt(c j , skid ) .Given a re-encrypted ciphertext cj, the delegatee can obtain the plaintext m by j

computing: m′ =

cj 2 eˆ(c j 1, H1 (Decrypt2 (c j 3 , skid ))) r

=

m ⋅ eˆ(g , H1 (X ))

eˆ(g r , H1(X )) = m.

j



B.1. Security Model We assume that the Proxy is semi-trusted in the following sense: it will honestly convert the delegator’s ciphertexts using the re-encryption key; however, it might act actively to obtain some information about the plaintexts for the delegator and the delegatee. As mentioned in (Chen, 2007), the key escrow problem (TA owns a master key which can be used to decrypt each encrypted data) can be avoided by applying some standard techniques (such as secret sharing) to the underlying scheme, hence, we skip

407

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

any further discussion in this paper. The delegatee may be curious in the sense that it may try to obtain some information about the plaintexts corresponding to the delegator’s ciphertexts which have not been re-encrypted by the Proxy. As a standard practice, we describe an attack game for modeling the semantic security against an adaptive chosen plaintext attack for the delegator (IND-ID-DR-CPA security) for our scheme. The INDID-DR-CPA game is carried out between a challenger and an adversary, where the challenger simulates the protocol execution and answers the queries from the adversary. Note that the allowed queries for the adversary reflect the adversary’s capability in practice. Specifically, the game is as follows: 1. Game setup: The challenger takes a security parameter k as input, runs the Setup1 algorithm to generate the public system parameter params1 and the master key mk1, and runs the Setup2 algorithm to generate the public system parameter params2 and the master key mk2. 2. Phase 1: The adversary takes params1 and params2 as input and is allowed to issue the following types of queries: a. Extract1 query with any identifier id: The challenger returns the private key sk corresponding to id. b. Extract2 query with any identifier id ′ : The challenger returns the private key sk ′ corresponding to id ′ . c. Pextract query with (id, id ′, t ) : The challenger returns the re-encryption key rkid →id ′ for the type t. d. Preenc† query with (m, t, id, id ′) : The challenger first computes c=Encrypt1(m,t,id) and then returns a new ciphertext c ′ which is obtained by applying the delegation key rkid →id ′ to c, where rkid →id ′ is issued for type t.

Once the adversary decides that Phase 1 is over, it outputs two equal length plaintexts m0, m1, a type t , and an identifier id*. At the end of Phase 1, there are three constraints here: *

a. id* has not been the input to any Extract1 query. b. For any id ′ , if (id * , id ′, t * ) has been the input to a Pextract query then id ′ has not been the input to any Extract2 query. c. If there is a Preenc† query with (m, t, id, id ′) , then (id, id ′, t ) has not been queried to Pextract. 3. Challenge: The challenger picks a random bit b∈{0,1} and returns c* = Encrypt1 (mb, t*, id*) as the challenge to the adversary. 4. Phase 2: The adversary is allowed to continue issuing the same types of queries as in Phase 1. At the end of Phase 2, there are the same constraints as at the end of Phase 1. 5. Guess (game ending): The adversary outputs a guess b ′ ∈ {0,1}. 1 | . Compared with 2 the CPA security formalizations in (Ateniese, Fu, Green and Hohenberger, 2006), in our case, we also take into account the categorization of messages for the delegator. The Preenc† query reflects the fact that a curious delegatee has access to the the delegator’s plaintexts. At the end of the game, the adversary’s advantage is defined to be | Pr[b ′ = b ] −

408

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

B.2. Security Proof Theorem 1.For the type-and-identity-based proxy re-encryption scheme described in Section B, any adversary’s advantage is negligible. Proof sketch. We suppose that the total number of queries issued to H1 and H2 is bounded by integer q1 and q2, respectively. Suppose an adversary A has the non-negligible advantage ε in the IND-ID-DR-CPA game. The security proof is done through a sequence of games. Game0: In this game, B faithfully answers the oracle queries from A. Specifically, B simulates the random oracle H1 as follows: B maintains a list of vectors, each of them containing a request message, an element of G (the hash-code for this message), and an element of Z*p . After receiving a request message, B first checks its list to see whether the request message is already in the list. If the check succeeds, B returns the stored element of G; otherwise, B returns gy, where y a randomly chosen element of Z*p , and stores the new vector in the list. A' simulates the random oracle H2 as follows: B maintains a list of vectors, each of them containing a request message and an element of Z*p (the hash-code for this message). After receiving a request message, B first checks its list to see whether the request message is already in the list. If the check succeeds, B returns the stored element of Z*p ; otherwise, B returns u which is a randomly chosen element of Z*p , and stores the new vector in the list. 1 |= ε . 2 Game1: In this game, B answers the oracle queries from A as follows.

Let δ0 = Pr[b ′ = b ] , as we assumed at the beginning, | δ0 −

• •

•

Game setup: B faithfully simulates the setup phase. Phase 1: B randomly selects j∈{1,2,…,q1+1}. If j=q1+1, B faithfully answers the oracle queries  and B answers the oracle queries from A from A. If 1≤j≤q1, we assume the j-th input to H1 is id as follows: Answer the queries to Extract1, Extract2, Pextract, and Preenc† faithfully, except that  is the input to a Extract query. B aborts as a failure when id 1 Challenge: After receiving (m0, m1, t*, id*) from the adversary, if one of the following events occurs, B aborts as a failure. a. id* has been issued to H1 as the i-th query and i≠j, b. id* has not been issued to H1 and 1≤j≤q1.

 is the input to j-th H Note that, if the adversary does not abort then either 1≤j≤q1 and id * = id 1 * query or j=q1+1 and id has not been the input to any H1 query. B faithfully returns the challenge. • •

Phase 2: B answers the oracle queries faithfully. Guess (game ending): The adversary outputs a guess b ′ ∈ {0,1} .

409

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

The probability that B successfully ends is execution is

1 , i.e. the probability that B does not abort in its q1 + 1

1 . Let δ1 = Pr[b ′ = b ] when B successfully ends, in which case |δ1=δ0|. Let θ1 be the q1 + 1

probability that B successfully ends and b ′ = b . We have θ1 =

δ1

. q1 + 1 Game2: In this game, B simulates the protocol execution and answers the oracle queries from A in the following way. α

Game setup: B faithfully simulates the setup phase. Recall that pk1 = g 1 . Phase 1: B randomly selects j∈{1,2,…,q1+1}. If j=q1+1, B faithfully answers the oracle queries from A. If 1≤j≤q1, B answers j-th query to H1 with gβ where β ∈R Z*p , and answers the oracle  . queries from A as follows. Suppose the input of the j-th query to H1 is id  is the input a. Answer Extract and Extract faithfully, except that B aborts as a failure when id

• •

1

2

to a Extract1 query.  , B returns the re-encryption key rk , where Pextract query with (id, id ′, t ) : If id = id id →id ′

gtid ′ ∈R G, Xtid ′ ∈R G1, rkid →id ′ = (t, gtid ′ , Encrypt2 (Xtid ′ , id ′)). Otherwise, B answers the query faithfully. If id ′ has been queried to Extract2, when Xtid ′ is queried

−1 to H1 then B returns gtid ′ ⋅ htid where htid ′ ∈R G . ′ †  , B returns Preenc query with (m, t, id, id ′) : If id = id

r ∈R Z*p , Xtid ′ ∈R G1, c ′ = (g r , eˆ(g r , H1 (Xtid ′ )), Encrypt2 (Xtid ′ , id ′)). Otherwise, B answers the query faithfully. •

Challenge: After receiving (m0, m1, t*, id*) from the adversary, if one of the following events occurs, B aborts as a failure. a. id* has been issued to H1 as the i-th query and i≠j, b. id* has not been issued to H1 and 1≤j≤q1.

 is the input to j-th H Note that, if the adversary does not abort then either 1≤j≤q1 and id * = id 1 query or j=q1+1 and id* has not been the input to any H1 query. In the latter case, B sets H1(id*)=gβ where β ∈R Z*p , and returns c * = (c1* , c2* , c3* ) as the challenge to the adversary, where: b ∈R {0,1}, r ∈R Z*p , T ∈R G1, c1* = g r , c2* = mb ⋅ T , c3* = t * .

410

Exploring Type-and-Identity-Based Proxy Re-Encryption Scheme

• •

Phase 2: B answers the oracle queries from A as in Phase 1. Guess (game ending): The adversary outputs a guess b ′ ∈ {0,1} . Let θ2 be the probability that B successfully ends and b ′ = b . We have θ2 =

1 since T∈RG1. 2(q1 + 1) α ⋅β

Let E1 be the event that, for some id ′ and t, the adversary issues a H2 query with the input g 1 || t or Xtid ′ is issued to H1 while id ′ has not been issued to Extract2. Compared with Game1, Game2 differs when E1 occurs. From the difference lemma (Shoup, 2006), we have |δ2-δ1|≤ε2 which is negligible in the random oracle model based on the BDH assumption. Note that (Setup2, Extract2, Encrypt2, Decrypt2) is one-way based on the BDH assumption and BDH implies CDH. 1 1 1 − θ1 |≤ ε2 . In addition, from | δ0 − |= ε , we have | From |θ2-θ1|≤ε2 and θ2 = 2 2(q1 + 1) 2(q1 + 1) ε1 ε ≤ + ε2 . Because εi (1≤i≤2) are negligible and ε is q1 + 1 q1 + 1 q1 + 1 assumed to be non-negligible, we get a contradiction. As a result, the proposed scheme is IND-ID-DRCPA secure based on the CDH assumption in the random oracle model, given that (Setup2, Extract2, Encrypt2, Decrypt2) is one-way. , |δ1-δ0|≤ε1 and θ1 =

δ1

, we have

This work was previously published in International Journal of Computational Models and Algorithms in Medicine (IJCMAM), Volume 1, Issue 2, edited by Aryya Gangopadhyay, pp. 1-21, copyright 2010 by IGI Publishing (an imprint of IGI Global).

411

412

Chapter 2.12

Integration of Clinical and Genomic Data for Decision Support in Cancer Yorgos Goletsis University of Ioannina, Greece Themis P. Exarchos University of Ioannina, Greece Nikolaos Giannakeas University of Ioannina, Greece Markos G. Tsipouras University of Ioannina, Greece Dimitrios I. Fotiadis University of Ioannina, Greece, Michaelideion Cardiology Center, Greece & Biomedical Research Institute, Greece

INTRODUCTION Computer aided medical diagnosis is one of the most important research fields in biomedical engineering. Most of the efforts made focus on diagnosis based on clinical features. The latest breakthroughs of the technology in the biomolecular sciences are a direct cause of the explosive growth of biological data available to the scientific community. New technologies allow for high volume affordable production and colDOI: 10.4018/978-1-60960-561-2.ch212

lection of information on biological sequences, gene expression levels and proteins structure on almost every aspect of the molecular architecture of living organisms. For this reason, bioinformatics is asked to provide tools for biological information processing, representing today’s key in understanding the molecular basis of physiological and pathological genotypes. The exploitation of bioinformatics for medical diagnosis appears as an emerging field for the integration of clinical and genomic features, maximizing the information regarding the patient’s health status and the quality of the computer aided diagnosis.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Integration of Clinical and Genomic Data for Decision Support in Cancer

Cancer is one of the prominent domains where this integration is expected to bring significant achievements. As genetic features play a significant role in the metabolism and the function of the cells, the integration of genetic information (proteomics-genomics) to cancer-related decision support is now perceived by many not as a future trend but rather as a demanding need. The usual patient management in cancer treatment involves several, usually iterative, steps consisting of diagnosis, staging, treatment selection, and prognosis. As the patient is usually asked to perform new examinations, diagnosis and staging status can change over time. On the other hand, treatment selection and prognosis depend on the available findings, response to previous treatment plan and, of course, clinical guidelines. The integration of these evolving and changing data into clinical decision is a hard task which makes fully personalised treatment plan almost impossible. The use of clinical decision support systems (CDSSs) can assist in the processing of the available information and provide accurate staging, personalised treatment selection, and prognosis. The development of electronic patient records and of technologies that produce and collect biological information have led to a plethora of data characterizing a specific patient. Although this might seem beneficial, it can lead to confusion and weakness concerning the data management. The integration of the patient data (quantitative) that are hard to be processed by a human decision maker (the clinician) further imposes the use of CDSSs in personalized medical care (Louie, Mork, Martin-Sanchez, Halevy, & Tarczy-Hornoch, 2007). The future vision—but current need—will not include generic treatment plans according to some naive reasoning, but totally personalised treatment based on the clinicogenomic profile of the patient. In this article, we address decision support for cancer by exploiting clinical data and identifying mutations on tumour suppressor genes. The goal is to perform data integration between medicine

and molecular biology by developing a framework where clinical and genomic features are appropriately combined in order to handle cancer diseases. The constitution of such a decision support system is based on (a) cancer clinical data and (b) biological information that is derived from genomic sources. Through this integration, real time conclusions can be drawn for early diagnosis, staging and more effective cancer treatment.

BACKGROUND Clinical Decision Support Systems are active knowledge systems which use two or more items of patient data to generate case-specific advice (Fotiadis, Goletsis, Likas, & Papadopoulos, 2006). CDSSs are used to enhance diagnostic efforts and include computer based programs that, based on information entered by the clinician, provide extensive differential diagnosis, staging (if possible), treatment, follow-up, and so forth. CDSSs consist of an inference engine that is used to associate the input variables with the target outcome. This inference engine can be developed based either on explicit medical knowledge, expressed in a set of rules (knowledge based systems) or on data driven techniques, such as machine learning (Mitchel, 2006) and data mining (intelligent systems) (Tan, Steinbach, & Kumar, 2005). CDSSs require the input of patient-specific clinical variables (medical data) and as a result provide patient specific recommendation. Medical data are observations regarding a patient, including demographic details (i.e., age, sex), medical history (i.e., diabetes, obesity), laboratory examinations (e.g., creatinine, triglyceride), biomedical signals (ECG, EMG), medical images (i.e., MRI, CT), and so forth. Demographic details, medical history, and laboratory data are the most easily obtained and recorded and, therefore, most commonly included in electronic patient records. On the other hand, biomedical signals and medical images require more effort in order to be acquired

413

Integration of Clinical and Genomic Data for Decision Support in Cancer

in a digital format and must be processed for useful feature extraction. Apart from these, several types of genomic data can be generated from laboratory examinations, that is, gene DNA or protein sequences, gene expression data, microarray images, and so forth. Genomic data can also be used for medical diagnosis, disease prevention, and population genetics studies. Although medical data are sufficient for the diagnosis of several diseases, recent studies have demonstrated the high information value of genomic data, especially in specific types of diseases, such as cancer diseases. The great amount and the complexity of the available genetic data complicates their analysis from conventional data analysis methods and requires higher order analysis methods such as data mining techniques. Lately, data mining has received much attention in bioinformatics and molecular biology (Cook, Lawrence, Su, Maglothin, & Jonyer, 2001). Data mining methods are usually applied in the analysis of data coming from DNA microarrays or mass spectrometry. Over the last few years, several scientific reports have shown the potential of data mining to infer clinically relevant models from molecular data and thus provide clinical decision support. The majority of papers published in the area of data mining for genomic medicine deals with the analysis of gene expression data coming from DNA microarrays (Jiang & Gruenwald, 2005; Shah & Kusiak, 2007; Walker et al., 2004) consisting of thousands of genes for each patient, with the aim to diagnose types of diseases and to obtain a prognosis which may lead to therapeutic decisions. Most of the research works are related to oncology (Louie et al., 2007), where there is a strong need for defining individualized therapeutic strategies. Another area where data mining has been applied is the analysis of genes or proteins, represented as sequences (Exarchos, Papaloukas, Lampros, & Fotiadis, 2006); sequential pattern mining techniques are suitable for analyzing these types of data (Zaki, 2000).

414

Several CDSSs for cancer have been proposed in the literature. Most of the approaches are based solely on clinical data and a few methods exist that provide cancer decision support using microarray gene expression data. The cancer CDSSs concern several different types of cancer and employ various techniques for their development. The majority of systems are still in a research level and only a few are being used in clinical practice. A CDSS which is already in clinical use is PAPNET (Boon & Kok, 2001) which deals with cervical cancer. PAPNET uses ANNs to extract abnormal cell appearances from vaginal smear slides and describe them in histological terms. Other CDSSs for cervical cancer concentrate on the evaluation of the benefits of the PAPNET system (Doornewaard, 1999; Nieminen, Hakama, Viikki, Tarkkanen, & Anttila, 2003). Colon cancer has also been studied using clinical data and fuzzy classification trees (Chiang, Shieh, Hsu, & Wong, 2005) or pattern analysis of gene expression levels (Alon et al., 1999). A CDSS that combines imaging data with pathology data for colon cancer has also been proposed (Slaymaker, 2006). CDSSs proposed for prostate cancer, employ prostate specific antigen (PSA) serum marker, digital rectal examination, Gleason sum, age, and race (Remzi et al., 2003). Another approach for decision support in prostate cancer is based on gene expression profiles (Singh et al., 2002). Regarding bladder cancer, a CDSS has been developed based on proteomic data (Parekattil, Fisher, & Kogan, 2003). Concerning breast cancer, the potential of microarray data has been analysed (Van’t Veer et al., 2002). Also, a recent CDSS has been developed that integrates data mining with clinical guidelines towards breast cancer decision support (Skevofilakas, Nikita, Templakesis, Birbas, Kaklamanos, & Bonatsos, 2005). It should be noted that all CDSSs mentioned above are just research attempts and only PAPNET is in clinical use.

Integration of Clinical and Genomic Data for Decision Support in Cancer

CLINICAL DECISION SUPPORT USING CLINICOGENOMIC PROFILES Methodology Conventional approaches for CDSS focus on a single outcome regarding their domain of application. A different approach is to generate profiles associating the input data (e.g., findings) with several different types of outcomes. These profiles include clinical and genomic data along with specific diagnosis, treatment and followup recommendations. The idea of profile-based CDSS is based on the fact that patients sharing similar findings are most likely to share the same diagnosis and should have the same treatment and follow-up; the higher this similarity is, the more probable this hypothesis holds. The profiles are created from an initial dataset including several patient cases using a clustering method. Health records of diagnosed and (successfully or unsuccessfully) treated patients, with clear follow-up description, are used to create the profiles. These profiles constitute the core of the CDSSs; each new case that is inserted is related with one (or more) of these profiles. More specifically, an individual health record containing only findings (and

maybe the diagnosis) is matched to the centroids. The matching centroids are examined in order to indicate potential diagnosis (the term diagnosis here refers mainly to the identification of cancer subtype). If the diagnosis is confirmed, genetic screening may be proposed to the subject and then, the clusters are further examined, in order to make a decision regarding the preferred treatment and follow-up. The above decision support idea is shown schematically in Figure 1.

General Description of the System Known approaches for the creation of CDSSs are based on the analysis of clinical data using machine learning techniques. This scheme can be expanded to include genomic information, as well. In order to extract a set of profiles, the integration of clinical and genomic data is first required. Then data analysis is realized in order to discover useful knowledge in the form of profiles. Several techniques and algorithms can be used for data analysis such as neural approaches, statistical analysis, data mining, clustering and others. Data analysis is a two stage procedure: (i) creation of an inference engine (training stage) and (ii) use of this engine for decision support. The type of

Figure 1. Decision support based on profiles extraction. Unknown features of patient are derived by known features of similar cases.

415

Integration of Clinical and Genomic Data for Decision Support in Cancer

analysis to be used greatly depends on the available information and the desired outcome. Clustering algorithms can be employed in order to extract patient clinico-genomic profiles. An initial set of records, including clinical and genomic data along with all diagnosis/treatment/follow-up information, must be available for the creation of the inference engine. The records are used for clustering and the centroids of the generated clusters constitute the profiles. These profiles are then used for decision support; new patients with similar clinical and genomic data are assigned to the same cluster, that is, they share the same profile. Thus, a probable diagnosis, treatment, and follow-up is selected. Both, the creation of the inference engine and the decision support procedure are presented in Figure 2.

Types of Data Clinical data that are contained in electronic patient records (i.e., demographic details, medical history, and laboratory data) are usually presented in a simple and structured format, thus simplifying the analysis. On the other hand, genomic data are not structured and, therefore, appropriate prepro-

cessing is needed in order to transform them into a more structured format. Three different kinds of biological data may be available to clinicians: (i) genomic data, often represented by a collection of single nucleotide polymorphisms (SNPs), DNA sequence variations that occur when a single nucleotide in the genome sequence is altered and since each individual has many SNPs, their sequence forms a unique DNA pattern for that person; (ii) gene expression data, which can be measured with DNA microarrays to obtain a snapshot of the activity of all genes in one tissue at a given time or with techniques that rely on a polymerase chain reaction (PCR) and real-time PCR when the expression of only a few genes needs to be measured with better precision; (iii) protein expression data, which can include a complete set of protein profiles obtained with mass spectra technologies, or a few protein markers which can be measured with ad hoc essays.

Data Processing Depending on the type of the available biological data, different preprocessing steps should be performed in order to derive structured biological

Figure 2. Representation of a general scheme for a CDSS, integrating clinical and genomic information

416

Integration of Clinical and Genomic Data for Decision Support in Cancer

information, while expert knowledge could favor the preprocessing steps. The processing stage is necessary in order to transform the genomic data into a more easy-to-analyse form, allowing their integration along with the clinical data into the data analysis stage. Also, the genomic data processing might take advantage of expert knowledge, that is, known genomic abnormalities. Finally, the integrated data (clinical and genomic) are analysed in order to discover useful knowledge that can be used for decision support purposes. This knowledge can be in the form of associations between clinical and genomic data, differential diagnosis, treatment, and so forth. The initial dataset (clinical or genomic) is defined by the experts and includes all features that according to their opinion are highly related with the domain at hand (clinical disease). After acquiring the integrated data, a feature selection technique is applied in order to reduce the number of features and remove irrelevant or redundant ones. Finally, the reduced set of features is used by a clustering algorithm. K-means (MacQueen, 1967), fuzzy k-means (Bezdek, 1981), and expectation maximization (Dempster, Laird, & Rubin, 1977) are known approaches for clustering and can be involved for profile extraction. The profiles are the output of the clustering procedure (centroids). A deficiency of several clustering algorithms is that the number of centroids (profiles) must be predefined; this is not always feasible. Thus, in order to fully automate the profile extraction process, a meta-analysis technique is employed, which automatically calculates the optimal number of profiles.

Application to Colon Cancer Colon cancer includes cancerous growths in the colon, rectum, and appendix. It is the third most common form of cancer and the second leading cause of death among cancers in the developed countries. There are many different factors involved in colon carcinogenesis. The associa-

tion of these factors represents the base of the diagnostic process performed by medics which can obtain a general clinical profile integrating patient information using his scientific knowledge. Available clinical parameters are stored together with genomic information for each patient to create (as much as possible) a complete electronic health record. Several clinical data that are contained in the electronic health records are related to colon cancer (Read & Kodner, 1999): age, diet, obesity, diabetes, physical inactivity, smoking, heavy alcohol consumption, previous colon cancer or other cancers, adenomatous polyps which are the small growths on the inner wall of the colon and rectum; in most cases, the colon polyp is benign (harmless). Also, other diseases or syndromes such as inflammatory bowel disease, ZollingerEllison syndrome, and Gardner’s syndrome are related to colon cancer. In the context of genomic data related with colon cancer, malignant changes of the large bowel epithelium are caused by mutations of specific genes among which we can differentiate (Houlston & Tomlinson, 1997): •

•

•

Protooncogenes. The most popular mutated protooncogenes in colon cancer are: K-RAS, HER-2, EGFR and c-MYC. Suppressor genes-anticogenes. In colorectal cancer the most important are DCC, TP53 and APC. Mutator genes. So far, six repair genes of incorrectly paired up bases were cloned from humans. Four are related to Hereditary Nonpolyposis Colon Cancer (HNPCC). These are: the hMSH2- homolog of yeast gene MutS, the hMLH1 - homolog of bacterial MutL, the hPMS1 and hPMS2 - from yeast equivalent.

An efficient way to process the above gene sequences is to detect Single Nucleotide Polymorphisms (SNPs) (Sielinski, 2005). SNPs data

417

Integration of Clinical and Genomic Data for Decision Support in Cancer

are qualitative data providing information about the genomic at a specific locus of a gene. An SNP is a point mutation present in at least 1% of a population. A point mutation is a substitution of one base pair or a deletion, which means the respective base pair is missing or an addition of one base pair. Though several different sequence variants may occur at each considered locus, usually one specific variant of the most common sequence is found, an exchange from adenine (A) to guanine (G), for instance. Thus, information is basically given in the form of categories denoting the combinations of base pairs for the two chromosomes, for example, A/A, A/G, G/G, if the most frequent variant is adenine and the

single nucleotide polymorphism is an exchange from adenine to guanine. According to previous medical knowledge, there are several SNPs with known relation to colon cancer. Some indicative SNPs already related to colon cancer, according to several sources in the literature, identified in TP53 gene, are presented in Table 1. The expert knowledge contains information about the position of the SNPs in the gene sequence (i.e., exon, codon position and amino acid position), the transition of the nucleotides and the translation of the mRNA to protein. Based on the list of known SNPs related to colon cancer, appropriate genomic information is derived, revealing the existence, or not, of these SNPs in the patient’s genes.

Table 1. Indicative SNPs transitions and positions in the TP53 gene, related with colon cancer mRNA pos.

Codon pos.

Amino acid pos.

Function

Transition

Protein residue transition

exon_10

1347

1

366

nonsynonymous

G/T

Ala [A]/Ser [S]

exon_10

1266

1

339

nonsynonymous

A/G

Lys [K]/Glu[E]

exon_9

1242

3

331

synonymous

A/G

Gln [Q]/Gln[Q]

exon_8

1095

1

282

nonsynonymous

T/C

Trp [W]/Arg[R]

exon_8

1083

1

278

nonsynonymous

G/C

Ala [A]/Pro[P]

exon_8

1069

2

273

nonsynonymous

A/G

His [H]/Arg[R]

Region

exon_7

1021

2

257

nonsynonymous

A/T

Gln [Q]/Leu[L]

exon_7

998

3

249

nonsynonymous

T/G

Ser [S]/Arg[R]

exon_7

994

2

248

nonsynonymous

A/G

Gln [Q]/Arg[R]

exon_7

984

1

245

nonsynonymous

A/G

Ser [S]/Gly[G]

exon_7

982

2

244

nonsynonymous

A/G

Asp [D]/Gly[G]

exon_7

973

2

241

nonsynonymous

T/C

Phe [F]/Gly[G]

exon_5

775

2

175

nonsynonymous

A/G

His [H]/Arg[R]

exon_5

702

1

151

nonsynonymous

A/T/C

Thr [T]/Ser[S]/Pro [P]

exon_5

663

1

138

nonsynonymous

C/G

Pro [P]/Ala [A]

exon_5

649

2

133

nonsynonymous

C/T

Thr [T]/Met [M]

exon_4

580

2

110

nonsynonymous

T/G

Leu [L]/Arg [R]

exon_4

466

2

72

nonsynonymous

G/C

Arg [R]/Pro [P]

exon_4

390

1

47

nonsynonymous

T/C

Ser [S]/Pro [P]

exon_4

359

3

36

synonymous

A/G

Pro [P]/Pro [P]

exon_4

353

3

34

synonymous

A/C

Pro [P]/Pro [P]

exon_2

314

3

21

synonymous

T/C

Asp [D]/Asp [D]

418

Integration of Clinical and Genomic Data for Decision Support in Cancer

Some of the described genes are acquired from the subjects and based on the SNP information regarding every acquired gene, such as SNPs in Table 1 for TP53 gene, new features are derived. These new features contain information regarding the existence or not of these SNPs in the patient’s gene sequence. The derived features along with the aforementioned clinical data that are related to colon cancer are the input to the methodology and, after following the above described inference engine creation methodology, clinicogenomic profiles are generated. These profiles are able to provide advanced cancer decision support to new patients.

FUTURE TRENDS There should be no doubt that several challenges remain regarding clinical and genomic data integration to facilitate clinical decision support. The opportunities of combining these two types of data are obvious, as they allow obtaining new insights concerning diagnosis, prognosis, and treatment. According to this, medical informatics are combined with bioinformatics towards biomedical informatics. Biomedical Informatics is the emerging discipline that aims to put these two worlds together so that the discovery and creation of novel diagnostic and therapeutic methods is fostered. A limitation of this combination is that although data exist, usually their enormous volume and their heterogeneity constitute their analysis and association a very difficult task. Another challenge is the lack of terminological and ontological compatibility, which could be solved by means of a uniformed representation. Besides new data models, ontologies are/have to be developed in order to link genomic and clinical data. Furthermore, standards are required to ensure interoperability between disparate data sources.

CONCLUSION Advances in genome technology are playing a growing role in medicine and health care. With the development of new technologies and opportunities for large-scale analysis of the genome, genomic data have a clear impact on medicine. Cancer prognostics and therapeutics are among the first major test cases for genomic medicine, given that all types of cancer are related with genomic instability. The integration of clinical and genomic data makes the prospect for developing personalized health care ever more realistic.

REFERENCES Alon, U., Barkai, N., Notterman, D. A., Gish, K., Ybarra, S., & Mack, D. (1999). Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Cell Biology, 96, 6745–6750. Bezdek, J. C. (1981). Pattern recognition with fuzzy objective function algorithms. Plenum Press, USA. Boon, M. E., & Kok, L. P. (2001). Using artificial neural networks to screen cervical smears: How new technology enhances health care. Clinical applications of artificial neural networks. Cambridge: Cambridge University Press (pp. 81-89). Chiang, I. J., Shieh, M. J., Hsu, J. Y., & Wong, J. M. (2005). Building a medical decision support system for colon polyp screening by using fuzzy classification trees. Applied Intelligence, 22(1), 61–75. doi:10.1023/B:APIN.0000047384.85823. f6 Cook, D. J., Lawrence, B. H., Su, S., Maglothin, R., & Jonyer, I. (2001). Structural mining of molecular biology data: A graph-based tool for discovering and analyzing biological patterns in structural databases. IEEE Engineering in Medicine and Biology, 4, 67–74. doi:10.1109/51.940050

419

Integration of Clinical and Genomic Data for Decision Support in Cancer

Dempster, A., Laird, N., & Rubin, D. (1977). Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society. Series B. Methodological, 39(1), 1–38. Doornewaard, H., van der Schouw, Y. T., van der Graaf, Y., Bos, A. B., Habbema, J. D., & van den Tweel, J. G. (1999). The diagnostic value of computerassisted primary cervical smear screening: A longitudinal cohort study. Modern Pathology, 12(11), 995–1000. Exarchos, T. P., Papaloukas, C., Lampros, C., & Fotiadis, D. I. (2006). Protein classification using sequential pattern mining. In Proceedings of the 28th Annual International Conference of the Engineering in Medicine and Biology Society, USA (pp. 5814-5817). Fotiadis, D. I., Goletsis, Y., Likas, A., & Papadopoulos, A. (2006). Clinical decision support systems. Encyclopedia of Biomedical Engineering, Wiley. Houlston, R. S., & Tomlinson, I. P. M. (1997). Genetic prognostic markers in colorectal cancer. Journal Clinical Pathology: Molecular Pathology, 50, 281–288. doi:10.1136/mp.50.6.281 Jiang, X. R., & Gruenwald, L. (2005). Microarray gene expression data association rules mining based on BSC-tree and FIS-tree. Data & Knowledge Engineering, 53(1), 3–29. doi:10.1016/ S0169-023X(04)00110-7 Louie, B., Mork, P., Martin-Sanchez, F., Halevy, A., & Tarczy-Hornoch, P. (2007). Data integration and genomic medicine. Journal of Biomedical Informatics, 40, 5–16. doi:10.1016/j.jbi.2006.02.007 MacQueen, J. B. (1967). Some methods for classification and analysis of multivariate observations. In Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability (Vol. 1, pp. 281-297).

420

Mitchell, T. (2006). Machine learning. Springer, McGraw-Hill Education (ISE Editions). Nieminen, P., Hakama, M., Viikki, M., Tarkkanen, J., & Anttila, A. (2003). Prospective and randomised public-health trial on neural networkassisted screening for cervical cancer in Finland: Results of the first year. International Journal of Cancer, 103(3), 422–426. doi:10.1002/ijc.10839 Parekattil, S. J., Fisher, H. A., & Kogan, B. A. (2003). Neural network using combined urine nuclear matrix protein-22, monocyte chemoattractant protein-1 and urinary intercellular adhesion molecule-1 to detect bladder cancer. The Journal of Urology, 169(3), 917–920. doi:10.1097/01. ju.0000051322.60266.06 Read, T. E., & Kodner, I. J. (1999). Colorectal cancer: risk factors and recommendations for early detection. American Family Physician, 59(11), 3083–3092. Remzi, M., Anagnostou, T., Ravery, V., Zlotta, A., Stephan, C., & Marberger, M. (2003). An artificial neural network to predict the outcome of repeat prostate biopsies. Urology, 62(3), 456–460. doi:10.1016/S0090-4295(03)00409-6 Shah, S., & Kusiak, A. (2007). Cancer gene search with data-mining and genetic algorithms. Computers in Biology and Medicine, 37(2), 251–261. doi:10.1016/j.compbiomed.2006.01.007 Sielinski, S. (2005). Similarity measures for clustering SNP and epidemiological data. Technical report, University of Dortmund. Singh, D., Febbo, P. G., Ross, K., Jackson, D. G., Manola, J., & Ladd, C. (2002). (2002). Gene expression correlates of clinical prostate cancer behavior. Cancer Cell, 1(2), 203–209. doi:10.1016/ S1535-6108(02)00030-2

Integration of Clinical and Genomic Data for Decision Support in Cancer

Skevofilakas, M. T., Nikita, K. S., & Templakesis, P. H. Birbas., K.N., Kaklamanos, I.G., & Bonatsos, G.N. (2005). A decision support system for breast cancer treatment based on data mining technologies and clinical practice guidelines. In Proceedings of the 27th Annual Conference of IEEE Engineering in Medicine and Biology (pp. 2429-2432), China. Slaymaker, M., Brady, M., Reddington, F., Simpson, A., Gavaghan, D., & Quirke, P. (2006). A prototype infrastructure for the secure aggregation of imaging and pathology data for colorectal cancer care. In Proceeding of IEEE Computer Based Medical Systems (pp. 63-68). Tan, P. N., Steinbach, M., & Kumar, V. (2005). Introduction to data mining. Addison Wesley. USA. Van‘T Veer, L.J., Dai, H., Van De Vijner, M.J., He, Y.D., Hart, A.A.M., et al. (2002). Gene expression profiling predicts clinical outcome of breast cancer. Nature, 415, 530–536. doi:10.1038/415530a Walker, P. R., Smith, B., Liu, Q. Y., Famili, A. F., Valdes, J. J., & Liu, Z. (2004). Data mining of gene expression changes in Alzheimer brain. Artificial Intelligence in Medicine, 31(2), 137–154. doi:10.1016/j.artmed.2004.01.008 Zaki, M. J. (2000). Sequence mining in categorical domains: Incorporating constraints. In Proceedings of the 9th international conference on information and knowledge management, USA (pp. 422-429).

KEY TERMS AND DEFINITIONS Cancer Staging: Knowledge of the extent of the cancer at the time of diagnosis. It is based on three components: the size or depth of penetration

of the tumour (T), the involvement of lymph nodes (N), and the presence or absence of metastases (M). Clinical Decision Support Systems (CDSS): Entities that intend to support clinical personnel in medical decision-making tasks. In more technical terms, CDSSs are active knowledge systems that use two or more items of patient data to generate case-specific advice. Cluster Analysis: The task of decomposing or partitioning a dataset into groups so that the points in one group are similar to each other and are as different as possible from the points in other groups. Data Integration: The problem of combining data residing at different sources and providing the user with a unified view of these data. This important problem emerges in several scientific domains, for example, combining results from different bioinformatics repositories. Data Mining: The analysis of observational datasets to find unsuspected relationships and to summarize the data in novel ways that are both understandable and useful to the data owner. Mutation: A change in the genetic material (usually DNA or RNA) of a living being. Mutations can happen for a lot of different reasons. They can happen because of errors during cell division, because of radiation, chemicals, and so forth. Single Nucleotide Polymorphism (SNP): A DNA sequence variation occurring when a single nucleotide—A, T, C, or G—in the genome differs between members of a species or between paired chromosomes in an individual. Tumour Suppressor Gene: A gene that reduces the probability that a cell in a multicellular organism will turn into a tumor cell. A mutation or deletion of such a gene will increase the probability of the formation of a tumor.

This work was previously published in Encyclopedia of Healthcare Information Systems, edited by Nilmini Wickramasinghe and Eliezer Geisler, pp. 768-776, copyright 2008 by Medical Information Science Publishing (an imprint of IGI Global).

421

422

Chapter 2.13

Module Finding Approaches for Protein Interaction Networks Tero Aittokallio University of Turku, Finland

ABSTRACT This chapter provides an overview of the computational approaches developed for exploring the modular organization of protein interaction networks. A special emphasis is placed on the module finding tools implemented in three freely available software packages, VisANT, Cytoscape and MATISSE, as well as on their biomedical applications. The selected methods are presented in the broader context of module discovery options, ranging from approaches that rely merely on topological properties of the underlying network to those that take into account also other compleDOI: 10.4018/978-1-60960-561-2.ch213

mentary data sources, such as the mRNA levels of the proteins. The author will also highlight some current limitations in the measured network data that should be understood when developing and applying module finding methodology, and discuss some key future trends and promising research directions with potential implications for clinical research.

INTRODUCTION Recent advances in experimental technologies and computational methods have made it possible to measure and predict protein-protein interactions on a global scale (see Chapters I-VI and

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Module Finding Approaches for Protein Interaction Networks

Shoemaker & Panchenko, 2007a, b). While the large-scale interaction datasets can provide an unprecedented glimpse into the cellular mechanisms underlying the behavior of various biological systems, the increasing sizes and densities of the protein interaction networks available today pose also many challenging computational problems. In particular, the inherent complexity of most biological processes and the large number of possible interactions involved can make it difficult to interpret and mine the interaction networks only by eye, even with the help of sophisticated visualization and layout tools available. Software packages that implement computational tools for more advanced network data mining can facilitate the explorative network analysis by identifying the key players and their interactions that contribute to the cellular processes of interest. This may allow e.g. to pinpoint errors in experimentally or computationally derived interaction links, identify proteins directly involved in the particular process, and to formulate hypotheses for follow-up experiments. It should be realized, however, that even if the rapidly developing complex network theory has successfully been applied to analysis of various social and technological networks, such as the Internet and computer chips, its impact on studying molecular interaction networks is still an emerging area of research and therefore all these methods should be considered as experimental. At the moment, the computational tools are best used together with network visualization and analysis software that enable interactive and fully-controlled mining of the complex protein interaction networks. One of the most fundamental properties found in many biological networks is their modular organization (Hartwell et al., 1999). Consequently, the decomposition of large networks into a hierarchy of possible overlapping sub-networks (so-called modules) has become a principal analytical approach to deal with the complexity of large cellular networks (Barabási & Oltvai, 2004). In protein interaction networks, a functional module refers

to a group of physically connected proteins that work together to carry out a specific cellular function in a particular spatio-temporal context. A large number of computational tools of increasing complexity have recently been developed for investigating the modular organization of interaction networks. These tools cannot only identify whether a given network is modular or not, but also detect the modules and their interrelationships in the underlying network. By relating the found sub-networks with complementary functional genomics or proteomics data, such as gene expression profiles from genome-wide microarray experiments or protein abundance measurements from mass-spectrometry-based assays, it is also possible to identify a hierarchy of connected groups of components that show coherent expression patterns. Such functionally organized modules cannot only emphasize the biological meaning of the modules discovered, but also allow us to gradually focus on the active subsystems of particular interest (active modules), which can lead to concrete hypotheses about the regulatory mechanisms and pathways most important for the given process (Ideker et al., 2002). Moreover, these modules can subsequently be used in predictive modeling studies, with the aim of suggesting new biological hypotheses, such as unexplored new interactions or the function of individual components, or even distinguishing different biomedical phenotypes (discriminative modules). While several computational approaches to discovering functional modules in protein interactions networks are available to molecular biologists, and new ones are constantly being developed by the methodologists, there is no general consensus on how to choose between the different strategies and definitions. The choice of the network mining approach is an important aspect which depends both on the network and questions under the analysis. Therefore, basic understanding of the graph-theoretic concepts and algorithms behind the software packages

423

Module Finding Approaches for Protein Interaction Networks

is required so that the specific questions can be addressed and biologically meaningful interpretations can be made. This chapter describes several computational strategies developed for investigating the modularity of protein interaction networks, together with their relative merits and potential pitfalls. Since this topic has witnessed great progress lately, only representative examples of the different approaches can be presented. Among the multitude of algorithms and methods available in the literature and Internet, I describe here those features and options available in three popular software tools, VisANT, Cytoscape and MATISSE, which support the discovery of functionally coherent network modules. These publicly available software were selected because they come with a convenient graphical user interface making the methods easily accessible also to the end-users without programming skills. A special emphasis is placed on four specific tools |E|=18implemented in these software, which represent typical computational strategies in the wide spectrum of module finding options, ranging from approaches that are based on topological network properties, such as nodes with a particular connectivity level (e.g. VisANT) or sub-networks that are highly connected (MCODE), to more sophisticated approaches that take into account also other related data sources, such as significant differences or co-expression patterns in the mRNA levels of the proteins (jActiveModules and JACS, respectively).

BACKGROUND AND MOTIVATION Many advanced network analysis software complement the visual network exploration with options that facilitate the characterization of topological and functional properties of the underlying network (comparison of the features available in the popular software tools can be found from Chapter XVI and Suderman & Hallett, 2007; Zhang et al., 2007; Cline et al., 2007). Typical

424

topological properties include, e.g., the degree distribution for determining hub nodes, clustering coefficient for detecting dense neighborhoods, and shortest paths for finding n-neighborhoods of selected nodes; see Figure 1 and Box 1 for the basic graph-theoretic concepts and their biological counterparts. Such measures, either alone or when compared to those of the whole network or its randomized counterpart, can be used to identify topologically coherent sub-networks, which may encode functional modules – i.e. groups of protein nodes and their interactions that can be attributed to a particular biological function (Hartwell et al., 1999). Several computational definitions of the topological properties of specific sub-networks, together with their biological interpretations, have been proposed during the past few years (Barabási & Oltvai, 2004), many of which – e.g. clusters, complexes and motifs – are described in detail in Chapters VIII and IX. In this chapter, the focus is on those network properties that have widely been recognized in many protein interaction networks: (1) modules are highly connected sub-networks (i.e. intra-module connectivity is higher than intermodule connectivity), and (2) protein interaction networks have hierarchical modular organization (modules are not isolated but they can interact and overlap at multiple levels). The first criterion reflects the fact that certain proteins are involved in common elementary biological functions, while the second implies that the same proteins can be active in multiple cellular processes at different points of time or under different conditions. In addition to such topological properties, requirement of functionally coherent modules often involves also some type of complementary information on the protein nodes or their interactions that can facilitate revealing the multi-level functional organization of the protein interaction networks. A useful source of complementary information on the activity of the protein nodes originates from the high-throughput transcriptomics and proteomics technologies. In particular, genomewide mRNA or protein expression measurements

Module Finding Approaches for Protein Interaction Networks

Figure 1. Basic graph-theoretic concepts. An example network graph G describing 18 binary interactions (edges) among 12 proteins (nodes), i.e. the cardinalities are |V|=12 and The neighborhood graph Na=(Va,Ea) of the node a is the set of nodes in G and their interactions that are directly connected to a; the left-hand circle in the figure identifies Na. By definition, the node a itself and its own connections are excluded from the sets Va and Ea, respectively. The degree d of a node is the number of its first neighbors, e.g. d(a)=|va|=5, making it a hub protein in the network. The path length between two nodes is the number of edges on a shortest path between them; for instance, node b belongs to the 4-neighborhood of a. This particular toy network is said to be connected, because there exists a path connecting each pair of the nodes. A sub-network is called a k-core if the minimal degree among its nodes is k. The density D of a graph G is the fraction of edges it has out of all possible node pairs; formally, D(G)=2|E|/|V|(|V|-1). The clustering coefficient c of a node defines the proportion of its first neighbors that are connected to each other; for instance, since there is only one edge connecting the neighbors of a, c(a)=D(Na)=2∙1/20. In other words, clustering coefficient quantifies how close the connectivity of the neighborhood of a node is to a sub-network in which every node is connected to every other node (so-called clique). For instance, the neighborhood of b in the right-hand circle forms a fully connected sub-network, and therefore c(b)=1.

using microarrays or mass-spectrometry, respectively, are increasingly being used to illuminate the molecular mechanisms involved in the control of cellular systems in various organisms. However, interpretation of the resulting lists of genes or proteins remains a labor-intensive and error-

prone task, because listing the individual elements alone can provide only a limited insight into the multitude of biological processes the individual elements participate in. To facilitate distinguishing elements directly involved in a particular process from bystander elements, whose expres-

Box 1. Network nomenclature and biological counterparts A graphG=(V,E) is a mathematical object that represents a protein interaction network. Conventionally, protein-protein interactions are modeled by undirected graphs G, in which V is a set of vertices (protein nodes), and E is a set of edges connecting pairs of nodes (binary interactions). The activity states of the proteins can be represented as node attributes (e.g. continuous weighting by the mRNA level of the gene encoding the particular protein under a specific condition), and the information how and when these biomolecules interact with each other as edge attributes (e.g. discrete labeling by the conditions under which the particular interaction is active). A module is a sub-network of G with specific properties. Typical topological properties include neighborhoods of highly connected genes (hubs), highly connected subnetworks (clusters), and highly-occuring sub-networks with particular connectivity patterns (motifs). Additionally, it can be required that the sub-networks show coherent expression changes (active modules), or allow distinguishing between distinct phenotypes (disriminative modules). In cell biology, such modules may correspond to functionally coherent groups of biomolecules, which act together in cellular mechanisms or processes, e.g., sub-units of protein complexes or biological pathways.

425

Module Finding Approaches for Protein Interaction Networks

sion have been altered by secondary effects or technical artifacts, the expression data can be mapped on a protein interaction networks to provide better understanding of the overall organization of the system in its entirety, rather than investigating individual genes, proteins or pathways separately. Moreover, instead of using the conventional pathway enrichment analysis among pre-defined gene/protein sets, which are limited to well-studied systems and known pathways only, identification of network modules that are active in the particular experiment is better targeted at finding novel interactions, cross-talks between the established pathways, and eventually the biological meaning of each individual component. Therefore, analyzing experimental data in the context of protein interaction networks can, in general, help to determine system-level cellular mechanisms explaining the experimental observations. Functional modules extracted from the human interaction networks, in particular, that are becoming available at an increasing rate, can provide systematic strategies to unravel the molecular basis of complex human diseases, and therefore enable identification of novel molecular markers for pharmaceutical and/or diagnostics developments for many disorders (Ideker & Sharan, 2008).

MODULE DISCOVERY ON THE BASIS OF NETWORK TOPOLOGY ALONE At the most general abstraction level, the protein interaction networks can be investigated and compared on the basis of their topology (i.e. the connection structure between the nodes). The representation of protein interaction networks as undirected graphs makes it possible to systematically investigate the structure-function relationships of these networks using well-understood graphtheoretical concepts and algorithms (Przulj et al., 2004). Global network measures, such as distribu-

426

tions of the node degrees, clustering coefficients, or path lengths, that can capture the overall network organization, have been proposed as topological correlates of the network evolution, function, stability or even dynamic responses (Yook et al., 2004; Albert, 2005). Alternatively, characterizing local interconnectivity patterns in terms of relative importance of individual nodes (centrality measures) or connecting paths (pathway redundancy), can provide more detailed functional information on the node neighborhoods and alternative pathways, and hence facilitate understanding of fundamental biological concepts, such as proteins’ essentiality and robustness of a particular system under internal or external perturbations (Estrada, 2006; Aittokallio & Schwikowski, 2006). Since the topological parameters can be rather sensitive to the chosen experimental setup and data analysis protocol (Hakes et al., 2008), it is essential to utilize these network measures in software packages, which allow multi-resolution network display and mining, to confirm that the properties discovered are not due to experimental or computational artifacts only.

Characterizing Network Connectivity Patterns Using VisANT VisANT software package is a visual analysis tool for biological networks and pathways, which provides a number of useful tools for both exploring and displaying the topological properties of a given protein interaction network (Hu et al., 2004; Hu et al., 2008). For instance, the user can easily display a scatter-plot or log-linear fit of the degree distribution of the connections in a separate window, which is directly linked to the main network view so that nodes with a particular level of connectivity can be readily identified. Such plot allows for distinguishing highlyconnected nodes (hubs) from weakly-connected nodes (leaves), whereas the degree exponent of the log-linear fit can be used to characterize the scale of the network (Barabási & Oltvai, 2004).

Module Finding Approaches for Protein Interaction Networks

Without entering to the debate whether or not protein interaction networks exhibit a scale-free topology, the degree distribution provides a quick summary of the overall connectivity structure and enables differentiation between distinct classes of network topologies (e.g. power-law type of distribution from a peaked one as seen in random network topologies). For identifying the modular organization of the network, VisANT provides similar plots for the distribution of the clustering coefficient over all of the nodes in the network, which can characterize the overall tendency of the nodes to form clusters or groups. Moreover, VisANT enables plotting of the average clustering coefficient of the nodes as a function of their degree k: C (k ) =

1 nk

∑ D(N

d (v )=k

v

),

where nk is the number of nodes with degree k. The function C(k) can be used to characterize the network’s hierarchical structure, and it allows the user to identify local neighborhood of nodes at a particular connectivity and density level. VisANT software also enables evaluation of shortest path lengths between nodes, which can help to further expand the selected modules, and the occurrence of sub-networks with a pre-defined connection structure (network motifs). One way to evaluate the modular organization of the network with VisANT is to exclude part of nodes according to their degree value. Interactive experimenting with increasing degree cut-off thresholds, coupled with dynamic navigation of the reduced networks, can reveal densely connected areas when sufficiently many low-degree regions have been removed. The downside of this rather crude multi-level filtering process is that it may also destroy the intermodule connections. Once the user has selected the node groups, based on the topology statistics or motif configurations, or imported existing groupings from external databases or from own

files, VisANT provides many options for displaying and manipulating these higher-level graph structures as meta-graphs, with the possibility to expand and collapse them iteratively to show reduced network views at multiple scales. Nested meta-nodes allow also one node to have multiple instances in the network and these instances are automatically tracked by the software.

Identifying Highly-Connected Sub-Networks Using MCODE Cytoscape software platform has a wide spectrum of tools and plug-ins for extracting the basic topological properties, such degree distributions, clustering coefficients, shortest paths, centrality statistics, and over-represented motifs, in addition to the advanced filtering and displaying options based on such attributes available in the core of the software package (Shannon et al., 2003; Cline et al., 2007). Cytoscape also includes specific plug-ins devoted for discovering modules – or putative complexes and pathways – in the imported networks. The molecular complex detection (MCODE) algorithm by Bader and Hogue (2003) was developed to enable an automated discovery of densely-connected sub-networks in large protein interaction networks. The requirement of high density originates from the observation that highly interconnected regions of protein interaction networks often correspond to known protein complexes and parts of biological pathways (Barabási & Oltvai, 2004). Like VisANT, MCODE considers local groups of nodes with a given connectivity and density level, but it can also expand these core modules automatically by recursively including nearby nodes that either can preserve or increase the sub-graph density. More formally, the operation of the module detection algorithm is organized through three stages. First, the node weighting is done on the basis of the density of the node’s neighborhood. Instead of using the standard clustering coefficient, the algorithm uses so-called core clustering coefficient of a node a,

427

Module Finding Approaches for Protein Interaction Networks

which is the density of the largest k-core Sa in its neighborhood (including also a itself): C k (a ) = D(Sa ),

where Sa = arg max {| S |: S ⊆ N a ∪ a, ∀v ∈ S : d (v ) ≥ k }

Depending on the chosen value of k, the definition of Ck(a) can emphasize the hub nodes, while giving smaller weights to the less-connected nodes. The final weight attached to the node a is the product of its core clustering coefficient Ck(a) and the size |Sa| of the highest k-core in its neighborhood. In the second stage, greedy optimization algorithm is used, which starts from the seed nodes with highest weights, and then recursively traverses their neighborhoods by including those nodes into the module whose weights deviate from the seed node’s weight by less than a user-defined cutoff value. This parameter controls the size and density of the resulting sub-networks. The third stage involves post-processing steps that can filter out or add nodes based on their connectivity. The output of the MCODE algorithm is a list of possible overlapping modules ranked according to their density and size values. The algorithm includes several useradjustable parameters, with different impacts on the results. The fast running time makes it possible to experiment with different parameter combinations, and their fine-tuning can be carried out according to existing biological knowledge by starting the optimization algorithm from known seed nodes. A further advantage over many traditional network clustering algorithms is that MCODE assigns to the modules only those nodes above a given local threshold density, whereas the rest of the nodes remain un-clustered (Figure 2).

MODULE DISCOVERY ON THE BASIS OF TOPOLOGY AND EXPRESSION One potential pitfall of the module finding approaches that rely solely on the topology of the networks is that false negative and positive inter-

428

actions can have severe effects on their results, regardless of whether one identifies protein hubs, clusters or network motifs. Since the current interaction networks are still limited both in accuracy and coverage, especially in more complex organisms like human, the topological structure alone may not always be sufficient for fully understanding of the complex cellular mechanisms and functions. To make the identification of functional modules more robust against experimental and modeling artifacts, a number of approaches have proposed the usage of protein interaction networks in conjunction with complementary information such as protein localization, expression or function (Tornow and Mewes, 2003; Lu et al., 2006; Lubovac et al., 2006; Cho et al., 2007). The presumption behind all such integrative module discovery approaches is that exploiting heterogeneous types of data sources in concert may be impacted less by the noise and biases associated with each individual data source if they were used separately. However, such integrative analysis also complicates the validation of the methods due to the overlap between complementary data sources and functional annotations, and thereby may increase the risk for circular reasoning. Here, I focus on those approaches only that combine global interaction networks with mRNA expression data, although any continuous type of state data on the nodes, such as protein abundance, could in principle be used instead. Most computational approaches that use both expression and network datasets are motivated by those early experimental studies which demonstrated that genes with similar expression profiles are more likely to encode interacting proteins (Ge et al. 2001). In particular, it has been shown that many stable protein complexes show significant co-expression in their mRNA levels (Jansen et al., 2002). Inspired by these observations, several computational studies have used genome-wide mRNA expression data to characterize and predict protein interactions (Tirosh & Barkai, 2005; Sprinzak et al., 2006; Shoemaker & Panchenko

Module Finding Approaches for Protein Interaction Networks

Figure 2. An example output from the MCODE. The selected sub-network corresponds to the 3rd highest ranked module in a sample interaction network of Cytoscape (galFiltered). The red intensity in the node color reflects its score, while the white nodes remain unscored and therefore unclustered. The shape of the nodes indicates their roles in the clustering process (seed node – square, core cluster member – circle and non-core cluster member – diamond). The default parameter values were used both in scoring and finding the core clusters. The Include Loops option controls whether or not also self-edges are used when calculating the neighborhood density. The Degree Cutoff parameter controls the minimum number of connections for a node to be scored. When the Haircut option is turned on, all the singly-connected nodes will be removed from the core cluster. The Fluff option enables expanding the cluster cores by one neighbor outwards. The Node Score Cutoff is the most influential parameter for the cluster size. New cluster members are added only if their node score deviates from the seed node’s score by less than the set cutoff percentage. The Size Threshold slider on the right allows the user interactively shrink or expand the core clusters by adjusting the threshold below or above its original value (^).The Node Attribute Enumerator provides a summary of the protein attributes and their frequency in the module (here the GO Biological Process annotation is selected). The K-Core parameter enables filtering out modules that do not contain an inter-connected sub-network of at least k degrees. The Max. Depth parameter controls how far from the seed node the search for new members can extend (virtually unlimited by default). For a more detailed and up-to-date parameter information, consult the MCODE website (http:www.baderlab.org/Software/MCODE) and the original article (Bader and Hogue, 2003).

2007b), or used these complementary data resources together to reveal dynamic organization of modularity in protein interaction networks (Han et al., 2004). Most network visualization software tools also allow mapping gene expression data on the global interaction networks to facilitate their interpretation and integrated analysis. For instance, VisANT 3.0 enables the user to select node groups based on expression ranges or profile

correlations, making it possible e.g. to discover new proteins with similar function to a given list of query proteins already known to have closely related function (Hu et al., 2007). Multiple experiments can be navigated using a sliding bar and the navigation process can be animated using color coding. However, VisANT currently lacks more automated module finding options based

429

Module Finding Approaches for Protein Interaction Networks

on both the networks connectivity and expression correlation structure.

Identifying Sub-Networks with Similar Expression Profiles Using MATISSE MATISSE software by Ulitsky & Shamir (2007) provides automated module detection tools that makes use of both the topology of interaction patterns as well as similarity sets of nodes inferred from expression data. In addition to several visualization and clustering options for network and expression datasets separately, it implements an efficient algorithm for identifying jointly active connected sub-networks (JACS) – i.e. connected sub-networks in the interaction data that show high internal similarity in their expression profiles (Ulitsky & Shamir, 2007). The algorithm finds such putative modules using statistical hypothesis testing framework, which evaluates the significance of the pairwise expression similarities Sij between all those node pairs (i, j) that belong to a connected sub-network U (candidate modules). Pearson correlation between the expression patterns of nodes i and j is used as a default Sij, but in principle any similarity measure could be used instead. More formally, let γijm = β Pi Pj be the probability that the nodes i and j are in the same group (so-called module hypothesis), where β is a parameter reflecting the general probability of any pair to group together and Pi is a prior probability describing how likely is it that the node i is transcriptionally regulated. Similarly, let γijn = pm Pi Pj be the probability that the nodes i and j are in the same group under random selection of pairs (the null hypothesis), where pm is the fraction of node pairs expected at random. The putative module U is then evaluated using the log-likelihood ratio score: L(U ) =

∏

i , j ∈U

430

γijm P (Sij | M ij ) + (1 − γijm )P (Sij | N ij ) γijn P (Sij | M ij ) + (1 − γinj )P (S ij | N ij )

This formula stems from the assumption that the pairwise similarities Sij are conditionally independent, and that they originate from a mixture of two Gaussian distributions, in which the two normal components P(Sij|Mij) and P(Sij|Nij) correspond to the node pairs (i,j) that have high or low expression similarity, respectively. These distributions, together with the prior probabilities under the two rivaling hypotheses, are automatically learned with the expectation-maximization algorithm. Using this scoring function, MATISSE identifies modules through a heuristic multi-stage optimization procedure. First, relatively small, high-scoring, connected node sets are detected as seeds. Second, a greedy iterative algorithm is used to optimize all the seeds simultaneously by considering the following four operations: (i) adding an unassigned node to a module, (ii) removing a node from a module, (iii) exchanging a node assignment between two modules, and (iv) merging two modules by taking their union. An operation is accepted if it improves the overall score calculated over all the modules and also preserves the connectivity of the modules. In the last stage, significance filtering of the resulting modules is performed by sampling random node sets of the same size and comparing their scores to those of the detected ones. While it assumes no prior knowledge on the number of modules (size constraints can be introduced), the optimization procedure forces the modules to be non-overlapping, which may hinder the assignment of proteins whose expression levels are a result of multiple cellular functions reflected in the experiment.

Identifying Sub-Networks with Coherent Expression Changes Using jActiveModules The usage of pairwise similarities, like Pearson correlation, in the identification of functional modules makes it possible to use various types of expression dataset, e.g., time-series or steady-state experiments. This comes with the potential limita-

Module Finding Approaches for Protein Interaction Networks

tion that such overall measures of co-expression may ignore the fact that a protein can be active in a particular process over a subset of experimentally measured conditions or time-points only – a phenomenon termed just-in-time assembly (De Lichtenberg et al., 2005). Experimental studies have indeed demonstrated that some components of protein complexes may show coherent expression changes under a few conditions only and those components often correspond to known sub-complexes with specific functions (Simonis et al. 2006). This motivates those computational approaches that search for connected sub-networks whose nodes show differential expression over a particular subset of conditions only. One such method by Ideker et al. (2002) is implemented in the Cytoscape software as a plug-in called jActiveModules. The algorithm requires as an input the condition-specific pa,c-values representing the significance of differential expression of a node a under each condition c. The method first converts the p-values to z-scores, using the inverse normal cumulative distribution function, that is, za,c=P-1(1-pa,c), and then derives an aggregate score Sc(A) for a sub-network A under the condition c by averaging the z-scores over the nodes in A. Using the simplifying assumption that these conditionspecific scores are independent, the significance of the jth highest scoring condition Sj(A) can be computed using the binomial order statistic: m   h m −h m p j (A) = ∑   1 − P (S j (A)) P (S j (A)) .      h  h=j 

The pj(A) -value gives the probability that at least j of the m conditions have scores above Sj(A), which reflects the activity of the sub-network A under conditions ranked 1 through j. After adjusting the scores for the rank and correcting for the background score distribution by sampling randomly node sets of similar size, independently of the underlying interactions, the scoring function can be use to rank both the sub-networks and

the subsets of conditions. For identifying highscoring connected sub-networks, jActiveModules enables using a stochastic optimization heuristic based on simulated annealing (Kirkpatrick et al., 1983). Briefly, in each iteration, a random node of the network is either removed or added to the module and the score for the resulting new subgraph is calculated. If the new score is larger than the previous one, then the move is accepted; otherwise the move is accepted with a probability proportional to the current temperature, which decreases geometrically with the iteration number. Finally, the temperature is set to zero, meaning that the optimization becomes greedy, until the algorithm reaches the closest local maximum. The algorithm can be adjusted by allowing the user to define the number of modules searched simultaneously and by reducing the effect of hub nodes on the search results (Ideker et al. 2002). To output smaller modules for visualization, one can repeat the search within each original sub-network using the local search option (see Figure 3). Several modifications to the original scoring function and search algorithm have been developed since then (Sohler et al., 2004; Rajagopalan & Agarwal, 2005; Cabusora et al., 2005; Nacu 2007; Guo et al., 2007; Dittrich et al. 2008).Figure 3. Network modules involved in HIV-1 reactivation. An example of integrated analysis using both gene expression profiling of human T-cell lines infected with human immunodeficiency virus (HIV), as well as curated human-human and human-HIV protein interactions from HPRD and HIV-1 interaction database, respectively. The active sub-networks during different stages of the process were identified using a local search, with parameters depth=2 and max depth=2, in the jActiveModules plugin to Cytoscape software (Ideker et al. 2002). The significance of differential expression at different activation stages was assessed using p-values from the t-test. The color of the nodes corresponds to the mean expression change and the shape indicates the species (HIV – diamond and human - circle). The sub-network shown in

431

Module Finding Approaches for Protein Interaction Networks

the inset was active during the early stages of the reactivation. This connected sub-network included five hub-proteins (Rb1, Polr2a, Jun, HSPca and Pten), each of which had significant numbers of differentially expressed neighbors (p 2, form a new class and take the average of all and store. STEP 7. Take the new reference from the remaining segments and go to step (ii) and determine the Euclidean distance in a similar manner; if it satisfies the threshold criteria, then MUAP class is determined accordingly.

For each matching point, calculate the normalized Euclidean distance (Nd), with the reference waveform.

It is observed that db-2 wavelets has proved to be the best suitable for MUAP separation. There is one advantage of this technique as the classification procedure is based on spectrum matching. The spectrum matching technique is considered to be more effective than the waveform matching technique in the time domain, especially when the interference is induced by baseline drift or by high-frequency noise. In both the cases, interference affects two small portions of the frequency spectrum.

PERFORMANCE COMPARISION AND DISCUSSIONS

DECOMPOSITION OF SUPERIMPOSED WAVEFORMS It was found after classification phase that superimposed waveform were also present in the segmented waveform. Therefore it is necessary to separate out these MUAPs from the superimposed waveform. The following steps are implemented for the decomposition of superimposed MUAP waveforms:

Nd =

2

M

∑ (x i=1

i

M

− yi ) / ∑ yi Nd< Th = 0.2 i =1

The general expected accuracy of the decomposition is dependent on the quality of data. The quality of the data is determined by its signal to noise ratio, stability, and actual separation of the motor unit class templates. If any one of these aspects of the data is degraded, accuracy of the process suffers. When there is increased recruitment or noise, one must acquire a large number of MUAP discharges for all above methods except wavelet technique and to identify a modest number of similar templates. Statistical pattern recognition and crosscorrelation techniques require recordings of the EMG signal that are more free of noise. Kohonen self-organizing feature technique is a fast method for extracting MUAPs that requires recordings of the EMG signal relatively free of noise. Overall Wavelet technique has the best performance. This study demonstrates the versatility of Kohonen technique. When the recordings are relatively free of background activity, the decomposition of EMG

473

Techniques for Decomposition of EMG Signals

signal by this method are similar to those obtained by the statistical pattern recognition and crosscorrelation techniques. When background activity is increased and statistical pattern recognition and cross-correlation techniques fail, decomposition of EMG signal are similar to Wavelet technique. Cross-correlation technique examines all possible combinations of templates and their waveforms shifts in each possible time location (exhaustive search scheme performed in the time domain) to determine the best template match. Obviously, any exhaustive search technique is usually time consuming and is normally applicable only when the number of available templates is small. The final task in MUAP classification is the selection of reference template. Identification of MUAP templates mainly relies on the number of MUAPs present in one set of data, which indicates the number of repetitions of a MUAP. Therefore, a group containing more MUAPs has a higher chance to be selected as a reference template. Training effort for the wavelet technique and Kohonen self-organizing feature map were significantly less as compared to the cross-correlation and statistical pattern recognition techniques. Decomposition performances of correlation approach and statistical technique were found to be similar. Statistical technique has the disadvantages of using a constant threshold for classification that make it less flexible, especially when the signal is noisy and with high variability. It is observed that number of classes which can be detected by the statistical technique is unlimited depending only on the number of actual MUAP classes existing in the signal. The Kohonen-NN performed better than the statistical pattern recognition technique and yielded a higher success rate. Wavelet spectrum of a spike describes not only frequency components of the spike but also their time locations. Thus, waveform matching in the wavelet preserves the best properties of matched filter techniques in both the time and frequency domains. The excitability and variability of MUAPs are easily evaluated with the help of Kohonen map and wavelet based

474

techniques. The success-rate of classification for NOR, MYO and MND cases is found to be highest using wavelet technique. When decomposition and clustering were performed on the acquired myoelectric signal, it was found that the number of action potentials were missing due to unresolved superpositions which occur at times when many neurons fire almost simultaneously. When the myoelectric signal is recorded, it is initially decomposed into individual action potentials. The decomposed signals are then clustered, using physical characteristics, into classes representing single motor neuron sources and the lumping of more than one neuron’s firings into a single class. The clustering step, however, is never fully successful due to unresolved signal superpositions and, less frequently, missed detection by the electrode. Number of action potential firings are therefore missed resulting in an incomplete data set. This often manifests itself in the data as an inter pulse interval (IPI) that is too large. It was also observed that in statistical pattern recognition and cross-correlation techniques, due to waveform variability, the MUAP classes coming from the same motor unit, although looked similar, were not grouped together. Merging of these classes can be achieved by using higher constant threshold. It is seen visually from the MUAPs of NOR, MYO and MND cases that there is a wide variation in some of the clinically important parameters. The duration measure is the key parameter being used in the clinical practice. Myopathy patients usually have MUAPs with short durations, low amplitudes and a small number of phases; whereas the MND patients have MUAPs with long durations, high amplitudes and a large number of phases. Sometimes waveform are more complex and have as many as six-phase reversal even in normal muscles. In complex EMG signals, where more active motor units were present, the physician was not able to confidently classify MUAPs based on a visual analysis of the data. The physician could only suggest how many motor units were active in a signal. The decomposition techniques

Techniques for Decomposition of EMG Signals

were able to confidently classify motor units in a complex EMG signal. The decision of reference MUAP is very important, because sometimes, it happens that when we are taking any one MUAP as a reference and find the members of the class and second time when we are taking the reference MUAP from this obtained class then the number of members change for the same class. This is due to location of the peak in the given MUAPs. In all above techniques, which are fully automatic, the data processing permits analysis without any loss or distortion of information.

CONCLUSION The following major conclusions are drawn on the basis of present work: 1. The performance of decomposition by Cross-Correlation approach is comparable to statistical pattern recognition technique.

The main problem of both the techniques is the large amount of computational time to process all the possible combinations of time shifted templates by taking different template as the reference. 2. The Kohonen-NN require smaller architecture and thus less computational effort and is found more appropriate for the classification of MUAPs because of their ability to adapt and to create complex classification boundaries. When Cross-Correlation approach is combined with the Kohonen-NN, then it results in an integrated approach for decomposition, which is fast, efficient, accurate and requires less computational time. 3. Wavelet technique is easily implementable and has better performance in respect of speed and accuracy. In the wavelet technique, it is observed that in all above five decompositions 4. 3rd , 4th and 5th decompositions show significant difference for MUAPs classification.

Figure 2. Performance comparison of techniques

475

Techniques for Decomposition of EMG Signals

5. If number of templates is small, then statistical pattern recognition technique and Cross-Correlation approach are good; and if number of templates is high, then KohonenNN and wavelet decomposition technique are very good. 6. Wavelet-based analysis provides the MUAP morphology in the time-frequency plane. It allows fast extraction of localized frequency components and there is no need of correction for baseline drift or high-frequency noise. Therefore, wavelet technique has proved to be the best for decomposition and classification of MUAPs. The outcome of this work will be very useful for diagnosing the neuromuscular disorders in planning aspects of curative and rehabilitation therapy and for evaluating response to treatment.

REFERENCES Chauvet, E., Fokapu, O., Hogrel, J.-Y., Gamet, D., & Duchene, J. (2003, November). Automatic Identification of Motor Unit Action Potential Trains from Electromyographic signals Using Fuzzy Techniques. Journal of Medical and Biological Engineering and Computing. SpringerLink, 41(6), 646–653. Christodoulos, I., & Pattichis, C. S. (1999, February). Unsupervided pattern recognition for the classification of EMG signals. IEEE Trans. on BME, 46(2), 169–178. doi:10.1109/10.740879 De Luca, C. J., Alexander, A., Wotiz, R., Gilmore, L. D., & Nawab, S. H. (2006). Decomposition of Surface EMG Signal. Journal of Neurophysiology, 96, 1646–1657. doi:10.1152/jn.00009.2006 Fever, R. S., & Luca, C. J. (1982, March). A procedure for decomposing the myoelectric signal into its constituent action potentials-part 1: Technique, Theory and Implementation. IEEE Trans. on BME, 29(3), 149–157. doi:10.1109/TBME.1982.324881 476

Fever, R. S., Xenakis, A. P., & Luca, C. J. (1982, March). A procedure for decomposing the myoelectric signal into its constituent action potentialspart 2: Execution and Test for Accuracy. IEEE Trans. on BME, 29(3), 158–164. doi:10.1109/ TBME.1982.324882 Gerber, A., Studer, R. M., Figueiredo, R. J. P., & Moschytz, G. S. (1984, December). A new framework and computer program for quantitative EMG signal analysis. IEEE Trans. on BME, 31(12), 857–863. doi:10.1109/TBME.1984.325248 Gill, K. C. M., Cummins, K. L., & Dorfman, L. J. (1985, July). Automatic decomposition of the clinical Electromyogram. IEEE Trans. on BME, 32(7), 470–477. doi:10.1109/TBME.1985.325562 Holobar, A., & Zazula, D. (2004). Correlationbased decomposition of surface electromyograms at low contraction forces. Journal of Medical and Biological Engineering and Computing. SpringerLink, 42(4), 487–495. Khandpur, R. S. (1987). Hand book of biomedical instrumentation. New York: McGraw-Hill Publishing Company Limited. Kleine, B. J., vanDijk, B., Lapatki, M., & Zwarts, D., Stegeman. (2007). Using two-dimensional spatial information in decomposition of surface EMG signals. Journal of Electromyography and Kinesiology, Elsevier, 17(5), 535–548. doi:10.1016/j. jelekin.2006.05.003 Loudon, G. H., Jone, N. B., & Sehmi, A. S. (1992, November). New signal processing techniques for the decomposition of EMG signals. Medical & Biological Engineering & Computing, 30(6), 591–599. doi:10.1007/BF02446790 Mallet, Y., Coomans, D., Kautsky, J., & Olivier, D. V. (1997, October). Classification using adaptive wavelets for feature extraction. IEEE Trans. on PAMI, 19(10), 1058–1066.

Techniques for Decomposition of EMG Signals

Nandedkar, S. D., & Sanders, D. B. (1989, November). Median averaging of electromyographic motor unit action potentials: comparison with other techniques. Medical & Biological Engineering & Computing, 27(6), 566–572. doi:10.1007/ BF02441637

Stashuk, D., & Bruin, H. D. (1988, January). Automatic decomposition of selective needle detected myoelectric signals. IEEE Trans. on BME, 35(1), 1–10. doi:10.1109/10.1330

This work was previously published in Biomedical Engineering and Information Systems: Technologies, Tools and Applications, edited by Anupam Shukla and Ritu Tiwari, pp. 187-197, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

477

478

Chapter 2.17

Prototype Based Classification in Bioinformatics Frank-M. Schleif University of Leipzig, Germany Thomas Villmann University of Leipzig, Germany Barbara Hammer Technical University of Clausthal, Germany

INTRODUCTION Bioinformatics has become an important tool to support clinical and biological research and the analysis of functional data, is a common task in bioinformatics (Schleif, 2006). Gene analysis in form of micro array analysis (Schena, 1995) and protein analysis (Twyman, 2004) are the most important fields leading to multiple sub omicsdisciplines like pharmacogenomics, glycoproteomics or metabolomics. Measurements of such studies are high dimensional functional data with few samples for specific problems (Pusch, 2005). This leads to new challenges in the data analysis. DOI: 10.4018/978-1-60960-561-2.ch217

Spectra of mass spectrometric measurements are such functional data requiring an appropriate analysis (Schleif, 2006). Here we focus on the determination of classification models for such data. In general, the spectra are transformed into a vector space followed by training a classifier (Haykin, 1999). Hereby the functional nature of the data is typically lost. We present a method which takes this specific data aspects into account. A wavelet encoding (Mallat, 1999) is applied onto the spectral data leading to a compact functional representation. Subsequently the Supervised Neural Gas classifier (Hammer, 2005) is applied, capable to handle functional metrics as introduced by Lee & Verleysen (Lee, 2005). This allows the classifier to utilize the functional

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Prototype Based Classification in Bioinformatics

nature of the data in the modelling process. The presented method is applied to clinical proteome data showing good results and can be used as a bioinformatics method for biomarker discovery.

BACKGROUND Applications of mass spectrometry (ms) in clinical proteomics have gained tremendous visibility in the scientific and clinical community (Villanueva, 2004) (Ketterlinus, 2005). One major objective is the search for potential classification models for cancer studies, with strong requirements for validated signal patterns (Ransohoff, 2005). Primal optimistic results as given in (Petricoin, 2002) are now considered more carefully, because the complexity of the task of biomarker discovery and an appropriate data processing has been observed to be more challenging than expected (Ransohoff, 2005). Consequently the main recent work in this field is focusing on optimization and standardisation. This includes the biochemical part (e.g. Baumann, 2005), the measurement (Orchard, 2003) and the subsequently data analysis (Morris, 2005)(Schleif 2006).

PROTOTYPE BASED ANALYSIS IN CLINICAL PROTEOMICS Here we focus on classification models. A powerful tool to achieve such models with high generalization abilities is available with the prototype based Supervised Neural Gas algorithm (SNG) (Villmann, 2002). Like all nearest prototype classifier algorithms, SNG heavily relies on the data metric d, usually the standard Euclidean metric. For high-dimensional data as they occur in proteomic patterns, this choice is not adequate due to two reasons: first, the functional nature of the data should be kept as far as possible. Second the noise present in the data set accumulates and likely disrupts the classification when taking a

standard Euclidean approach. A functional representation of the data with respect to the used metric and a weighting or pruning of especially (priory not known) irrelevant function parts of the inputs, would be desirable. We focus on a functional distance measure as recently proposed in (Lee, 2005) referred as functional metric. Additionally a feature selection is applied based on a statistical pre-analysis of the data. Hereby a discriminative data representation is necessary. The extraction of such discriminant features is crucial for spectral data and typically done by a parametric peak picking procedure (Schleif, 2006). This peak picking is often spot of criticism, because peaks may be insufficiently detected and the functional nature of the data is partially lost. To avoid these difficulties we focus on a wavelet encoding. The obtained wavelet coefficients are sufficient to reconstruct the signal, still containing all relevant information of the spectra, but are typically more complex and hence a robust data analysis approach is needed. The paper is structured as follows: first the bioinformatics methods are presented. Subsequently the clinical data are described and the introduced methods are applied in the analysis of the proteome spectra. The introduced method aims on a replacement of the classical three step procedure of denoising, peak picking and feature extraction by means of a compact wavelet encoding which gives a more natural representation of the signal.

BIOINFORMATIC METHODS The classification of mass spectra involves in general the two steps peak picking to locate and quantify positions of peaks and feature extraction from the obtained peak list. In the first step a number of procedures as baseline correction, denoising, noise estimation and normalization are applied in advance. Upon these prepared spectra the peaks have to be identified by scanning all local maxima. The procedure of baseline cor-

479

Prototype Based Classification in Bioinformatics

rection and recalibration (alignment) of multiple spectra is standard, and has been done here using ClinProTools (Ketterlinus, 2006). As an alternative we propose a feature extraction procedure preserving all (potentially small) peaks containing relevant information by use of the discrete wavelet transformation (DWT). The DWT has been done using the Matlab Wavelet-Toolbox (see http://www.mathworks.com). Due to the local analysis property of wavelet analysis the features can still be related back to original mass position in the spectral data which is essential for further biomarker analysis. For feature selection the Kolmogorov-Smirnoff test (KS-test) (Sachs, 2003) has been applied. The test was used to identify features which show a significant (p < 0.01) discrimination between the two groups (cancer, control). In (Waagen, 2003) also a generalization to a multiclass experiment is given. The now reduced data set has been further processed by SNG to obtain a classification model with a small ranked set of features. The whole procedure has been cross-validated in a 10-fold cross validation.

WAVELET TRANSFORMATION IN MASS SPECTROMETRY Wavelets have been developed as powerful tools (Rieder, 1998) used for noise removal and data compression. The discrete version of the continuous wavelet transform leads to the concept of a multi-resolution analysis (MRA). This allows a fast and stable wavelet analysis and synthesis. The

analysis becomes more precise if the wavelet shape is adapted to the signal to be analyzed. For this reason one can apply the so called bi-orthogonal wavelet transform (Cohen, 1992), which uses two pairs of scaling and wavelet functions. One is for the decomposition/analysis and the other one for reconstruction/synthesis, giving a higher degree of freedom for the shape of the scaling and wavelet function. In our analysis such a smooth synthesis pair was chosen. It can be expected that a signal in the time domain can be represented by a small number of a relatively large set of coefficients from the wavelet domain. The spectra are reconstructed in dependence of a certain approximation level L of the MRA. The denoised spectrum looks similar to the reconstruction as depicted in Figure 1. One obtains approximation- and detail-coefficients (Cohen, 1992). The approximation coefficients describe a generalized peak list, encoding primal spectral information. For linear MALDITOF spectra a device resolution of 500−800Da can be expected. This implies limits to the minimal peak width in the spectrum and hence, the reconstruction level of the Wavelet-Analysis should be able to model corresponding peaks. A level L = 4 is appropriate for our problem (see Figure 1). Applying this procedure including the KS-test on the spectra with an initial number of 22306 measurement points per spectrum one obtains 602 wavelet coefficients used as representative features per spectrum, still allowing a reliable functional representation of the data. The coefficients were used to reconstruct the spectra and the final functional representation of the signal.

Figure 1. Wavelet reconstruction of the spectra with L = 4, 5, x-mass positions, y-arbitrary unit. Original signal - solid line. One observes for L = 5 (right plot) the peak approximate is to rough.

480

Prototype Based Classification in Bioinformatics

PROTOTYPE CLASSIFIERS Supervised Neural Gas (SNG) is considered as a representative for prototype based classification approaches as introduced by Kohonen (Kohonen, 1995). Different prototype classifiers have been proposed so far (Kohonen, 1995) (Sato, 1996) (Hammer, 2005) (Villmann, 2002) as improvements of the original approach. The SNG has been introduced in (Villmann, 2002) and combines ideas from the Neural Gas algorithm (NG) introduced in (Martinetz, 1993) with the Generalized learning vector quantizer (GLVQ) as given in (Sato, 1996). Subsequently we give some basic notations and remarks to the integration of alternative metrics into Supervised Neural Gas (SNG). Details on SNG including convergence proofs can be found in (Villmann, 2002). Let us first clarify some notations: Let cvin L be the label of input v, L a set of labels (classes). Let V in RDV be a finite set of inputs v. LVQ uses a fixed number of prototypes (weight vectors, codebook vectors) for each class. Let W={wr} be the set of all codebook vectors and cr be the class label of wr. Furthermore, let Wc={wr|cr = c} be the subset of prototypes assigned to class c in L. The task of vector quantization is realized by the map Ψ as a winner-take-all rule, i.e. a stimulus vector vin V is mapped onto that prototype s the pointer ws of which is closest to the presented stimulus vector v, measured by a distance dλ (v,w). dλ (v,w) is an arbitrary differentiable similarity measure which may depend on a parameter vector λ. For the moment we take λ as fixed. The neuron s (v) is called winner or best matching unit. If the class information of the weight vector is used, the above scheme generates decision boundaries for classes (details in (Villmann, 2002)). A training algorithm should adapt the prototypes such that for each class c in L, the corresponding codebook vectors Wc represent the class as accurately as possible. Detailed equations and cost function for SNG are given in (Villmann, 2002). Here it is sufficient to

keep in mind that in the cost function of SNG the distance measure can be replaced by an arbitrary (differentiable) similarity measure, which finally leads to new update formulas for the gradient descent based prototype updates. Incorporation of a functional metric to SNG As pointed out before, the similarity measure dλ (v,w) is only required to be differentiable with respect to λ and w. The triangle inequality has not to be fulfilled necessarily (Hammer, 2005). This leads to a great freedom in the choice of suitable measures and allows the usage of non-standard metrics in a natural way. For spectral data, a functional metric would be more appropriate as given in (Lee, 2005). The obtained derivations can be plugged into the SNG equations leading to SNG with a functional metric, whereby the data are functions represented by vectors and, hence, the vector dimensions are spatially correlated. Common vector processing does not take this spatial order of the coordinates into account. As a consequence, the functional aspect of spectral data is lost. For proteome spectra the order of signal features (peaks) is due to the nature of the underlying biological samples and the measurement procedure. The masses of measured chemical compounds are given ascending and peaks encoding chemical structures with a higher mass follow chemical structures with lower masses. In addition, multiple peaks with different masses may encode parts of the same chemical structure and, hence, are correlated. Lee proposed an appropriate norm with a constant sampling period τ: 1

D p L (v ) = ∑ (Ak −1 (v) + Ak +1 (v))p    fc p

k =1

with τ    2 | vk | Ak (v) =   τ vk2     2 |vk |+|vk −1 |

 τ | v | 2 k Bk (v)=   τ vk2  if 0 > vk vk − 1    2 |vk |+|vk +1 | if 0 ≤ vk vk − 1

if 0 ≤ vk vk + 1 if 0 > vk vk + 1

481

Prototype Based Classification in Bioinformatics

are respectively of the triangles on the left and right sides of xi. Just as for Lp, the value of p is assumed to be a positive integer. At the left and right ends of the sequence, x0 and xD are assumed to be equal to zero. The derivatives for the functional metric taking p = 2 are given in (Lee, 2005). Now we consider the scaled functional norm where each dimension (0, 1], vi is scaled by a parameter λi > 0 and all λi sum up to 1: 1

D p L (λv) = ∑ (Ak −1(λv) + Ak +1(λv))p    fc p

k =1

with τ  τ λ | v | if 0 ≤ v v + 1   2 k k k k  2 λk | vk | if 0 ≤ vk vk − 1 Ak (λ v) =  τ Bk (λ v)=  τ λk2vk2 λk2vk2   e lse else    2 λk |vk |+λk +1 |vk +1 |  2 λk |vk |+λk −1 |vk −1

The prototype update changes to: ∂δ22 (x,y, λ) ∂xk

=

τ2 (2 −U k −1 −U k +1 )(Vk −1 + Vk +1 )∆k 2

cancer, 50 control samples). Sample preparation and profile spectra analysis were carried out using the CLINPROT system (Bruker Daltonik, Bremen, Germany [BDAL]). The preprocessed set of spectra and the corresponding wavelet coefficients are then analyzed using the SNG extended by a functional metric. We reconstructed the spectra based upon the discriminative wavelet coefficients determined by the Kolmogorov-Smirnoff test as explained above and used corresponding intensities as features. We used all features for the parameterized functional norm i.e. all λi = 1. The original signal with approx. 22000 sampling points had been processed with only 600 remaining points still encoding the significant parts of the signal relevant for discrimination between the classes. The SNG classifier with functional metric obtains a crossvalidation accuracy of 84% using functional metric and 82% by use of standard Euclidean metric. The results from the wavelet processed spectra are slightly better than using standard peak lists, with 81% crossvalidation accuracy.

FUTURE TRENDS with 0 if 0 ≤ ∆ ∆   k k +1  0 if 0 ≤ ∆k ∆k −1 2 2 U k −1 =  , U =   λk +1∆k +1 λk −1∆k −1 k +1   else  λk |∆k |+λk +1 |∆k +1 |   λk |∆k |+λk −1 |∆k −1 |   1λ if 0 ≤ ∆ ∆  1λk if 0 ≤ ∆k ∆k −1 k k +1  k  Vk −1 =  ,Vk +1 =   λk |∆k | λk |∆k |   else else   λk |∆k |+λk +1 |∆k +1 |   λk |∆k |+λk −1|∆k −1 |

(

)

(

)

And ∆k=xk-yk using this parameterization one can emphasize/neglect different parts of the function for classification.

ANALYSIS OF PROTEOMIC DATA The proposed data processing scheme is applied to clinical ms spectra taken from a cancer study (45

482

The proposed method generates a compact but still complex functional representation of the spectral data. While the bior3.7 wavelet gives promising results they are still not optimal, due to signal oscillations, leading to negative intensities in the reconstruction. Further, the functional nature of the data motivates the usage of a functional data representation and similarity calculation but there are also spectra regions encoded which do not contain meaningful biological information but measurement artefacts. In principle it should be possible to remove this overlaying artificial function from the real signal. Further it could be interesting to incorporate additional knowledge about the peak width, which is increasing over the mass axis.

Prototype Based Classification in Bioinformatics

CONCLUSION The presented interpretation of proteome data demonstrate that the functional analysis and model generation using SNG with functional metric in combination with a wavelet based data pre-processing provides an easy and efficient detection of classification models. The usage of wavelet encoded spectra features is especially helpful in detection of small differences which maybe easily ignored by standard approaches as well as to generate a significant reduced number of points needed in further processing steps. The signal must not be shrinked to peak lists but could be preserved in its functional representation. SNG was able to process high-dimensional functional data and shows good regularization. By use of the Kolmogorov-Smirnoff test we found a ranking of the features related to mass positions in the original spectrum which allows for identification of most relevant feature dimensions and to prune irrelevant regions of the spectrum. Alternatively one could optimize the scaling parameters of the functional norm directly during classification learning by so called relevance learning as shown in (Hammer, 2005) for scaled Euclidean metric. Conclusively, wavelet spectra encoding combined with SNG and a functional metric is an interesting alternative to standard approaches. It combines efficient model generation with automated data pre-treatment and intuitive analysis.

REFERENCES Baumann, S., Ceglarek, U., Fiedler, G. M., & Lembcke, J. (2005). Standardized approach to proteomic profiling of human serum based magnetic bead separation and matrix-assisted laser esorption/ionization time-of flight mass spectrometry. Clinical Chemistry, 51, 973–980. doi:10.1373/ clinchem.2004.047308

Cohen, A., Daubechies, I., & Feauveau, J.-C. (1992). Biorthogonal bases of compactly supported wavelets. Communications on Pure and Applied Mathematics, 45(5), 485–560. doi:10.1002/ cpa.3160450502 Hammer, B., Strickert, M., & Villmann, T. (2005). Supervised neural gas with general similarity measure. Neural Processing Letters, 21(1), 21–44. doi:10.1007/s11063-004-3255-2 Haykin, S. (1999). Neural Networks (2nd ed.). Englewood Cliffs, NJ: Prentice Hall. Ketterlinus, R., Hsieh, S.-Y., Teng, S.-H., Lee, H., & Pusch, W. (2005). Fishing for biomarkers: analyzing mass spectrometry data with the new clinprotools software. BioTechniques, 38(6), 37–40. doi:10.2144/05386SU07 Kohonen, T. (1995). Self-Organizing Maps, volume 30 of Springer Series in Information Sciences. Springer, Berlin, Heidelberg, (2nd Ext. Ed. 1997). Lee, J., & Verleysen, M. (2005) Generalizations of the lp norm for time series and its application to self-organizing maps. In Marie Cottrell, editor, 5th Workshop on Self-Organizing Maps, volume 1, pages 733–740. Mallat, S. (1998) A wavelet tour of signal processing. San Diego, CA: Academic Press. Martinetz, T., Berkovich, S., & Schulten, K. (1993). ’Neural-gas’ network for vector quantization and its application to time-series prediction. IEEE Transactions on Neural Networks, 4(4), 558–569. doi:10.1109/72.238311 Morris, J., Coombes, K., Koomen, J., Baggerly, K., & Kobayashi, R. (2005). Feature extraction and quantification for mass spectrometry in biomedical applications using the mean spectrum. Bioinformatics (Oxford, England), 21(9), 1764–1775. doi:10.1093/bioinformatics/bti254

483

Prototype Based Classification in Bioinformatics

Orchard, S., Hermjakob, H., & Apweiler, R. (2003). The Proteomcs Standards Initiative. Proteomics, 3, 1274–1376. Petricoin, E. F., Ardekani, A., Hitt, B., & Levine, P. (2002). Use of proteomic patterns in serum to identify ovarian cancer. Lancet, 359, 572–577. doi:10.1016/S0140-6736(02)07746-2 Pusch, W., Flocco, M., Leung, S. M., Thiele, H., & Kostrzewa, M. (2003). Mass spectrometrybased clinical proteomics. Pharmacogenomic, 4, 463–476. doi:10.1517/phgs.4.4.463.22753 Ransohoff, D. F. (2005). Lessons from controversy: ovarian cancer screening and serum proteomics. Journal of the National Cancer Institute, 97, 315–319. Rieder, A. Louis, A.K. & Maaß, P. (1998) Wavelets: Theory and Applications. Wiley. Sachs, L. (2003) Angewandte Statistik. Springer. Sato, A., & Yamada, K. (1996) Generalized learning vector quantization. In D. S. Touretzky, M. C. Mozer, and M. E. Hasselmo, editors, Advances in Neural Information Processing Systems 8. Proceedings of the 1995 Conference, pages 423–9. MIT Press, Cambridge, MA, USA. Schena, M., Shalon, D., Davis, R. W., & Brown, P. O. (1995). Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science, 270(5235), 467–470. doi:10.1126/science.270.5235.467 Schleif, F.-M. (2006) Prototype based Machine Learning for Clinical Proteomics. Technical University Clausthal, PhD-Thesis. Twyman, R. M. Principles of proteomics BIOS Scientific Publishers, NY,2004. Villanueva, J., Philip, J., Entenberg, D., & Chaparro, C. A. (2004). Serum peptide profiling by magnetic particle-assisted, automated sample processing and maldi-tof mass spectrometry. Analytical Chemistry, 76, 1560–1570. doi:10.1021/ac0352171

484

Villmann, T., & Hammer, B. (2002) Supervised neural gas for learning vector quantization. In D. Polani, J. Kim, and T. Martinetz, editors, Proc. of the 5th GermanWorkshop on Artificial Life (GWAL-5), pages 9–16. Akademische Verlagsgesellschaft - infix - IOS Press, Berlin. Waagen, D. E., Cassabaum, M. L., Scott, C., & Schmitt, H. A. (2003) Exploring alternative wavelet base selection techniques with application to high resolution radar classification. In Proc. of the 6th Int. Conf. on Inf. Fusion (ISIF’03), pages 1078–1085. IEEE Press.

KEY TERMS AND DEFINITIONS Bioinformatics: Generic term of a research field as well as a set of methods used in computational biology or medicine to analyse multiple kinds of biological or clinical data. It combines the disciplines of computer science, artificial intelligence, applied mathematics, statistics, biology, chemistry and engineering in the field of biology and medicine. Typical research subjects are problem adequate data pre-processing of measured biological sample information (e.g. data cleaning, alignments, feature extraction), supervised and unsupervised data analysis (e.g. classification models, visualization, clustering, biomarker discovery) and multiple kinds of modelling (e.g. protein structure prediction, analysis of expression of gene, proteins, gene/ protein regulation networks/interactions) for one or multidimensional data including time series. Thereby the most common problem is the high dimensionality of the data and the small number of samples which in general make standard approach (e.g. classical statistic) inapplicable. Biomarker: Mainly in clinical research one goal of experiments is to determine patterns which are predictive for the presents or prognosis of a disease state, frequently called biomarker. Biomarkers can be single or complex (pattern)

Prototype Based Classification in Bioinformatics

indicator variables taken from multiple measurements of a sample. The ideal biomarker has a high sensitivity, specificity and is reproducible (under standardized conditions) with respect to control experiments in other labs. Further it can be expected that the marker is vanishing or changing during a treatment of the disease. Clinical Proteomics: Proteomics is the field of research related to the analysis of the proteome of an organism. Thereby, clinical proteomics is focused on research mainly related to disease prediction and prognosis in the clinical domain by means of proteome analysis. Standard methods for proteome analysis are available by Mass spectrometry. Mass Spectrometry: An analytical technique used to measure the mass-to-charge ratio of ions. In clinical proteomics mass spectrometry can be applied to extract fingerprints of samples (like blood, urine, bacterial extracts) whereby semiquantitative intensity differences between sample cohorts may indicate biomarker candidates Prototype Classifiers: Are a specific kind of neural networks and related to the kNN classifier. The classification model consists of so called

prototypes which are representatives for a larger set of data points. The classification is done by a nearest neighbour classification using the prototypes. Nowadays prototype classifiers can be found in multiple fields (robotics, character recognition, signal processing or medical diagnosis) trained to find (non)linear relationships in data. Relevance Learning: A method, typically used in supervised classification, to determine problem specific metric parameter. With respect to the used metric and learning schema univariate, correlative and multivariate relations between data dimensions can be analyzed. Relevance learning typically leads to significantly improved, problem adapted metric parameters and classification models. Wavelet Analysis: Method used in signal processing to analyse a signal by means of frequency and local information. Thereby the signal is encoded in a representation of wavelets, which are specific kinds of mathematical functions. The Wavelet encoding allows the representation of the signal at different resolutions, the coefficients contain frequency information but can also be localized in the signal.

This work was previously published in Encyclopedia of Artificial Intelligence, edited by Juan Ramón Rabuñal Dopico, Julian Dorado and Alejandro Pazos, pp. 1337-1342, copyright 2009 by Information Science Reference (an imprint of IGI Global).

485

486

Chapter 2.18

An Integrated System for E-Medicine: E-Health, Telemedicine and Medical Expert Systems Ivan Chorbev Ss. Cyril and Methodius University, Republic of Macedonia Boban Joksimoski European University, Republic of Macedonia

ABSTRACT This chapter presents an overview of an integrated system for eMedicine that the authors propose and implement in the Republic of Macedonia. The system contains advanced medical information systems, various telemedicine services supported by modern telecommunication technologies, and decision support modules. The authors describe their telemedicine services that use wireless broadband technologies (WiMAX, 3G, Wi-Fi).

A significant part of the chapter presents a web based medical expert system that performs self training using a heuristic rule induction algorithm. The data inserted by medical personnel while using the e-medicine system is subsequently used for additional learning. The system is trained using a hybrid heuristic algorithm for induction of classification rules that we developed. The SA Tabu Miner algorithm (Simulated Annealing and Tabu Search based Data Miner) is inspired by both research on heuristic optimization algorithms and rule induction data mining concepts and principles.

DOI: 10.4018/978-1-60960-561-2.ch218

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

An Integrated System for E-Medicine

INTRODUCTION E-medicine can be viewed as a symbiosis between medicine, informatics and telecommunication technologies. Basically, e-medicine incorporates the use of computer technologies, multimedia systems and global networking in the provision of medical services. It is an area of great scientific and research interest, followed by fast implementation of novel commercial functionalities. A common simple definition describes e-medicine as the use of multimedia technologies like text, pictures, speech and/or video for performing medical activities. The goal of our research is to define a prototype of an integrated system for e-medicine that enables application of information and communication technologies over a wide spectrum of functionalities in the health sector including medical personnel, diagnostics, therapy, managers, medical insurance and patients. Additionally we aim at incorporating artificial intelligence in various modules of the system making it a useful partner to all entities using the system. We present algorithms for building medical decision support and expert systems as part of the e-medicine system. The chapter is organized as follows. The first section gives a short overview of e-medicine, telemedicine and medical expert systems. The second part explains in detail our model of a system for e-medicine, its main modules, the used technologies and the implemented functionalities. The third section gives as overview of the medical expert subsystem we implemented, along with the SA Tabu Miner rule induction algorithm for classification that we developed for that purpose.

BACKGROUND E-medicine (sometimes referred to as e-medicine or eHealth) is a rather new term for describing the medical care that is supported by modern electronic processes and modern telecommunications. It is sometimes used to describe the use of computers in

health institutions, for providing medical services via Internet or simply a new name for telemedicine. In fact, it is used to describe a wide spectrum of services that are part of the medical practice supported by the aid of information technology. The provided services include: •

•

•

•

•

The use of Electronic Medical Records (EHR) for easy storing, retrieving and sharing data between medical personnel (doctors, pharmacists, therapists etc). Telemedicine, as a way to provide medical services remotely and as a way for providing teleconsultations and assistance to other doctors Public Health Informatics, where the population and/or patients could get informed about relevant medical information. Management of medical information and medical knowledge, using the data in data mining research Mobile e-medicine, a field that includes the use of mobile devices for various purposes including real-time monitoring of patients, diagnosis, gathering and providing data for the doctors and mobile telemedicine

Medical Information Systems Information systems have been developing rapidly through the past decades, and we have now means of managing and organizing large quantities of data, methods of validating the data, and ways of processing the data for retrieving valuable information as well as learning from the data. However, for practical implementations, there are a lot of requirements that should be satisfied in order to make a healthcare information system usable. A lot of the tasks are concerned with gathering and manipulation of data provided by patients, doctors and insurance companies. The medical information is critical, and should be accessible, up to date and coherent at all times. Also, the data must be secure, confidential and protected from

487

An Integrated System for E-Medicine

unauthorized access. Security of medical data is regulated by various requirements, like HIPAA and European Commission’s Directive, and all systems should strictly implement them. The evolution of Healthcare Information Systems has gone through different phases (Vogel & Perreault, 2006). The first phase was marked by development of specific applications that would help healthcare workers in a specific area (like patient registration or laboratory results). Access to data was mostly at department level, and there was the problem of transferring data between departments in a hospital. To overcome the inoperability between different systems, an approach was taken to interconnect them. The second phase was the development of appropriate interface engines to make interconnection possible. The goal is to make data accessible wherever it is required. Progress is still made in interconnecting different countries or regions and it has been regarded as a challenging task.

Telemedicine Initially telemedicine was defined by Bird (1971) as “the practice of medicine without the usual physician-patient confrontation …via an interactive audio-video communications system” (p.67). Telemedicine basically provides options for neutralizing the geographical distance between the users. It is useful to spare the ill patients from the discomfort and expenses of traveling. Saving the medical personnel from travelling to remote areas leaves more time for the medical problems. The transfer of data from site to site ensures better judgment and informed decisions. Application areas of telemedicine expand to almost all the fields of medicine (Zielinski & Duplaga, 2006). Applications of telemedicine include: Consultation, Diagnostic Consultation, Monitoring, Education, Disaster management (Olariu et al., 2004), Virtual Microscopy (Fontelo et al., 2005), Homecare, Diagnosis, Treatment and Therapy (Psychology). Along the development of telemedicine, new

488

terms like eHealth have emerged. According to (Maheu, 2001), “E-health refers to all forms of electronic healthcare delivered over the Internet, ranging from informational, educational, and commercial “products” to direct services offered by professionals, nonprofessionals, businesses or consumer themselves” (p. 3). The history of telemedicine shows evident correlation with the developments in communication technology and IT software development. Researches categorize the telemedicine history into three eras based on the technological development (Bashshur et al., 2000; Tulu & Chatterjee, 2005). All the definitions during the first era of telemedicine focused on medical care as the only function of telemedicine. The first era can be named as telecommunications era of the 1970s (Bashshur et al., 2000). Applications in this era were dependent on broadcast and television technologies. Telemedicine was not integrated with any other clinical data. It was based on transmission of a TV and audio signal both ways, and lacked the ability to connect other devices and transmit data automatically. Hence, the use was limited to basic teleconsultations and low quality video communication with patients. The second era of telemedicine evolved with the digitalization in telecommunications (Bashshur et al., 2000). The transmission of data was supported by various communication mediums ranging from telephone lines to Integrated Service Digital Network (ISDN) lines. However, the high costs attached to the communication mediums that can provide higher bandwidth became an important bottleneck for telemedicine. This era overcame a lot of the problems in the first era, like patient feedbacks, data transfer, but was not applicable for general use, mostly because the expensive services needed (ISDN, satellite communication). The third era has been named “Internet era”, where the main burden for interconnecting is transferred to the Internet network, as a robust system that is accessible, already deployed, relatively

An Integrated System for E-Medicine

cheap and constantly growing. Also, the technologies to access the Internet are constantly evolving for better performance and higher speeds. Regardless of the technology used to access the network, previous services still function without modifications, even faster and more capable (Bashshur et al., 2000). The development of mobile wireless Internet access, by using 3G wireless networks and high-speed WiMAX provides new ways of implementing telemedicine services. Other telemedicine projects are based on alternative ways of establishing communication, like satellite-based telemedicine (Healthware project - http://healthware.alcasat.net/), and although this provides effective large scale connectivity, its deployment and maintenance is difficult and thus its services are expensive. The wireless networks are ultimately the best cost-effective solution for fast deployment and large area covering, and WiMAX as the leading technology for fast, broadband Internet access.

Expert Systems Expert Systems belong to the broader class of Decision Support Systems, and are viewed as consultants to human decision makers. The use of expert systems in medical problems has always been a topic of interest, and an integrated system for e-medicine must include a decision support system in order to gain the advantages that artificial intelligence can offer. Some of the reasons why decision support systems are necessary include the following challenges that the medical personnel is facing: • • •

•

Classification – deciding who is treated first, and who is sent back Treatment - the decisions for treatment must be informed and expert consulted The physicians’ knowledge in a particular area might be outdated, and the time for reading new publications is always limited Physicians go into diagnostic habits

•

• • •

The physician might be inexperienced with diagnosing or treating particular rare diseases The treatment might be too expensive and the doctor is hesitating to prescribe Time constraints to make a decision The doctor might be unaware of the existence of some new drugs or their side effects

Medical expert systems have been an area of research and implementation since the 1970s. A major part of the research in Artificial Intelligence was focused on expert systems. Various programming languages like LISP and Prolog were introduced for the purpose of declarative programming and knowledge representation, later to be used for development of expert systems. Famous medical expert systems include Mycin®, designed to identify bacteria causing severe infections and to recommend antibiotics, CADUCEUS®, embracing all internal medicine, also Spacecraft Health Inference Engine (SHINE)®, STD Wizard® etc. There are two fundamental approaches for knowledge base construction for the expert system: knowledge acquisition from a human expert, and empirical induction of knowledge from collections of training samples. In our research we choose the later. We used an algorithm for rule induction that we developed to create a medical expert system.

MODEL OF AN INTEGRATED SYSTEM FOR E-MEDICINE Services in E-Medicine are growing rapidly with the development of the underlying telecommunication and computer technologies. The developed applications quickly get replaced by newer services that use the bandwidth and accessibility of the newest communication technologies. Still, by analyzing the incoming technologies and anticipated commercial needs, the medical services,

489

An Integrated System for E-Medicine

architectures and design of the e-medicine systems of the future can be predicted to some extent. The system must be based on reliable communication technologies. Cable and optical networks will always be faster, more secure and reliable than wireless technologies, hence ensuring their infrastructural function as backbones. Wireless networks have the drawbacks of providing less resource in terms of data traffic and bandwidth, but add the very significant mobility to the users. Speeds in wireless networks are constantly growing, and with standards like WiMAX of 4G mobile telephony, the implementation of cheaper and faster e-medical services is ensured. Integrating different multiplatform services is the crucial phase of the development of an integrated system for e-medicine. IT Systems in general are in constant development, and as such should have modular design. Older modules should be replaced with newer without interfering with the overall functioning of the system. The Healthcare Information System should be designed to provide complete and reliable storage of electronic records, implementing the concerns of privacy, confidentiality and security. Furthermore, the use of unique electronic identifications in form of chip cards, proximity cards or RFID Figure 1. Diagram of a modular e-health system

490

can help improve patient care by identifying and getting medical records from patients that can’t communicate (unconscious patients). The diagram of an integrated system is shown on Figure 1. Users of the system can use various devices to access parts of the system. With the implementation of encrypted communication with XML web services, the modules do not depend on the telecommunication technology or the devise used for access (Notebook, PDA, mobile phone or other terminal). Because of modularity, newer services can be deployed when devices that support that service will be available. Of course, the new devices should support older services that are still in use. The integrated and coherent patient data, through wireless network, are accessible to personnel in hospitals, ambulance vehicles, in a medical campus and even in homes. Data that can be transmitted can be in form of text, patient records, video streaming, audio recordings, real-time data from sensors etc. The services can also be used for management purposes, like allocation of human resources, checking supplies of medical equipment, making appointments, control of patients etc.

An Integrated System for E-Medicine

Implementing an E-Medicine System in a Developing IT Society In the Republic of Macedonia there is no integrated health information and communication system. There are only individual multiplatform systems at some of the hospitals, usually based on Microsoft SQL Server databases supporting various Windows forms or recently web applications. The analysis of the current solutions show significant differences among hospitals, but the overall conclusion is serious lack of Information and Communication Technology (ICT) and nonexistence of integrated hospital information systems, with small number of exceptions. Also, it is evident that information systems that operate in some facilities are challenging to integrate and extremely difficult to enable information exchange. While researchers in developed countries can have different goals, our objectives locally had to be scalable, ranging from establishment of basic telemedicine services up to advanced up to date functionalities. The main concepts that our system was based on included creation of necessary basic Medical Information Systems (MISs) where hospitals had none, interfaces for various MISs, using modern telecommunication technologies for creating an integrated MIS and provision of advanced medical services at remote locations and other telemedicine applications. (Chorbev & Mihajlov, 2008) There are many prerequisites for the integrated e-medicine system to be implemented. Organizational prerequisites include cooperation and communication between actors in the complex organizational structures. Human resources need to be trained to a level of understanding and familiarity with ICT. Legal prerequisites include regulating patient related information. The Republic of Macedonia already integrates Diagnosis Related Groups (DRG) into patient’s medical data, but the records have so far been distributed among various hospitals. In developed

integrated medical information systems there are numerous standardized records of patient data. With some initial organized data we already started research with diseases that are endemic for the region. By using data mining techniques, we developed modules that serve as diagnosis consultants. We used already available data in form of blood tests from 70000 patients in the Ohrid Orthopedic hospital. This data serves as training data input in our research where our rule induction algorithm SA TABU Miner (Chorbev et al., 2009) is used for generating classification rules. Other aspects of artificial intelligence, like combinatorial optimization, can also be integrated in the system. Using optimization tools to obtain optimal drug amounts in treatment can reduce the burden on the state health insurance fund. The limited number of existing surgery teams and equipment, kidney dialysis sets, MRI devices can be scheduled for use with combinatorial optimization, achieving maximal utilization according to specific patient needs and urgency. However, scattered results in research must be integrated in an e-medicine system for the benefits to be readily available. The modularity will enable every new diagnostic tool to be quickly integrated and delivered to physicians and patients. All developed modules of the integrated MIS need to be readily available to both patients and physicians throughout the country. By designing the system as a web and PDA application, available through statewide wireless network and secured with adequate authentication and encryption methods, it is expected that it will increase severely the quality of medical service.

Wireless Infrastructure In order to implement our telemedicine system we used the backbone network of a fast growing privately owned data communication provider. The backbone network consists of some fiber optic connections in the city limits of Skopje and among some major cities and mostly 802.16 (WiMAX)

491

An Integrated System for E-Medicine

base stations throughout the country. The optic fiber connections are used for provision of fast bandwidth services where possible. The WiMAX antennas are used for connecting hospitals where the optic fiber has not reached yet and 802.11 hotspots are used for wireless devices (PDAs, notebook PCs etc.) Due to the sufficient bandwidth of WiMAX, it is used to cover most of the needs of our telemedicine system. WiMAX is a telecommunication technology aimed at providing broadband wireless data connectivity over long distances. It is based on the IEEE 802.16 standard. The high bandwidth and increased reach of WiMAX make it suitable for providing a wireless alternative to cable and DSL for last mile broadband access. We tested the performance of both fixed outdoor and fixed indoor WiMAX antennas and the results are very promising, since both provided robust connections. Latest systems built using 802.16e2005 and the OFDMA PHY™ as the air interface technology are called “Mobile WiMAX” and are expected to provide broadband connections while the client is moving. Within the city limits of the capital Skopje there is a functional fiber optic Metro Ethernet network. The fiber optical connection enables Figure 2. Optical network in Macedonia

492

fast and robust connectivity for provision of advanced telemedicine services like high quality video streaming of surgical procedures, medical visualization etc. Even when the fiber optical lines are used for communication, the WiMAX wireless lines could be used for backup in case of disrupted cable communication. While cables can be physically cut, the WiMAX connections are stable even in severe weather conditions. A wireless backbone network is established throughout the country, and hospitals in different cities are (or will be) connected to the network. Antennas are placed on hills overseeing cities, and coverage with the radio signal is good and robust. The backbone network is depicted in Figure22, while one antenna in Skopje.

Implemented Services and Functionalities In the initial stages we included two hospitals in the pilot project: The Institute for respiratory diseases in children-Kozle and the University clinical center in Skopje. Due to the lack of a modern Medical Information Systems (MIS) in the hospitals, we developed a prototype of a modular MIS that can later be distributed to all the hospi-

An Integrated System for E-Medicine

tals. The main database is centralized, since the data for approximately 2.000.000 inhabitants in the Republic of Macedonia can easily be handled by the contemporary database engines. Because of the idea for centralization of the database and since some hospitals cannot afford to maintain an IT department, the MIS is hosted on the Internet Service Provider’s (ISP) servers. Knowing that connectivity speeds are high enough when using WiMAX, there is no need to host the MIS locally at the hospital. The MIS is developed as a web application that can be accessed by a common Internet browser. Querying data in the web based MIS is possible using multiple criteria. Data can be searched from other patients with similar symptoms in order to learn from other previous experiences. Entire patient history is accessible online, with strong regard to privacy issues. While patient identity details are available to the physician in charge of the particular case, for other medical personnel with lower access privileges, only medical information is available, without disclosing the identity of the particular patient. The developed system includes software components specialized for use by PDA devices. Both patients and staff can wirelessly access different software modules. Physicians can access patient’s data, results from laboratory analyses, forums and chats, web sites with medical scientific papers. Patients can access their results from different analyses, make appointments, and check the availability of certain physicians. We paid great attention to the usability of the user interface in the PDA applications. Due to the resolution and dimension limitations, significant effort was made to maximize the utilization of the given space on the small screens and to enable easy navigation through the user interface. We adopted a policy of gradual increase of details presented on demand, since scrolling and navigating large texts is unpleasant on a PDA device. The system includes a Short Message Service (SMS) gateway that is used for SMS notifications for both physicians and patients. Current func-

tionalities include confirmation of appointments for patients, notification for completed laboratory analyses, SMS emergency calls for physicians on stand-by etc. The system can even notify the patient for the upcoming time for therapy or treatments. WiMAX is also used for Voice over IP (VoIP) services. PSTN telephone bills are drastically reduced as a result of the use of VoIP for communication among hospitals. A vital part of a telemedicine system is the sharing of knowledge, experience and expertise. The implemented MIS includes a forum and a virtual chat room where physicians can consult each other. The forum enables posting various laboratory results and even video and audio sequences from various diagnostic procedures for the consultations to be supported by appropriate information. Since the system is centralized, consultations are possible among physicians from all hospitals included in the system. Posted materials in the virtual chat room cannot be connected to the patient identity. Our system incorporates modules that enable laboratory results and other analyses to be submitted for review to the specialists. Physicians working in smaller towns can access the system using their accounts and can submit questions along with supporting materials electronically. Special web application software modules are developed for submitting images (MRI, X-Ray, CAT scan) from remote hospitals in the country to the specialist working in the capital. Also results from blood analysis are filled in online forms. Specialists review the results and can post their reply to the sender. This system enables reduction of transport costs, response times are drastically smaller and patients do not have to suffer through long trips to the specialist. We introduced a system of grading each submitted material giving it different priority according to the contents and level of urgency demanded by the sender. Extremely urgent submissions can even cause the SMS gateway to notify the specialist for the incoming request. We tested streaming video through the WiMAX

493

An Integrated System for E-Medicine

connections and we set up a system where experts from one hospital can oversee complex surgical operations performed by the surgeons elsewhere. Also, students at the University hospital in Skopje will be able to learn from the live feed from surgeries performed elsewhere. We performed several video streaming experiments from a paramedic’s vehicle to the hospital. In the first experiment we used a vehicle equipped with a WiMAX antenna and an MPEG coding device. A continual video stream was established, and used for transmitting live feed from the patient in the ambulance to the hospitals. The video link enables specialists to give advice to first aid workers on the scene of an accident, based on real time video feed from the patient’s condition. Paramedics could be supervised by experienced medical personnel while performing necessary life support interventions. In the second experiment we used a 3G mobile telephony device for the same use, but the bandwidth was insufficient for a higher quality video. Due to current limitations of WiMAX, the ambulance must not move while being connected online. However new equipment based on Mobile WiMAX (802.16e-2005) is expected to overcome this issue. The equipment used in the experiment was SCOM® MPEG-2 Digital Video Encoder/ Decoder. The used WiMAX antennas support 2-10 Mbit/s. The particular experiment used 2 Mbit/s, but an acceptable video quality is achieved even with a 512 Kbit/s connection. A third experiment was conducted using a personal computer instead of a specialized MPEG coding device; however a noticeable delay was evident in the video stream. The later architecture is applicable for a smaller spectrum of services. The small indoor antennas were also used for video telephony experiment. We tested a scenario where an older woman suffering from strong pain in the back and almost immobilized, had to communicate with her doctor for consultation. Since transportation of the patient was difficult and painful, we brought the WiMAX antennas and IP video phones at both locations (the patient’s and

494

doctor’s) and established a video link that they used for consultation. We also used the video phones for establishing sign language communication for patients with impaired hearing. We used Leadtek® IP broadband videophones (BVP8882). They use H.323 protocol for high performance and good quality video communication. The quality of the video stream using only 256 Kbit/s was sufficient for the common sign language to be used and understandable by the communicating parties. The video signal that we used in most of the testing originated from a digital video camera. Another even more import feature is streaming of digitalized video signals received from analogous endoscopy equipment. We worked on digitalization of an analogous signal from a fluoroscopic camera using a Plextor® MPEG encoder. The digital output from the encoder was easily streamed. The received live video could be used to consult subspecialists not present at the location where the exam is performed. Using VoIP and chat on PDA devices, the specialist could provide feedback and guidance to the person performing the exam in the field or in the remote hospital. The implementation of the system consists of three main parts: the database, the online web and PDA applications and a standalone application that performs batch data processing and performs scheduled jobs and maintenance functionalities. Most of the applications are developed in Microsoft® .NET technology, using SQL Server 2005® as a database engine, and some are coded in PHP and hosted on Apache servers using MySQL databases.

OUR MEDICAL EXPERT SYSTEM A crucial module in the integrated system was a decision support subsystem that would provide useful advice to users based on information gathered from the medical information system in use. For that purpose we developed several expert subsystems, the main of which is a diagnostic module.

An Integrated System for E-Medicine

The goal of the web medical expert subsystem presented here is to serve as a consultant to physicians when setting a diagnosis. The physician logs in the system and chooses the appropriate input form that suite the test data that he/she is going to enter. We have previously collected training data from various patient histories from hospitals and training datasets from the UCI machine learning repository™ (http://archive.ics.uci.edu/ml/datasets.html). By using the training data we trained classifiers using the SA Tabu Miner algorithm. After the doctor inputs the patient’s data and submits the form, the classifier assigns a class (disease) to the patient. The classifier also presents the percent of its predictive certainty, hence the certainty of its diagnosis. The data entered by the physician is stored. Later, when the diagnosis is confirmed and the entered data is rechecked, the confirmed records are added to the training data for a new training cycle of the classifiers, each time with more data. The training cycle is repeated in different intervals for different classifiers, since not all diagnostic forms are used with the same frequency. The tendency is to run the training when new training cases exceed 5% of the previous number of training records. The training process is based on a well-known 10-fold cross-validation procedure (Weiss & Kulikowski, 1991). Each data set is divided into 10 mutually exclusive and exhaustive partitions and the algorithm is run once for each partition. Each time a different partition is used as the test set and the other 9 partitions are grouped together and used as the training set. The predictive accuracies (on the test set) of the 10 runs are then averaged. Eventually the training is performed with the entire dataset so that the generated rules are based on the entire knowledge available. These final rules have significantly more impact in the deciding process when the system is in use. The rules generated in each iteration are stored. When deciding, the different groups of rules vote for the final classification of the new case with different

impact, dependant on the predictive accuracy. The system adopts the weighted majority vote approach to combine the decision of the rule groups. The average predictive accuracy is necessary to estimate the reliability of the system when used as a diagnosis consultant.

Extracting Knowledge in Expert Systems, Rule Induction Some efforts were made to implement an expert system entirely using the ID3 algorithm (Mingers 1986, Quinlan 1986), while other expert systems use combination of data mining and human knowledge verification (Holmes & Cunningham, 1993). Empirically derived knowledge is commonly represented by classification rules, gained using specific algorithms. A number of induction methods were devised for extracting knowledge from the data, and most common known are decision trees (Breiman et al., 1984), rough set theory (Tsumoto et al., 1995), artificial neural networks, etc. This text describes an algorithm for rule induction called SA Tabu Miner (Simulated Annealing and Tabu Search based Data Miner). The goal of this rule induction algorithm is a type of data mining that aims to extract knowledge in form of classification rules from data. Simulated Annealing (SA) (Kirkpatrick et al., 1982) and Short-term Tabu Search (TS) algorithm (Zhang & Sun, 2002; Sait & Youssef, 1999; Glover, 1989) are used to develop the algorithm since the rule discovery problem is NP-hard. To the best of our knowledge, the use of SA and TS algorithms for discovering classification rules in data mining is still unexplored research area. Despite the accuracy of the discovered knowledge, it is equally important for the derived rules to be comprehensible for the user (Fayyad et al., 1996; Freitas & Lavington, 1998). Comprehensibility is very important especially when the discovered knowledge will be used for supporting a decision made by a human user. Comprehensible knowledge can be interpreted and validated by a

495

An Integrated System for E-Medicine

human. Validated knowledge will be trusted and actively used for decision making. When deriving rules for classification, comprehensibility is achieved with discovering rules that contain less terms in the “if” part. Also, fewer rules are easier to comprehend.

An Overview of Rule Induction Algorithms Several methods have been proposed for the rule induction process such as ID3 (Quinlan, 1986), C4.5 (Quinlan, 1993), CN2 (Clark & Boswell, 1991), CART(Breiman et al., 1984), AQ15 (Holland, 1986) and Ant Miner (Parepinelli et al., 2002). All mentioned algorithms can be grouped into two broad categories: sequential covering algorithms and simultaneous covering algorithms. Simultaneous covering algorithms like ID3 and C4.5 generate the entire rule set at once, while the sequential covering algorithms like AQ15 and CN2 learn the rule set in an incremental fashion. The algorithm ID3 (Iterative Dichotomiser 3) is used to generate a decision tree. The ID3 algorithm takes all unused attributes and evaluates their entropy. It chooses the attribute for which the entropy is minimal and makes a node containing that attribute. Always using the attribute with minimal entropy rule can lead to trapping in local optima. C4.5 is an extended version of ID3. It implements a “divide-and-conquer” strategy to create a decision tree through recursive partitioning of a training dataset. The final tree is transformed into an equivalent set of rules, one rule for each path from the root to a leaf of the tree. Creating decision trees means using quite a lot of memory which grows exponentially when the number of attributes and classes increases. CN2 works by finding the most influential rule that accounts for part of the training data, adding the rule to the induced rule set, removing the data it covers, and then iterating this process until no training instance remains. The most in-

496

fluential rule is discovered by a beam search, a search algorithm that uses a heuristic function to evaluate the promise of each node it examines. Always using the most influential rule can also lead to trapping in local optima.

Heuristic Search Algorithms in Data Mining and Rule Induction Ant Colony Optimization (ACO) has recently been successfully used for rule induction by Parepinelli et al. (2002). He developed an ACO based algorithm that managed to derive classification rules with higher predictive accuracy than CN2. Also, the ACO derived rules were quite simpler and more comprehensible for humans. Some general use of Simulated Annealing and Tabu Search in data mining applications can be found in literature. Zhang has used TS with short-term memory to solve the optimal feature selection problem (Zhang et al. 2002). Tahir et al. (2005) also propose a feature selection technique using Tabu Search with an intermediate-term memory. TS is used (Bai, 2005) for developing a Tabu Search enhanced Markov Blanket (TS/MB) procedure to learn a graphical Markov Blanket classifier from data sets with many discrete variables and relatively few cases that often arise in health care. Johnson and Liu (2006) have used the traveling salesman approach for predicting protein functions. Unlike SA and TS, Genetic algorithms (GA) can be found related to classification more often. Bojarczuk et al. (2000) use genetic programming for knowledge discovery in chest pain diagnosis. Weise et al. (2007) developed a GA based classification system to participate in the 2007 DataMining-Cup Contest, proving that combinatorial optimization heuristic algorithms are emerging as an important tool in data mining. Other cases of algorithms for deriving classification rules using Genetic Algorithms are referenced in Gopalan et al., (2006); Otero et al., (2003); Yang et al., (2008); Podgorelec (2005).

An Integrated System for E-Medicine

SA Tabu Miner - Algorithm for Rule Induction The SA Tabu Miner algorithm incrementally constructs and modifies a solution - a classification rule of the form: IF < term1 AND term2 AND ...> THEN

Each term is a triple . Since the operator element in the triple is always “=”, continuous (real-valued) attributes are discretized in a preprocessing step using the C4.5 algorithm (Quinlan, 1993). A high level description of the SA Tabu Miner algorithm is shown in Algorithm 1. The algorithm creates rules incrementally, performing a sequential process to discover a list of classification rules

Algorithm 1. SA Tabu Miner, the rule induction algorithm. TrainingSet = {all training cases}; DiscoveredRuleList = [ ]; /*initialized with an empty list*/ While (TrainingSet > Max_uncovered_cases) Calculate entropy and hence probability Start with an initial feasible solution S ∈ Ω. Initialize temperature While (temperature > MinTemp) Generate neighborhood solutions V* ∈ N(S). Update tabu timeouts of recently used terms Sort by (quality/tabu order) desc sol-s S* ∈ V* S* = the first solution ∈ V* While(move is not accepted or V* is exhausted) If metrop(Quality(S) - Quality(S*)) then Accept move and update best solution. Update tabu timeout of the used term break while End if S* = next solution ∈ V* End while Decrease temperature End While Prune rule S Add discovered rule S in DiscoveredRuleList TrainingSet = TrainingSet - {cases covered by S}; End while where • Ω is the set of feasible solutions, • S is the current solution, • S∗ is the best admissible solution, • Quality(S) is the objective function, • N(S) is the neighborhood of solution S, • V∗ is the sample of neighborhood solutions. 497

An Integrated System for E-Medicine

covering as many as possible training cases with as big quality as possible. At the beginning, the list of discovered rules is empty and the training set consists of all the training cases. Each iteration of the outer WHILE loop of SA Tabu Miner, corresponding to a number of executions of the inner WHILE loop, discovers one classification rule. After the rule is completed, it is pruned from the excessive terms, to exclude terms that were wrongfully added in the construction process. The created rule is added to the list of discovered rules, and the training cases covered by this rule are removed from the training set. This process is iteratively performed while the number of uncovered training cases is greater than a user-specified number called Max_uncovered_cases, usually 5% of all cases. The selection of the term to be added to the current partial rule depends on both a problemdependent heuristic function (entropy based probability), a tabu timeout for the recently used attribute values and the metropolitan probability function based on the Boltzman distribution of probability. The algorithm keeps adding one term at a time to its partial rule until one of the following two stopping criteria is met:

able values in the rule terms, but a combination of less probable ones. The entropy based probability guides and intensifies the search into promising areas (attribute values that have more significance in the classification), therefore intensifying the search. The tabu timeouts that recently used attribute values are given, discourage their repeated use, therefore diversifying the search. The metropolitan probability function controls the search, allowing greater diversification and broader search at the beginning, while the control parameter temperature is big, and later in the process, when the control parameter temperature is low, it intensifies the search only in promising regions. An important step in the rule construction process is the neighborhood function that generates the set V* of neighborhood solutions of the current rule S. The neighborhood contains as many rule proposals as there are values of attributes in the dataset. Let termij be a rule condition of the form Ai = Vij, where Ai is the i-th attribute and Vij is the j-th value of the domain of Ai. The selection of the attribute value placed in the term is dependent on the probability given as follows:

•

Pij =

•

Any term to be added to the rule would make the rule cover a number of cases smaller than a user-specified threshold, called Min_cases_per_rule. The control parameter “temperature” has reached its lowest value.

Every time an attribute value is used in a term added to the rule, its tabu timeout is reset to the number of values of the particular attribute. In the same time, all other tabu timeouts of the other values for the particular attribute are decreased. This is done to enforce the use of various values rather than the most probable one, since often the difference in probability between the most probable one and the others is insignificant. Therefore the final solution might not include the most prob-

498

φij λ ×TabuTimeoutij



(1)

where φij =

log2 k − H ij bi

∑ (log j =1

2



(2)

k − H ij )

where: • • • •

bi is the number of values for the attribute i. k is the number of classes. Hij is the entropy H(W|Ai=Vij). TabuTimeoutij is a parameter for the attribute value Vij, reset to the number of values

An Integrated System for E-Medicine

•

bi of the attribute Ai, when Vij is used in a term. λ is a multiplier of the tabu timeout used to increase/decrease the influence of the tabu timeout to the probability of using the particular attribute value.

For each termij that can be “added” to the current rule, SA Tabu Miner computes the value Hij of a heuristic function that is an estimate of the term quality, with respect to its ability to improve the predictive accuracy of the rule. This heuristic function is based on the Information Theory (Cover & Thomas, 1991). More precisely, the value of Hij for termij involves a measure of the entropy (or amount of information) associated with that term. The entropy of each termij (of the form Ai=Vij,) is given in the formula: H ij ≡ H (W Ai = Vij ) = −∑ (P (w Ai = Vij ) ⋅ log2 P (w Ai = Vij )) k

w =1

(3)

where: •

•

W is the class attribute (i.e., the attribute whose domain consists of the classes to be predicted). P(w|Ai=Vij) is the empirical probability of observing class w conditional on having observed Ai=Vij.

The higher the value of the entropy H(W|Ai=Vij), the classes are more uniformly distributed and so, the smaller the probability that termij would be part of the new solution. If the entropy and the tabu timeout are smaller, the more likely the attribute’s value is going to be used. H(W|Ai=Vij) of termij is always the same, for a constant dataset. Therefore, to save computational time, the H(W|Ai=Vij) of all termij is computed as a preprocessing step to every while loop. If the value Vij of attribute Ai does not occur in the training set, then H(W|Ai=Vij) is set to its maximum value of log2k. This corresponds to

assigning the lowest possible predictive power to termij. Second, if all cases belong to the same class then H(W|Ai=Vij) is set to 0. This corresponds to assigning the highest possible predictive power to termij. This heuristic function used by SA Tabu Miner, the entropy measure, is the same kind of heuristic function used by decision-tree algorithms such as C4.5 (Quinlan, 1993). The main difference between decision trees and SA Tabu Miner, with respect to the heuristic function, is that in decision trees the entropy is computed for an attribute as a whole, since an entire attribute is chosen to expand the tree, whereas in SA Tabu Miner the entropy is computed for an attribute-value pair only, since an attribute-value pair is chosen to expand the rule. The tabu timeout given to recently used attribute values servers as diversifier of the search, forcing the use of unused attribute values. Each time an attribute value Vij is used, its tabu timeout TabuTimeoutij is reset to the number of values bi of the attribute Ai. In the same time, the tabu timeouts of all remaining values of the attribute Ai are decreased by 1. The parameter λ serves to increase or decrease the influence of the tabu timeouts to the probability of using the particular attribute value. The usual value of λ is one, but in certain datasets, its value may vary to achieve better results. Once a solution proposal is constructed, it is evaluated using the quality measure of the rule. The quality of a rule, denoted by Q, is computed by the formula: Q = sensitivity • specificity (Lopes et al., 1998), defined by: Q=

TP TN ⋅ TP + FN FP + TN

(4)

where: TP - true positives, FP - false positives, FN false negatives, TN - true negatives. Q´s value is within the range 0 < Q < 1 and, the larger the value of Q, the higher the quality of the rule.

499

An Integrated System for E-Medicine

The quality of the rule is the “energy” parameter of the SA metropolitan function that decides if the new solution proposal will be accepted as the next solution. As soon as our algorithm completes the construction of a rule, the rule pruning procedure is invoked. The term whose removal most improves the quality of the rule is effectively removed from it, completing each iteration. This process is repeated until there is no term whose removal will improve the quality of the rule. The computational complexity of SA Tabu MinerThe most outer while loop of the algorithm will execute r times, r being the number of discovered rules. This number is highly variable depending on the particular dataset. Before a new rule is constructed, the algorithm calculates the probability of using each attribute value as a preprocessing step. This step has the complexity of O(n×a), where n is the number of instances in the training dataset, and a is the number of attributes. The inner while loop will execute t times, where t is the number of temperature decrements. Its contents have the complexity of: choosing a probable attribute value of each attribute - O(a×v2) and calculating the coverage and quality of each solution proposal - O(a×n×a), bubble sorting the solution proposals and choosing the most probable - O(a2). The rule pruning initially consists of evaluating k new candidate rules, derived by removing each one of the k terms from the rule. The next pruning iteration will evaluate k-1 new candidates etc., until pruning makes no improvement. k2 is the worst case scenario. Every evaluation has the complexity of O(n×k), hence the pruning process has the complexity of O(n×k3). Therefore the entire algorithm complexity is O(r(n×a+t×a×v2+t×n×a2+t×a2+n×k3)). Since the average number of values per attribute v is insignificant compared to n, it can be excluded. Some of the other components can collapse, so the final complexity is O (r×n×[t×a2+k3]).

500

Experimental Results and Discussion We compared the performance and results of our SA Tabu miner with the results derived by CN2, C4.5 and Ant Miner (Parepinelli et al., 2002). We used the CN2 version integrated in the WEKA™ package (http://www.cs.waikato.ac.nz/ml/weka), and for the Ant Miner we developed our source code based on the description in the published materials. We coded versions in C#.NET and in Pyton. The performances were evaluated using public-domain datasets from the UCI (University of California at Irvine) machine learning repository™ (http://archive.ics.uci.edu/ml/datasets.html). The comparison was performed across two criteria, the predictive accuracy of the discovered rule lists and their simplicity (hence comprehensibility). Predictive accuracy was measured by a well-known ten-fold cross-validation procedure (Weiss & Kulikowski, 1991). The comparison of predictive accuracies after the 10-fold cross-validation procedure is given in table 1. As shown in the Table 1, SA Tabu Miner achieved better predictive accuracy than CN2 in the Dermatology dataset, ANN, BUPA, Echocardiogram, Haberman and better accuracy than Ant Miner in tic-tac-toe, AllBP, BUPA New Tyroid, Echocardiogram and Haberman sets. In the other cases, the predictive accuracy is almost equal or slightly smaller than the other two. Despite the insignificantly smaller accuracy on some of the datasets, the advantage of SA Tabu Miner is in the robustness of the search, and its independence from the dataset size and number of attributes. While other algorithms that build a decision tree, might need exponential time or extremely large portions of memory to work, SA Tabu Miner can perform its search quickly, utilizing modest memory resources. Figure 3 shows a graphical representation of the results in Table 1. An important feature of classification algorithm is the simplicity of the discovered rule list, mea-

An Integrated System for E-Medicine

Table 1. The predictive accuracy of the SA Tabu Miner algorithm compared with the predictive accuracy of CN2 and Ant miner. Predictive accuracies (%) Data Sets Ljubljana breast cancer

SA Tabu Miner 65,1

CN2 67,69

Ant Miner 75,28

C45 73,22

Wisconsin breast cancer

90,3

94,88

96,04

93,28

tic-tac-toe

84,7

97,38

73,04

61,81

Dermatology

91,3

90,38

94,29

91,43

Hepatitis

89,2

90,00

90,00

83,33

AllBP

96,91

97,2

94,99

97,33

ANN

92,71

92,07

92,72

93,64

BUPA

63,70

58,53

57,23

59,12

New Tyroid

92,44

93,79

87,96

91,43

Echocardiogram

54,40

53,33

54,36

53,00

Haberman

74,86

66,33

73,32

72

Mamography

79,4

80,84

83,23

81,58

Transfusion

75,56

77,3

74,33

74,87

Nursery

43,7

74,08

86,38

82,72

Pima

68,4

67,5

67,96

66,58

PostOp

68,9

73,75

58,89

73,61

Heart disease

52,55

53

56,26

57,24

Parkinson

88,65

88

80,08

90

sured by the number of discovered rules and the average number of terms (conditions) per rule. Simplicity is very important for any use of the rules by humans, for easier verification and implementation in praxis. The results comparing the simplicity of the rule lists discovered by SA TabuMiner, CN2 and Ant Miner are reported in Table 2. SA Tabu Miner achieved significantly smaller number of simpler rules in all datasets compared with CN2, while deriving simpler rules than Ant Miner in the case of Wisconsin breast cancer, dermatology, Hepatitis, AllBP, ANN, New Tyroid, Haberman etc. In the Ljubljana breast cancer and tic-tac-toe, the simplicity is very similar with the one achieved by Ant Miner. Figure 4 shows a graphical representation of the results in Table 2. The parameters of the meta-heuristic algorithms used (SA, TS) have significant impact on

the result quality. The given results are achieved by using a SA starting temperature of n×a/4, n being the number of training samples and a being the number of attributes. The temperature decrement and the final temperature were set to 1. The dependency of the SA parameters from the dataset size in not linear and aside from some nonlinear functions that we derived, different values can be derived empirically for each dataset.

CONCLUSION This chapter describes the basic framework of an integrated system for e-medicine. The prototype of the system is described along with its most essential modules. Using the contemporary communication and software technologies we tried to define ways to significantly improve medicine and

501

An Integrated System for E-Medicine

Figure 3. Graphical representation of the predictive accuracy of SA Tabu Miner algorithm compared with the CN2, Ant miner and C4.5

health care. Most of the essential modules were developed, implemented and practically tested. Also, the performances of the used telecommunication technologies were put to the test. The proposed framework especially applies to a growing information society like the one in the Republic of Macedonia. In such places, e-medicine should follow step along with other growing IT areas in order to close the gap with developed countries. The framework proposed here and the steps already taken to implement it promise a fast trip toward a modern system that could enhance

502

the quality of medical services, reduce costs and increase patients satisfaction and health. The quality of mobility offered by modern wireless communication technologies like WiMAX enforce their use for various applications. They enable the provision of telemedicine services to places previously unreachable by landlines. Web services and XML enable integration of various Medical Information Systems into an Integrated System for E-Medicine. High bandwidth and reliability of WiMAX helps the integration with bringing remote hospitals ever closer.

An Integrated System for E-Medicine

Table 2. Simplicity of rules discovered by the compared algorithms. The number of rules and terms per rule # of Rules; Conditions per Rule Data Set

SA Tabu miner

CN2

Ant Miner

C45

Ljubljana breast cancer

8,55;1,70

55,40;2,21

7,10;1,28

9,7;2,56

Wisconsin breast cancer

6,10;2,15

34,67;1,94

12,6;1,07

37,9;1,85

tic-tac-toe

8,62;1,30

39,70;2,90

8,50;1,18

95;5,76

Dermatology

6,92;4,08

18,50;2,47

7,30;3,16

31;5,1

Hepatitis

3,21;2,54

7,20; 1,58

3,40;2,41

8;1,88

AllBP Thyroid disease

1:1

66,00;2,29

12,00;2,9

13;1,92

ANN Thyroid disease

1;1

86;3,45

11;2,99

27;3,19

BUPA liver disorders

9,9;0,95

98;3,03

8,2;0,98

24,4;2,91

New Tyroid gland data

5,8;0,87

24,7;1,85

7,2;1,07

13,5;1,40

Echocardiogram

8,3;1,25

40,7;2,33

6;1,55

11;1,91

Haberman’s survival

6,4;0,85

76,7;2,40

6,6;0,85

3,8;1,37

PostOp

7;1,23

29,4;2,68

5;1,80

19,89;0,35

Nursery

9,2;1,39

298,5;5,09

5,4;1,00

328;7,43

Pima

9,8;0,90

166,2;2,94

8,6;2,76

47,4;2,91

Mamography

8,5;0,89

117,3;2,87

9;0,88

17,9;1,86

Transfusion

8,8;0,90

43,2;2,43

6,3;1,21

12,9;1,79

Heart disease

8,3;1,06

132;3,28

11,1;1,84

94,20;4,11

Parkinson

7,80;0,9

23,1;2,05

5,9;1,14

33,8;3,92

We also present a web based medical expert system that performs self training using a heuristic rule induction algorithm. The system has a self training component since data inserted by medical personnel while using the integrated system for e-medicine is subsequently used for additional learning. For the purpose of training, the system uses a hybrid heuristic algorithm for induction of classification rules that we previously developed. The SA Tabu Miner is inspired by both research on heuristic optimization algorithms and rule induction data mining concepts and principles. We have compared the performance of SA Tabu Miner with CN2, C45 and Ant miner algorithms, on public domain data sets. The results showed that, concerning predictive accuracy, SATabu Miner obtained similar and often better results than the other approaches. Since comprehensibility is important whenever discovered knowledge will be used for supporting a decision made by a human user, SA Tabu Miner

often discovered simpler rule lists. Therefore, SA Tabu Miner seems particularly advantageous. Furthermore, while CN2 and C4.5 have its limitations when large datasets with big number of attributes are in question, SA Tabu Miner can still be applicable and will obtain good results due to the heuristic local search. Important directions for future research include extending SA Tabu-Miner to cope with continuous attributes, rather than requiring that this kind of attribute be discretized in a preprocessing step.

REFERENCES Bai, X. (2005). Tabu Search Enhanced Markov Blanket Classifier for High Dimensional Data Sets. In The Next Wave in Computing (Vol. 29, pp. 337–354). Optimization, and Decision Technologies.

503

An Integrated System for E-Medicine

Figure 4.Graphical representation of the rule simplicity of SA Tabu Miner algorithm compared with CN2 and Ant miner

Bashshur, R. L., Reardon, T. G., & Shannon, G. W. (2000). Telemedicine: A New Health Care Delivery System. Annual Review of Public Health, 21, 613–637. doi:10.1146/annurev.publhealth.21.1.613 Bird, K. T. (1971). Telemedicine; a new health information exchange system. In Annual Report no. 5. Washingon: Veterans administration. Bojarczuk, C. C., Lopes, H. S., & Freitas, A. A. (2000). Genetic programming for knowledge discovery in chest pain diagnosis. IEEE Engineering in Medicine and Biology Magazine, 19(4), 38–44. doi:10.1109/51.853480 504

Breiman, L., Friedman, J. H., Olshen, R. A., & Stone, C. J. (1984). Classification and Regression Trees. Belmont, CA: Wadsworth. Brewlow, L. A., & Aha, D. W. (1997). Simplifying decision trees: a survey. The Knowledge Engineering Review, 12(1), 1–40. doi:10.1017/ S0269888997000015 Chorbev, I., Mihajlov, D., & Jolevski, I. (2009). Web Based Medical Expert System with a Self Training Heuristic Rule Induction Algorithm. In DBKDA 2009, The First International Conference on Advances in Databases, Knowledge, and Data Applications, Cancun, Mexico: IEEE

An Integrated System for E-Medicine

Chorbev, I., & Mihajlov, M. (2008). Wireless Telemedicine Services as part of an Integrated System for E-Medicine. In MELECON08, The 14th IEEE Mediterranean Electrotechnical Conference (pp. 264-269), Ajaccio, France: IEEE Clark, P., & Boswell, R. (1991). Rule Induction with CN2: Some Recent Improvements. In Y. Kodratoff (Ed.), Fifth European Conference on Machine Learning (pp. 151-163). Berlin, Springer-Verlag. Fayyad, U. M., Piatetsky-Shapiro, G., & Smyth, P. (1996). From data mining to knowledge discovery: an overview. In Fayyad, U. M., Piatetsky-Shapiro, G., Smyth, P., & Uthurusamy, R. (Eds.), Advances in Knowledge Discovery & Data Mining (pp. 1–34). Menlo Park, CA, US: American Association for Artificial Intelligence. Fontelo, P., DiNino, E., Johansen, K., Khan, A., & Ackerman, M. (2005). Virtual Microscopy: Potential Applications in Medical Education and Telemedicine in Countries with Developing Economies. 38th Hawaii International Conference on System Sciences. Freitas, A. A., & Lavington, S. H. (1998). Mining Very Large Databases with Parallel Processing. London, UK: Kluwer. Glover, F. (1989). Tabu search - Part I. ORSA Journal on Computing, 1(3), 190-206. Gopalan, J., Alhajj, R., & Barker, K. (2006). Discovering Accurate and Interesting Classification Rules Using Genetic Algorithm. In International Conference on Data Mining, DMIN 2006 (pp. 389-395). Las Vegas, Nevada: CSREA Press Holland, J. H. (1986). Escaping brittleness: the possibilities of general purpose algorithms applied to parallel rule-based systems. In Michalski, R. S., Carbonell, J. G., & Mitchell, T. M. (Eds.), Machine Learning, an AI Approach (Vol. 2, pp. 593–623). San Mateo, California: Morgan Kaufmann.

Holmes, G., & Cunningham, S. J. (1993). Using data mining to support the construction and maintenance of expert systems. In Artificial Neural Networks and Expert Systems. (pp. 156-159). First New Zealand International Two-Stream Conference. Johnson, O., & Liu, J. (2006). A traveling salesman approach for predicting protein functions. Source Code for Biology and Medicine, 1(3), 1–7. Kirkpatrick, S., Gelatt, C. D. Jr, & Vecchi, M. P. (1983). Optimization by Simulated Annealing. Science, 220(4598), 671–680. doi:10.1126/science.220.4598.671 Lopes, H. S., Coutinho, M. S., & Lima, W. C. (1998). An evolutionary approach to simulate cognitive feedback learning in medical domain. In Sanchez, E., Shibata, T., & Zadeh, L. A. (Eds.), Genetic Algorithms and Fuzzy Logic Systems: Soft Computing Perspectives (pp. 193–207). Singapore: World Scientific. Maheu, M. M., Whitten, P., & Allen, A. (2001). E-Health, Telehealth, and Telemedicine. John Wiley and Sons. Mingers, J. (1986). Expert Systems-Experiments with Rule Induction. The Journal of the Operational Research Society, 37(11), 1031–1037. Olariu, S., Maly, K., Foudriat, E. C., & Yamany, S. M. (2004). Wireless support for telemedicine in disaster management.Tenth International Conference on Parallel and Distributed Systems (pp. 649-656). Otero, F. E. B., Silva, M. M. S., Freitas, A. A., & Nievola, J. C. (2003). Genetic Programming for Attribute Construction in Data Mining. In Genetic Programming (pp. 384–393). Berlin, Heidelberg: Springer. doi:10.1007/3-540-36599-0_36

505

An Integrated System for E-Medicine

Parepinelli, R. S., Lopes, H. S., & Freitas, A. (2002). An Ant Colony Algorithm for Classification Rule Discovery. In Abbass, H. A., Sarker, R. A., & Newton, C. S. (Eds.), Data Mining: Heuristic Approach (pp. 191–208). Hershey, PA: Idea Group Publishing. Podgorelec, V., Kokol, P., Molan Stiglic, M., Heričko, M., & Rozman, I. (2005). Knowledge Discovery with Classification Rules in a Cardiovascular Database. Computer Methods and Programs in Biomedicine, 80, S39–S49. doi:10.1016/ S0169-2607(05)80005-7 Quinlan, J. R. (1986). Induction of deciscion trees. Machine Learning, 1(1), 81–106. doi:10.1007/ BF00116251 Quinlan, J. R. (1987). Generating production rules from decision trees. In International Joint Conference on Artificial Intelligence: Vol. 1. Knowledge Representation (pp. 304-307). San Francisco: CA: Morgan Kaufmann. Quinlan, J. R. (1993). C4.5: Programs for Machine Learning. San Francisco, CA: Morgan Kaufmann. Sait. S. M., & Youssef, H. (1999). General Iterative Algorithms for Combinatorial Optimization. Los Alamitos, California, USA: IEEE Computer Society. Shortliffe, E. H., & Cimino, J. J. (Eds.). (2006). Biomedical Informatics: Computer Applications in Health Care and Biomedicine (Health Informatics). Springer. Tahir, M. A., Bouridane, A., Kurugollu, F., & Amira, A. (2005). A Novel Prostate Cancer Classification Technique Using Intermediate Memory Tabu Search. EURASIP Journal on Applied Signal Processing, (14): 2241–2249. doi:10.1155/ ASP.2005.2241

506

Tulu, B., & Chatterjee, S. (2005). A Taxonomy of Telemedicine Efforts with respect to Applications, Infrastructure, Delivery Tools, Type of Setting and Purpose. 38th Hawaii International Conference on System Sciences (pp. 147.2). Vogel, L. H., & Perreault, L. E. (2006). Management of Information in Healthcare Organizations. In Edward, H., & Cimino, J. J. (Eds.), Biomedical Informatics; Computer Applications in Health Care and Biomedicine (pp. 476–510). Berlin, Heidelberg: Springer. Weise, T., Achler, S., G¨ob, M., Voigtmann, C., & Zapf, M. (2007). Evolving Classifiers – Evolutionary Algorithms in Data Mining. [KIS]. Kasseler Informatikschriften, 2007, 1–20. Weiss, S. M., & Kulikowski, C. A. (1991). Computer Systems that Learn. San Francisco, CA: Morgan Kaufmann. Yang, Y. F., Lohmann, P., & Heipke, C. (2008). Genetic algorithms for multi-spectral image classification. In Schiewe, J., & Michel, U. (Eds.), Geoinformatics paves the Highway to Digital Earth:Festschrift zum 60 (pp. 153–161). Geburtstag von Prof. M. Ehlers. Zhang, H., & Sun, G. (2002). Feature selection using tabu search method. Pattern Recognition, 35(3), 701–711. doi:10.1016/S00313203(01)00046-2 Zielinski, K., Duplaga, M., & Ingram, D. (2006). Information Technology Solutions for Health Care, Health Informatics Series. Berlin: SpringerVerlag.

ADDITIONAL READING Chorbev, I., & Mihajlov, D. (2007). Integrated system for eMedicine in a developing information society. In International Multiconference. Ljubljana, Slovenija: Information Society

An Integrated System for E-Medicine

Cooke, C. D., Santana, C. A., Morris, T. I., DeBraal, L., Ordonez, C., & Omiecinski, E. (2000). Validating expert system rule confidences using data mining ofmyocardial perfusion SPECT databases. [Cambridge, MA, USA]. Computers in Cardiology, 2000, 785–788.

Okuyama, F., Hirano, T., Nakabayasi, Y., Minoura, H., Tsuruoka, S., & Okayama, Y. (2006). Telemedicine Imaging Collaboration System with Virtual Common Information Space. Sixth IEEE International Conference on Computer and Information Technology

Cover, T. M., & Thomas, J. A. (1991). Elements of Information Theory. New York: John Wiley & Sons. doi:10.1002/0471200611

Olariu, S., Maly, K., Foudriat, E. C., & Yamany, S. M. (2004). Wireless support for telemedicine in disaster management. Tenth International Conference on Parallel and Distributed Systems (pp. 649- 656)

Dhillon, H., & Forducey, P. G. (2006). Implementation and Evaluation of Information Technology in Telemedicine. 39th Hawaii International Conference on System Sciences. Holopainen, A., Galbiati, F., & Voutilainen, K. (2007). Use of smart phone technologies to offer easy-to-use and cost-effective telemedicine services. First International Conference on the Digital Society (p. 4). Hu, P. J. (2003). Evaluating Telemedicine Systems Success: A Revised Model. 36th Hawaii International Conference on System Sciences Kohavi, R., & Sahami, M. (1996). Error-based and entropy-based discretization of continuous features. In 2nd International Conference Knowledge Discovery and Data Mining (pp. 114-119). Menlo Park, CA: AAAI Press. Lach, J. M., & Vázquez, R. M. (2004). Simulation Model Of The Telemedicine Program, Winter Simulation Conference (Vol. 2, pp. 2012-2017). LeRouge, C., & Hevner, A. R. (2005). It’s More than Just Use: An Investigation of Telemedicine Use Quality. 38th Hawaii International Conference on System Sciences (pp. 150b - 150b). Maia, R. S., von Wangenheim, A., & vNobre, L. F. (2006). A Statewide Telemedicine Network for Public Health in Brazil. 19th IEEE Symposium on Computer-Based Medical Systems (pp. 495-500).

Paul, D. L. (2005). Collaborative Activities in Virtual Settings: Case Studies of Telemedicine. 38th Hawaii International Conference on System Sciences. Sadat, A., Sorwar, G., & Chowdhury, M. U. (2006). Session Initiation Protocol (SIP) based Event Notification System Architecture for Telemedicine Applications. 5th IEEE/ACIS International Conference on Computer and Information Science and 1st IEEE/ACIS International Workshop on Component-Based Software Engineering, Software Architecture and Reuse (ICIS-COMSAR’06) (pp. 214-218). Yamauchi, K., Chen, W., & Wei, D. (2004). 3G Mobile Phone Applications in Telemedicine - A Survey. The Fifth International Conference on Computer and Information Technology (CIT’05). J. Mauricio Lach, Ricardo M. Vázquez. Simulation Model Of The Telemedicine Program, 2004 Winter Simulation Conference (pp. 956-960). Zielinski, K., Duplaga, M., & Ingram, D. (2006). Information Technology Solutions for Health Care. Health Informatics Series. Berlin: SpringerVerlag.

This work was previously published in Ubiquitous Health and Medical Informatics: The Ubiquity 2.0 Trend and Beyond, edited by Sabah Mohammed and Jinan Fiaidhi, pp. 104-126, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global). 507

508

Chapter 2.19

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy Simulation Analysis Mahendran Maliapen University of Sydney, Australia, National University of Singapore, Singapore and UCLAN, UK Alan Gillies UCLAN, UK

ABSTRACT This paper uses simulation modelling techniques and presents summarized model outputs using the balanced scorecard approach. The simulation models have been formulated with the use of empirical health, clinical and financial data extracted from clinical data warehouses of a healthcare group. By emphasising the impact of strategic financial and clinical performance measures on healthcare institutions, it is argued that hospitals, in particular, DOI: 10.4018/978-1-60960-561-2.ch219

need to re-focus cost-cutting efforts in areas that do not impact clinicians, patient satisfaction or quality of care. The authors have added a real time component to business activity monitoring with the executive dashboards shown as graphs in this paper. This study demonstrates that it is possible to understand health policy interactions and improve hospital performance metrics through evaluation using balanced scorecards and normalized output data. Evidence from this research shows that the hospital executives involved were enthusiastic about the visual interactive interface that pro-

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

vides the transparency needed to isolate policy experimentation from complex model structures that map strategic behaviour.

cators such as patient satisfaction, clinical utilization and outcomes, financial performance as shown in Table 1:

INTRODUCTION

• • •

The provision of health care is a complicated activity requiring a multitude of skills, experiences and technologies. No one person or discipline can be responsible for poor or excellent performance. Similarly, hospitals are complex organizations that cannot be measured on a single dimension of performance. A balanced scorecard includes financial measures that capture the organisation’s ability to survive and grow. However, it complements financial measures with operational measures on customer satisfaction, internal processes, and the organization’s innovation and improvement activities (Caldwell, 1995). If well chosen, these operational measures capture the organisation’s operating performance, which is the ultimate driver of both current and future financial performance. The power of the balanced scorecard derives from its ability to present a succinct yet multifaceted picture of an organization to top management and a board of directors. A “balanced scorecard” for measuring the multiple dimensions of hospital performance is shown as four quadrants in Table 1. The objectives of this research paper was to develop, test and evaluate a sustainable hospital performance model for Senior Executives that would use both qualitative and quantitative indi-

• • •

•

•

Cash flow in the private hospital; Net cash balance; Patient satisfaction with hospital and clinician services; Clinician satisfaction with hospital management; Hospital bed occupancy; Deviations between the national average length of stay (NLOS) and the hospital’s LOS by Diagnostic Related Group (DRG) for patient admissions; Gap in available bed days comparing NLOS and hospital’s LOS for the patient admissions; and Average marginal costs per patient.

These cardinal dimensions are visually represented in radar chart format so that the simulated outcomes across the dimensions under different combination of hospital policies and scenarios can be visually compared against a reference baseline for these metrics. The Balanced Scorecard approach to represent the results was then integrated the simulation outputs so that each time a policy variation or scenario was tested, users could see the differences across all dimensions simultaneously the impact of the policy variations.

Table 1. The four quadrants’ of hospital performance Patient Satisfaction Examines patient perceptions of their hospital experience including the overall quality of care, outcome of care and unit-based care

Clinical Utilization and Outcomes Describes the clinical performance of PCH and refers to access to hospital services, clinical efficiency and quality of care

Financial Performance and Condition Describes how PCH manages the financial and human resources. Refers to the hospital’s financial health, efficiency, management practices and HR allocations

System Integration and Change Describes PCH’s ability to adapt, including how clinical information technologies, work processes and hospital-community relationships function within the hospital system.

509

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

RESEARCH METHODS AND METHODOLOGY The model development was conducted in consultation with key hospital executives, who identified the following key influencing factors as the enablers of strategic hospital business success. We illustrate how these factors have influenced the development and prototyping of the systems dynamics model, and in particular the development of the dynamic hypothesis for the Clinician & PCH Patient Market domain is described here. The feedback loops for other domains such as Clinical Governance, Financial Management and Hospital Management were developed applying the same methodology. Five hospital focus group members were actively involved in the model development process. The focus group, which included two clinicians, helped to identify the key trails in the flows of patients and monetary transactions taking a “process-orientated” view through the organization (Vissers & Beech, 2005). In order to understand hospital throughput and contributing drivers to workload it is necessary to understand the linkages between patient referrals, clinician interactions, hospital performance and costs. A visual walkthrough of the business processes was also conducted. The dynamic relationships and linkages between the various variables were identified for each sector. The model structure for each was developed based on these information linkages and the dynamic hypotheses and delays in response to actions have been included where appropriate. The relationships in model structure were developed based upon the experience of hospital professionals, researched literature references and evidence of documented hospital practices. The key factors as identified by the hospital focus group that would impact the Clinician and Patient Market domain in the typical Australian healthcare market were identified as:

510

• • • • • • • • •

Clinician / Hospital Alliances Clinician Consultation Charges Capitation & Private Insurance/Hospital Alliance Waiting Times and Delays Case mix Capitation and Hospital Fees Preference for “No-Gap” in Fees Co-location and Service Partnerships Outsourcing Targeted Market Growth

Clinician / Hospital Alliances Clinicians are pursuing various initiatives designed to increase their revenues and market share that put them in competition with hospital systems. A variety of clinician initiatives such as ‘focused-factory’ specialty hospitals, ambulatory care centres, disease management initiatives have mushroomed according to Dickinson et al. (1999). These are designed to compete directly for a share of the most lucrative hospital and ancillary revenue sources. As a collaborative strategy, hospitals are increasing using incentive schemes or gain sharing models to secure the loyalty of clinicians. These schemes are designed to provide incremental compensation to the participating clinician for achieving agreed upon reductions of inpatient costs or improvements in quality of care by following improved processes and standardized protocols. The economic linkage between the hospital and the clinician can be used to focus on hospital cost reduction and quality improvement (Chilingerian, 1992). The hospital focus group unanimously agreed the hospital should carefully evaluate its market strengths and vulnerabilities in light of the relative indispensability of targeted clinicians. Dickinson (1999) regards the balance of power or indispensability as the ability for the hospital to retain patients in the face of a clinician threat. Clinician and hospital alliances through various

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

mechanisms such as gain sharing, and collaboration can help to improve patient supply. The main source of the referrals to a clinician is the general practitioner (GP). Hence, the relationship between Clinicians and GP’s alliance is a key factor in determining the volume of patients referred for treatment. Successful treatment regimes and reputation of clinicians are created through the network of GP’s, which strengthens the relationships. The reinforcing loop R1 in Figure 1 exemplifies this monotonic tendency. The wider the GP network the greater the patient referrals and cash flow. The focus group participants felt that any strategic moves by the Board to increase the polarity of these relationships by providing clinicians in house residency with medical clinics at the hospital supports vertical downward integration to patient supply sources and would significantly increase patient admissions. Clinicians with high historical patient admissions should be targeted. Clinicians tend to form practice provider groups and consortiums to allow doctors and locums much more clout and equity in the delivery system. Practice consolidation also provides

larger practice volume and dominance in the marketplace. Clinicians have a one-to-many relationship with hospitals and therefore are at liberty to refer patients to hospitals of their choice. The objective of gaining patient dominance enhances the possibility of providing patient services in cardiology, general surgery, oncology, obstetrics and paediatrics. Partnerships with employers and workmen compensation groups to manage their medical components in multiple geographies and postcodes pose a serious threat to clinicians without such alliances or partnerships. The group felt that the Hospital should take cognizance of consortiums in their strategic analysis, as they are a threat to the supply of patients. The rate of disease incidence or epidemiological rate determines the number of cases occurred in the past. It is a mandatory requirement for GP’s and clinicians to report certain notifiable diseases that are likely to cause an epidemic. Public health statistics in primary care provides such analysis in an effort to control the growth of communicable diseases such as HIV/AIDS. Mortality rates, which are taken into account in computing inci-

Figure 1. Dynamic hypothesis for clinician & PCH patient market

511

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

dence rates, also provide useful indicators of the occurrence of heart failure, coronary thrombosis, diabetes, kidney disease and various forms of cancer in the community. Clinician caseload, long consultation hours and case complications can have both positive and negative effects on the quality of care and productivity of clinicians. Based on actual patient/ clinician admission ratios for PCH in 1998/1999 and 1999/2000, the focus group estimated that the clinician market share is estimated to be at least twice the number of PCH admissions. The backlog, thus created, in turn has the effect of reducing patient satisfaction and increasing waiting times for acute care. The prospect of delays due imaging, diagnostic and/or pathological investigations further reduces patient satisfaction.

Clinician Consultation Charges In a well-functioning competitive market, doctors’ charges would tend to reflect their relative costs (i.e., resource inputs). However, it is generally accepted that in Australia the market for medical services is imperfect. Factors such as the supply of doctors within the various disciplines, the impact of medical insurance, and the capacity of patients to pay and the ability of doctors to influence the demand for their services act to distort the market and therefore influence the pricing of services. In the context of a market approach if the going price for a particular surgical procedure is, say, AU$1,200 and a cost based relative value study indicates that the Medicare fee should be reduced from say AU$900 to AU$600 then the “gap” to the patient would widen. On the other hand, those specialists in relative oversupply could find themselves with no alternative but to accept prices dictated by third parties such as health funds, hospitals or both. In the private sector, market forces regulate the pricing mechanisms for patient consultations and clinicians have some discretion to set their consultation and procedure fees in accordance to

512

the AMA’s guidelines. Increase in consultation and anaesthetist fees above the Commonwealth Medical Benefits (CMB) scheduled rates will increase the patient or payer gap for the procedure and hence patient dissatisfaction. The higher the payment gap the more likely it is that patients will seek alternatives such as less costly clinicians for treatment or join a public hospital waiting list depending of the severity of the illness and degree of pain. In 1998, private healthcare insurance organizations, being well aware of the negative effects of clinician consultation and procedure fees, and in a bid to help their consumers (patients) alleviate the financial pressure, initiated the “no-gap” clinician’s scheme. Clinicians can therefore partner with insurance organizations for practice in designated hospitals to create a price cap for medical and surgical procedures. In return for clinician participation in the no-gap scheme, the hospital fees for theatre and facility usage by the clinician are reduced. In general, the implications of the price cap are to encourage more patient referrals to clinicians, which have the cascading effects of higher hospital admissions. This results in longer patient waiting times for surgery and patient dissatisfaction with the clinician (Martin et al., 1995).

Capitation & Private Insurance/ Hospital Alliance The growth of hospital capitation has been slow because most hospitals have not reached their “capitation threshold”. In the Australian context, capitation arises from the use of standardized cost and service weights to provide case mix adjusted revenue streams to hospital. In addition, most markets have not provided private healthcare insurance organizations with adequate incentives to share risk. Conditions that will drive private healthcare insurance organizations to adopt hospital capitation include true integration of care delivery systems and information systems,

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

sustained medical losses, and strong market position of hospitals. Hospital capitation has not become a prevalent payment mechanism, despite expectations that it would become the predominant method for controlling healthcare expenditures. Private healthcare insurance organizations will continue to be reluctant to share risk unless sufficient market pressure dictates a move to capitation contracts. Therefore, the motivation for hospital management to seek alliances with private healthcare insurance funds is a critical success factor. Arrangements with private insurance organizations to cap hospital fees have a significant impact on patient referrals for admission. Private insurance provides the second tier refund to the patient for the use of hospital beds. The accommodation charge refund is fixed for per day of overnight hospital stay. Day procedures performed at a hospital do not accrue an accommodation fee refund. The amount of refund is a function of the insurance premium paid by the patient (insurance table) and the level of healthcare cover held by the patient. Spiralling hospitals costs, on the one hand and the imposition of government control on insurance refunds on the other, create an unhappy marriage between the partners. Partnerships with private healthcare insurers and other managed care organizations are needed to keep the refunds at current levels as shown in the Balancing Loop, B3 in Figure 1. The net effect of

higher hospital accommodation charges is patients turning away to the public system.

Waiting Times and Delays The proportion of elective surgery patients waiting longer than the accepted standard is a nationally recognized indicator of access to acute care hospitals (Health Department of WA, 1998). Hospitals also collect waiting time data for internal management purposes. From a patient’s perspective, the relevant question is, ‘If I need surgery, what is the likelihood that I will have to wait longer than is considered desirable?’ Clinical judgments about need for surgery, and allocations by surgeons into the three categories of urgency, would need to be consistent across the hospital. As a private hospital, waiting times and delays are relatively short compared to the public system but this is an important factor for patient and clinician satisfaction (Australian Institute of Health and Welfare, 1995). In Figure 2, there are 3 balancing loops labelled B1, B2 and B3 and 2 reinforcing loops labelled R1 and R2. Loop B1, a balancing loop with delay, an archetype, where clinician referrals to the PCH hospital increases the proportion of the clinicians’ workload for PCH. The higher volume of referrals results in longer patients waiting times for PCH admission, which in turn increases patient dissatisfaction and that, slows down the rate at which patients are referred to that clinician. The opposing

Figure 2. The balanced scorecard: a potent tool (source: adapted from Chow et al., 1998)

513

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

effect in the loop is from link 10, which implies shorter waiting times reduce patient dissatisfaction and this characteristic behaviour has been observed in patient surveys conducted at PCH. The turn away loop is a balancing loop, shown as B2 in Figure 2, which dominates the rates of PCH admissions. Delays in pathological assessment exacerbate patients with critical and serious illnesses, in particular, who are placed on the waiting list. Clinicians or GP’s therefore, prefer to admit these episodes to other hospitals. As a consequence, it reinforces PCH hospital’s loss in market share. The same feedback loop occurs if the delay was due to a lack of operating theatre time slots or availability of beds at the hospital. This loop clearly demonstrates the leverage clinicians have in admitting patients and confirms the focus group’s belief that PCH has “no control” over the patient supply network. Also, in Loop B2, the higher patient volume results in the offer of medical clinics from PCH management to house the prospective high volume clinician. Such a facility inducement provides outpatients visiting their clinicians, with a “onestop service centre”. Such clinicians have full access to the diagnostic and pathological facilities at hospital. The proximity to hospital facilitates shortens waiting times for diagnostic procedures to be completed. The convenience offered to patients especially to geriatrics and patients with limited mobility increases patient satisfaction. The prompt service leads to higher referrals to clinicians with premises at PCH medical clinics from the goodwill and ‘word of mouth’ of satisfied patients.

Case Mix Capitation and Hospital Fees Hospitals that accept both case mix capitation and fee-for-service payments have contradictory performance incentives that make it difficult to determine optimal strategies. Case mix capitation promotes decreases in the frequency and intensity

514

of healthcare services, whereas fee-for-service rewards utilization increases with additional revenue. Consequently, hospitals’ profits suffer if they receive a portion of their revenues from capitation while maintaining fee-for-service as the primary payment mechanism. As the hospital’s percentage of gross revenue from capitation increases, the negative financial impact of misaligned incentives also increases. Private healthcare insurance organizations often stipulate that hospitals provide complete, “seamlessly” coordinated healthcare services across the continuum of care, cover an extensive geographic area, and be affiliated with a broad panel of primary care clinicians and specialists. To meet these requirements, many hospitals have expanded their service offerings in an attempt to create systems capable of delivering comprehensive, coordinated care. However, many “systems” actually are loosely aligned ambulatory, inpatient, home care services, and other business units that, although financially linked, do not offer truly coordinated care. A hospital, with a significant market presence (30 percent market share or more) with minimal excess capacity in the market, is in a strong negotiating position with private healthcare insurance organizations (Timothy et al., 2000). The combination of the hospital’s marketability with employers and the market’s capacity constraints will position the hospital as a “must have” in a private healthcare insurance organization’s network.

Preference for “No-Gap” in Fees In Australia, the CMB Schedule is a government benchmark reference for clinician consultation charges. Patients are concerned about “gaps” between fees and rebates. Most doctors would consider that equity in relation to the fees they consider as appropriate is synonymous with equity for patients in relation to their rebates. The Government expects reasonable efficiency from expenditure on health and reasonable assurance

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

that the fees charged by doctors will not leave patients facing substantial out-of-pocket costs. In Loop B3, the lesson for PCH management is that patients show a preference for hospitals with lower fees and that are fully covered (reimbursable) by private medical insurance. Hospitals that form alliances with private insurers to fully refund the cost of the accommodation charge thereby minimize any patient “out of pocket” cost component are more likely to meet the patients’ affordability and therefore increase patient satisfaction. The reinforcing loop, R1 in Figure 1, represents the typical clinician preference for the “no-gap” scheme launched by private medical funds such as Medibank Private, Health Benefits Fund (HBF) or Medical Benefits Fund (MBF). The scheme has gained popularity and the patient can proactively inquire at the private insurer’s website for such “no-gap” clinicians by the postal district and can request their GP’s to refer them to such clinicians. The market diversion and shift of patients is a factor, which increases the desire for clinicians to strengthen their alliances with the private healthcare insurance organizations. Reinforcing loop, R2 in Figure 1, represents the escalating effects resulting from market potential of PCH to accept or turn away patients either by virtue of capacity or operating theatre resources or availability of ‘state of art’ medical technology. Clinicians treat patients at PCH based on the choices that can be offered to patients by the hospital facility and therefore can opt to admit paying patients to other hospitals with better facilities. In addition the following other factors that we discovered through focus group investigations which were also supported in the literature and were developed as part of the integrated simulation model.

Co-Location and Service Partnerships In a hospital research case study in New York, Timothy et al. (2000) support the view that most hospital facilities have joined or formed a network to enhance market power and achieve efficiencies. Hospital executives were particularly concerned with the one-sided nature of their relationships with private healthcare organizations. They believed that these mergers and affiliations enable them to bring more leverage and more bargaining power when negotiating with these organizations. They obtained better prices for their services as a group. Co-location also enables hospitals to combine functional tasks such as purchasing, lab services, laundry and housekeeping. In most cases, these administrative efficiencies achieve cost savings almost immediately. Some hospital executives expressed the view that while consolidation was attractive, they feared losing patients who might choose to go to another facility due to a location change of the traditional facility.

Outsourcing Hospitals maintain several key ancillary facilities in a bid to maintain a “one-stop” service to clinicians and patients. However, under-utilization of such resources is a key decision factor in the consideration for the facility to be privatized or the whole service outsourced and managed by healthcare contractors. In many cases, the demand for services is too small to generate a market yet the pathological need is real. Similarly, contracts are awarded through competitive bidding for major purchases of medical supplies and pharmaceuticals. Specifications for the purchase of these products are presented and price solicitations are sought form vendors. The process strives to obtain the lowest price for products and reduce the possibility of kickbacks in the bidding process. Tight expenditure controls restrict the freedom to operate can reduce

515

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

significant opportunity costs to productivity and often focus strictly on the nominal value of the bid. In cases where professional services are being procured, procurement may be difficult in a competitive context due to the need to disqualify irresponsible or unreliable contractors (Lynch, 1995; Block, 1993).

Targeted Market Growth In order to build a new direction for the hospital, the focus group members have to realize and learn more about the private patient profile and its competitive value. Customers not only included patients but staff members, clinicians and private healthcare insurance organizations such as insurance companies and employers. In the communities surrounding the geography of the hospital, the fastest growing population segment was the age group over 55 years in age. Analysis of the case mix data showed that 59% of 1995 revenue for hospital under investigation came from inpatients above the age of 61 years from the outer suburbs. The losses came from the inner suburbs in the age groups between 19 and 47 years of age. Demographics of the patient population demonstrate the untapped market for the hospital in the primary and secondary catchments. According the recent census and with a population that is aging, those in the 50 – 60 age group are indeed projected to be the fastest growing cohort over the next 5 years and hospital’s target market.

MODEL EVALUATION AND VALIDATION The focus group participants from PCH were interviewed to investigate their perceptions about learning and applying the system dynamics (SD) models for decision-making. Issues investigated during interviews included: the understanding of the model design and its purpose; the value

516

of the visual user interface to the decision maker and its contribution to understanding the model structure; the decision maker’s learning curve with the visual interface; the impact of the visual interface on decision making; interpretation of the balanced scorecards, the benefits and the perceived advantages of simulation models with balanced scorecards (radar charts). A questionnaire, consisting of 40 questions, was developed to probe hospital managers’ views of SD-based visual interface models. The questionnaire was applied to the 5 focus group members and the 30 participants independently. All respondents were provided with the questionnaires two weeks prior to the interviews. The focus group was regarded as the reference group of modelbuilders’ as they were involved in all phases of model development, evaluation and analysis. The responses were analyzed in a descriptive fashion across four dimensions namely model characteristics, manipulation, evaluation and performance. The differences between the focus group members’ responses and the hospital managers’ responses were determined. The similarity in the responses helps build confidence that other managers within the hospital population held concurrent views as the focus group towards the interpretation of simulation results with balanced scorecards. The reference mode was fundamental to model structure evaluation - for the purposes of discovering patterns and model validation, analysis of simulated policies and scenarios. The model was tuned until it produced acceptable error rates in replicating historical financial performance, bed occupancies and seven other key performance indicators. The model was validated against two years of historical case mix data using actual patient arrival patterns. Clinicians and hospital administrators validated the model responses in the case study and compared the simulated results against the actual financial performance - called the Reference Mode.

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

However, it needs to be pointed out that occurrence of events in the past alone cannot be interpreted to mean that events in the future will follow accordingly. While both historical financial behaviour and a reference mode can be expressed in either quantitative or descriptive terms, a reference mode is essentially a qualitative and intuitive concept since it represents a pattern rather than a precise description of a series of events (Saeed, 1998). A reference mode also subsumes past history, extended experience and a future inferred from projecting the inter-related past trends based on the experience of the experts. Also, a reference mode will not contain random noise normally found in historical trends, as noise represents problem behaviour rather than the system behaviour. The caveat is with the exclusive use of historical case mix data to construct the reference mode.

BALANCED SCORECARD APPLICATION Kaplan and Norton (1992) found that neither financial performance measures nor operational performance measures alone provided senior managers in leading manufacturing firms the information they needed to operate complex organizations in globally competitive markets. Converting to a clinical setting, managers increasingly relied on what is best described as a comprehensive business picture of organizational effectiveness, including metrics for financial performance, market share, clinical outcomes, patient safety, customer satisfaction and community benefit. In their study of health care organizations and their use of integrated performance measurement systems, Chow et al. (1998) found that the design and implementation of a balanced scorecard is intimately intertwined with planning and control processes. Figure 2 illustrates the interrelationship between the design and use of a balanced scorecard and an organization’s planning and control

processes. In particular, this can be divided into four stages: • • • •

Translating the vision and gaining consensus; Communicating the objectives, setting the goals, and linking strategies; Setting targets, allocating resources, and establishing milestones; and Feedback and learning.

As Lockamy and Cox (1994) observe, a performance measurement system includes performance criteria, standards, and measures. Further, the performance criteria, standards, and measures of local business processes must be linked to other processes and also to the strategic objectives of the organization as a whole in order to support management’s decision-making needs.

HOSPITAL POLICY ANALYSIS AND DECISION MAKING In New Zealand, an elaborate study at Central Health concluded that the active co-operation of clinicians and other health professionals is required in the operation of case mix based accounting and control systems designed to monitor medical activity (Doolin, 1999). Managerial rationality underlies the notion that comparative information can be constructed as a management tool from which changes in the way clinicians manage their practice will and naturally follow. The appropriate levels of resource use that occur are often debated between managerial and clinical cultures (Parker, 1997). For example, managers at Central Health viewed the use of DRGs in their case mix from a funding and costing perspective, while clinicians tended to view the DRG framework as a medical classification system. The belief that peer review is influential in modifying clinician behaviour is well established

517

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

(Symes, 1992). However, in most cases of viewing case mix analysis, whether by financial manager, clinicians or healthcare professionals, only a static snapshot of the data has been considered in a specific time frame. The impact of simulating events and occurrences in future time periods with the aid of simulation technology using case mix data has not been published to date in the literature. This is the most significant paradigm shift, for clinicians and hospital managers, in analyzing case mix data and the major objective of this research study. Hospital executives involved in this research decided that strategic business decisions in a hospital that would significantly alter the trajectory of the hospital into “best clinical practices” and healthy surpluses could include following: •

•

•

518

Cost Reduction Policy- The aim here is to identify and consistently reduce the occurrences of high length of stay outliers and high cost DRGs and is of importance to the strategic financial management of hospitals. Although hospitals cannot refuse admission, they can redirect patients to appropriate institutions specializing in these high cost DRGs. It also requires stringent fiscal control policies, efficient manpower resource use monitoring systems and an increase in standardization practices across clinical, hotel services, pharmaceuticals, medical consumables and products. Market Growth and Intervention Policy - Negotiated schemes with clinicians and private insurers that would introduce an umbrella of private health insurance cover that reduces or eliminates the patient or payer gap. The strategy would be to increase market share, as most patients would have the ‘best of both worlds’, clinicians of their choice and no cash component or gap to pay for the treatment. Outsourcing or Co-location Policy – There are many options to consider with management of facilities and services in a private

hospital which could include sharing facilities (co-location) with another provider for pathology, radiology, surgical and other medical related services (Greco, 1996). For non-medical type services such as laundry and food catering, building and equipment maintenance, the outsourcing option is considered contracting and removes the burdens of unionized staff with workplace agreements. The opportunity for streamlining purchases from a single supplier who manages all the distribution and delivery to various hospitals can effectively cut costs and overheads through supply chain management (Greco, 1996). •

•

Clinical Pathways Policy – Clinical Pathways describes the usual way of providing multidisciplinary care for a particular type of patient and allows annotation of deviations from the norm for the purpose of improvement (Hindle, 1998). Clinical pathways can provide reasons for variance and empower management to redress the system. Analysis of individual clinician variance could possibly result in suggestions for changes in care, economic punitive action (such as withdrawal of accreditation), or demands that a clinical pathway be adhered to as a condition of future accreditation or agreement between clinician and hospital. With a simulation model, it possible to understand the “most cost effective clinical practice” within the hospital for a particular DRG profiled against the national benchmarks for that DRG. Clinicians’ Succession Policy – Clinicians and specialists who contract or commit their services to hospitals also negotiate utilization targets for the use of expensive hospital facilities such as operating theatres, a practice prevalent in the US. Based on these committed targets, the hospital management can forecast the revenue

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

bandwidth that a specialist would fit into in future production time periods. A clear understanding for revenue streams would allow for strategic resource allocation and facilities planning. Likewise, a specialist who is planning to retire or leave can be replaced with an equally reputed professional who is likely to bring the same level of commitment and revenue as the incumbent to the hospital. •

General Practitioner Network Policy – The General Practitioners (GP’s) form the ‘backbone’ of the patient supply chain in Australia. However, few GP’s have admitting rights to hospitals but the GP usually makes a recommendation or selection as to the clinician or specialist who could treat the patient based on their suspected aliment.

The GP network can be managed to ensure high hospital occupancy levels. Vertical integration into the GP network is a market supply policy issue that would affect long term planning decisions for hospitals.

POLICY EXPERIMENTS The simulation model is used to assess the effectiveness of each of the policies in isolation to determine the value of implementing such a policy in the organization. The effectiveness of the policy is gauged against the performance metrics discussed earlier in this paper. A strategic switch in the simulation model controls each policy and activation of the switch in the iThink Visual Interactive User Interface launches the implementation of the policy in the model.

Market Growth and GP Networks Policy High performance in the private sector patient market can be achieved by increased GP network alliance and consortium partnerships and the concurrent availability of PCH medical clinics to tap this market. The policy activates the PCH market potential for patient admissions by shifting the clinician’s market to the PCH hospital. An increase in the strength of the GP networks exhibits itself as increased patient growth. Before the policy switch is activated, the “Baseline Case” situation in Figure 3 is considered. Plots of actual patient admissions (Plot 1) and simulated patient admissions (Plot 2) have the same magnitude and pattern. However, in Figure 4, after the market growth policy implementation, Plot 2, the simulated admissions, peaks at week 60 with about 500 admissions per week compared to the base case average of 127 admissions. The market growth then declines, as the hospital capacity is not able to cope with the excessive growth, shown by the occupancy plot in Figure 6. So increases in the GP networking put an upper limit on patient admissions. Cash flow increases as patient arrivals increase and erodes with loss of admissions in week 60 as shown in Figure 5. Marginal costs as seen in Figure 7 are fairly consistent and on average have decreased from the base case of AU$1284 per patient to AU$1021 (consistent with economies of scale) indicating that there is some reduction in cost growth per patient as patient volume increases with the implementation of the market policy.

OUTSOURCING POLICY The fundamental assumption for this policy is to effectively control the hiring of healthcare contractors at an organizational level. PCH experts’ group hypothesized that there be many redundant

519

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 3. Market policy –base case

Figure 4. Market policy – increase in GP networking

contractors in the system as hiring decisions are currently made at a departmental level. The decision to hire or layoff contractors in the model is a global decision based on the total consumables cost and the costs of operating theatre and ancillary services such as diagnostic imaging and pathology provided to support patient care in the hospital. Price competitiveness and the desire for outsourcing services are based on the derived average cost per episode of these services. Figure 8 shows the level of contracting with high (Plots 2 & 4) and low (Plots 1 & 3) patient growth re-

520

spectively. The marked decrease in contractor use over a period of two years with the outsourcing policy in effect can be observed in Plots 2 & 4. The sharp increase in price competitiveness is observable in Plots 1 & 2 in Figure 9 with the increase in admissions volume, an indication that the cost of providing the service with full time salaried staff rather than with contractors becomes a viable and economically sensible option. The utilization gap (unused capacity) for the use of operating theatres and ancillary hospital facilities in Figure 10 shows a decrease with increased

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 5. Market policy – cash flow and mean cash balancest

Figure 6. Market policy – occupancy plot

Figure 7. Market policy – marginal costs plot

521

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 8. Outsourcing policy –level of contracting

patient admissions (Plots 1 & 2) as expected. This implies that Outsourcing Policy has not impact on the utilization of available hospital bed capacity. A sharp reduction in simulated overhead costs in Figure 11 (Plot 4) coincides with the reduction in contract workforce in Figure 8 (Plot 4). It can be seen that the outsourcing policy effectively curtails the engagement of contractors with high throughput of patients, as the price effectiveness factor regulates contractor use based Figure 9. Outsourcing policy – price competitiveness

522

on the average cost of service provision. And the converse is true when the price effectiveness decreases.

CLINICAL PATHWAYS POLICY The Clinical Pathways policy is the result of PCH management’s desire to reduce medical errors and deviations in clinical practice and also to reduce

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 10. Outsourcing policy – utilization gap

Figure 11. Outsourcing policy – overheads

clinicians with low case volumes. A simulation switch labelled ‘Clinical Governance Policy’ ensures the same patient volume to allow realistic experimentation of the effectiveness of Clinical Pathway (CP) policies. Each DRG or disease group has its own CP documenting the details of the recommended procedures, expected patient outcomes, pharmaceuticals, clinical investigations, prognosis monitoring and instrumentation for the surgical procedure. The formalization of CP in any hospital environment is a tedious and challenging implementation and

requires many hours of education, protocol explanation, monitoring and control against targets. There are benefits for such a long-term view as this clinical methodology introduces standardization in medical practice. In the simulation model, the desire to minimize clinicians with low practice volume is PCH management’s central objective. In applying the CP policy, it was decided that patient admissions be held constant throughout the experimentation with the CP variables such as the rate of CP implementation, the desire to minimize medical

523

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 12. Clinical pathways policy – net hospital costs variation

procedure deviations and the level of CP varied across the DRGs. The simulation results demonstrate (Figure 12) that hospital costs decrease with increased CP implementation. It brings with it a potential increase in capacity to treat more patients (deFigure 13. Clinical pathways policy – mean cp level

524

scribed by variations in occupancy) resulting from the overall reductions in length of stay (due to adherence to recommended CP procedures). The different levels of CP across the DRGs are shown in Figure 13. It is clear that the model represents PCH management’s desire to reduce

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 14. Clinical pathways policy – bed occupancy

Figure 15. Clinical pathways policy – variations in occupancy

clinical and treatment variances related to the divergent number of clinicians practicing at the hospital. Clinical treatment variations from a large number of clinicians dissipate the efforts of the clinical pathways policy to streamline clinical practices. The number of practicing clinicians for DRG G45B (Gastroscopy) was reduced to test this policy. A smaller and more manageable group of clinicians with a focus on implementing the CP system will bring better benefits to the organization.

It is also worth noting that the overall bed occupancy of the hospital peaked much earlier (week 79 in Plot 1, Figure 14) for the same volume of patients than the actual performance over two years. PCH management indicated that, from their experience, CP implementation actually has a regulatory effect on the use of shared resources within the hospital. The variations in bed occupancies enable the treatment of more patients, a side effect of overall shorter lengths of stay and more efficient clinical care, the desired effect from CP imple-

525

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 16. Clinical pathways policy – quality of care

Figure 17. Clinical pathways policy – hospital cash flow

mentation is shown in Figure 15 for each of the simulations. A notable and desirable consequence of the CP implementation is the albeit increase in quality of care, even though it is delayed until week 82 seen in Figure 16 on Plot 4. As observed the full implementation of CP is the result of the organizational and cultural change in the clinician

526

and nurses attitudes with greater due diligence in patient management and care. Finally, the CP implementation results in an increase in cash flow less the cost of CP implementation. A comparison of Plots 1 & 2 and Plots 3 & 4 separately in Figure 17 shows that the reductions in low volume clinicians for DRG G45B

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Figure 18. Radar chart of reference modes

did not have a significant impact on cash flow as their patient volume is low. In Figure 18, the balanced scorecards of the base case models are compared against the Actual Financial Performance of the hospital. The nine cardinal performance indicators are Occupancy, NLOS Gap (Gap in Available Bed days), NLOS Deviation (Deviation from National Average LOS), Marginal Costs, Cash flow per patient, Cash Balance per patient, Clinician Satisfaction, Patient Satisfaction and Quality of Care. A numerical summary of the output values are provided in Table 2 along with a comparison against

the actual performance metrics of the hospital over two years. The values of these indices in Table 2 were in different dimensions and these have been normalised with no dimensions and indexed against the Actual Financial Performance base value of 1 (-1 for NLOS Deviation) for the purposes of comparison in Figure 18. So, for example, the average cash balance (in Table 2) under the Actual Financial Performance, Average Cash Balance, (ACB) i.e. (ACB AFP) was $37, 56773 and in the simulated model (ACB ST) the value was $36, 44426. Thus the normalised value would be (ACB ST)/(ACB AFP) which would be 0.97 as plotted on the radar chart in Figure 18. High values of these indicators demonstrate their enhanced strategic value and contribution for all except the Marginal Costs and the NLOS Gap. For these two indicators, the smaller the value the better as it means that the actual gaps in available bed days and marginal costs are significantly lower than the national average. Lower marginal costs per patient and lower gaps in available bed days are highly desirable. In particular, these two indicators show significant differences in Figure 18 from the Actual Financial Performance in the radar chart. It indicates that the simulation models and their variants have better base case performance characteristics

Table 2. Numerical values of simulation outputs against actual performance measures PCH Performance Indicators Bed Occupancy (max 1) NLOS Available Beds Gap Deviation from NLOS Average Marginal Costs per Head (AU$)

Actual Financial Performance

Simulation model (Systems Thinking)

0.28

0.29

-97.08

-143.89

1.08

3.29

1738.20

2557.61

Sum of Cash flow (AU$)

894378

981605

Average of Cash Balances (AU$)

3756773

3644426

Clinician Satisfaction (max 1)

0.60

0.60

Patient Satisfaction (max 1)

0.65

0.56

Quality of Care (max 1)

0.70

0.60

527

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Table 3. Model policy variable settings for different scenarios (ST) Scenario Type/ Policy Market Growth Policy

Variable

Middle of the Road

‘Doldrums’

0.15

10% Increase to 0.165

0.15

10% Decrease to 0.135

Consortium Partnerships

0.2

10% Increase to 0.22

0.2

10% Decrease to 0.18

Market Growth Fraction

0.1

0.1

0.1

10% Decrease to 0.09

Policy Not Implemented

Implemented

Implemented

Implemented

Desire to Reduce Medical Deviations

0.11

Increase to 1 (max)

Increase to 1 (max)

Decrease to 0 (min)

Desire to reduce low volume clinicians

0.17

Increase for DRG G45B only to mean value of 7.76

Increase for DRG G45B only to mean value of 7.76

Decrease to 0 for DRG G45B only to mean value of 0.07

than the Actual Financial performance for the two fiscal years.

SCENARIO ANALYSIS Scenario analysis is applied by combining the implementation of a combination of market growth, outsourcing and clinical pathway policies. The simultaneous implementation of the 3 policies based on a combination of model input parameters as defined in Table 3 produced the scenarios namely the Base Case; Nightingale; Middle of the Road; Doldrums. The strength and influence of each policy (by fine-tuning simulation model parameters) was developed by the PCH hospital executive team such that the four classic scenarios namely Base Case, Nightingale (best case), Middle of the Road and Doldrums (worst case) were used to study the strategic combination of the above policies. The experimental evidence from these simulated scenarios suggests that a combination of market growth, clinical governance and managed contracting policies are necessary for business success. In the Nightingale scenario, the hospital experiences high patient growth with strong ele-

528

‘Nightingale’

Strength of GP Alliance

Outsourcing Policy Clinical Pathways Policy

Base Case

ments of partnerships with clinicians because of the implementation of clinical governance. The scenario shows that high cash flow can be expected in spite of both increased internal net hospital cost variations and reductions in simulated overheads from significant contractor workforce reduction and increased facilities utilization. But the efforts to sustain cash flow declines as capacity and workforce reductions are limiting factors. It is important to recognize that quality of care drops Figure 19. Radar chart of scenarios in Table 3

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

for some time due to excessive patient admissions in this scenario. The Middle of the Road scenario predicts that cash flow will fall below the base case if only clinical pathways and outsourcing polices is implemented without any serious efforts to improve patient supply. The Doldrums scenario is the worst case scenario and reports that cash flow will fall even below the Middle of the Road scenario level if PCH management opts to reduce the contractor workforce by implementing an outsourcing policy without commensurate actions to implement clinical pathways and increase market growth. The simulation (ST) model scenarios (1–4) developed are summarized in Figure 19, showing the effectiveness of the Nightingale scenario. This suggests that, by selectively implementing clinical pathways and reducing low volume clinicians, cash flow per patient rises. More significantly the normalized value of NLOS deviation increases. As a result of normalization (used for the purposes of comparison), a higher value is indicative of shorter lengths of stay. This shows that with the Nightingale strategies in place, there is greater alignment to the critical success factors that drive successful hospital performance. Clearly, the effect of increasing the pressure to manage staff and clinicians’ recruitment policies with greater management scrutiny has a significant impact on revenue with its goal seeking behaviour. The hospital has the responsibility of weighting these indicators. Preference for the Nightingale strategy is a management choice but the simulation model indicates performance above current levels.

IMPLICATIONS AND CONCLUSION The use of the balanced scorecard approach with radar charts enabled hospital executives to visually compare the simulation model outputs across all the nine performance metrics and subsequently

benchmark the strategies and scenarios that offered the best performance for the hospital. While the simulation models permitted detailed analysis of individual policy variables and behaviour over a two year horizon, it was useful to fine tune policy parameters to understand the influence of the variables and other parametric changes. The use of a balanced scorecard provided the snapshot of strategy performance, averaged over the 2-year simulation period, the hospital executives needed to understand the interaction of the various policies. The simulation model results presented in balanced scorecard format is an opportunity for “experimenting” with new ideas about hospital management policies without endangering their existing operations. Hospital managers could learn by trial and error the policies that would have a better success rate in this rapidly changing environment. It is a “no risk” tool for strategic thinking that most hospital managers, clinicians and administrators have little experience with. Qudrat-Ullah et al. (2007) report a research study for Changi General Hospital, Singapore where a balanced scorecard was developed and deployed. Researchers identified the need to have decision support systems that are sufficiently robust to incorporate a better understanding of the external environment (demand for services and the market competition) on the internal processes of the hospital. They also developed a broader understanding of organizational performance by identifying areas that had not received attention previously. Most notably these were procedures and methods for enhancing feedback and learning across several dimensions; for example, deferring patients with the consequent linkages to organizational performance indicators. The current research takes this further by linking the portrayal of a balanced scorecard with the outputs from a dynamic model of hospital operations. A similar use in a business management context has recently been made available (Bianchi, 2008) who quotes the originators of the Balanced

529

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Scorecard: “The Balanced Scorecard can be captured in a system dynamics model that provides a comprehensive, quantified model of a business’s value creation process” (Kaplan & Norton, 1996, p. 67) and “…dynamic systems simulation would be the ultimate expression of an organisation’s strategy and the perfect foundation for a Balanced Scorecard” (Norton, 2000, pp. 14-15). Balanced scorecards can be employed in a framework for learning how to think systemically and in particular to understand system feedback. It was clear that hospital managers typically took action based on direct impacts that were expected. But any action is likely to ricochet and sometimes even boomerang affecting other parts of the hospital with far-reaching implications. The methodology improves the understanding that actions can have indirect as well as direct consequences and there is a need to think systemically when taking action. There is also a need for collaboration to successfully achieve financial and clinical outcomes since the pressures of change tend to promote a focus on hospital managers’ selfinterest. The methodology offers the opportunity to understand surprisingly large differences in the “mental models” among executives, clinicians, nurses and administrators (Cavana et al., 1999). It generates a shared vision amongst participants, providing policy convergence and consensus.

REFERENCES Andersen, D. F., Richardson, G. P., & Vennix, J. A. M. (1997). Group model building: Adding more science to the craft. System Dynamics Review, 13(2), 187–201. doi:10.1002/ (SICI)1099-1727(199722)13:23.0.CO;2-O Australian Institute of Health and Welfare (AIHW). (1995). Australian Hospital Statistics 1995; Canberra. Retrieved from http://www. aihw.gov.au

530

Bianchi, C. (2008). Enhancing Strategy Design and Planning in Public Utilities through ‘Dynamic’ Balanced Scorecards: insights from a project in a city water company. System Dynamics Review, 24(2), 175–213. doi:10.1002/sdr.395 Block, P. (1993). Stewardship: Choosing Service over Self Interest. San Francisco, CA: BerretKoehler Publishers. Caldwell, C. (1995). Mentoring Strategic Change in Healthcare. Seattle: Quality Press. Cavana, R. Y., Davies, P. K., Robson, R. M., & Wilson, K. J. (1999). Drivers of quality in health services: different worldviews of clinicians and policy managers revealed. System Dynamics Review, 15(3), 331–340. doi:10.1002/ (SICI)1099-1727(199923)15:33.0.CO;2-G Chilingerian, J. (1992). New directions for hospital strategic management: The market for efficient care. Health Care Management Review, 17(4), 73–80. Chow, C. W., Garulin, D., Tekina, O., Haddad, K., & Williamson, J. (1998). The Balanced Scorecard: A potent tool for energizing and focusing healthcare organization management. Journal of Healthcare Management, 43, 263–280. Cisneros, R. J. (1998). Are providers constructing towers of Babel? Managed Healthcare, 21-23. Deloitte & Touche. (2000). U.S. Hospitals and the Future of Health Care Survey. Modern Healthcare, 30(26). Dickinson, R. A., Thomas, S. M., & Naughton, B. B. (1999). Rethinking specialist integration strategies. Journal of Healthcare Financial Management Association, 53(1), 42–47. Doolin, B. (1999). Case mix Management in a New Zealand Hospital: Rationalization and Resistance. Financial Accountability & Management, 15(3-4).

The Integration of Systems Dynamics and Balanced Scorecards in Strategic Healthcare Policy

Greco, J. (1996). Physicians heal themselves. The Journal of Business Strategy, 17(6). doi:10.1108/ eb039817 Health Department of Western Australia. (1998). Annual Report 1997-1998. Certification of Performance Indicators. Hernandez, S. R. (2000). Horizontal and Vertical Healthcare Integration: Lessons Learned from the United States. Healthcare Papers, 1(1). Hindle, D. (1998). Classifying the care needs and services received by Home and Community Care (HACC) clients. Aged and Community Care Service Development and Evaluation Reports. Kaplan, R., & Norton, D. (1992, January). The balanced scorecard-measures that drive performance. Harvard Business Review, 71–79. Kaplan, R., & Norton, D. (1996). Linking the Balanced Scorecard to Strategy. California Management Review, 39(1), 53–79. Lockamy, A., & Cox, J. F. (1994). Reengineering performance measurement: How to align systems to improve processes, products, and profits. Burr Ridge, IL: Irwin. Lynch, T. D. (1995). Public Budgeting in America. Upper Saddle River, NJ: Prentice Hall. Martin, S., & Smith, P. (1995). Modeling waiting times for elective surgery (Tech. Rep.). York, UK: University of York, Center for Health Economics. Norton, D. (2000). Is Management finally ready for the ‘Systems’Approach? Balanced Scorecard Report.

Parker, M. (1997). Dividing Organizations and Multiplying Identities. In Hetherington, K., & Munro, R. (Eds.), Ideas of Difference (pp. 112–136). Oxford, UK: Blackwell. Qudrat-Ullah, H., Chow, C., & Goh, M. (2007). Towards a dynamic balanced scorecard approach: the case of Changi General Hospital in Singapore. Int. J. Enterprise Network Management, 1(3), 230–237. doi:10.1504/IJENM.2007.012756 Richmond, B., & Peterson, S. (1997). An Introduction to Systems Thinking: IThink Analysis Software. Hanover, NH: High Performance Systems Inc. Saeed, K. (1998). Defining a problem or constructing a reference mode. In Proceedings of the 16th System Dynamics Conference, Quebec, Canada. Symes, D. (1992). Resource Management in the National Health Service. In Pollitt, C., & Harrison, S. (Eds.), Handbook of Public Services Management (pp. 194–204). Oxford, UK: Blackwell. Timothy, S. P., Haslanger, K., Delia, D., Fass, S., Salit, S., & Cantor, J. C. (2000). Hospital Markets, Policy Change and Access to Care for Low-Income Populations in New York. New York: Rutgers. Vissers, J., & Beech, R. (2005). Health Operations Management. London: Routledge.

ENDNOTE 1



PCH is an acronym for the hospital participating in the study

Nunamaker, J. F., Chen, M., & Purdin, T. D. M. (1990). System development in information systems research. Journal of Management Information Systems, 7(3), 89–106. This work was previously published in International Journal of Healthcare Delivery Reform Initiatives (IJHDRI), Volume 2, Issue 2, edited by Matthew W. Guah, pp. 10-34, copyright 2010 by IGI Publishing (an imprint of IGI Global).

531

532

Chapter 2.20

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems and Ubiquitous Computing Devices Upon Health Professional Work Elizabeth M. Borycki University of Victoria, Canada Andre W. Kushniruk University of Victoria, Canada

ABSTRACT Health information systems, and in particular ubiquitous computing devices (UCD), promise to revolutionize healthcare. However, before this can be widely achieved UCD need to be adapted to fit the information, workflow and cognitive needs of users of such devices. Indeed systems and devices that are not developed appropriately may inadvertently introduce error in healthcare (“technology-induced error”). This chapter deDOI: 10.4018/978-1-60960-561-2.ch220

scribes an approach to applying clinical simulations to evaluate the impact of health information systems and ubiquitous computing devices on health professional work. The approach allows for an assessment of “cognitive-socio-technical fit” and the ability to modify and improve systems and devices before they are released into widespread use. The application of realistic clinical simulations is detailed, including the stages of development of such simulations (from the creation of representative clinical environments to subject selection and data collection approaches). In order

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

to ensure the success and widespread adoption of UCD, it is argued that greater emphasis will need to be placed on ensuring such systems and devices have a high degree of fit with user’s cognitive and work processes.

INTRODUCTION Health information systems (HIS) and in particular ubiquitous computing devices (UCD) have promised to revolutionize healthcare. Early research in this area demonstrated the value of integrating varying HIS and UCD into clinical settings in terms of improving the quality of patient care and reducing medical error rates (e.g. Bates et al., 1999; Chaudry et al., 2006). More recent research has called into question some of these findings. Some researchers have identified that poor cognitive, social and technical fit may influence some implementations of HIS and UCD and be the cause of health professional adoption and appropriation failures (Borycki et al., 2009a). In healthcare the hospital and clinic costs associated with implementing and re-implementing a HIS and UCD can be significant. There is a need to develop new evaluation methodologies that allow for testing of differing constellations or groupings of HIS and UCD prior to their implementation in clinical settings to reduce the likelihood of costs being associated with health professional adoption and appropriation failure. The authors of this book chapter will introduce a novel methodology that can be used to evaluate the cognitive-socio-technical fit of HISs and UCD prior to their implementation in real world clinical settings. This may prevent poor adoption and appropriation of these technologies. The authors will begin this chapter by first introducing the reader to the literature involving HIS and UCD successes and failures. This will be followed by a brief introduction to the theory of cognitive–sociotechnical fit as applied to HIS and UCD in clinical settings. Following this, the authors of the book

chapter will describe the new and emerging area of HIS evaluation involving the use of clinical simulations. This section of the book chapter will not only describe the background and rationale for using clinical simulations to evaluate HIS and UCD, but will also describe the steps in the evaluation method and provide examples from the authors’ work. The authors will use examples from their previous research to illustrate the use of clinical simulations as an evaluation methodology for HIS and UCD. The application of clinical simulations to the evaluation of HIS and UCD was pioneered by the authors and builds upon their previous work in the areas of clinical simulation and usability engineering (e.g. Borycki et al., 2009b; Kushniruk et al., 1992). Clinical simulations as applied to HIS and UCD evaluation in health informatics have their origins in the medical education and usability engineering literatures. Clinical simulations, when applied to the evaluation of a technology(ies), borrow from medical education, where clinical simulations are used to train physicians and evaluate their competency as health care practitioners (Issenberg et al., 1999). The author’s work also extends the usability literature to clinical simulation by using many of the methods and analysis approaches employed by usability engineers to collect, record and analyze data involving HIS and UCD. Unlike the clinical simulation and usability literatures, clinical simulation when applied to the evaluation of HIS and UCD involves a more holistic evaluation of technology’s impact upon the cognitive-socio-technical fit of health professionals’ work. Lastly, this work also adds to the scientific knowledge in health informatics by identifying a new methodological approach that can be used by health informaticians to evaluate cognitive-socio-technical fit prior to HIS and UCD release and implementation. This approach has not been described elsewhere in the mainstream health informatics evaluation literature (e.g. Anderson & Aydin, 2005; Friedman & Wyatt, 2006; Shortliffe & Cimino, 2006) and is emergent in nature.

533

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

UNDERSTANDING THE NEED FOR APPLYING CLINICAL SIMULATIONS TO THE EVALUATON OF HEALTH INFORMATION SYSTEMS AND UBIQUITOUS COMPUTING DEVICES The published research literature demonstrating the ability of HIS and UCD to improve patient and health care system processes and outcomes has been mixed. For example, Chaudry et al. (2006) in his systematic review of the health informatics literature found that HIS in conjunction with varying devices: (a) improved physician use of guidelines, (b) monitoring and surveillance of patient disease, and (c) decreased medication error rates (but only in healthcare organizations where home grown systems were being used). There were no proven improvements in the quality of patient care associated with the use of commercial HIS. As well, Chaudry and colleagues found that research on the cost-effectiveness of HIS/UCDs was limited and the findings that had been published in the literature were inconclusive. Eslami and colleagues in their 2007 and 2008 systematic reviews of the literature had similar findings. Eslami et al. (2007; 2008) found that implementing a computerized physician order system (i.e. the ordering of medications via computer) increased physician adherence to guidelines in outpatient settings, but that there was a lack of evidence to support the system’s ability to improve patient safety and reduce the costs of providing patient care in outpatient care settings. Lastly, Ammenwerth et al. in her 2008 systematic review found that HIS/UCD may reduce the risk of medical errors when used by physicians, but at the same time acknowledged there is a need for more research in this area as the quality of the studies published to date are variable. Other researchers found that some types of HIS and UCD may facilitate medical errors (e.g. Koppel et al. 2005). Koppel et al. (2005) found HIS system features and methods of implementing could facilitate medical errors in real-world

534

settings if not designed and implemented properly. Kushniruk et al. (2005) suggested the interaction between HIS and UCD (i.e. handheld devices) could potentially lead to technology-induced errors (e.g. where interface design features may induce users to unknowingly prescribe medications incorrectly and UCD features may influence decision making). Lastly, Schulman et. al. (2005) found that while some types of errors were reduced by HIS, new types of errors were also introduced or emerged with the introduction of the technology. This has lead some researchers (e.g. Borycki & Kushniruk, 2008; Koppel & Kreda, 2009) and healthcare organizations (e.g. The Joint Commission for Health Care Quality, 2008) to suggest there is a need to exercise caution and apply rigorous evaluation when implementing HIS and UCD in healthcare organizations. In response to these suggestions researchers have begun to call for the evaluation of differing constellations of HIS and UCD cognitive-socio-technical fit to prevent medical errors involving patients and health professionals from occurring (Borycki & Kushniruk, 2005; Borycki et al., 2009a). As Berg (1999), Kushniruk et al. (1992) and Koppel et al. (2005) suggest improving this fit may reduce the likelihood of unintended consequences. When a healthcare organization (e.g. hospital or clinic) identifies that a HIS and UCD affects the quality of patient care, the organization’s leadership typically tries to address these HIS and UCD issues (Koppel et al., 2005). However, there are high costs associated with modifying a HIS and/or replacing a UCD after a hospital or clinic has implemented it. Yet, in healthcare, HISs and/or UCDs often require modification and/or reimplementation in order to address the above mentioned issues and improve the cognitive-sociotechnical fit of the system with health professional work (i.e. physician, nurse and other allied health professional work) (Ash et al., 2004; Borycki et al., 2009b). In some cases, where systems have not achieved full cognitive-socio-technical fit with health professional work, physicians and

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

nurses have chosen to boycott the HIS and/ or UCD (LaPoint & Rivard, 2005; LaPoint & Rivard, 2006). In other cases health professionals have chosen to not use many of the HIS and UCD features and functions that would have lead to significant reductions in medical errors or improvements in patient outcomes (Granlien et al., 2008; Kushniruk et. al., 2006). As a result, some organizations have “turned off” HISs and/ or removed some types of UCD from the clinical setting, instead attempting to redesign HIS or select new UCD and then re-implement them at a later date and time (thereby improving fit and health professional adoption and appropriation) (Ash et. al., 2004). Some health informaticians have suggested there is a 65% failure rate associated with implementing HIS and their associated devices (Rabinowitz et al., 1999). Others have suggested this failure rate may be higher when one considers the number of HIS and/or UCD that are not used to their fullest extent (i.e. in terms of their features and functions) (Granlien et al., 2008). Therefore, there is a need for new approaches that can be used to evaluate the cognitive-socio-technical fit of HIS and UCD. As well, poor software and hardware designs (including UCD), and poor implementations can lead to significant healthcare organizational costs (e.g. hospital or clinic) (LaPoint & Rivard, 2005; LaPoint & Rivard, 2006). Costs arise from HIS/UCD device removal, redesign and/ or reimplementation (including costs associated with retraining health professional staff) when such failures occur. In a time where there is a scarcity of healthcare resources (i.e. financial and human) to provide patient care such expenditures can have a significant impact upon regional health authority and federal healthcare budgets. This has led some policy makers and researchers to also conclude that there is a need for new evaluation methods that can be used to inform systems/UCD design and implementation prior to deployment in order to reduce the need for costly redesign and

reimplementation (Ash et al., 2004; Borycki et al., 2009a; Borycki & Kushniruk, 2006).

THE THEORY OF COGNITIVESOCIO-TECHNICAL FIT, HEALTH INFORMATION SYSTEMS AND UBIQUITOUS COMPUTING DEVICES Assessing cognitive-socio-technical fit involves examining the degree of integration between cognitive, social and technical aspects of work. The effects of poor cognitive-socio-technical fit are often significant. The process of evaluating cognitive-socio-technical fit involves taking an integrated, holistic approach towards understanding how differing constellations of HIS and UCD interact in healthcare settings and identifying those HIS and UCD constellations that best match work processes. Cognitive-socio-technical fit addresses the cognitive aspects of fit involving health professionals’ cognition, the social aspects of fit including social interactions that take place between health professional, patient and technology and the use of the technology to provide patient care. Research suggests poor cognitive-socio-technical fit often leads to unintended consequences that were not expected by either the technology’s designer or the organization that procured and implemented the HIS and/or UCD (Borycki et al., 2009a). Poor cognitive-socio-technical fit may result in health professional process and outcomes changes. Process changes include those changes that alter how work is done when a new technology(ies) is implemented. As outcome changes arise from process changes (McLaughlin & Kaluzny, 2006), outcome changes include those changes that affect a patient’s health and wellness and are as a direct result of interactions between the health professional, the HIS/UCD that supports work and the social systems of the organization (e.g. hospital or clinic) (Borycki et al., 2009a). Research has found cognitive-socio-technical fit affects processes and has led to health profes-

535

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

sionals bypassing some HIS and UCD functions (Kuwata et. al., 2006; Patterson et al., 2002) or experiencing increased cognitive load (Borycki et. al. 2009b; Shachuk et al., 2009), and/or experiencing cumbersome workflows (Kuwata et al., 2006; Patterson et al., 2002). For example, research has shown that that introducing a computer workstation into a physician’s office affects the nature of physician-patient communication and physician workflow (Ludwick & Doucette, 2009; Shachuk et al., 2009). Ludwick and Doucette (2009) found that the positioning of a computer workstation in typical physician office led some physicians to modify their workflow and in some cases affected their ability to conduct sensitive communications about health issues with their patients. In such cases, the physicians chose to not document in the electronic patient record on the workstation in the examination room, instead choosing to document on a paper patient record. In another study Shachuk et al., (2009, p.345) found that 92% of physicians found that having an electronic medical record in the exam room “disturbed” their communication with patients in some way. In order to overcome some of these issues physicians developed new skills (i.e. used blind typing) or used templates to improve the workflow that emerged from interactions between the workstation, electronic medical record and the patient. Shachuk et al. (2009) also noted the positioning of a computer workstation in a clinic setting (i.e. social context) influenced workflow and increased cognitive work. These findings are not unique to the physician office setting. Poor workflows emerging from interactions between UCD-HIS and organizational context (i.e. environment) are also present in the hospital setting. For example, wireless carts that provide ubiquitous access for health professionals to electronic patient records may provide timely patient information, but when used to provide certain types of patient care (i.e. medication administration), they may lead health professionals to develop unplanned workarounds. Publications by Granlien et al. (2008), Kushniruk et al. (2006),

536

Kuwata et al., (2005) and Patterson et al., (2004) reveal wireless carts in conjunction with medication administration systems (i.e. systems that support nurse administration and documentation of medications administered to patients), bar code scanners, patients, and the hospital environment may lead clinicians to undertake significant workarounds and bypassing of some system functions in order to provide timely patient care. Each of these studies has identified that there is a need to better understand the cognitive-socio-technical fit between the worker and system/UCD in order to optimize fit and improve adoption and appropriation of these technologies (Borycki et al., 2009a; Granlien et al., 2008).

Cognitive-Socio-Technical Fit and Healthcare Processes and Outcomes The theory of cognitive-socio-technical fit involving HIS and UCD influences healthcare processes and subsequent outcomes. According to the healthcare quality improvement literature, process improvements may lead to significant improvements in outcomes (McLaughlin & Kaluzny, 2006). Along these lines, process changes may also lead to workarounds or bypassing of system functions (if not effectively changed), thereby leading to poor or unintended outcomes such as no reductions in health professional error rates or improvements in patient health status) (Koppel et al., 2005; Kushniruk et al., 2005). Healthcare process changes arising from poor cognitivesocio-technical fit include the development of new unintended interactions (not expected by the technology’s designer nor the adopting organization) between a HIS, the UCD and the health professional that in turn influences patient outcomes. For example, in some cases HIS have failed to reduce medication error rates projected by procuring organizations (Koppel et al., 2005). In other cases researchers have reported that the attributes of UCD (e.g. small screen size) have served as a barrier to clinicians use of the UCD

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

(e.g. a handheld device, cellular phone) in the clinical practice setting (Chen al., 2004). Lastly, still other researchers have found that interactions between HIS and UCD can introduce a new class of “unintended” consequences (Ash et al., 2004) such as technology-induced errors (Kushniruk et al., 2005). These unintended consequences have their origins in the interactions between the HIS and the UCD within a given healthcare organizational context. In these cases HIS and UCD have introduced new types of errors to the organization. Such instances of poor cognitive-socio-technical fit need to be identified and addressed to prevent harm from occurring to patients prior to HIS and UCD implementation in real-world settings (Borycki & Kushniruk, 2008). In summary poor-cognitive-socio-technical fit involving HIS, UCD, the healthcare organization and the patient can have significant impact upon healthcare processes and patient outcomes. The substantive costs associated with addressing “unintended” process and outcome changes arising from poor cognitive-socio-technical fit following the introduction of HIS and UCDs in healthcare requires that researchers develop new methods of identifying potential process and outcome changes before a HIS/UCD is implemented to prevent unintended consequences from occurring. According to Boehm’s seminal research (1976), the relative costs of repairing software errors increase exponentially the further the software product moves through the design, coding, testing and implementation phases of the software development lifecycle (SDLC). Although this work focuses on software products, the quality improvement literature in healthcare identifies introducing new processes using HIS and UCD can also lead to significant costs - costs associated with modifying the HIS to meet UCD specifications or the need to identify a new UCD that has high cognitivesocio-technical fit with health professional work (McLaughlin & Kaluzny, 2006). Lastly, the adverse events literature in healthcare identifies the human cost of death and disability arising from software

and UCD failure during medical treatment (Levenson, 1995). In summary it is important to detect unintended consequences associated with using HIS and UCD early in the SDLC, rather than during the operations and maintenance phase (i.e. when HIS and UCD are implemented in a hospital or clinic) and they may lead to medical errors. In the next section of the book chapter the authors will discuss the use of clinical simulations, a methodology for evaluating the impact of HIS and UCD upon health professional work.

CLINICAL SIMULATIONS: A METHODOLOGY FOR EVALUATING THE IMPACT OF HEALTH INFORMATION SYSTEMS AND UBIQUITOUS COMPUTING DEVICES UPON HEALTH PROFESSIONAL WORK One emerging methodology that may be an effective approach to evaluating the impact of a HIS and UCD upon health professional work processes and patient outcomes is clinical simulation as applied to the evaluation of HIS and UCD in healthcare. Clinical simulations involve the use of real-world examples of complex clinical work activities involving health professionals (e.g. doctors or nurses) being observed as they carry out a work task involving a HIS and UCDs in a hospital or clinic. In the next section of this book chapter, the authors will describe the development and history of clinical simulations in healthcare, their subsequent application and use in health informatics in the evaluation of HIS and UCD, and their use in detecting “unintended” consequences associated with poor cognitive-socio-technical fit.

History of the Use of Clinical Simulations is Healthcare Clinical simulations have been used in the healthcare industry for many years. Several researchers

537

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

have successfully developed and utilized clinical simulations as a way of preparing medical and nursing students for real-world practice. Initially, instituted as a method for providing clinicians with opportunities to practice clinical skills and procedures in preparation for work involving live patients, clinical simulation has become an accepted methodology for training physicians and nurses prior to their real-world encounters with patients (Issenberg et al. 2005; Kneebone et al., 2005). Research examining the effectiveness of clinical simulation as a training methodology has shown that student participation in clinical simulations can improve the quality of their clinical decision-making and practice. Clinical simulation can also provide opportunities for students to develop expertise in a given clinical area as well as learn about how to manage a patient’s states of health and illness in a safe environment where there are no live patients and instructors can provide constructive feedback (Issenberg et al., 1999; Issenberg et al., 2005; Kneebone et al., 2005). Clinical simulations may take many forms. For example, a clinical simulation may involve live actors who are “playing” the part of an ill patient – acting out the signs and symptoms of a disease such as pain or melancholy. In such clinical simulations health professional students such as medical students have the opportunity to practice skills such as patient interviewing while an instructor observes and later provides feedback. Instructors review the medical students’ performance and provide feedback aimed at improving the health professionals’ diagnostic and procedural skills. Simulations may also take the form of interactions with a mannequin (that represents a patient) in a laboratory setting. The health professional has the opportunity to perform procedures on the mannequin that would typically be done on a patient in the clinical setting. Procedures are practiced until the health professional achieves competence in performing the activity, skill or procedure (Issenberg et al., 2005; Kneebone et al., 2005).

538

Clinical Simulations and Health Professional Education In the laboratory setting instructors can observe students applying theory to practice using such clinical simulations. For example, an instructor can observe a student performing an activity or a procedure that was discussed in the classroom. Once the student has completed the activity or procedure, the instructor can provide the student with constructive feedback and thereby help the student to achieve a higher level of knowledge and improve the student’s ability to perform the activity or procedure. More recently, the tools used in training health professionals (e.g. physicians and nurses) have improved significantly. Instead of using unresponsive mannequins’, health professional training programs are using computerized mannequins and actors to help build clinician skills. Mannequins have been developed that can simulate real patient signs and symptoms of disease over time. Furthermore, these computerized mannequins can be used to train health professionals to respond to life critical events (such as heart attacks) and to perform complex procedures within the context of simulated, life threatening events. This gives the health professional student an opportunity to experience events that are more representative of the real-world and an opportunity to learn about how to respond to these events over time as the patient’s condition deteriorates or improves in response to student interventions. Once competencies are achieved, health professionals can move to assessing actors playing the role of a patient. Following this students can learn to manage real patients and perform procedures on real patients in real-world clinical settings under the supervision of a health professional (having practiced these activities on a mannequin and with actors) (Cioffi, 1997; Issenberg et al., 2005).

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

APPLICATION OF CLINICAL SIMULATIONS TO THE EVALUATION OF HEALTH INFORMATION SYSTEMS AND UBIQUITOUS COMPUTING DEVICES Researchers have begun to apply key learning’s gained from clinical simulation research involving medical and nursing students to the testing and evaluation of HIS and UCDs prior to their implementation in real-world clinical settings. Initial research in this area has employed clinical simulations as a methodology for evaluating the impact of systems and UCDs upon physician cognition and work. Early research in this area took place in the 1990’s. Cognitive researchers attempted to determine what effect a HIS deployed on a desktop computer would have upon physician-patient interviews and a physician’s cognition. The researchers simulated a typical physician office and asked physician subjects to conduct an interview with an actor, while using the HIS. The researchers found the introduction of a HIS had significant cognitive implications for physicians, as the HIS altered the nature of the patient-physician interview, caused some physicians to become “screen driven” (Kushniruk et al., 1992) and in some cases altered diagnostic reasoning processes (Patel et al., 2000). More recent research has extended this work to evaluate the impact of HIS deployed on differing UCDs (i.e. hand held UCDs). In one line of research, researchers asked physician subjects to perform a number of prescribing tasks in a simulated clinical environment using a prescription writing system on a personal digital assistant (PDA). The researchers studied the effects of the HIS/UCD upon physician prescribing error rates. In the clinical simulation the researchers were able to observe interactions between a HIS and a hand held computing UCD upon physician prescribing and subsequent medical error rates. The researchers observed that HIS and UCD usability problems led to technology-induced errors.

They were able to identify a number of hand held UCD and HIS interface features that could lead to errors. In addition to this, the researchers noted that the size of the UCD and its responsiveness influenced subject error rates, demonstrating an interaction effect between the HIS and UCD. This research was successful in predicting differing types of technology-induced errors prior to HIS/UCD deployment and implementation (Kushniruk et al., 2005). In a subsequent study the researchers extended their work to additional health professionals (i.e. nurses and physicians) to evaluate the use of a mobile wireless medication cart with a laptop and a barcode scanner. Subjects were asked to administer medications (i.e. oral medications, injectible medications and intravenous medications) to a mannequin who was wearing a bar-coded bracelet for identification. In this study the researchers again conducted a clinical simulation where health professionals were asked to administer the medications and document their administration on a HIS using a laptop on a mobile cart. The researchers noted the interactions between the health professional, HIS, UCDs (i.e. laptop and bar code scanner) and the mannequin (who took the place of the patient). The researchers observed that the medication administration task involving the UCD and system in the clinical environment led to the development of cumbersome workflows involving the UCD, system and patient. In addition to this, the researchers found the interactions between the subject, UCD, system and patient could lead to increased cognitive load. In some cases subjects engaged in cumbersome and confusing workarounds in order to be able to successfully complete the task of medication administration (including documentation) (Borycki et al., 2009a; Kushniruk et. al. 2006). The researchers were able to provide the healthcare organization that was implementing the HIS/UCDs with feedback about how the system/UCDs could be modified, how to simplify some of the emergent workflows arising from using systems/UCDs and how to improve

539

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

health professional training so that health professionals spent less time learning about how to use the integrated system (Kushniruk et al., 2006). In summary the use of clinical simulations in the evaluation of interactions between health professionals, HIS, UCDs and patients has demonstrated the value of using clinical simulations to improve health professional work processes while at the same time identifying key ways outcomes can be improved when using HISs/UCDs (i.e. reducing the likelihood of technology-induced errors or unintended consequences). In the next section of this chapter the authors will provide an overview of the clinical simulation methodology.

The Clinical Simulation Methodology In order for clinical simulations to be effectively used in evaluating the process and outcome impacts of implementing a HIS and its associated UCDs, researchers need to ensure that clinical simulations have a high level of ecological validity. Ecological validity refers to the ability to create clinical simulations that are representative of the real-world settings so that the results of the clinical simulations are generalizable (Vogt, 1999). Therefore, ecological validity ensures the quality and value of the clinical simulation results when they are used to: (1) redesign a system/UCD, (2) procure a system, (3) understand the effects of system customization in conjunction with UCD selection and (4) as an aid to providing technology trainers with information about how to train individuals to use such systems/devices to increase the speed of learning how to use the HIS/UCD in a given clinical setting. In the next section of the chapter we will discuss a sequence of steps that can be used to create clinical simulations that can test aspects of cognitive-socio-technical fit in more detail. This involves creating representative clinical environments (that would typically be found in healthcare organizations where the HIS/UCD would be designed for or deployed). This involves: (1) using representative equipment

540

from that organization, (2) selecting a range of representative tasks health professionals would typically engage in, (3) developing a range of representative scenarios, and (4) inviting a range of representative health professional subjects to participate in the clinical simulations as part of the evaluation.

Creating Representative Clinical Environments The complexity of representative clinical environments required for clinical simulations in health informatics where a HIS/UCD is being evaluated varies from low to high fidelity. A low fidelity clinical environment somewhat approximates the real world. In a high fidelity clinical simulation every attempt is made to imitate the real world. In creating high or low fidelity environments there is a need to attend to ecological validity (as outline earlier). As part of creating an ecologically valid environment, the essential features of a real-world environment would need to be re-created so that the subject that is participating in the clinical simulation responds as they would in the realworld when confronted with a similar situation. Therefore, creating a representative organizational or clinical environment is critical to ensuring the ecological validity of the clinical simulation and the validity and value of the results. High ecological validity will lead to representative results in terms of observed impacts of systems and UCDs upon health professional processes and outcomes where as low ecological validity will not.

Tangible and Intangible Aspects of Creating Representative Clinical Environments Clinical environments (e.g. hospital rooms, nursing stations) have both tangible and intangible components. Tangible aspects of organizations include those visible structures that can be easily replicated such as the physical layout of a hospital

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

room, desks, chairs, hospital beds, intravenous equipment, oxygen equipment etc. Intangible aspects of an organization include those that are not readily visible such as underlying organizational policies, procedures, social norms and culture. Intangible aspects of the clinical environment are difficult to identify and replicate, yet they have a significant impact upon how work is done within an organizational context (Shortell & Kaluzny, 2006). Tangible aspects of a clinical environment should be evaluated and judged by experts to be representative of typical clinical settings to ensure the setting’s ecological validity (Haimson & Elfenbein, 1985; Shortell & Kaluzny, 2006). Experts with clinical, management and health information technology backgrounds should be used to evaluate the ecological validity of a clinical environment in conjunction with health informaticians (Shamian, 1988). Clinical and management experts can be used to judge the representativeness of the simulated clinical environment in terms of layout, furniture and equipment. Information technology and health informatics experts can be used to judge the representativeness of the systems/UCDs that will be used or are currently being used in a “typical” clinical environment. These experts will also be expected to select the “ideal” or planned UCDs for use in the clinical simulation and integrating the system/UCDs for the simulation. For example, if the intent of a clinical simulation is to re-create a representative hospital environment that is “typical”(where the system/UCD would be implemented) of a hospital room with healthcare equipment, this can be done in a laboratory setting. The results of this clinical simulation would be generalizable to most typical hospital environments, but may not be typical of the organization where the HIS/UCDs would actually be implemented (Borycki et al., 2009b; Haimson & Elfenbein, 1985). Alternatively, if the intent of the clinical simulation is to create a representative hospital environment that is ecologically valid and to ensure the results of the clinical simulation are representative

of the local healthcare organization (e.g. hospital) every effort should be taken to conduct the clinical simulation in the local organization (Haimson & Elfenbein, 1985; Kushniruk & Borycki, 2005; Shortell & Kaluzny, 2006). For example, if a hospital is implementing a specific HIS/UCD that has been customized for the hospital, an empty hospital room can be used with the HIS/UCD to be tested (Kushniruk & Borycki, 2005). As a result, both the tangible and intangible aspects of the organizational environment are replicated in using an empty hospital room within the local organization. In conducting the clinical simulation in the organizational environment where the HIS/UCD will be deployed there will be full replication of the tangible aspects of the organization in the clinical simulation (e.g. room layout, hospital equipment etc.). As well, the intangible aspects of the organizational environment will also be present in the clinical simulation (e.g. organizational culture, social norms in behaviour etc.). The health professionals who participate in the simulation within the context of their own local organization may engage in those organizational and practice norms that underlie all health professional practice and are unique to that organizational context. It must be noted that if a health professional participates in the clinical simulation and they are not familiar with the intangible aspects of the organization they may not practice according to local organizational practices and thereby influence the quality of the clinical simulation results for the local organization. Alternatively, if the health professional participates in a clinical simulation outside their organization (i.e. in a laboratory setting) they may not use the same organizational cultural and behavioural norms. This will also influence the quality of the clinical simulation results (Haimson & Elfenbein, 1985; Shortell & Kaluzny, 2006). Therefore, ecological validity is necessary in order to ensure the process and outcomes changes that are observed during the clinical simulation are representative of those involving the HIS/UCD. Furthermore, there is a

541

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

need to attend to the tangible and intangible aspects of an organization. Tangible and intangible aspects of organizations may influence health professional activities, practices and behaviours during a clinical simulation and therefore, the quality of the simulation outcomes are affected.

Use of Representative Equipment The complexity of the equipment used in clinical simulations varies depending on whether the clinical simulation is low or high fidelity. As a low fidelity clinical simulation roughly approximates the real-world, the equipment is often not complex and may be limited to only the HIS/UCDs that are to be evaluated. Alternatively, as we increase the fidelity of the clinical simulation the equipment that is added to the clinical simulation is more representative of the clinical environment. For example, a clinical simulation in an empty hospital room would also include the use of equipment that would typically be used by health professionals in managing a patient’s care such as a hospital bed, bedside table, IV poles, oxygen equipment etc. As well, the HIS/UCD that is planned for use would also be used in the clinical simulation environment. This might include the use of a computer workstation placed beside the patient’s bed with a copy of the patient’s electronic patient record available for viewing during the clinical simulation (Kushniruk et al., 2006; Kuwata et al., 2005).

Use of Representative Patients and Other Health Professionals The use of representative patients and other health professionals will also range from low to high fidelity. In a low fidelity clinical simulation a mannequin may be used to represent the patient. As the fidelity of the clinical simulation increases a computerized mannequin (that simulates the signs and symptoms of disease) may be used to represent the patient. The fidelity may further increase with instructors controlling the signs

542

and symptoms of disease that the mannequin is experiencing and providing the mannequin with a “voice” via a speaker or instructor from the control room. Lastly, a high fidelity situation may be constructed where the health professional interacts with an actor playing the part of a patient, other health professionals as well as the systems/UCDs that will be evaluated. The authors of the book chapter have found that one can create clinical simulations that can be used to evaluate the HIS/UCD. Low and high fidelity simulations have been created by the authors. Low and high fidelity simulations involving a health professional, HIS and UCD can be easily created for as little as $2000 (to evaluate cognitive-socio-technical fit). These simulations involve using real-world equipment and health professionals that can be found in any healthcare facility globally (see Kushniruk & Borycki, 2006). Simulations involving computerized mannequins can also be constructed in facilities that employ such technology to train medical and nursing students as they are now an important part of the training these health professionals globally. Here, such simulations can be conducted in a medical school, nursing school or government organization that uses such technology to train health professionals or a teaching hospital where such equipment resides (Issenberg et al. 2005; Kneebone et al., 2005).

Use of Scenarios Scenarios are an important component of clinical simulations. Careful attention should be paid to the attributes or qualitative dimensions of each scenario used in the clinical simulation. Scenarios are used to select or create the environment where the clinical simulations will take place (e.g. a hospital room, an operating room, a hallway). Scenarios are also used to set out the context for the clinical simulation. Scenarios should include varying levels of complexity and urgency. For example, if the researcher wishes to examine the

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

ability of a system to address subject information needs in time limited, demanding environments, he or she may observe the subject attempting to acquire information from the system given varying time constraints and urgency (e.g. 5 minutes use of the system to address an information need about a patient’s condition versus 20 minutes of possible system use to make a decision about a patient’s condition after a clinic visit). Scenarios should take into account routine and atypical cases. For example, in conducting evaluations involving medication order entry systems, the evaluator may include routine cases (e.g. the prescribing of medication in a physician’s office to a patient with a chronic illness) (as illustrated in Kushniruk et al, 2005) to atypical cases (e.g. a physician giving medications immediately (i.e. stat) during a critical, life threatening event such as a patient experiencing a myocardial infarction (see Kushniruk et al., 2006 for more details). Scenarios should be representative of a range of situations that would be encountered by subjects. They should also outline possible sequences of activities that will take place. This will serve as a comparison point between existing processes and outcomes and those that are expected as a result of introducing the technology and those that emerge during a scenario as unintended consequences once the HIS/UCD are introduced (Borycki et al., 2009b; Orlinkowski, 1992). For example, when studying the adequacy of a HIS in addressing information needs, the researcher must first select scenarios that stimulate subjects’ need for information such as the need to assess a patient’s condition (e.g. Borycki & Lemieux-Charles, 2009). Then the researchers might compare what is routinely done by health professionals in a paper environment to that done to complete the same work in a HIS/UCD environment. Scenarios can involve low fidelity, simple written medical case descriptions that are given to subjects to read and respond to. Higher fidelity clinical simulation scenarios can involve more complex scenarios involving computer controlled

mannequins that display preset programmed signs and symptoms of disease over time for health professionals to respond to. Even higher fidelity simulations may involve actors with specifically developed scripts that engage the health professional and involve the HIS/UCD to be evaluated. For example, a simple clinical simulation may involve presenting physicians with a short written case description of a patient and asking them to enter the information about the patient into a patient record system. A high fidelity simulation would reproduce more closely the real world situation being studied. This might involve actors playing the roles of patients and staff in a clinic exam room in a study of how doctors’ use of an electronic patient record system on a hand held UCD. Such a study may involve multiple recordings of UCDs to precisely document all subject interactions (e.g. audio and video recording of all verbalizations, computer activities and the hospital room or clinic environment as a whole to document actions). For example, in studying the impact of hybrid electronic-paper environments upon novice nurse information seeking, Borycki et al., (2009b) recreated a typical nursing unit environment that would be found in a hospital (i.e. desk, telephone, chairs, workstation, software and paper and electronic sources of nursing and medical information about a typical patient). In order to capture the effects of the hybrid environment upon novice, nurse information seeking the researcher recorded each nurses’ verbalizations while reviewing their patient’s information by using an audio recorder. The nurses’ actions and interactions with their environment were recorded by using a video camcorder. Lastly, the nurses’ interactions with the software and hardware (i.e. the workstation) were recorded using Hypercam®, a screen recording program.

Selecting Representative Tasks This stage involves the selection of representative tasks users are expected to undertake when

543

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

using the system and UCD under study. A range of tasks should be selected (Kushniruk & Patel, 2004). Ideally, tasks should be in keeping with the environment where the task will be performed and the scenario that has been constructed (Borycki et al., 2005). For example, a nurse may be asked to administer routine medications to a patient in a hospital room. Tasks that are part of this scenario include: logging on to the electronic medication administration system (eMAR), searching for a patient that will be receiving the medication in the electronic medication administration system (eMAR), electronically identifying the patient, identifying the medications that need to be administered at that time, scanning the patients’ identifying bar-coded bracelet with a bar code scanner, scanning the medication package, pouring the medication, verifying the patient’s identity verbally, giving the medication to the patient and then documenting the administration of the medication in the eMAR that is provided on a laptop mounted on a portable medication administration cart (Altman, 2004; Patterson et al., 2002). This process ensures the medication that is being given by the nurse is given to the correct patient, is the correct medication, is in the correct dose, and is provided by the correct route (e.g. by mouth). This process also ensures the administration of the medication is documented in the eMAR (Altman, 2004).

Inviting Representative Participants This step involves the identification and selection of representative subjects for the simulation. Subjects should be representative of end users of the system under study in terms of level of disciplinary, domain and technology expertise. Research suggests health professional background can have a significant impact upon processes that involve patient care, systems and UCDs and patient and health professional outcomes (e.g. medical error rates). For example, health professionals of differing disciplinary backgrounds may

544

respond differently to the same scenario (Johnson & Turley, 2006). Research has also demonstrated that a domain of expertise will influence health professional behaviours and outcomes (Lammond et al., 1996). For example, a medical nurse will attend to differing types of information as compared to a surgical nurse when presented with the same task - searching an electronic patient record. Organizational experience has also been shown to have an effect upon worker behaviours (Jablin & Putnam, 2001). Lastly, more recent research by Patel and colleagues (2000) reveals computer expertise can have a significant impact upon health professional use of systems. In addition to this, research suggests experience with differing UCDs may have an impact upon health professional behaviours and outcomes. Evaluators must take into account disciplinary, domain, organizational and technology expertise when selecting representative participants. Ideally a range of participants should be selected to understand differing users’ responses to the system/UCD. If the intent is to study user ability to learn about the new HIS/UCD constellation or group of software and devices, then novice users are best to be invited to participate in the clinical simulations (Borycki et al., 2009b).

Data Collection The authors of this book chapter recommend that audio and video data be collected. Individuals may be asked to “think aloud” as they work with the HIS/UCD individually. Groups of health professionals working with the system/UCD should have their conversations recorded. Audio data of individual’s thinking aloud or the conversations of groups of users working with the system/UCD is essential. Audio data provides insights as to what information in the system representative participants are attending to as well as data about potential usability issues arising from system and UCD attributes while the individual or group is working with the HIS and UCD constellation

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

(Kushniruk & Patel, 2004). Other forms of data may include video recordings of subjects interacting with systems, UCDs, other health professionals and simulated patients. Video data allows the researcher to observe emergent workflows arising from interactions between system, UCD and technology (Borycki & Kushniruk, 2005). Computer screen recordings of participants interacting with the system allow the researchers to identify sources of poor or cumbersome workflows. Screen recordings also illuminate human computer interaction issues and issues arising from poor system usability (Kushniruk et al., 2006). Video data and computer screen recordings can be triangulated with the audio data. Increasingly, the role of computer screen recordings and video data has been found to inform and contextualize information (Borycki et al., 2009). These two types of data have provided researchers with additional insights and understanding of the underlying cognitive processes of subjects and the effects of systems and UCDs upon them. For example, in a recent study by the author that examined the relationship between medical error and system usability, video and computer screen data was used to inform audio transcripts of subjects’ “thinking aloud” while interacting with a system on a hand held device. During analysis, audio data indicated subjects were under the impression they had entered the correct prescription data when using an electronic prescribing program on a hand held device. Corresponding video data and computer screen recordings revealed subjects had not entered the correct information and they experienced usability issues. As a result, subjects unknowingly entered the wrong prescription (Borycki & Kushniruk, 2006; Kushniruk et al., 2005). As described above, clinical simulations as a methodological approach can be used to study the effect of systems and UCDs upon health professional work processes and outcomes. Clinical simulation unlike other evaluation approaches in health informatics (e.g. randomized clinical control trials) allow one to observe clinicians (e.g.

physicians or nurses) interacting with HIS/UCDs under simulated conditions. Clinical simulations allow one to study the effects of the system, UCD, health professional and patient interactions upon clinical processes and outcomes. Clinical simulations can be conducted early in the software development lifecycle to inform design, development, customization, procurement, implementation, operation and maintenance of software and hardware. It must be noted that clinical simulations conducted in a laboratory setting offer significant insights into interactions between UCD, systems and health professionals within an organizational environment (e.g. a hospital). In conducting clinical simulations (as outlined above) all aspects of the interaction are recorded. Researchers can make observations about cognitive-socio-technical fit and changes can be made to the UCD/system to improve fit (Borycki & Kushniruk, 2006). Recordings allow multiple researchers to review the collected data. Inter-rater reliability can be calculated as a result. Researcher reviews of video and audio recordings can prevent researcher and subject recall bias and researcher recording bias. Subject’s may experience recall bias – being unable to remember all the interactions between the system, UCD and organizational environment, if other forms of data collection are used such as semi-structured interviews or focus groups. In such cases subjects may not be able to recall all the events that occurred in their interactions with the UCD and system within in a given organizational environment (Jackson & Verberg, 2006). Researcher bias involves a researcher not being able to physically record all of the information when taking notes or being selective about the information that is recorded during observations of subjects interacting between the UCD and technology. Some researchers have attempted to conduct these types of studies in naturalistic settings involving real-world interactions between patients, health professionals, systems, UCDs and organizational contexts. Many of these studies (e.g. Ash

545

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

et al., 2004) have involved semi-structured oneon-one interviews and focus groups with key informants (i.e. health professionals who have been affected by these UCD/HIS implementations). As well, researchers have conducted observational analyses of subjects working in these environments. Researchers observed subjects while they were performing tasks and took notes.

Data Analysis Data analysis, where clinical simulations are concerned, is both qualitative and quantitative in nature. Data can be qualitatively coded using empirically accepted coding schemes to identify subject patterns of interactions with a system, workflow issues, usability problems and recurring system issues. Audio, video and computer screen data can be qualitatively coded. Data can be coded using inductive, deductive and mixed method approaches. Deductive approaches involving the coding of qualitative data involve the use of existing empirically accepted coding schemes, model(s) or theory (ies) (Ericsson & Simon, 1993). Alternatively, when inductive approaches are used, data drives coding scheme development (e.g. grounded theory approaches) (Kuziemsky et al., 2007). Mixed method approaches towards the analysis of qualitative data combine both inductive and deductive methods. Here inductive and deductive approaches are combined. Empirically validated coding schemes from the research are used while at the same time new codes can emerge from the data as prior empirical work may not be sufficient to fully inform the coding of the data. In combining the approaches, transcripts of audio, video and computer screen recordings are initially coded using existing empirically validated coding schemes, models or theories. Then, inductively extensions are made to the empirically validated coding schemes, models or theories (Borycki et al., 2009b; Ericsson & Simon, 1993; Kushniruk & Patel, 2004; Patel et al., 2000).

546

Qualitative data are not limited to coded audio, video or computer screen data. Video data and computer screen recordings can be reviewed. Workflow and data flow diagrams can be constructed to observe for differences in before and after HIS/UCD use. Such diagrams allow the evaluator to identify cumbersome workflows as well as interactions between HIS/UCDs that increase the complexity of work and cognitive load (Borycki et. al., 2009a). Qualitative data (i.e. coded verbal transcripts, actions on the computer screens and observed behaviours) can be quantified (Borycki & Lemieux-Charles, 2008). Coded verbal data may be reduced to individual items “intended to mean on thing” (Sandelowski, 1999, p. 253). Frequencies can be tabulated for each code in the transcribed data. Then, inferential statistics can be done (Borycki & Lemieux-Charles, 2008; Ericsson & Simon, 1993). In summary clinical simulations in health informatics involve imitating real-world situations in a laboratory setting or a organizational environment (e.g. a clinic, hospital room or operating room). Clinical situations are used to assess the impact of new software, UCDs and interactions between the two upon health professional work. In health informatics clinical simulations allow health informaticians to observe health professionals with differing backgrounds (e.g. nurses or physicians) while they perform tasks typical of their profession as they would perform them in real-world settings while engaging with other health professionals and patients. In addition to aiding in the assessment of these UCDs, clinical simulations allow one to assess differing constellations of HIS and UCDs in laboratory or real-world environments to be tested and evaluated prior to their real-world or organization wide implementation - in the process reducing the risk to the healthcare organization by minimizing the likelihood of an institution wide redesign and reimplementation arising from poor cognitive-socio-technical fit. In undertaking clinical simulations in health informatics, HIS and UCD designers as well as the organization’s where

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

these systems and UCDs are implemented are able to assess the impact of the system and UCD upon health professional processes and outcomes without patient involvement and without placing the health professional in a situation where the new system and UCD may impact upon their ability to work with other health professionals or with a patient. In this case, the health professional encounters difficulties while working with the HIS and UCD when using the system/device in the context of a “safe” laboratory environment and clinical situation (i.e. not involving patients).

FUTURE RESEARCH DIRECTION Future research involving the use of clinical simulations will involve applying the methodology to new and emerging types of UCD and HIS. For example, many healthcare organizations are attempting to determine the UCDs that would best meet the needs of clinicians in the hospital and clinical setting. In addition to this as UCDs have become more prevalent in healthcare, HIS that have historically been made available to clinicians via desktop workstations will need to be evaluated in terms of their ability to support the cognitive and social aspects of health professional work using varying types of UCD/HIS in combination. This will help to identify the best constellations of HISs and UCDs that lead to improvements in health care processes and the quality of patient care while at the same time reducing the likelihood of a medical error occurring. The authors are currently working on developing advanced simulations to assess potential issues in the integration of electronic health records with a range of physical devices (e.g. monitoring devices, bar coded patient bracelets etc). Lastly, there will be a need to identify the range of routine to urgent and typical to atypical situations that can be used to test HIS and UCD.

CONCLUSION Clinical simulations in health informatics have allowed software and hardware vendors as well as healthcare organizations to better customize HIS to the local organizational environments prior to implementation. Clinical simulation also allows organizations to assess the impact of HIS and UCD constellations upon health professionals cognitivesocio-technical fit. Such knowledge is significant as it influences the design and development of software and UCDs at the vendor level. Such knowledge also influences the customization of software by organizations and their procurement of software and UCDs for use in the clinical setting. Lastly, such information offers organizations the opportunity to develop training strategies for health professionals that take into account the cognitive-socio-technical learning needs of health professionals as they learn to use the new HIS and the UCDs that will be used in conjunction with the HIS in a real-world setting. The outcomes of using clinical simulations to evaluate HIS and UCDs are significant. Clinical simulations have been effectively used to evaluate the impact of a technology (i.e. HIS and UCDs) upon health professional processes such as patient interviews (involving actors who are playing the role of patient) (Kushniruk et al., 1996), information seeking (Borycki et al., 2009b), decision-making (Patel et al., 2000), knowledge organization, reasoning (Patel et al., 2001), workflow (Borycki et. al., 2009a), and patient safety (Kushniruk et al., 2005). They have also been used to evaluate the impact of interactions between UCDs, HIS and health professionals where patient care processes and outcomes are concerned (Kushniruk et. al., 2005; Patel et. al., 2000). Clinical simulations when applied to the evaluation of cognitive-socio-technical fit involving HIS and UCDs can provide significant benefit to technology designers and organizations who are procuring these systems and UCDs. They allow technology designers to evaluate the impact

547

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

of systems/UCDs upon health professional work prior to systems implementation (thereby preventing possible system/UCD implementation failures or failures to fully use HIS/UCD functionality). From an organizational perspective, clinical simulations allow organizations to evaluate the interactions between HIS, UCD, organizational environment and health professional. The use of clinical simulations ensures that organizations achieve a strong cognitive-socio-technical fit for health professionals. This lessens the likelihood of implementation failure and poor health professional adoption and appropriation of technology(ies). Such use of clinical simulation is needed to evaluate HIS/UCD cognitive-socio-technical fit with health professional work. Such evaluation is important to prevent the need for the modification of HIS/UCDs after implementation when the cost of implementation is significantly higher. The use of clinical simulations in the evaluation of HIS/UCDs is a significant advancement over an evaluation of HIS/UCD after implementation using semi-structured interviews or observation of clinicians in the healthcare work environment (as both these methodologies are prone to recall and researcher recording bias) and may not allow for full documentation of the interactions. This is a new and emerging methodology that extends work in the area of usability engineering by borrowing methods of recording data and applying those recording methods to the evaluation of cognitive-socio-technical fit involving HIS and UCDs. This work also borrows methods from the training of health professionals and extends them for use in evaluating the effects of HIS/UCDs upon cognitive-socio-technical fit involving organizational processes and outcomes.

REFERENCES Altman, G. (2004). Delmar’s fundamental and advanced nursing skills (2nd ed.). Toronto: Thompson Delmar Learning.

548

Ammenwerth, E., Schnell-Inderst, P., Machan, C., & Siebert, U. (2008). The effect of electronic prescribing on medication errors and adverse drug events: A systematic review. Journal of the American Medical Informatics Association, 15(5), 585–600. doi:10.1197/jamia.M2667 Anderson, J. G., & Aydin, C. E. (2005). Evaluating the organizational impact of healthcare information systems. New York: Springer Verlag. doi:10.1007/0-387-30329-4 Ash, J. S., Berg, M., & Coiera, E. (2004). Some unintended consequences of information technology in health care: The nature of patient care information system related errors. Journal of the American Medical Informatics Association, 11(2), 121–124. Bates, D. W., Teich, J. M., Lee, J., Seger, D., & Kuperman, G. J., Ma’Luf, N., Boyle, D., Leape, L. (1999). The impact of computerized physician order entry on medication error prevention. Journal of the American Medical Informatics Association, 6(4), 313–321. Berg, M. (1999). Patient care information systems and health care work: A sociotechnical approach. International Journal of Medical Informatics, 55(2), 87–101. doi:10.1016/S13865056(99)00011-8 Boehm, B. W. (1976). Software engineering. IEEE Transactions on Computers, C-25. Borycki, E. M., & Kushniruk, A. W. (2006). Identifying and preventing technology-induced error using simulations: Application of usability engineering techniques. Healthcare Quarterly (Toronto, Ont.), 8, 99–105. Borycki, E. M., & Kushniruk, A. W. (2008). Where do technology induced errors come from? Towards a model for conceptualizing and diagnosing errors caused by technology. In Kushniruk, A. W., & Borycki, E. M. (Eds.), Human and Social Aspects of Health Information Systems (pp. 148–166). Hershey, PA: IGI Global.

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

Borycki, E. M., Kushniruk, A. W., Kuwata, S., & Kannry, J. (2006). Use of simulation in the study of clinician workflow. AMIA Annual Symposium Proceedings (pp. 61-5). Borycki, E. M., Kushniruk, A. W., Kuwata, S., & Watanabe, H. (2008). Using low-cost simulation approaches for assessing the impact of a medication administration system on workflow. Studies in Health Technology and Informatics, 136, 567–572. Borycki, E. M., Kushniruk, A. W., Kuwata, S., & Watanabe, H. (2009a). Simulations to assess medication administration systems. In Staudinger, B., Hob, V., & Ostermann, H. (Eds.), Nursing and Clinical Informatics (pp. 144–159). Hershey, PA: IGI Global. Borycki, E. M., & Lemieux-Charles, L. (2008). Does a hybrid electronic-paper environment impact on health professional information seeking. Studies in Health Technology and Informatics, 136, 505–510. Borycki, E. M., Lemieux-Charles, L., Nagle, L., & Eysenbach, G. (2009b). Evaluating the impact of hybrid electronic-paper environments upon novice nurse information seeking. Methods of Information in Medicine, 2, 1–7. Chaudhry, B., Wang, J., Wu, Sh., Maglione, M., Mojica, W., & Roth, E. (2006). Systematic review: Impact of health information technology on quality, efficiency and costs of medical care. Annals of Internal Medicine, 144, E-12–E22. Chen, E. S., Mendonca, E. A., McKnight, L. K., Stetson, P. S., Lei, J., & Cimino, J. J. (2004). PalmCIS: A wireless handheld application for satisfying clinician information needs. Journal of the American Medical Informatics Association, 11(1), 19–28. doi:10.1197/jamia.M1387

Elmes, D. G., Kantowicz, B. H., & Roediger, H. L. (1981). Research methods in psychology (2nd ed.). St. Paul: West Publishing Company. Ericsson, KA, Simon, HA. Protocol analysis: Verbal reports as data (2nd ed.). Cambridge, Massachusetts: MIT Press. Eslami, S., Abu-Hanna, A., & deKeizer, N. F. (2007). Evaluation of outpatient computerized physician medication order entry systems: A systematic review. Journal of the American Medical Informatics Association, 14(4), 400–406. doi:10.1197/jamia.M2238 Eslami, S., deKeizer, N. F., & Abu-Hanna, A. (2008). The impact of computerized physician medication order entry in hospitalized patients: A systematic review. International Journal of Medical Informatics, 77, 365–376. doi:10.1016/j. ijmedinf.2007.10.001 Friedman, C. P., & Wyatt, J. C. (2006). Evaluation methods in biomedical informatics. New York: Springer. Granlien, M. F., Herzum, M., & Gudmundsen, J. (2008). The gap between actual and mandated use of an electronic medication record three years after deployment. Studies in Health Technology and Informatics, 136, 419–424. Haimson, B. R., & Elfenbein, M. H. (1985). Experimental methods in psychology. Toronto: McGraw-Hill. Issenberg, S. B., McGaghie, W. C., Hart, I. R., Mayer, J. W., Felner, J. M., & Petrusa, E. R. (1999). Simulation technology for health care professionals skills training and assessment. Journal of the American Medical Association, 282(9), 861–866. doi:10.1001/jama.282.9.861

Cioffi, J., & Markham, R. (1997). Clinical decision making by midwives: Managing case complexity. Journal of Advanced Nursing, 23, 265–272. doi:10.1046/j.1365-2648.1997.1997025265.x

549

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

Issenberg, S. B., McGaghie, W. C., Petrusa, E. R., Gordon, D. L., & Scalese, R. J. (2005). Features and uses of high-fidelity medical simulations that lead to effective learning: A BEME systematic review. Medical Teacher, 27(1), 10–28. doi:10.1080/01421590500046924 Jablin, F. M., & Putnam, L. L. (2001). The new handbook of organizational communication: Advances in theory, research and methods. New York: Sage. Jackson, W., & Verberg, N. (2006). Methods: Doing social research (4th ed.). Toronto: Pearson Prentice Hall. Johnson, C. M., & Turley, J. P. (2006). The significance of cognitive modeling in building healthcare interfaces. International Journal of Medical Informatics, 75, 163–172. doi:10.1016/j. ijmedinf.2005.06.003 Joint Commission for Health Care Quality. (2008). Safely implementing health information and converging technologies. Retrieved April 14, 2009 from http://www.jointcommission.org/ SentinelEventAlert/sea_42.htm Kneebone, R. (2005). Evaluating clinical simulation for learning procedural skills: A theory-based approach. Academic Medicine, 80(6), 549–553. doi:10.1097/00001888-200506000-00006 Koppel, R., & Kreda, D. (2009). Health care information technology vendors’ “hold harmless” clause: Implications for clinicians. Journal of the American Medical Association, 301(12), 1276–1278. doi:10.1001/jama.2009.398 Koppel, R., Metlay, J. P., Cohen, A., Abaluck, B., Localio, R., Kimmel, S. E., & Strom, B. L. (2005). Role of computerized physician order entry systems in facilitating medication errors. Journal of the American Medical Association, 293(10), 1197–1203. doi:10.1001/jama.293.10.1197

550

Kushniruk, A. W., Borycki, E., Kuwata, S., & Kannry, J. (2006). Predicting changes in workflow resulting from healthcare information systems: Ensuring the safety of healthcare. Healthcare Quarterly (Toronto, Ont.), 9, 114–118. Kushniruk, A. W., & Borycki, E. M. (2006). Low-cost rapid usability engineering: Designing and customizing usable healthcare information systems. Healthcare Quarterly (Toronto, Ont.), 9(4), 98–100, 102. Kushniruk, A. W., Kaufman, D. R., Patel, V. L., Levesque, Y., & Lottin, P. (1992). Assessment of a computerized patient record system: A cognitive approach to evaluating an emerging medical technology. M.D. Computing, 13(5), 406–415. Kushniruk, A. W., & Patel, V. L. (2004). Cognitive and usability engineering methods for the evaluation of clinical information systems. Journal of Biomedical Informatics, 37, 56–76. doi:10.1016/j. jbi.2004.01.003 Kushniruk, A. W., Triola, M. M., Borycki, E. M., Stein, B., & Kannry, J. L. (2005). Technology induced error and usability: The relationship between usability problems and prescription errors when using a handheld application. International Journal of Medical Informatics, 74, 519–526. doi:10.1016/j.ijmedinf.2005.01.003 Kuwata, S., Kushniruk, A., Borycki, E., & Watanabe, H. (2006). Using simulation methods to analyze and predict changes in workflow and potential problems in the use of bar-coding medication order entry systems. AMIA Annual Symposium Proceedings, 994. Kuziemsky, C. E., Borycki, E. M., Black, F., Boyle, M., Cloutier-Fisher, D., Fox, L., et al. (2007). A framework for interdisciplinary team communication and its implications for information systems design. wcmeb. Eighth World Congress on the Management of eBusiness (pp. 14).

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

Lammond, D., Crow, R., Chase, J., Doggen, K., & Sinkels, M. (1996). Information sources used in decision making: Considerations for simulation development. International Journal of Nursing Studies, 33(1), 47–57. doi:10.1016/00207489(95)00064-X Lapointe, L., & Rivard, S. (2005). Clinical information systems: Understanding and preventing their premature demise. Healthcare Quarterly (Toronto, Ont.), 3(4), 92–99. Lapointe, L., & Rivard, S. (2006). Getting physicians to accept new information technology: Insights from case studies. Canadian Medical Association Journal, 174(11), 1573–1578. doi:10.1503/cmaj.050281 Levenson, N. (1995). Safeware: System safety and computers. Reading, Massachusetts: AddisonWesley. Ludwick, D. A., & Doucette, J. (2009). The implementation of operational processes for Alberta Electronic Health Record: Lessons for Electronic Medical Record Adoption in primary care. Healthcare Quarterly (Toronto, Ont.), 7(4), 103–107.

Patel, V. L., Kushniruk, A. W., Yang, S., & Yale, J.-F. (2000). Impact of a computer-based patient record system on data collection, knowledge organization, and reasoning. Journal of the American Medical Informatics Association, 7(6), 569–585. Patterson, E. S., Cook, R. I., & Render, M. L. (2002). Improving patient safety by identifying side effects from introducing bar coding in medication administration. Journal of the American Medical Informatics Association, 9(5), 540–553. doi:10.1197/jamia.M1061 Rabinowitz, E. (1999). Is there a doctor in the house? Managed Care (Langhorne, Pa.), 8(9), 42–44. Sandelowski, M. (1991). Combining qualitative and quantitative sampling, data collection and analysis techniques in mixed-method studies. Research in Nursing & Health, 23, 2046–2255. Shachak, A., Hadas-dayagi, M., Ziv, A., & Reis, S. (2009). Primary care physicians’ use of an electronic medical record system: A cognitive task analysis. Journal of General Internal Medicine, 24(3), 341–348. doi:10.1007/s11606-008-0892-6

McLaughlin, C. P., & Kaluzny, A. D. (2006). Continuous quality improvement in healthcare: Theory, implementations and applications (3rd ed.). Toronto: Jones and Bartlett.

Shamian, J. (1988). The effect of the teaching of decision analysis on student nurses’ clinical intervention decision making. Unpublished doctorial dissertation. Case Western Reserve University, Clevland, Ohio.

Orlikowski, W. J. (1992). The duality of technology: Rethinking the concept of technology in organizations. Organization Science, 3(3), 398–427. doi:10.1287/orsc.3.3.398

Shortell, S. M., & Kaluzny, A. D. (2006). Health care management: Organization design and behaviour (5th ed.). Canada: Thompson Delmar Learning.

Patel, V. L., & Groen, G. J.The general and specific nature of medical expertise: A critical look. In Ericsson, K. A., & Smith, J. (Eds.), Toward a general theory of expertise prospects and limits. New York (pp. 93–125). Cambridge.

Shortliffe, E. H., & Cimino, J. J. (2008). Biomedical Informatics: Computer Applications in Healthcare and Biomedicine. New York: Springer.

551

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

Shulman, R., Singer, M., Goldstone, J., & Bellingan, G. (2005). Medication errors: A prospective cohort study of hand-written and computerized physician order entry in the intensive care unit. Critical Care (London, England), 9(8), 516–521. doi:10.1186/cc3793 Vogt, W. P. (1999). Dictionary of statistics and methodology (2nd ed.). Thousand Oaks: Sage.

ADDITIONAL READING Ammenwerth, E., Schnell-Inderst, P., Machan, C., & Siebert, U. (2008). The effect of electronic prescribing on medication errors and adverse drug events: A systematic review. Journal of the American Medical Informatics Association, 15(5), 585–600. doi:10.1197/jamia.M2667 Ash, J. S., Berg, M., & Coiera, E. (2004). Some unintended consequences of information technology in health care: The nature of patient care information system related errors. Journal of the American Medical Informatics Association, 11(2), 121–124. Borycki, E. M., & Kushniruk, A. W. (2008). Where do technology induced errors come from? Towards a model for conceptualizing and diagnosing errors caused by technology. In Kushniruk, A. W., & Borycki, E. M. (Eds.), Human and Social Aspects of Health Information Systems (pp. 148–166). Hershey, PA: IGI Global. Borycki, E. M., Kushniruk, A. W., Kuwata, S., & Watanabe, H. (2009a). Simulations to assess medication administration systems. In Staudinger, B., Hob, V., & Ostermann, H. (Eds.), Nursing and Clinical Informatics (pp. 144–159). Hershey, PA: IGI Global.

552

Borycki, E. M., Lemieux-Charles, L., Nagle, L., & Eysenbach, G. (2009b). Evaluating the impact of hybrid electronic-paper environments upon novice nurse information seeking. Methods of Information in Medicine, 2, 1–7. Chaudhry, B., Wang, J., Wu, Sh., Maglione, M., Mojica, W., & Roth, E. (2006). Systematic review: Impact of health information technology on quality, efficiency and costs of medical care. Annals of Internal Medicine, 144, E-12–E22. Eslami, S., Abu-Hanna, A., & deKeizer, N. F. (2007). Evaluation of outpatient computerized physician medication order entry systems: A systematic review. Journal of the American Medical Informatics Association, 14(4), 400–406. doi:10.1197/jamia.M2238 Eslami, S., deKeizer, N. F., & Abu-Hanna, A. (2008). The impact of computerized physician medication order entry in hospitalized patients: A systematic review. International Journal of Medical Informatics, 77, 365–376. doi:10.1016/j. ijmedinf.2007.10.001 Koppel, R., Metlay, J. P., Cohen, A., Abaluck, B., Localio, R., Kimmel, S. E., & Strom, B. L. (2005). Role of computerized physician order entry systems in facilitating medication errors. Journal of the American Medical Association, 293(10), 1197–1203. doi:10.1001/jama.293.10.1197 Kushniruk, A. W., & Borycki, E. M. (2006). Low-cost rapid usability engineering: Designing and customizing usable healthcare information systems. Healthcare Quarterly (Toronto, Ont.), 9(4), 98–100, 102. Kushniruk, A. W., Triola, M. M., Borycki, E. M., Stein, B., & Kannry, J. L. (2005). Technology induced error and usability: The relationship between usability problems and prescription errors when using a handheld application. International Journal of Medical Informatics, 74, 519–526. doi:10.1016/j.ijmedinf.2005.01.003

Use of Clinical Simulations to Evaluate the Impact of Health Information Systems

KEY TERMS AND DEFINITIONS Clinical Simulation: A re-creation or imitation of the key characteristics of a real-world healthcare environment or setting (e.g. hospital room, emergency room, physician office) for the purpose of evaluating a group of HIS/UCD. Cognitive-Socio-Technical Fit: Cognitivesocio-technical fit refers to the cognitive aspects of fit involving health professionals’ cognition, the social aspects of fit including social interactions that take place between the health professional(s), patient and technology and the use of the technology to provide patient care. Computerized Physician Order Entry (CPOE): Computerized physician order entry is a health information system that allows for the electronic entry of physician orders for patients under a physician’s care. These orders are sent over a computer network to other health professionals so that they may be fulfilled. Ecological Validity: Ecological validity refers to the ability to create clinical simulations that are representative of real-world healthcare settings (e.g. hospitals and clinics) so that the

results of a clinical simulation are generalizable to the real-world. Electronic Medication Administration Record (eMAR): An electronic medication administration record is an electronic health information system that uses bar code reader technology, an electronic medication record system and a wireless medication cart to aid in the process of medication administration. Intended Consequences: Those real-world processes and outcomes that HIS and UCD are expected to change by the technology’s designers or the organization that implemented the HIS/UCD after the HIS/UCD has been implemented. Technology-Induced Error: A new type of medical error that arises from the introduction of a technology or from the interactions that take place between a technology, a health professional and a real-world healthcare environment. Unintended Consequences: Those realworld processes and outcomes of HIS and UCD that were not predicted and that did not change as was originally expected by the technology’s designers or the organization that implemented the HIS/UCD.

This work was previously published in Ubiquitous Health and Medical Informatics: The Ubiquity 2.0 Trend and Beyond, edited by Sabah Mohammed and Jinan Fiaidhi, pp. 552-573, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

553

554

Chapter 2.21

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems A. Janß RWTH Aachen University, Aachen, Germany W. Lauer RWTH Aachen University, Aachen, Germany F. Chuembou Pekam RWTH Aachen University, Aachen, Germany K. Radermacher RWTH Aachen University, Aachen, Germany

ABSTRACT Studies concerning critical incidents with technical equipment in the medical/clinical context have found out, that in most of the cases non-ergonomic and non-reliable user interfaces provoke use deficiencies and therefore hazards for the patient and the attending physician. Based on these studies, the authors assume that adequate and powerful tools for the systematic design of error-tolerant and ergonomic Human-Machine-Interfaces for DOI: 10.4018/978-1-60960-561-2.ch221

medical devices are missing. In this context, the Chair of Medical Engineering (mediTEC) has developed the new software-based tool mAIXuse in order to overcome these difficulties and to support designers as well as risk assessors. Based on two classical formal-analytical approaches, mAIXuse provides a high-performance modelling structure with integrated temporal relations in order to visualise and analyse the detailed use process, even with complex user interfaces. The approach can be used from the very early developmental stages up to the final validation process. Results

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

of a comparative study with the new mAIXuse tool and a conventional process-FMEA (Failure Mode and Effect Analysis) show, that the new approach clearly outperforms the FMEA technique.

INTRODUCTION The major objective of the rapidly evolving technological progress and automation in the field of medical devices and systems is an enhancement of the efficiency and effectiveness of diagnostic, therapeutic procedures. This is partially associated with fundamental changes of the HumanMachine-Interaction characteristics. To ensure safe and reliable user interfaces not only ergonomic but also error-tolerant interface design has to be taken into consideration by the design engineer. In this context, a new software-based tool mAIXuse for formal-analytical usability evaluation and use-oriented risk analysis of medical devices and systems has been developed at the Chair of Medical Engineering and will be presented in this chapter. The in depth analysis of human error in the clinical context of use are of major importance particularly for the introduction of new technical equipment into the medical work system. State of the art methods and tools in usability engineering and risk analysis still have some problems and bottlenecks related to a systematic a-priori as well as a-posteriori review and analysis of human induced risks in risk sensitive work processes. As the complexity and speed of development of medical devices is increasing (more than 50% of medical products typically are less than 2 years on the market) and as the incidents of human error in medicine increases, more sophisticated tools for human error risk analysis seems to be mandatory. The main focus of this chapter is on the development and evaluation of the mAIXuse approach for model-based usability evaluation and use-oriented risk analysis.

BACKGROUND The overall risk for manufacturers of risk-sensitive products (e.g. in aeronautics, nuclear engineering, medical technology and pharmaceuticals) has increased in recent years. Expensive products can only be lucrative if they are distributed to a large number of customers worldwide. This increases not only the cumulative potential damage, but also the potential consequences. Especially in the medical field of diagnostic and therapeutic systems undetected remaining failures or residual risks, which occur in the development process, often cannot be tolerated. Bringing defective and erroneous products to the market means a high potential of severe consequences for the manufacturers. Apart from ethical considerations, related costs can endanger the livelihood of enterprises. It ranges from warranty and goodwill costs to product recalls as well as liability for consequential damages. The medical branch is, similar to almost all manufacturing industries, under a high cost and time pressure, which is characterised by globalisation and dynamic markets, innovation and shorter product life cycles. The increase of the complexity of products and production processes is another factor determining the situation of the risk (Doubt, 2000). The necessity to respond to the risks, is also reflected by the fact that the introduction of systematic risk management processes including the early application of an usability engineering process in accordance with established national and international standards (EN ISO 14971, IEC 62366, MDD (Medical Device Directive), FDA (Food and Drug Administration)) are obligatory for medical device manufacturers. The risk management process can be seen as a “systematic application of management principles, procedures and practices to identify, analyse, treat and monitor understanding of risks, allowing organisations to minimise losses and maximise opportunities” (Hoffmann, 1985). However, these goals are often missed in practice (Hansis, 2002).

555

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

There are about 40.000 malpractice complaints and the evidence of more than 12.000 malpractice events per year in Germany (Kindler, 2003). More than 2.000 cases of this so-called “medical malpractice” can be traced to medical and surgical causes, leading in the end to the death of the patient (Hansis, 2001). The expert’s opinion on the development of health services since the year 2007 is that in Germany approximately 17.000 deaths each year are due to “preventable adverse events”. In addition to false decisions and medication errors, use errors are a major cause of deaths in the medical context (Merten, 2007). Harvard Medical Practice Studies confirmed the important role of human factors in safety aspects of Human-Machine-Interaction in clinical environment. Following their statement, 39.7% of avoidable mistakes in operating rooms arise from “carelessness” (Leape, 1994). Avoidable mistakes, especially in combination with the use of technical equipment in the medical context, are, with a disproportionately high rate, due to human-induced errors (Leape, 1994). Results from the Harvard Medical Practice Study were confirmed by a number of other epidemiological studies. In the retrospective study by Gawande et al. regarding 15000 surgically treated patients in 3% of the cases user-based injuries have been found, of which 54% have been preventable errors (Gawande, Thomas Brennan and Zinner, 1999). According to incident reports in orthopaedic surgery avoidable mistakes with technical equipment are in 72% due to human error (Rau et al., 1996). Thus, human-oriented interface design plays an important role in the introduction of new technology and specific devices into medical applications. To ensure safe and reliable user interfaces not only ergonomic but also error-tolerant interface design has to be taken into consideration for the acceptance as well as for the routine application of medical devices (Radermacher, 2004). With the increasing complexity of tasks and systems, reliability can only be guaranteed by an optimized Human-Machine Interaction (HMI)

556

(Giesa and Timpe, 2000). From the cognitive psychological point of view, inadequate coordination of the design of the work equipment and environment with the task and situation, the erroneous interpretation of information in user interfaces and situational factors as well as external environmental factors and extrinsic stress are the main causes of “human error” in medicine (Bogner, 1994). Statistics on malpractice incidents and the handling of medical devices seem to prove this fact (Kohn, Corrigan and Donaldson 2000). Related studies state that an inadequate design of working task and situation with respect to working equipment and environment can be regarded as a major trigger for user errors (Hyman 1994). The advances in automation abets the number and complexity of technical components in medicine in recent years (e.g. in the clinical context or in home care). In general, this provides more effective and efficient treatment options, but devices with a broad range of functions and complex user interfaces, additionally, potentially bring along use problems. This trend is reflected e.g. in the risk-sensitive work system operating room (OR) characterized by a high amount of Human-Machine- and HumanHuman-Interaction. Increasing functionality and complexity of new technical medical equipment in the OR – such as multi modal image data acquisition and processing, navigation telemanipulators and robot systems - can implicate deficiencies in the use process, bringing along high potential for hazardous human-induced failure implicating higher risk for the patient and the attending physician. Nowadays, in addition to conventional medical tasks, physicians have to cover a multitude of technical monitoring and control duties. Moreover, the boundary conditions are characterized by various potentially adverse performance shaping factors (climate, light and sterile working conditions, clothing, risk of infection and irradiation, time pressure…), which amongst others can lead to a higher physical and mental workload as well as to a deprivation of situation awareness

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

and therefore to potential human failure (Cook et al., 1996, Radermacher, 1999). When performing under extreme time pressure, the combination of high physical effort and multimodal information processing (e.g. interpretation of several X-ray images while perceiving acoustic alarms and tactile feedback concerning tissue manipulation) brings along an increased risk potential for the patient and the surgical team. The progressive change in the interaction characteristics between surgical staff and computer/ machine (e.g. increase of control and monitoring functions, the change in focus concerning operational area vs. imaging and navigation systems etc.) and subsequent use errors induce a significantly increasing risk potential (Radermacher et al., 2004). Last but not least, task analysis in surgery show, that task sequences to be supported by Computer Aided Surgery (CAS) techniques, represent only a relatively small fraction of time in daily clinical routine work of a surgeon. This implies on one hand that, time efficiency of novel technical procedures introduced into the common workflow, is of major importance. On the other hand, practical use of these techniques may be adversely affected by routine deficiencies as well as due to the fact, that time consuming technical training can only be realized to a very limited amount. This especially holds for technical equipment explicitly dedicated to support rare and complicated cases. Nevertheless, even in situations with high workload in the OR (e.g. in emergency situations) a reliable performance of the operation must be ensured. To guarantee patient and user safety, when applying medical technical components, manufacturers of medical devices have to fulfill specific legislative and regulatory requirements. Essential requirements according to the risk management process are in terms of ergonomic analysis and optimization of Human-Machine Interaction (HMI), thus in assuring the usability and reliability of new medical products. To assure high reliability of medical risk-sensitive systems, e.g. in the work-

ing field OR, it is necessary to conduct a usability evaluation in relation to the use context and the user. Since 2006, as part of the risk management for medical devices, the harmonization of the DIN EN 60601-1-6 standard requires a documented usability assessment of medical electrical equipment. Since 2007, the international standard IEC 62366 describes a process for the establishment and implementation of usability engineering for all medical devices. A systematic usability evaluation and risk analysis enables the development team to identify risks and hazards in products and (manufacturing and usage) processes early developmental phases and to treat them efficiently in order to avoid serious adverse events in the field and to minimise in the end harm for patients, users or third parties. It is meaningful to conduct the risk analysis (which includes the identification and the evaluation of risks) using appropriate methods and tools, which, particularly in the area of diverse and complex Human-Machine-Interaction, are inevitable for the performance of a systematic and scientifically-grounded investigation.

Tools and Methods for Risk Analysis and Usability Assessment The effective implementation of risk management requires not only a systematic approach concerning management principles but also an appropriate methods and tools for the different stages of implementation. Based on the Failure Mode and Effects Analysis (FMEA), a product or a process can be systematically investigated in the early development stages in order to initiate prevention measures (Pfeifer, 2001). There are two forms of the FMEA, the product-FMEA and the process FMEA, which analyses possible errors in the process of product manufacturing (VDA, 2006). In addition, there are special developments such as the Human FMEA analysis, which focuses on potential human errors (Algedri, 2001). The disadvantages of the method-

557

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

ology, mentioned by many practitioners, are high effort and the complexity of the analysis, which increases significantly with a rise in the range of functionality of the analysed system (Pfeifer, 1999). A partial reduction of the effort can be achieved by using appropriate software support. With regard to the triple “error, consequence, cause”, which defines a sequential analysis, the FMEA doesn’t allow to model complex causal failure chains within the investigation. Besides the FMEA, the fault tree analysis (FTA) is one of the most commonly used methods in risk and quality management. The core of the method is the development of the fault tree, based on a previously performed system analysis. In this inductive approach, initially on the basis of potential error, all possible consequences and failure combinations are analysed. Finally the detected errors or malfunctions are rated with regard to the probability of occurrence (Pfeifer, 2001). As with the FMEA, complex relationships between failures, which seem to be time and space independent in their appearance at first sight, cannot be identified. The calculation of component-based error probabilities can be done in product-based experiments, but the difficulty to evaluate humaninduced errors in a realistic dimension to capture and quantify them in a sufficient way, however, remains unanswered. In particular, studies concerning Human-Machine-Interaction of medical experts in the specific clinical context are often only conducted during the clinical testing at the end of the product development - and therefore too late for effective risk management. Due to the complexity of cause-effect relationships, especially in Human-Machine-Interaction, these methods are always associated with compromises. Risk models - like all models - are inevitably simplistic, often isolated considerations of the system and therefore not able to detect inherent influence on the particular error or hazard. As a result of above-described problems, in particular ergonomic aspects have to be already considered in the prospective identification and

558

analysis of risks. Hereby, especially the estimation of error probabilities and influencing factors as well as the availability of suitable tools and methods for effective and efficient risk assessment and the derivation of appropriate countermeasures mean serious problems. The aim of a clinical trial of medical products in the framework of the accreditation process is the gathering of empirical data, which is difficult to obtain in a controlled laboratory environment. Previous attempts to identify human-induced risk potentials in the use of medical devices can be devided in the personal and the systemic approach (Giesa, 2000). In the personal-based approach the error tendency is seen as the individual status of a special person. Troubleshooting therefore focuses on identifying the user’s action leading to the failure. Empirically, however, it has been proven that under the same conditions with different actors similar error patterns occur, therefore the hypothesis of personal accident proneness is not sustainable (Hacker, 1998). In contrast to the personal approach the system-oriented approach defines errors as consequences from a number of systemic triggers, such as an inadequate design of the workplace and work organization (Bubb, 1992). The system-oriented approach by Reason (Reason, 1990) distinguishes active failures of the actors and latent error-promoting system states. The goal in system design must therefore be the avoidance of conditions which provoke human errors. It is shown, however, that the users’ actions are designed to minimise the workload and to test out the limits of acceptable behavior (Rasmussen, Pejtersen and Goodstein, 1994). With the harmonisation of the IEC 62366 (Medical devices - Application of usability engineering to medical devices) standard, which has been coming into effect in the year 2007, the experimental and criteria-oriented usability-evaluation of medical systems are receiving increasing attention. The design engineer has to examine and determine the usability of an envisioned system already in the very early developmental phases,

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

often without being able to interact with a first mock-up. Experimental usability testing and heuristic analysis (conducted by usability experts) offer many advantages for manufacturers, but are still connected to method-related disadvantages. The experimental usability testing provides, in contrast to formal analytic procedures, a potentially high validation level as, in addition to tasks and systemspecific characteristics, also (partly implicit) complex environment and user-specific factors (such as stress, fatigue, awareness, temperature, climate) can be included within the laboratory and field tests by the analysis of e.g. video analysis, physiological data, observations and interviews. However, the existence of an interactive mockup (also necessary for heuristic analysis), and the presence of a representative user group are necessary for the conduction of these tests, which usually cause high time-related and infrastructural costs. Results from usability tests are often acquired too late for a proper integration in the ongoing development process without inducing high costs. Moreover, in general realistic contextspecific stress or emergency situations are in the experimental laboratory setup poorly mapped. In various stages of the developmental process a multiplicity of appropriate methods for the support of usability evaluation can be applied. Model-based approaches can be utilized in the definition and specification phase (Kraiss, 1995). Approaches for modelling user, interaction and system behavior are e.g. cognitive modelling and task analysis. Due to the lack of efficient and easyto-use modelling tools, cognitive architectures are currently mainly used for fundamental research in the area of cognitive psychology. Pertaining to the (cognitive) task analysis, there are several methodologies (e.g. hierarchical task analysis) which can be applied depending on the specific questions and cases. The application of task analysis for supporting user interface design of envisioned and redesigned interactive systems can be a fundamental contribution to enhance human-

centered development (Diaper, 2004). Regarding model-based usability examination approaches, there is often the lack of an application-oriented software tool, although the theoretical framework behind shows well-developed syntax models and structures. The majority of commercially available software tools concerning (cognitive) task analysis enables interactive data storing and provides different task modelling options. Unfortunately none of these tools offers a subsequent failure analysis, in order to analyse potential risks in the use process and to derive accurate information concerning proper design of the user interface. Furthermore, with existing software-based tools/methodologies the detailed modelling of all possible interaction tasks with a complex human-computer-interface (e.g. with reference to the task taxonomy and time dependencies) is not accomplishable. To integrate a model of the user, the system and the interaction tasks and their dependencies, a model-based usability-evaluation software tool has to comply with a lot of methodological and application-specific requirements. In reference to the future of the medical and clinical field (e.g. integration and interoperability in the OR), it is of great importance being able to model the comprehensive functional spectrum of modern (graphical) user interfaces. Furthermore, Human-Human-Interaction has to be taken into account with regards to multi-disciplinary and multi-personnel (OR) work systems. The main focus lies within the detailed modelling of cognitive information processing (e.g. OR personnel and the surgeons during the intraoperative workflow). In conjunction with this, assured information flow (e.g. verbal instructions) and coordination are indispensable as well as observing and controlling various signals from discriminative sources (e.g. bio-physical and image data in the OR) (Woods, 1995). Based on these inspections, evaluation criteria have to be predefined, in order to discover and examine contradictions in the allocation of hu-

559

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

man cognitive resources. Finally, a model-based usability-evaluation tool should be able to model external performance shaping factors (e.g. environmental conditions occurring in the OR). To apply the standards IEC 60601-1-6 for medical electrical equipment and the IEC 62366 for all medical products in the framework of usability evaluation, methods on the basis of specific models are important tools for prospective usability evaluation. These approaches should support the analysis of the stress/strain on the patient and practitioner as well as the resulting risks. By the use of formal-analytic procedures the focus on potential ergonomic deficiencies can be set at the early developmental stage. The iterative process of usability engineering and the effort related to usability experiments can potentially be reduced. In future this may represent a significant cost and time factor for manufacturers when developing new safety-critical (medical) products -especially when qualified subjects are needed for interaction-centered user testing- and thus facilitate the accreditation process.

A NEW MODEL-BASED USABILITY ASSESSMENT TOOL Based on an initial concept, developed in the framework of the BMWi funded project INNORISK (run-time: 07.2006-12.2008), a software-based usability and risk analysis tool has been created at the Chair of Medical Engineering at the RWTH Aachen University. This method supports designers and development engineers of complex medical equipment/systems in conducting a formal-analytical usability evaluation and risk analysis, which can be applied from the specification and definition phase up to the validation phase. The methodology and the corresponding software tool mAIXuse, which are permanently developed further, are based on a twofold modelbased approach. Additionally, a systematic error analysis concerning human-induced failures has

560

been integrated. The novel tool has been created on the basis of several standards for the development of medical devices (e.g. ISO 14971:2006 - Application of risk management to medical devices and IEC 62366:2007 - Application of usability engineering to medical devices) The aim of the formal-analytical approach is to provide an overview of the Human-MachineInteraction process, being helpful for the prospective evaluation of a new system/interface as well as for the redesign process of an existing system. Task models describe the necessary activities of the user, the system and the interaction steps to achieve a specific goal. Complex tasks can be hierarchically divided into sub-tasks. Based on an initial risk analysis (e.g. according to DIN EN ISO 14971 for medical devices) use-oriented process, sequences and their potential impact on the overall process are initially evaluated. The proposed formal-analytic mAIXuse procedure is based on a twofold strategy (Figure 1).

High-Level Task Analysis The first part of our procedure is a methodological extension of the ConcurTaskTree approach (Paternò, Mancini and Meniconi 1997), which provides a graphical syntax for the specification of task models, taking into account defined task categories and temporal relations. Within this initial task analysis, the usability investigator gains a systematic overview of the high-level operations of the user, the system and the interactions which are required to achieve a specific task objective within the task fulfilment. As part of the graphical notation, it is possible to describe different task categories and types as well as various attributes and objects. Within the mAIXuse modelling, there are five task categories. Human-systeminteraction tasks are subdivided into the classical information processing sequence “Perception”, “Cognition” and “Action” in combination with an analysis of human-human-interactions (“HumanHuman-Interaction“). The before-mentioned task

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Figure 1. mAIXuse method for model-based usability assessment and prospective risk analysis

categories describe operations in which human errors can potentially occur. System tasks (“System”) however, are carried out independently by the system/technical product. The category “Abstraction” could include all the previously listed types of tasks and provides a higher-ranking task type (mathematically a parenthesis). Unlike traditional task analysis (Diaper, 2004), within the mAIXuse modelling not only top-down dependencies are presented, but with the help of temporal relations (e.g. sequence, concurrency, disabling, sharing, choice, etc.) coequal tasks are linked with temporal interconnections. The descriptions refer to the form “Task A [operator] Task B”. The operators are used to specify the relationships and dependencies between tasks. Examples for temporal relations and their corresponding symbols are shown in Table 1. The relationship between tasks and their subtasks are modelled in a net-like structure. The number of subtasks per task is arbitrary. Each sub-task is assigned to a higher-level task, but

may also have subtasks. Using the above-mentioned modelling syntax, process states arise, which define the task execution order of subtasks by various temporal operators. If the interaction process information has been entirely modelled, it is necessary to determine whether a “parent” task or a “child” subtask can be affected by potential errors and which consequences could result. It must be examined within each sub-task, how a deviation from the planned course of action may cause an error-prone Table 1. Excerpt of temporal relations and corresponding coding symbols Concurrency with information exchange

T1 |[ ]| T2

Sequence with information exchange

T1 []>> T2

Choice

[T]

Return Iteration

T1↖ T2

Finite Iteration

Ti

T*

561

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

situation. To identify human-induced errors the task categories “Perception”, “Cognition” and “Human-Human-Interaction” have to be analysed. The task category “System”, if necessary, may be analysed in a product-related risk analysis (e.g. FMEA, FTA), in order to identify and assess potential system-inherent technical errors or failures in the manufacturing process. For the usability evaluation, however, we will assume that the system/device works according to its technical requirements. To detect potential errors early in the user interface design process a table of potential human errors (classification according to its external appearance) is given. A classification of defects due to the external appearance is defined as an error that occurs when directly interacting with the system (in the category “Action”). The occurring errors are classified into omission, execution and additional failures, whereas execution failures are further subdivided into selection, time, qualitative, sequential and mix-up failures. To identify potential human-induced errors in Human-Machine-Interaction and Human-Human Interaction, first the motor actions in the task category “Action” are filtered and analysed regarding errors between different tasks (inter-task error) or within the respective tasks (intra-task error). Furthermore, potential causes of the abovementioned failures are defined for the cognitive information processing and the perception process (information acquisition). A possible cause of error e.g. lies within the cognitive information processing. For this purpose, a failure checklist has been derived on the basis of the cognitive modelling system called the Generic Error Modelling System. The Generic Error Modeling System (GEMS) is a framework for understanding error cause types and designing error prevention strategies. This model is based on the cognitive psychological approach by Reason (Reason, 1987). Here, the cognitive information process of human beings is divided into three levels of regulation (skill-, rule- and knowledge-based). The skill-based performance

562

mode is characterized by routine actions in familiar settings. The rule-based performance mode is characterized by pre-defined actions performed because of the recognition of a similar situation. The rule-based performance mode requires an individual to use analytical skills and judgment to complete a task in an unfamiliar situation. Each cognitive processing level is analysed for potential failures in the specific task step of the use process. Potential error causes in a cognitive regulation level can then semi-automatically be integrated into an FMEA form sheet and finally evaluated. Additionally, on the basis of relevant standards (e.g. 9241-12) a checklist for the analysis of potential error causes in the acquisition and perception of information has been developed. Due to the provision of this checklist potential sources of errors and guidelines for the interface design can be derived. The procedure of analysing potential errors and failure causes in the information processing of the human being are shown in Figure 2, additionally exemplary modelling with the software tool is presented here. An exemplary excerpt of the comprehensive modelling and analysis of the use process with a surgical planning and navigation system, developed at the Chair of Medical Engineering, is shown in Figure 3. Figure 4 shows an excerpt of the model of the use process with two dialogues of the planning and navigation system for hip surface replacement. Here the different task categories and the temporal relations between various tasks are depicted. Next to the different process steps potential humaninduced failures are integrated and listed.

Low-Level Task Analysis In the second part of the mAIXuse analysing process, a cognitive task model based on the initial high-level task analysis is created. Here, parallel tasks concerning cognitive, perceptual and motor low-level operators are decomposed.

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Figure 2. Systematic failure analysis on the basis of derived checklists concerning the information processing and visualization of modelling structure

Figure 3. Excerpt of the mAIXuse modelling and analysis using the example of the use process with two GUI dialogues of a planning and navigation system for hip surface replacement

Breaking down the previously detected high-level decompositions into low-level operators eases the acquisition of cognitive processing steps involved. However, the success of this step is depending on the accuracy of the previous high-level task modelling.

Following this approach, a methodological extension of CPM-GOMS (Cognitive Perceptual Motor – Goals Operators Methods Selection Rules) (John and Gray, 1995), a cognitive task analysis, based on the Model Human Processor (MHP) (Card, Moran and Newell, 1983), can be

563

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Figure 4. Excerpt of the modelling of the use process with two dialogues of the planning and navigation system for hip surface replacement

conducted subsequently to the high-level task analysis. The MHP architecture is developed on the basis of a computer-oriented architecture and integrates various categories of repository (visual and auditory cache, working and long-term storage). Furthermore, the average processor cycle times are specified by several rules (e.g. Hick’s law, power law of practice, Fitt’s law etc.). Here, concurrent-working processors are able to model low-level operators (perceptual, cognitive and motor-driven) in a parallel mode, in order to generate an exact and detailed overview of the information processing tasks of the human operators. Especially the parallelism between the processors/operators can be helpful in order to model and analyse multimodal Human-MachineInteractions, typically occurring in the context of clinical work systems. Dependencies between the users’ perceptual, cognitive and motor-driven activities are mapped out in a schedule chart (PERT - Program Evaluation Review Technique), where the critical path represents the minimum required execution time. John and Gray have provided templates for cognitive, motor-driven and perceptual activities alongside their dependencies under various conditions (John and Gray, 1995). In project Ernestine

564

CPM-GOMS has been validated to real-world task performance by assessing Human-MachineInterfaces of telephone companies’ workstations (Gray, 1992). Additionally, within this cognitive task analysis approach different cognitive task steps are matched with appropriate cognitive processing levels. In the course of these analyses potential contradictions and conflicts in concurrent cognitive resource utilisation can be detected and several risks and hazards in Human-Machine-Interaction can be identified and assessed. Supplementary, the second part of the methodology developed includes, in contrast to original CPM-GOMS, the distinction of skill-, rule- and knowledge-based behavior according to Rasmussen (Rasmussen, 1983) in the PERTChart. Moreover, external performance shaping factors are currently being integrated, in order to provide supplementary qualitative evaluation criteria specifically taking into account the clinical context of use. Although, time performance predictions can only be made for primitive skill-based operators without special contextual PSF’s, there is a great benefit concerning the detailed overview. In addition, a process bar is integrated into the PERT Chart in order to transcribe human and system

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

activities (obtained in the high level analysis) to facilitate decomposition of high-level tasks and to create additional help for the user. The inclusion of these new features into the low level task modelling provides more detailed qualitative evaluation criteria and an extensive overview of the complete interaction. In combination with existing quantitative statements on learning and operating time, as well as on failure rates and efficiency, the inclusions mean a powerful extension of CPM-GOMS. In addition, a failure analysis on the basis of above-mentioned failure taxonomies will also be possible after the cognitive task analysis.

Experiences with the New Approach The mAIXuse method and the corresponding software tool have been evaluated in several workshops with industrial partners. Systems investigated in these workshops have been e.g. a shockwave therapy system, planning and navigation systems for orthopaedic- and neurosurgery, a robotic system for orthopaedic surgery, a pacemaker and the corresponding pacemaker system analyser in the context of cardio-electrotherapy systems as well as a semiautomatic safety trepanation system for neurosurgery. The workshops (participants have been employees from different departments: risk management, quality management, design, verification, validation, marketing, etc.) have initially been paper-based. Further details have been reported in (Schmidt et al., 2009). Meanwhile, a first software version has been implemented and the partners used this software tool during the workshops. As a result of the very positive experience with the mAIXuse methodology and the corresponding software tool several partners continued to use the mAIXuse tool for the evaluation and design of human-machineinterfaces. Furthermore, the mAIXuse method and tool have been investigated in comparison to classic risk analysis methods concerning efficiency, ef-

fectiveness and usability aspects in the framework of the risk analysis with an existing planning and navigation system for hip resurfacing in orthopaedic surgery. This tool is currently being developed at the Chair of Medical Engineering. The results of the mAIXuse application in a two days workshop with different test groups have been compared to the results with a conventional risk analysis method, the Process-FMEA (Failure Mode and Effect Analysis). The hypothesis was that with the new method more and risk-sensitive failures should be found within a risk investigation in the same time in comparison to the conventional method. Two test groups, each including three test subjects (age between 24-30), had to apply the Process-FMEA and the mAIXuse tool on the same system/interface in varying order, so as to exclude dependencies and correlations between working time and results. Application time for each analysis has been set up to 2½ hours. Two consecutive dialogues of the computer-assisted planning of the implant-based resurfacing of an affected femoral head had been investigated regarding potential use errors. In the assessed computer-assisted orthopaedic intervention, a high precision planning and execution is essential in order to achieve a sustained therapeutic outcome. The corresponding planning and navigation system is currently being developed at the Chair of Medical Engineering. The dialogues “Positioning of the safety zone for the femoral neck” and “Positioning of the implant”, which are conducted interactively using a graphical user interface, have been modelled and analysed concerning (especially risk-sensitive) potential human failures. Finally the subjects had to answer a questionnaire in order to evaluate both methods relatively and absolutely concerning user-satisfaction, understandability and learnability. Objective and subjective test data are shown in Table 2. Results show the advantages of the mAIXuse method in contrast to the conventional Process-

565

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Table 2. Evaluation results of the comparison between the mAIXuse and the Process-FMEA application Group A

Group B

Criteria

FMEA

mAIXuse

mAIXuse

FMEA

Detected risks

14

27

29

16

Critical risks

5

9

12

7

Application time

2½ hours

2½ hours

2½ hours

2½ hours

Evaluation (1 = very good, 2 = good, 3 = satisfactory, 4 = adequate, 5 = poor) User satisfaction

3-4

2

1-2

4

Understandability

2-3

1-2

2

3

Intuitiveness

3-4

2

1-2

4

Learnability

3

2

1-2

3

FMEA. The new modelling and evaluation tool shows better results regarding all subjective and objective criteria. Usability aspects as user satisfaction, understandability, intuitiveness and learnability have been rated much better for the mAIXuse approach then for the FMEA. Concerning detected (critical) risks mAIXuse clearly outperforms the FMEA. Surprisingly, when the subjects had worked in the first session with the mAIXuse tool, afterwards they were not even able to find an equal number of risks respectively critical risks with the process FMEA, although the subjects then in summary worked twice as much time with the planning and navigation system. Analysis of the questionnaires and the discussion with the subjects revealed, that the predefined lowest level of the modelling technique and the use of temporal relations in the mAIXuse approach are responsible for its detailed modelling analysis and therefore for its higher effectiveness and efficiency. The results of the questionnaire and the objective test data show that mAIXuse supports the user and eases the application process. Objective results are corroborated by the results of the questionnaire. Even the cost benefit evaluation has been rated as “good”, because almost two times more potential system-inherent failures have been

566

detected with the mAIXuse method than with the Process-FMEA. Additionally, the application of the software-assisted tool doesn’t show a reduction of the communication or the collaboration of the test subjects. This estimated negative effect has not been observed during the evaluation. On the contrary, the software tool supports the efficient and effective analysis and the comprehensive overview of the modelling. The described new method for prospective usability-evaluation and the corresponding software tool have proven their applicability and the practicability in various investigations. After a short introduction concerning the application of the methodology all test subjects have been able to successfully use the mAIXuse method and the corresponding software tool on their own. A comparison of the modelling results of the interaction process concerning the same user interface with different test groups shows nearly identical outcome. This leads to the hypothesis that, because of the predefined lowest level of modelling and the integration of temporal relations, a standardised modelling structure has been developed. The way of coding and presenting is intended to support designers and engineers with the internal communication and to ease the understanding of the specific Human-Machine-In-

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

teraction. Furthermore the mAIXuse tool supports medical device developers and manufacturers in conducting a model-based usability-evaluation and a usage-oriented risk analysis on their own. The standardised documentation required for the risk management and the usability-engineering process, is automatically generated in a FMEA form-sheet. When using the mAIXuse technique for modelling a prototypical interface an interesting and productive side-effect has been observed. In order to model the current state of the use process regarding the prototypal system with the mAIXuse tool, the test subjects have shown explorative test behaviour. The combination and integration of explorative user tests in the framework of the mAIXuse modelling process show a promising approach for an enhanced usability evaluation method, extending the conventional Cognitive Walkthrough approach.

FUTURE RESEARCH Effective and efficient usability engineering and risk management will have a growing impact on the success of medical device manufacturers in the future. The increasing complexity and functionality of medical devices together with the need for interoperability and integration of modern work systems (e.g. in modern operating theaters) bring along additional system-inherent risks and potential use problems. Thus, cognitive engineering is becoming more and more a part of medical equipment design, enabling the improvement of patient safety measures. A systematic consideration of cognitive information processing in Human-Machine-Interaction regarding medical work systems, particularly within the scope of multifunctional applications, is a special requirement for usability assessment in the future. Particularly, ergonomic quality and reliability will become important factors for reducing safety-critical risks in the medical field.

A prospective systematic and methodological acquisition of all task steps/types and subsequently of potential human errors in a formal-analytical modelling and simulation can be expedient and useful for medical device manufacturers, as it allows a more comprehensive and earlier overview on potential risks and bottlenecks than user-interactive tests with the final medical device. An early usability evaluation can be a significant advantage with regard to saving time and resources. Therefore, easy-to-handle methods and all-purpose tools are needed in order to fulfill the requirements of normative and legislative regulations in usability engineering and risk management. In general, user-centered design and cognitive engineering will become more and more important especially in the medical field. Particularly in this area, there will be an increased emphasis on research, trying to understand and predict the specific contextual requirements and needs of medical users of (risk-sensitive) systems. Early cognitive modelling and simulation of an envisioned system or of a prototypical interface will become more important in order to minimize additional time and financial costs related to summative usability tests. Emerging nations are developing quickly also on the medical technology market and there will be a rapid evolution in business. In this context usability and safety will become selling points of increasing importance.

CONCLUSION The usability engineering process as a basis for developing usable and error-tolerant user interfaces should start already during the definition and specification phases and continue through every phase of the product life cycle. User-centered usability tests (e.g. in special usability labs) are a mandatory element of the approval process for medical products. There is no doubt that a lot

567

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

of potential hazards and derived solutions have been identified through the use of such labs and that these tests are an important part of usability improvement. However, the costs of running such tests sometimes significantly increase the overall cost of the system/device development. One possible solution to reduce these costs could be the early and consistent application of software-based modelling and simulation tools. With a-priori data, the usability assessment and reporting can be improved and expedited. Due to the lack of efficient and easy-to-use modelling tools, cognitive architectures have been mainly used for fundamental research in the area of cognitive psychology. Future research in human-centred design will claim the development of easy-to-apply cognitive architectures, in order to model and simulate the interaction process in detail. The motivation for cognitive engineering and for developing new prospective usability-evaluation tools is to provide adequate models of human performance for designing and assessing user interfaces already in the very early developmental stage. In this framework the developed software tool mAIXuse provides useful information (e.g. overview of tasks and relations, potential use errors and design rules) in the specification, definition and validation phase. Taking into account the mental workload e.g. in clinical interventions, the mAIXuse approach is intended to be a useful approach for modelling cognitive information processing in Human-Machine-Interaction, especially in complex risk-sensitive medical work systems. Nevertheless, interface and technical designers shall be supported methodologically with the help of a reliable as well as easy-to-use analysis tool, in order to systematically identify potential hazards related to Human-Machine-Interaction. Particularly, manufacturers of medical devices and especially small and medium-sized enterprises (SME), which represent the majority of enterprises in the branch of medical device industry could benefit from such a standardised coding and modelling approach concerning the design

568

process of medical device interfaces. Thus, SME shall be supported within the application of the usability-engineering and risk management process as crucial elements of the device approval.

REFERENCES Algedri, J., & Friehling, E. (2001). Human-FMEA - Menschliche Handlungsfehler erkennen und vermeiden. Carl Hanser Verlag München Wien. Bogner, M. S. (1994). Human Error in Medicine. Mahwah, NJ: Lawrence Erlbaum Associates. Bubb, H. (1992). Menschliche Zuverlässigkeit. Landsberg/Lech: ecomed. Card, S. K., Moran, T. P., & Newell, A. (1983). The psychology of Human-Computer-Interaction. Hillsdale, NJ: Lawrence Erlbaum Associates. Cook, R. I., & Woods, D. D. (1996). Adapting to the new technology in the operating room. Human Factors, 38, 593–613. doi:10.1518/001872096778827224 Diaper, D., & Stanton, N. (2004). The Handbook of Task Analysis for Human-Computer Interaction. London: Lawrence Erlbaum Associates. Gawande, A. A., Thomas, E. J., Zinner, M. J., & Brennan, T. A. (1999). The incidence and nature of surgical adverse events in Colorado and Utah in 1992. Surgery., 126(1), 66–75. doi:10.1067/ msy.1999.98664 Giesa, H. G., & Timpe, K. P. (2000). Technisches Versagen und menschliche Zuverlässigkeit: Bewertung der Zuverlässigkeit in Mensch-MaschineSystemen. In Timpe, K.-P., Jürgensohn, T., & Kolrep, H. (Eds.), Mensch-Maschine-Systemtechnik. New York: Symposion Publishing. Gray, W. D., John, B. E., & Atwood, M. E. (1992). The precis of Project Ernestine or an overview of a validation of GOMS. Proceedings of CHI,pp.307-312. New York: ACM.

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Hacker, W. (1998). Allgemeine Arbeitspsychologie. Bern: Huber. Hansis, M. L. (2002). Medizinische Behandlungsfehler in Deutschland. Schleswig-Holsteinisches Ärzteblatt, 06(02), 51–59. Hansis, M. L., Hart, D., Hansis, D. E., & BeckerSchwarze, K. (2001). Gesundheitsberichterstattung des Bundes. Heft 04/01: Medizinische Behandlungsfehler. Robert-Koch-Institut. Hoffmann, K. (1985). Risk Management – Neue Wege der betrieblichen Risikopolitik. Karlsruhe: Verlag Versicherungswirtschaft. Hyman, W. A. (1994). Errors in the Use of Medical Equipment. In Bogner, M. S. (Ed.), Human Error in Medicine. Hillsdale, NJ: Lawrence Erlbaum. John, B. E., & Gray, W. D. (1995). CPM-GOMS - An Analysis Method for Tasks with Parallel Activities. In: Conference companion on Human factors in computing systems, pp. 393-394. New York: ACM Press. Kindler, M. (2003). Risikomanagement von medizintechnischen Geräten bei Entwicklung, Zulassung und Anwendung. Weiterentwicklung der Arbeitsprozesse in der Krankenhaustechnik. Strategien, Administration, Kooperation. Kongress TK. Kohn, L. T., Corrigan, J., & Donaldson, M. S. (2000). To Err Is Human: Building a Safer Health System. Washington, DC: National Academy Press. Kraiss, K.-F. (1995). Modellierung von MenschMaschine Systemen. In Willumeit, H.-P. und H. Kolrep (Eds.). ZMMS-Spektrum, Band 1, Verlässlichkeit von Mensch-Maschine-Systemen. S. 15–35. Pro Universitate Verlag. Leape, L. L. (1994). The Preventability of Medical Injury. In Bogner, M. S. (Ed.), Human Error in Medicine. Hillsdale, NJ: Erlbaum Publ.

Merten, M. (2007). Risikomanagement: Den Ursachen auf der Spur. Deutsches Ärzteblatt. Deutscher Ärzte-Verlag GmbH. Paternò, F., Mancini, C., & Meniconi, S. (1997). ConcurTaskTrees: A Diagrammatic Notation for Specifying Task Models. Proc. of IFIP Int. Conf. on Human-Computer Interaction Interact ‘97, pp. 362–369. London: Chapman & Hall Pfeifer, T. (2001). Qualitätsmanagement: Strategien, Methoden, Techniken. München, Germany: Carl Hanser Verlag. Radermacher, K. (1999). Computerunterstützte Operationsplanung und -ausführung mittels individueller Bearbeitungsschablonen in der Orthopädie. In Rau, G. (Ed.), Berichte aus der Biomedizinischen Technik. Aachen: Shaker. Radermacher, K., Zimolong, A., Stockheim, M., & Rau, G. (2004). Analysing reliability of surgical planning and navigation systems. In Lemke, H.U., & Vannier, M.W. (Eds.), International Congress Series 1268, 824-829. Rasmussen, J. (1994). Skills, Rules, Knowledge: Signals, Signs, Symbols and other Distinctions in Human Performance Models. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3, 257–267. Rasmussen, J., Pejtersen, A. M., & Goodstein, L. P. (1994). Cognitive Systems engineering. New York: Wiley-Interscience Publications. Rau, G., Radermacher, K., Thull, B., & v. Pichler, C. (1996). Aspects of Ergonomic System Design Applied to Medical Worksystems. In Taylor, R. H. (Ed.), Computer-integrated surgery: technology and clinical applications (pp. 230–221). Cambridge, MA: MIT Press. Reason, J. (1990). Human Error. Cambridge, UK: Cambridge University Press.

569

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Reason, J. T. (1987). Generic error-modelling system (GEMS). A cognitive framework for locating human error forms. In Rasmussen, J., Duncan, K., & Leplat, J. (Eds.), New technology and human error. London: Wiley.

Hollnagel, E. (1991). The phenotype of erroneous actions: Implications for HCI design. In Weir, G. W. R., & Alty, J. L. (Eds.), Human-computer interaction and complex systems. New York: Academic Press.

Schmitt, R. H., Rauchenberger, J., Radermacher, K., Lauer, W., & Janß, A. (2009). Neue Wege für das Risikomanagement bei der Entwicklung risikosensitiver Produkte am Beispiel der Medizintechnik. In FQS (Eds.) FQS-DGQ Band 88-04, Beuth Verlag.

Hollnagel, E. (1993). Human reliability analysis: Context and control. New York: Academic Press.

VDA. (2006). Sicherung der Qualität vor Serieneinsatz - Produkt- und Prozess-FMEA. Henrich Druck und Medien. Woods, D. D., Cook, R. I., & Billings, C. E. (1995). The impact of technology on physician cognition and performance. Journal of Clinical Monitoring, (11): 5–8. doi:10.1007/BF01627412 Zimolong, A., Radermacher, K., Zimolong, B., & Rau, G. (2001). Clinical Usability Engineering for Computer Assisted Surgery. In Stephanides, C. (Eds.), Universal Access in HCI – Towards an Information Society for All, pp. 878-882. Mahwah, NJ:Lawrence Erlbaum Ass. Publ., Zweifel, P., & Eisen, R. (2000). Versicherungsökonomie. Berlin:Springer.

ADDITIONAL READING

Hollnagel, E. (1998). Cognitive reliability and error analysis method: CREAM. New York: Elsevier. Holzinger, A. (2005). Usability Engineering for Software Developers. Communications of the ACM, 48(1), 71–74. doi:10.1145/1039539.1039541 Kirwan, B. (1994). A Guide to Practical Human Reliability Assessment. New York: Taylor & Francis. Kirwan, B., & Ainsworth, L. (Eds.). (1992). A guide to task analysis. New York: Taylor & Francis. Nielsen, J. (1993). Usability Engineering. Boston: Academic Press. Norman, D. (1988). The psychology of everyday things. New York: Basic Books. Paternò, F. (2000). Model-based design and evaluation of interactive applications. New York: Springer. Rasmussen, J. (1986). Information processing and human-machine interaction: An approach to cognitive engineering. New York: Wiley.

Davies, J. B., Ross, A., Wallace, B., & Wright, L. (2003). Safety Management: a Qualitative Systems Approach. New York: Taylor and Francis. doi:10.4324/9780203403228

Reason, J. (1990). Human error. Cambridge, UK: Cambridge University Press.

Gaba, D. M., Fish, K. J., & Howard, S. K. (1998). Zwischenfälle in der Anästhesie – Prävention und. Management. München, Germany: Elsevier.

Vicente, K. J. (1999). Cognitive Work Analysis: Toward safe, productive, and healthy computer based work. Mahwah, NJ: Erlbaum.

Hacker, W. (1995). Arbeitstätigkeitsanalyse Analyse und Bewertung psychischer Arbeitsanforderungen. Heidelberg: Roland Asanger Verlag.

570

Salvendy, G. (1997). Handbook of human factors and ergonomics. New York: Wiley.

Using New Model-Based Techniques for the User Interface Design of Medical Devices and Systems

Williams, J. C. (1985) HEART – A proposed method for achieving high reliability in process operation by means of human factors engineering technology. Proceedings of a Symposium on the Achievement of Reliability in Operating Plant, Safety and Reliability Society. NEC, Birmingham. Woods, D. D. (1990). Modeling and predicting human error. In Elkind, J., Card, S., Hochberg, J., & Huey, B. (Eds.), Human performance models for computer-aided engineering (pp. 248–274). New York: Academic Press. Woods, D. D., Johannesen, L., Cook, R., & Sarter, N. (1994). Behind human error: Cognitive systems, computers, and hindsight. CSERIAC SOAR Report 94-01. Crew Systems Ergonomics Information Analysis Center, Wright-Patterson Air Force Base, Ohio.

KEY TERMS AND DEFINITIONS Cognitive Engineering: Cognitive engineering is a multidisciplinary field concerned with the analysis, design, and evaluation of HumanMachine-Interfaces. It combines knowledge and

experience from cognitive science, human factors, human-computer interaction design and systems engineering. Cognitive Modelling: Cognitive modelling is an area of computer science which deals with simulating human problem solving and mental task processes in a computerized model. Such a model can be used to simulate or predict human behavior or performance on tasks. Cognitive models can be used in human factors engineering and user interface design. Human-Machine-Interaction: An interdisciplinary field which focuses on the interactions between users and systems, including the user interface and the inherent processes. Use Error: The use error occurs due to latent error-promoting system states. Use-Oriented Risk Analysis: A risk analysis, which supports designers and assessors to investigate use deficiencies with Human-Machine-Interfaces, in contrast to system and (manufacturing) process risk analysis. User-Centered Design: User-centered design involves simplifying the structure of tasks, making things visible, getting the mapping right, exploiting the powers of constraint, and designing for error.

This work was previously published in Human-Centered Design of E-Health Technologies: Concepts, Methods and Applications, edited by Martina Ziefle and Carsten Röcker, pp. 234-251, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

571

572

Chapter 2.22

The European Perspective of E-Health and a Framework for its Economic Evaluation Paola Di Giacomo University of Udine, Italy

ABSTRACT E-health is a priority of the European i2010 initiative, which aims to provide safe and interoperable information systems for patients and health professionals throughout Europe. Moreover, the use of electronic storing and transmission of data to patients is increasing while through the deployment of e-health applications, health care is improved in terms of waiting time for patients. The concentration results from the cumulative DOI: 10.4018/978-1-60960-561-2.ch222

incidence of chronic-degenerative pathologies, the greater utilization of biomedical technologies, and the increased health services demand. Finally, the interest towards electromechanical systems means the realization of tools of small dimensions, which have tremendous advantages thanks to their invasivity and greater diagnostictherapeutic effectiveness. Therefore, an economic analysis has to take into consideration the use of biomedical technology, the analysis of alternatives, the selection of the economic evaluation technique, and the identification and quantification of the costs and benefits.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

The European Perspective of E-Health and a Framework for its Economic Evaluation

INTRODUCTION Innovation is essential to improve accessibility, effectiveness and efficiency of healthcare delivery. E-health promises these improvements, provided we comply with fundamental requirements with respect to quality and safety. E-health must be implemented thoughtfully to provide the maximum advantage of the innovation. However, there exists no structured framework of the basic requirements of quality and safety issues. This often hampers development, implementation and usage. Therefore, a framework of quality and safety requirements must evolve to support and encourage innovation. The analysis of an innovation, which contributes to quality, safety and efficiency, leads to an evaluation process, which should be followed to assess an e-heath application. The issue of minimum requirements is a political one while the requirements themselves are the subject of research. Furthermore, one problem in addressing this issue is the mere fact that e-health is under development and that it can assume various forms and sizes. It is, therefore, hardly likely to formulate all requirements in advance. For each service offered to the market, it is anticipated that different aspects will have to be assessed while a rigorous program of requirements will make innovation difficult. It is worthwhile mentioning that e-health is a priority of the European i2010 initiative, which aims to provide safe and interoperable information systems for patients and health professionals throughout Europe. Based on a pan-European survey on electronic services in healthcare, 87% of general practitioners use a computer and 48% have a broadband connection. Moreover, the use of electronic storing and transmission of data to patients is increasing while through the deployment of e-health applications, health care is improved in terms of waiting time for patients. There is, however, considerable space for improvement since ICT assists many aspects of the doctor-patient

relationship. This includes remote monitoring services (used in Sweden, the Netherlands and Iceland), electronic prescriptions (used by only 6% of EU general practitioners, and only three Member States, which include Denmark, Netherlands and Sweden), and medical care, practiced by only 1% of general practitioners, with the highest percentage in the Netherlands (5%). The survey also shows that e-health services are used where the broadband Internet use is widespread. In Denmark, for example, where the penetration of broadband Internet is the highest in Europe, 60% of doctors are currently exchanging emails with patients. On the contrary, the EU average is stuck at 4%. Overall, despite considerable progress, there are still notable differences between different countries. The main barriers to adoption of new technologies in health care include the lack of training and technical support. To encourage the routine use of ICT in health care and accelerate the adoption of appropriate strategies at the country level, the Commission adopted an action plan for e-health, under the title ‘Lead Market Europe’ (Commission of the European Communities, 2007). The analysis aimed at the cost-benefit of the patient treatment, an essential part of the quality evaluation, is widely documented in healthcare literature. The focus results from the cumulative incidence of chronic-degenerative pathologies, the greater utilization of biomedical technologies, and the increased health services demand. The method is based on the formulation of testable criteria and standard values with respect to the particular parts of the service. Therefore, it can determine the criteria, which are not met, leading to a provisional acceptance of the e-health application. Methods of comparative analysis, like the cost-benefit approach have to be adapted to the various aspects of the healthcare process, in order to determine the advantage of a traditional model of care with respect to an integrated one, which is based on new computer science technologies. Among the new methods of book keeping that

573

The European Perspective of E-Health and a Framework for its Economic Evaluation

operate in this direction, the most widely used is the Activity Based Costing (ABC). ABC is a method of imputation of the costs incurred by the activities of the cost centers. Therefore, every reengineering decision can be isolated, and estimated in relation to the cost that it incurs. In this chapter, we will provide an overview of the European strategy in e-health, the research programs and initiatives of the European Union, as well as a framework for the economic evaluation of e-health applications (Drummond et al., 2005; Scrivens, 1997; Hughes & Humphrey, 1990).

BACKGROUND Commissioner for Health and Consumer Protection Markos Kyprianou said: ‘e-health can improve health care. Even more important, it is possible to reduce medical errors and save lives. We need a partnership between the ministers of health, technology providers, patient associations and NGOs to realize the full potential of e-health in Europe’. ‘The EU-US e-health is important because there are large economic areas with the same characteristics, such as the aging population. We need to coordinate the development of standards and interoperability in this area,’ said the General Director of the Commission for Information Society, Fabio Colasanti. He also clarified that the technology itself is not enough. It needs to be accompanied by an appropriate legal environment and education of health professionals, said Frans de Bruin of the EU Directorate General for Information Society: ‘The e-health should no longer be a subject of specific conferences, but simply the normal way of doing health care,’ said Petra Wilson, Director for the public health sector and Cisco Internet Business Solutions Group. ‘We need to give incentives to doctors to encourage them to adopt and use technology.’ Peter Langkafel of SAP agrees that the incentive of physicians is

574

important, stressing that it is not enough: ‘You have to demonstrate value and a business case for e-health - a better quality of care and patient safety, as well as improved use of resources in health care (human resources, processes). The business case must be presented with the costbenefit analysis’. ‘Why should not patients have the same type of service in health care such as in the banking sector? (...) Patients should contact their health care providers and start putting pressure on them claiming greater use of technology’, said Baldur Johnsen, Hewlett Packard Director of Development of health services in the e-health market. . ‘The one devoted to health is the segment with the highest growth rate in our society,’ according to Charles Scatchard, Vice-president of health sciences at Oracle, ‘... and we also know that all new health technologies need a period of 17 years to finally be usable’. ‘The e-health helps to overcome the obstacles related to the distance of the organization and the delivery of health services and is, therefore, an important tool for both rural and urban areas. The development of these technologies contributes to the development of a global economic region, attracting firms specializing in this area and the creation of new employment opportunities, ‘said Agneta Granström, Advisor of Norrbotten. ‘We must begin with the strengthening of cooperation between hospitals and between other sectors of health care across Europe. We do not need a single system, the same for everyone, but interoperable systems. Both universities and companies should fully appreciate the enormous potential of e-health and provide the necessary tools along with the regions that should work for the development of e-health’, he added. The Group of Pharmacists of the European Union (PGEU) calls the national and European authorities to help pharmacists to understand the advantages and opportunities of the greater use of data from the Internet and the development of e-health applications. PGEU promotes the development of European standards and the certification of P2P

The European Perspective of E-Health and a Framework for its Economic Evaluation

applications, including the necessary standards for the electronic transmission of prescriptions. The standards should help all stakeholders to expand existing services and draw medical attention to the security of the Internet applications. Finally, in 2007, the European Data Protection Statement approved a working document that discusses the legal requirements and application parameters for the structure and management of a national system of electronic health records. The document is open for public consultation to solicit contributions and comments.

THE EUROPEAN STRATEGY IN E-HEALTH ‘E-health’ implies e-health applications of Information and Communication Technologies (ICT) for health, which encompass hospital doctors, patients and social security data processing. The goal is to improve the quality, access and efficiency of health services for all. In the EU, e-health plays a key role in the strategy for action on innovation, which was launched in 1999. A series of action plans followed in 2002 and in 2005(eEurope 2005: An information society for all). The Action Plan i2010 - A European Information Society for growth and employment, was followed by the Communication ‘Preparing Europe’s digital future i2010: Mid-Term Review’ in April 2008. Regional networks of health, electronic health records, primary care and deployment of electronic health cards have contributed to the rise of ehealth, which has the potential to become the third largest industry, after the pharmaceutical industry and the medical imaging. In particular, for the period 2006-2009, Member States and the Commission promoted interoperability and encouraged greater standardization. They then set the target to establish the basis of European E-health services by the end of 2009.

The key stages included: December 2005: E-health stakeholder’s group meets for the first time. The group is an organization that brings together the leading European actors involved in the standardization (such as CEN - European Committee for Standardization, CEN / ISSS - Information Society Standardization System, CENELEC - Comité Européen de Normalization Electrotechnique, ETSI - European Telecommunications Standards Institute), industry associations, and user groups. June 2006: the Assembly of European Regions (AER) is organizing an international conference and presents a session on the e-health (Thematic Dossier of the Assembly of European Regions). June 2006: The Commission for Health Information Technology adopted a new strategy (ICT for Health and 2010, ‘Transforming the European healthcare landscape: Towards a strategy for ICT for Health’) to promote the transformation of the European health systems. The strategy argues that in order to meet the challenges of aging, Europe needs a new model of health care, which should be based on prevention and people-centeredness. The approach involves the E-health Action Plan and includes research under the Seventh Framework Program for Science and Research. The strategy is also in line with the communication of the new framework of i2010 (i2010 Annual Information Society Report 2007), which promotes the application of ICT to improve social inclusion, public services and quality of life. October 2006: The results of the study on the impact e-health are published (economic benefits of the e-health implementation in several European sites). April 2007: Conference and exhibition on e-health April 2007: A report argued that Member States have made considerable progress in implementing the strategy for e-health, but had failed to establish educational and socioeconomic guidelines for matters falling under their responsibility.

575

The European Perspective of E-Health and a Framework for its Economic Evaluation

May 2007: Workshop on e-health, which was organized by the European Commission and the U.S. Department of Health & Human Services, in collaboration with the European-American Business Council. The main issues included the health practices which were supported by information technology, interoperability, certification and improvement of patient safety July 2007: Presentation of a draft recommendation on interoperability December 2007: The Task Force on e-health publishes a report on Accelerating the Development of the E-health Market in Europe. December 2007: Publication of a kick-off study by the EU Commission on financing e-health. The Commission adopts a Communication on the action guide for the European market (‘A lead market initiative for Europe’), with e-health as one of the top six markets. April 2008: A study provides an overview on e-health emerging markets regulation and a report on the use of ICT among general practitioners in Europe. May 2008: The EU Portorož Declaration on the e-health is adopted. It reaffirms the need to share experiences and collaborations carried out in Europe, and the terms of interoperability for cross-border care. June 2008: Two European pilot projects are launched to check the European cooperation in the use of data relating to emergencies and prescriptions. In 2008, the Commission issued a Recommendation on cross-border e-health systems. By the end of the year, the Commission had also planned to publish a Communication on telemedicine and innovative ICT solutions for the management of chronic diseases. It is worthwhile mentioning that the Action Plan for 2004-2010 focused on three priority areas: •

576

Dealing with common challenges and creating an appropriate framework to support e-health, such as the interoperability

•

•

of health information systems, the electronic health records, and patient and staff mobility Accelerating the implementation of the of e-health use in the areas of health education and disease prevention, and promoting the use of electronic health cards linking state efforts in monitoring, benchmarking and dissemination of best practices

European E-Health Research Projects The EU is supporting research in e-health for nearly two decades. The technologies developed in hundreds of successful projects have helped to improve health care provision in many different fields. Research on e-health is a priority in the Seventh Framework Program, which is valid until 2013. The Seventh Framework Program for Research (presented in the Decision of December 18, 2006) will continue to this direction until 2013, with a total budget more than 50 billion euros. It aims to create an ‘intelligent environment’ in which the state of health of each individual can be monitored and managed continuously. The key points of the program on e-health are the following: • • •

•

Personal Health Systems (PHS) Patient Safety (PS) Software tools, which help health professionals to find the available information each time they have to draw conclusions. Virtual Physiological Human (VPH) Framework

Multidisciplinary networks of researchers in the field of bioinformatics, genomics and neuroinformatics, which create a new generation of e-health systems.

The European Perspective of E-Health and a Framework for its Economic Evaluation

Research Projects Bepro Enabling Best Practices for Oncology (IST2000-25252 BP) The project focused on a multinational network of oncology and assessed the effectiveness of oncology services. The project involved three phases: establishment, medical evaluation, and assessment. The results were disseminated to the most influential European medical communities and, in some cases, to standardization bodies.

Diaffoot Remote Monitoring of Diabetic Foot (IST2001-TR 33281) Diabetes mellitus is a growing problem in European countries with a high impact both on the quality of life of millions of citizens, and the financial situation of national health systems. The technical objective of the project was to test the use of a remote system, capable of measuring data on clients and sending data to the hospital for further analysis and monitoring of patients with diabetes mellitus. Another objective was to compare current procedures and techniques in order to redefine the protocol for the treatment of diabetes mellitus and propose an operational procedure.

age and communication. The digital microscopy could then allow the integration of PACS (Picture Archiving and Communication System), HIS (Hospital Information System) and information available in hospitals.

Ihelp Electronic remote operating room (IST-200133269 BP) The project Ihelp analyzed the state of art of ehealth platforms and investigated the methodology for establishing best practices for online support, 24 hours, anywhere in the world, with the latest systems in neurosurgery. The project aimed to ensure the proper use of the new technologies by the surgeons and the drastic reduction of complications. The service is integrated with treatment systems already in use for the operations of neurosurgery.

Prideh Privacy in data management for e-health (IST2001-TR 32647) Technologies for securing privacy, using encryption methods, are complex. Prideh takes advantage of the experience gained from its partners (Custodix as a supplier and as Wren’s) for the implementation of services in the pharmaceutical and health industries.

E-Scope

Reshen

Digital microscopy for diagnostics and data integration in the hospital (IST-2001-33294 BP) The last decade, the technology of a pathology diagnosis takes advantage of multimedia systems, which include standards for images (DICOM Visible Light), electronic forms of reporting, storage systems and support procedures for accreditation and certification. The project, therefore, aimed to prepare a digital microscopy to effectively integrate the standard for medical imaging, stor-

Security in Regional Networks of Health (IST2000-25354 BP) The project aimed at the demonstration of best practices in health care information exchange. The goal was to develop a business case for information security in health care and highlight the best practices, which ensure communication and information exchange between all participants (providers of health care services, end users).

577

The European Perspective of E-Health and a Framework for its Economic Evaluation

Screen-Trial Regional Secure Healthcare Networks (IST2001-TR 33439) Millions of women perform a mammography screening in Europe. The new technologies of digital mammography and Soft-Copy Reading (SCR) will have an impact on radiology and the quality of service. SCR allows the use of computer-aided detection, as well as the use of digital images and digital communications. SCR is the key to the integration of screening programs in the e-health while it will accelerate the adoption of a SCR system, which was developed in the Screen project.

Spirit Priming The Virtuous Spiral for Healthcare: Implementing an Open Source Approach to Accelerate the Uptake of Best Practice and Improvement, Regional Healthcare Network Solutions (IST-2000-26162 BP) Spirit is a pioneering initiative, which accelerates the adoption of regional health network solutions. The project aimed to establish best practices in Open Source Business software for health care. This will enable the implementation, use and dissemination of regional networks for health. The project created a partnership between the worldwide community of software developers and healthcare professionals to share, enhance, and create innovative solutions for health care.

Stemnet Information Technology for the transplantation of stem cells (IST-2000-26117 BP) The project aimed to demonstrate best practices in a network database of stem cells donors. The project facilitates the sharing of best European practices, including standards, the best laboratory practices, and quality control procedures.

578

This translates into greater efficiency and quality, reducing waiting time and cost.

Woman European network of services for the health of women (IST-2001-TR 32672) Woman II, a project funded by the EC in 19982000, has created an innovative approach for the protection of women’s health, with the combination of a Web portal and the Electronic Patient Record (EPR). The multilingual portal provides access to women and professionals with respect to women’s health information. The project’s objective was to increase the number of Web sites, in order to create a European network working for the collection and compilation of data, and promote the use of the Web portal among citizens and experts. The work ahead is adaptation, personalization and improvement of the results of Woman.

The Economic Evaluation of E-Health Health care expenditures are increasing in the industrialized countries (Scrivens, 1997; Hughes & Humphrey, 1990; Ranci Ortigiosa, 2000). Therefore, European countries spend, in the health sector, 8-10% and the U.S.A 15% of their Gross Domestic Product. The interest towards electromechanical systems (MEMs - Micro-ElectroMechanical to you Systems) means the realization of tools of small dimensions, which have remarkable advantages thanks to their invasivity and greater diagnostic-therapeutic effectiveness. Moreover, telemedicine requires a methodology, which analyzes its potential effects, taking into account the different actors of the health industry. Therefore, an economic evaluation has to take into consideration the following aspects/variables (Gerhardus, 2003; Luke et al., 2004; Zuckerman & Coile, 2003): •

Use of biomedical technology: The perspective of evaluation is limited to a single

The European Perspective of E-Health and a Framework for its Economic Evaluation

•

•

•

•

service or is extended to the entire health system. Therefore, it is necessary to assess the effects of change, estimating the total economic impact. That way, we estimate the cost of the changes, produced from the new technology, on the entire health system; Identification of the point of view: The financial resources for the patient admission are usually provided by a public system. However, although the main reference is the entire society it is necessary to identify the gains and losses of the single groups. Analysis of the alternatives: The scarce resources make choice necessary. Therefore, we have to compare the new and existing technology. Selection of the economic evaluation technique: Levels of different sophistication exist in the economic analysis, all deriving from the approach cost-benefits (cost analysis, cost minimization, cost-effectiveness, cost-utility, cost-benefits) that differ essentially because of the different ways they estimate benefits. However, the technique of evaluation is tied to the question we must answer. Identification and quantification of the costs and benefits: The studies reported in the literature do not show the same amplitude and variety of costs and benefits. For example, the capital costs are often omitted while the complexity in quantifying benefits is increasing. Normally, for the economic evaluation we use market prices or proxies, which are referred as prices, and we have to take into account the different aspects of outcome evaluation comparing the existing and the new model (Walston & Bogue, 1999; Coile, 2000; Bolon, 1997).

CONCLUSION Healthcare supply and the public health concept are being modified according to new standards. Nevertheless, although the budget constraints, e-health represents an interesting opportunity for the exchange, integration and sharing of information among health care providers and patients (Grimson et al., 2001; Maceratini et al., 1995; Leisch et al., 1997).

REFERENCES Bolon, D. S. (1997). Organizational citizenship behavior among hospital employees: a multidimensional analysis involving job satisfaction and organizational commitment. Hospital & Health Services Administration, 42(2), 221–241. Coile, R. (2000). E-health: Reinventing healthcare in the information age. Journal of Healthcare Management, 45(3), 206–210. Commission of the European Communities. (2007). A lead market initiative for Europe. Brussels: Commission of the European Communities. Drummond, M., Sculpher, M., Torrance, G., O’Brien, B., & Stoddart, G. (2005). Methods for the Economic Evaluation of Health Care Programmes. Oxford, UK: Oxford University Press. Gerhardus, D. (2003). Robot-Assisted Surgery: The Future Is Here. Journal of Healthcare Management, 48(4), 242–251. Grimson, J., Stephens, G., Jung, B., Grimson, W., Berry, D., & Pardon, S. (2001). Sharing HealthCare Records over the Internet. IEEE Internet Computing, 5(3), 49–58. doi:10.1109/4236.935177 Hughes, J., & Humphrey, C. (1990). Medical Audit in General Practice: Practice Guide to the literature. London: King’s Fund Centre.

579

The European Perspective of E-Health and a Framework for its Economic Evaluation

Leisch, E., Sartzetakis, S., Tsiknakis, M., & Orphanoudakis, S. C. (1997). A framework for the integration of distributed autonomous healthcare information systems. Informatics for Health & Social Care, 22(4), 325–335. doi:10.3109/14639239709010904 Luke, R., Walston, S., & Plummer, P. (2004). Healthcare strategy: in pursuit of competitive advantage. Chicago, IL: Health Administration Press. Maceratini, R., Rafanelli, M., & Ricci, F. L. (1995). Virtual Hospitalization: reality or utopia? Medinfo, 2, 1482–1486.

Ranci Ortigiosa, E. (2000). La valutazione di qualità nei servizi sanitari. Milano, Italy: Franco Angeli Edizioni. Scrivens, E. (1997). Accreditamento dei servizi sanitari: Esperienze internazionali a confronto. Torino, Italy: Centro Scientifico Editore. Walston, S. L., & Bogue, R. J. (1999). The effects of reengineering: Fad or competitive factor? Journal of Healthcare Management, 44(6), 456–474. Zuckerman, A., & Coile, R. (2003). Competing on Excellence: Healthcare Strategies for a Consumer-Driven Market. Chicago, IL: Health Administration Press.

This work was previously published in E-Health Systems Quality and Reliability: Models and Standards, edited by Anastasius Moumtzoglou and Anastasia Kastania, pp. 28-36, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

580

581

Chapter 2.23

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders Benjamin I. Rapoport Massachusetts Institute of Technology, USA & Harvard Medical School, USA Rahul Sarpeshkar Massachusetts Institute of Technology, USA

ABSTRACT Algorithmically and energetically efficient computational architectures that operate in real time are essential for clinically useful neural prosthetic devices. Such architectures decode raw neural data to obtain direct motor control signals for external devices. They can also perform data compression and vastly reduce the bandwidth and consequently power expended in wireless transmission of raw data from implantable brain–machine interfaces. DOI: 10.4018/978-1-60960-561-2.ch223

We describe a biomimetic algorithm and micropower analog circuit architecture for decoding neural cell ensemble signals. The decoding algorithm implements a continuous-time artificial neural network, using a bank of adaptive linear filters with kernels that emulate synaptic dynamics. The filters transform neural signal inputs into control-parameter outputs, and can be tuned automatically in an on-line learning process. We demonstrate that the algorithm is suitable for decoding both local field potentials and mean spike rates. We also provide experimental validation of our system, decoding discrete reaching

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

decisions from neuronal activity in the macaque parietal cortex, and decoding continuous head direction trajectories from cell ensemble activity in the rat thalamus. We further describe a method of mapping the algorithm to a highly parallel circuit architecture capable of continuous learning and real-time operation. Circuit simulations of a subthreshold analog CMOS instantiation of the architecture reveal that its performance is comparable to the predicted performance of our decoding algorithm for a system decoding three control parameters from 100 neural input channels at microwatt levels of power consumption. While the algorithm and decoding architecture are suitable for analog or digital implementation, we indicate how a micropower analog system trades some algorithmic programmability for reductions in power and area consumption that could facilitate implantation of a neural decoder within the brain. We also indicate how our system can compress neural data more than 100,000-fold, greatly reducing the power needed for wireless telemetry of neural data.

INTRODUCTION Brain–machine interfaces have proven capable of decoding neuronal population activity in real time to derive instantaneous control signals for prosthetics and other devices. All of the decoding systems demonstrated to date have operated by analyzing digitized neural data (Chapin, Moxon, Markowitz, & Nicolelis, 1999; Hochberg et al., 2006; Jackson, Mavoori, & Fetz, 2006; Musallam, Corneil, Greger, Scherberger, & Andersen, 2004; Santhanam, Ryu, Yu, Afshar, & Shenoy, 2006; Taylor, Tillery, & Schwartz, 2002; Wessberg et al., 2000). Clinically viable neural prosthetics are an eagerly anticipated advance in the field of rehabilitation medicine, and development of brain–machine interfaces that wirelessly transmit neural data to external devices will represent an important step toward clinical viability. The general model for such devices has

582

two components: a brain–implanted unit directly connected to a multielectrode array collecting raw neural data; and a unit outside the body for data processing, decoding, and control. Data transmission between the two units is wireless. A 100-channel, 12-bit-precise digitization of raw neural waveforms sampled at 30 kHz generates 36 Mbs-1 of data; the power costs in digitization, wireless communication, and population signal decoding all scale with this high data rate. Consequences of this scaling, as seen for example in cochlear-implant systems, include unwanted heat dissipation in the brain, decreased longevity of batteries, and increased size of the implanted unit. Recent designs for system components have addressed these issues in several ways, including micropower neural amplification (Holleman & Otis, 2007; Wattanapanitch, Fee, & Sarpeshkar, 2007); adaptive power biasing to reduce recording power in multielectrode arrays (Sarpeshkar et al., 2008); low-power data telemetry (Ghovanloo & Atluri, 2007; Mandal & Sarpeshkar, 2007, 2008; Mohseni, Najafi, Eliades, & Wang, 2005); ultralow-power analog-to-digital conversion (Yang & Sarpeshkar, 2006); low-power neural stimulation (Theogarajan et al., 2004); energy-efficient wireless recharging (Baker & Sarpeshkar, 2007); and low-power circuits and system designs for brain–machine interfaces (Sarpeshkar et al., 2008; Sarpeshkar et al., 2007). Power-conserving schemes for compressing neural data before transmission have also been proposed (Olsson III & Wise, 2005). However, almost no work has been done in the area of power-efficient neural decoding (Rapoport et al., 2009). Direct and power-efficient analysis and decoding of analog neural data within the implanted unit of a brain–machine interface could facilitate extremely high data compression ratios. For example, the 36 Mbs-1 required to transmit raw neural data from 100 channels could be compressed more than 100,000-fold to 300 bs-1 of 3-channel motoroutput information updated with 10-bit precision at 10 Hz. Such dramatic compression brings

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

concomitant reductions in the power required for communication and digitization of neural data. Ultra-low-power analog preprocessing prior to digitization of neural signals could thus be beneficial in some applications. Related considerations arise in the design of cochlear implants, and prior work in that field has demonstrated that powerefficient, analog preprocessing before digitization can be used to achieve high data compression ratios, leading to order-of-magnitude reductions in power consumption relative to fully digital data-processing schemes (Sarpeshkar, Baker et al., 2005; Sarpeshkar, Salthouse et al., 2005). Such power savings have been achieved while preserving programmability, as well as robustness to multiple sources of noise and transistor mismatch. Importantly, a processor based on this low-power design paradigm functioned successfully in a deaf patient, enabling her to understand speech on her very first attempt (Sarpeshkar, 2006). In this chapter we describe an approach to neural decoding using low-power analog preprocessing methods that can handle large quantities of high-bandwidth analog data, processing neural input signals in a slow-and-parallel fashion to generate low-bandwidth control outputs. Parallel architectures for data compression constitute an area in which analog systems perform especially well relative to digital systems, as has been demonstrated analytically (Sarpeshkar, 1998), and as exemplified by biological systems such as the retina and the cochlea (Mead, 1989; Sarpeshkar, 2006). Multiple approaches to neural signal decoding have been implemented successfully by a number of research groups, as mentioned in the section entitled “Implementation of the Decoding Algorithm in Analog Circuitry.” All of these have employed highly programmable, discrete-time, digital algorithms, implemented in software or microprocessors located outside the brain. We are unaware of any work on continuous-time analog decoders or analog circuit architectures for neural decoding (Rapoport et al., 2009). The neural signal decoder

we present here is designed to complement and integrate with existing approaches. Optimized for implementation in micropower analog circuitry, it sacrifices some algorithmic programmability to reduce the power consumption and physical size of the neural decoder, facilitating use as a component of a unit implanted within the brain. We demonstrate a method of designing highly power-efficient adaptive circuit architectures for neural decoding in two stages. We first generalize a discrete-time algorithm for adaptive-filter decoding to a continuous-time analog decoder. We then provide an approach to translating this decoder into a circuit architecture. In this chapter we show that our algorithm and analog architecture offer good performance when learning to interpret real neural cell ensemble codes in the rat thalamus, from which we decode continuous head direction trajectories, and the macaque monkey parietal cortex, from which we decode discrete reach decisions. We further show that the system is sufficiently flexible to accept local field potentials or mean spike rates as inputs. Circuit simulations of a subthreshold analog CMOS implementation of our circuit architecture suggest that a spike processor and decoder with 100 input channels and 3 output control parameters could be built with a total power budget of approximately 42 μW; this would reduce the power expended in wireless telemetry to approximately 300 nW and the total power consumption of the implanted unit would be under 43 μW. Our decoding scheme sacrifices the flexibility of a general-purpose digital system for the efficiency of a special-purpose analog system. This tradeoff may be undesirable in some neural prosthetic devices. Therefore, our proposed decoder is meant to be used not as a substitute for digital signal processors but rather as an adjunct to digital hardware, in ways that synergistically fuse the efficiency of embedded analog preprocessing options with the flexibility of a general-purpose external digital processor. The schematic for such a combined system is shown in Figure 1. First,

583

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

an external digital processor transiently analyzes raw, high-bandwidth neural data to determine spike thresholds and DAC-programmable analog parameters for the implanted analog decoder. Such parameters could control which input channels are turned on or off, set filter time constants for converting spikes to mean firing rates, and establish amplitude thresholds that ensure optimal spike sorting. Such analysis requires a great deal of flexible programmability and so is best done digitally, by an external processor, taking into account all available raw neural data. The analysis generates configuration settings that can be downloaded over a low-bandwidth link to reconfigure the implanted unit, including a low-power analog decoder, digitally. The analog system then learns, decodes, and outputs low-bandwidth information in a low-power fashion for most of the time that the brain–machine interface is operational. Digital analysis and recalibration of the analog system may be done infrequently (perhaps once a day, as in early reports of a clinically useful brain–machine interface (Hochberg et al., 2006)) so that the average power consumption is always low even though transient power consumption of the external digital system may be relatively high during recalibration. The architecture of Figure 1 permits exploration of other decoding algo-

rithms in the external unit, allowing high-power operation when necessary to achieve sufficient programmability. For clinical neural prosthetic devices, the necessity of highly sophisticated decoding algorithms remains open to question, since both animal (Carmena et al., 2003; Musallam et al., 2004; Taylor et al., 2002; Velliste, Perel, Spalding, Whitford, & Schwartz, 2008) and human (Hochberg et al., 2006) users of even first-generation neural prosthetic systems have proven capable of rapidly adapting to the particular rules governing the control of their brain–machine interfaces. In the present work we focus on an architecture to implement a simple, continuous-time analog linear (convolutional) decoding algorithm. The approach we present here can be generalized to implement analog-circuit architectures of more general Bayesian algorithms; examples of related systems include analog probabilistic decoding circuit architectures used in speech recognition and error correcting codes (Lazzaro, Wawrzynek, & Lippmann, 1997; Loeliger, Tarköy, Lustenberger, & Helfenstein, 1999). Related architectures can be adapted through our mathematical approach to design circuits for Bayesian neural decoding. This chapter is organized as follows: The section entitled “Methods of Designing and Testing

Figure 1. Schematic diagram of a brain–machine interface system incorporating an implantable analog neural decoder

584

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

a Micropower Neural Decoding System” begins by presenting a derivation of our adaptive-kernel decoder; a corresponding description of an example instantiation in subthreshold micropower analog VLSI is presented in the subsection “Implementation of the Decoding Algorithm in Analog Circuitry.” “Methods of Testing the Neural Signal Decoding System” describes methods for testing the algorithm and corresponding circuit architecture in several decoding tasks: continuous trajectory decoding from simulated local field potential (LFP) input signals, continuous headdirection trajectory decoding from spike-train inputs recorded from the thalamus of a freely moving rat, and discrete arm-reach intentions decoded from spike-train inputs recorded from the parietal cortex of a macaque monkey. The results of these tests are subsequently presented in the section entitled “Evaluating the Performance of the Neural Decoding System.” The following section, “Evaluating Design Tradeoffs in the Construction of an Implantable Neural Decoder,” discusses the simulated performance characteristics of our decoding system and alternative digital implementations from the perspectives of reducing power consumption for implantable brain–machine interfaces. The chapter concludes with a summary of our work and anticipated directions for future research in the field of neural decoding in the context of brain–machine interfaces.

METHODS OF DESIGNING AND TESTING A MICROPOWER NEURAL DECODING SYSTEM An Algorithm for Adaptive Convolutional Decoding of Neural Cell Ensemble Signals The function of neuronal population decoding is to map neural signals onto higher-level cognitive processes to which they correspond. This section presents the mathematical foundations of an adap-

tive convolutional decoder whose kernels can be modified according to a gradient-descent–based learning algorithm that optimizes decoding performance in an on-line, real-time fashion. In convolutional decoding of neural cell ensemble signals, the decoding operation takes the form   M (t ) = W (t )  N (t ) n

M i (t ) = ∑Wij (t )  N j (t )



(1.1)

j =1

 where i∈{1,…,m}; N (t ) is an n-dimensional vector containing the neural signal (n input channels of neuronal firing rates, analog signal values, or local field potentials, for example) at time t;  M (t ) is a corresponding m-dimensional vector containing the decoder output signal (which in the examples presented here corresponds to motor control parameters, but could correspond as well to limb or joint kinematic parameters or to characteristics or states of nonmotor cognitive processes); W is a matrix of convolution kernels Wij(t) (formally analogous to a matrix of dynamic synaptic weights), each of which depends on a set of p modifiable parameters, Wijk , k ∈ {1,…,p}; and ∘ indicates convolution. Accurate decoding requires first choosing an appropriate functional form for the kernels and then optimizing the kernel parameters to achieve maximal decoding accuracy. Since the optimization process is generalizable to any choice of kernels that are differentiable functions of the tuning parameters, we discuss the general process first. We then explain our biophysical motivations for selecting particular functional forms for the decoding kernels; appropriately chosen kernels enable the neural decoder to emulate the real-time encoding and decoding processes performed by biological neurons. Our algorithm for optimizing the decoding kernels uses a gradient-descent approach to

585

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

minimize decoding error in a least-squares sense during a learning phase of decoder operation. ˆ During this phase the correct output M (t ) , and   ˆ hence the decoder error e (t ) = M (t ) − M (t ) , is available to the decoder for feedback-based learning. We design the optimization algorithm to evolve W(t) in a manner that reduces the squared decoder error on a timescale set by the parameter τ, where the squared error is defined as t

) ∫ e (u )

(

E W (t ), τ =

t −τ m

t

2

du 2

= ∑ ∫ ei (u ) du

(1.2)

i =1 t −τ m

≡ ∑ Ei , i =1

and the independence of each of the m terms in Equation (1.2) is due to the independence of the m sets of n × p parameters Wijk , j∈{1,…,n}, k∈{1,…,p} associated with generating each component Mi(t) of the output. Our strategy for optimizing the matrix of decoder kernels is to modify each of the kernel parameters Wijk continuously and in parallel, on a timescale set by τ, in proportion to the negative gradient of E(W(t),τ) with respect to each parameter:

 ∂E −∇kij E W (t ), τ ≡ − ∂Wijk

(

)

  t   n   ∂Wij (u )     N j (u ) = − ∫ du  2 M i (u ) − ∑Wij (u )  N j (u ) × − k    ∂ W    = 1 j  t −τ ij     t  ∂W (u )   ij = 2 ∫ ei (u )   N j (u ) du.  ∂W k  t −τ   ij

(1.3)

The learning algorithm refines W in a contin uous-time fashion, using −∇E (t ) as an error feedback signal to modify W(t), and incrementing each of the parameters Wijk (t ) in continuous time  by a term proportional to −∇kij E W (t ) (the

(

586

)

proportionality constant, ε, must be large enough to ensure quick learning but small enough to ensure learning stability). If W(t) is viewed as an array of linear filters operating on the neural input  signal, the quantity −∇kij E W (t ), τ used to

(

)

increment each filter parameter can be described as the product, averaged over a time interval of length τ, of the error in the filter output and a secondarily filtered version of the filter input. The error term is identical for the parameters of all filters contributing to a given component of the output, Mi(t). The secondarily filtered version of the input is generated by a secondary convolution ∂Wij (u ) , which depends on the funckernel, − ∂Wijk tional form of each primary filter kernel and in general differs for each filter parameter. Figure 2 shows a block diagram for an analog circuit architecture that implements our decoding and optimization algorithm. Many functional forms for the convolution kernels are both theoretically possible and practical to implement using low-power analog circuitry. Our approach has been to emulate biological neural systems by choosing a biophysically inspired kernel whose impulse response approximates the postsynaptic currents biological neurons integrate when encoding and decoding neural signals in vivo (Arenz, Silver, Schaefer, & Margrie, 2008). Combining our decoding architecture with the choice of a first-order low-pass decoder kernel enables our low-power neural decoder to implement a biomimetic, continuous-time artificial neural network. Numerical experiments have also indicated that decoding using such biomimetic kernels can yield results comparable to those obtained using optimal linear decoders (Eliasmith & Anderson, 2003). But in contrast with our on-line optimization scheme, optimal linear decoders are computed off-line after all training data have been collected. We have found that this simple choice of kernel

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 2. Block diagram of an analog circuit architecture for linear convolutional decoding and learning

offers effective performance in practice, and so we confine the present analysis to that kernel. We also offer a heuristic justification for using first-order low-pass kernels. Kernels for decoding continuous trajectories should be designed to anticipate low-frequency variation in the input and output signals. Two-parameter first-order low-pass filter kernels account for trajectory continuity by exponentially weighting the history of neural inputs: Wij =

Aij τij

e

t − τij

Wijk =2 = τij , the decay time over which past inputs  N (t '), t ' < t , influence the present output es  timate M (t ) = W  N (t ) . The filters used to tune the low-pass filter kernel parameters can be implemented using simple and compact analog circuitry. The gain parameters are tuned using low-pass filter kernels of the form ∂Wij (t ) ∂Wijk =1

,

t

1 − τij = e , τij

(1.5)

(1.4)

where the two tunable kernel parameters are Wijk =1 = Aij , the low-pass filter gain, and

while the time-constant parameters are tuned using band-pass filter kernels:

587

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

∂Wij (t ) ∂Wijk =2

=

Aij τij2

e

−

t τij

t     − 1 .   τij

(1.6)

When decoding discontinuous trajectories, such as sequences of discrete decisions, we use the limiting case of this kernel as τij → 0: Wij (t ) = Wijk =1δ (t ) = Aij δ (t ) .

(1.7)

Such a decoding system, in which each kernel is a zeroth-order filter characterized by a single tunable constant, performs instantaneous linear decoding, which has successfully been used by others to decode neuronal population signals in the context of neural prosthetics (Hochberg et al., 2006; Wessberg & Nicolelis, 2004). With kernels of this form, W(t) is analogous to matrices of synaptic weights encountered in artificial neural networks, and our optimization algorithm resembles a ‘delta-rule’ learning procedure (Haykin, 1999).

Implementation of the Decoding Algorithm in Analog Circuitry In this section we describe the implementation of our adaptive decoder in an analog circuit architecture. This approach represents a divergence from conventional approaches to neural signal decoding. The ease of implementing sophisticated algorithms in digital systems has led to a proliferation of effective approaches to decoding and learning. Some of the most popular such algorithms in the context of neural decoding involve construction of optimal linear filters (Serruya, Hatsopoulos, Fellows, Paninski, & Donoghue, 2003; Warland, Reinagel, & Meister, 1997); training of artificial neural networks (Sanchez, Erdogmus, Principe, Wessberg, & Nicolelis, 2005); use of Kalman filters (Wu, Black et al., 2004; Wu, Gao, Bienenstock, Donoghue, & Black, 2006; Wu, Shaikhouni, Donoghue, & Black, 2004) and adaptive Kalman filters (Wu & Hatsopoulos, 2008); estimation

588

based on Bayesian inference techniques (Brockwell, Rojas, & Kass, 2004; Srinivasan, Eden, Mitter, & Brown, 2007) and point-process models (Eden, Frank, Barbieri, Solo, & Brown, 2004); and decoding on the basis of frequency-domain data (such as the spectral content of local field potentials) (Musallam et al., 2004; Pesaran, Pezaris, Sahani, Mitra, & Andersen, 2002; Shenoy et al., 2003) or wavelet decompositions of neural signals (Musallam et al., 2004). However, to our knowledge the present work represents the first description of an analog circuit architecture for neural signal decoding. We separate the present discussion into three parts. The first part treats the preprocessing of the two principal classes of neural input signals, local field potentials (LFPs) and action potentials (‘spikes’). We then address the decoding architecture itself. Finally, we describe our estimates of power consumption by each module of the decoder architecture. Figure 3 is a diagram of the circuit modules required to implement a single adaptive kernel of the decoder. These modules fall into six functional classes: (1) Adaptive filters corresponding to the kernels Wij(t); (2) Parameter-

{ ( ) } = {A , τ } ;

learning filters to tune the Wij

k

ij

ij

(3) Biasing circuits for the parameter-learning filters; (4) Multipliers; (5) Adders and subtracters; and (6) Memory units for storing learned parameter values. The second part of this section presents the design for each functional subunit of our decoding architecture in turn. Note that the descriptions given here correspond to an unoptimized, proof-of-concept circuit implementation. Current-mode techniques, circuit optimizations not described here, and noise-robust analog biasing techniques such as those described in (Sarpeshkar, Salthouse et al., 2005) will be necessary to ensure robust, programmable, and efficient operation in a practical implementation.

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 3. Block diagram indicating the functional component circuits required to implement a single adaptive kernel of the convolutional decoder

Input Signals for the Neural Decoder Local-Field-Potential–Based Decoding Local field potentials encode information about well defined cognitive states and can also be used as control signals for neural prostheses. Furthermore, it has been found that particular classes of information are encoded in distinct frequency bands of the LFP power spectrum (Pesaran et al., 2002). Information encoded in this manner can be extracted from the LFP by passing the raw LFP signal through a band-pass amplifier tuned to the spectral band of interest, rectifying the output of the band-pass filter, and passing the result through a peak-detection circuit to generate an envelope waveform. Prior work solved this design problem in the context of ultra-low-power bionic-ear (cochlear implant) processors (Sarpeshkar, Baker et al., 2005; Sarpeshkar, Salthouse et al., 2005) for which micropower band-pass amplifier (Salthouse & Sarpeshkar, 2003) and envelope detector circuits (Zhak, Baker, & Sarpeshkar, 2003) were built that can be employed to process neural signals at lower frequencies. Spike–Based Decoding Neuronal action potential voltage spikes typically have widths on the order of 1 ms, corresponding to frequencies in the kilohertz range. Spike-

based inputs are transformed to lower-frequency signals (of order 1–10 Hz) through time-domain averaging and the resulting spike rates are used as input signals for the decoder. Such averaging can be implemented by low-pass interpolation filters. The simplest such filter is a first-order low-pass filter with frequency-domain transfer 1 , and cutoff frequency function H 1 (s ) = 1 + τ1s fc = (2πτ1)-1 Hz. Smoother interpolation can be obtained by cascading first-order filters. The analog implementation of such filters can be achieved using a Gm–C design: Gm refers to the transconductance of an operational transconductance amplifier (OTA) component, while C denotes 1 Gm , a filter capacitance. In such a filter fc = 2π C so a low cutoff frequency requires C to be large or Gm to be small. Circuit layout area restrictions will constrain the maximum value of C to approximately 4 pF, so a low fc requires Gm to be small. Wide-linear-range transconductors with subthreshold bias currents (Sarpeshkar, Lyon, & Mead, 1997) allow Gm to be small enough to achieve such low corner frequencies in a reliable fashion. We convert raw neural input waveforms into smooth time-averaged spike rates (Figure 5) using the circuitry shown in Figure 4, which uses dual thresholding to detect action potentials on

589

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 4. Analog preprocessing circuitry for converting raw neural signals into smooth analog inputs for an analog decoder

Figure 5. Analog preprocessing of raw neural signals to generate smooth analog inputs to an implantable analog decoder. The raw neural signal (top trace) is thresholded to generate a spike train (middle trace), which is converted to a smooth time-averaged spike rate (bottom trace) using a second-order low-pass filter

spectively. As a result, power consumption of the preprocessing stages will be dominated by the requirements of the comparator; a comparator of the kind described in (Yang & Sarpeshkar, 2006), operating at 30 kHz, requires approximately 240 nW.

Functional Circuit Subunits of the Decoder Architecture Primary Adaptive Filters Each of the m × n tunable kernels used to implement our decoding architecture can be understood as an adaptive filter whose frequency-domain transfer function is obtained from the Laplace transform of the time-domain kernel in Equation (1.4): Wij (s ) =

each input channel and then smoothens the resulting spike trains to generate mean firing rate input signals. Circuit simulations indicate that the charge pump and smoothing filter modules consume approximately 1.1 nW and 40 pW of power from a 1 V supply in 0.18 μm CMOS technology, re-

590

Aij 1 + τij



(1.8)

where the gain Aij can be positive or negative. This transfer function can be obtained from a filter having the topology shown in Figure 6, which contains four standard, nine-transistor wide-range operational transconductance amplifiers (OTAs) of the form described in (Mead, 1989). Every OTA is operated subthreshold such that its transconductance is linear in its bias current. The gain of the filter is determined by the three (A ) (A ) (R ) OTAs, Gm +ij , Gm −ij , and Gm ij , which have trans-

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 6. Adaptive filter with tunable parameters for learning the optimal convolution kernels for neural signal decoding

(A ) κI + ij

(A ) κI − ij

(R ) κI ij , reVT VT VT spectively, where κ denotes the gate-coupling coefficient of the MOS transistor, and in this analysis κ = 0.7 is assumed for all transistors; and VT = kT/q, where k denotes the Boltzmann constant, q denotes the electron charge, and T denotes the Kelvin temperature. By applying Kirchoff’s Current Law (KCL) at node Vx, we obtain the voltage gain from Vin to Vx as

conductances

(Aij )

,

, and

1 . Cτ ij 1+s τ Gmij

(1.11)

As a result, we can express the overall transfer function from Vin to Vout as (A ) (A ) 2I + ij − I totij (s ) = (R ) Vin I ij

Vout

1 1+s

Cτ

.

(1.12)

ij

τ

Gmij

=

(1.9)

(Aij )

(Aij )

If we set the sum of I + and I − equal to a (A ) constant current I totij , the expression in Equation (1.9) can be reduced to (A ) (A ) 2I + ij − I totij = . (R ) Vin I ij

Vx

Vx

(s ) =

(Aij )

Gm + − Gm − (R ) Vin Gm ij (A ) (A ) I + ij − I − ij . = (R ) I ij Vx

Vout

(1.10)

The transfer function from Vx to Vout can be expressed as

Comparing Equation (1.12) to Equation (1.8), we obtain expressions for the gain and time constant parameters: (A ) (A ) 2I + ij − I totij Aij = (R ) I ij Cτ ij . τij = (τij ) Gm

(1.13)

(A ) (A ) Since I + ij can be adjusted from 0 to I totij , (A ) (R ) (A ) (R ) Aij can vary from −I totij / I ij to I totij / I ij .

591

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

The time constant τij can be adjusted by tuning (τ ) the bias current in Gm ij . Secondary Parameter Tuning Filters Real-time gradient-descent–based optimization of the convolution kernels can be achieved through

{ ( )}

tuning the filter parameters Wij = {Aij , τij }  (k ) using signals proportional to −∇ij E . Construction of such signals requires convolution kernels ∂Wij (t ) . These convolution proportional to (k ) ∂Wij k

kernels can be implemented by ‘parameterlearning filters’ of the kind shown in Figure 7. Figure 7(a) shows a first-order low-pass Gm–C 1 A filter with transfer function Wij ij (s ) = 1 + τij s to be used as a ‘gain-learning filter.’ Figure 7(b) shows a second-order band-pass Gm–C filter with τij s τ transfer function Wij ij (s ) = to be used 2 1 + τ s ( ij ) as a ‘time-constant–learning filter.’ The time Cτ ij for the two parameterconstant τij = (τij ) Gm

learning filters is identical to that of the adaptive filter, as described in the section entitled “Primary adaptive filters.” Correspondingly, the bias (τ ) currents and therefore the transconductances Gm ij in all three types of filter are identical, so the time constants of all filters in the learning architecture are updated simultaneously. Note that the actual transfer function of the time-constant–learning filter need only be proportional to Wij ij (s ) , so τ

the factor of τij/Aij between the transfer function of Equation (1.8) and the filter shown in Figure 7(b) is acceptable. Implementation of the negation required to adapt the time constants is addressed in the section entitled “Multipliers.” Multipliers The multipliers that perform the operations ∂Wij (t ) ei (t ) × required by the gradient-descent (k ) ∂Wij algorithm, denoted by the symbol × in Figure 2 and Figure 3, can be implemented using widerange four-quadrant Gilbert multipliers of the kind described in (Mead, 1989), which accept four voltage inputs Vi, i ∈ {1, 2, 3, 4} and a bias current Ib, and generate an output current; low-power performance can be obtained by operating the

Figure 7. Parameter-learning filters for tuning adaptive filter parameters based on error-signal feedback. 1 A to be used as a ‘gain(a) A first-order low-pass Gm–C filter with transfer function Wij ij (s ) = 1 + τij s learning filter.’ (b) A second-order band-pass Gm–C filter with transfer function Wij ij (s ) = τ

τij s 2

(1 + τ s ) ij

to be used as a ‘time-constant–learning filter’

592

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

multiplier circuit with all transistors in the subthreshold regime. The input-output characteristic of the Gilbert multiplier is given by I out = I b tanh

κ (V1 −V2 ) 2VT

2

tanh

κ (V3 −V4 ) 2VT



 κ   (V1 −V2 ) (V3 −V4 ), ≈ I b   2VT 

(1.14)

where the approximation of Equation (1.14) is valid in the intended operating region, where V1 ≈ V2 and V3 ≈ V4. Noninverting multiplication, as required for adapting the Aij, can be implemented ∂Wij (t ) into V3, and setting by feeding ei into V1, (A ) ∂Wij ij V2 and V4 to a constant reference voltage. On the other hand, inverting multiplication, as required for adapting the τij, can be implemented by inter∂Wij (t ) changing the roles of V3 and V4, feeding (τ ) ∂Wij ij into V4, and setting V3 to a constant reference voltage. The output of each multiplier is a current, Iout, so the integrations required in Equation (1.3) for

{ ( ) } = {A , τ } are

updating the parameters Wij

k

ij

ij

conveniently implemented by linear capacitors,  (k )  Wij  C

as indicated in Figure 3. The voltages V the capacitors C

W (k )   ij 

(k )

on

and used in adapting the

parameter values Wij are therefore given by W (k )   ij 

VC

=

Ib W (k )   ij 

C

κ2 4VT2

 ∂W u   ij ( )   e u N u  ( ) ( )  du j ∫u=0 i  ∂W (k )   ij t

(1.15)

which has the form required by Equation (1.3). The filter parameters therefore vary continuously in

time, and the time variation of the control voltages can be obtained by differentiating Equation (1.15). Biasing Circuits As discussed in the section entitled “Multipliers,” the filter parameters Aij and τij defining the transfer function of adaptive filter Wij are stored on the (A ) (τ ) capacitors C ij and C ij , respectively. Furthermore, as indicated in the section entitled “Primary adaptive filters,” the values of the filter parameters can be tuned by adjusting the bias (A ) (τ ) currents that determine Gm ij and Gm ij . Since IA ( ij ) the gain Aij depends on ∝ I A , while the ( ij ) IR ( ij ) Cτ 1 ij ∝ , realtime constant τij depends on (τ ) Iτ Gm ij ( ij ) time adaptive parameter tuning requires a scheme (A ) for modifying I A in proportion to VC ij and ( ij ) (τ ) I τ in inverse proportion to VC ij . ( ij ) Tuning I A in proportion to variations in the ( ij ) (A ) capacitor voltage VC ij can be accomplished by (A ) converting VC ij into a current proportional to (A ) VC ij and then using a current mirror to generate a copy of that current that is in turn used to set (A ) the transconductance Gm ij of the adaptive filter. (A ) The conversion of VC ij into a current propor(A ) tional to VC ij can be performed by a wide-linearrange transconductance amplifier (WLR) of the form described in (Sarpeshkar et al., 1997). Figure 8(a) shows a schematic of the gain-biasing (A ) circuit used to generate I A ∝ VC ij . This biasing ( ij ) (A ) (A ) (A ) circuit is intended to make I + ij + I − ij = I totij . Assuming that all of the NMOS transistors are well matched, the current in M3 must be equal to A

that in M2, which is I +ij . Using KCL at the drain

593

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 8. Circuits for setting the bias currents and transconductances that determine the adaptive filter parameters: (a) Bias-current–setting circuit for filter gains, (b) Bias-current–setting circuit for filter time constants

(A ) (A ) of M3, the current in M4 is thus I totij − I + ij , which (A ) (A ) (A ) is equal to the current in M5, so I − ij = I totij − I + ij as required. Tuning I τ in inverse proportion to variations ( ij ) (τ ) in the capacitor voltage VC ij can be accomplished using the circuit shown in Figure 8(b), which (τ ) operates as follows. First, VC ij is converted into

a proportional current I pτ as in the gain-biasing ( ij ) circuit. A translinear circuit, formed by the four well matched MOS transistors M1–M4, is then I2 used to invert I pτ , producing I τ = ps , where ( ij ) ( ij ) Iτ ( ij )

594

Is is a current reference that scales the inversion. A mirror copy of I τ is then used as the bias ( ij ) (τ ) current that sets transconductance Gm ij . Adders and Subtracters Each adder, denoted by the symbol Σ in Figure 2 and Figure 3, sums the n outputs of each set {Wij}, j ∈ {1, …, n} of adaptive filters contributing to ˆ (t ) . The adders can be W (t )  N (t ) = M i

(

)

i

implemented using a follower-aggregation circuit of the kind described in (Mead, 1989). The corresponding error signal, ei(t), is generated by ˆ (t ) . This performing the subtraction M i (t ) − M i

operation can be implemented by another adder,

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

with a unity-gain inverting amplifier negating the ˆ (t ) . adder input from M i Parameter Memory A completely analog implementation of our decoder could include analog memory units for storing parameter values when learning ends. Such units could consist of analog memory elements operating in a switched sample-and-hold scheme to permit memoryless adaptation during the learning phase and parameter storage as soon as learning terminates. Analog memory circuits with 8-bit hold times of 3.9 hours and 12-bit hold times of 14.5 minutes have been developed (O’Halloran & Sarpeshkar, 2004, 2006) and could be used in this context. Output from the memory and biasing circuits could be multiplexed onto the adaptive filter nodes whose voltages correspond to the adaptive filter parameters, using a CMOS transmission gate. This scheme is indicated in Figure 3. However, even using digital memory elements does not increase total power consumption significantly, since the termination of a learning phase is a rare event and therefore writing to memory, with its associated power cost, occurs only infrequently. Power Consumption of the Decoder Table 1 shows an estimate of the power consumed by each circuit block needed to implement our neural decoder, based on a SPICE simulation of the decoder, the performance of which is shown in Figure 11. The supply voltage of the entire system is 1 V. The bias current of each module is chosen so that the circuit has enough bandwidth to process input signals band-limited to 1 kHz (note that this choice of input signal bandwidth is more than sufficient because the bandwidth of meanfiring-rate changes is typically much smaller than the reciprocal of a refractory period. As indicated in Table 1, the total power consumption of one decoding module is approximately 54 nW. In order to ensure robust, stable decoding, the learned (optimized) filter parameters for the neu-

ral decoder will be programmed using digital-toanalog converters (DACs) as indicated in the “Introduction” section. These DACs do not appreciably increase system power consumption. One DAC is required per filter parameter to set the bias current for each filter constant, so a system decoding 3 motor parameters from 100 neural input channels would require 600 current DACs. The static current in a current DAC is approximately equal to that of a transistor whose gate is connected to the associated capacitors in Figure 3, which is approximately 100 pA. The total power consumed by all 600 DACs, when operated from a 1.8-V supply, is therefore only approximately 110 nW.

Methods of Testing the Neural Signal Decoding System In this section we describe three approaches to testing the ability of our system to decode neural signals. The first experiment is based on a simulation of local field potential input signals, which are used to decode continuous-time motor trajectories. The second experiment involves continuous-time decoding of head direction from spike train data recorded from a small set of neurons in the thalamus of an awake, behaving rat. In the third experiment we decode discrete arm-reach Table 1. Power consumption in decoder circuit modules Portion of Channel

Power Consumption (nW)

Tunable Decoding Filter

16

Gain-Learning Filter

2

Time-Constant–Learning Filter

2

Multipliers (2)

2

Gain Biasing Circuit

10

Time Constant Biasing Circuit

10

Analog Memories (2)

5

Total (One Decoder Module)

54 nW

595

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

intentions from spike train data recorded using a multielectrode array in the posterior parietal cortex of a macaque monkey.

normalized mean squared error, η, of the estimated trajectory, defined as

Testing the Ability to Decode Trajectory Information from Multichannel Local Field Potential Neural Input Signals A recent set of experiments has shown that parietal cortex neurons tend to exhibit electrical activity predictive of arm or eye movement in a single preferred direction. Increases in γ -band (25–90 Hz) spectral activity of such tuned parietal neurons anticipate movements in the preferred directions of those neurons, so a potentially useful signal for decoding intended movement from neural activity is therefore an envelope curve describing the modulated amplitude of the power transmitted in the gamma band (Andersen, Musallam, & Pesaran, 2004; Musallam et al., 2004; Pesaran et al., 2002). As a preliminary demonstration of the ability of our convolutional decoder to interpret LFP-type neural input signals, we therefore generated simulated γ -band power envelopes in order to model the local field potentials recorded by a set of n neural recording electrodes. We modeled γ-band power envelopes using a set of sinusoids with randomized amplitudes and phases and a constant offset term, and stored the corresponding  waveforms in the vector N (t ) , which was used

1 η≡ T

as the input to the decoder. We then randomly generated an m×n matrix, W*, and used it to construct a vector of m motor control parameters,  M (t ) . The decoder was permitted to observe  M (t ) during a learning period of variable length, over which it sought to optimize its m×ndimensional convolution kernel W(t). The parameter ε was set to 0.1 during these simulations, and both the Ni(t) and the Mi(t) were transformed using a hyperbolic-tangent normalization to constrain them to the interval [-1, 1]. We evaluated decoder performance using a scale-invariant and dimensionless conventional figure of merit, the

596

∫

t2 =2T

t1 =T

2

 M (t ) − M ˆ (t )  i i  , (1.16) dt ∑    Li  i =1   m

where Li denotes the maximum extent of excursions permitted to Mi(t) and the time T denotes the length of the training interval; η(1) is used to denote the average value of η for a single output dimension. It is possible to consider other figures of merit, including ones based on correlation  rather than absolute error between M (t ) and ˆ M (t ) , but other authors have agreed that η-like figures of merit tend to reflect decoding system performance most reasonably (Wu et al., 2006). The results of these simulations are provided in the section entitled “Model Motor Trajectory Decoding from Simulated Multichannel Local Field Potential Inputs.” Testing the Decoding of Continuous Trajectories Using Neuronal Spike Recordings from the Thalamus of an Awake Behaving Rat Head direction cells of the rat thalamus are neurons known to exhibit receptive fields tuned to specific orientations of the head relative to the environment (Taube, 1995). We explored the ability of our system to decode the temporal firing patterns of such cells in real time. The spike trains used as input signals to the decoder in this set of experiments were derived from tetrode recordings made in the thalamus of a laboratory rat that rat ran back and forth between two points on a circular maze for food reward over a 30-minute period. The position and head direction of the animal were continuously tracked using a pair of head-mounted light-emitting diode (LED) arrays imaged at a sampling frequency of 30 Hz using a 300 × 300-pixel charge-coupled device (CCD)

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

array. Data from the imager were time-stamped in order to ensure synchronization with the neural recordings, and were used to generate a target ˆ output signal M (t ) for the decoder. In order to avoid learning the discontinuity associated with a mod-2π–based definition of the head direction ˆ angle θ(t), M (t ) was constructed as a two-dimensional vector having components M1(t) ≡ cosθ(t) and M2(t) ≡ sinθ(t). Spike-sorting analysis of the tetrode-derived waveforms isolated n = 6 single units, and the activity of each of these units was converted to a normalized spike rate Ni(t), i  ∈ {1, …, n = 6} so that N (t ) could be used as an n-channel analog input to the decoder. As indicated in the section entitled “Input Signals for the Neural Decoder,” the conversion of spike trains to analog input signals was achieved by treating each spike train as a train of pulses having uniform amplitude and pulse width, then passing the pulse trains through a third-order interpolation filter with transfer function H 3 (s ) =

1 3

(1 + τ s )



(1.17)

1

with τ1 = 500 ms. The filter output was then rescaled and recentered about a zero-offset in order to generate the normalized spike rates Ni(t). The performance of the convolutional decoding algo  rithm in learning to map N (t ) to M (t ) was studied using a software implementation of the decoder, as well as a SPICE analog-circuit simulation of the convolutional decoder, the details of which are described in the section entitled “Functional Circuit Subunits of the Decoder Architecture.” We discuss the results of these simulations in the section entitled “Continuous Real-Time Decoding of Head Direction from Neuronal Spike Activity in the Rat Thalamus.”

Testing the Decoding of Discrete Decisions Using Neuronal Spike Recordings from the Posterior Parietal Cortex of a Macaque Monkey Engaged in an Arm-Reaching Task Neurons in the posterior parietal cortex have been shown to encode intention to execute limb movements in both humans (Connolly, Andersen, & Goodale, 2003) and nonhuman primates (Snyder, Batista, & Andersen, 1997). The ability to decode signals from this region has provided proof of principle for neural prosthetic devices based on cognitive (as opposed to explicitly motor) control signals (Musallam et al., 2004). Using neural spike trains recorded during these proof-of-principle experiments, we investigated the performance of our system in the real-time decoding of discrete arm-reaching decisions. The spike trains used as input signals to the decoder in this set of experiments were derived from recordings by a multielectrode array chronically implanted in the medial intraparietal area (within the ‘parietal reach region’) of a macaque monkey. The animal had previously been trained to perform a standard stimulus-response task involving center-out arm-reaching movements between visual targets. This task, which has been described in detail elsewhere (Musallam et al., 2004), was designed to isolate the neural correlates of motor intention from those of actual movement. The monkey initiated each iteration of the task by touching a central cue target and looking at a nearby visual fixation point at t = – 800 ms (its gaze was monitored using an eye-tracking device, and cues were presented on a touch-sensitive screen). After a delay of 500 ms a peripheral cue target was flashed from t = – 300 to t = 0 ms at one of four locations displaced up, right, down, or left from the starting point. The animal was rewarded if it touched the indicated target at the end of a memory period of 1500±300 ms. Neural activity as monitored by the implanted electrode array was recorded continuously during each trial. A spike-sorting algorithm isolated 54 units from the recorded signals, and spike trains from

597

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

each unit were smoothed to spike rate waveforms using a filter of the form provided in Equation (1.17), with τ1 = 50 ms. Filter output was then rescaled and recentered about a zero-offset in order to generate the normalized spike rate input signals Ni(t) for each of the isolated units indexed by i ∈ {1, …, n – 1 = 54}; the (n = 55)-dimen sional neural input signal N (t ) contained an

additional constant-offset component Nn(t) = 1. The task of the decoder was to predict imminent arm movement on the basis of intention-related neuronal activity. Therefore, in order to ensure the  absence from N (t ) of residual neuronal activity

corresponding to actual arm movement (such artifact signals are sometimes present at the beginning and end of a memory interval), only the segment of  N (t ) from t ∈ [200, 1100] ms was fed into the  decoder for each reach. The motor output M (t )

was defined as a two-dimensional position vector corresponding to the target to which the monkey reached at the end of a corresponding memory period. Reach target positions were encoded as follows: Analog outputs generated by the decoder were thresholded according to Mi→sgnMi,

(1.18)

so that positive and negative outputs were interpreted as +1 and –1, respectively. We studied the performance of a software implementation of our convolutional decoding system in learning to use neural signals from the parietal reach region to predict the direction of subsequent reaching movements. We discuss the Table 2. Encoding vectors for reach targets Direction

M1

M2

Up

+1

+1

Right

+1

–1

Down

–1

–1

Left

–1

+1

598

results of these simulations in the section entitled “Decoding Discrete Arm Reaches from Neuronal Spike Activity in the Macaque Monkey Posterior Parietal Cortex.”

EVALUATING THE PERFORMANCE OF THE NEURAL DECODING SYSTEM Model Motor Trajectory Decoding from Simulated Multichannel Local Field Potential Inputs Parts (a) and (b) of Figure 9 respectively illustrate the training and post-training phases of decoder operation as the system learns to trace a threedimensional trajectory in real time during a simulation of the kind described in the section entitled “Testing the Ability to Decode Trajectory Information from Multichannel Local Field Potential Neural Input Signals,” with (n, m) = (10,  3). The figure illustrates qualitatively that M (t ) ˆ converges toward M (t ) reasonably quickly on the timescale set by full-scale variations in the trajectory. Figure 10 presents the results of a set of computations of η(1) for the performance of the neural decoding system in simulations of the kind described in the section entitled “Testing the Ability to Decode Trajectory Information from Multichannel Local Field Potential Neural Input Signals.” The system was trained for intervals of varying length up to one minute, T ∈ [0, 60] s, and the value of η(1) was computed for each of 50 trials at each value of T. In a fraction f of trials at each value of T, the decoding performance as reflected by η(1)(T) was significantly worse than in the remaining fraction, 1–f, of cases. In such cases markedly improved decoding, comparable to that achieved in the (1–f)-majority of cases, could be achieved by randomly reinitializing the

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 9. Trajectory decoding from simulated local field potential input signals. (a) The correct trajecˆ tory, M (t ) , is plotted in light gray over the initial segment t ∈ [0, 10] s of a 40-second training interval,   while the decoded trajectory, M (t ) ≡ W  N (t ) , is plotted as a darker line whose shade lightens as  ˆ time progresses; M (t ) converges toward M (t ) . (b) For t ∈ [40, 80] s following training, the correct trajectory is plotted in light gray while the decoded trajectory evolves with time from dark to lighter shades

(k )

parameters Wij and decoding again. The data presented in Figure 10 were obtained by setting f = 0.1. Error in trajectory estimation by the decoder decreases rapidly as the training interval increases. At T = 30 s, for example, 〈η(1)〉≈0.008±0.008, as compared with a baseline value of 〈η(1)〉≈0.118±0.106 computed for an untrained system (T = 0) over 1000 trials.

Continuous Real-Time Decoding of Head Direction from Neuronal Spike Activity in the Rat Thalamus Head direction was decoded from the activity of the n = 6 isolated thalamic neurons according to the method described in the section entitled “Testing the Decoding of Continuous Trajectories Using Neuronal Spike Recordings from the Thalamus of an Awake Behaving Rat.” The adap-

{ ( ) } = {A , τ } were

tive filter parameters Wij

p

ij

ij

optimized through gradient descent over training intervals of length T during which the decoder error, ˆ (t ) ei (t ) = M i (t ) − M i  M (t ) = cos θ (t ), sin θ (t ) ,

(

)

(1.19)

(where θ denotes the head direction angle) was made available to the adaptive filter in the feedback configuration described in the section entitled “An Algorithm for Adaptive Convolutional Decoding of Neural Cell Ensemble Signals” for t ∈ [0, T]. Following these training intervals feedback was discontinued and the performance of the decoder was assessed by comparing the decoder  ˆ output M (t ) with M (t ) for t>T. Figure 11(a) compares the output of the decoder to the measured head direction over a 240s interval. The filter parameters were trained over the interval t ∈ [0, T = 120] s. The figure shows 599

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 10. Mean squared trajectory prediction error as a function of training time for the convolutional decoding algorithm

 ˆ M (t ) (thick, light line) tracking M (t ) (thin, dark line) with increasing accuracy as training progresses, illustrating that while initial predictions are poor, they improve with feedback over the course of the training interval. Feedback is discontinued at t = 120 s. Qualitatively, the plots on the interval t ∈ [120, 240] s illustrate that the output of the neural decoder reproduces the shape of the correct waveform, predicting head direction on the basis of neuronal spike rates. Figure 11(b) displays output over a brief interval from the neurons in the ensemble whose activity was decoded, showing individual action potentials as raster plots below the corresponding smoothened firing rate curve for each cell, as well as a plot of head direction over the same time interval; individual neurons evidently have distinct receptive fields in head-direction space, and their spatial selectivity makes decoding both possible and, in this case, intuitive. The performance of the decoder in predicting head direction was assessed quantitatively using the normalized mean-squared error measure η(1) (in this context Li = 2, t1 = T, and t2 = T + 60 s). In

600

order to quantify the accuracy of head direction decoding as a function of training time T, η(1) was computed for a set of training and decoding trials with increasingly long training periods, averaging over randomized initial settings of the filter parameters and different choices of training interval. The results of this computation are displayed in Figure 12, which shows improving accuracy of head direction decoding with increased training.

Decoding Discrete Arm Reaches from Neuronal Spike Activity in the Macaque Monkey Posterior Parietal Cortex The input signals to the neural decoder when decoding discrete reaches were piecewise-constant, as described in the section entitled “Testing the Decoding of Discrete Decisions Using Neuronal Spike Recordings from the Posterior Parietal Cortex of a Macaque Monkey Engaged in an Arm-Reaching Task.” This form of input reflects a qualitative difference between the decoding problem in this experiment, which requires the decoder to make a series of decisions from among

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 11. (a) Continuous decoding of head direction from neuronal spiking activity. (b) Spiking activity in head direction cells and corresponding head direction plotted as functions of time. The paired plots illustrate neuronal receptive fields and the distribution of their peaks over the range of possible head direction angles

a finite set of options, and the decoding problems framed in the sections entitled “Testing the Ability to Decode Trajectory Information from Mul-

tichannel Local Field Potential Neural Input Signals” and “Testing the Decoding of Continuous Trajectories Using Neuronal Spike Recordings

Figure 12. Performance improvements in head direction decoding with increasing training times

601

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

from the Thalamus of an Awake Behaving Rat,” and simulated in the sections entitled “Model Motor Trajectory Decoding from Simulated Multichannel Local Field Potential Inputs” and “Continuous Real-Time Decoding of Head Direction from Neuronal Spike Activity in the Rat Thalamus,” which require the decoder to estimate smooth trajectories as functions of time. While the gradient-descent least-squares approach is applicable to both kinds of problem, the convolution kernel chosen to implement the neural decoder, Wij =

Aij τij

e

−

t τij

, is designed to exploit the

predictive value of past input signals. The degree  to which past inputs N (t ' < t ) have predictive value is reflected by the value of the time constant τij, and as τij → 0 the time interval over which  N (t ' < t ) contributes significantly to the present time output M (t ) correspondingly vanishes. In this experiment the signal to be decoded corresponds to a time series of discrete decisions made every Δt = w; consecutive reach cues were guaranteed to be independent through experimental  design. Consequently, N (t ') is completely uncor  related from N (t ) and M (t ) for t − t ' ≥ w . (In concrete terms, since successive reaches are independent, neural activity preceding one reach contains no predictive information concerning the direction of the next reach.) As a result, meaningful decoding requires τij = 0, which we enforce by taking τij to be a small, unmodifiable constant. Therefore, our decoding scheme as applied to the reach-intention neural data reduces to an instantaneous linear decoder, analogous to a singlelayer artificial neural network implemented in continuous time and trained with continuous-time feedback. The neural data used in the experiments reported here were first obtained and analyzed in connection with a previously reported set of experiments (Musallam et al., 2004). The decod-

602

ing method used in those experiments involved an analysis of variance to preselect a subset of neurons exhibiting the greatest directional tuning, followed by Bayesian inference on the mean firing rates and higher-order Haar wavelet coefficients of the signals obtained from the selected neurons. We consequently had the opportunity to compare the performance of the adaptive-filter decoder to that of the Bayesian decoder. The principal performance measure reported using the Bayesian decoder was a 64.4% success rate in predicting the correct one of four allowed reach directions (Musallam et al., 2004). Under corresponding training conditions the neural decoding system described in the present work generated accurate predictions in 65% ± 9% of trials (the uncertainty figure preceded by the ± symbol indicates the magnitude of one standard deviation). Improved decoding performances were demonstrated in (Musallam et al., 2004) by considering higher-order coefficients in the Haar wavelet decomposition of the neural input signals, but in the present study we considered only the zeroth-order coefficients, corresponding to mean firing rates. The decoding scheme originally used to decode the data analyzed in this section was based in part on the known tendency of direction-sensitive neurons to ‘tune’ to preferred directions in the sense that only movement in certain preferred directions induces such neurons to modulate their firing rates away from a baseline (Cohen & Andersen, 2002). In preparing the Bayesian decoder, an off-line analysis of variance on the training set of spike trains (corresponding to arm reaches in each direction for each isolated neuron) was required to rank the isolated neurons by degree of directional sensitivity. The computational intensity of this decoding scheme was sufficiently high that inputs from only a subset of isolated neurons were used in decoding after the learning period ended, and this ranking provided a means of prioritizing neurons for use as decoder inputs. By contrast, the adaptive-filter decoder described here easily handles all 54 neuronal inputs in computer

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

simulations of real-time decoding; in a certain sense the adaptive-filter decoder automatically learns which neural inputs are the most highly directionally tuned (Rapoport, 2007). Moreover, the analog-circuit–based implementation of the decoder described here processes all neuronal inputs in parallel, so the computational intensity of the decoding task does not constrain the number of neuronal inputs the system can handle. Consequently, the adaptive-filter approach to decoding scales favorably with the number of neuronal inputs to the system. This is an important virtue of adaptive-filter decoding, as decoding accuracy typically improves and more complex decoding tasks can be performed without sacrificing accuracy as more neuronal inputs are used (Chapin, 2004). Furthermore, improvements in multielectrode neural recording methodologies and technologies continue to facilitate recording from increasing numbers of neurons (Harrison et al., 2007; M. A. L. Nicolelis et al., 2003; Suner, Fellows, Vargas-Irwin, Nakata, & Donoghue, 2005; Wise, Anderson, Hetke, Kipke, & Najafi, 2004). The ability of the adaptive-filter decoder to handle large numbers of neurons provides an opportunity to explore a decoding regime less amenable to the Bayesian analysis of (Musallam et al., 2004). Figure 13 plots decoder performance as a function of the number of neuronal inputs (the horizontal line across the plot indicates the 0.25 threshold corresponding to unbiased guessing). While the corresponding performance curves for the Bayesian decoding algorithm were originally analyzed for up to sixteen neurons (Musallam et al., 2004), the computational efficiency of the adaptive filter enables performance to be evaluated for considerably larger numbers of neuronal inputs; the computation illustrated in Figure 13 is limited only by the total number of neuronal inputs available. The computation was performed under the standard training condition of 30 trials, and the error bars indicate the magnitude of a standard deviation after averaging over sets of randomized initial conditions and training inputs.

As expected, decoder performance increases from just above the 25% chance threshold to the maximum of approximately 65% reported earlier in this section. The lower curve corresponds to random selections of the input neurons, while the upper curve corresponds to preselection of neurons in order of decreasing variance in the mean firing rates over the four reach directions (higher variance indicates greater directional selectivity). The latter curve suggests that decoding input signals from the subset of neurons transmitting the greatest amount of directional information results in performance nearly equivalent to that obtained from using the full set of available signals. The 65% success rate of the decoding system, while comparable to that of other decoding algorithms tested on the same data (Musallam et al., 2004), indicates that there is considerable room for improvement. An outstanding question in the field of neural prosthetics concerns the degree to which intelligent users can compensate for imperfect decoding through biofeedback. Marked improvements in performance along these lines have been observed over time in both monkeys and humans (Carmena et al., 2003; Hochberg et al., 2006; Musallam et al., 2004; Taylor et al., 2002), but such contributions from biological learning are evidently insufficient. The work of (Musallam et al., 2004) indicates that performance can be improved by considering the temporal structure of the input neural signals at higher than zeroth order, and this might be achieved through changing the form of the filter kernels Wij.

EVALUATING DESIGN TRADEOFFS IN THE CONSTRUCTION OF AN IMPLANTABLE NEURAL DECODER Power Efficiency Simulations using the circuit designs presented in the section entitled “Implementation of the Decoding Algorithm in Analog Circuitry” indicate

603

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Figure 13. Decoding performance as a function of neuron number for randomly selected neurons (dark) and neurons selected on the basis of directional selectivity (light)

that a single decoding module (corresponding to an adaptive kernel Wij and associated optimization circuitry, as diagrammed in Figure 3) should consume approximately 54 nW from a 1-V supply in 0.18 μm CMOS technology (and require less than 3000 μm2 in area). This low power consumption is achieved through the use of subthreshold bias currents for transistors in the analog filters and other components. Analog preprocessing of raw neural input waveforms can be accomplished by dual thresholding to detect action potentials on each input channel and then smoothing the resulting spike trains to generate mean firing rate input signals. Simulations of the circuits presented in the section entitled “Spike–based decoding” indicate that each analog preprocessing module should consume approximately 241 nW from a 1-V supply in 0.18 μm CMOS technology. A full-scale system with n = 100 neuronal inputs  comprising N (t ) and m = 3 control parameters  comprising M (t ) would require m × n = 300 decoding modules and consume less than 17 μW in the decoder and less than 25 μW in the preprocessing stages.

604

Savings in Telemetry Power An advantage of our system is that its suitability for power- and area-efficient analog implementation can enable decoding in the implanted unit of a brain–machine interface, saving power by obviating the need for analog-to-digital conversion of neural signals before decoding, and by compressing data before wireless transmission. In this section we compare the power costs of internal analog decoding to those of external digital decoding. System power will depend on the bandwidth required to transmit digitized neural data to an external unit, so we consider three alternatives for digitization and telemetry, reflecting three approaches to trading algorithmic flexibility in decoding for total system power: (1) Transmission of digitized raw neural waveforms, preserving the greatest amount of neural information and requiring the greatest bandwidth; (2) Transmission of threshold-crossing events from neural waveforms, preserving only spike timing information and requiring intermediate bandwidth; (3) Transmission of control-parameter outputs from an implantable analog decoder, requiring minimal bandwidth

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

and internal power consumption but sacrificing algorithmic flexibility. In cases (1) and (2) initial 8-bit digitization of the input data at 30 kHz would require approximately 1 μW per channel (Yang & Sarpeshkar, 2006), contributing 100 μW in total. In these cases we consider implementing the external decoding system using a highly power-efficient digital signal processor (DSP) (“TMS320C55x Technical Overview (Literature Number SPRU393),” 2000) and analog-to-digital converters (A/Ds) (Yang & Sarpeshkar, 2006). (1) The first alternative requires transmitting all of the digitized data, corresponding to a bandwidth of 24 Mbs-1 for all n = 100 channels; at a rate of 1 mW per Mbs-1 (assuming typical link geometries and efficient topologies) (Mandal & Sarpeshkar, 2007, 2008) wireless data telemetry would consume 24 mW. (2) The second alternative requires dual thresholding for spike detection, followed by transmission of one bit per channel indicating the presence of a detected spike. Two comparison operations per period at 30 kHz across n = 100 input channels constitutes 6 × 106 operations per second. In practice (due to chip and board level parasitics) the most power-efficient DSPs operate at efficiencies of several hundred microwatts per MIPS (Millions of Instructions Per Second) (Verret, 2003); assuming an efficiency of 250 μW/MIPS, the power associated with these thresholding operations would be 1.5 mW. The bandwidth required to transmit the thresholded waveforms, assuming a maximal spike rate of 1 Hz per channel, would be 0.1 Mbs-1 for all n = 100 channels, corresponding to a transmission power of 0.1 mW. So the total power consumed by a system adopting this alternative for data digitization, compression and transmission would be approximately 1.7 mW. In estimating the power consumed in cases (1) and (2) we have not included the overhead of a DSP in excess of the power consumed in computations. In practice such overhead costs can be significant, contributing several milliwatts or more, depending on the processor used and its operating

settings (Verret, 2003). However, it is important to note that all DSPs incur static power costs due to CMOS leakage currents, independent of device activity and operating frequency. Leakage power depends primarily on operating voltage and temperature, and the processing core of one highly power-efficient DSP dissipates approximately 180 μW at 1.6 V and room temperature (25°C); this leakage power increases to approximately 277 μW at body temperature (37°C) (Verret, 2003). Therefore, our power estimates for DSP-based processing represent lower bounds on what would be consumed in a practical implementation. When our decoding algorithm is executed by a power-efficient DSP, the power expended in performing computations is negligible relative to the leakage power of the processor core. In a digital implementation of our decoder the adaptive filter kernels are transformed into discrete-time filters. A discrete-time implementation of the −

t

filter kernel Ae τ has a two-parameter z-transform, so each update cycle requires two multiplications. An update frequency of 10 Hz requires 20 multiplications per second, and scaling by n × m = 300 yields the computational rate required to execute the algorithm, 0.006 × 106 multiplications per second. At an efficiency of 250 μW/ MIPS, the power cost of the algorithm itself would therefore be only approximately 1.5 μW. (3) As discussed in the section entitled “Power Efficiency,” our simulations indicate that a lowpower-analog implementation of our first-order adaptive kernel decoder with (n, m) = (100, 3) could be built with a power budget of approximately 17 μW (54 nW per decoding module) and that an additional 25 μW is required for the preprocessing stages. Transmission of m = 3 decoded parameters for real-time control of an external device with 10-bit precision and an update frequency of 10 Hz requires a transmission rate of 300 bs-1 and an associated power of 300 nW. (Here we have assumed an impedance modulation telemetry scheme similar to the one reported in (Mandal &

605

A Biomimetic Adaptive Algorithm and Micropower Circuit Architecture for Implantable Neural Decoders

Sarpeshkar, 2007) and (Mandal & Sarpeshkar, 2008), which operates at an efficiency of 99.95% The transfer rate should be at least ≥ 3.4 kbps [Design requirement] The data transmission must be without information loss or artifact gain (Bit Error Rate 22 dB [Design requirement]) The system should have high performance (Delay < 7.5 s application-level Round Trip Time, Dropped Call Rate < 1.8%, Call Success Rate > 95%, Packet Error Rate < 10-4, Packet Lost Rate < 10-4) (Melero et al., 2002; Romero, Martinez, Nikkarinen & Moisio, 2002).

Figure 3. A simplified sketch of a telemedicine system

722

TELEMEDICINE AND MOBILE HEALTHCARE SYSTEM AND TECHNOLOGY A telemedicine system, in general, is composed of four main parts; 1) Patient unit /client side, 2) Communication network, 3) Receiver unit / server side, and 4) Presentation unit / user interface. Figure 3 is a simplified sketch of a general telemedicine system. A patient unit collects information / vital data from a patient, sends it as an analogue signal or performs A/D conversion and error correction, controls the data flow, and performs data transmission, using a transmission media. The type of the information that might be sent could be: audio, data, motion video, still images or fax. The network is responsible for data security and data transmission from patient side to hospital / server side. The architecture of the communication network could contain one or a combination of telecommunication services such as: Plain Old Telephone Service (POTS), Integrated Services Digital Network (ISDN), Asynchronous Transfer Mode (ATM), Global System for Mobile communication (GSM), Universal Mobile Telecommunications System (UMTS), Microwaves radio system and Satellite link. A server unit receives the data from the communication network, processes and converts it into a well defined format and then sends it to a presentation unit. The received data could be analogue or digital, and could be in a bit stream or data packets. A presentation unit receives the data from the server and transforms it into either a text or a graphical format, and presents the data to the user via a user interface (e.g. at the hospital).

Sensing of Vital Signs and Transmission Using Wireless Networks

No matter which technology or network a telemedicine system has, patient safety, data integrity and, privacy, system’s performance and reliability, data quality, applicability and interoperability are common aspects which should be considered, during a telemedicine system design and setup. A number of trials in telemedicine field such as modeling and system simulation, design and development, have been carried out by different research groups. These found a number of valuable issues to deal with, from which the following were selected. Istepanian et al. (Istepanian, Woodward, Gorilas & Balos, 1998) modeled and simulated a complete cellular digital telemedicine system, using cellular telephone channels. The system was studied using both Interim Standard 54 (IS54) and GSM cellular telephone standard. The feasibility of transmitting and receiving Photoplethesmography (PPG) data using simulated GSM and IS-54 cellular channels was investigated. The results showed a successful PPG transmission over both simulated standards. They found that the performance of the mobile telemedicine system was dependent on the degree of multi-path, noise and interference, which were presented in the specified communication channel. To test the performance of the GSM standard, the input PPG data were tested for different power delay profiles (rural environment, hilly environment, urban environment or flat fading), mobile speeds, and noise and interference conditions. In another study (Freedman, 1999), investigated two important issues in a telemedicine system, namely, timeliness and ease of ECG transmission, using a cellular phone (Nokia 9000) and GSM standard. To study the mentioned issues experimentally, an ECG transmission system to a mobile phone was developed. The system was an off-line (saves and forward) using the mobile phones inbuilt fax software and modem. The results showed that the system could solve the above mentioned issues. Namely, it required

very little effort for acquiring the ECG and the transmission to the fax mail-box was rapid, in this way the healthcare process was not delayed. Shimizu (Shimizu, 1999), proposed a concept providing a mobile telemedicine system for emergency care in moving vehicles. An experimental system for transmission of color images, an audio signal, three-channel ECG and blood pressure was developed. The system used a satellite link, and both a fixed and a cellular communication network for real-time medical data transmission. Some problems with the used technique were identified (the problems are not mentioned in his paper), but the theoretical analysis verified the feasibility of the proposed technique. Later on in 2001 Woodward et al. (Woodward, Istepanian & Richards, 2001), presented the design of a prototype integrated mobile telemedicine system that was compatible with the existing mobile telecommunication networks. The system utilized an Infrared Data Association standard (IrDA), GSM and ISDN. The system was in its first phase of development and the research was carried out producing a working system that allowed the transmission of only one parameter of data (ECG). The system was simulated but not fully developed. In 2003, Kyriacou et al. (Kyriacou et al., 2003) designed and tested a Multi-purpose HealthCare Telemedicine System using satellite, POTS system and GSM mobile communication link support, which combined real-time, and store and forward facilities. It seems that the final system was installed and used in Greece and Cyprus later on. They reported that the results from the system application were very promising using the GSM system. Zhao et al. (Zhao, Fei, Doarn, Harnett & Merrell, 2004) designed and developed a Telemedicine System for Wireless Home Healthcare Based on Bluetooth and the Internet. The system uses Bluetooth as wireless technology connecting the client side to the server side and conversely via

723

Sensing of Vital Signs and Transmission Using Wireless Networks

internet using dial-up modem, Digital Subscriber Line, or cable modem. Medical information and data were transmitted over short-range interface (Universal Serial Bus (USB) and RS232), wireless connection and the internet. The system was tested and the results showed promising performance for the wireless Bluetooth connection. It was revealed that the latency could be reduced as bandwidth increases. They found that the major source of data error arose from the serial port communication, when working in high noise environments. Jasemian and Arendt-Nielsen (2005a), have designed and implemented a telemedicine system using Bluetooth and GSM/GPRS, for real time remote patient monitoring. They evaluated the system applying a number of well defined tests and experiments. The results showed that the system fulfilled most of the requirements. They concluded that the implemented system was reliable, was functioning with a clinically acceptable performance, and was transferring medical data with a reasonable quality, even though the system was tested under totally uncontrolled circumstances during the patients’ daily activities (Jasemian & Arendt-Nielsen, 2005a & 2005b). They concluded that the system was applicable in clinical practice, since the patients as well as the involved nurses expressed their confidence and acceptance in using it. Therefore, they suggested that the real-time remote monitoring system might be generalized in clinical practice e.g. cardiology, and with small adjustment could be applied for monitoring other patient categories e.g. epileptic and diabetic patients. In spite of the differences in system setups, materials and methods, the encountered issues such as feasibility, applicability, bandwidth, throughput, performance, reliability and data quality of the telemedicine system, were common investigation issues in all the above mentioned studies. They are also the concern of this present chapter.

724

The State of the Art in Telemedicine Inevitably, healthcare goes mobile. Recently developed mobile healthcare (i.e., m-health) services allow healthcare professionals to monitor mobile patient’s vital signs and provide feedback to them anywhere at any time. The recent studies in the telemedicine field showed that wireless and mobile telemedicine systems are needed and are in focus at the present time. In this relation cellular networks and communication technologies such as GSM cellular telephone, infrared, Bluetooth, Satellite link, TCP/IP over the GSM/GPRS network or internet connection, Wireless Local Access Network (WLAN) have been utilized. Thus, using wireless and mobile technologies utilizing the existing public cellular network for telemedicine system design and setup is the stateof-the-art at the present time. Due to the nature of current supporting mobile service platforms, M-health services are delivered with a best-effort, i.e., there are no guarantees on the delivered Quality of Service (QoS) yet.

MOBILE HEALTHCARE SYSTEM STRUCTURE Mobile Healthcare (M-Health) can be regarded as a medical assistance service that integrates the technologies of medical sensors, mobile computing, information technologies and wireless communications into one complete remote system. Advances in technology have led to development of various sensing, computing and communication devices that can be woven into the physical environment of our daily lives. Such systems enable on-body and mobile health-care monitoring, can integrate information from different sources, and can initiate actions or trigger alarms when needed. An M-health system could be composed of four main parts; (1) Wireless patient units including wireless Network Access Point, (2) Mobile communication network, (3) Server unit, and (4)

Sensing of Vital Signs and Transmission Using Wireless Networks

Presentation unit. Figure 4 illustrates a simplified sketch of an M-Health system.In the successive subsections, those four mentioned parts will be described using the scenario in Figure 4.

Wireless Patient Units and Network Access Point Wireless patient units are a set of wireless devices that monitor patient’s vital signs and transmit measured data wirelessly to a mobile gateway or access point. In the scenario illustrated in Figure 4 four types of patient unit are applied. 1. A Bluetooth Pedometer or Step Counter; it is a Bluetooth enabled wireless device, that counts each step a patient takes by detecting the motion of his/her hips. Pedometers are inexpensive body-worn motion sensors that can easily be used by researchers and practitioners to assess and motivate physical activity behaviors (Tudor-Locke & Bassett, 2004). The device determines the patient’s physical activity indices and transmits those data via a Bluetooth connection to a mobile

access point, which the patient has on him/ her. 2. A Bluetooth enabled monitoring device for home telemonitoring of patients with chronic diseases. This mobile device is able to measure: Blood Oxygen saturation (SpO 2), Pulse rate, Breath rate and Body acceleration. The measured vital signs are transmitted to the mobile access point by Bluetooth connection. 3. A digital Bluetooth enabled device which monitors, measures and transmits the patients Blood Pressure over a Bluetooth connection to the mobile access point. 4. A Bluetooth enabled Precision Health Scale. The device is able to measure patient’s weight with a reasonable precision, and it is capable of wireless communication by Bluetooth to the mobile access point. A Bluetooth and GSM/GPRS enabled Personal digital assistant (PDA) device. The device acts as master in an ad-hoc network, which communicates with all patients’ units and establishes GSM/GPRS connection to a public mobile network when it is

Figure 4. The structure of an M-Health system; (1) Bluetooth Pedometer or Step Counter. (2) Bluetooth Monitoring Device for home telemonitoring of patients with chronic diseases. The device is able to measure: Blood Oxygen saturation, Pulse rate, Breath rate and Body acceleration. (3) Bluetooth Blood Pressure Monitor. (4) Bluetooth Precision Health Scale. (5) Bluetooth and GSM/GPRS enabled Network Access Point e.g. a smart mobile phone or a personal digital assistant (PDA) device applying mobile communication network, remote server at hospital and a monitoring device (presentation unit).

725

Sensing of Vital Signs and Transmission Using Wireless Networks

necessary. The availability of compact IP stacks has made it possible for Bluetooth to adopt the protocol suite which drives the Internet, and it is the Personal Area Network (PAN) profile which lays out the ground rules for doing that. The PAN profile provides the rule for carrying IP traffic across Bluetooth connection. A PAN user can connect to a Network Access point (NAP) in order to access a remote public network. Thus, the NAP provides a bridge to a public network. This NAP can be mobile (A) or fixed (B), as it is illustrated in Figure 5. These two ways of connection require three different roles: (1) Network Access Point (NAP) – A device acting as a bridge to connect a Pico net to an IP network. It forwards data packets to and from the network composed of PAN user. (2) Group ad-hoc network (GN) - A NAP device which connects to one or more PAN users, forwarding data packets between more than one connected PAN users. (3) PAN user – A client which uses the Group ad-hoc Network or NAP service (Bray & Sturman, 2002). In the scenario (Figure 4) a NAP is applied, which is a Bluetooth enabled access point. This access point automatically detects and connects Bluetooth devices to the Bluetooth network, confirming and connecting users with security and access rights configurations specified for each user. The current Bluetooth technology provides a data

transfer rate of 1Mbps, with a personal area range of up to 10m in client-to-client open air (5m in a building). In terms of client-to-access point, the current range is 100m in the open air and 30m in buildings. The present scenario utilizes client-toaccess point within the range of 30-100 m.

An Overview of Bluetooth Architecture Bluetooth is a wireless technology with a universal radio interface within the globally available 2.4 GHz frequency band, which makes the wireless data and telephony communication, in both fixed and mobile environment, possible. Bluetooth employs the globally available, license free Industrial Scientific Medical (ISM) frequency bands. Bluetooth technology is based on a low cost short distance wireless connection, which eliminates the necessity of cable connections. It provides movement freedom, via a wireless connection. It has a number of different security algorithms, such as authentication and encryption. By Bluetooth, it is easy to establish an ad hock connection for a local network. It is versatile system, since it can interact with different devices regardless of manufacturer. It is reliable within an open frequency band, ISM. Bluetooth does not exactly match the well known Open System Interconnect (OSI) standard reference model, which is an ideal

Figure 5. (A) Bluetooth enabled Precision Health Scale and a Bluetooth Blood Pressure monitor using mobile access point to communicate with a public communication network and/or Internet. (B) The same Personal Area Network (PAN) establishing Bluetooth connection to a fixed access point (AP) to communicate with a public communication network and/or Internet.

726

Sensing of Vital Signs and Transmission Using Wireless Networks

reference model for a communication protocol stack. Figure 6 illustrates an OSI reference model contra Bluetooth protocol stack for comparison. In OSI model, the Physical Layer is responsible for the electrical interface to the communications media, including modulation and channel coding. This covers the Radio and a part of Baseband tasks in Bluetooth. The Data Link Layer in OSI is responsible for transmission, framing, and error control over a particular link. This overlaps the control part of the Baseband including error checking and correction and the Link Controller in Bluetooth. The Network Layer in OSI is responsible for data transfer across the network. This covers the higher end of the Link Controller (which is setting up and maintaining multiple links), and also covers most of the Link Manager (LM) task in Bluetooth. In OSI, The Transport Layer is responsible for reliability and multiplexing of data transfer across the network to the level provided by the application. It covers the higher end of LM and the Host Controller Interface (HCI) in the Bluetooth. HCI indeed provides the actual data transport mechanisms. The Session Layer in OSI is responsible for the management and data flow control services. This is provided by Logical Link Control and Figure 6. OSI reference model contra Bluetooth protocol stack

Adaptation Protocol (L2CAP) and the lower end of the RFCOMM (protocol for RS-232 serial cable emulation)/SDP (Service Discovery Protocol) in the Bluetooth protocol. Finally, the Application Layer in OSI is responsible for managing communications between host applications, which covers the Applications in the Bluetooth. In spite of the mentioned differences, Bluetooth covers all necessary communication protocol layers for a reliable wireless connection.

Bluetooth Device, Discovery and Ad Hock Connection Bluetooth is able to establish an ad-hock wireless connection to a variety of Bluetooth enabled mobile and fixed devices in surroundings. The connection establishment can be arranged to happen selectively and automatically. In order to establish a connection, both ends of the link have to be willing to connect. Thus, a connection can not be forced to accept if it is not selected or not in the correct mode. Figure 7 illustrates the steps when a Bluetooth enabled telemetry device (patient’s unit) finds the correct NAP, in order to establish a connection to the monitoring centre at hospital for medical data transmission. The PDA here acts as a Network Access Point (NAP) using the Dial-Up Networking (DUN) profile. It periodically scans the surroundings to see if any device wants to use it. When the patient’s unit wants to establish a connection to the remote monitoring centre it needs a DUN connection to connect to NAP. To do that, it employs the DUN profile of the Bluetooth protocol. The first stage in such connection is to find out if any Bluetooth enabled device in surroundings supports the DUN profile. The Bluetooth enabled patient’s unit performs an inquiry, by sending a set of inquiry packets to look for NAP devices in the neighborhood. The NAP replies with a Frequency Hop Synchronization (FHS) packet which contains all the information that the patient’s unit needs for a

727

Sensing of Vital Signs and Transmission Using Wireless Networks

Figure 7. The necessary steps to take when a patient’s unit using the Bluetooth protocol finds the correct NAP in order to establish a connection to the remote monitoring centre for medical data transmission

connection establishment to the NAP. The information packet contains also the device class. Every existing Bluetooth enabled device in the neighborhood that scans for inquiries, will respond with a similar FHS packet, thus the Bluetooth enabled patient’s unit accumulates a list of devices. However, the patient’s unit automatically chooses the NAP device with pre-known address among the found devices. The next stage is to find out if the NAP supports the DUN profile. To do that, the patient’s unit uses the Service Discovery Protocol (SDP). First the patient’s unit pages the NAP, using the gathered information during the inquiry. The NAP responds if it is scanning for pages, then an Asynchronous Connection Less (ACL) baseband connection is set up to transfer data (control and configuration data) between these two entities. After an ACL connection establishment, a Logical Link Control and Adaptation Protocol (L2CAP) is set up and used. The patient’s unit uses the L2CAP channel to set up a connection to the SDP server in the NAP.

728

The SDP client in the patient’s unit asks the SDP server in the NAP to send all the information it has about the DUN profile. The SDP server in the NAP searches its database and returns the attributes relating to the DUN. This SDP information provides the patient’s unit all needed information to connect to the DUN service on the NAP. The patient’s unit configures the link using the Link Management Protocol (LMP) to meet its requirement, which in this case is DUN connection. Once the ACL and L2CAP connection are set up, an RFCOMM (an RS-232 emulator layer) connection it also set up, and then the DUN profile uses the RFCOMM. Each protocol has its own channel number. In this case, the NAP’s channel number for DUN was previously sent to the patient’s unit in the DSP packet. So, the patient’s unit knew in advance which number it should be used for setting up a RFCOMM connection. Thus, the DUN is set up using the RFCOMM connection and the patient’s unit starts to use the DUN service of the NAP.

Sensing of Vital Signs and Transmission Using Wireless Networks

Finally, the patient’s unit uses the NAP to perform connection across the public mobile network.

Bluetooth Security Since anyone, possibly, could eavesdrop wireless transmission, security has become an important key issue for wireless communication systems. Bluetooth uses SAFER+ cipher, which generates 128-bit cipher keys from a 128-bit pain text input, for link encryption and authentication in order to insure that the device at the two ends of the communication are who they claim to be. The SAFER+ has been designed by Cylink Corporation as a candidate for the U.S. Advanced Encryption Standard (AES) (Bray et al., 2002). The basic objective of security arrangements in Bluetooth is to provide means for a secure link layer, which encompasses entity authentication (facilitate access control and “Hardware” identification), and link privacy (eavesdropping is not easy). The applied key types are Link Keys (128 bit) and Encryption keys (8-128 bit). For Pairing it establishes secret keys and authentication. For encryption algorithm uses a stream cipher algorithm. The encryption engine is initiated with a random number using a random number machine. When the initiation is succeed the encryption engine uses four inputs for the encryption process, namely, (1) A number to be encrypted or decrypted

(this is the data being passed between devices), (2) The Master’s Bluetooth device address (the device address of the initiator device), (3) the Master’s Bluetooth slots clock, (4) A secret key which is shared by both devices. Additional to the above mentioned security arrangement the high speed pseudo-random frequency hopping algorithm in Bluetooth makes it very difficult to eavesdrop a wireless connection (Bray et al., 2002).

MOBILE COMMUNICATION NETWORK, SERVICES AND WIRELESS TECHNOLOGIES This section explores the most applied wireless and cellular technologies in most part of the world, in order to get an overview on the key enabling technologies for a mobile healthcare system. Figure 8 summarizes frequency allocations for mobile phones, cordless phones, and Wireless Local Access Network (WLAN) technologies in Europe, USA and Japan (Schiller, 2000).

Global System for Mobile Communication (GSM) GSM is a most successful digital mobile telecommunication system. It is the most popular digital system in Europe, approximately 100 million individuals use GSM, more than 130 countries use

Figure 8. frequency allocations for some of most used mobile phone, cordless phone, Wireless Local Access Network (WLAN) in Europe, USA and Japan

729

Sensing of Vital Signs and Transmission Using Wireless Networks

this system, and it has 40% of the market share. GSM allows integration of different telephony and communication services, and it adapts to the existing network. A GSM mobile device has approximately 0.125 – 0.25 W maximum output power. It has full duplex, synchronous, 1.2, 2.4, 4.8 and 9.6 kb/s data rate, and full duplex asynchronous, from 300 to 9.6 kb/s data rate (Schiller, 2000; Melero, wigard, Halnen & Romero, 2002). It has 9.6 kb/s and 14.4 kb/s Bandwidth. GSM interplays with PSTN, ISDN, and Public Switched Packet Data Network (PSPDN).

High Speed Circuit Switched Data (HSCSD) The High Speed Circuit Switched Data (HSCSD) is designed to enhance the data transmission capabilities in GSM system. HSCSD is circuit switched as basic GSM is, and it is based on connection-oriented traffic channel of 9.6 kbps each channel. By combining several channels, HSCSD increases the bandwidth and enhances the data transmission capabilities. A Mobile Station, in theory, could use all eight time slots within a Time Division Multiple Access (TDMA) frame to achieve an air interface user rate of 115.2 kbps (Schiller, 2000; Melero et al., 2002). HSCSD seems to be attractive at first glance, but HSCSD exhibits some major disadvantages. It applies connection-oriented mechanism of GSM, Figure 9. Channel arrangement in GSM and GPRS

730

which means the connection setup is performed before information transfer take place. In this way, transferring a large amount of bursty data (e.g. computer data traffic) may require all channels reserved. This channel allocation will be reflected in the service cost directly. Furthermore, for n channel allocation, HSCSD requires n time signaling for connection setup, connection release and during handover, as in HSCSD each channel is treated separately. Thus, the probability of blocking or service degradation increases during handover and connection setup or release, as Base Station Controller (BSC) has to check the resources for n channels. HSCSD might be an attractive solution for higher bandwidth and constant traffic (e.g. file download) in spite of being a costly solution.

General Packet Radio Service (GPRS) GPRS is a data communication service for a GSM system. As it is illustrated in Figure 9, for one GPRS radio channel the GSM can allocate 1-8 timeslots within one Time Division Multiplexing frame (TDM). The time slots are allocated on demand and are not fixed. All time slots can be shared by the active users. This means that each time slot is multiplexed for up to 8 active users. Allocation of the time slots is based on current

Sensing of Vital Signs and Transmission Using Wireless Networks

load and operator preferences (Schiller, 2000; Romero, Martinez, Nikkarinen & Moisio, 2002). The traffic channel is dependent on the coding scheme. Allocating all time slots using the coding for 14.4 kbps traffic channel, results in 115.2 kbps channel. GPRS is independent of channel characteristic and channel type. It does not set any limitation regarding the maximum data rate (just the GSM transport system limits the transmission rate). GPRS keeps an open connection, staying online to receive and send data at all times when it is needed. In GPRS, data is sent in packets, and up to three time slots can be combined to provide the necessary bandwidth, up to 39.6 kbps for receiving data, depending on coding scheme (Schiller, 2000; Romero et al., 2002). GPRS is an IP-based connection, which means that a high transmission capacity is only used when needed. This makes it possible to stay connected via GPRS, whereas keeping a constant circuit switched connection would be more expensive. Using GPRS, this provides data and internet/ intranet access, for a PC, PDA or any handheld device connected via Bluetooth wireless technology, infrared or cable.

Enhanced Data Rates for GSM Evolution (EDGE) EDGE is an enhanced version of GPRS, as it combines digital Time Division Multiple Access (TDMA) and GSM. It is anticipated that by using EDGE you can reach 85% of the world, using dual-mode handsets. EDGE applies enhanced modulation schemes and another technique, using the same 200 kHz wide carrier and the same frequency as GSM. In an arrangement of 48-69.2 kbps per time slot it offers up to 384 kbps data rate (Schiller, 2000; Hakaste, Nikula & Hamiti, 2002).

Universal Mobile Telecommunications System (UMTS) The European proposal for IMT-2000 prepared by ESTI is called Universal Mobile Telecommunications System (Dasilva 1997 and Ojanperä 1998). IMT-2000 is called future public land mobile telecommunication system. The major aims of UMTS are personal mobility, removing any distinctions between mobile and fixed networking and supporting the Universal Personal Telecommunication concept of ITU. UMTS offers both narrowband (2 Mbps) and broadband (> 100 Mbps in 60 GHz band) type of services (Schiller, 2000). By UMTS implementation any user will be able to approach any fixed or mobile UMTS terminal nationally or internationally. UMTS intends to provide several bearer services, real-time and non real-time services, circuit and packet switched transmission, and many different data rates (Schiller, 2000). High Speed Downlink/Uplink Packet Access (HSDPA and HSUPA) are advanced data services that are now being deployed on the UMTS networks worldwide. Together, HSDPA and HSUPA offer reduced latency and much higher data rates on the downlink and uplink. They are expected to help users in a mass market for mobile IP multimedia services. UMTS was not fully implemented at the time of the study; therefore it was not possible to investigate its communication general aspects in practice.

Digital Enhanced Cordless Telecommunication (DECT) Digital enhanced cordless telecommunication (DECT) standard was developed in 1992 within the European Telecommunication Standard Institute (ETSI). DECT technology is successor technology for CT2 and CT3 digital wireless telephone systems in Europe. Comparing DECT with Bluetooth, DECT has lack of spontaneous and ad hock network management mechanism. Furthermore, Bluetooth is more protected in

731

Sensing of Vital Signs and Transmission Using Wireless Networks

Figure 10. Properties of digital enhanced cordless telecommunication compared to the properties of Bluetooth

respect to the data security management. Figure 10 compares Bluetooth contra DECT properties (Schiller, 2000). DECT is dependent on a fixed infrastructure while Bluetooth is not. This makes Bluetooth more feasible for the implementation of an M-health system comparing with DECT.

Infra Red Data Association (IrDA) Infra Red Data Association (IrDA) with 900 nm wavelengths provides data communication system based on infrared light. The technology is simple and cheap and is integrated in almost all mobile devices today. Electrical devices do not interfere with infrared transmission (Schiller, 2000). The main disadvantage is that infrared is quite easy shielded (transmission can not penetrate walls or obstacles). Infrared contra Bluetooth properties are compared in Figure 11. IrDA creates only data communication by applying infrared light, while Bluetooth creates its connection by Radio Frequency (RF). IrDA is limited to an optical directional wireless connection, while Bluetooth is Omni-directional. IrDA is not able to penetrate furniture and wall, while Bluetooth does that. Bluetooth and IrDA are both

short distance wireless technologies, but IrDA has much less range than Bluetooth. Bluetooth has more complete network architecture, compared to IrDA which has no capability of internet working, media access, or other enhanced communication mechanisms. IrDA is typically limited to only two participants (point-to-point connection). Data security in Bluetooth is more protected than in IrDA.

Wireless LAN Wireless Local Access Networks (LAN) offer much higher access data rates than do cellular mobile networks, but provide limited coverage – typically up to 50 meters from the radio transmitter, while GSM/GPRS and WCDMA offer widespread – typically nationwide – coverage. Ericsson provides IEEE 802.11b based 2.4GHz WLAN enterprise systems today, and will continue to update this service with new capabilities. WLAN IEEE 802.11 technology offers infrared transmission in addition to radio transmission, whereas High Performance Local Area Network (HIPERLAN) and Bluetooth rely only on radio transmission (Schiller, 2000).

Figure 11. Bluetooth compared with Infra red data association (IrDA)

732

Sensing of Vital Signs and Transmission Using Wireless Networks

Figure 12. Bluetooth properties compared with the properties of WLAN IEEE 802.11

In this respect, WLAN compared to Bluetooth has a limitation concerning patient’s mobility, as it needs an infrastructure and it depends on a fixed access point in the patient’s house and partly line-of-sight (LOS) between sender and receiver. Bluetooth does not need a fixed infrastructure and is not restricted to LOS. Therefore Bluetooth provides more mobility when integrated in a device using a mobile cellular network. Figure 12, summarizes the difference between Bluetooth and WLAN 802.11.

Terrestrial Trunked Radio (TETRA) TETRA is one of a number of digital wireless communication technologies standardized by European Telecommunications Standards Institute (ETSI). ETSI Technical Committee (TC) Terrestrial Trunked Radio (ETSI TC TETRA) shall produce standards for a frequency spectrum for Professional Mobile Radio (PMR) and Public Access Mobile Radio (PAMR) Operators to support voice and data services using techniques such as Trunking, Time Division Multiple Access (TDMA) methods in narrow band RF channels and/or a variety of efficient modulation schemes in multiple allocations of narrow band RF channels for increased data throughput. TETRA applies another arrangement for wireless data transmission. This system uses many different carriers but assigns only one specific carrier to a certain user for a short period of time according to a demand. This radio system offer interfaces to a fixed telephone network, i.e., voice

and data, but is not publicly accessible. It is reliable and cheap, but it does not have nationwide coverage. TETRA offers bearer services up to 28.8 kbps and 9.6 kbps. It offers two standards, namely, Voice + Data (V+D) and Packet Data Optimized (Schiller, 2000).

Zigbee Specification for High Level Communication Protocols ZigBee is a low-cost, low-power, wireless standard which applies digital radio based on the IEEE 802.15.4 standard. ZigBee uses self-organizing mesh networks that can be used for industrial control, embedded sensing, medical data collection, smoke and intruder warning, building automation, home automation, etc. The low cost attribute allows the technology to be widely deployed in wireless control and monitoring applications. The low power-usage allows longer life with smaller batteries, and the deployment of mesh networking characteristic provides high reliability and larger range (ZigBee Standards Organization, 2008). ZigBee operates in the industrial, scientific and medical (ISM) radio bands; 868 MHz in Europe, 915 MHz in countries such as USA and Australia, and 2.4 GHz in most other countries. The technology is intended to be simpler and cheaper than other Wireless Personal Area Networks (WPANs) such as Bluetooth (ZigBee Standards Organization, 2008). The raw, over-the-air data rate is 250 Kbit/s per channel in the 2.4 GHz band, 40 Kbit/s per channel in the 915 MHz band, and 20 Kbit/s in the 868 MHz band. Transmission range is between 10

733

Sensing of Vital Signs and Transmission Using Wireless Networks

Figure 13. Bluetooth properties compared with the properties of ZigBee IEEE 802.15.4

and 300 meters, however it is heavily dependent on the particular environment (ZigBee Standards Organization, 2008). The maximum output power of the radios is generally 0 dBm (1 mW). Figure 13 summarizes the differences between ZigBee and Bluetooth technologies.

Wi-Fi Wi-Fi is the global brand name across all markets for any 802.11-based wireless LAN products. Wi-Fi stands for Wireless Fidelity and is meant to be used generically when referring of any type of 802.11 network, whether 802.11b, 802.11a, dual-band, etc. The term is formally proclaimed by the Wi-Fi Alliance. Any products tested and approved as “Wi-Fi Certified” (a registered trademark) by the Wi-Fi Alliance are certified as interoperable with each other, even if they are from different manufacturers. Formerly, the term “Wi-Fi” was used only in place of the 2.4GHz 802.11b standard, in the same way that “Ethernet” is used in place of IEEE 802.3. The Wi-Fi Alliance expanded the generic use of the term in an attempt to stop confusion

about wireless LAN interoperability. Typically, any Wi-Fi product using the same radio frequency (for example, 2.4GHz for 802.11b or 11g, 5GHz for 802.11a) will work with any other, even if not “Wi-Fi Certified. As Wi-Fi is widely used WLAN, some of its important features are compared with ZigBee, and Bluetooth in Figure 14.

Server Unit A server unit receives medical data from the communication network, processes and converts it into a well defined format and then sends it to a presentation unit. The server unit could be e.g. a Pentium 4 computer, 2GHz or faster running on Windows 2000 or XP operating system, with minimum 1 GB RAM, 80 GB hard disk with a built in database. It could be connected either to a mobile modem or to the internet, and could be situated at a hospital, care/ health centre or alarm centre. The server software carries out the following functions: •

It allows the registration of client terminals to make them able to communicate to the

Figure 14. the properties of Bluetooth IEE 802.15.1, ZigBee IEEE 802.15.4 and WI-Fi IEEE 802.11b are compared

734

Sensing of Vital Signs and Transmission Using Wireless Networks

•

•

•

remote server. The process uses the stored information in the database of the server to identify each user; telephone number, device address and IP address assigned by e.g. the GPRS network. It detects the communication request from a caller trying to communicate with the server. If the caller is not registered in the server, the caller will be warned that a conversation is not possible. When the server accepts the incoming call of a user and the communication channel is established, the server drives the data traffic between the client and the server. It receives the medical information as data packets, interprets and converts them to a synchronized data stream, before sending it to the monitoring/presentation unit.

Presentation Unit A presentation unit receives medical data from the server and transforms it into either a text or a graphical format, then presents the data to the user via a user interface (e.g. at the hospital). The presentation unit could be a computer, applying a user application program to convert the received medical data to a user-friendly graphical image or text format. The application program controls and detects true alarms and ignores false alarms. The presentation unit can monitor many patients simultaneously. Each patient has a unique

identification number in addition to his/her name. Through the application program, the caregiver can communicate with the patient at any time either via text massage or by phone (e.g. IP-telephony). The patient also can communicate with the health personals either via Short Message Service (SMS) or by phone.

Patient’s Safety (Electromagnetic Waves and Specific Absorption Rate) Electromagnetic waves have a spectrum range from 30 Hz with a wavelength of approximately the half of the earth’s diameter to more than 1022 Hz high-frequency cosmic rays with a wave length smaller than the nucleus of an atom (Goleniewski, 2002; Schiller, 2000). Figure 15 summarizes the designation of the frequency bands and the corresponding wave length in this spectrum. In mobile telephony, radio waves within 450 – 2200 MHz frequency range are used (Schiller, 2000), which are a part of microwave frequencies. In fact, all modern communications are based on manipulation and controlling signals within this electromagnetic spectrum. The power density or intensity of radio signals emitted from a mobile telephone or a base station, wanes quickly following the Inverse square law (Parsons, 2000). Radio waves’ tendency to lose power intensity as distance from the antenna increases is considered to be useful from the safety point of view. Furthermore, in order to serve more

Figure 15. the designation of the frequency bands in electromagnetic spectrum and the corresponding wave lengths

735

Sensing of Vital Signs and Transmission Using Wireless Networks

subscribers, network providers develop networks of higher density. This higher density networks make it possible to keep the transmission power low, this property has been investigated and considered as an advantage in relation to an M-health system (Jasemian & Arendt-Nielsen, 2005a). The absorption of electromagnetic energy causes temperature rise in a tissue (Bernardi, Cavagnaro, Pisa & Piuzzi, 2000). Specific Absorption Rate (SAR) is defined as the mass averaged rate of energy absorption in tissue, and can be considered as a measure of the local heating rate. Testing the SAR of a mobile phone is an effective method of quantifying the amount of energy absorbed by biological tissue, particularly in the human body. To test the SAR value, standardized methods are utilized, with the phone transmitting at its highest certified power level (2 W) in all used frequency bands (Poljak, 2004; Ramqvist, 1997). Many different devices are tested using the same method and not only mobile handsets. Wireless LAN and Bluetooth are just a few of the available technologies, which have had SAR measurements taken. Patient safety is investigated in a previous research work (Jasemian et al., 2005a), where the applied mobile phone and Bluetooth module had 0.89 W/Kg maximum SAR value. This is less than the recommended SAR value (< 2.0 W/Kg in Europe, and < 1.6 W/ Kg in USA) (ICNIRP, 1996; ANSI/IEEE C95.11992, NJ 08854-1331], 1991- 1999; IRPA, 1988; WHO, 1993).

medical data should be impersonalized and access controlled. Data security in such M-health systems has also been investigated by the author (Jasemian et al., 2005a) using telecommunication technologies and services such as Bluetooth, GSM and GPRS. The applied technologies offer Access Control, Authentication, Data Encryption, and User Anonymity for data security and access arrangements.

MOBILE HEALTHCARE SYSTEMS AND STANDARDS Mobile healthcare and telemedicine systems like other fields must, of course, obey standard rules and regulations. Indeed, telemedicine benefits from a large number of standards such as medical informatics, digital images, messaging, telecommunications, equipment specifications and networking (Sosa-Iudicissa, Luis & Ferrer-Roca, 1998). Most used telemetry devices obey the following standards: European Telecommunications Standards Institute (ETSI 300 220), International Electro technical Commission (IEC 529), International standard for Electromagnetic compatibility (EMC) and testing of Medical Electrical Equipment (EN60601-1 and EN60601-1-2). The most used mobile phone and Bluetooth modules obey the ETSI as well as the Institute of Electrical and Electronics Engineers (IEEE) standards.

DATA SECURITY

GENERAL DISCUSSION, IMPLICATION AND CONCLUSION

Data security is an essential aspect in telemedicine, particularly in mobile healthcare systems, as the patient is moving from one location to another, and is not bound to a specific place. A mobile healthcare system applies wireless and cellular technologies. Wireless communication contrary to a cable connection is more likely to be exposed to eavesdropping. Therefore, the transmitted

The present chapter dealt with a comprehensive investigation of feasibility of wireless and cellular telecommunication technologies and services in a real-time M-health system. The chapter based its investigation, results, discussion and argumentation on a remote patient monitoring scenario (figure 4). GSM allows 8 timeslots allocation for one GPRS radio channel within one Time Division

736

Sensing of Vital Signs and Transmission Using Wireless Networks

Multiplexing frame (TDM), which results in an effective resource utilization for up to 8 active user and consequently higher bandwidth (115.2 kbps), however allocation of the time slots is based on current load and operator preferences (Schiller, 2000; Romero et al., 2002). GPRS is an IP-based connection, keeps an open connection, staying online to receive and send data at all times when it is needed, which make the service less expensive in respect to a standard GSM circuit switched connection. Using GPRS provides data and internet/intranet access, for a PC, PDA or any handheld device connected via Bluetooth wireless technology, infrared or cable. These properties make GPRS more attractive for implementation of an M-health system. The investigation shows that EDGE is an enhanced version of GPRS, which offers up to 384 kbps data rate as it combines digital Time Division Multiple Access (TDMA) and GSM. This can be a promising enhancement of, and an alternative to, the GPRS service. Although HSCSD appeared to be an attractive and promising alternative to GPRS, it exhibited some disadvantages, such as higher service costs comparing to a traditional GSM data service. Moreover, HSCSD required more signaling during each handover, connection setup and release, thus the probability of blocking or service degradation increased during each handover, and as a consequence the QoS in general was reduced. The HSCSD is an ideal alternative to the GPRS if the patient is monitored only at home, however this is an essential limitation for realization of an M-health system. DECT acquires a higher bandwidth and communication range; however Bluetooth has more protected data security arrangements compared to DECT. Bluetooth supports spontaneous and ad hock networking while DECT does not. This makes Bluetooth more feasible for the implementation of an M-health system comparing with DECT.

TETRA offers bearer services op to 28.8 kbps and 9.6 kbps. It offers two standards, namely, Voice + Data (V+D) and Packet Data Optimized. Its radio system offer interfaces to fixed telephone network, i.e., voice and data, it is reliable and cheap, it does not have nationwide coverage and it is not publicly accessible. Therefore, it doesn’t fulfill the requirement of an M-health system. ZigBee is a low-cost, low-power, wireless standard, uses self-organizing mesh networks that can be used for embedded sensing and medical data collection. The low cost attribute allows the technology to be widely deployed in M-health and monitoring applications, the low power-usage allows longer life with smaller batteries, and the deployment of mesh networking characteristic provides high reliability and larger range. The technology is simpler and cheaper than other Wireless Personal Area Networks (WPANs) such as Bluetooth, however the raw, over-the-air data rate (250 Kbit/s per channel) is much less than Bluetooth data rate. The technology is more attractive compared with Bluetooth as long as medical application requires data transmission with low data rate. Despite the high bandwidth of Wi-Fi which is the most widely used WLAN, it has essential limitations compared with ZigBee and Bluetooth. It requires more memory space, its battery life is very much shorter and it is dependent on a fixed infrastructure. However, its bandwidth is much higher than ZigBee and Bluetooth. Therefore, Wi-Fi is more suitable for home healthcare but not for an M-health solution which are intended to be used outdoors. The Universal Mobile Telecommunications System (UMTS) as a third generation of GSM mobile network was also considered as an alternative network. UMTS was not fully implemented at the time of the study; therefore it was not possible to investigate its communication general aspects in practice.

737

Sensing of Vital Signs and Transmission Using Wireless Networks

It has been shown that development of an M-health system, utilizing Bluetooth, GSM and GPRS is entirely possible (Jasemian et al., 2005a & 2005b & 2005c). To implements a real-time M-health system a wireless communication platform, utilizing Bluetooth protocol was designed, implemented and tested (Jasemian et al., 2005a). It has been shown that there is interplay between high throughput and high QoS (Jasemian et al., 2005a). To increase the probability of errorless packet transmission, and increase the QoS, the TCP/IP packet size was decreased, and by this the transmission traffic was reduced, as the transmission of a long packet generates more traffic. On the other hand, the error correction mechanism in GPRS and RFCOMM reduces the bandwidth, as it was dealt with a distributed data line. Thus, TCP/IP was triggered to retransmit a packet, which was stuck in the buffer waiting for retransmission; this increased the traffic unnecessarily. This issue appeared to be one of the limitations in the implementation process. The limited memory capacity in the patient unit or the Network Access Point (NAP) was another limitation; however it is believed that a patient unit with a higher memory capacity would easily overcome this problem. Furthermore, with dedication of telemedicine transmission channels by network providers, an additional system improvement can be achieved. It has also been demonstrated (Jasemian et al., 2005a) that a Bluetooth enabled M-health service combined with GSM/GPRS solution was applicable, but while promising, further work needs to be carried out before reaching ideal reliability and performance. The M-health system was tested and evaluated on healthy volunteers and heart patients while they were in their daily environments. Generally, the test results showed that the system, in a more realistic environment, had a reasonable reliability, performance and quality (Jasemian et al., 2005b; Jasemian et al., 2005c), compared to the system behavior in the laboratory (a more controlled circumstance/environment).

738

Comparing these two conditions, there was no significant difference in throughput, but there were clear differences in PER, PLR and Up-time. These variations can probably be due to the fact that in the more realistic (uncontrolled) conditions, volunteers and patients attended to different indoor and outdoor activities. These included moving in vehicles at varying speeds, being in different landscapes and building environments and possibly being close to different sources of interference. All factors which can significantly influence the reliability and the performance of the system. This drawback can probably be avoided by asking the patients to avoid specific circumstances, which may influence the systems behaviors during the monitoring period. The quality of the transmitted data was generally good, and only 9.6% of the transferred ECG, in the more realistic environments, had an unacceptable quality. It is believed that the data quality could be improved by using the most suitable disposable electrodes, and by instructing the patients in skin preparation, since most of the impaired quality originated from a bad skin-electrode connection. The M-health system has been evaluated and validated by a number of well defined tests and experiments. Comparing the characteristics of the designed system with the system requirements, it can be concluded that the designed and implemented system fulfils the requirements. The suggested system is reliable, functions with a clinically acceptable performance, and transfers medical data with a reasonable quality, even though the system was tested under totally uncontrolled circumstances during the patients’ daily activities. Thus, it can be concluded that the system is applicable in clinical practice. Both the patients and the involved healthcare personnel expressed their confidence in using it. Therefore, it can also be concluded that the M-health real-time remote monitoring system might be generalized in clinical practice e.g. cardiology.

Sensing of Vital Signs and Transmission Using Wireless Networks

FUTURE RESEARCH ON AN M-HEALTH SYSTEM When implementing a mobile or wireless healthcare solution, a complicated set of choices must be faced. Combinations of devices, network, software, infrastructure, application design and security can make or break a mobile project. Thus, the system requirements should be based upon a set of real scenarios. Knowledge about the existing technologies, services and their advantages and disadvantages, in addition to knowing what is practical, ease the choice and aid its success. The scenarios may be built upon two different conditions: 1. Providing mobility by M-health system only for monitoring vital medical signs on patients at home (Home Healthcare Monitoring) 2. Providing mobility by M-health system for monitoring vital medical signs on patients everywhere at any time (Pervasive Healthcare) In the first case it would be advantageous to apply a fixed Access Point (AP), Figure 5b, TCP/ IP backbone and internet connection for long rage communication, in order to use the power line to supply the fixed AP power, as there are always battery life limitations in an M-health solution. Based on the nature of the monitored signal, the required measuring frequency and the required bandwidth one of the globally applied wireless technologies can be chosen; namely, Bluetooth or ZigBee for short range communication. The choice should be made as a trade-off between what is feasible and what is necessary. In the second case (Figure 5a), it is suggested to take advantage of The UMTS properties for long distance communication. Again based on the nature of the monitored signal, the required measuring frequency and the required bandwidth one of the globally used wireless technologies can be chosen; namely, Bluetooth or ZigBee for short range communication. In this case power

consumption and battery life are the two main issues to deal with, as both the measuring unit and the mobile AP are battery dependent. Recent advances in embedded computing systems have led to the emergence of wireless sensor networks, consisting of small, batterypowered units with limited computation and radio communication capabilities. Sensor networks permit data gathering and computation to be deeply embedded in the physical environment. This technology has the potential to allow vital signs to be automatically collected and fully integrated into the patient care record and used for real-time triage, correlation with hospital records, and long-term observation. During the past few years, wireless sensor technology has shown great potential as an enabler of the vision of ubiquitous computing. One promising application of wireless sensor networks (WSNs) is healthcare. Recent advances in sensor technology have enabled the development of small, lightweight medical sensors such as pulse oximeters and electrocardiogram leads that can be worn by the patient while wirelessly transmitting data. This frees the patient from the confinement of traditional wired sensors, increasing his or her comfort and quality of life. However, wireless sensor networks generally have limited bandwidth and high data loss rate. However, based on the nature of the monitored signal, the required measuring frequency, the required bandwidth, power consumption and sufficient memory space are still the main issues. By reducing the amount of signal processing in the patient unit or the WSN and letting most of the required signal processing be done on the remote server, longer battery life can be achieved. In any case the most fundamental challenge is the security and privacy of sensitive patient data. Encryption must be done to ensure the confidentiality of the data. At the same time, a sensor receiving a query from a base station (and likewise a base station receiving data from a sensor) needs to have some way of verifying the identity of the

739

Sensing of Vital Signs and Transmission Using Wireless Networks

other party, and of ensuring that the data has not been altered from its source. Hence, mechanisms must exist for data authenticity and integrity. What makes security uniquely challenging for WSNs is that the computational, memory and bandwidth costs must be carefully balanced against the limited resources of the individual nodes. As it has been explored and described in the present section, the future trend is toward the patient’s involvement in her/his own treatment and health monitoring in own natural everyday’s environment. Thus, the patient needs to be equipped with mobile health care devices that provide him/her full mobility in a secure manner. To fulfill this need, the device must be user friendly, lightweight and battery driven. This requires a low power conception device using longer lifetime batteries. The sensors need to be wireless, small, and lightweight and preferably integrated in the patient’s cloths or should be worn easily. Moreover, the mobile health device needs to be equipped with high security arrangements. Thus, the trend of the future research will be development of embedded software design with minimum memory requirement for M-health services. The devices should have a build in intelligent data flow control mechanism with higher data security that is integrated in a cost-effective wireless sensor network. Moreover, there is an insisting demand for development of lightweight batteries with longer life-time. All these will enhance the M-health services significantly in the future, ensuring the patients, their relatives and the healthcare providers and elevate their compliance to M-health services.

REFERENCES Administration on Aging. (2002). Profile of older Americans. Administration on Aging, U.S. Department of Health and Human Services.

740

Agha, Z., Schapira, R. M., & Maker, A. H. (2002). Cost effectiveness of telemedicine for the delivery of outpatient pulmonary care to a rural population. Telemedicine Journal and e-Health, 8, 281–291. doi:10.1089/15305620260353171 Bai, J., Zhang, Y., Shen, D., Wen, L., Ding, C., & Cui, Z. (1999). A portable ECG and Blood Pressure Telemonitoring System. IEEE Engineering in Medicine and Biology, 18(4), 63–70. doi:10.1109/51.775490 BASUMA Project. Berek, B., & Canna, M. (1994). Telemedicine on the move: healthcare heads down the information superhighway. AHA Hospital Technology Series, 13, 1–55. Bernardi, P., Cavagnaro, M., Pisa, S., & Piuzzi, E. (2000). Specific Absorption Rate and Temperature Increases in the Head of a Cellular-Phone User. IEEE Trans. MTT, 48(7). Bray, J., & Sturman, C. F. (2002). BLUETOOTH 1.1: Connect Without Cables. In J. Bray C. F. Sturman, B. Goodwin (Eds.), (pp. 12, 33, 52-53, 112, 224-230, 320-328). New Jersey. Coyle, G., Boydell, L., & Brown, L. (1995). Home telecare for the elderly. Journal of Telemedicine and Telecare, 1, 183–185. Dansky, K. H., Palmer, L., Shea, D., & Bowles, K. H. (2001). Cost analysis of telehomecare. Telemedicine Journal and e-Health, 7, 225–232. doi:10.1089/153056201316970920 European Commission Information Society Directorate-General. (2001). Telemedicine Glossary, 3rd Edition, Glissary of standard, concepts, technologies and users. Unit B1: Applications relating to Health, working Document. Fields, J., & Casper, L. M. (2001). America’s families and living arrangement population characteristics. P20-537. U.S. Bureau of the Census, U.S. Department of Commerce, Washington, D.C.

Sensing of Vital Signs and Transmission Using Wireless Networks

Freedman, S. B. (1999). Direct Transmission of Electrocardiograms to a Mobile Phone for Management of a Patient with Acute Myocardial Infarction. Journal of Telemedicine and Telecare, 5, 67–69. doi:10.1258/1357633991932315 Goleniewski, L. (2002). Telecommunications Essentials. P. Education (Ed.), (pp. 28-30 & 444449). Bosyon: Addition-Wesley Gott, M. (1995). Telematics for health: The role of Telemedicine in Homes and Comminities. Oxford, UK: Radcliffe Med. Press. Hakaste, M., Nikula, E., & Hamiti, S. (2002). GSM/EDGE Standard Evolution. In T. Halonen, J. Romero, & J. Melero, (Eds.), GSM, GPRS and EDGE performance, Evolution Towards 3G/UMTS (pp. 3-63). West Sussex: John Wiley & Sons LTD. He, W., Sengupta, M., Velkoff, V. A., & DeBarros, K. A. (2005). 65+ in the United States: 2005. Special report issued by the U.S. Department of Health and Human Services and the U.S. Department of Commerce. Washington, DC. ICNIRP. (1996). Health issues related to the use of hand-held radiotelephones and base transmitters. International Commission on Non-Ionizing Radiation Protection, Health Physics (pp. 70-4, 587–593). IEEE (1991- 1999). Standard for Safety levels with respect to human exposure to radio frequency electromagnetic fields (3 kHz to 300 GHz). ANSI/ IEEE C95.1-1992, NJ 08854-1331. New York. IRPA. (1988). Guidelines on limits of exposure to radio frequency electromagnetic fields in the frequency range from 100 kHz to 300 GHz. International Radiation Protection Association. Health Physics, 54, 115–123.

Istepanian, R. H., Woodward, B., Gorilas, E., & Balos, P. A. (1998). Design of Mobile Telemedicine Systems using GSM and IS-54 Cellular Telephone Standards. Journal of Telemedicine and Telecare, 4(Supplement 1), 80–82. doi:10.1258/1357633981931579 Jasemian, Y. (December 2006). Security and privacy in a wireless remote medical system for home healthcare purpose. 1st International Conference on Pervasive Computing Technologies for Healthcare [CD-ROM] s. 3, & Proceedings in IEEE Xplore, peer reviewed conference article, 29 November-1, Innsbruck, Austria. Jasemian, Y. (2008). Elderly Comfort and Compliance to Modern Telemedicine System at home. 2nd International Conference on Pervasive Computing Technologies for Healthcare, Proceedings, peer reviewed conference article, ISBN 978-9639799-15-8. Jasemian, Y., & Arendt-Nielsen, L. (2005a). Design and implementation of a telemedicine system using BLUETOOTH and GSM/GPRS, for real time remote patient monitoring. The International Journal of Health Care Engineering, 13, 199–219. Jasemian, Y., & Arendt-Nielsen, L. (2005b). Evaluation of a real-time, remote monitoring telemedicine system, using the Bluetooth protocol and a mobile phone network. Journal of Telemedicine and Telecare, 11(5), 256–260. doi:10.1258/1357633054471911 Jasemian, Y., & Arendt-Nielsen, L. (2005c). Validation of a real-time wireless telemedicine system, using Bluetooth protocol and a mobile phone, for remote monitoring patient in medical practice. European Journal of Medical Research, 10(6), 254–262. Kammann, J. (2001). Proposal for a Mobile Service Control Protocol. Institute for Communications and Navigation German Aerospace Centre (DLR), PDCS Anaheim (USA),

741

Sensing of Vital Signs and Transmission Using Wireless Networks

Kyriacou, E., Pavlopoulos, S., Berler, A., Neophytou, M., Bourka, A., Georgoulas, A., et al. (2003). Multi-purpose HealthCare Telemedicine System with mobile communication link support. BioMedical Engineering OnLine. 1-12. Lin, C. C., Chiu, M. J., Hsiao, C. C., Lee, R. G., & Tsai, Y. S. (2006). Wireless health care service system for elderly with dementia. IEEE Transactions on Information Technology in Biomedicine, 10(4), 696–704. doi:10.1109/TITB.2006.874196 Lo, B. P. L., & Yang, G. Z. (2005). Key Technical Challenges and Current Implementations of Body Sensor Networks. In Proc. of 2nd Intl. Workshop on Wearable and Implantable Body Sensor Networks (BSN 2005) (p. 1). London, UK. Magrabi, F., Lovell, N. H., & Celler, B. G. (1999). A Web-based approach for electrocardiogram monitoring in the home. International Journal of Medical Informatics, 54, 145–153. doi:10.1016/ S1386-5056(98)00177-4 Melero, J., Wigard, J., Halonen, T., & Romero, J. (2002). Basics of GSM Radio Communication and Spectral Efficiency. In T. Halonen, J. Romero, & J. Melero (Eds.), GSM, GPRS and EDGE performance, Evolution Towards 3G/UMTS (pp. 145- 190). West Sussex: John Wiley & Sons LTD. MobiHealth Project (last update, 2008), Orlov, O. I., Drozdov, D. V., Doarn, C. R., & Merrell, R. C. (2001). Wireless ECG monitoring by telephone. Telemedicine Journal and e-Health, 7, 33–38. doi:10.1089/153056201300093877 Parsons, J. D. (2000). Fundamental of VHF and UHF Propagation. In J. D. Parsons (Ed.), the Mobile radio propagation channel (pp. 15-18). West Sussex: John Wiley & Sons LTD. Patel, U., & Babbs, C. (1992). A computer based, automated telephonic system to monitor patient progress in home setting. Journal of Medical Systems, 16, 101–112. doi:10.1007/BF00996591

742

Perednia, D. A., & Allen, A. (1995). Telemedicine technology and clinical applications. Journal of the American Medical Association, 273, 483–488. doi:10.1001/jama.273.6.483 Philips Medical Systems. (February 2005). Striving for cable less monitoring. In C. Ragil (Ed.), Philips Medical Perspective Magazine, 8, 24-25. Poljak, D. (2004). Human Exposure to Electromagnetic Fields. In D. Poljak (Ed.), (pp. 11-40 & 49-52). Southampton: WIT press. Ramqvist, L. (1997). Health and Safety in Mobile Telephony. EN/LZT 123 4060, Ericsson Radio Systems AB, 164 80 Stockholm, 1-14. Rezazadeh, M., & Evans, N. (1990). Multichannel physiological monitor plus simultaneous fullduplex speech channel using a dial-up telephone line. IEEE Transactions on Bio-Medical Engineering, 37, 428–432. doi:10.1109/10.52352 Romero, J., Martinez, J., Nikkarinen, S., & Moisio, M. (2002). GPRS and EGPRS Performance. In T. Halonen, J. Romero, & J. Melero (Eds.), GSM, GPRS and EDGE performance, Evolution Towards 3G/UMTS (pp. 230-248). West Sussex: John Wiley & Sons LTD. Romero, J., Martinez, J., Nikkarinen, S., & Moisio, M. (2002). GPRS and EGPRS Performance. In T. Halonen, J. Romero, & J. Melero (Eds.), GSM, GPRS and EDGE performance, Evolution Towards 3G/UMTS (p. 291). West Sussex: John Wiley & Sons LTD. Satava, R., Angood, P. B., Harnett, B., Macedonia, C., & Merrell, R. (2000). The physiologic cipher at altitude: telemedicine and real-time monitoring of climbers on Mount Everest. Telemedicine Journal and e-Health, 6, 303–313. doi:10.1089/153056200750040165

Sensing of Vital Signs and Transmission Using Wireless Networks

Scalvini, S., Zanelli, E., Domenighini, D., Massarelli, G., Zampini, P., & Giordano, A. (1999). Telecardiology community: A new approach to take care of cardiac patients. Cardiologia (Rome, Italy), 921–924. Schiller, J. H. (2000). Mobile Communications. In J. H. Schiller (Ed.) (pp. 24, 27, 83-92, 86, 105113, 119-125). London, UK: Addition-Wesley. Shimizu, K. (1999). Telemedicine by Mobile Communication. IEEE Engineering in Medicine and Biology, (pp. 32-44). Sosa-Iudicissa, M., Luis, J., & Ferrer-Roca, O. (1998). Annex I: Standardisation Bodies. In O. Ferrer-Roca & M. Sosa-Iudicissa (Eds.), Handbook of telemedicine (pp. 197-209). Amesterdam: IOS press, Amesterdam. Tudor-Locke, C., & Bassett, D. R. Jr. (2004). How many steps/day are enough? Preliminary pedometer indices for public health. Sports Medicine (Auckland, N.Z.), 34(1), 1–8. doi:10.2165/00007256-200434010-00001

Uldal, S. B., Manankova, S., & Kozlov, V. (1999). Choosing a PC-based ECG System for a Mobile Telemedicine unit. The Health News magazine (pp. 30-31). Woodward, B., Istepanian, R. S. H., & Richards, C. I. (2001). Design of a Telemedicine System Using a Mobile Telephone, IEEE, (pp. 13-15). World Health Organization WHO. (1993). Environmental Health Criteria 137, Electromagnetic Fields (300 Hz to 300 GHz), Geneva. World Wide Web. (July 2002). Zhao, X., Fei, D., Doarn, C. R., Harnett, B., & Merrell, R. (2004). A Telemedicine System for Wireless Home Healthcare Based on Bluetooth and the Internet. Telemedicine Journal and e-Health, 10(supplement 2), 110–116. doi:10.1089/1530562042632038 ZigBee Standards Organization (January 17, 2008). Copyright © 2007 All rights reserved, ZigBee Specification, ZigBee Document 053474r17.

This work was previously published in Mobile Health Solutions for Biomedical Applications, edited by Phillip Olla and Joseph Tan, pp. 180-207, copyright 2009 by Medical Information Science Publishing (an imprint of IGI Global).

743

744

Chapter 3.8

Neural Networks in Medicine: Improving Difficult Automated Detection of Cancer in the Bile Ducts Rajasvaran Logeswaran Multimedia University, Malaysia

ABSTRACT Automatic detection of tumors in the bile ducts of the liver is very difficult as often, in the defacto non-invasive diagnostic images using magnetic resonance cholangiopancreatography (MRCP), tumors are not clearly visible. Specialists use their experience in anatomy to diagnose a tumor by absence of expected structures in the images. Naturally, undertaking such diagnosis is very difficult for an automated system. This chapter DOI: 10.4018/978-1-60960-561-2.ch308

proposes an algorithm that is based on a combination of the manual diagnosis principles along with nature-inspired image processing techniques and artificial neural networks (ANN) to assist in the preliminary diagnosis of tumors affecting the bile ducts in the liver. The results obtained show over 88% success rate of the system developed using an ANN with the multi-layer perceptron (MLP) architecture, in performing the difficult automated preliminary detection of the tumors, even in the robust clinical test images with other biliary diseases present.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Neural Networks in Medicine

INTRODUCTION There are a large number of algorithms and applications that have been and are actively being developed to assist in medical diagnosis. As medical problems are biological in nature, it is expected that nature-inspired systems would be appropriate solutions to such problems. Among the most popular of such nature-inspired tools is the artificial neural network (ANN), which has lent itself to applications in a variety of fields ranging from telecommunications to agricultural analysis. There is a large amount of literature on the use of ANN in medical applications. Some examples of medical systems developed employing neural networks include those for screening of heart attacks (Furlong et al., 1991) and coronary artery disease (Fujita et al., 1992), facial pain syndromes (Limonadi et al., 2006), diabetes mellitus (Venkatesan & Anitha, 2006), psychiatric diagnosis (NeuroXL, 2003), seizure diagnosis (Johnson et al., 1995), brain injuries (Raja et al., 1995), and many more. There is an increasing number of cancer cases in most countries, with an increasing variety of cancers. Over the years, ANN has been actively employed in cancer diagnosis as well. ANN systems have been developed for cancer of the breast (Degenhard et al., 2002), skin (Ercal et al., 1994), prostate (Brooks, 1994), ovaries (Tan et al., 2005), bladder (Moallemi, 1991), liver (Meyer et al., 2003), brain (Cobzas et al., 2007), colon (Ahmed, 2005), lung (Marchevsky et al., 2004), eyes (Maeda et al., 1995), cervix (Mango & Valente, 1998) and even thyroid (Ippolito et al., 2004). ANN has also been used for cancer prognosis and patient management (Naguib & Sherbet, 2001). Although there has been extensive development of ANN systems for medical application, there are still many more diagnostic systems for diseases and organs that would be able to gain from this nature-inspired technology. This chapter proposes a multi-stage nature-inspired detection

scheme that mimics the radiologist’s diagnosis strategy, where most of the algorithms employed are themselves nature-inspired. The scheme is augmented with the nature-inspired neural networks to improve the system performance in tackling automatic preliminary detection of a difficult and much less researched set of tumors affecting the bile ducts, using the defacto diagnostic imaging technology for the liver and pancreato-biliary system.

BACKGROUND Bile is used in the digestion and absorption of fat-soluble minerals and vitamins in the small intestines. In addition, it also has the function of removing soluble waste products from the body, including cholesterol. Diseases affecting the biliary tract cause distension (swelling) in the bile ducts, blockages, swelling of the liver and build up of toxic waste in the body, which can be fatal. Tumor of the bile ducts, medically known as cholangiocarcinoma, is the second most common primary malignant tumor of the liver after hepatocellular carcinoma and comprises approximately 10% to 15% of all primary hepatobiliary malignancies (Yoon & Gores, 2003). The incidence of this disease has been on the rise in recent decades (Patel, 2002). It is highly lethal as most tumors are locally advanced at presentation (Chari et al., 2008). These tumors produce symptoms by blocking the bile duct, often seen in clinical diagnosis as clay colored stools, jaundice (yellowing of the skin and eyes), itching, abdominal pain that may extend to the back, loss of appetite, unexplained weight loss, fever, chills (UCSF, 2008) and dark urine (Chari et al., 2008). The clinical diagnosis of the biliary tumor depends on appropriate clinical, imaging, and laboratory information (Yoon & Gores, 2003). Generally, after taking the medical history and performing a physical examination, the doctor would order one or more tests to get a better view

745

Neural Networks in Medicine

of the affected area before making the diagnosis. The common follow-up test for biliary tumors include non-invasive medical imaging tests such as computed tomography (CT) scans, magnetic resonance imaging (MRI) scans or ultrasound; (minimally) invasive imaging tests such as endoscopic retrograde cholangiopancreatography (ERCP), endoscopic ultrasound (EUS) or percutaneous transhepatic cholangiography (PTC); or even a bile duct biopsy and fine needle aspiration (UCSF, 2008). Treatments for this disease include surgery, liver transplantation, chemotherapy, radiation therapy, photodynamic therapy and biliary drainage (Mayo, 2008). Medical imaging has become the vital source of information in the diagnosis of many diseases, work-up for surgery, anatomical understanding of the body’s internal systems and functions etc. It is also the preferred technological method in aiding diagnosis as most of the medical imaging techniques are either non-invasive or minimallyinvasive, thus causing minimal discomfort to the patient and does not involve long (or any) hospitalization for recovery. Magnetic resonance cholangiopancreatography (MRCP) is a special sequence of magnetic resonance imaging (MRI) that is used to produce images of the pancreatobiliary region of the abdomen. This area covers the biliary system within and just outside the liver, as shown in Figure 1 (NIDDK, 2008). MRCP is now the preferred imaging test for biliary diseases as it is non-invasive, non-ionizing, has no side-effects nor requires any hospitalization. However, as MRI uses electromagnets, this technique is not applicable to any patients with implants affected by magnets. So far, there has been very little work done on developing automated computer-aided diagnosis (CAD) systems for tumor detection in MRCP images due to the high complexity in accurately identifying such diseases in a relatively noisy image. Although there are articles on clinical work, case studies and on the MRCP technology, the only automated biliary tumor detection scheme using MRCP

746

Figure 1. Location of the biliary tract in the abdomen

images readily available in the literature to date is the one developed by Logeswaran & Eswaran (2006). The performance of the ANN model proposed in this chapter will be compared against the published non-ANN model, in the Results section. To illustrate the complexity faced in automatic detection of tumors affecting the bile ducts, let us look at an example. Figure 2 shows a sample MRCP image containing a tumor at the indicated area. It is observed that there is no apparent identifiable structure in the area of the tumor, nor is the tumor boundary distinguishable from the background. Secondly, it should be observed that the image is noisy and affected by a lot of background tissue. Very often MRCP images are severely influenced by the presence of other organs (e.g. stomach and intestines) and tissue (e.g. liver). In addition, MRCP images are also prone to artifacts and bright spots. The orientation of the image during acquisition may also not be optimal in detecting a tumor, as there may be parts of the biliary structure itself overlapping the tumor. The intensity ranges of MRCP images are very susceptible to parameter settings, which are routinely tweaked by the radiographer in an attempt to best visualize the regions of interest.

Neural Networks in Medicine

Figure 2. MRCP image containing tumor of the bile duct (indicated by the arrow)

The signal strength or signal to noise ratio (SNR) decreases with the thickness of the image slices, but thicker slices are subject to greater partial volume effect (where larger amounts of the 3D volume is squashed into a 2D image causing loss of depth, and thus, structural information).

ANN Tumor Detection System With the complexities described in the previous section, conventional image processing techniques for structure detection and modeling would be hard-pressed to detect the presence of the tumor. As such, it was decided that it would be best to attempt to mimic as much of the processes involved in the natural methodology of diagnosis of such diseases by the specialists, and adapt them for implementation in computer systems. Through collaboration with a leading medical institution in Malaysia, which is also the liver diseases referral center of the country and a modern paperless hospital, radiologists were interviewed and observed on the diagnosis patterns used for identifying tumors in these MRCP images. Their diagnoses stemmed from their educational background and experience, the human visual system and biological neural network system in the brain, allowing

for powerful noise compensation and making associations within structures in an image. The important aspects observed that were vital in the diagnosis were basic knowledge of identifying characteristics, the ability to seek out important structures through the image noise and artifacts in the MRCP images, being able to compensate for inter-patient variations and inhomogeneity in intra-patients images (e.g. different orientations), in addition to the issues raised in the previous section. Based on the observed radiologist diagnosis methodology, a nature-inspired scheme comprising various nature-inspired components is proposed in an attempt to adapt computer systems, albeit in a rudimentary fashion, to mimic parts of their diagnosis method. The computer-aided system to be developed should be automatic as far as possible for ease of use and to allow automatic detection of the tumors. It is not meant to replace the medical experts in diagnosis but instead to assist them in screening large numbers of images and tagging those with potential tumors. Such a system can then be extended to highlight, navigate and manipulate pre-processed information, structures and parts of images, to aid the medical practitioners further in their diagnosis. Enabling an interactive interface also allows such a system to operate in a semi-automatic fashion, thus allowing

747

Neural Networks in Medicine

the medical practitioner to correct or modify the parameters and results at various stages of the implementations to overcome weaknesses of the automated system, for instance when faced with the inability to handle a very difficult pathology. Essentially, the algorithm involves several parts, as shown in Figure 3. Firstly, some image preparation or pre-processing is required to highlight the region(s) of interest and reduce the influence of noise in the image. Then, segmentation is required for preliminary abstraction of the objects from the background. Structure identification would be required for refining the segmentation process and labeling the biliary structures of interest. This will then be followed by the tumor detection phase, and finally enhanced with decision-making via the ANN implementation. Although at first glance the above steps may appear to be standard computer-aided methodology, each step taken consists of nature-inspired components. Some of the steps above are taken for granted by many as our natural human visual system undertakes them automatically. For instance, noise reduction, brightness and contrast compensation and image enhancement are all handled by our visual cortex such that when a radiologist studies MRCP images with such Figure 3. Flowchart of algorithm for tumor detection

748

“problems”, he/she will not face much difficulty or conscious effort in overcoming them and analyzing the important information within the images. Similarly, although the method of segmentation differs, when observing a picture we do naturally break it up into smaller regions to which we can assign some meaning, in order to understand the different parts of the picture and get a better sense of the entire picture. The segmentation algorithm employed in this work is natureinspired by rainfall on mountains, as will be described later. Structure identification is usually a conscious task as we identify the structures of interest (in this case the biliary ducts and tract) from other structures in the image (blood vessels, fat, tissue, artifacts, noise, organs etc.). The algorithm used in the structure identification consists of modified region-growing, which is inspired by fluid spreading to areas with desired characteristics (e.g. down a stream). From these structures, the next step, naturally, is to shortlist the potential tumor areas for further analysis. The final decision is made using the ANN. Although now more associated with their mathematical algorithms (Sarle, 1994), ANNs were originally nature-inspired after a crude model of the brain. It is used in this scheme to incorporate some intelligence,

Neural Networks in Medicine

tolerance to data inconsistencies, and as a tool to incorporate “learning” and storage of adaptive knowledge through training. In the interest of faster processing and reduced complexity, one major aspect of the natural diagnosis is not undertaken in the proposed scheme, which is the use of multiple MRCP slice images to conduct the diagnosis. A typical MRCP examination consists of a number of series of images, composed of locator slices, axial T2 sequence series, axial in-phase (fat saturation) series, MRCP thin slice series and MRCP thick slab images (Prince, 2000), and additional series as deemed necessary by the radiologist. Often, the total number of images acquired by the radiographer during a single examination may be in excess of a hundred images. In the event of doubt and for better understanding of the anatomy, a radiologist would scan through a number of the images, if not all of them. Implementing such ability in a CAD system is very tedious due to the large amount of variations in orientations and even sequences that may be used in a single examination, in addition to the need to approximate volumetric information lost due to the partial volume phenomenon during acquisition. Furthermore, it would be very processing intensive, taking up a lot of resources and time. Although not undertaken in this implementation, some recent work towards more rigorous 3D reconstruction of the biliary tract has been done in Robinson (2005) and with sufficient resources such as GPU (graphic processing unit) programming, may be undertaken in the near future to enhance the system proposed here. Such considerations will be described in the Discussion and Future Trends sections later in this chapter.

Image Preparation Image preparation essentially consists of two parts. Firstly, the region of interest (ROI) in the images should be determined, and then the image needs to be normalized it so that it is more consistent with the other images. ROI identification

may be undertaken in several ways. A popular manual method is by a rectangular selection box. Another method, in the case that the ROI is spread throughout most of the image, is to remove the parts that are not of interest, such as other organs. The image shown in Figure 2 is the result of a freehand selection area (not box), with obvious organs and artifacts removed from the boundary areas. For the ROI stage, it will be assumed that only appropriate MRCP images focusing on the biliary tract would be presented to the system. The images should have good orientation where the structures of interest are not unduly obstructed. Ideally, only perfectly oriented, clear, high quality images would be presented to the system, but that would be an unrealistic expectation even if most of the images were manually selected. So the system and consecutive stages in the algorithm will be developed to handle various imperfections in the images presented. The most popular method of normalization, which often reveals acceptable noise reduction, is via thresholding. However, due to the nature of MRCP, fixed values cannot be used as there is often a lot of inter-image variation, especially in terms of brightness and contrast. Instead, the thresholds need to be dynamically determined in each of the non-normalized MRCP images due to their varying intensity ranges and parameter settings. When a radiologist analyzes MRCP images with different intensity distributions (i.e. change in brightness and/or contrast), the relative information within the intensities is usually used, as opposed to the absolute intensity value. The human visual system compensates for this, and also fills in missing details to reconstruct the whole picture. A similar facility has to be implemented in the proposed system. First, histogram analysis is used to collect information relating to the image. Robinson (2005) studied typical MRCP images and found that the patterns in the histogram provided good estimates of the intensity range of the biliary structures in the image. The frequency histogram of the intensities

749

Neural Networks in Medicine

in an MRCP image is shown in Figure 4(a). From an analysis of a large number of images, it was found that the first global peak (P1) in an MRCP image corresponded to the intensity of most of the air present in the abdominal area, whilst the second peak (P2) corresponded to the intensity of the soft tissues (which includes the lower intensity biliary ducts). As such, the intensity distribution, if not the exact values, could be used to normalize the brightness and contrast between images. The minimum point between the two peaks (i.e. the trough, X) is used as the threshold intensity (the value of which varies for each image) to remove most of the air. The rest were found to be different intensities of the various parts corresponding to the biliary tract (as well as some noise, artifacts, background and other structures). The very high intensities (after Z) tend to be a few very bright spots, which make the rest of the image appear dull in comparison. Adapting these thresholds, by truncating to the right of Z, the biliary structures could be enhanced. To improve the enhancement, the middle area between Y (a point estimated 2/3 ways down the slope, considered as the beginning intensity of the more significant parts of the bile ducts) and Z are stretched so that the histogram between X and Z stretches to cover the original x-axis, enhancing the intensities of the main parts of the biliary ducts in the image. In practical implementations, the x-axis (intensity) would be scaled down from the original 12 bits per pixel (bpp) representation (i.e. 212

intensity range) in the images to the 8 bpp resolution of the conventional computer monitor (i.e. 28 intensities, ranging from 0-255). The processed histogram is shown in Figure 4(b). This represents the normalized intensities correctly on the conventional monitor. If the scaling is not done, there is a tendency for the monitors and programs to represent the 12 bpp intensities as cycles of the 0-255 intensity, i.e. intensity 256 in the 12 bit format would be displayed as 0 on the conventional monitor, thus wrongly representing the image. The outcome of this stage on a sample MRCP image is shown in a later section (in Figure 5).

Segmentation Good segmentation allows an image to be broken up into meaningful homogenous sections. Although infamous for its over-segmentation problem where images tend to be broken up into too many segments, the watershed algorithm (Vincent & Soille, 1991) does produce relatively accurate segmentation of edges. This morphological (shape based algorithm) is also nature-inspired as it was derived from the idea of watersheds formed in mountain ranges during rainfall. The topography (elevation) of the mountain represents the image intensity. The flow of the water down the gradients is the inspiration behind how the intensity flows in the image is analyzed. The water catchment basins (which are very often lakes) is the central

Figure 4. Dynamic histogram thresholding of an MRCP image

750

Neural Networks in Medicine

Figure 5. Effects of image preparation, segmentation and biliary structure detection

idea of the segments. Taking the intensity gradient image, the watershed algorithm produces segment boundaries at the steepest points of the gradient. This allows the image to be separated into semihomogeneous segments. Once the image has been segmented, as is expected with watershed, small spurious segments (cf. puddles) may occur in various parts of the image, most notably in the background. Also, as the intensity of the biliary structures do fluctuate throughout its length and width, which often makes it difficult to differentiate parts of the structure from the background, the spurious and very small segments need to be consolidated into the larger structures (either background or bile ducts). In order to overcome inter-pixel variations in a segment, the segments (instead of pixels) are considered as the minimal unit, and each segment’s average intensity is calculated, and all pixels within the segment are assigned this average intensity value. This is then followed by region merging, where small segments that are similar in terms of average intensity are merged to form larger, more meaningful segments.

The segments that merge most in this phase are those of the background. Ideally, the segmentation identifies all the background with a single label 0 and removes it. The intensity averaging and merging, although very useful in simplifying processing and overcoming intra-segment inconsistencies, should be taken with utmost caution as it could lead to elimination of parts of the biliary structure. In practice, using a test set of 256x256 MRCP images, it was found that the segments that generally fit the criteria for merging were those that were less than 30 pixels in size, and are merged to other segments with intensities within 10% of their own.

Biliary Structure Identification In the case of most of the tumors, the bile ducts that are of interest (in nearby locations) are those that are distended (enlarged) from the bile buildup due to the blockage caused by the tumor. As such, the first clue to the possible existence of a tumor (and for that matter, many of the diseases affecting the biliary ducts) is the presence of uncommonly large biliary structures in the MRCP image. The

751

Neural Networks in Medicine

segmentation should, ideally, have identified each meaningful segment as a separate biliary structure. However, due to the nature of the MRCP images, partial volume problems and intensity inconsistencies, some parts of the background may appear to be bile ducts and vice versa. Several strategies including scale-space analysis (another nature-inspired algorithm based on the human visual system) were attempted. Scalespace examines the same image at different scales (implemented as different levels of blurring of the image) in order to discern the significance of the different parts of the image. This is likened to looking at a picture of a fruit tree and first just seeing the overall shape of the tree, before focusing on the fruits, type of leaves etc. This method is effective in hierarchical analysis of an image, but only moderate results were achieved as the blurring process (even anisotropic blurring) caused too much loss of information. A related scheme using multiresolution analysis through directional wavelets such as contourlets also failed to produce sufficiently accurate results. A more promising yet simple approach of merging the segments meaningfully, and discarding non-biliary segments, was segment growing (i.e. an adaptation of the region growing algorithm at the segment level). The region-growing strategy too is nature-inspired as it mimics the natural expansion to areas or materials of similar characteristics. Such a growing strategy requires identification of appropriate starting source or seed segments. As the proposed system is to be able to undertake automatic detection, automatic seed selection is required. To identify the characteristics for determining the appropriate seed selection, test MRCP images were segmented and manually labeled into background and bile ducts, to collect the statistics and identify the characteristics for merging. Out of the many features analyzed in Logeswaran (2005), it was found that only average intensity, location and size influenced the correct selection of seed segments. A seed segment, it was found, is usually

752

a segment within the top 20% highest intensity, close to the middle of the image (i.e. the focus of the region of interest) and not too small (minimum 10 pixels in size). Using the characteristics above, all seeds meeting the criteria are selected automatically and grown. Many seeds are used due to the disjoint nature in tumor affected biliary ducts, where in a typical 2D MRCP image, the tumor obscures parts of the biliary tree structure, causing them to appear disjoint. Also, in the case of patients with other biliary diseases and prior biliary surgeries, as well as to compensate for possible low intensity areas of the biliary structures in the MRCP image, several seeds allow for all potential biliary structure areas to be identified. To grow the segments, once again appropriate characteristics had to be identified. These characteristics were discerned from the previously manually labeled images. Grown segments had to meet the criteria of having similar high average intensities (i.e. intensity greater than 200 with a intensity difference of less than 20%), small segment size (less than 100 pixels as larger segments were usually background), located nearby (maximum distance between the centers of gravity of the two corresponding segments was 10 pixels apart) and sharing a small border (maximum 30 pixels). Large segments with large borders often tend to be background or noise. For clarity, the effects of steps above until the structure detection of part of the biliary tract in an MRCP image are shown in Figure 5 (Logeswaran, 2006). Take note that the MRCP image in Figure 5(a) is that of a 50mm thick slab (i.e. 50mm volume squashed into 1 pixel thickness in a 2D image). Figure 5(c) shows the segments with each pixels assigned its average segment intensity. Through the various steps, the algorithm managed to successfully eliminate most of the background and even provide better 3D information approximation. The lower intensities usually represent parts further away; although lower intensities also represent structures deeper in the body and those influenced by some obstructions (e.g. fat).

Neural Networks in Medicine

It should be observed that although most of the background was removed, some still remains between the biliary tree branches on the right side of Figure 5(d). Increasing the thresholds or making the segment growing more stringent will not be able to completely remove such influences, as minor branches of the biliary tree are already adversely affected by the processing (see missing minor ranches on the left side of the same image).

Tumor Detection A tumor affects the bile ducts by impeding the flow of bile in the ducts, causing the mentioned swelling or distention in the biliary structures around the tumor. The tumor itself is not seen clearly in a T2 MRCP image, as it is solid and appears to be black (fluid is represented by higher intensities in such images). In radiological diagnosis, the presence of a tumor is suspected if the distended biliary structure appears disjoint (i.e. a black object overlapping part of the biliary structure). In the typical radiological diagnosis, the suspicion is confirmed by examining the appropriate series of MRCP images, other series (different orientations and even different sequence) and possibly ordering more tests and/or images, if necessary. As the proposed system is a preliminary detection system to aid in the diagnosis of the tumor, without implementing sophisticated and rigorous processing or requiring intensive processing power, the system is required at this stage to only shortlist images with potential tumors for further analysis, whilst discarding those that do not fit the profile. The algorithm looks for two or more disjoint biliary structures in moderately close proximity. The distance information to be incorporated is very subjective as the orientation of the images in the 3D plane heavily influences the way disjoint sections are viewed in a 2D plane. For example, two disjoint sections as seen in the coronal (front) view may appear joined in the sagital (side) view, with decreasing distance between the disjoint sections in the rotational angles between the coronal and

sagital views. A sample of this problem is given in the Discussion section later. As such, only the worst case scenario for distance, i.e. maximum distance, is used. From the test images, the sizes of the tumors were found to be less than 50 pixels wide, and consequently the maximum distance is set likewise. Each structure has to be sufficiently large to be considered as distended. As such, structures less than 50 pixels in size were not considered as a significant biliary structure. In the case of Figure 5(d), the small patch above the biliary structure could be eliminated this way. Ideally, the duct thickness (or a simple approximation of area divided by length) would be used as a measure of the distension, but it was found that the distension of the biliary structures in the case of biliary tumors were not significant enough (see Figure 2) to be incorporated as a criteria, as opposed to other biliary diseases such as cyst. So, the area (i.e. total number of pixels in a biliary segment) is used as the criterion. The actual structure analysis implementation in the developed system labels each biliary segment (i.e. after the segment merging and growing above) with different numbers. The size (in pixels) of each biliary segment is then calculated. If two or more sufficiently large biliary segments exist in the MRCP image, the image is considered as possibly containing a tumor.

ANN Implementation Issues Unfortunately, assumption of the size of disjoint segments can vary depending on the orientation of the acquired image. Furthermore, it is also greatly influenced by the size of the focus area in the image. To add to the complication, not all disjoint structures necessarily indicate cancer. An ANN is employed to improve the automatic learning and decision-making performance of the system. Inspired by the crude model of the brain, the ANN is powerful in detecting patterns in data and predicting meaningful output. For the purposes of this work, a training set of images are ascer-

753

Neural Networks in Medicine

tained, both with tumors and without. However, medical diagnosis is rarely that straightforward, as patients often suffer from more than one disease, and even preliminary detection should be able to exclude other common diseases present in the type of images being analyzed. To make the implementation realistic and applicable in the clinical environment, MRCP images with other biliary diseases present (such as stones, cyst and non-tumor swelling in the bile duct) are also included in the training and test sets. An ANN essentially approximates functions, and the characteristics pivotal to its performance rely very much on its architecture or network topology. Lately, there have also been many architectures of ANN introduced for specific problems. In addition to the very popular multi-layer perceptron (MLP) network, linear networks, recurrent networks, probabilistic networks, self-organizing maps (SOM), clustering networks, Elman, Kohonen and Hopfield networks, regression networks, radial basis functions, temporal networks and fuzzy systems, are just a few of the many existing ANN architectures that have been developed and incorporated into numerous applications to handle different classes of problems. Choosing the best network topology for the ANN allows it to produce the best results, but this is rarely obvious and certainly not easy. Literature such as (Sima & Orponen, 2003) and (Ripley, 1995) may be referred for a more detailed discussion on choosing an appropriate ANN architecture for specific problems, while there are numerous books and even Wikipedia (2008) providing information on the architectural configuration, learning algorithms and properties of the ANN. In the case of multi-layered architectures, the number of hidden layers corresponds to the complexity of problems that the network is able to process, while the number of neurons generally denotes the number of patterns that may be handled. Many of these networks are able to ascertain patterns in the input automatically, through supervised or unsupervised learning algorithms,

754

making them a powerful tool in data analysis and prediction. Decision-making information is usually stored as weights (coefficients) and biases that influence the neurons and the connections between them. There are many different learning algorithms as well as activation / transfer / squashing functions that may be used in the neurons. The learning or training algorithms determine how the appropriate values of the weights and biases are assigned to each neuron and connection, based on the data it is given during training (or at run-time, in the case of unsupervised or adaptive networks). The speed and quality of training is determined by a learning rate, whilst a learning goal (i.e. and error rate) is set to indicate that the network should be trained until the error achieved is below the set training goal. An important aspect with regards to the training of an ANN is that over-training an ANN makes it rigid and hinders its ability to handle unseen data. The error tolerance property of the ANN is very much dependent on its generalization ability, which is the strength of the ANN. As such, caution must be exercised during training such that over-zealous training to achieve high precision output must not render the ANN rigid. At the same time, the choice of training data is significant, and a good distribution of data covering typical and atypical scenarios improves the ANN’s ability in decision-making for a larger group of unseen data. In many cases, a maximum number of iterations (epochs) is also set to forcibly terminate training, thus preventing it from going on indefinitely, even when the desired training goal is not achieved. Bender (2007) states that when testing a large number of models, especially when using flexible modeling techniques like ANN, one that fits the data may be found by pure chance (Topliss & Edwards, 1979). A discussion on this issue, with regards to ANN can be found in Livingstone & Salt (2005). It is recommended that feature selection be used and refined until the “best” model is found. Good classification or regression result may be obtained by shortlisting possible features,

Neural Networks in Medicine

generating many models, conducting validation, and repeating the process until the “best” model based on the user’s preference is found. The rational is that very often, the actual feature may not be known at the onset, and will only be realized through such a refining process. The initial features in this work were selected manually through the observations described in the previous section. The model significance, often measured through metrics such as the “F measure” has been shown to vary with feature selection (Livingstone & Salt, 2005) and thus, such correlation coefficients and RMSE by themselves may not reflect the performance of the model. The necessary factors to be taken into account to reflect the model’s performance include the number of descriptors (parameters), the size of the dataset, the training/test/validation set split, the diversity of structures (in this work reflect the cancerous bile ducts), and the quality of the experimental data (Bender, 2007). Model validation requires splitting the data into 3 sets: training (to derive the parameters), validation (assess model quality) and testing (assess model performance). Such splitting is undertaken in this work. Alternatives to this method, especially for small datasets, include using cross-validation (split data to multiple equal parts – use one for testing and the rest for training) or leave-multiple-out splits. K-fold validation or receiver operating characteristic (ROC) curve could be used as well, to evaluate the statistical significance of a classifier. Ideally a clean dataset, measured by a single, well-defined experimental procedure, should be used in any experiment. This is not possible in this work as when dealing with typical medical data, it involves a variety of radiographers, patients, diseases and their many inconsistencies. However, the training, validation and test sets will each encompass a representative sample of the structure diversity of the data.

ANN Setup and Training The issues discussed in the previous section are those that should be taken into account when developing an optimal ANN system. However, very often, some compromises are made depending on the objectives of the work, available resources and other compounding factors. The objective of this chapter is not to produce the best bile duct tumor detection system, as that requires much more indepth research and implementation, but rather to show that the nature-inspired ANNs could be a possible solution in the ultimate development of a CAD system for that purpose using MRCP images. As such, this section concentrates on employing a popular but relatively simple ANN to the task and showing that even such an implementation with low resource requirements could provide good results, as will be discussed further in the following sections. Although facing many valid critiques in recent times, such as those in Roy (2000), Thornton (1994) and Smagt & Hirzinger (1998), the feedforward MLP ANN is still arguably the most popular ANN. It has, over the decades, provided good performance in a large number of applications with a broad base of data types. Easy availability, convenient and intuitive setup, automatic setting of weights and biases, and ability to be instructed on expected outcomes through supervised training, are among the reasons the MLP is set up as the final component of the proposed system to make the binary selection between tumor and non-tumor cases. The reader, however, is advised that alternative architectures and ANN setups may be experimented with for improved results. Some possible alternatives that may be implemented in the near future are discussed later in the chapter. The ANN used in this implementation is set up as shown in Figure 6. Figure 6(a) shows the basic architecture of a single perceptron model taking in j number of input streams or parameters. Each input is multiplied by its weight coefficient (w) acting on that stream, before being sent to the

755

Neural Networks in Medicine

perceptron node (i.e. the black circular node). The node itself will be subjected to a constant bias (b), producing the result f(x) as given by the formula in the figure. The output is dependent on the activation function that is used in the node. The output of a traditional hardlimiter (step activation function) perceptron is binary, so O(x) would be a 0 or 1 depending on a set threshold of the result f(x). Note that the output is usually denoted by f(x) itself as that is the activation function, but we are splitting it here and denoting a separate term O(x) to describe the calculation and the output individually. An MLP is made up of multiple layers of one or more perceptrons in each layer, as shown in Figure 6(b) for a fully connected MLP architecture. The nodes of the input layer in Figure 6(b) are shown in a different color (grey instead of black) as these are usually not perceptrons but merely

an interface to accept the input values in their separate streams. The input layer of the MLP used for this work consists of a neuron for each of the parameters identified in the previous section, namely, the average segment intensities and sizes of the two largest biliary segments, thus four neurons. Additional sets of input neurons could be used if more than two biliary segments are to be analyzed, although this was found to be unnecessary in the test cases. Depending on the stringent control of images being presented to the system, further input neurons to incorporate parameters including distance between the midpoint of the segments, thickness of each segment (can be approximated by dividing the area of the segment by its length), as well as shape characteristics, could be added. This was not undertaken in the proposed system as the available images were retrieved from the radiological archives spanning

Figure 6. Fully-connected feedforward MLP ANN with one hidden layer

756

Neural Networks in Medicine

a number of years. The acquisitions were by various radiographers under the supervision of various radiologists, and thus, were not taken under a stringent enough control environment for a more precise study. A priority in the setup is low computing power requirements and fast processing time. Since the features used as input to the system were few and of low complexity, a single hidden layer sufficed. 10 neurons were used to delineate the network in this layer, the number found experimentally to be suitable in producing the best results. Lesser number of hidden neurons caused the overall detection rate to deteriorate, whilst increasing the number of neurons (only up to 20, as the complexity was to be kept low) did not cause any appreciation of the results. As for the output, only a single neuron was required as the output would be binary, i.e. 0 indicating negative tumor detection or 1 indicating possible tumor detection. All neurons used were standard. For generalization and function approximation, the activation functions used in the neurons were sigmoidal and linear, in the hidden and output layers, respectively. No activation function is used in the input layer, but the input is normalized to a range of 0.0-1.0, where the maximum size would be the size of the image and the maximum intensity 255. As the ANN used is a supervised network, a set of training data and confirmed diagnosis results are required. 20% (120 images) of the acquired 593 MRCP images (see next section) were used as the training set. The images were selected pseudorandomly to include a distribution comprising proportionately of images consider as normal (healthy), diagnosed as bile duct tumor, and those with other diseases. The confirmed diagnoses for these images were used to determine the binary expected output which was fed to the MLP during training, labeled as 1 (containing tumor) or 0 (no tumor), for the corresponding image. The MLP uses the backpropagation learning algorithm to adjust the coefficients (weights) on each neuron interconnection (lines in the figure), such that the

error between the actual results produced by the network and the expected result is minimized. There are several variants but the basic gradient descent backpropagation was used in this work. The system is trained until the training converges to the training goal (mean squared error of 0.05) or maximum limit of training cycles (30,000 epochs) to prevent the training from going on indefinitely if it does not converge. In handling robust data, as the case was in this work, some cases of nonconvergence is to be expected. The maximum training limit was set to 30,000 as there was no observable improvement in the convergence beyond that. Based on past experience, a learning rate of 0.15 was used in this work as it was found that minor changes in the learning rate did not affect the outcome for these images, whilst large learning rates produced poor results.

RESULTS Before the system is tested in a robust clinical environment, it should be first tested for proof of concept. For this test, the algorithm is evaluated on its ability to differentiate images containing tumor from images of healthy patients. It must be remembered that the tumor may be present in different parts of the bile ducts, and that in biological systems, even the images of healthy patients can differ considerably. The choice of parameters, intensity thresholds, values of the segment sizes for merging and growing, and biliary segment size for tumor detection were varied and tested. The performance of the proposed system is very dependent on the choices made, which is in turn dependent on the quality of MRCP images obtained. The configuration and results presented in this chapter are for the setup that performed the best for the available 2D MRCP images. All the images used in this work, inclusive of the training and test data, were obtained from the archives of the collaborating medical institution. As the images were acquired there, and the

757

Neural Networks in Medicine

patients were also treated at the same place, the actual medical records and diagnosis information (in some cases further confirmed through ERCP or biopsy) were available. The presence and absence of tumors and other biliary diseases was further confirmed by the expert medical consultant, who is the senior radiologist heading the digital imaging department where the clinical and radiological diagnoses were made. The proposed ANN system is trained with 20% of the clinical 2D MRCP images acquired from the collaborating hospital, containing a distribution of diseases. The results obtained for 55 images in this first test using the trained ANN above are given in Table 1. As is observed from the table, the accuracy level is very good when validating the detection against normal healthy images. In cases where the classification was inaccurate, the description of the reason for the error is given in the table. In one of the cases where the tumor was not identified correctly, it was due to lack of distention which made the structures appear to be that of a healthy patient, whilst in the other case, the orientation made the biliary structures appear joint, indicating other biliary diseases. The normal image that was detected wrongly was not detected as a tumor, but merely as distended (dilated), as the biliary structure appeared unusually large, possibly due to some zooming during image acquisition. Such cases are unavoidable through a simple detection system using a single image. Table 1. Results of the tumor detection validation testing Medically Diagnosed Detected Accuracy % Error Description of Error

758

Normal

Tumor

27

28

26

26

96.3%

92.8%

1

2

Large biliary structures

1 normal, 1 non-tumor dilation

Next, for realistic medical diagnosis, all the available images (including those of healthy patients as well as those of patients with various biliary diseases) were used in the second test. Many patients suffer from more than one biliary affliction, and as such, many of the test images were influenced by more than one disease. This kind of testing allows for robust evaluation of the proposed system in the clinical environment for which it is targeted. This set of 593 images contained 248 healthy (normal), 61 tumor and the remainder images with other biliary diseases such as stones, cyst, residual postsurgery dilation, Klatskin’s disease, Caroli’s disease, strictures, and obstruction from other liver diseases. In some of the images with more than one disease, some of the visual characteristics were very similar between the different afflictions (e.g. dilation / swelling / distention is common in most biliary diseases). The results obtained are given in Table 2. As this is a robust testing, the evaluation criterion known as “Overall accuracy” is calculated as the percentage of the sum of the true positives and true negatives over the total test set. This overall accuracy takes into account how well the system performed in detecting the correct disease, as well as in rejecting the wrong diagnosis. For comparison, results from the earlier published work on automated tumor detection using MRCP images is given in the last row. As is observed from the table, the ANN implementation successfully improved the detection results for both the cases with tumor as well as for the healthy normal images. In terms of specificity (i.e. the non-tumor images correctly identified as not containing tumor), the ANN implementation performed well too, achieving 94.70% (an improvement from 89.85% without ANN). The algorithm prototype was developed in IDL as it allowed for fast development with readily available tools and libraries, without incorporating any significant optimization. In terms of timing performance, the entire process from pre-

Neural Networks in Medicine

Table 2. Results of robust testing using multidisease clinical MRCP images Normal Test images

Tumor

Others

Total

248

61

284

593

Overall accuracy (ANN)

83.64%

88.03%

-

-

Overall accuracy (Non-ANN)*

76.90%

86.17%

-

-

* Based on results from (Logeswaran & Eswaran, 2006)

processing until ANN output for each image on a standard Pentium IV desktop personal computer took less than a minute. IDL is a matrix-based software, and programs running in its run-time environment would not be as fast as those compiled into machine language. Thus, the timing performance could be dramatically enhanced by changing the platform and using optimized code in a language that allowed compilation to machine code, e.g. C or C++. The timing performance enhancement falls out of the scope of this chapter and is left for the reader to experiment with. The classifier’s performance could also be analyzed using machine learning research tools such as MLC++, WEKA etc.

DISCUSSION The results obtained show good overall accuracy for the ANN system, proving that the natureinspired strategy employed has merit. Although errors were present, the accuracy achieved is amicable taking into account that even visual diagnosis using single 2D MRCP images is very difficult and impossible in certain cases. This is compounded by the fact the test data contained a large number of images with various diseases, in which accurate detection algorithms had not been developed. Very often, even experienced radiologists would look through a series of images before confidently making a diagnosis.

In most of the failed cases, the influence of other diseases and inconsistency in image acquisition by various radiography technologists affected the detection. Examples of some of the failed cases are shown in Figure 7. The image in Figure 7(a) was acquired in the very same MRCP examination of the patient in Figure 2, taken at a different orientation that makes it impossible for the developed system and even a human observer to conclusively diagnose the tumor. In the case of Figure 7(b), the signal strength for the acquisition was poor (a norm with MRCP thin slices) and the liver tissue intensity masked the biliary structures. Again, such images would defeat the human as well. Although the proposed algorithm does incorporate several features at various stages to minimize the influence of a certain amount of weaknesses in the acquired images, problems due to cases as shown in Figure 7 are too severe. Overcoming such problems requires strict adherence of image acquisition standards and uniformity, which is beyond the scope of the work conducted. Some information relating to the MRCP protocol, examination and relevant information can be found at (Prince, 2000). What is important in the presented results is that even when faced with a difficult test set of mostly unseen data, a simplified implementation of a non-optimal ANN was able to perform well through its nature-inspired learning and processing capabilities, as the improvement in results obtained against the system in the literature relied entirely on the ANN’s learning abilities from the parameters fed into it. The non-ANN implementation used experimental hard-coded decisionmaking. Further benchmarking may be possible once such automated cholangiocarcinoma detection using MRCP implementations using other algorithms (possibly those employing advanced classifiers such as Support Vector Machine, Adaboost etc.) are available. Of course, to be fair for the purposes of benchmarking, an optimally selected and configured ANN topology with appropriate

759

Neural Networks in Medicine

Figure 7. Example MRCP images that caused tumor detection inaccuracies

customizations should be the used in the tests. Performance comparisons would also need to be undertaken under a controlled environment, implemented on the same platform, and using the same set of images. Of interest to this discussion is also the appropriate and correct use of the ANN in medical applications. There have been issues raised on the misuse of ANN in oncology / cancer in the past, as highlighted by Schwarzer et al. (2000). In the article, the authors studied literature published between 1991-1995 where ANN were used in oncology classifications (43 articles) and identified several common shortcomings in those articles. These include: •

•

•

•

760

mistakes in estimation of misclassification probabilities (in terms of biased estimation due to neglecting the error rate, and inefficient estimation due to very small test sets). fitting of implausible functions (often leading to “overfitting” the ANN with too many hidden nodes) incorrectly describing the complexity of a network (by considering the complexity to only cover only one part of the network, leading to underestimating the danger of overfitting), no information on the complexity of the network (when the architecture of the net-

•

• •

work used in terms of number of hidden layers and hidden nodes are not identified), use of inadequate statistical competitors (feedforward ANN are highly flexible classification tools and as such, it should only be compared against statistical tools of similar flexibility), insufficient comparison with statistical methods, and naïve application of ANN to survival data (building and using ANN with shortcomings to the data and situations being modeled, such as not guaranteeing monotonicity of estimated survival curves, omission of censored data, or not providing data with the proper relationships).

The above problems are still very common and should be taken into account for better understanding and use of ANNs. Care must always be taken, especially in developing medical applications, such that they are rigorously tested for accuracy and robustness before being adapted to the clinical environment. Pertinently, it must always be remembered that no matter how good a tool is, it is just a tool to facilitate the medical practitioner in obtaining the necessary information so that he/ she can made the diagnosis. A tool, such as the proposed system, is never to be used to replace the medical practitioner’s responsibility in making the final diagnosis. The objective of such a tool as

Neural Networks in Medicine

the one proposed here is to help flag suspicious images so that the medical practitioner may proceed with further analysis as necessary.

FUTURE TRENDS From its humble beginnings of a single layered, hardlimiter / step activation function and binary output perceptron, the ANN has evolved impressively over the years. In the Introduction section of this chapter, a very small number of examples of the vast work that has been done in employing ANN in medical applications were mentioned. From this, it is obvious that ANN is already a very established nature-inspired informatics tool in medical applications, in addition to its contributions in many other fields. More interesting is the development of hybrid networks that allow the strengths of various architectures and algorithms to be combined into a single network, giving it the ability to handle a greater variety of data and highly complex pattern recognition, apt for the increasingly complex modern systems. Such optimized ANNs boast of superior performance to traditional architectures. In addition to the ANN, there has been a marked increase in development of other nature-inspired algorithms and tools as well. Some of the popular ones include simulated annealing (mimics the crystallization process), genetic algorithms (mimics population growth), particle swarm optimization (mimics social behavior of birds flocking), ant colony algorithm and many more. These stateof-the-art algorithms often provide solutions to difficult problems. In this author’s opinion, many such algorithms can be adapted to collaboratively work with ANNs, to develop even more powerful tools. As such, there is definitely a trend for nature-inspired informatics in general, and neural networks specifically, to be used in larger proportions in the foreseeable future. A majority of implementations of ANN systems treat the ANN as a “black box” system where the

ANNs, through the training routine, determine its internal coefficient values. The setup proposed in this chapter is an example of that. Lately, there have been efforts for possible rule extraction from the trained network. Once the rules are known, they may be further optimized and/or be applied for other similar problems. Once perfected, such rule extraction becomes a powerful companion to the ANN systems. There has been a lot of work done in the handling of 3D data. Although most of the frontiers in this technology appear in the computer gaming as well as the film and animation worlds, the medical field has also begun employing more of such technology. Conventional MRI equipment visualizes 3D volume as a static series of 2D images, such as those used in this work. Although this is still the norm at most medical institutions worldwide, technological advancement has seen the introduction of many real-time and 3D systems such as 3D ultrasound, 3D CT, fMRI and many others. Furthermore, computer processing power, technology and memory are increasing dramatically. New paradigms in programming, such as those for the graphics processing units (GPU), are becoming popular. All these enable more 3D volume reconstruction and manipulation. The future in medical diagnosis is in 3D and as such, the next step is to upgrade systems such as the proposed one to become a complete 3D graphical user interface (GUI) ANN CAD system.

CONCLUSION This chapter started off by providing a brief introduction to the increasing efforts in incorporating automated and computerized technology that employs nature-inspired artificial neural networks (ANN) to undertake difficult diagnosis of various diseases in different organs. It then proceeded to the disease of interest, i.e. cancer, which is fast becoming a serious threat and cause of death in many nations. It is expected that there will a

761

Neural Networks in Medicine

significant increase in cancer deaths in the years to come. The central focus of the chapter was to introduce a nature-inspired algorithm mimicking the natural diagnosis strategies of medical specialists, aided by the nature-inspired methods and technology, for the preliminary detection of tumors affecting the bile ducts. The algorithm developed consisted of image enhancement, watershed segmentation, region growing, diagnosis-based evaluation, and ANN pattern recognition for detection. Almost all steps employed were inspired in one way of another by natural systems. The results obtained prove that such a nature-inspired approach can provide improved detection of difficult to observe tumors in 2D MRCP images in a clinically robust multi-disease test set, even when using a non-optimized simplistic setup. From the trends observed, it is expected that many more successful systems would be developed in the future, gaining from a broader outlook at various natural systems. With the formidable track record, ANNs are sure to lead the way to many more accomplishments. It is hoped that this chapter contributes ideas and examples that propagate such advancement to nature-inspired systems, especially in developing and improving much needed medical applications that would benefit from computer-aided technology.

ACKNOWLEDGMENT This work is supported by the Ministry of Science, Technology and Innovation, Malaysia, the Academy of Sciences Malaysia, and the Brain Gain Malaysia program. The author would like to express his appreciation to Dr. Zaharah Musa and Selayang Hospital, Malaysia for the medical consultation and clinical data (MRCP images and medical diagnosis) used in this research work.

762

REFERENCES Ahmed, F. E. (2005). Artificial neural networks for diagnosis and survival prediction in colon cancer. Molecular Cancer, 4, 29. http://www.molecularcancer.com/content/4/1/29. doi:10.1186/14764598-4-29 Bender, A. (2007, October). A Primer on Molecular Similarity in QSAR and Virtual Screening. Part III – Connecting descriptors and experimental measurements – model generation. QSARWorld Strand Life Sciences, (pp. 1-4). Brooks, A. C. (1994, December). Prostate cancer: diagnosis by computer - neural network trained to identify men with prostate cancer and to predict recurrence. Brief ArticleScience News. Retrieved February 21 2008, from http://findarticles.com/p/ articles/mi_m1200/↜is_n23_v146/ai_15972193. Chari, R. S., Lowe, R. C., Afdhal, N. H., & Anderson, C. (2008). Clinical manifestations and diagnosis of cholangiocarcinoma. UpToDate. Retrieved May 24 2008, from http:// www.uptodate.com/patients/content/topic. do?topicKey=gicancer/23806. Cobzas, D. Birkbeck, N., Schmidt, M., Jagersand, M., & Murtha, A. (2007). 3D Variational Brain Tumor Segmentation using a High Dimensional Feature Set. Workshop on Mathematical Methods in Biomedical Image Analysis, (pp. 1-8). Degenhard, A., Tanner, C., Hayes, C., Hawkes, D. J., Leach, M. O., & Study, T. U. M. B. S. (2002). Comparison between radiological and artificial neural network diagnosis in clinical screening. Physiological Measurement, 23(4), 727–739. doi:10.1088/0967-3334/23/4/311 Ercal, F., Chawla, A., Stoecker, W. V., Lee, H.-C., & Moss, R. H. (1994). Neural network diagnosis of malignant melanoma from color images. IEEE Transactions on Bio-Medical Engineering, 41(9), 837–845. doi:10.1109/10.312091

Neural Networks in Medicine

Fujita, H., Katafuchi, T., Uehara, T., & Nishimura, T. (1992). Application of Artificial Neural Network to Computer-Aided Diagnosis of Coronary Artery Disease in Myocardial SPECT Bull’s-eye Images. Journal of Nuclear Medicine, 33(2), 272–276. Furlong, J., Dupuy, M., & Heinsimer, J. (1991). Neural Network Analysis of Serial Cardiac Enzyme Data. American Journal of Clinical Pathology, 96(1), 134–141. Ippolito, A. M., Laurentiis, M. D., Rosa, G. L. L., Eleuteri, A., Tagliaferri, R., & Placido, S. D. (2004). Immunostaining for Met/HGF Receptor May be Useful to Identify Malignancies in Thyroid Lesions Classified Suspicious at Fine-Needle Aspiration Biopsy. Thyroid, 14(12), 1065–1071. doi:10.1089/thy.2004.14.1065 Johnson, M. A., & Kendall, G. Cote, P.J., & Meisel, L.V. (1995). Neural Networks in Seizure Diagnosis. Report no. A926592 (Army Armament Research Development and Engineering Center Watervliet NY Benet Labs). Retrieved February 21 2008, from http://www.stormingmedia. us/92/9265/A926592.html. Limonadi, F. M., McCartney, S., & Burchiel, K. J. (2006). Design of an Artificial Neural Network for Diagnosis of Facial Pain Syndromes. Stereotactic and Functional Neurosurgery, 84(5-6), 212–220. doi:10.1159/000095167 Livingstone, D. J., & Salt, D. W. (2005). Judging the significance of multiple linear regression models. Journal of Medicinal Chemistry, 48(3), 661–663. doi:10.1021/jm049111p Logeswaran, R. (2005). Scale-space Segment Growing For Hierarchical Detection of Biliary Tree Structure. International Journal of Wavelets, Multresolution, and Information Processing, 3(1), 125–140. doi:10.1142/S0219691305000750

Logeswaran, R. (2006). Neural Networks Aided Stone Detection in Thick Slab MRCP Images. Medical & Biological Engineering & Computing, 44(8), 711–719. doi:10.1007/s11517-006-0083-8 Logeswaran, R., & Eswaran, C. (2006). Discontinuous Region Growing Scheme for Preliminary Detection of Tumor in MRCP Images. Journal of Medical Systems, 30(4), 317–324. doi:10.1007/ s10916-006-9020-5 Maeda, N., Klyce, S. D., & Smolek, M. K. (1995). Neural Network Classification of Corneal Topography Preliminary Demonstration. Investigative Ophthalmology & Visual Science, 36, 1327–1335. Mango, L. J., & Valente, P. T. (1998). Comparison of neural network assisted analysis and microscopic rescreening in presumed negative Pap smears. Acta Cytologica, 42, 227–232. Marchevsky, A. M., Tsou, J. A., & Laird-Offringa, I. A. (2004). Classification of Individual Lung Cancer Cell Lines Based on DNA Methylation Markers Use of Linear Discriminant Analysis and Artificial Neural Networks. The Journal of Molecular Diagnostics, 6(1), 28–36. Mayo Clinic. (2008). Bile Duct Cancer. Retrieved May 24 2008, from http://www.mayoclinic.org/ bile-duct-cancer/. Meyer, C. R., Park, H., Balter, J. M., & Bland, P. H. (2003). Method for quantifying volumetric lesion change in interval liver CT examinations. Medical Imaging, 22(6), 776–781. doi:10.1109/ TMI.2003.814787 Moallemi, C. (1991). Classifying Cells for Cancer Diagnosis Using Neural Networks. Intelligent Systems and Their Applications, 6(6), 8–12. Naguib, R. N. G. (Ed.) & Sherbet, G.V. (Ed.) (2001). Artificial Neural Networks in Cancer Diagnosis, Prognosis, and Patient Management (Biomedical Engineering Series). CRC.

763

Neural Networks in Medicine

Neuro, X. L. (2003). Neural networks for medical and psychiatric diagnosis. Retrieved February 21 2008, from http://www.neuroxl.com/neural_networks_psychiatry.htm. NIDDK – National Institute of Diabetes and Digestive and Kidney Diseases. (2008). Digestive diseases dictionary A-D: biliary track. National Digestive Diseases Information Clearinghouse (NDDIC). Retrieved February 21 2008, from http://digestive.niddk.nih.gov/ddiseases/pubs/ dictionary/pages/a-d.htm. Patel, T. (2002). Worldwide trends in mortality from biliary tract malignancies. [PMID 11991810.]. BMC Cancer, 2, 10. doi:10.1186/1471-2407-2-10 Prince, M. R. (2000). MRCP Protocol. Retrieved 2008 May 25, from http://www.mrprotocols. com/MRI/Abdomen/MRCP_Dr.P_Protocol.htm. Raja, A., Meister, A., Tuulik, V., & Lossmann, E. (1995). A neural network approach to EEG classification in brain chemical injuries diagnosis. MEDICON’95: VII Mediterranean Conference on Medical & Biological Engineering (IFMBE) (pp. 133–133), Jerusalem, Israel.

Sima, J., & Orponen, P. (2003). General purpose computation with neural networks: A Survey of Complexity Theoretic Results. Neural Computation, 15, 2727–2778. doi:10.1162/089976603322518731 Tan, T. Z., Quek, C., & Ng, G. S. (2005). Ovarian cancer diagnosis by hippocampus and neocortexinspired learning memory structures. Neural Networks, 18(5-6), 818–825. doi:10.1016/j. neunet.2005.06.027 Thornton, C. (1994). The worrying statistics of connectionist representation. Technical Report CSRP 362, Cognitive and computing sciences. University of Sussex. United Kingdom. Topliss, J. G., & Edwards, R. P. (1979). Chance factors in studies of quantitative structure-activity relationships. Journal of Medicinal Chemistry, 22(10), 1238–1244. doi:10.1021/jm00196a017 UCSF Medical Center. (2008). Cholangiocarcinoma. Retrieved May 24 2008, from http:// www.ucsfhealth.org/adult/medical_services/ gastro/cholangiocarcinoma/conditions/cholang/ signs.html.

Ripley, B. D. (1995). Statistical ideas for selecting network architectures. Neural Networks: Artificial Intelligence and Industrial Application (pp. 183190). Springer.

v Smagt, P. d., & Hirzinger, G. (1998). Why feedforward networks are in bad shape. Proceedings 8th International Conference on Artificial Neural Networks (pp. 159-164).

Robinson, K. (2005). Efficient pre-segmentation. PhD thesis, Dublin, from http://www.eeng.dcu. ie/~robinsok/pdfs/Robinson PhDThesis2005.pdf

Venkatesan, P., & Anitha, S. (2006). Application of a radial basis function neural network for diagnosis of diabetes mellitus. Current Science, 91(9), 1195–1198.

Roy, A. (2000). Artificial neural networks – a science in trouble. Special Interest Group on Knowledge Discovery and Data Mining, 1, 33–38. Schwarzer, G., Vach, W., & Schumacher, M. (2000). On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology. Statistics in Medicine, 19, 541–561. doi:10.1002/ (SICI)1097-0258(20000229)19:43.0.CO;2-V

764

Vincent, L., & Soille, P. (1991). Watersheds in digital spaces: an efficient algorithm based on immersion simulations. Transactions on Pattern Analysis and Machine Intelligence, 13, 583–598. doi:10.1109/34.87344

Neural Networks in Medicine

Wikipedia (2008). Artificial neural network. Retrieved September 17 2008, updated September 14 2008, from http://en.wikipedia.org/wiki/ Artificial_neural_network.

Yoon, J.-H., & Gores, G. J. (2003). Diagnosis, staging, and treatment of cholangiocarcinoma. Current Treatment Options in Gastroenterology, 6, 105–112. doi:10.1007/s11938-003-0011-z

This work was previously published in Nature-Inspired Informatics for Intelligent Applications and Knowledge Discovery: Implications in Business, Science, and Engineering, edited by Raymond Chiong, pp. 144-165, copyright 2010 by Information Science Reference (an imprint of IGI Global).

765

766

Chapter 3.9

Image Registration for Biomedical Information Integration Xiu Ying Wang BMIT Research Group, The University of Sydney, Australia Dagan Feng BMIT Research Group, The University of Sydney, Australia & Hong Kong Polytechnic University, Hong Kong

ABSTRACT The rapid advance and innovation in medical imaging techniques offer significant improvement in healthcare services, as well as provide new challenges in medical knowledge discovery from multi-imaging modalities and management. In this chapter, biomedical image registration and fusion, which is an effective mechanism to assist

medical knowledge discovery by integrating and simultaneously representing relevant information from diverse imaging resources, is introduced. This chapter covers fundamental knowledge and major methodologies of biomedical image registration, and major applications of image registration in biomedicine. Further, discussions on research perspectives are presented to inspire novel registration ideas for general clinical practice to improve the quality and efficiency of healthcare.

DOI: 10.4018/978-1-60960-561-2.ch309

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Image Registration for Biomedical Information Integration

INTRODUCTION With the rapid advance in digital imaging techniques and reduction of cost in data acquisition, widely available biomedical datasets acquired from diverse medical imaging modalities and collected over different imaging sessions are becoming essential information resources for highquality healthcare services. Anatomical imaging modalities such as Magnetic Resonance (MR) imaging, Computed Tomography (CT) and X-ray mainly provide detailed morphological structures. Functional imaging modalities such as Positron Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT) primarily reveal information about the underlying biochemical and physiological changes. More recently, the combination of functional and anatomical imaging technologies into a single device, PET/CT and SPECT/CT scanners, has widened the array of biomedical imaging approaches and offered new challenges in efficient and intelligent use of imaging data. Since each of these imaging technologies can have their own inherent value for patient management, and ideally all such imaging data would be accessible for the one individual when they are required, huge volumes of biomedical imaging datasets are generated daily in the clinical practice (Wang et al, 2007). However, these ever-increasing huge amounts of datasets unavoidably cause information repositories to overload and pose substantial challenges in effective and efficient medical knowledge management, imaging data retrieval, and patient management. Biomedical image registration is an effective mechanism for maximizing the complementary and relevant information embedded in various image datasets. By establishing spatial correspondence among the multiple datasets, biomedical image registration enables seamless integration and full utilization of heterogeneous image information, thereby providing a more complete insight into medical data (Wang, Feng, 2005) to

facilitate knowledge discovery and management of patients with a variety of diseases. Biomedical image registration has important applications in medical database management, for instance, patient record management, medical image retrieval and compression. Image registration is essential in constructing statistical atlases and templates to capture and encode morphological or functional patterns across a large specific population (Wang and Feng, 2005). The automatic registration between patient datasets and these available templates can be used in the automatic segmentation and interpolation of structures and tissues, and the detection of pathologies. Registration and fusion of information from multiple, diverse imaging resources is critical for accurate clinical decision making, treatment planning and assessment, detecting and monitoring dynamic changes in structures and functions, and is important to minimally invasive treatment (Wang, Feng, 2005). Due to its research significance and crucial role in clinical applications, biomedical image registration has been extensively studied during last three decades (Brown, 1992; Maintz et al., 1998; Fitzpatrick et al. 2000). The existing registration methodologies can be catalogued into different categories according to criteria such as image dimensionality, registration feature space, image modality, and subjects involved (Brown, 1992). Different Region-of-Interests (ROIs) and various application requirements and scenarios are key reasons for continuously introducing new registration algorithms. In addition to a large number of software-based registration algorithms, more advanced imaging devices such as combined PET/CT and SPECT/CT scanners provide hardware-based solutions for the registration and fusion by performing the functional and anatomical imaging in the one imaging session with the one device. However, it remains challenging to generate clinically applicable registration with improved performance and accelerated computation

767

Image Registration for Biomedical Information Integration

for biomedical datasets with larger imaging ranges, higher resolutions, and more dimensionalities.

Concepts and Fundamentals of Biomedical Image Registration Definition Image registration is to compare or combine multiple imaging datasets captured from different devices, imaging sessions, or viewpoints for the purpose of change detection or information integration. The major task of registration is to search for an appropriate transformation to spatially relate and simultaneously represent the images in a common coordinate system for further analysis and visualization. Image registration can be mathematically expressed as (Brown, 1992): I R (X R ) = g(I S (T (X S )))

(1)

where IR and IS are the reference (fixed) image and study (moving) image respectively; T:(XS)→(XR) is the transformation which sets up spatial correspondence between the images so that the study image XS can be mapped to and represented in the coordinate system of reference image XR; g:(IS)→(IR) is one-dimensional intensity calibration transformation. Figure 1. Framework of Image Registration

768

Framework As illustrated in Figure 1, in registration framework, the study dataset is firstly compared to the reference dataset according to a pre-defined similarity measure. If the convergence has not been achieved yet, the optimization algorithm estimates a new set of transformation parameters to calculate a better spatial match between the images. The study image is interpolated and transformed with the updated transformation parameters, and then compared with the reference image again. This procedure is iterated until the optimum transformation parameters are found, which are then used to register and fuse the study image to the reference image.

MAJOR COMPONENTS OF REGISTRATION Input Datasets The characteristics of input datasets, including modality, quality and dimensionality, determine the choice of similarity measure, transformation, interpolation and optimization strategy, and eventually affect the performance, accuracy, and application of the registration.

Image Registration for Biomedical Information Integration

The input images for registration may originate from identical or different imaging modality, and accordingly, registration can be classified as monomodal registration (such as CT-CT registration, PET-PET registration, MRI-MRI registration) or multimodal registration (such as CT-MRI registration, CT-PET registration, and MRI-PET registration). Monomodal registration is required to detect changes over time due to disease progression or treatment. Multimodal registration is used to correlate and integrate the complementary information to provide a more complete insight into the available data. Comparatively, multimodal registration proves to be more challenging than monomodal, due to the heterogeneity of data sources, differing qualities (including spatial resolution and gray-level resolution) of the images, and insufficient correspondence. Registration algorithms for the input datasets with different dimensionalities are used for other applications and requirements. For instance, two-dimensional image registration is applied to construct image mosaics to provide a whole view of a sequence of partially overlapped images, and is used to generate atlases or templates for a specific group of subjects. Although threedimensional image registration is required for most clinical applications, it is challenging to produce an automatic technique with high computational efficiency for routine clinical usages. Multidimensional registration is demanded to align a series of three-dimensional images acquired from different sessions for applications such as tumor growth monitoring, cancer staging and treatment assessment (Fitzpatrick et al, 2000).

Registration Transformations The major task of registration is to find a transformation to align and correlate the input datasets with differences and deformations, introduced during imaging procedures. These discrepancies occur in the form of variations in the quality, content, and information within the images, therefore posing a

significant obstacle and challenge in the area of image registration. In addition, motions, either voluntary or involuntary, require special attention and effort during the registration procedure (Wang et al., 2007; Fitzpatrick et al., 2000). Rigid-body Transformations include rigid and affine transformations, and are mainly used to correct simple differences or to provide initial estimates for more complex registration procedures. Rigid transformation is used to cope with differences due to positional changes (such as translations and rotations) in the imaging procedure and is often adopted for the registration of brain images, due to the rigid nature of the skull structure. In addition, affine transformation can be used to deal with scaling deformations. However, these transformations are usually limited outside of the brain. Deformable Transformations are used to align the images with more complex deformations and changes. For instance, motions and changes of organs such as lung, heart, liver, bowel, need to be corrected by more comprehensive non-linear transformations. The significant variance between subjects and changes due to disease progression and treatment intervention require nonlinear transformations with more degrees of freedom. In deformable registration, a deformation field which is composed of deformation vectors, needs to be computed. One displacement vector is decided for each individual image element (pixel for 2D or voxel for 3D). Compared to rigid-body transformations, the complexity of deformable transformations will slow down the registration speed, and efficient deformable registration remains a challenging research area.

Interpolation Algorithms Interpolation is an essential component in registration, and is required whenever the image needs to be transformed or there are resolution differences between the datasets to be registered. After the transformation, if the points are mapped

769

Image Registration for Biomedical Information Integration

to non-grid positions, interpolation is performed to approximate the values for these transformed points. For the multimodal image registration, the sample space of the lower-resolution image is often interpolated (up-sampled) to the sample space of the higher-resolution image. In the interpolation procedure, the more neighbouring points are used for the calculation, the better accuracy can be achieved, and the slower the computation. To balance interpolation accuracy and computational complexity, bilinear interpolation technique which calculates the interpolated value based on four points, and trilinear interpolation, are often used in registration (Lehmann et al., 1999).

Optimization Algorithms Registration can be defined as an iterative optimization procedure (Equation 2) for searching the optimal transformation to minimize a cost function for two given datasets: Toptimal = arg min f (T (X S ), X R ) (T )

(2)

where T is the registration transformation; f is the cost function to be minimized. Gradient-based optimization methods are often used in registration, in which the gradient vector at each point is calculated to determine the search direction so that the value of the cost function can be decreased locally. For instance, Quasi-Newton methods such as Broyden-FletcherGoldfarb-Shanno (BFGS), has been investigated and applied in medical image registration (Unser, Aldroubi, 1993; Mattes et al., 2003). Powell optimization (Powell, 1964) is another frequently adopted searching strategy in registration. Powell method performs a succession of one-dimensional searches to find the best solution for each transformation parameter, and then the single-variable optimizations are used to determine the new search direction. This

770

procedure is iterative until no better solution or further improvement over the current solution can be found. Because of no derivatives required for choosing the searching directions, computational cost is reduced in this algorithm. Downhill Simplex optimization (Press et al., 1992) does not require derivatives either. However, compared with Powell algorithm, Downhill Simplex is not so efficient due to more evaluations involved. For a given n-dimensional problem domain, the Simplex method searches the optimum solution downhill through a complex n-dimensional topology through operations of reflection, expansion, contraction, and multiple contractions. Because it is more robust in finding the optimal solution, Simplex optimization has been widely used in medical registration. Multi-resolution optimization schemes have been utilized to avoid being trapped into a local optimum and to reduce computational times of the registration (Thévenaz, Unser, 2000; Borgefors, 1988; Bajcsy, Kovacic, 1989; Unser, Aldroubi, 1993). In multi-resolution registration, the datasets to be registered are firstly composed into multiple resolution levels, and then the registration procedure is carried out from low resolution scales to high resolution scales. The initial registration on the global information in the low resolution provides a good estimation for registration in higher resolution scales and contributes to improved registration performance and more efficient computation (Wang et al, 2007). Spline based multi-resolution registration has been systematically investigated by Thévenaz (Thévenaz, Unser, 2000), and Unser (Unser, Aldroubi, 1993). In these Spline-based registration, the images are filtered by B-spline or cubic spline first and then down-sampled to construct multi-resolution pyramids. Multi-resolution Bspline method provides a faster and more accurate registration result for multimodal images when using mutual information as a similarity measure.

Image Registration for Biomedical Information Integration

BIOMEDICAL IMAGE REGISTRATION METHODOLOGIES AND TECHNIQUES Registration methods seek to optimize values of a cost function or similarity measure which define how well two image sets are registered. The similarity measures can be based on the distances between certain homogeneous features and differences of gray values in the two image sets to be registered (Wang et al, 2007). Accordingly, biomedical image registration can be classified as feature-based or intensity-based methods (Brown, 1992).

Feature-Based Registration In feature-based registration, the transformation required to spatially match the features such as landmark points (Maintz, et al, 1996), lines (Subsol, et al, 1998) or surfaces (Borgefors, 1988), can be determined efficiently. However, in this category of registration, a preprocessing step is usually necessary to extract the features manually or semi-automatically, which makes the registration, operator- intensive and dependent (Wang et al, 2007).

Landmark-Based Registration Landmark-based registration includes identifying homologous points which should represent the same features in different images as the first step, and then the transformation can be estimated based on these corresponding landmarks to register the images. The landmark points can be artificial markers attached to the subject which can be detected easily or anatomical feature points. Extrinsic landmarks (fiducial markers), such as skin markers, can be noninvasive. However, skin markers cannot provide reliable landmarks for registration due to elasticity of human skin. The invasive landmarks such as stereotactic frames are able to provide robust basis for registration,

and can be used in Image Guided Surgery (IGS) where registration efficiency and accuracy are the most important factors. Since they are easily and automatically detectable in multiple images to be registered, extrinsic landmarks can be used in both monomodal and multimodal image registration. Intrinsic landmarks can be anatomically or geometrically (such as corner points, intersection points or local extrema) salient points in the images. Since landmarks are required to be unique and evenly distributed over the image, and to carry substantial information, automatic landmark selection is challenging task. Intensive user interaction is often required to manually identify the feature points for registration (Wang et al, 2007). Iterative closest point (ICP) algorithm (Besl, MaKey 1992) is one of most successful landmark-based registration methods. Because no prior knowledge on correspondence between the features is required, ICP eases the registration procedure greatly.

Line-Based Registration and Surface-Based Registration Line-based registration utilizes line features such as edges and boundaries extracted from images to determine the transformation. “Snakes” or active contours (Kass et al, 1988) provide effective contour extraction techniques and have been widely applied in image segmentation and shape modelling, boundary detection and extraction, motion tracking and analysis, and deformable registration (Wang and Feng 2005). Active contours are energy-minimizing splines, which can detect the closest contour of an object. The shape deformation of an active contour is driven by both internal forces, image forces and external forces. To handle the difficulties of concavity and sensitivity during initialization, classic snakes, balloon model (Cohen and Cohen 1993) and gradient vector flow (GVF) (Xu and Prince 1998) were proposed.

771

Image Registration for Biomedical Information Integration

Surface-based registration uses the distinct surfaces as a registration basis to search the transformation. The “Head-and-Hat” algorithm (Chen, et al, 1987) is a well-known surface fitting technique for registration. In this method, two equivalent surfaces are identified in the images to be registered. The surface extracted from the higher-resolution images, is represented as a stack of discs, and is referred to as “head”, and the surface extracted from the lower-resolution image volume, is referred to as “hat”, which is represented as a list of unconnected 3D points. The registration is determined by iteratively transforming the hat surface with respect to the head surface, until the closest fit of the hat onto the head is found. Because the segmentation task is comparatively easy, and the computational cost is relatively low, this method remains popular. More details of surface-based registration algorithms can be found in the review by Audette et al, 2000.

Intensity-Based Registration Intensity-based registration can directly utilize the image intensity information without segmentation or intensive user interaction required, and thereby can achieve fully automatic registration. In intensity-based registration, a similarity measure is defined on the basis of raw image content and is used as a criterion for optimal registration. Several well-established intensity-based similarity measures have been used in the biomedical image registration domain. Similarity measures based on intensity differences including sum of squared differences (SSD) (Equ. 3) and sum of absolute differences (SAD) (Equ.4) (Brown, 1992), are the simplest similarity criteria which exhibit minimum value for perfect registration. As these methods are too sensitive to the intensity changes and significant intensity differences may lead to false registration, SAD and SSD are limited in application and as such, are mainly used to register monomodal images (Wang et al, 2007).

772

2

N

SSD = ∑ (I R (i) − T (I S (i)))

(3)

i

SAD =

1 N

N

∑I i

R

(i ) − T (I S (i ))

(4)

where IR(i) is the intensity value at position i of reference image R and IS(i) is the corresponding intensity value in study image S; T is geometric transformation. Correlation techniques were proposed for multimodal image registration (Van den Elsen, et al, 1995) on the basis of assumption of linear dependence between the image intensities. However, as this assumption is easily violated by the complexity of images from multiple imaging devices, correlation measures are not always able to find optimal solution for multimodal image registration. Mutual information (MI) was simultaneously and independently introduced by two research groups of Collignon et al (1995) and Viola and Wells (1995) to measure the statistical dependence of two images. Because of no assumption on the feature of this dependence and no limitation on the image content, MI is widely accepted as multimodal registration criterion (Pluim et al., 2003). For two intensity sets R={r} and S={s}, mutual information is defined as: I (R, S ) = ∑ pRS (r , s ) log r ,s

pRS (r , s ) pR (r ) ⋅ pS (s )



(5)

where pRS(r,s) is the joint distribution of the intensity pair (r,s); pR(r) and pS(s) are the marginal distributions of r and s.Mutual information can be calculated by entropy: I (R, S ) = H (R) + H (S ) − H (R, S ) = H (R) − H (R | S ) = H (S ) − H (S | R)

(6)

Image Registration for Biomedical Information Integration

H(S|R) is the conditional entropy which is the amount of uncertainty left in S when R is known, and is defined as:

Figure 2. Registration for Thoracic CT volumes from different subjects

H (S | R) = ∑ ∑ pRS (r , s ) log pS |R (s | r ) r ∈R s ∈S

(7)

If R and S are completely independent, pRS (r , s ) = pR (r ) ⋅ pS (s ) and I(R,S)=0 reaches its minimum; if R and S are identical, I(R,S)=H(R)=H(S) arrives at its maximum. Registration can be achieved by searching the transformation parameters which maximize the mutual information. In implementation, the joint entropy and marginal entropies can be estimated by normalizing the joint and marginal histograms of the overlapped sections of the images. Although maximization of MI is a powerful registration measure, it cannot always generate accurate result. For instance, the changing overlap between the images may lead to false registration with maximum MI (Studholme, et al., 1999).

Case Study: Inter-Subject Registration of Thorax CT Image Volumes Based on Image Intensity The increase in diagnostic information is critical for early detection and treatment of disease and provides better patient management. In the context of a patient with non-small cell lung cancer (NSCLC), registered data may mean the difference between surgery aimed at cure and a palliative approach by the ability to better stage the patient. Further, registration of studies from healthy lung and the lung with tumor or lesion is critical to better tumor detection. In the example (Figure 2), the registration between healthy lung and the lung with tumor is performed based on image intensity, and normalized MI is used as similarity measure. Figure 2 shows that affine registration is not able to align the common structures from different subjects

correctly and therefore deformable registration by using spline (Mattes, et al 2003) is carried out to further improve the registration accuracy.

Hardware Registration Although continued progress in image registration algorithms (software-based registration) has been achieved, the software-based registration might be labor intensive, computationally expensive, and with limited accuracy, and thus is impractical to be applied routinely (Townsend, et al, 2003). Hardware registration, in which the functional imaging device, such as PET is combined with an anatomical imaging device such as CT in the one

773

Image Registration for Biomedical Information Integration

instrument, largely overcomes the current limitations of software-based techniques. The functional and anatomical imaging are performed in the one imaging session on the same imaging table, which minimizes the differences in patient positioning and locations of internal organs between the scans. The mechanical design and calibration procedures ensure that the CT and PET data are inherently accurately registered if the patient does not move. However, patient motion can be encountered between the CT and PET. This not only results in incorrect anatomical localization, but also artifacts from the attenuation correction based on the misaligned CT data (Wang et al, 2007). Misalignment between the PET and CT data can also be due to involuntary motion, such as respiratory or cardiac motion. Therefore, although the combined PET/CT scanners are becoming more and more popular, there is a clear requirement for software-registration to remove the motions and displacements from the images captured by the combined imaging scanners.

APPLICATIONS OF BIOMEDICAL IMAGE REGISTRATION Biomedical image registration is able to integrate relevant and heterogeneous information contained in multiple and multimodal image sets, and is important for clinical database management. For instance, registration is essential to mining the large medical imaging databases for constructing statistical atlas of specific disease to reveal the functional and morphological characteristics and changes of the disease, and to facilitate a more suitable patient care. The dynamic atlas in turn is used as a pattern template for automated segmentation and classification of the disease. Image registration is also critical for medical image retrieval of a specific type of disease in a large clinical image database, and in such a scenario, the functional or anatomical atlas provides prior knowledge and is used as a template for early-

774

stage disease detection and identification (Toga, Thompson, 2001). Registration has a broad range of clinical applications to improve the quality and safety of healthcare. Early detection of tumors or disease offers the valuable opportunity for early intervention to delay or halt the progression of the disease, and eventually to reduce its morbidity and mortality. Biomedical image registration plays an important role in detection of a variety of diseases at an early stage, by combining and fully utilizing complementary information from multimodal images. For instance, dementias are the major causes of disability in the elderly population, while Alzheimer’s disease (AD) is the most common cause of dementia (Nestor, et al, 2004).. Registration of longitudinal anatomical MR studies (Scahill, et al, 2003) allows the identification of probable AD (Nestor, et al, 2004) at an early stage to assist an early, effective treatment. Breast cancer is one of major cause of cancer-related death. Registration of pre- and post-contrast of a MR sequence can effectively distinguish different types of malignant and normal tissues (Rueckert, et al, 1998) to offer a better opportunity to cure the patient with the disease (Wang et al, 2007). Biomedical image registration plays an indispensable role in the management of different diseases. For instance, heart disease is the main cause of death in developed counties (American Heart Association 2006) and cardiac image registration provides a non-invasive method to assist in the diagnosis of heart diseases. For instance, registration of MR and X-ray images is a crucial step in the image guided cardiovascular intervention, as well as in therapy and treatment planning (Rhode, et al, 2005). Multimodal image registration such as CT-MR, CT-PET allows a more accurate definition of the tumor volume during the treatment planning phase (Scarfone et al 2004). These datasets can also be used later to assess responses to therapy and in the evaluation of a suspected tumor recurrence (Wang et al, 2007).

Image Registration for Biomedical Information Integration

FUTURE TRENDS Image registration is an enabling technique for fully utilizing heterogeneous image information. However, the medical arena remains a challenging area due to differences in image acquisition, anatomical and functional changes caused by disease progression and treatment, variances and differences across subjects, and complex deformations and motions of internal organs. It is particularly challenging to seamlessly integrate diverse and complementary image information in an efficient, acceptable and applicable manner for clinical routine. Future research in biomedical image registration would need to continuously focus on improving accuracy, efficiency, and usability of registration. Deformable techniques are in high demand for registering images of internal organs such as liver, lung, and cardiac. However, due to complexity of the registration transformation, this category of registration will continuously hold research attention. Insufficient registration efficiency is a major barrier to clinical applications, and is especially prevalent in the case of whole-body images from advanced imaging devices such as the combined PET/CT scanners. For instance, whole-body volume data may consist of more than 400 slices for each modality from the combined PET/CT machine. It is a computationally expensive task for registering these large data volumes. With rapid advance in medical imaging techniques, greater innovation will be achieved, for instance, it is expected that a combined MRI/PET will be made available in near future, which will help to improve the quality of healthcare significantly, but also pose a new set of challenges for efficiently registering the datasets with higher resolution, higher dimensionality, and wider range of scanning areas. Multi-scale registration has the potential to find a more accurate solution with greater efficiency. Graphics Processing Units (GPUs) may provide a high performance hardware platform

for real-time and accurate registration and fusion for clinical use. With its superior memory bandwidth, massive parallelism, improvement in the programmability, and stream architecture, GPUs are becoming the most powerful computation hardware and are attracting more and more research attention. The improvement in floating point format provides sufficient computational accuracy for applications in medical areas. However, effective use of GPUs in image registration is not a simple issue (Strzodka, et al 2004). Knowledge about its underlying hardware, design, limitations, evolution, as well as its special programming model is required to map the proposed medical image registration to the GPU pipeline and fully utilize its attractive features. Medical image registration, particularly for high-dimensional data, which fully utilizes the outstanding features of graphics hardware to facilitate fast and cost-saving real-time clinical applications, is new and yet to be fully explored.

CONCLUSION Registration of medical images from multiple imaging devices and at multiple imaging times is able to integrate and to facilitate a full utilization of the useful image information, and is essential to clinical diagnosis, treatment planning, monitoring and assessment. Image registration is also important in making the medical images more ready and more useful to improve the quality of healthcare service, and is applicable in a wide array of areas including medical database management, medical image retrieval, telemedicine and e-health. Biomedical image registration has been extensively investigated, and a large number of software-based algorithms have been proposed alongside the developed hardware-based solutions (for instance, the combined PET/CT scanners). Among the comprehensive softwarebased registration, the feature-based techniques are more computationally efficient, but require

775

Image Registration for Biomedical Information Integration

a preprocessing step to extract the features to be used in registration, which make this category of registration user-intensive and user-dependent. The intensity-based scheme provides an automatic solution to registration. However, this type of registration is computationally costly. Particularly, image registration is a data-driven and caseorientated research area. It is challenging to select the most suitable and usable technique for specific requirement and datasets from various imaging scanners. For instance, although maximization of MI has been recognized as one of the most powerful registration methods, it cannot always generate accurate solution. A more general registration is more desirable. The combined imaging devices such as PET/CT provide an expensive hardwarebased solution. However, even this expensive registration method is not able to always provide the accurate registration, and software-based solution is required to fix the mis-registration caused by patient motions between the imaging sessions. The rapid advance in imaging techniques raises more challenges in registration area to generate more accurate and efficient algorithms in a clinically acceptable time frame.

ACKNOWLEDGMENT This work is supported by the ARC and UGC grants.

REFERENCES American Heart Association. (2006). Heart and stroke statistical update. Http://www.american heart.org Audette, M., Ferrie, F., & Peters, T. (2000). An algorithm overview of surface registration techniques for medical imaging. Medical Image Analysis, 4(4), 201–217. doi:10.1016/S13618415(00)00014-1

776

Bajcsy, R., & Kovacic, S. (1989, April). Multiresolution elastic matching. Computer Vision Graphics and Image Processing, 46, 1–21. doi:10.1016/ S0734-189X(89)80014-3 Besl, P. J., & MaKey, N. D. (1992). A method for registration of 3-D shapes. IEEE Trans. PAMI, 14(2), 239-256. Borgefors, G. (1988). Hierarchical Chamfer Matching: A Parametric Edge Matching Algorithm. IEEE Transactions on Pattern Analysis and Machine Intelligence, 10, 849–865. doi:10.1109/34.9107 Brown, L. G. (1992). A survey of image registration techniques. ACM Computing Surveys, 24(4), 325–376. doi:10.1145/146370.146374 Chen, C., Pellizari, C. A., Chen, G. T. Y., Cooper, M. D., & Levin, D. N. (1987). Image analysis of PET data with the aid of CT and MR images. Information Processing in Medical Imaging, 601–611. Cohen, I., & Cohen, I. (1993, November). Finiteelement methods for active contour models and balloons for 2-D and 3-D images. IEEE Pattern Anal. Machine Intelligence, 15, 1131–1147. doi:10.1109/34.244675 Collignon, A., Maes, F., Delaere, D., Vandermeulen, D., Suetens, P., & Marchal, G. (1995). Automated multimodality image registration based on information theory. In Proc. 14th International Conference of Information Processing in Medical Imaging 1995, vol.3, (Bizais, Y., Barillot, C. and Di Paola, R. eds.), Ile Berder, France, pp. 263–274, June 1995. Fitzpatrick, J. M., Hill, D. L. G., & Maurer, C. R. (2000). Handbook of medical imaging, (pp. 375-435). Bellingham, WA: SPIE Press. Kass, M., Witkin, A., & Terzopoulos, D. (1988). Snakes: active contour models. International Journal of Computer Vision, 321–331. doi:10.1007/ BF00133570

Image Registration for Biomedical Information Integration

Lehmann, T. M., Gönner, C., & Spitzer, K. (1999). Survey: Interpolation methods in medical image processing. IEEE Transactions on Medical Imaging, 18(11), 1049–1075. doi:10.1109/42.816070 Maintz, J. B. A., van den Elsen, P. A., & Viergever, M. A. (1996). Evaluation of ridge seeking operators for multimodality medical image registration. IEEE Trans. PAMI, 18(4), 353–365. Maintz, J. B. A., & Viergever, M. A. (1998). A Survey of Medical Image Registration. Medical Image Analysis, 2(1), 1–36. doi:10.1016/S13618415(01)80026-8 Mattes, D., Haynor, D. R., Vesselle, H., Lewellen, T. K., & Eubank, W. (2003). PET-CT image registration in the chest using free-form deformations. IEEE Transactions on Medical Imaging, 23(1), 120–128. doi:10.1109/TMI.2003.809072 Nestor, P. J., Scheltens, P., & Hodges, J. R. (2004, July). Advances in the early detection of Alzheimer’s disease. Nature Reviews. Neuroscience, 5(Supplement), S34–S41. doi:10.1038/nrn1433 Pluim, J. P., Maintz, J. B. A., & Viergever, M. A. (2003, August). Mutual-information-based registration of medical images: a survey. IEEE Transactions on Medical Imaging, 22(8), 986–1004. doi:10.1109/TMI.2003.815867 Powell, M. J. D. (1964). An efficient method for finding the minimum of a function of several variables without calculating derivatives. The Computer Journal, 7, 155–163. doi:10.1093/ comjnl/7.2.155 Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. (1992). Numerical Recipes in C. Cambridge Univ. Press, Cambridge, U.K. Rhode, K. S., Sermesant, M., Brogan, D., Hegde, S., Hipwell, J., Lambiase, P., & Rosenthsal, E., bucknall, C., Qureshi, S. A., Gill, J.S., Razavi, R., & Hill, D. L.G. (2005, November). A system for real-time XMR guided cardiovascular intervention. IEEE Transactions on Medical Imaging, 24(11), 1428–1440. doi:10.1109/TMI.2005.856731

Rueckert, D., Hayes, C., Studholme, C., Summers, P., Leach, M., & Hawkes, D. J. (1998). Non-rigidregistration of breast MR images using mutual information. MICCAI’98 lecture notes in computer science, Cambridge, (pp.1144-1152). Scahill, R. I., Frost, C., Jenkins, R., Whitwell, J. L., Rossor, M. N., & Fox, N. C. (2003, July). A longitudinal study of brain volume changes in normal aging using serial registered magnetic resonance imaging. Archives of Neurology, 60(7), 989–994. doi:10.1001/archneur.60.7.989 Scarfone, C., Lavely, W. C., Cmelak, A. J., Delbeke, D., Martin, W. H., Billheimer, D., & Hallahan, D. E. (2004, Apr). Prospective feasibility trial of radiotherapy target definition for head and neck cancer using 3-dimensional PET and CT imaging. Journal of Nuclear Medicine, 45(4), 543–552. Strzodka, R., Droske, M., & Rumpf, M. (2004). Image registration by a regularized gradient flow a streaming implementation in DX9 graphics hardware. Computing, 73(4), 373–389. doi:10.1007/ s00607-004-0087-x Studholme, C., Hill, D. L. G., & Hawkes, D. J. (1999). An overlap invariant entropy measure of 3D medical image alignment. Pattern Recognition, 32, 71–86. doi:10.1016/S0031-3203(98)00091-0 Subsol, G., Thirion, J. P., & Ayache, N. (1998). A scheme for automatically building three dimensional morphometric anatomical atlases: application to a skull atlas. Medical Image Analysis, 2(1), 37–60. doi:10.1016/S1361-8415(01)80027-X Thévenaz, P., & Unser, M. (2000). Optimization of mutual information for multiresolution registration. IEEE Transactions on Image Processing, 9(12), 2083–2099. doi:10.1109/83.887976 Toga, A. W., & Thompson, P. M. (2001, september). The role of image registration in brain mapping. Image and Vision Computing Journal, 19, 3–24. doi:10.1016/S0262-8856(00)00055-X

777

Image Registration for Biomedical Information Integration

Townsend, D. W., Beyer, T., & Blodgett, T. M. (2003). PET/CT Scanners: A Hardware Approach to Image Fusion. Seminars in Nuclear Medicine, XXXIII(3), 193–204. doi:10.1053/ snuc.2003.127314 Unser, M., & Aldroubi, A. (1993, November). A multiresolution image registration procedure using spline pyramids. Proc. of SPIE 2034, 160170, Wavelet Applications in Signal and Image Processing, ed. Laine, A. F. Van den Elsen, P. A., Maintz, J. B. A., Pol, E.-J. D., & Viergever, M. A. (1995, June). Automatic registration of CT and MR brain images using correlation of geometrical features. IEEE Transactions on Medical Imaging, 14(2), 384–396. doi:10.1109/42.387719 Viola, P. A., & Wells, W. M. (1995, June) Alignment by maximization of mutual information. In Proc. 5th International Conference of Computer Vision, Cambridge, MA, 16-23. Wang, X., Eberl, S., Fulham, M., Som, S., & Feng, D. (2007). Data Registration and Fusion, Chapter 8 in D. Feng (Ed.) Biomedical Information Technology, (pp.187-210), Elsevier Publishing

Wang, X., & Feng, D. (2005). Biomedical Image Registration for Diagnostic Decision Making and Treatment Monitoring. Chapter 9 in R. K. Bali (Ed.) Clinical Knowledge Management: Opportunities and Challenges, (pp.159-181), Idea Group Publishing Xu, C., & Prince, J. L. (1998, March). Snakes, shapes, and gradient vector flow. IEEE Transactions on Image Processing, 7, 359–369. doi:10.1109/83.661186

KEY TERMS AND DEFINITIONS Image Registration: The process to search for an appropriate transformation to spatially align the images in a common coordinate system. Intra-Subject Registration: Registration for the images from same subject/person. Inter-Subject Registration: Registration for the images from different subjects/persons. Monomodal Images: Refers to images acquired from same imaging techniques. Multimodal Images: Refers to images acquired from different imaging.

Wang, X., & Feng, D. (2005). Active Contour Based Efficient Registration for Biomedical Brain Images. Journal of Cerebral Blood Flow and Metabolism, 25(Suppl), S623. doi:10.1038/ sj.jcbfm.9591524.0623 This work was previously published in Data Mining and Medical Knowledge Management: Cases and Applications, edited by Petr Berka, Jan Rauch and Djamel Abdelkader Zighed, pp. 122-136, copyright 2009 by Medical Information Science Publishing (an imprint of IGI Global).

778

779

Chapter 3.10

Cognition Meets Assistive Technology: Insights from Load Theory of Selective Attention Neha Khetrapal University of Bielefeld, Germany

ABSTRACT This chapter deals with the issue of treating disorders with the help of virtual reality (VR) technology. To this end, it highlights the concept of transdiagnostic processes (like cognitive biases and perceptual processes) that need to be targeted for intervention and are at the risk of becoming atypical across disorders. There have been previous theoretical attempts to explain such common processes, but such theoretical exercises have DOI: 10.4018/978-1-60960-561-2.ch310

not been conducted with a rehabilitative focus. Therefore, this chapter urges greater cooperation between researchers and therapists and stresses the intimate links between cognitive and emotional functioning that should be targeted for intervention. This chapter concludes by providing future directions for helping VR to become a popular tool and highlights issues in three different areas: (a) clinical, (b) social and (c) technological. Coordinated research efforts orchestrated in these directions will be beneficial for an understanding of cognitive architecture and rehabilitation alike.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Cognition Meets Assistive Technology

INTRODUCTION

BACKGROUND

This chapter will begin with background on the concept of cognitive rehabilitation and will serve to illustrate a recently successful popular example of it. It will then describe cognitive models that explain cognitive and emotional functioning and how these could give rise to disorders. A major focus of the chapter is to highlight the manner in which various disorders could be treated in a similar manner and how technology could aid this process—bringing in the concept of transdiagnostic approach—the basic tenet of which is to emphasize that the processes that serve to maintain disorders cut across these disorders and hence could be dealt with a single appropriately built technology. Though there has not been much research in this direction because therapists prefer to specialize in particular treatment approaches and disorders, this kind of work has picked up momentum (due to the recent scientific focus on an interdisciplinary framework). This chapter will make an initial attempt in this direction by describing how cognitive theories could be applied in understanding the transdiagnostic processes like attentional biases and perceptual processing. This chapter will also attempt to describe the merger of cognitive architecture, specially the transdiagnostic processes and recent rehabilitative tools. Since there remains much work to be done in this direction, this chapter will highlight the areas that require much needed research attention, and at the same time, will provide future directions for embarking upon this process. This chapter will provide an important resource for understanding the transdiagnostic process in terms of assistive technology to psychologists, cognitive scientists, teachers, parents, students of psychology, neuroscientists and rehabilitation professionals.

Cognitive Rehabilitation

780

The aim of rehabilitation is to maintain an optimal level of functioning in domains like physical, social and psychological (McLellan, 1991). Therefore, a rehabilitation program is designed for a particular individual and is conducted over a period of time based on the nature of impairment. The basic goal is not to enhance performance on a set of standardized cognitive tasks, but instead to improve functioning in the day-to-day context (Wilson, 1997). Models of cognitive rehabilitation stress the need to address cognitive and emotional difficulties in an integrated manner and not as isolated domains (Prigatano, 1999). Therefore, cognitive training could be of immense help in this endeavor. Cognitive tasks could thus be designed to deal with cognitive functions like memory, attention, language, and so on, and the level of difficulty could also be varied to suit individual specification (Clare & Woods, 2004).

Techniques for Cognitive Rehabilitation An exciting new development in the field of cognitive rehabilitation is the use of virtual reality (VR). Virtual environments (VE) could be built by keeping in mind the needs of the individual. Examples include presenting a specific number of stimuli to an autistic child that can be gradually increased as the treatment progresses (Max & Burke, 1997) or virtual wheelchair training for people afflicted by physical disabilities (Stephenson, 1995). Schultheis and Rizzo (2001) define VR for behavioral sciences as, “an advanced form of human-computer interface that allows the user to interact with and become immersed in a computergenerated environment in a naturalistic fashion.” Virtual reality could also be viewed as an excellent example of assistive technology (AT) because it can be used to build upon the existing strengths

Cognition Meets Assistive Technology

of an individual who in turn helps in offsetting the disability or, in other words, provides an alternate way of completing a task which also helps in compensating for the disability (Lewis, 1998). Virtual reality technology has yielded promising results in terms of cognitive functioning (Rose, Attree, & Johnson, 1996), social benefits (Hirose, Taniguchi, Nakagaki, & Nihei, 1994), and has proved to be less expensive than the real-world simulators. The previous discussions show that VR as AT could be fruitfully employed to treat disabilities. But it is also important to take into account the functioning of human cognitive systems while designing the VR/VE or any other AT and rehabilitation program. So far in the scientific literature there have been discussions about both cognitive psychology and AT, but each one has remained isolated from the other. Due to the significant contributions from both fields, it becomes essential that both are discussed in relation to each other so that each of these fields could be utilized to maximize the benefits that the other can confer. To begin with, one needs to adopt a good working definition of deficient cognitive components that require rehabilitative attention and which also cut across various disabilities. An important concept in this regard is the concept of “transdiagnostic approach.” According to this approach, behavioral and cognitive processes that serve to maintain disorders are transdiagnostic in nature (Mansell, Harvey, Watkins, & Shafran, 2008). The transdiagnostic approach has many advantages to itself and these include a better understanding about comorbidity—generalization of knowledge derived from cognitive model(s) to explain a particular disorder. Therefore, when the processes are seen as cutting across the disorders, it becomes easier to generalize one explanation to other processes that are similar in nature. The next advantage is the development of treatment approaches. If the processes are assumed to be common, then it becomes easier and even cost-effective to treat various disorders. Stud-

ies in cognitive psychology indeed support the transdiagnostic approach. For instance, attention to external or internal concern related stimuli have been found to be common across various psychological disorders like social phobia, panic disorder, depression, eating disorder, psychotic disorder, posttraumatic stress disorder and so on. Other processes that are also transdiagnostic are memory, thought, reasoning and the like (Mansell et al., 2008). But how exactly are transdiagnostic processes implicated in disorders? How could such processes serve as targets for rehabilitation? The following discussion on cognitive models will make this clearer.

COGNITIVE MODELS FOR EMOTIONAL PROCESSING Any program of cognitive rehabilitation is built upon a comprehensive understanding of the cognitive and behavioral processes/architecture. Cognitive models that explain cognitive functioning and behavior are good candidates on which rehabilitation endeavors could be built. Current scientific theorizing proposes a more intimate link between cognition and emotion than has been proposed before. Therefore, it may be useful as rehabilitation programs are planned to keep both cognitive and emotional processing in mind because of the intimate interaction. Describing all such theories is outside the scope of this paper; however, the following discussion will touch upon some of these theories.

A Cognitive Model for Selective Processing in Anxiety Mathews and Machintosh (1998) proposed a cognitive model to explain selective processing in anxiety. Anxiety is usually the experience of unpleasant feelings of tension and worry in reaction to unacceptable wishes or impulses. A popular finding in anxiety research is attentional

781

Cognition Meets Assistive Technology

bias towards anxiety relevant concerns only under competitive conditions (where the competition is between neutral and anxiety relevant stimuli). The Mathews and Macintosh (1998) model provides a parsimonious explanation for this finding. The model essentially explains that stimuli attributes are processed in a parallel manner and compete for attention due to the involvement of two different routes. A threat evaluation system (TES) is in place that provides help in easing the competition between the two routes by interacting with the level of anxiety and consequently strengthens activations of anxiety relevant attributes. Within their model, a slower route is used when the threat value is appraised by the consciously controlled higher level, such as in a novel anxious situation. Repeated encounters with similar situations will store the relevant cues in the TES and, therefore, a later encounter will produce anxiety through the shorter route bypassing the slower one. As a result, attention will be captured automatically in a competing situation (where the competition is between neutral and anxiety relevant stimuli). Cues encountered in a novel situation that resemble attributes already stored in TES will also tend to elicit anxiety automatically and receive attentional priority. In the model, some danger related attributes are innate while the others are learned. At the same time, other neurobiological evidence suggests that the threat value of a stimulus is evaluated and determined in two distinct ways (LeDoux, 1995). One is a shorter and quicker route that directly runs from the thalamus to the amygdale and the other is a slower route, which is mediated by higher cortical level resources. The proposal of two routes within the model proposed by Mathews and Machintosh (1998), implies that a threatening cue that matches the current concern of anxiety disordered people will sufficiently attract attention under competing situations due to the involvement of the shorter route and will counteract the functioning of the higher level; but following treatment, these patients would no longer show the attentional effects mediated by

782

the shorter route. This is because in the later case, the higher level counteracts the processing of the faster route. Their model also implies that when people encounter threatening cues (no neutral cue), the most threatening one will capture attention and the least threatening one will be inhibited. The model is also applicable in an evolutionary framework because it is certainly more adaptive to process the most potent source of danger and its consequence as a result of mutual inhibition within the TES. The model proposed by Mathews and Machintosh (1998) deals exclusively with selective processing in anxiety and even though it has implications for treatment, it does not give a detailed account of the how treatment will counteract the effects of anxiety. Their model entails a link between cognitive and emotional processing, but a different model proposed by Power and Dalgleish (1997) serves to document a closer relationship between cognitive and emotional processing. The model proposed by Power and Dalgleish (1997) deals with explaining the processing of five basic emotions, such as sadness, happiness, anger, fear, disgust, as well as complex emotions. Since this model is more comprehensive in nature it will be detailed here rather than other models designed to explain processing in specific disorders.

The SPAARS Approach SPAARS (Schematic, Propositional, Analogical and Associative Representational Systems) is the integrated cognitive model of emotion proposed by Power and Dalgleish (1997). It is a multilevel model. The initial processing of stimuli occurs through specific sensory systems that are collectively termed the analogical processing system. This system can play a crucial role in emotional disorders, for instance, where certain sights, smells, noise, etc. become inherent parts of a traumatic event. The output from this system then feeds into three semantic representation systems. These systems operate in parallel. At

Cognition Meets Assistive Technology

the lowest level is the associative system which takes the form of a number of modularized connectionist networks. The intermediate level has the propositional system that has language like representation though it is not language specific. There is no direct route from intermediate level to emotions, but they feed either through appraisals at the schematic level or directly through the associative system. The highest level is called the Schematic Model level. It has the merit of storing information in a flexible manner along with the traditional schema approach. At this level, the generation of emotion occurs through the appraisal process. Appraisal refers to the evaluation of meaning of affective stimuli and is considered causal in the generation of an emotional response. Different types of appraisals exist for eliciting the five basic emotions of sadness, happiness, anger, fear and disgust. An appraisal for sadness would focus on the loss (actual or possible) of some valued goal to an individual and pathological instances, for sadness appraisal could be termed as depression. An individual will feel happy when he or she successfully moves towards the completion of some valued goal. When an appraisal of physical or social threat to self or some valued goal is done by the person, then he or she will experience fear and, when such an appraisal is done for a harmless object, it could result in instances of phobia or anxiety. The appraisal of blocking or frustration of a role or goal through an agent leads to feelings of anger. A person will feel disgust when he or she appraises elimination from a person, object or idea, repulsive to the self or some valued goal (Power & Dalgleish, 1997). These appraisals provide the starting point for complex emotions or a sequence of emotions. In this scheme, complex emotions as well as the disorders of emotions are derived from the basic emotions. A second important feature of emotional disorders is that these may be derived from a coupling of two or more basic emotions or appraisal cycles that further embroider on the

existing appraisals through which basic emotions are generated or through the integration of appraisals which include the goals of others. Examples include the coupling of happiness and sadness, which can generate nostalgia. Indignation can result from the appraisal of anger combined with the further appraisal that the object of anger is an individual who is inferior in the social hierarchy. Empathy results from sadness when combined with the loss of another person’s goal. The model acknowledges the need for two routes for the generation of emotions and this need is based in part on the fact that basic emotions have an innate pre-wired component; additionally, certain emotions may come to be elicited automatically. These two routes are not completely separable. Thus genetics provides a starting psychological point, though the subsequent developmental pathways may be different for each individual. An additional way in which emotions might come to be generated through the direct route is from repeated pairings of certain event-emotion sequences that eventually leads to the automatization of the whole sequence. This repetition bypasses the need for any appraisal. An example of the involvement of the direct route in an emotional disorder, which is an instance of phobia or anxiety where the automatic processing of the objects is anxiety provoking even though it is non-threatening—but due to previous encounter with the object in an individual’s past always in an anxiety provoking situation—it comes to be associated with anxiety. The two routes, for example, can also sometimes generate conflicting emotions as in when the individual may appraise a situation in a happy way, while the direct route is generating a different emotion. The therapeutic technique for working with emotional disorders varies depending on which route the emotion is involved in the disorder. For instance, the person can be provided with a new schematic model for the appraisal of events. Once this model has been accepted the recovery is faster.

783

Cognition Meets Assistive Technology

This type of therapy will work in situations where the schematic level is involved in the disorder. This is an example of fast change processes occurring in therapy. But the patient may continue to experience the maladaptive emotions through the activation of the direct route that is slow to change and is an example of slow processes in recovery. In such cases, exposure-based techniques (as used in the case of phobias) can be helpful. There may also be cases in which a combination of the two techniques will be most effective. Therapies that try to focus on the propositional level of representation only may not be successful if the higher schematic models are flawed. The description of the SPAARS approach shows various similarities with the model proposed by Mathews and Machintosh (1998). Both the models posit two different routes for emotion generation. The models are parsimonious since they advance the same explanation for both normal and disordered cognition, though the SPAARS approach has a broader scope and also gives more detailed specification for treatment choices.

ASSISTIVE TECHNOLOGY AND HUMAN COGNITION Current rehabilitative tools and assistive technologies could be significantly improved by considering the architecture of human cognition during the design. The principles derived from cognitive models described, if incorporated into AT tools will help the tools to serve effectively with the target population. Before embarking on the process of merging the nature of human cognition with assistive tools, I will discuss another popular theory of how selective attention operates—load theory of selective attention, which could be utilized to explain the nature of emotional and cognitive functioning in both order and disorder.

784

Load Theory of Selective Attention Goal-directed behavior requires focusing attention on goal relevant stimuli. The load theory of selective attention proposes two mechanisms for selective control of attention (Lavie, Hirst, Fockert, & Viding, 2004). The first is a perceptual selection mechanism, which is passive in nature and ensures the exclusion of distractors from perception under high perceptual load (Lavie, 1995). The distractors are not perceived under high perceptual load as the target absorbs all the available processing capacity. But under conditions of low perceptual load, spare processing capacity left over from processing task relevant stimuli “spills-over” to irrelevant stimuli that are processed accordingly (Lavie & Tsal, 1994). Loading perception would require either adding more items to the task at hand or a more perceptually demanding task on the same number of items. The second mechanism of attentional control is more active in nature and is evoked for the purpose of rejecting distractors that have been perceived under low perceptual load. This type of control depends on higher cognitive functions, like working memory. Therefore, loading higher cognitive functions that maintain processing priorities result in increased distractor processing. The effect occurs because the reduced availability of control mechanisms in turn reduces the ability to control attention according to the processing priorities. Supporting the theory, Lavie and Cox (1997) have shown that an irrelevant distractor failed to capture attention under high perceptual load conditions as compared to low perceptual load. The load was manipulated by either increasing the number of stimuli among which the target had to be detected or by increasing the perceptual similarity between the target and distractors making the task increasingly perceptually demanding in nature. This result was cited as a support for passive distractor rejection in contrast to active inhibitory mechanisms that are employed for

Cognition Meets Assistive Technology

the purpose of rejecting distractors under low perceptual load conditions.

Load Theory and Disorders The two mechanisms of selective attention could be distributed in disorders and hence the load theory of selective attention could serve as an aid while describing the attentional deficits encountered in both cognitive and emotional disorders. A complete description of all the disorders and how these attentional deficits are present in each are out of the scope of this chapter, but for illustrative purpose, a few disorders will be chosen here to further illustrate the transdiagnostic processes. Consistent with the theories previously introduced, Bishop, Jenkins, and Lawrence (2007) showed that anxiety modulated the amygdalar response to fearful distractors that interfered with the task performance only under low perceptual load conditions. But this effect was observed for state anxiety (current anxious state) rather than trait anxiety (a permanent personality feature). Trait anxiety, on the other hand, correlated with reduced activity in brain regions responsible for controlled processing under low perceptual load. This result implies that trait anxiety is associated with poor attentional controls. Therefore, state and trait anxiety potentially produce interactive effects and disturb task performance because of the disturbed passive mechanisms and faulty attentional control which in turn does not prevent irrelevant emotional distractors from capturing attention under conditions of load. Deficient attentional control was also observed for aged participants by Lavie (2000; 2001). Maylor and Lavie (1998) investigated the role of perceptual load in aging. They showed that distractor processing was decreased for older participants at lower perceptual loads as compared to the younger ones. Similarly high level affective evaluation (appraisals that are necessary for emotion generation as described in the SPAARS approach) requires attention and working memory, and as a result,

is disrupted under high cognitive load. Kalisch, Wiech, Critchley, and Dolan (2006) varied cognitive load, while at the same time, anxiety was induced with the help of anticipation of an impending pain. They observed no change in subjective and physiological indices of anxiety expectations under conditions of load. They did obtain reductions in the activity of brain areas responsible for controlled processing under conditions of high load indicating that high level appraisal was suppressed. Their results did not only show dissociation between brain areas responsible for higher and lower level appraisals, but also how these interact with the manipulations of load.

Merging Technology and Cognition Having described the intimate role between cognitive and emotional processing in both order and disorder and their interaction with perceptual and cognitive load, what should be the next step if we need to plan a rehabilitative program considering the aforementioned principles of human cognition? Therapy with VR, as previously described, has shown promising results. For instance, VR has been employed effectively for the treatment of phobias, that are usually described as intense and irrational fears of objects or events, like acrophobia (Emmelkamp, Krijn, Hulsbosch, de Vries, Schuemie, & van der Mast, 2002), fear of flying (Rothbaum, Hodges, Smith, Lee, & Price, 2000), spider phobia (Garcia-Palacios et al., 2002) and social phobia (Roy, Légeron, Klinger, Chemin, Lauer, & Nugues, 2003). Clinicians also consider phobias as part of anxiety disorders. VR as a potential tool for dealing with phobias has several advantages. Because the essential component in the treatment of phobias is exposure to the threat related object (like spiders in case of spider phobia) either in the form of imagery or in vivo (the latter involves graded exposure), VR as a treatment device could be employed effectively. When working with VR/VE, the therapist can control feared situation and graded exposure with a significant

785

Cognition Meets Assistive Technology

degree of safety. VR thus turns out to be more effective than the imagination techniques/sessions, where the patients are simply left to themselves to imagine the feared object. Under the imagination procedure the therapist not only lacks control on the imagination of the patient, but it also becomes hard for the therapist to determine whether the patient is actually following the imagination procedure leading to poor treatment generalization outside the treatment clinics. On the other hand, real exposure to the feared object could lead the patient to be traumatized, making him/her more fearful of it. Consequently VR could be employed fruitfully to overcome the difficulties of both the imagination techniques and real exposure. The other most important advantage that VR confers on the treatment process is the opportunity for interoceptive exposure (Vincelli, Choi, Molinari, Wiederhold, & Rive, 2000). This becomes important given the fact that bodily sensations are interpreted as signs of fear in the presence of feared object. Virtual reality also turns out to be effective when higher level distorted cognitions need to be challenged (Riva, Bacchetta, Baru, Rinaldi, & Molinari, 1999).

FUTURE RESEARCH DIRECTIONS Future research efforts on VR as a successful application for rehabilitation should concentrate on three major issues and associated problems: (a) clinical, (b) social and (c) technological issues.

Clinical Issues The previous discussion clearly shows that VR as a rehabilitative tool has shown promising results and, therefore, has implications for further improvement. Virtual reality could be better suited to rehabilitate a range of disorders by meshing it with the functioning of human cognition. Much remains to be done in order to pinpoint the specific transdiagnostic processes that cuts across disorders

786

and are also found to be deficient. A promising direction in this regard is the application of load theory of selective attention. Though the studies conducted by Bishop et al. (2007) and Kalisch et al. (2006) show that atypical cognitive bias interacts with behavior and neural responses under differing conditions of load, these kinds of results still await to be incorporated into a rehabilitative VR endeavor. As we have previously stated, VR has been used successfully for treating various phobias like acrophobia (Emmelkamp et al., 2002), fear of flying (Rothbaum et al., 2000), spider phobia (Garcia-Palacios et al., 2002) and social phobia (Roy et al, 2003). What the literature lacks currently is an intimate link between cognitive architecture and the basis for VR successes. Cognitive psychologists, rehabilitative therapists, and VR professionals will stand to gain much if more studies are planned in this direction. For instance, VR is a good choice in exposure techniques for phobias, but since the SPAARS framework and the model proposed by Mathews and Machintosh (1998) show that there could be two routes to emotions—and exposure technique is useful when the faster route that runs from thalamus to amygdala is involved—it will be fruitful to plan future VR studies as was done by Bishop et al. (2007). If such studies show improved attentional control under different conditions of load and prevent anxiety from modulating amygdalar response to anxiety relevant distractors (that disrupt task performance under low perceptual load) with VR treatment, then this will strengthen the link between cognitive models and rehabilitation. The prior theorizing also shows that for successful treatment, practitioners need to provide the patient with a new schematic model for the appraisal of events other than exposure techniques. Once this model has been accepted, recovery is faster. This type of therapy will work in situations where the schematic level is involved in the disorder. This is an example of fast change processes occurring in therapy. For the future, VR could be

Cognition Meets Assistive Technology

used in conjunction with brain imaging techniques to study the brain responses along with behavioral responses before and after treatment. Researchers need to meticulously plan such studies by manipulating cognitive load in participants to study the effect of treatment on cognitive appraisals as was done by Kalisch et al. (2006). Once such future endeavors show successful results for anxiety treatment, practitioners will be more confident about the transdiagnostic processes that become atypical and give rise to cognitive biases. How do researchers and practitioners know which route to emotion (faster or slower) is involved in atypical functioning before embarking on such endeavors? This again calls for stronger links with neuropsychology and thorough assessments before chalking out a treatment plan. Finally, if both the routes are involved, then a mixture of techniques can be used. If the decision is to concentrate on both the routes, then it is essential to increase the load on perceptual and cognitive processes parametrically in an orthogonal manner; this is a very important concept because if both kinds of load were to be increased simultaneously, then it would become difficult to discern the effect of each individually. Moreover, since VR allows for interoceptive exposure, which becomes important given the fact that bodily sensations are interpreted as signs of fear in the presence of feared object, it would make sense to study the effect of treatment on the schematic level while the bodily responses are also monitored. If the treatment also shows improvement in bodily responses, then one can be even more confident of the VR intervention.

Social Issues Before VR could become a part of mainstream use, researchers and practitioners need to overcome several social obstacles. In many traditional schools of therapies, a personal relationship between the therapist and the client is given a high degree of importance. For some, VR could be

viewed as disruptive to this relationship. This issue is even more important for a culture that does not emphasize individualism, for instance in some Eastern societies. In this scenario, it becomes important to consider even technologically less developed societies. Apart from this hindrance, any new therapy initially faces resistance from the broader clinical society. This was even true for behavioral therapy when it was introduced, and hence in the field of mental health, there are other issues that determine the acceptance of a new rehabilitative method rather than just documented efficacy. Until the relevant social problems connected to VR are solved, the best course of action might be to adopt VR in conjunction with other traditional modes of rehabilitation.

Technological Issues Research on social and clinical issues is not enough to promote VR; it is also essential to concentrate on the technological aspects of it. Currently, VR devices and protocols lack standardization, while many others are developed for a specific context making generalization poor (Riva, 2005). Though VR systems cost less than real world simulators, VR is expensive considering that many are built for specific purposes only. In addition, VR is very time-consuming to build.

CONCLUSION Given that VR research proceeds along three different directions, the future of VR as a rehabilitative tool is promising. Current research efforts and scientific discussions do focus on VR and human cognition, but these have so far remained isolated from each other. But the advent of cognitive science and multidisciplinary frameworks calls for a better cooperation between the two. First, fruitful research direction in this regard will be to focus on transdiagnostic processes that cut across various disorders and need to be targeted with rehabili-

787

Cognition Meets Assistive Technology

tative efforts. This will bring down the cost of building rehabilitative tools for specific contexts and will also save precious time. Next, once the transdiagnostic processes have been examined, practitioners would apply these as models of human cognition that explain typical and atypical cognition. There have been few theories, and some popular ones have been described, but still a lot work needs to be done to develop them further and make them the basis of cognitive rehabilitation with VR. Once such efforts are in place, we will truly be able to understand comorbidity, generalize knowledge, and bring down the cost of treating various disorders. The day is not far off when mass rehabilitation over the Internet would be possible with such exciting tools!

REFERENCES Bishop, S. J., Jenkins, R., & Lawrence, A. D. (2007). Neural processing of fearful faces: Effects of anxiety are gated by perceptual capacity limitations. Cerebral Cortex, 17(7), 1595–1603. doi:10.1093/cercor/bhl070 Clare, L., & Woods, R. T. (2004). Cognitive training and cognitive rehabilitation for people with early-stage Alzheimer’s disease: A review. Neuropsychological Rehabilitation, 14(4), 385–401. doi:10.1080/09602010443000074 Emmelkamp, P. M., Krijn, M., Hulsbosch, A. M., de Vries, S., Schuemie, M. J., & van der Mast, C. A. (2002). Virtual reality treatment versus exposure in vivo: A comparative evaluation in acrophobia. Behaviour Research and Therapy, 40(5), 509–516. doi:10.1016/S0005-7967(01)00023-7 Garcia-Palacios, A., Hoffman, H., Carlin, A., Furness, T. A., & Botella, C. (2002). Virtual reality in the treatment of spider phobia: A controlled study. Behaviour Research and Therapy, 40(9), 983–993. doi:10.1016/S0005-7967(01)00068-7

788

Hirose, M., Taniguchi, M., Nakagaki, Y., & Nihei, K. (1994). Virtual playground and communication environments for children. IEICE Transactions on Information & Systems. E (Norwalk, Conn.), 77D(12), 1330–1334. Kalisch, R., Wiech, K., Critchley, H. D., & Dolan, R. J. (2006). Levels of appraisal: A medial prefrontal role in high-level appraisal of emotional material. NeuroImage, 30(4), 1458–1466. doi:10.1016/j.neuroimage.2005.11.011 Lavie, N., & Tsal. (1994). Perceptual load as a major determinant of the locus of selection in visual attention. Perception & Psychophysics, 56(2), 183–197. Lavie, N. (1995). Perceptual load as a necessary condition for selective attention. Journal of Experimental Psychology. Human Perception and Performance, 21(3), 451–468. doi:10.1037/00961523.21.3.451 Lavie, N. (2000). Selective attention and cognitive control: Dissociating attentional functions through different types of load. In Monsell, S., & Driver, J. (Eds.), Control of cognitive processes: Attention & performance XVIII (pp. 175–194). Cambridge, MA: MIT Press. Lavie, N. (2001). The role of capacity limits in selective attention: Behavioural evidence and implications for neural activity. In Braun, J., & Koch, C. (Eds.), Visual attention and cortical circuits (pp. 49–68). Cambridge, MA: MIT Press. Lavie, N., & Cox, S. (1997). On the efficiency of visual selective attention: Efficient visual search leads to inefficient distractor rejection. Psychological Science, 8(5), 395–398. doi:10.1111/j.1467-9280.1997.tb00432.x Lavie, N., Hirst, A., Fockert, J. W. D., & Viding, E. (2004). Load theory of selective attention and cognitive control. Journal of Experimental Psychology. General, 133(3), 339–354. doi:10.1037/0096-3445.133.3.339

Cognition Meets Assistive Technology

LeDoux, J. E. (1995). Emotion: clues from the brain. Annual Review of Psychology, 46, 209–235. doi:10.1146/annurev.ps.46.020195.001233

Riva, G. (2005). Virtual reality in psychotherapy [Review]. Cyberpsychology & Behavior, 8(3), 220–230. doi:10.1089/cpb.2005.8.220

Lewis, R. B. (1998).Assistive technology and learning disabilities: Today’s realities and tomorrow’s promises. Journal of Learning Disabilities, 31(1), 16–26, 54. doi:10.1177/002221949803100103

Riva, G., Bacchetta, M., Baru, M., Rinaldi, S., & Molinari, E. (1999). Virtual reality based experiential cognitive treatment of anorexia nervosa. Journal of Behavior Therapy and Experimental Psychiatry, 30(3), 221–230. doi:10.1016/S00057916(99)00018-X

Mansell, W., Harvey, A., Watkins, E. R., & Shafran, R. (2008). Cognitive behavioral processes across psychological disorders: A review of the utility and validity of the transdiagnostic approach. International Journal of Cognitive Therapy, 1(3), 181–191. doi:10.1521/ijct.2008.1.3.181 Mathews, A., & Machintosh, B. (1998). Cognitive model of selective processing in anxiety. Cognitive Therapy and Research, 22(6), 539–560. doi:10.1023/A:1018738019346 Max, M. L., & Burke, J. C. (1997). Virtual reality for autism communication and education, with lessons for medical training simulators. In Morgan, K. S., Hoffman, H. M., Stredney, D., & Weghorst, S. J. (Eds.), Studies in health technologies and informatics, 39. Burke, VA: IOS Press.

Rose, F. D., Attree, E. A., & Johnson, D. A. (1996). Virtual reality: An assistive technology in neurological rehabilitation. Current Opinion in Neurology, 9(6), 461–467. Rothbaum, B. O., Hodges, L., Smith, S., Lee, J. H., & Price, L. (2000). A controlled study of virtual reality exposure therapy for the fear of flying. Journal of Consulting and Clinical Psychology, 68(6), 1020–1026. doi:10.1037/0022-006X.68.6.1020 Roy, S., Légeron, P., Klinger, E., Chemin, I., Lauer, F., & Nugues, P. (2003). Definition of a VR−based protocol for the treatment of social phobia. Cyberpsychology & Behavior, 6(4), 411–420. doi:10.1089/109493103322278808

Maylor, E. A., & Lavie, N. (1998). The influence of perceptual load on age differences in selective attention. Psychology and Aging, 13(4), 563–573. doi:10.1037/0882-7974.13.4.563

Schultheis, M. T., & Rizzo, A. A. (2001). The application of virtual reality technology in rehabilitation. Rehabilitation Psychology, 46(3), 296–311. doi:10.1037/0090-5550.46.3.296

McLellan, D. L. (1991). Functional recovery and the principles of disability medicine. In Swash, M., & Oxbury, J. (Eds.), Clinical Neurology (Vol. 1, pp. 768–790). London: Churchill Livingstone.

Stephenson, J. (1995). Sick kids find help in a cyberspace world. Journal of the American Medical Association, 274(24), 1899–1901. doi:10.1001/ jama.274.24.1899

Power, M., & Dalegleish, T. (1997). Cognition and emotion: From order to disorder. London: The Psychology Press.

Vincelli, F., Choi, Y. H., Molinari, E., Wiederhold, B. K., & Rive, G. (2000). Experiential cognitive therapy for the treatment of panic disorder with agoraphobia: Definition of a clinical protocol. Cyberpsychology & Behavior, 3(3), 375–385. doi:10.1089/10949310050078823

Prigatano, G. P. (1999). Principles of neuropsychological rehabilitation. New York: Oxford University Press.

789

Cognition Meets Assistive Technology

Wilson, B. A. (1997). Cognitive rehabilitation: How it is and how it might be. Journal of the International Neuropsychological Society, 3(5), 487–496.

ADDITIONAL READING Baumgartner, T., Speck, D., Wettstein, D., Masnari, O., Beeli, G., & Jäncke, L. (2008). Feeling present in arousing virtual reality worlds: Prefrontal brain regions differentially orchestrate presence experience in adults and children. Frontiers in Human Neuroscience, 2(8). doi:.doi:10.3389/ neuro.09.008.2008 Buxbaum, L. J., Palermo, M. A., Mastrogiovanni, D., Read, M. S., Rosenberg-Pitonyak, E., Rizzo, A. A., & Coslett, H. B. (2008). Assessment of spatial attention and neglect with a virtual wheelchair navigation task. Journal of Clinical and Experimental Neuropsychology, 30(6), 650–660. doi:10.1080/13803390701625821 Capodieci, S., Pinelli, P., Zara, D., Gamberini, L., & Riva, G. (2001). Music-enhanced immersive virtual reality in the rehabilitation of memory related cognitive processes and functional Abilities: A case report. Presence (Cambridge, Mass.), 10(4), 450–462. doi:10.1162/1054746011470217 Glantz, K., Durlach, N. I., Barnett, R. C., & Aviles, W. A. (1996). Virtual reality (VR) for psychotherapy: From the physical to the social environment. Psychotherapy (Chicago, Ill.), 33(3), 464–473. doi:10.1037/0033-3204.33.3.464 Harvey, A. G., Watkins, E. R., Mansell, W., & Shafran, R. (2004). Cognitive behavioral processes across psychological disorders: A transdiagnostic approach to research and treatment. Oxford, UK: Oxford University Press. Khetrapal, N. (2007a). Antisocial behavior: Potential treatment with biofeedback. Journal of Cognitive Rehabilitation, 25(1), 4–9.

790

Khetrapal, N. (2007b). SPAARS Approach: Integrated cognitive model of emotion of Attention Deficit/Hyperactivity Disorder. Europe’s Journal of Psychology. Khetrapal, N. (in press). SPAARS Approach: Implications for Psychopathy. Poiesis & Praxis: International Journal of Technology Assessment and Ethics of Science. Lavie, N., & Fockert, J. W. D. (2005). The role of working memory in attentional capture. Psychonomic Bulletin & Review, 12(4), 669–674. LeDoux, J. E. (1996). The emotional brain. New York: Simon & Schuster. McGee, J. S., van der Zaag, C., Buckwalter, J. G., Thiebaux, M., Van Rooyen, A., & Neumann, U. (2000). Issues for the Assessment of Visuospatial Skills in Older Adults Using Virtual Environment Technology. Cyberpsychology & Behavior, 3(3), 469–482. doi:10.1089/10949310050078931 Parsons, T. D., & Rizzo, A. A. (2008). Initial validation of a virtual environment for assessment of memory functioning: Virtual reality cognitive performance assessment test. Cyberpsychology & Behavior, 11(1), 17–25. doi:10.1089/ cpb.2007.9934 Renaud, P., Bouchard, S., & Proulx, R. (2002). Behavioral avoidance dynamics in the presence of a virtual spider. Information Technology in Biomedicine. IEEE Transactions, 6(3), 235–243. Riva, G. (1998). From toys to brain: Virtual reality applications in neuroscience. Virtual Reality (Waltham Cross), 3(4), 259–266. doi:10.1007/ BF01408706 Riva, G., Botella, C., Légeron, P., & Optale, G. (Eds.). (2004). Cybertherapy: Internet and virtual reality as assessment and rehabilitation tools for clinical psychology and neuroscience. Amsterdam: IOS Press.

Cognition Meets Assistive Technology

Riva, G., Molinari, E., & Vincelli, F. (2002). Interaction and presence in the clinical relationship: Virtual Reality (VR) as communicative medium between patient and therapist. IEEE Transactions on Information Technology in Biomedicine, 6(3), 1–8. doi:10.1109/TITB.2002.802370

Strickland, D., Marcus, L., Mesibov, G. B., & Hogan, K. (1996). Brief report: Two case studies using virtual reality as a learning tool for autistic children. Journal of Autism and Developmental Disorders, 26(6), 651–660. doi:10.1007/ BF02172354

Riva, G., Wiederhold, B. K., & Molinari, E. (Eds.). (1998). Virtual Environments in Clinical Psychology and Neuroscience. Amsterdam: IOS Press.

Williams, J. M., Watts, F. N., MacLeod, C., & Mathews, A. (1997). Cognitive psychology and emotional disorders (2nd ed.). Chichester, UK: John Wiley & Sons.

Srinivasan, N., Baijal, S., & Khetrapal, N. (in press). Effects of emotions on selective attention and control. In Srinivasan, N., Kar, B. R., & Pandey, J. (Eds.), Advances in cognitive science (Vol. 2). New Delhi: SAGE.

This work was previously published in Handbook of Research on Human Cognition and Assistive Technology: Design, Accessibility and Transdisciplinary Perspectives, edited by Soonhwa Seok, Edward L. Meyen and Boaventura DaCosta, pp. 96-108, copyright 2010 by Medical Information Science Publishing (an imprint of IGI Global).

791

792

Chapter 3.11

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis Mario Tesconi University of Pisa, Italy Enzo Pasquale Scilingo University of Pisa, Italy Pierluigi Barba University of Pisa, Italy Danilo De Rossi University of Pisa, Italy

INTRODUCTION Posture and motion of body segments are the result of a mutual interaction of several physiological systems such as nervous, muscle-skeletal, and sensorial. Patients who suffer from neuromuscular diseases have great difficulties in moving and walking, therefore motion or gait analysis are widely considered matter of investigation by the clinicians for diagnostic purposes. By means of DOI: 10.4018/978-1-60960-561-2.ch311

specific performance tests, it could be possible to identify the severity of a neuromuscular pathology and outline possible rehabilitation planes. The main challenge is to quantify a motion anomaly, rather than to identify it during the test. At first, visual inspection of a video showing motion or walking activity is the simplest mode of examining movement ability in the clinical environment. It allows us to collect qualitative and bidimensional data, but it does not provide neither quantitative information about motion performance modalities (for instance about dynamics and muscle activity),

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

nor about its changes. Moreover, the interpretation of recorded motion pattern is demanded to medical personnel who make a diagnosis on the basis of subjective experience and expertise. A considerable improvement in this analysis is given by a technical contribution to quantitatively analyse body posture and gesture. Advanced technologies allow us to investigate on anatomic segments from biomechanics and kinematics point of view, providing a wide set of quantitative variables to be used in multi-factorial motion analysis. A personal computer enables a real-time 3D reconstruction of motion and digitalizes data for storage and off-line elaboration. For this reason, the clinicians have a detailed description of the patient status and they can choose a specific rehabilitation path and verify the subject progress. In this context, the Gait Analysis has grown up and currently provided a rich library about walking ability. In a typical Motion Analysis Lab (MAL), multiple devices work together during walking performance, focusing on different motion aspects, such as kinematics, dynamics, and muscular activity. The core of the equipment used in the MAL is a stereophotogrammetric system that measures 3D coordinates of reflecting markers placed on specific anatomic “repere” on the body (Capozzo, Della Croce, Leardini, & Chiari, 2005). This instrumentation permits to compute angles, velocities, and accelerations. The MAL is also equipped with force platforms, which reveal force profiles exchanged with ground on landing. The electromyographic unit is deputed to measure muscle contraction activity by using surface electrodes. Pressure distributions can be obtained by means of baropodometric platforms, made up of a matrix of appropriately shaped sensors. A unique software interface helps the operator to simply manage data from the complex equipment.

BACKGROUND Although its remarkable advantages, the Quantitative Gait Analysis require large spaces, appropriate environments, and cumbersome and expensive equipment that limit the use to restricted applications. Moreover, the stereophotogrammetric system requires a pretest calibration and complex procedure which consists in the placement of reflecting markers on the subject body. Electromyography may be obtrusive if needle electrodes are used to investigate deeper muscles or single motor unit activity. Because of this, although it is a widely accepted methodology, Gait Analysis is still quite difficult to be used in clinical contexts. For these motivations, in the last few years technological research accepted the challenge of transferring any support and high level information close to the patient, better on the patient himself, by creating a user friendly patient-device interface. This means that health services can be brought out of their classic places with a remarkable improvement in health service and in health care based on prevention. Patient can be monitorized for longer time, ensuring the same high service quality at lower costs. A profitable combination of expertise and know how from electronic engineering, materials science and textile industry led to develop effective Wearable Health Care Systems (or simply Wearable Systems) (Lymberis & De Rossi, 2004). These latter ones had electronics integrated on traditional clothes, like on a breadboard. In order to allow subjects to freely move without constraints, improvements in materials manufacture, microelectronics, and nanotechnologies allowed us to develop a new generation of wearable sensing systems characterized by multifunctional fabrics (referred to as e-textiles or smart textiles), where electronic components are made up of polymeric materials, weaved or impressed on the fabric itself, without losing their original mechanical and electrical properties.

793

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

The main advantages of this advanced technology consist of: • •

• •

Strict contact of fabric with skin, the natural human body interface Flexibility and good adherence to the body, which helps avoiding motion artefacts affecting the measurement Washability to be reused Lightness and unobtrusivity, so the patient can naturally move in his usual environment

By means of wearable systems, a large set of physiological variables (ECG, blood pressure, cardiac frequency, etc.) can constantly be monitored and quantitative parameters can be computed, from motion sensing, by dedicated software. Important applications related to health care deal with Ergonomics, Telemedicine, Rehabilitation, and any health service (elderly or impaired people).

MATERIALS AND METHODS As described in the introductive section, Gait Analysis is usually performed for measuring kinematic variables of anatomic segments with tracking devices, which have their main disadvantages in being complex, expensive, and not easy to be used in clinical environments. Thanks to the development of wearable systems, a new class of smart Figure 1. Sensorized shoe for marking the gait cycle

794

textiles (De Rossi, Santa, Mazzoldi, 1999) allows us to design and produce special garments using electrically conductive elastomer composites (CEs) (Post, Orth, Russo, & Gershenfeld, 2000). CEs show piezoresistive properties when a stress is applied, and in several applications they can be integrated into fabrics or other flexible substrates acting as strain sensors (Scilingo, Lorussi, Mazzoldi, & De Rossi, 2003). Such a sensors are just applied in sportive medicine for rehabilitation in soccer post-strokes, for athletic gesture analysis, for angles measurement, and for lower limb force analysis. We used a wearable prototype based on CEs sensors for the first time in a clinical environment. It consists of a sensorized fabric bend wrapped around the knee, very light, well adherent and elastic, hence very suitable for the chosen application. A sensing garment has been realized deposing the CEs over a sublayer of cotton Lycra® and building a sensorized shoe devoted to the gait cycle synchronism. The sensorized shoe consists of a pressure sensor realized of the same material of the knee-band (KB), and is placed on the sole near the heel, as shown in Figure 1. The CEs mixture is smeared on the fabric previously covered by an adhesive mask cut by a laser milling machine. The mask is designed according to the shape and the dimension desired for sensors and wires. After smearing the solution, the mask is removed. Then the treated fabric is placed in an oven at a temperature of about 130 centigrade degrees. During this phase, the cross-

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

linking of the solution speeds up, and in about 10 minutes the sensing fabric is ready to be employed. This methodology is used both for sensing areas and connection paths, thus avoiding employing metallic wires to interconnect sensors or link them to the electronic acquisition unit (Lorussi, Rocchia, Scilingo, Tognetti, & De Rossi, 2004). In this way, no rigid constraints or linkages are present and movements are unbounded (Tognetti et al., 2005). Section (a) of Figure 2 shows the mask for the realization of the KB; the bold black track represents four sensors connected in series and covers the knee joint. The thin tracks represent the connection between the sensors and the electronic acquisition system. Being the thin tracks are made of the same piezorestive mixture, they undergo a not negligible (and unknown) change in their resistance when the knee bends. To face this inconvenience, the analog front-end of the electronic unit is designed to compensate the resistance variation of the thin tracks during the deformations of the fabric. The electric scheme is shown in section (b) of Figure 2. While a generator supplies the series of sensors Si with a constant current I, the acquisition system is provided by a high input impedance stage realized by instrumentation amplifiers and represented in section (b) of Figure 2 by the set of voltmeters. The voltages measured by the instrumentation

amplifiers are equal to the voltages which fall on the Si that is related to resistances of the sensors. In this way, the thin tracks perfectly substitute the traditional metallic wires and a sensor consisting of a segment of the bold track between two thin tracks can be smeared in any position to detect the movements of a joint. The KB acquires information on the flexion-extension of the knee from four sensors spread on a leotard and a step-signal from the sensorized shoe (Morris & Paradiso, 2002). Moreover, the prototype consisted of an “on body unit” dedicated to the acquisition of signals from the KB and the Bluetooth transmission to a remote PC as shown in Figure 3.

EXPERIMENTAL PROTOCOL The prototype was used in a clinical application consisting of the gait monitoring on subjects affected by venous ulcers localized in the lower limbs. A test for discriminating between pathologic and normal walking was performed by the acquisition of flexion-extension signals from the knee joint during movement. Next, experimental data were validated in order to assess goodness of the methodology. Nine subjects, five females and four males, volunteered to participate in the study. The “normal” sample was made of four voluntaries, three males and one female, belong-

Figure 2. (a) The mask used for the sensorised KB. (b) The equivalent electric scheme of the KB.

795

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

Figure 3. The overall prototype consisting of a KB, sensorized shoe and an “on body unit.” Front (left side) and back (right side) views

ing to university context with a mean age of 32 years. Each subject was required to fill in a suitable anamnesis questionnaire in order to verify they did not undergo lower-limb injuries, disease, or trauma. The “pathologic” sample consisted of five patients, four females and one male (mean age about 73 years), in-patient in a clinic specialized in vascular diseases, with special competences in cutaneous ulcers treatment (the medical partner in this work). The presence of the pathology under study was certified by consulting clinical folders. After the measurement system was correctly applied, taking care of arranging the sensors upon the knee articulation, each subject was required to walk freely on a level ground. By means of a wireless communication system, based on Bluetooth protocol, data were transferred in real time from an on-body unit to a personal computer, where a software interface let the operator monitorize signals from the sensorized garment, including diagnostics and current settings. In a five-gait cycle observation interval the flexion-extension signals were acquired and the correspondent files stored for off-line elaboration.

796

DATA ANALYSIS In the typical signal from a sensor of KB (continuous curve in Figure 4), resistance increases during flexion and decreases while the articulation extends, according to the piezoresistive effect. Voltage shows the same behaviour because the supply current was fixed for this kind of measurement. The synchronized matching with stepsignal acquired from the sensorized shoe (dashed line) allowed us to detect the “gait cycle,” or the elemental reference interval of the analysis. A single step-signal is an “on-off” signal where the low-level indicates that the foot is in contact with the ground, while the high-level indicates lifting. In order to extract significant variables for discrimination, two different approaches were studied. The first one, based on the evaluation of flexion-extension capability, was referred to as the width excursion parameter E, defined in Equation (1). (| V −V |)  B E = VM −  A +VA    2

(1)

VM represents the maximum voltage width in the maximum flexion instant; VA is the volt-

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

Figure 4. Flexion-extension of the central sensor of KB (continuous curve) and signal coming from the shoe sensor (dashed curve) which marks the gait cycle

Figure 5. Graphical representation of the width excursion

Em =

EC + EC 1

Em = L

2

2

C

EL + EL 1

2

2



(2)



(3)

ECi (i = 1, 2) indicates the width excursion regarding to central sensor i, while ELi represents the width excursion related to lateral sensor i. IR parameters were computed by “nondistortion” method such as shown in Equation (4):

IR = age value when flexion starts; VB indicates the voltage value in the maximum extension instant such as shown in Figure 5. The second approach proposes to analyze gait discontinuity during the observed interval. To do this we referred to the standard deviation over a sample made up of average width excursion (E) values, on the five-gait cycles. This parameter called irregularity (IR) was calculated both for central (IRc) and lateral (IRL) sensors. In order to integrate information coming from each couple of sensors, a standard deviation was calculated with respect to their average value:

2  _   ∑ x − x    (n − 1)

(4)

where x represents the sample arithmetical mean, while n is the dimension of the sample.

RESULTS According to graphic representation of both IRc and IRL parameters over the entire recruited population (see Figure 6), several important issues have to be addressed: •

IR values showed fundamentally to be higher for pathologic subjects than for normal ones, confirming how they reveal an

797

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

Figure 6. Graphical representation of IRc parameter

• •

•

irregular knee flexion-extension during normal walking. A good discrimination between the two populations was obtained. IR values is higher in patients 5 and 1, which have a larger ulcer size, as it is known in clinical folder. Figure 6 reports on the discrimination on the basis of the irregularity of the mean value of central sensors IRc and Figure 7 shows the graphical discrimination of subjects according to the irregularity of the mean value of lateral sensors IRL. The graph IRc reports a normal subject (normal 4) classified as belonging to the pathologic group, while in IRL graph he is rightly recognized as normal. This means that, for our application, lateral sensors of KB resulted more specific than the central ones.

TEST EVALUATION The test validity has to be intended as the capability by the test to discriminate between normal and pathological subjects within the recruited population. Receiver Operative Characteristic (R.O.C.) curves analysis provided its scientific support in

798

this evaluation, mainly to decide the best cut-off value of parameters for discrimination. A R.O.C. curve is a graphic representation of sensitivity vs. false positive rate (FPR) for different cut-off levels of the measurement variable. The best decisional value ensures maximum of sensitivity to the minimum FPR. By this method, optimal cut-off value of irregularity was detected both for central sensors and for lateral ones (see Figure 8) and the test characteristic parameters were computed, in terms of sensibility, specificity and positive predictive value (PPV), as shown in the Table 1. Positive predictive value is directly proportional to disease prevalence in the study population. For a good test a high PPV is more significant if prevalence is high too. In our study, prevalence was high enough (55%), so PPV values of 80% and 100% gave to the test a good validity score, together with high sensibility and specificity. Lateral sensors showed to be equally sensitive but better specific and predictive than central ones. However these results should be verified over a wider population.

CONCLUSION In this article, we reported on the possibility of using a wearable kinesthetic system for moniFigure 7. Graphical representation of IRL parameter

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

Figure 8. Central sensor and Lateral R.O.C. curves

Table 1. Sensitivity, specificity, ppv and vpn value of the sensors central and lateral Central sensors

Lateral Sensors

Sensitivity

80%

Sensitivity

80%

Specificity

75%

Specificity

100%

PPV

80%

PPV

100%

VPN

75%

VPN

80%

toring through gait analysis the improvement of patients suffering from venous ulcers after undergoing surgical operation. As a validation tool, a preliminary pilot test has been realized, which reports a good capability of discriminating health from injured subjects. Furthermore, results demonstrated the possibility of identifying the severity of the disease in the pathologic group by analyzing of walking activity. Commonly in the clinical practice, these patients are monitored by means of obtrusive instrumentation aimed at evaluating how the circumference of a thigh increases over time. We used an alternative approach to investigate the illness progress. As patients suffering from this illness exhibit an irregular gait, we customized a sensing garment in order to analyze the gait discontinuity.

Therefore this prototype could be a means that allows clinicians to monitor patients without causing any discomfort during the recovery process after a surgical operation. Next, developments aim at extending this application to the study of hip motion. Finally, it has been pointed out that the use of these sensorized garments can be considered a valid alternative and comfortable instrumentation applicable in several rehabilitation areas, such as in sport disciplines and in multimedia fields.

REFERENCES Capozzo, A., Della Croce, U., Leardini, A., & Chiari, L. (2005). Human movement analysis using stereophotogrammetry part 1: Theoretical background. Gait & Posture, 21(21), 186–196. De Rossi, D., Lorussi, F., Scilingo, E. P., Carpi, F., Tognetti, A., & Tesconi, M. (2004). Artificial kinesthetic systems for telerehabilitation. In A. Lymberis & D. de Rossi, (Eds.), Wearable ehealth systems for personalised health management - state of the art and future challenges (pp. 106-114). Amsterdam: IOS Press.

799

Wearable Kinesthetic System for Joint Knee Flexion-Extension Monitoring in Gait Analysis

De Rossi, D., Santa, A. D., & Mazzoldi, A. (1999). Dressware: Wearable hardware. Materials Science and Engineering C, 7, 31. doi:10.1016/S09284931(98)00069-1 Lorussi, F., Rocchia, W., Scilingo, E. P., Tognetti, A., & De Rossi, D. (2004, December). Wearable redundant fabric-based sensors arrays for reconstruction of body segment posture. IEEE Sensors Journal, 4(6), 807–818. doi:10.1109/ JSEN.2004.837498 Morris, S. J., & Paradiso, J. A. (2002). Shoeintegrated sensor system for wireless gait analysis and real-time feedback. In Proceedings of the 2nd Joint Meeting of IEEE EMBS and BMES, Houston, TX (pp. 2468-2469). Post, E. R., Orth, M., Russo, P. R., & Gershenfeld, N. (2000). Design and fabrication of textile-based computing. IBM Systems Journal, 39(3-4). Scilingo, E. P., Lorussi, F., Mazzoldi, A., & De Rossi, D. (2003). Sensing fabrics for wearable kinaesthetic-like systems. IEEE Sensors Journal, 3(4), 460–467. doi:10.1109/JSEN.2003.815771

KEY TERMS AND DEFINITIONS Anatomical Repere: A prominent structure or feature of the human body that can be located and described by visual inspection or palpation at the body’s surface; used to define movements and postures. Also known as anatomical landmark. Conductive Elastomer Composites: A rubberlike silicone material in which suspended metal particles conduct electricity. Piezoresistive Effect: The changing electrical resistance of a material due to applied mechanical stress. Quantitative Gait Analysis: Useful in objective documentation of walking ability as well as identifying the underlying causes for walking abnormalities in patients with cerebral palsy, stroke, head injury, and other neuromuscular problems Wearable Systems: Devices that are worn on the body. They have been applied to areas such as behavioral modeling, health monitoring systems, information technologies, and media development.

Tognetti, A., Lorussi, F., Bartalesi, R., Tesconi, M., Quaglini, S., & Zupone, G. (2005, March). Wearable kinesthetic system for capturing and classifying upper limb gesture in post-stroke rehabilitation. Journal of Neuroengineering and Rehabilitation, 2(8).

This work was previously published in Encyclopedia of Healthcare Information Systems, edited by Nilmini Wickramasinghe and Eliezer Geisler, pp. 1390-1397, copyright 2008 by Medical Information Science Publishing (an imprint of IGI Global).

800

801

Chapter 3.12

Wearable Systems for Monitoring Mobility Related Activities:

From Technology to Application for Healthcare Services Wiebren Zijlstra University Medical Center Groningen, The Netherlands Clemens Becker Robert Bosch Gesellschaft für medizinische Forschung, Germany Klaus Pfeiffer Robert Bosch Gesellschaft für medizinische Forschung, Germany

ABSTRACT Monitoring the performance of daily life mobility related activities, such as rising from a chair, standing and walking may be used to support healthcare services. This chapter identifies available wearable motion-sensing technology; its (potential) clinical application for mobility assessment and monitoring; and it addresses the need to assess user perspectives on wearable monitoring systems. Given the basic requirements for apDOI: 10.4018/978-1-60960-561-2.ch312

plication under real-life conditions, this chapter emphasizes methods based on single sensor locations. A number of relevant clinical applications in specific older populations are discussed; i.e. (risk-) assessment, evaluation of changes in functioning, and monitoring as an essential part of exercise-based interventions. Since the application of mobility monitoring as part of existing healthcare services for older populations is rather limited, this chapter ends with issues that need to be addressed to effectively implement techniques for mobility monitoring in healthcare.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Wearable Systems for Monitoring Mobility Related Activities

INTRODUCTION One of the major challenges in health care is the ability to timely initiate interventions that prevent loss of functional abilities and maintain or improve quality of life. The individual capacity for safe locomotion is a major indicator for independent functioning in older people. However, within the growing population of older people, safe and independent mobility can be at risk due to age-related diseases such as osteo-arthritis, Parkinson’s or Alzheimer’s disease, and stroke. In addition, older people may become inactive and develop frailty without overt pathology. The latter increases the incidence and impact of falls which are a major threat for health related quality of life in older people (Skelton and Todd 2004). In the next decades, Europe will face a sharp increase, in both relative as well as absolute terms, in the number of older adults. This development is a result of an increasing number of older adults and an average ageing of the population. In 2008, less than 15% of the Dutch population was aged 65 or older; by 2040 this percentage will have increased and reached its peak at approximately 26% (CBS, 2009). Estimates of ageing in other European countries, such as Germany and Italy, are even higher (Eurostat, 2008). The demographic trend towards an ageing society poses social as well as economic challenges. While the demands on health care services are steadily increasing, the (relative) number of persons to give care and to finance health care decreases. Thus, there is a need to adapt health care services. New technologies may aid in providing solutions. Effective interventions are needed to maintain functioning and prevent the loss of independent mobility in older people. Wearable technology for monitoring the performance of daily life mobility related activities, such as lying, rising from a chair, standing and walking may be used to support interventions, which aim to maintain or restore independent mobility. However, at present the routine-use of movement monitoring for the

802

clinical management of care in older populations is limited. Therefore, this chapter aims to identify the potential relevance of wearable systems for monitoring mobility for exercise-based interventions and healthcare services by addressing: available wearable motion sensing technology and its application in methods for mobility assessment and monitoring; clinical applications of mobility monitoring; user perspectives on mobility monitoring; and present shortcomings that prevent an effective implementation of wearable solutions for mobility monitoring in health care services.

AVAILABLE WEARABLE TECHNOLOGY FOR MONITORING HUMAN MOVEMENTS A general principle underlying studies of human movements is to consider the human body as a set of rigid bodies (e.g. foot, shank, or thigh), interconnected by joints (e.g. ankle, or knee). Human movement analyses thus require measuring the kinematics of one or more body segments, e.g. by a camera-based system for position measurements or by different motion sensors. The resulting kinematic data are input into further analyses. Depending on research aims, measurements may be simple (e.g. head or trunk position to study walking distance and speed), or highly complex (e.g. full-body measurements to study inter-segmental dynamics). Recent developments in the miniaturization of movement sensors and measurement technology have opened the way for wearable motion sensing technology (Bonato 2005) and the ambulatory assessment of mobility related activities (e.g. Aminian & Najafi 2004, Zijlstra & Aminian 2007). The recent advances even allow full-body ambulatory measurements by motion sensors. However, since monitoring techniques need to be applied over long durations and under real-life conditions, a number of feasibility criteria should be taken into account. These criteria encompass technical criteria (e.g.

Wearable Systems for Monitoring Mobility Related Activities

the degree to which energy consumption limits the use of multiple sensors) as well as criteria which relate to user acceptance (e.g. the methods need to be non-obtrusive and easy-to-use). The latter aspects will be addressed in a subsequent section, in this section an a-priori choice will be made which facilitates technical and practical feasibility; namely a focus on single-sensor approaches to monitoring mobility. The next sub-sections give a short evaluation of available sensors, signal analysis aspects, and methods for mobility assessment and monitoring. The latter subsection will primarily address current possibilities and limitations of methods using single sensor locations, such as on trunk or leg segments.

Wearable Movement Sensors Goniometers can be used to directly measure changes in joint angle. Regardless of sensor type, the basic principle consists of attaching the two axes of the goniometer to the proximal and distal segments of a joint and measuring the angle between these two axes. Drawbacks of using goniometers are sensor attachments which may hinder habitual movement, limited accuracy, and vulnerability of the sensors. Accelerometers measure accelerations of body segments. Piezo-electric type accelerometers measure accelerations only. They are mainly used for quantifying activity related accelerations. Other accelerometers, i.e. (piezo)resistive or capacitive type accelerometers, measure accelerations (a) and the effect of gravitation (g). In the absence of movement, they can be used to calculate the inclination of the sensor with respect to the vertical. Thus, by attaching accelerometers on one or more body segments (e.g. trunk, thigh, and shank), the resistive type can be used to detect body postures at rest (e.g. standing, sitting, and lying). During activities, movements of body segments induce inertial acceleration, and a variable gravitational component depending on the change of segment inclination with respect to the vertical axis. Both

acceleration components are superimposed and their separation is necessary for a proper analysis of the movement. Miniature gyroscopes are sensitive to angular velocity. Although their use for analyzing human movements is still rather new, gyroscopes are promising since they allow the direct measurement of segment rotations around joints. Unlike resistive type of accelerometers, there is no influence of gravitation on the signal measured by gyroscopes. The gyroscope can be attached to any part of a body segment: as long as its axis is parallel to the measured axis, the angular rotation is still the same along this segment. Rotation angles can be estimated from angular velocity by simple numeric integration. However, due to integration drift, it is problematic to estimate absolute angles from angular velocity data. Earth magnetic field sensors or magnetometers measure changes in the orientation of a body segment relative to the magnetic North. Earth magnetic field sensors can be used to determine segment orientation around the vertical axis. Hence, they provide information that cannot be determined from accelerometers or by the integration of gyroscope signals. However, a disadvantage of magnetometers is their sensitivity to nearby ferromagnetic materials (e.g. iron) and magnetic fields other than that of the earth magnetic field (e.g. electro-magnetic fields produced by a TV screen). Pressure sensors or foot switches attached to the shoe or foot sole can be used to detect contact of the foot with the ground. These sensors allow the identification of different movement phases during walking (i.e. swing, stance and bipedal stance phases), or they can provide pressure distribution of the foot during stance phases. These techniques allow the measurement of longer periods of walking with many subsequent stride cycles. A barometric pressure sensor technically is not a movement sensor, but its sensitivity to air pressure can be used to estimate movement characteristics. Changes in altitude are reflected

803

Wearable Systems for Monitoring Mobility Related Activities

in a different air pressure, thus changes in the height of a body segment can be estimated from the measured pressure signal. Use of a Global Position System (GPS) offers the possibility to locate (changes in) the position of body segments. However, the system can only be used outside and the accuracy of determining position is limited to ca. 0.3 m. The latter accuracy can be improved by use of so called differential GPS (dGPS). At present, the most relevant movement sensors for use in wearable systems for mobility monitoring are (resistive type) accelerometers, gyroscopes and foot switches. Sometimes combinations of sensors are used (i.e. hybrid sensing) to overcome limitations of specific sensor types and therewith obtain optimal kinematic data of body segments. A well-known example of such a hybrid sensor consists of a combination of three-dimensional accelerometers, gyroscopes and earth-magnetic field sensors.

Analysis of Data from Motion Sensors Unlike conventional camera-based methods for movement analyses, most motion sensors do not provide information about position of body segments, therefore the use of these sensors requires intelligent signal processing and appropriate methods to obtain relevant movement parameters. Advanced signal processing is necessary to extract relevant information from movement sensors. To begin with, appropriate signal processing methods are required in order to estimate body kinematics with an acceptable accuracy. Often the data analysis requires that kinematic data are transformed from a local to an inertial (i.e. gravity- or earth-oriented) reference frame. Accelerometers and gyroscopes measure within a local frame of reference, and the orientation of this local reference frame depends on the orientation of the body segment to which the sensor is attached. To transform the sensor data from a local segment

804

oriented reference frame to an inertial reference frame requires knowledge of sensor orientation in space. This orientation can be estimated using data from different sensor types and adequate signal processing algorithms. For example, it has recently been demonstrated that the use of hybrid sensors, consisting of accelerometers, gyroscopes and magnetometers, and signal processing algorithms using a Kalman filter allows for an accurate sensor-based analysis of movements in an inertial frame of reference (e.g. see Luinge & Veltink 2005, Roetenberg et al. 2005). Another recent development is the possibility to combine hybrid sensors for analyzing movements of body segments and GPS for overall changes in position (e.g. Terrier & Schutz 2005). Since the kinematics of body segments are task dependent, the analysis of time varying properties of segment kinematics requires methods that vary in dependence of the specific movement task (e.g. sit-to-stand, standing, or walking). In addition, the choice for signal processing of sensor data depends on the specific goal of the measurements. Not all analyses of human activities need a complete description of the movements of all body segments. The same activity, for example gait, may be analyzed differently depending on whether overall measures (e.g. duration of walking and estimated walking distance) or specific measures (e.g. gait variability, joint rotations) are required. As will be seen in the next section, data acquisition and signal analyses can be simplified if a-priori knowledge of a specific movement is available.

Methods for Mobility Assessment and Monitoring Sensor-based assessments of mobility related activities can target both quantitative and qualitative aspects of movement performance. To quantify movement performance requires methods to detect specific postures, activities, or events from data measured by motion sensors. The qualitative

Wearable Systems for Monitoring Mobility Related Activities

analysis of mobility related activities requires algorithms, which extract relevant biomechanical parameters of movement performance from sensor data. As will be seen in the next sub-sections, a wide variety of different sensor configurations (i.e. the specific type of sensors and the location of sensors on body segments) has been used to extract quantitative and qualitative parameters of movement performance. The detection of frequency and duration of specific body postures and activities from sensor signals, in the literature often indicated as “Activity Monitoring”, is at the heart of any method which aims to monitor mobility over long durations. The underlying principle for the detection of different postures is that during static conditions (e.g. lying, sitting or standing) the orientation of body segments can be determined from the static component in the measured acceleration signal. Thus, it is possible to deduce whether a person is standing, sitting or lying from the orientation of leg and trunk segments. Using (video-based) observations as a reference, several papers have demonstrated the validity of discerning different postures and activities based on accelerometers on trunk and leg segments (e.g. Bussmann et al. 1995, Veltink et al. 1996, Aminian et al. 1999). A sensitivity and specificity higher than 90% has been reported for the detection of different postures and activities based on multiple sensors. However, it should be noted that available validation studies typically are based on data-sets which have been obtained under conditions which are different from activity patterns in daily life. Whereas the first activity monitoring studies were all based on the use of sensors on multiple body segments (mostly trunk and leg), recent studies have shown possibilities to obtain similar information about frequency and duration of activities based on single sensor approaches. For example, based on the detection of transitions between postures by a hybrid sensor which combines a two-dimensional (2D) accelerometer and a gyroscope on the ventral thorax (Najafi et al. 2003) or based on a 3D ac-

celerometer at the dorsal side of the lower trunk (Dijkstra et al. 2008, Dijkstra et al. 2010) postures and activities can be detected. Fall detection can be considered a special case within activity monitoring. Sensor-based fall detectors are highly relevant as the basis for reliable fall reports and automatic fall alarm systems. The advantage of sensor-based fall alarm systems over existing alarm systems is that in the event of a serious fall, the alarm can be triggered automatically. Thus, services can be initiated immediately even if a person is unable to trigger an alarm manually, or otherwise call for help. The available literature on sensor-based fall detection methods presents different approaches, which most often are based on accelerometers at body locations such as head, trunk, or wrist (e.g. Lindemann et al. 2005, Noury et al. 2006, Karantonis et al. 2006, Zhang et al. 2006, Bourke et al. 2007, Kangas et al. 2008). However, at present, there is little published information about the real-life validity of sensor-based fall detectors. The validity of fall detection methods is determined by their sensitivity to falls (i.e., “Does the method detect all falls?”) and specificity (i.e., “Does the method correctly classify those situation where no falls occurred?”). The available studies often have followed a validation approach in which the sensitivity and specificity of fall detection algorithms are determined based on data measured in healthy subjects who simulate different type of falls. Although this approach is an essential first step, the real validation should come from real-life falls obtained in the target populations for applying a fall detection method. In the last decades a steadily increasing number of papers report the use of sensor-based methods for a qualitative analysis of the performance of mobility related activities such as standing, walking, sit-to-stand movement (e.g. see Zijlstra & Aminian 2007). A considerable part of these papers are based on measurements at a single sensor location. The latter simplifies the procedures for data-acquisition, and is possible when a-priori

805

Wearable Systems for Monitoring Mobility Related Activities

knowledge exists about the characteristics of the movements to be measured. For example, many studies are based on measuring the kinematics of the lower trunk at a position close to the body’s centre of mass, which during static standing conditions is located within the pelvis at the level of the second sacral vertebrae. The subsequent data-analyses and interpretations are then based on the assumption that the measured kinematics are similar to the movement patterns of the body’s centre of mass. A conventional approach to determine postural stability during standing is to analyze how well a subject is able to maintain his body position without changing the base of support. Data measured by a force plate under both feet allows for an analysis of the changes in centre of pressure position under the feet. Often these analyses are based on measures of the amplitude and frequency content of changes in the centre of pressure. During quiet standing conditions, the accelerations of the body’s centre of mass are proportional to the changes in centre of pressure (Winter 1995). Thus, accelerations measured at the height of the body’s center of mass should yield similar information about postural stability as the data measured by a force plate. The latter assumption seems to be confirmed by accelerometry based studies of trunk sway during different standing conditions (e.g. Mayagoitia et al. 2002, Moe-Nilssen & Helbostad 2002). In addition to accelerometry-based approaches, the use of gyroscopes to measure trunk rotations during standing has shown to be sensitive for difference in postural stability (e.g. Allum et al. 2005). Spatio-temporal gait parameters such as walking speed, step length and frequency, duration of swing and stance phases can be determined from sensors on leg or trunk segments. Hausdorff et al. (1995) used foot switches to measure (the variability of) temporal gait parameters. Aminian et al. (2002) used gyroscopes on shank and thigh segments of both legs in combination with a geometrical model of the leg to estimate spatial

806

parameters in addition to temporal gait parameters. Based on knowledge of the mechanical determinants of trunk movements during walking (i.e. inverted pendulum characteristics (see Zijlstra & Hof 1997, Zijlstra & Hof 2003)), Zijlstra (2004) estimated spatio-temporal gait parameters using 3D accelerometry at a lower trunk position close to the body’s centre of mass. Sabatini et al. (2005) used knowledge of specific constraints for human walking (i.e. repeated stance phases where the foot has no forward velocity) to overcome typical limitations of sensors and determine spatio-temporal gait parameters from a 2D hybrid sensor on the foot. When more than one sensor location is used, the basic spatio-temporal gait parameters can be complemented with additional parameters which depend on the exact sensor configurations (e.g. joint kinematics (Dejnabadi et al. 2005, 2006), trunk angles (Zijlstra et al. 2007), or joint dynamics (Zijlstra & Bisseling 2004). Performance of the Sit-to-Stand (STS) transfer requires the ability to maintain balance while producing enough muscle force to raise the body’s centre of mass from a seated to a standing position (e.g. see Schenkman et al. 1999, Ploutz-Snyder et al. 2002, Lindemann et al. 2003). Recent studies demonstrated that hybrid sensors can be used to obtain temporal measures of STS (e.g. Najafi et al. 2002), and estimations of the power to lift the body’s center of mass during the STS movement (Zijlstra et al. 2010). Usually, the latter analyses of muscle power are restricted to a laboratory-based approach using cycle- or rowing-ergometers, or opto-electronic camera systems and force plates. However, based on laboratory validations, the recent study demonstrated that motion sensors can be used to obtain measures of muscle strength and power during the Sit-to-Stand (STS) movement in young and older (70+) subjects. To summarize this section: the present literature indicates the availability of wearable technology and suitable methods for assessment of relevant quantitative and qualitative aspects of mobility related activities. As a general rule,

Wearable Systems for Monitoring Mobility Related Activities

the use of hybrid sensors and complex sensor configurations on the body offer better analytical solutions. However, even complex configurations of multiple hybrid sensors are limited with regard to the analysis of (changes in) position of body segments in relation to the support surface or objects in the environment. Furthermore, practical considerations may necessitate the choice for as few sensors as possible. It should be noted that the wealth of sensor-based approaches to mobility assessment and monitoring has not yet resulted in a standardization of procedures for data-acquisition, data-analysis and the validation of assessment methods.

EMERGING HEALTHCARE SERVICES BASED ON MOBILITY MONITORING The World Health Organization’s International Classification of Functioning (i.e. the ICF-model (see references)) presents a framework for studying the effects of disease and health conditions on human functioning. The model indicates that insight in the underlying mechanisms of functional decline requires insight in the complex relationships between body functions (e.g. joint flexibility, muscle force), activities (e.g. gait & balance capacity), and actual movement behavior in daily life. The ICF model underlines the need of assessing both capacity (“What a person can do under standardized conditions”) and performance (“What a person really does in his or her daily life”). During daily life a person may use specific assistive devices and services (facilitators), or vice versa may encounter specific problems (barriers) in his or her personal situation. Thus, capacity and performance measures yield different information and the ICF model very clearly indicates the need to assess capacity and to monitor the performance of mobility related activities in real-life. The ICF model is widely used as a framework for clinical research, but the model does not

specify how capacity and performance should be measured. At present, the extent to which the ICF model is systematically used as part of the daily work routines of those disciplines which address impaired motor functioning and mobility seems rather limited. One reason for this situation is that tools to systematically address capacity and performance aspects of mobility have not been available for application in clinical practice. The tools which are based on wearable technology, as described in the preceding section, are based on recent developments, and there still is the need to resolve many issues before these new tools may become a standard part of clinical routine. First of all, an undisputed evidence-base is needed for the clinical relevance of monitoring based health-care services, and secondly the monitoring approaches need to be acceptable to the potential users. Thus, it is of paramount importance to not only have technical solutions available, it is also necessary to evaluate the clinical validity of tools for mobility assessment and monitoring, and to assess and incorporate user perspectives in the development of monitoring-based health-care services. The next sub-sections will address some of the issues relating to clinical relevance by focusing on aspects of current clinical practice and the potential contribution of mobility monitoring techniques in new health care services. Examples will be given in relation to different conditions, which may lead to mobility impairments. This section will be followed by a next section, which specifically addresses the user perspective on wearable technology.

Current Clinical Practice When aiming to provide optimal care to a patient, health-care professionals are typically confronted with a number of recurrent issues; e.g. the need to diagnose certain conditions, the need to evaluate outcome of an intervention, and the need to predict the risk of future adverse events. All these issues require decisions, which are based on adequate

807

Wearable Systems for Monitoring Mobility Related Activities

assessment tools. The quality of available tools for clinical assessment varies strongly between different clinical disciplines, and at present the clinical fields, which address (potential) mobility problems in different populations still need to further develop assessment tools of acceptable quality. At present, the clinical assessment of mobility related activities mainly encompasses the use of field tests, observational methods, and self-report instruments. Rarely are objective quantitative methods used as part of the routine clinical assessments. The available clinical instruments that are used can be disease specific or generic, but generally, there are serious limitations to their use. For example, at present, there is little information to give evidence-based recommendations for specific existing field tests for balance and mobility in frail older persons (see the Prevention of Falls Network (www.ProFaNE.eu.org) and see Gates et al. 2008). Although many conventional clinical tests for balance and mobility (such as the Timed Up & Go (TUG), or Berg Balance Scale (BBS)) are easy-to-use, and hence allow for integration in clinical practice, they do have serious disadvantages. Most of the present field tests lack a conceptual fundament for balance assessment (e.g. Huxham et al. 2001); the scores which result from these tests are usually based on observation and simple time measures (e.g. time one one-leg, time required for standing up from a chair walking 3m, turning walking back and sitting down); and at present there is only limited evidence for the clinical validity of these measures for diagnosis, prognosis, or outcome evaluation. Similarly, there are serious limitations to the use of self-report methods (such as diaries, questionnaires, or interviews) for obtaining information about the quantity of mobility related activities in daily life. Recalling physical activity is a complex cognitive task and the available self-report instruments vary in their cognitive demands. Older adults in particular may have memory and recall skill limitations, and generally people tend

808

to overestimate their physical activity levels. In addition, the methods do not give a detailed profile of type, duration, frequency, and intensity of the most common daily physical activities. Although out-of-home activities, such as leisure time and sportive activities can be assessed by questionnaire, it is hard to assess non-exercise activities with thermogenesis (NEATs). In older people, more than 70% of physical activities occur at home and this should be included in an appropriate analysis. Particularly in sedentary frail older people, self-reports are insensitive to (small) changes in patterns of physical activity. Currently, the available figures about the incidence of falls primarily depend on oral reports of the subjects themselves or their proxies. Thus, also the evaluation of fall prevention methods, in terms of a reduction in number of falls, depends on subjective reports. These reports can be biased, for example due to difficulties in defining a fall, not-remembering a fall, or due to the cognitive status of the subject who fell (e.g. see Zecevic et al. 2006, Hauer et al. 2006). In addition, the circumstances of a fall and the fall characteristics are seldom reported in a structured manner (Zecevic 2007). The lack of objective fall data seriously hampers the development and evaluation of effective fall prevention programs.

Potential Contribution of Mobility Monitoring to New Health Care Services A large number of studies have demonstrated the potential clinical relevance of sensor-based mobility assessment and monitoring in older people (e.g. see Zijlstra & Aminian 2007). The shortlist of available sensor-based methods in the previous section clearly indicates that a variety of clinically relevant applications seem possible. However, at present there are no standardized sensor-based clinical tests for the assessment of balance and mobility. Furthermore, a recent systematic review (de Bruin et al. 2008) indicates that the majority

Wearable Systems for Monitoring Mobility Related Activities

of available studies, which have used wearable systems for monitoring mobility related activities in older populations have merely addressed the technical validity of monitoring methods. The number of studies that actually applied monitoring methods over long durations in support of clinical goals was (very) limited. Thus, the review unfortunately demonstrates that there is still the need to develop evidence-based applications of the available monitoring technology. At present, a number of clinically relevant applications of monitoring techniques can be identified. First of all, a generic aspect of exercise-based interventions in different target groups is setting (individual) target levels for physical activity. To control whether target levels are reached (or not) requires appropriate monitoring information. These can be obtained by different monitoring techniques. Second, the quality of care to specific patient groups with mobility problems may be improved based on the systematic use of monitoring techniques. Third, the use of monitoring techniques can be used to identify persons with an increased risk of falling. Fourth, when a high fall risk has been established, reliable fall detection methods may be used to automatically detect a fall and initiate alarm services when necessary. These four examples of clinically relevant application of monitoring techniques will be elaborated in the next subsections.

Mobility Monitoring to Support Interventions Aiming to Stimulate Physical Activity An increasing body of evidence demonstrates the importance of physical activity for general health in older adults (e.g. see position stand by the American College of Sports Medicine (ACSM), 2009). Regular physical activity has proven to be effective in the primary and secondary prevention of several chronic conditions and is linked to a reduction in all-cause mortality (e.g. U.S. Department of Health and Human Services

1996, Warburton et al. 2006). In addition, regular physical activity can enhance musculoskeletal fitness (Warburton et al. 2001, 2001), and cognitive functioning (Kramer et al. 2006, Hillman et al. 2008) which both are major prerequisites for functional autonomy. Physical inactivity, on the other hand, is associated with premature death and specific chronic diseases such as cardiovascular disease, colon cancer, and non-insulin dependent diabetes (Warburton et al. 2006). It is widely recognized that, in western societies, many older adults do not achieve sufficient levels of physical activity according to guidelines by the ACSM and the American Heart Association (cf. Nelson et al. 2007, Haskell et al. 2007). However, most of the present knowledge about daily physical activity patterns has been obtained through selfreport based methods, and there still is a lack of objective data about physical activity in different older populations. Recent data collected on 7 consecutive days by a body-fixed sensor in community-dwelling older adults demonstrate that on average the older persons spent ca. 5 hours in an upright position (i.e. standing and walking), the average cumulative walking time was approximately 1.5 hours. However, a strong inter-individual, and intraindividual day-to-day variability was observed (Nicolai et al. 2010). Monitoring variables such as cumulative walking and standing time can be used as a robust indicator of overall physical activity. This allows for self-management or remote counseling of older adults and specific patients groups as part of exercise-based interventions. Particularly when more specific information about qualitative aspects of movement performance can be monitored it will be possible to initiate remotely supervised exercise-programs (e.g. see Hermens & Vollenbroek-Hutten 2008). An individual personal coaching based on objective monitoring data might be used to optimize levels of performance. In specific patient groups, monitoring might also prove to be useful for the

809

Wearable Systems for Monitoring Mobility Related Activities

early detection of clinical deterioration and the management of chronic diseases.

Mobility Monitoring to Support Interventions in Specific Patient Groups In several ways, the monitoring of quantitative and qualitative aspects of mobility related activities in patients with impaired mobility is likely to result in a better disease management and counseling. Not only can patients be monitored to see whether target levels for physical activity are reached (cf. preceding sub-section), monitoring may also contribute to a better diagnosis, prognosis, and evaluation of treatment effects. Thus, the treatment and guidance can be tuned to the individual needs of a patient. The following examples demonstrate potential clinical applications in specific diseases with a high prevalence, severe mobility problems, and considerable associated medical costs: •

810

Arthrosis: A very common age-associated problem is hip or knee arthrosis. In these conditions, the early course of the disease usually consists of pharmaceutical and/ or exercise-based treatment. In the more advanced stages, hip or knee replacement are the most common treatment options to regain pain relief, functional improvement and normal participation levels. Recent studies demonstrated systematic improvements in gait parameters (van den AkkerScheek et al. 2007), and a poor relationship between self-reported and performancebased measures of physical functioning in patients after total hip arthroplasty (van den Akker-Scheek et al. 2008). Thus, the use of mobility monitoring seems ideally suited to document physical activity patterns and the treatment response of the patients in an objective and valid manner. The monitoring can demonstrate the improvement in walking distances and duration after the application of pain treatment as well as the

•

treatment response to an operative procedure. Although Morlock et al. (2001) used an ambulatory system to investigate the duration and frequency of every day activities in total hip patients, mobility monitoring has, to our knowledge, not yet been used to systematically document treatment effects in patients with arthrosis. Chronic Obstructive Pulmonary Disease (COPD): COPD refers to chronic lung conditions in which the airways become narrowed; i.e. chronic bronchitis and emphysema. This results in a limited air flow to and from the lungs causing a shortness of breath and, thus, a limited capacity for physical activities. The disease usually is progressively getting worse over time. Medication can improve lung function and reduce the risk of pulmonary inflammations and thus enhance exercise tolerance. In addition it has been demonstrated that pulmonary rehabilitation can enhance exercise tolerance, improve symptoms and reduce exacerbations. Increasing exercise tolerance and the quantity of daily life physical activities is of crucial importance for COPD patients, because it has been shown that regular physical activity reduces hospital admissions and mortality (Garcia-Aymerich et al. 2006). A recent monitoring-based study confirmed that physical activity indeed is reduced in COPD patients (Pitta et al 2005). A subsequent study then showed that whereas pulmonary rehabilitation can improve exercise capacity, muscle force, and functional status after 3 months of rehabilitation, the improvements in daily life physical activities were first demonstrated after 6 months (Pitta et al. 2008). The latter finding demonstrates the need to develop effective strategies to enhance daily life physical activities in COPD patients. It can be expected that mobility monitoring could be used

Wearable Systems for Monitoring Mobility Related Activities

•

to support such strategies and thus enhance functioning and life-expectancy in COPD. Parkinson’s Disease (PD): This neurodegenerative disease is common, its prevalence in the population above 65 is approximately 1-2 percent. Due to the increasing percentage of old people as part of the total population, its prevalence is increasing. Disease progression is mainly characterized by increasing postural instability, balance problems, and mobility impairment (cf. Hoehn & Yahr 2001), which also lead to an increased risk for falls and fall related injuries. In addition, PD patients suffer from symptoms such as a slowing of movement (bradykinesia) and tremors. Particularly in the later stages of the disease, additional difficulties may arise like emotional problems and cognitive deficits. The early stages of PD can be treated effectively with medication, but, after some years many patients develop disease related problems that can no longer fully be treated by medication. In these later stages, additional therapeutic strategies, such as deep brain stimulation may be initiated (Fahn, 2010).

During the early course of the disease, mobility monitoring can be used to document the effect of the medication on movement patterns, and to reach an optimal individual dosage of medication. This is highly relevant since the numbers of patient visits with the specialist usually are few and brief. The observed movement performance during these visits is not necessarily representative for movement performance in daily life. Thus, remote monitoring might lead to modifications in pharmacological treatment, which else would not have been possible. Due to the specific patho-physiology of PD, patients may profit from external information, which facilitates their movements. Provided that adequate algorithms and actuators are available, mobility

monitoring in PD may be used as a foundation for initiating feedback-based interventions, or auditory and/or visual cues that facilitate movement performance. Anecdotal evidence and some studies indicate that PD patients do profit from such interventions. Recently, these possible treatment strategies have been addressed in research projects financed by the European Commission (e.g. see www.rescueproject.org, www.sensaction-aal.eu). In later stages of the disease, PD patients may develop motor problems such as hyperkinesias, freezing or off-periods that are intermittent and cannot be documented during the visit. Here, the movement monitoring allows the development of patient profiles to document changes due to the course of the disease or the changes in treatment. In addition, monitoring may be used to detect gait arrhythmia, which indicates an increased risk of falling that should be discussed with the patient and care-givers.

Identification of Persons at an Increased Risk for Falls Regarding fall risk, a decreased gait speed (e.g. Lamb 2009), an increased variability in temporal gait parameters such as stride duration (e.g. Hausdorff 1997, 2001), variability in lower trunk accelerations (Moe-Nilssen & Helbostad 2005) and frontal plane trunk rotations during walking (de Hoon et al. 2003), and the duration and variability of body transfers such as during sit-to-stand (Najafi et al. 2002), have all shown to be sensitive in distinguishing older fallers from non-fallers, or frail from fit elderly. Using wearable sensors, it is possible to monitor such parameters and other parameters, such as instabilities during missteps or tripping, over long duration. However, the predictive validity of these and similar measures still needs to be demonstrated in larger cohorts. Ideally, a monitoring approach would allow to evaluate whether a person is at an increased risk for falling before a fall occurs. Thus, a timely initiation of preventive measures could avoid future

811

Wearable Systems for Monitoring Mobility Related Activities

falls and injuries. Such preventive measures could encompass exercise-based interventions to optimize balance (i.e. ‘capacity’) and daily physical activity (i.e. ‘performance’), counseling strategies, home-adaptations, or compensating interventions (for example the use of assistive devices such as a cane or a walker, or wearing hip protectors).

Monitoring of Mobility and Fall Incidents in High-Risk Groups The epidemiology of falls is threatening. Approximately 30% of persons older than 65 years fall each year, and after the age of 75 fall rates are even higher (e.g. O’Loughlin et al. 1993). In Germany more than 5 millions falls occur each year, with a substantial part of these falls leading to hospital admissions. It has been estimated that the fall-related costs in European countries, Australia and the USA are between 0.85% and 1.5% of the total health care expenditures (Heinrich et al. 2009). Given the huge impact of falls, strategies that effectively reduce the number of falls or the consequences of falls are highly relevant. Thus, the development of reliable monitoring approaches which monitor parameters, which indicate fall-risk (cf. the previous sub-section), physical activity patterns and the occurrence of falls is one of the most urgent needs in health care services to older people. Given its relevance, it is no surprise that in the last two decades an increasing effort is made to develop reliable fall detection methods. At present, a number of different approaches, either based on ambient technology (i.e. smart homes) or body worn sensors (e.g. N´I Scanaill et al. 2006), have been reported (see also in earlier section). However, it is astonishing, how little real-life falls data are available and it must be stressed that currently there are no sensor-based fall detection systems available that have been tested in large scale trials. Therefore, the development of adequate sensors and fall-detection algorithms, and their real-life validation in different older populations continues to be an enormous challenge.

812

The development of automatic fall-alarms needs to encompass an analysis of the fall and its impact as well as the post-fall phase. Sensorbased signals during the falling and impact phase are needed to identify a fall. Depending on sensor type and location, information on body acceleration, jerk, impact height, fall direction, protective mechanisms and body position while hitting the ground can be derived from the data. These can be used to develop algorithms to detect a fall. Figure 1 presents an example of accelerations measured at the lower trunk before, during and after a real-life fall. In this specific example the circumstances of the fall and the nature of the fall itself were well-documented. It should be noted that, given the variety of falls and their pre-conditions, the categorization and recording of different fall types is highly relevant for the development of sensitive fall detection methods in different populations. The movement pattern in the time period following the impact phase reflects the consequences of the fall. Consequently, their interpretation is crucial for the decision whether or not to initiate an automated alarm response. The post-fall situation might be life threatening (e.g. syncope or epilepsy), very serious (fracture), or without serious consequences. From a user perspective it is crucial that a major event is recognized immediately (e.g. calling an emergency) and that a minor event does not result in an overreaction of the social environment.

THE USER PERSPECTIVE In respect to monitoring solutions over longer durations, questions about the users’ acceptance become more and more relevant next to aspects of effectiveness and (social) significance. The successful application of technologies in this field is strongly influenced by factors that relate to the fit between the demands of the technology and the specific capabilities of the user, the obtrusiveness of the monitoring method, and not least by intrinsic

Wearable Systems for Monitoring Mobility Related Activities

Figure 1. An example of acceleration signals in vertical (x-axis), medio-lateral (y-axis) and sagittal (z-axis) direction during a real-life fall with impact at t = 30 sec

and extrinsic motivational factors. Furthermore non-acceptance and non-usage can be regarded partly as a consequence of the failure of designs and operational procedures to respond to the wishes and feelings (of this very heterogeneous group) of older people (Fisk 1998). This seems to be a very important issue because even very simple and already widely used systems like wearable alarm-buttons seem to be poorly used for their intended purpose: the case of emergency. In a recent study it was shown that alarmbuttons were not used in most cases where a fall was followed by long period of lying on the floor (Fleming & Brayne 2008). It was shown that 94% of the person who were living in institutional settings, 78% in the community, and 59% in sheltered accommodations did not use their call alarm to summon help after they felt alone and were unable to get up without help. 97% of the person lying on the floor for over one hour did not use their alarm to summon help. “Barriers to using alarms arose at several crucial stages: not seeing any advantage in having such a system, not developing the habit of wearing the pendant even if the system was installed, and, in the event of a fall, not activating the alarm – either as a conscious decision or as a failed attempt” (Fleming & Brayne 2008). One reason probably could be found in the fact that

35% of the participants were severely cognitive impaired. Another aspect of these results may show the danger in the overemphasis of clinical objectives within service frameworks for tele-care in people’s own homes. Technologies in the home may be viewed as more intrusive (and, therefore, less acceptable) than in other, more institutional contexts (Fisk 1997). For a successful extension of (commercial) monitoring technology and its integration into tele-health systems a better understanding of acceptance and non-acceptance for specific target groups is indispensable. The following briefly outlines three theoretical constructs, which address important key aspects relevant to this question.

The Human Factors Approach The human factors approach refers to “the role of humans in complex systems, the design of equipment and facilities for human use, and the development of environments for comfort and safety” (Salvendy 1997). The aim of this approach is to match the demands of a system to the capabilities of the user; as illustrated in Figure 2, described by Rogers & Fisk 2003 and originally adopted by the Center for Research on Aging and Technology Enhancement (Czaja et al. 2001). “The system

813

Wearable Systems for Monitoring Mobility Related Activities

imposes certain demands on the user as a function of the characteristics of the hardware, software, and instructional support that is provided for it. The operator of the system has certain sensory/ perceptual, cognitive, and psychomotor capabilities. The degree of fit between the demands of the system and the capabilities of the user will determine performance of the system as well as attitudes, acceptance, usage of the system, and self-efficacy beliefs about one’s own capabilities to use that system” (Rogers & Fisk 2003). To develop good fitting technologies, various approaches like exact analysis of the necessary tasks when using the technology, observation of users, manipulation of the different designs, and the development of proper training and instructional support are recommended by the authors.

The Obtrusiveness Concept The obtrusiveness concept in tele-health was defined by (Hensel et al. 2006a) “as a summary evaluation by the user based on characteristics or effects associated with the technology that are

perceived as undesirable and physically and/or psychologically prominent”. This concept (see Figure 3; Hensel et al 2006a) was constructed inductively by 22 categories found in the literature. It adds some further and more specific aspects compared to the more general “Human factors approach”. In a secondary analysis within two studies based on focus groups and interviews with residential care residents 16 of the postulated 22 subcategories could be confirmed (*= confirmed in 1 study, **= in both studies; Courtney, Demiris, & Hensel 2007). Restrictively, it has to be mentioned that only two of the 29 participants of the two studies were using information-based technologies at the time of the interview.

The Expectations-Confirmation Theory From a consumer behavior perspective the expectations-confirmation theory (ECT) adds a further aspect to user satisfaction with including initial expectations of a specific product or service prior to purchase or delivery. This theoretical construct

Figure 2. Matching system demands to user capabilities (see text for further explanation)

814

Wearable Systems for Monitoring Mobility Related Activities

Figure 3. The obtrusiveness concept (see text for further explanation)

posits that the expectations lead, after a period of initial experience, to post-purchase satisfaction. The primary reason that we experience disappointment or delight is because our expectations have not been met (negative disconfirmation) or they have been superseded (positive disconfirmation) and we have noticed and responded to the discrepancy (Figure 4; Oliver 1980; Spreng et al. 1996).

Integration of User Perspectives in the Development of Monitoring Approaches When summarizing the previous three concepts, user satisfaction with wearable activity monitoring systems should be seen in the context of a good fit between the demands of the device and

Figure 4. The expectations-confirmation theory (see text for further explanation)

815

Wearable Systems for Monitoring Mobility Related Activities

the capabilities of the user, and furthermore the meeting or exceeding of the user’s expectations in regard of a favorite balance between the perceived benefit and obtrusive aspects. Considering aspects of user satisfaction is becoming more recognized as an important criterion for the development and application of long term and unsupervised monitoring approaches. However, there still seems to be a lack of pertinent data. In a recent systematic review of studies using wearable systems for mobility monitoring in older people (de Bruin et al. 2008), it was noted that only few studies had actually applied monitoring methods over long durations, and that the number of studies that addressed feasibility and user acceptance aspects was even smaller. The identified feasibility and adherence aspects mainly related to the reliability of various devices in unsupervised measurement settings and the acceptance of devices by the populations under study. The user acceptance seemed to vary with sensor location and attachment method. Our own experience in recent or ongoing studies indicates a reasonable user acceptance of singlesensor monitoring techniques. The adherence was above 90% for short periods (24-48 hours) of monitoring with a sensor on the trunk. However, with repeated and extended measurements the acceptability decreased to 50-60%, and in patients with severe cognitive impairment, or during hot periods, adherence can be considerably lower. These experiences indicate that there is a need for sensor miniaturization and/or the integration of small sensors in clothing. Ultimately, monitoring systems need to be evaluated over longer periods and by taking into account different aspects of user perspectives. To our knowledge, there currently is no comprehensive assessment tool available that evaluates user satisfaction with monitoring technologies. Therefore, we have started to develop a questionnaire with six dimensions and 30 items in total. The five-point level Likert scaled items are phrased with regard to sensor based applications and comprise the following six dimensions: (perceived) benefit, usability, self-concept, privacy &

816

loss of control, quality of life, wearing comfort. The questionnaire is currently under validation in a number of ongoing studies. Further information is available under www.aktiv-in-jedem-alter.de.

CONCLUSION Demographic changes confront most industrialized countries with the challenge of an ageing population together with a decreasing work force. Innovative solutions are needed to maintain the quality of our health care systems. Monitoring based health-care services can potentially play a role in decreasing the workload of health care professionals. For example by increasing the effectiveness of existing work routines, or by supporting effective preventive strategies that avoid the extensive use of regular health care services. Given the huge challenges in the next decades, the latter preventive medicine approach deserves increased attention. The present contribution has shown that basic technical solutions for wearable systems for monitoring mobility are available, and that relevant clinical applications can be identified for different older populations. However, from the preceding sections, it will be clear that there still is a gap to bridge before mobility monitoring approaches can become an accepted part of regular healthcare services. Without claiming to be complete, we think at least the following issues need to be resolved to facilitate the successful application of wearable systems for mobility monitoring as part of healthcare services. First of all, the present evidence-base for the clinical relevance of existing clinical assessment tools for mobility related activities is insufficient, and, at this time, this shortcoming is also true for the new sensor-based assessment tools. Although several studies suggest a high potential for relevant applications in older people, the new sensor-based approaches for mobility assessment and monitoring still need to be applied in appropriately designed clinical studies to unequivocally demonstrate their

Wearable Systems for Monitoring Mobility Related Activities

clinical validity for specific purposes. Evidencebased clinical applications of new monitoring techniques should rely on epidemiological studies with non-selected cohorts as well as in well-defined patient groups. The additional benefit of monitoring should be documented in controlled trials. Secondly, there is little doubt that ongoing and future research in the next years will only further increase the abundance of different sensor-based assessment approaches. This will lead to further improvements and innovations, but it will also increase the need for standardization of instruments and procedures for data-acquisition and data-analysis. In addition, it can be expected that there will be a need to define standards for the clinical validation procedures, in order to be able to compare different approaches. At present, such considerations have hardly received attention. Third, the application of monitoring based health-care services requires the development of monitoring approaches which are acceptable to its potential users. Thus, it is essential to assess and incorporate user perspectives in the development process. It should be noted that these user perspectives must not only comprise the patient perspective, but also the perspectives of health care professionals who are involved in using the system. Both types of users will need to be convinced of the potential benefits of the monitoring techniques. It is likely that the willingness to accept will not only depend on an existing evidence-base for the clinical relevance, but also on the practicability of the systems in the home environment and in the clinical environment.

REFERENCES Allum, J. H. J., & Carpenter, M. G. (2005). A speedy solution for balance and gait analysis: angular velocity measured at the centre of body mass. Current Opinion in Neurology, 18, 15–21. doi:10.1097/00019052-200502000-00005

American College of Sports Medicine, ChodzkoZajko, W.J., Proctor, D.N., Fiatarone Singh, M.A., Minson, C.T., Nigg, C.R., Salem, G.J. & Skinner, J.S. (2009). American College of Sports Medicine position stand. Exercise and physical activity for older adults. Medicine and Science in Sports and Exercise, 41(7), 1510–1530. Aminian, K., & Najafi, B. (2004). Capturing human motion using body-fixed sensors: outdoor measurement and clinical applications. Computer Animation and Virtual Worlds, 15(2), 79–94. doi:10.1002/cav.2 Aminian, K., Najafi, B., Bula, C., Leyvraz, P. F., & Robert, P. (2002). Spatio-temporal Parameters of Gait Measured by an Ambulatory System Using Miniature Gyroscopes. Journal of Biomechanics, 35(5), 689–699. doi:10.1016/ S0021-9290(02)00008-8 Aminian, K., Robert, Ph., Buchser, E., Rutschmann, B., Hayoz, D., & Depairon, M. (1999). Physical activity monitoring based on accelerometry: validation and comparison with video observation. Medical & Biological Engineering & Computing, 37, 304–308. doi:10.1007/BF02513304 Bonato, P. (2005). Advances in wearable technology and applications in physical medicine and rehabilitation. Journal of Neuroengineering and Rehabilitation, 2, 1–4. doi:10.1186/17430003-2-2 Bourke, A. K., O’Brien, J. V., & Lyons, G. M. (2007). Evaluation of a threshold-based triaxial accelerometer fall detection algorithm. Gait & Posture, 26, 194–199. doi:10.1016/j. gaitpost.2006.09.012 Bourke, A. K., van de Ven, P. W. J., Chaya, A. E., Olaighin, G. M., & Nelson, J. (2008). Testing of a long-term fall detection system incorporated into a custom vest for the elderly. Proceedings of 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (pp. 2844-2847).

817

Wearable Systems for Monitoring Mobility Related Activities

Bussmann, J. B. J., Veltink, P. H., Koelma, F., van Lummel, R. C., & Stam, H. J. (1995). Ambulatory monitoring of mobility-related activities: the initial phase of the development of an activity monitor. European Journal of Physical Medicine and Rehabilitation, 5(1), 2–7.

Dijkstra, B., Kamsma, Y. P. T., & Zijlstra, W. (2010). Detection of gait and postures using a miniaturized tri-axial accelerometer-based system: accuracy in community-dwelling older adults. Age and Ageing, 39(2), 259–262. doi:10.1093/ ageing/afp249

Courtney, K. L., Demiris, G., & Hensel, B. K. (2007). Obtrusiveness of information-based assistive technologies as perceived by older adults in residential care facilities: a secondary analysis. Journal of Medical Internet, 32(3), 241–249.

Dijkstra, B., Zijlstra, W., Scherder, E., & Kamsma, Y. (2008). Detection of gait episodes and number of steps in older adults and patients with Parkinson’s disease: accuracy of a pedometer and an accelerometry based method. Age and Ageing, 37(4), 436–441. doi:10.1093/ageing/afn097

Czaja, S. J., Sharit, J., Charness, N., Fisk, A. D., & Rogers, W. A. (2001). The Center for Research and Education on Aging and Technology Enhancement (CREATE): A program to enhance technology for older adults. Gerontechnology (Valkenswaard), 1, 50–59. doi:10.4017/gt.2001.01.01.005.00 de Bruin, E. D., Hartmann, A., Uebelhart, D., Murer, K., & Zijlstra, W. (2008). Wearable systems for monitoring mobility related activities in older people. Clinical Rehabilitation, 22, 878–895. doi:10.1177/0269215508090675 de Hoon, E. W., Allum, J. H., Carpenter, M. G., Salis, C., Bloem, B. R., Conzelmann, M., & Bischoff, H. A. (2003). Quantitative assessment of the stops walking while talking test in the elderly. Archives of Physical Medicine and Rehabilitation, 84(6), 838–842. doi:10.1016/S0003-9993(02)04951-1 Dejnabadi, H., Jolles, B. M., & Aminian, K. (2005). A new approach to accurate measurement of uniaxial joint angles based on a combination of accelerometers and gyroscopes. IEEE Transactions on Bio-Medical Engineering, 52(8), 1478–1484. doi:10.1109/TBME.2005.851475 Dejnabadi, H., Jolles, B. M., Casanova, E., Fua, P., & Aminian, K. (2006). Estimation and Visualization of Sagittal Kinematics of Lower Limbs Orientation using Body-Fixed Sensors. IEEE Transactions on Bio-Medical Engineering, 53, 1385–1393. doi:10.1109/TBME.2006.873678

818

Fahn, S. (2010). Parkinson’s disease: 10 years of progress, 1997-2007. Movement Disorders, 25(Suppl 1), 2–14. doi:10.1002/mds.22796 Fisk, M. J. (1997). Telecare equipment in the home. Issues of intrusiveness and control. Journal of Telemedicine and Telecare, 3(Suppl. 1), 30–32. doi:10.1258/1357633971930274 Fisk, M. J. (1998). Telecare at home: factors influencing technology choices and user acceptance. Journal of Telemedicine and Telecare, 4, 80–83. doi:10.1258/1357633981931993 Fleming, J., & Brayne, C. (2008). Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90. British Medical Journal, 337, a2227. doi:10.1136/bmj.a2227 Garcia-Aymerich, J., Lange, P., Benet, M., Schnohr, P., & Anto, J. M. (2006). Regular physical activity reduces hospital admission and mortality in chronic obstructive pulmonary disease: a population based cohort study. Thorax, 61, 772–778. doi:10.1136/thx.2006.060145 Gates, S., Smith, L. A., Fisher, J. D., & Lamb, S. E. (2008). Systematic review of accuracy of screening instruments for predicting fall risk among independently living older adults. Journal of Rehabilitation Research and Development, 45, 1105–1116. doi:10.1682/JRRD.2008.04.0057

Wearable Systems for Monitoring Mobility Related Activities

Giannakouris, K. (2008). Eurostat - Statistics in focus. Population and social conditions. 72/2008: Luxemburg. Haskell, W. L., Lee, I. M., Pate, R. R., Powell, K. E., Blair, S. N., & Franklin, B. A. (2007). Physical activity and public health: updated recommendation for adults from the American College of Sports Medicine and the American Heart Association. Medicine and Science in Sports and Exercise, 39, 1423–1434. doi:10.1249/mss.0b013e3180616b27 Hauer, K., Lamb, S. E., Jorstad, E. C., Todd, C., & Becker, C. (2006). PROFANE-Group. Systematic review of definitions and methods of measuring falls in randomised controlled fall prevention trials. Age and Ageing, 35, 5–10. doi:10.1093/ ageing/afi218 Hausdorff, J. M., Ashkenazy, Y., Peng, C. K., Ivanov, P. C., Stanley, H. E., & Goldberger, A. L. (2001). When human walking becomes random walking: fractal analysis and modeling of gait rhythm fluctuations. Physica A, 302(1-4), 138–147. doi:10.1016/S0378-4371(01)00460-5 Hausdorff, J. M., Edelberg, H. K., Mitchell, S. L., Goldberger, A. L., & Wei, J. Y. (1997). Increased gait unsteadiness in community-dwelling elderly fallers. Archives of Physical Medicine and Rehabilitation, 78, 278–283. doi:10.1016/S00039993(97)90034-4 Hausdorff, J. M., Ladin, Z., & Wei, J. Y. (1995). Footswitch system for measurement of the temporal parameters of gait. Biometrical Journal. Biometrische Zeitschrift, 28(3), 347–351. Heinrich, S., Rapp, K., Rissmann, U., Becker, C., & König, H. H. (2009). Cost of falls in old age: a systematic review. [Epub ahead of print]. Osteoporosis International, (Nov): 19. Helbostad, J. L., & Moe-Nilssen, R. (2003). The effect of gait speed on lateral balance control during walking in healthy elderly. Gait & Posture, 18(2), 27–36. doi:10.1016/S0966-6362(02)00197-2

Hensel, B. K., Demiris, G., & Courtney, K. L. (2006). Defining obtrusiveness in home telehealth technologies: a conceptual framework. Journal of the American Medical Informatics Association, 13, 428–431. doi:10.1197/jamia.M2026 Hermens, H. J., & Vollenbroek-Hutten, M. M. (2008). Towards remote monitoring and remotely supervised training. Journal of Electromyography and Kinesiology, 18(6), 908–919. doi:10.1016/j. jelekin.2008.10.004 Hillman, C.H., & Erickson, K.I., Kramer, & A.F. (2008). Be smart, exercise your heart: exercise effects on brain and cognition. Nature Reviews. Neuroscience, 9(1), 58–65. doi:10.1038/nrn2298 Hoehn, M. M., & Yahr, M. D. (2001). Parkinsonism: onset, progression, and mortality. 1967. Neurology, 57(10Suppl 3), S11–S26. Huxham, F. E., Goldie, P. A., & Patla, A. E. (2001). Theoretical considerations in balance assessment. The Australian Journal of Physiotherapy, 47, 89–100. International Classification of Functioning. Disability and Health (ICF). (n.d.). World Health Organisation, Geneva. Retrieved from www.who. int/classification/icf. Kangas, M., Konttila, A., Lindgren, P., Winblad, I., & Jämsä, T. (2008). Comparison of low-complexity fall detection algorithms for body attached accelerometers. Gait & Posture, 28(2), 285–291. doi:10.1016/j.gaitpost.2008.01.003 Karantonis, D. M., Narayanan, M. R., Mathie, M., Lovell, N. H., & Celler, B. G. (2006). Implementation of a real-time human movement classifier using a triaxial accelerometer for ambulatory monitoring. IEEE Transactions on Information Technology in Biomedicine, 10(1), 156–167. doi:10.1109/TITB.2005.856864

819

Wearable Systems for Monitoring Mobility Related Activities

Kramer, A. F., Erickson, K. I., & Colcombe, S. J. (2006). Exercise, cognition, and the aging brain. Journal of Applied Physiology, 101(4), 1237– 1242. doi:10.1152/japplphysiol.00500.2006

Moe-Nilssen, R., & Helbostad, J. L. (2002). Trunk accelerometry as a measure of balance control during quiet standing. Gait & Posture, 16(1), 60–68. doi:10.1016/S0966-6362(01)00200-4

Lamb, S. E., McCabe, C., Becker, C., Fried, L. P., & Guralnik, J. M. (2008). The optimal sequence and selection of screening test items to predict fall risk in older disabled women: the Women’s Health and Aging Study. The Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 63(10), 1082–1088.

Morlock, M., Schneider, E., Bluhm, A., Vollmer, M., Bergmann, G., Müller, V., & Honl, M. (2001). Duration and frequency of every day activities in total hip patients. Journal of Biomechanics, 34, 873–881. doi:10.1016/S0021-9290(01)00035-5

Lindeman, U., Claus, H., Stuber, M., Augat, P., Muche, R., Nikolaus, T., & Becker, C. (2003). Measuring power during the sit-to-stand transfer. European Journal of Applied Physiology, 89, 466–470. doi:10.1007/s00421-003-0837-z Lindemann, U., Hock, A., Stuber, M., Keck, W., & Becker, C. (2005). Evaluation of a fall detector based on accelerometers: a pilot study. Medical & Biological Engineering & Computing, 43(2), 548–551. doi:10.1007/BF02351026 Luinge, H. J., & Veltink, P. H. (2005). Measuring orientation of human body segments using miniature gyroscopes and accelerometers. Medical & Biological Engineering & Computing, 43(2), 273–282. doi:10.1007/BF02345966 Mathie, M. J., Coster, A. C., Lovell, N. H., & Celler, B. G. (2004). Accelerometry: providing an integrated, practical method for long-term, ambulatory monitoring of human movement. Physiological Measurement, 25(2), R1–R20. doi:10.1088/0967-3334/25/2/R01 Mayagoitia, R. E., Lötters, J. C., Veltink, P. H., & Hermens, H. (2002). Standing balance evaluation using a triaxial accelerometer. Gait & Posture, 16(1), 55–59. doi:10.1016/S09666362(01)00199-0

820

N’I Scanaill, C., Carew, S., Barralon, P., Noury, N., Lyons, D. & Lyons, G.M. (2006). A Review of Approaches to Mobility Telemonitoring of the Elderly in Their Living Environment. Annals of Biomedical Engineering, 34, 547–563. doi:10.1007/s10439-005-9068-2 Najafi, B., Aminian, K., Loew, F., Blanc, Y., & Robert, P. (2002). Measurement of stand-sit and sit-stand transitions using a miniature gyroscope and its application in fall risk evaluation in the elderly. IEEE Transactions on Bio-Medical Engineering, 49(8), 843–851. doi:10.1109/ TBME.2002.800763 Najafi, B., Aminian, K., Paraschiv-Ionescu, A., Loew, F., Bula, C., & Robert, Ph. (2003). Ambulatory system for human motion analysis using a kinematic sensor: monitoring of daily physical activity in elderly. IEEE Transactions on Bio-Medical Engineering, 50(6), 711–723. doi:10.1109/TBME.2003.812189 Najafi, B., Büla, C., Piot-Ziegler, C., Demierre, M., & Aminian, K. (2003). Relationship between fear of falling and spatio-temporal parameters of gait in elderly persons. From Basic Motor Control to Functional Recovery III, Chapter II: From Posture to Gait, 152-158. Ecole Polytechnique Fédérale Lausanne: MCC2003 Book.

Wearable Systems for Monitoring Mobility Related Activities

Nelson, M. E., Rejeski, W. J., Blair, S. N., Duncan, P. W., Judge, J. O., & King, A. C. (2007). Physical activity and public health in older adults: recommendation from the American College of Sports Medicine and the American Heart Association. Medicine and Science in Sports and Exercise, 39(8), 1435–1445. doi:10.1249/ mss.0b013e3180616aa2 Nicolai, S., Benzinger, P., Skelton, D. A., Aminian, K., Becker, C., & Lindemann, U. (2010). Day-today variability of physical activity of older adults living in the community. Journal of Aging and Physical Activity, 18(1), 75–86. Noury, N., Fleury, A., Rumeau, P., Bourke, A. K., Laighin, G. O., Rialle, V., & Lundy, J. E. (2007) Fall detection principles and methods. Conference Proceedings - IEEE Engineering in Medicine and Biology, 1663-1666. O’Loughlin, J. L., Robitaille, Y., Boivin, J. F., & Suissa, S. (1993). Incidence of and risk factors for falls and injurious falls among the communitydwelling elderly. American Journal of Epidemiology, 137(3), 342–354. Oliver, R. (1980). A cognitive model of the antecedents and consequences of satisfaction decisions. JMR, Journal of Marketing Research, 17, 460–469. doi:10.2307/3150499 Pitta, F., Troosters, T., Probst, V. S., Langer, D., Decramer, M., & Gosselink, R. (2008). Are patients with COPD more active after pulmonary rehabilitation? Chest, 134(2), 273–280. doi:10.1378/ chest.07-2655 Pitta, F., Troosters, T., Spruit, M. A., Probst, V. S., Decramer, M., & Gosselink, R. (2005). Characteristics of physical activities in daily life in chronic obstructive pulmonary disease. American Journal of Respiratory and Critical Care Medicine, 171(9), 972–977. doi:10.1164/rccm.200407-855OC

Ploutz-Snyder, L. L., Manini, T., Ploutz-Snyder, R. J., & Wolf, D. A. (2002). Functionally relevant thresholds of quadriceps femoris strength. Journal of Gerontology: Medical Sciences, 57A, M144–M152. Roetenberg, D., Luinge, H. J., Baten, C. T., & Veltink, P. H. (2005). Compensation of magnetic disturbances improves inertial and magnetic sensing of human body segment orientation. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 13(3), 395–405. doi:10.1109/ TNSRE.2005.847353 Rogers, W. A., & Fisk, A. D. (2003). Technology Design, Usability, and Aging: Human Factors Techniques and considerations. In Charness, N., & Schaie, K. W. (Eds.), Impact of technology on successful aging (pp. 1–14). New York: Springer. Sabatini, A. M., Martelloni, C., Scapellato, S., & Cavallo, F. (2005). Assessment of Walking Features From Foot Inertial Sensing. IEEE Transactions on Bio-Medical Engineering, 52(3), 486–494. doi:10.1109/TBME.2004.840727 Salvendy, G. (1997). Handbook of human factors and ergonomics. New York: Wiley. Schenkman, M., Hughes, M., Samsa, G., & Studenski, S. (1996). The relative importance of strength and balance in chair rise by functionally impaired older individuals. Journal of the American Geriatrics Society, 44(12), 1441–1446. Skelton, D. A., & Todd, C. (2004). What are the main risk factors for falls amongst older people and what are the most effective interventions to prevent these falls? How should interventions to prevent falls be implemented? Health Evidence Network. Denmark: World Health Organisation. Spreng, R. A., MacKenzie, S. B., & Olshavsky, R. W. (1996). A reexamination of the determinants of consumer satisfaction. Journal of Marketing, 60, 15–32. doi:10.2307/1251839

821

Wearable Systems for Monitoring Mobility Related Activities

Terrier, P., & Schutz, Y. (2005). How useful is satellite positioning system (GPS) to track gait parameters? A review. Journal of Neuroengineering and Rehabilitation, 2, 2–28. doi:10.1186/17430003-2-28 U.S. Department of Health and Human Services. (1996). Physical activity and health: a report of the Surgeon General (pp. 146–148). Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion. van den Akker-Scheek, I., Stevens, M., Bulstra, S. K., Groothoff, J. W., van Horn, J. R., & Zijlstra, W. (2007). Recovery of gait after short-stay total hip arthroplasty. Archives of Physical Medicine and Rehabilitation, 88(3), 361–367. doi:10.1016/j. apmr.2006.11.026 van den Akker-Scheek, I., Zijlstra, W., Groothoff, J. W., Bulstra, S. K., & Stevens, M. (2008). Physical functioning before and after total hip arthroplasty: perception and performance. Physical Therapy, 88(6), 712–719. doi:10.2522/ptj.20060301 Van Duin, C. (2009). Bevolkingsprognose 2008–2050: naar 17,5 miljoen inwoners. Centraal Bureau voor de Statistiek 2009. http://www.cbs. nl/NR/rdonlyres/D49C0D9A-7540-42E9-948D794EDD3E8443/0/2010bevolkingsprognose200 92060art.pdf Veltink, P. H., Bussmann, H. B. J., de Vries, W., Martens, W. L. J., & van Lummel, R. C. (1996). Detection of static and dynamic activities using uniaxial accelerometers. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 4(4), 375–385. Warburton, D. E., Gledhill, N., & Quinney, A. (2001). The effects of changes in musculoskeletal fitness on health. Canadian Journal of Applied Physiology, 26, 161–216.

822

Warburton, D. E., Gledhill, N., & Quinney, A. (2001). Musculoskeletal fitness and health. Canadian Journal of Applied Physiology, 26, 217–237. Warburton, D. E., Nicol, C. W., & Bredin, S. S. D. (2006). Health benefits of physical activity: the evidence. Canadian Medical Association Journal, 174, 801–809. doi:10.1503/cmaj.051351 Winter, D. A. (1995). Human balance and posture control during standing and walking. Gait & Posture, 3, 193–214. doi:10.1016/09666362(96)82849-9 Zecevic, A. A., Salmoni, A. W., Speechley, M., & Vandervoort, A. A. (2006). Defining a fall and reasons for falling: Comparisons among the views of seniors, health care providers, and the research literature. The Gerontologist, 46, 367–376. Zhang, T., Wang, J., Liu, P., & Hou, J. (2006). Fall Detection by Embedding an Accelerometer in Cellphone and Using KFD Algorithm. International Journal of Computer Science and Network Security, 6(10), 277–284. Zijlstra, W. (2004). Assessment of spatio-temporal parameters during unconstrained walking. European Journal of Applied Physiology, 92, 39–44. doi:10.1007/s00421-004-1041-5 Zijlstra, W., & Aminian, K. (2007). Mobility assessment in older people, new possibilities and challenges. European Journal of Ageing, 4, 3–12. doi:10.1007/s10433-007-0041-9 Zijlstra, W., & Bisseling, R. (2004). Estimation of Hip Abduction Moment based on Body Fixed Sensors. Clinical Biomechanics (Bristol, Avon), 19(8), 819–827. doi:10.1016/j.clinbiomech.2004.05.005 Zijlstra, W., Bisseling, R. W., Schlumbohm, S., & Baldus, H. (2010). A body-fixed-sensor based analysis of power during sit-to-stand movements. Gait & Posture, 31(2), 272–278. doi:10.1016/j. gaitpost.2009.11.003

Wearable Systems for Monitoring Mobility Related Activities

Zijlstra, W., & Hof, A. L. (1997). Displacement of the pelvis during human walking: experimental data and model predictions. Gait & Posture, 6, 249–262. doi:10.1016/S0966-6362(97)00021-0 Zijlstra, W., & Hof, A. L. (2003). Assessment of spatio-temporal Gait Parameters from Trunk Accelerations during Human Walking. Gait & Posture, 18(2), 1–10. doi:10.1016/S09666362(02)00190-X

ADDITIONAL READING American College of Sports Medicine, ChodzkoZajko, W.J., Proctor, D.N., Fiatarone Singh, M.A., Minson, C.T., Nigg, C.R., Salem, G.J. & Skinner, J.S. (2009). American College of Sports Medicine position stand. Exercise and physical activity for older adults. Medicine and Science in Sports and Exercise, 41(7), 1510–1530.

Hermens, H. J., & Vollenbroek-Hutten, M. M. (2008). Towards remote monitoring and remotely supervised training. Journal of Electromyography and Kinesiology, 18(6), 908–919. doi:10.1016/j. jelekin.2008.10.004 International Classification of Functioning. Disability and Health (ICF).(n.d.). World Health Organisation, Geneva (www.who.int/classification/icf). Luinge, H. J. (2002). Inertial sensing of human movement. PhD thesis. The Netherlands: Twente University Press. Mathie, M. J., Coster, A. C., Lovell, N. H., & Celler, B. G. (2004). Accelerometry: providing an integrated, practical method for long-term, ambulatory monitoring of human movement. Physiological Measurement, 25(2), R1–R20. doi:10.1088/0967-3334/25/2/R01

Aminian, K., & Najafi, B. (2004). Capturing human motion using body-fixed sensors: outdoor measurement and clinical applications. Computer Animation and Virtual Worlds, 15(2), 79–94. doi:10.1002/cav.2

N’I Scanaill, C., Carew, S., Barralon, P., Noury, N., Lyons, D. & Lyons, G.M. (2006). A Review of Approaches to Mobility Telemonitoring of the Elderly in Their Living Environment. Annals of Biomedical Engineering, 34, 547–563. doi:10.1007/s10439-005-9068-2

Bonato, P. (2005). Advances in wearable technology and applications in physical medicine and rehabilitation. Journal of Neuroengineering and Rehabilitation, 2, 1–4. doi:10.1186/17430003-2-2

Noury, N., Fleury, A., Rumeau, P., Bourke, A. K., Laighin, G. O., Rialle, V., & Lundy, J. E. (2007) Fall detection principles and methods. Conference Proceedings - IEEE Engineering in Medicine and Biology, 1663-1666.

de Bruin, E. D., Hartmann, A., Uebelhart, D., Murer, K., & Zijlstra, W. (2008). Wearable systems for monitoring mobility related activities in older people. Clinical Rehabilitation, 22, 878–895. doi:10.1177/0269215508090675

Roetenberg, D. (2006). Inertial and magnetic sensing of human motion. PhD thesis. The Netherlands: University of Twente.

Hausdorff, J. M. (2005). Gait variability: methods, modeling and meaning. Journal of Neuroengineering and Rehabilitation, 2, 19. doi:10.1186/17430003-2-19

Skelton, D. A., & Todd, C. (2004). What are the main risk factors for falls amongst older people and what are the most effective interventions to prevent these falls? How should interventions to prevent falls be implemented? Health Evidence Network. Denmark: World Health Organisation.

823

Wearable Systems for Monitoring Mobility Related Activities

Terrier, P., & Schutz, Y. (2005). How useful is satellite positioning system (GPS) to track gait parameters? A review. Journal of Neuroengineering and Rehabilitation, 2, 2–28. doi:10.1186/17430003-2-28 Winter, D. A. (2009). Biomechanics and Motor Control of Human Movement (4th ed.). New York: John Wiley & Sons. doi:10.1002/9780470549148 Zijlstra, W., & Aminian, K. (2007). Mobility assessment in older people, new possibilities and challenges. European Journal of Ageing, 4, 3–12. doi:10.1007/s10433-007-0041-9

KEY TERMS AND DEFINITIONS Movement: Movement is a change of position, or a rotation, of an object or (parts of) the body. Physical Activity: Physical activity is any body movement produced by skeletal muscles that results in energy expenditure.

Mobility (as used in this contribution): Mobility relates to voluntary movements that involve a change of overall body position, for example rising from a chair, or walking. A person with a mobility impairment may have difficulty with walking, standing, lifting, climbing stairs, carrying, or balancing. Motion Sensor: An instrument that senses and measures movement. Assessment (as used in this contribution): The systematic collection of information about an individual’s motor and/or cognitive functioning. Monitoring (as used in this contribution): Repeated assessments of aspects of an individual’s motor and/or cognitive functioning. Fall: An unexpected event in which a person comes to rest on the ground, floor, or a lower level. fall detection: An automated recognition of the occurance of a fall, for example based on video data or based on motion sensors.

This work was previously published in E-Health, Assistive Technologies and Applications for Assisted Living: Challenges and Solutions, edited by Carsten Röcker and Martina Ziefle, pp. 244-267, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

824

825

Chapter 3.13

RFID in Hospitals and Factors Restricting Adoption Bryan Houliston Auckland University of Technology, New Zealand

ABSTRACT Hospitals are traditionally slow to adopt new information systems (IS). However, health care funders and regulators are demanding greater use of IS as part of the solution to chronic problems with patient safety and access to medical records. One technology offering benefits in these areas is Radio Frequency Identification (RFID). Pilot systems have demonstrated the feasibility of a wide range of hospital applications, but few have been fully implemented. This chapter investigates the factors that have restricted the adoption of RFID technology in hospitals. It draws on related DOI: 10.4018/978-1-60960-561-2.ch313

work on the adoption of IS generally, published case studies of RFID pilots, and interviews with clinicians, IS staff and RFID vendors operating in New Zealand (NZ) hospitals. The chapter concludes with an analysis of the key differences between RFID and other IS, and which RFID applications have the greatest chance of successful implementation in hospitals.

INTRODUCTION In 1989 management guru Peter Drucker described hospitals as prototypical knowledge-based organisations (Drucker, 1989). Considering the variety and volume of information that hospitals,

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

RFID in Hospitals and Factors Restricting Adoption

and other health care organisations, deal with, it is easy to see why they might be expected to be early adopters of IS. Kim and Michelman (1990) identify the components of a typical integrated Hospital Information System (HIS): General accounting and budgeting; Staff payroll; Patient demographic information and medical records; Nursing care plans; Treatment orders; Test results; Surgery and resource schedules; and Databases of clinical information relating to Radiology, Pharmacology, Pathology, and other specialist departments. Yet research suggests that in comparison to other industries, the health care sector invests relatively little in IS. Twelve years on from Drucker’s statement, a British survey of annual IS spending per employee found that the health care sector spent approximately one-third that of the manufacturing sector, one-fifth that of the distribution sector, and one-ninth that of the financial sector (Wallace, 2004). This low level of investment has led various stakeholders to demand greater use of IS in the health care sector. Two key areas in which significant potential benefits have been identified are improving patient safety, and sharing electronic medical records amongst all the health care organisations that may treat a patient. Patient safety is, of course, paramount in health care. ‘First, Do No Harm’ is the fundamental principle of the medical profession. Yet each year medical mistakes take a heavy toll in both human life and health care resources. For example, errors in administering drugs, known as Adverse Drug Events (ADEs), are believed to result in tens of thousands of deaths, many more serious injuries, and to cost the health care sector tens of billions of dollars (Classen, Pestotnik, Evans, Lloyd, & Burke, 1997; Davis et al., 2003; Johnson & Bootman, 1995; Wilson et al., 1995). The US Institute of Medicine (IOM) strongly advocate the use of IS to reduce the incidence of ADEs (Institute of Medicine, 2001). Regulatory agencies, such as the Food and Drug Administration (FDA) and Joint Commission on Accreditation on Healthcare

826

Organisations (JCAHO), have mandated the use of barcode technology in US hospitals to improve identification of medications and patients (Merry & Webster, 2004). The NZ Ministry of Health has recently announced plans to spend NZ$115 million to implement systems such as Computerised Physician Order Entry (CPOE) and Barcoded Medication Administration (BCMA) (Johnston, 2007). The leading cause of ADEs is the prescription of unsuitable drugs (Bates, Cullen, & Laird, 1995; Leape, Bates, & Cullen, 1995). Unsuitable prescriptions result primarily from clinicians lacking ready access to patients’ medical records, and thus being unaware of drug allergies, existing conditions, and current prescriptions. Storing medical records in electronic form, in a centralised database, enables timely access to such information for all clinicians who may treat a patient. In the UK, the government is planning to spend around £6 billion on an Electronic Patient Records (EPR) system for the National Health Service (NHS) as part of the ‘Connecting for Health’ initiative (Wallace, 2004). In NZ, the WAVE (Working to Add Value through E-information) Advisory Board to the Director-General of Health recommended a similar system (WAVE Advisory Board, 2001), which has been included in the country’s Health Information Strategy (Health Information Strategy Steering Committee, 2005). In Australia a non-profit company created by federal and state governments, the National e-Health Transition Authority (NEHTA), invests in IS that supports sharing of EPRs. The same approach has been taken in Canada, with the Health Infoway corporation. In the US, major insurers, such as Medicare, require hospitals to provide details of treatment in electronic form (Jonietz, 2004). One technology that has been gaining attention in the health care sector for its potential to address these issues is RFID. It offers very similar functionality to barcode technology, but with a number of advantages (Schuerenberg, 2007). Most notably, RFID allows multiple labels to be scanned

RFID in Hospitals and Factors Restricting Adoption

simultaneously without requiring a line-of-sight between the scanner and the label. A range of RFID applications, from real-time stocktaking, to tracking patients, staff and equipment to storing patient data, have been successfully trialled by hospitals around the world (Wasserman, 2007). Based on published case studies, relatively few hospitals have fully implemented such systems. At the time of writing, NZ hospitals are just beginning to RFID pilots. This chapter seeks to investigate the factors that have restricted the adoption of RFID technology in hospitals. The next section, RFID and Hospital IS, describes the various RFID applications piloted in hospitals. It then reviews research on the adoption of IS generally by hospitals, identifying a number of key organisational and technological factors. To gain greater insight of these factors in the NZ context, a series of interviews were conducted with clinicians, IS staff and RFID vendors operating within the NZ hospital environment. The section Interviews and Findings briefly describes the interview process and presents the results. The Discussion section considers some implications of this research for RFID pilots in NZ hospitals, and some possible areas for further research.

RFID AND HOSPITAL IS This section describes the various RFID applications piloted in hospitals. It then reviews research on the adoption of IS generally by hospitals, and concludes with the aims of the current research.

RFID Applications for Hospitals Hospitals all around the world have successfully trialled RFID applications. These applications can be broadly organised into four types. Identification applications involve a single reader interrogating a single tag. Data storage applications are those where a single reader reads and writes to a single tag. Location-based applications involve a single

reader reading multiple tags. Tracking applications involve multiple readers reading single or multiple tags. Examples of each application are given below.

Identification Applications Identification of people, objects and locations is an obvious application of RFID technology. Security systems that use RFID proximity cards to control access to restricted areas are perhaps the most common RFID application in use today. Even in developing countries, such as Bangladesh, hospitals are using RFID-enabled staff ID cards (Bacheldor, 2008). Other hospitals have extended the use of staff ID cards to control access to computers and applications (Bacheldor, 2007d). Patient identification appears to be the next most common application, after access control. A survey of US health care organisations found that 29% expected to be using RFID wristbands for patient identification by the end of 2007 (Malkary, 2005). One example is the US Navy, whose medics in Iraq have trialled RFID dog-tags and wristbands in the field, attaching them to injured servicemen, refugees and prisoners of war. The tags can be read even in harsh desert and battlefield conditions (Greengard, 2004). Drug identification is an application attracting growing interest, as it is the basis for both medication administration and e-pedigrees. Intended to combat the counterfeiting, dilution and diversion of drugs, e-pedigrees are mandated in the US under the Prescription Drug Marketing Act. The FDA, which will administer the law, advocates RFID as their preferred technology for keeping e-pedigrees. Drug manufacturers such as Pfizer, GlaxoSmithKline, and Johnson & Johnson have trialled RFID ‘e-pedigrees’ for their products (Herper, 2004) At the Georgia Veteran’s Medical Centre RFID has been used for location identification (Ross & Blasch, 2002). Patients with visual impairments can navigate around the hospital by reading these

827

RFID in Hospitals and Factors Restricting Adoption

location tags with using a cane containing an RFID reader.

Location-Based Applications Once patients, staff, and drugs and other medical consumables are identified by RFID tags, it becomes possible to automate a number of manual processes that happen at specific locations. One example is prevention of retained surgical items. Sometimes known as the NoThing Left Behind (NTLB) policy, this involves checking that surgical implements and supplies used during an operation have been removed from the patient before they leave theatre. More than 30 hospitals around the US already employ a system that allows surgeons to use an RFID wand to detect tagged surgical sponges left inside patients (Japsen, 2008). Medication administration is another example. At the University of Amsterdam Medical Centre, patients awaiting blood transfusion are issued RFID wristbands. The blood products are also tagged. By reading the two RFID tags prior to starting the transfusion, the risk of a patient receiving incorrect blood is reduced (Wessel, 2007b). Similar systems apply the same principle to patients receiving general, anaesthetic, and chemotherapy drugs. ‘Smart’ cabinets, are already in use in a number of hospitals (Collins, 2004). Drugs and other consumables are tagged, allowing RFID readers inside the cabinets to read them. Inventory levels can then be monitored in real time, and warnings given about expired items or items low stock levels. The cabinets may also read the RFID badges of staff, allowing only authorised people to access restricted drugs. Smart cabinets are relatively expensive, and tagging large numbers of items can be costly and time-consuming. As a result, some companies have started to provide and monitor cabinets as a managed service, with monthly fees rather than a large up-front cost (Bacheldor, 2007a, 2007f).

828

RFID makes it safer to take inventory of sterilised items, hazardous chemicals, and other objects with which people should have minimal contact. Sokol and Shah (2004) describe how a US surgical supplies company have embedded RFID tags in over 1 million re-usable surgical gowns and drapes. These are cleaned, assembled into surgical packs, and sterilised. As a final step in the process an RFID reader scans the contents of the pack, verifying that it is complete without compromising the sterilisation. National University Hospital and Alexandra Hospital in Hong Kong used a location-based application for quarantine management during the SARS epidemic of 2002/03 (Roberti, 2003). Two areas were set aside for patients with SARS-like symptoms. All patients, staff and visitors entering these areas were issued with RFID-enabled cards and their movements recorded. Had any patients within these areas been diagnosed with SARS, it would have been possible to identify all the people they had been in contact with during the incubation period.

Tracking Applications RFID technology is being used to track thousands of wheelchairs, beds, IV pumps and other pieces of expensive medical equipment at several hospitals, including Bielefeld City Clinic in Germany (Wessel, 2007a), Lagos de Moreno General Hospital in Mexico (Bacheldor, 2007c), and the Walter Reed Army Medical Centre in the US (Broder, 2004). Each item is tagged and can be located in real time via a network of RFID readers throughout the hospital. Some such systems, often referred to as Real Time Location Systems (RTLSs), also keep track of whether the equipment is available, in use, awaiting cleaning, or undergoing maintenance. If patients and staff are tagged for identification, then patient tracking and staff tracking are also possible. Systems are already available commercially for tracking babies. Osborne Park Hospital in Australia places RFID bands on the

RFID in Hospitals and Factors Restricting Adoption

legs of newborn babies (Deare, 2004), following a kidnapping from its maternity ward. Readers around the ward allow the location of each baby to be tracked. Readers at the exits set off an alarm if they detect a baby being removed from the ward without the presence of an authorised staff member. Alerts are also raised if the ankleband is cut, or if it remains stationary for some time. This may indicate that the band has been removed, or that there is something wrong with the baby. By collecting tracking data over time, hospitals can also use workflow analysis to gain insights into, and identify possible improvements to, their procedures. Huntsville Hospital (Bacheldor, 2007e) is using a system to analyse patients and staff movements up to and following surgery. Ospdedale Traviglio-Caravaggio in Italy has been using RFID to analyse the progress of patients from first arrival in the Emergency Department (ED) (Swedberg, 2008). The cost of tracking systems can be significant, depending on the area to be covered, the number of items to be tracked, and the accuracy required. This cost can be reduced by employing existing wireless networks. For example, Lagos de Moreno General Hospital’s system operates over its WiFi (802.11) network, using the access points as RFID readers and simplified network cards as tags. Walter Reed Army Medical Centre’s RTLS works over a Zigbee (802.15) network. RTLS systems are also available as a managed service, with monthly fees rather than a large up-front cost (Bacheldor, 2007b).

Data Storage Applications The applications described above typically read only the RFID tag’s identifier, and retrieve or update data across a network. But more advanced tags also permit data to be stored on the tag itself. Georgetown University Hospital has trialled a system where the patient’s RFID wristband stores their blood type, allergies and medications (Schuerenberg, 2007). This allows the data to be

viewed quickly even when network access is not available.

IS Adoption in Hospitals The case studies above contain individual hospitals’ experiences with RFID technology. But there is little research on the overall uptake of RFID in the health care sector. In comparison, there is a more significant body of research on adoption of IS generally. The research suggests that hospitals, and the health care sector as a whole, spend relatively little on IS. A recent survey of US hospitals shows that almost 50% spend between 1% and 2.5% of their budget on IS, with 20% spending less and 30% spending more (Morrissey, 2004). According to the IOM, the health care sector ranks 38th out of 53 industries in IS spending per employee, at about 1/25th the level of securities brokers (Institute of Medicine, 2001). A more recent survey in the UK showed a similar pattern, with IS spending per employee in the health care sector approximately one-ninth of that in the financial industry (Wallace, 2004). There are a number of reasons why measuring IS investment per employee may give low figures for the health care sector. One reason is the nature of work in a hospital. Hospitals generally operate 24/7. Medical staff spend a large part of their day on rounds or in theatre, rather than at a desk (Reddy & Dourish, 2002). So while a personal computer (PC) in an office may be used by one person for eight hours a day, a hospital PC is likely to be used by three shifts a day and multiple people during each shift. A second reason is that in comparison to a financial institution, hospitals employ a lot of support staff. Cleaners, caterers, orderlies and the like typically have less need for IS than, for instance, securities brokers. IS has also frequently proven to be a bad investment for hospitals in the past. Rosenal et al (Rosenal, Paterson, Wakefield, Zuege, & LloydSmith, 1995) state that “the set of all successful

829

RFID in Hospitals and Factors Restricting Adoption

HIS implementations is only slightly larger than the null set”. Littlejohns et al (Littlejohns, Wyatt, & Garvican, 2003) suggest that around 75% of hospital IS projects fail. Heeks et al (Heeks, Mundy, & Salazar, 1999) offer a brighter picture, giving a failure rate of only 50%. The work of Heeks et al is one of the few pieces of research to analyse a number of HIS projects and attempt to generalise some theory as to why they fail. They conclude that the probability of failure is proportional to the size of ‘ConceptReality’ gaps. The larger the gap between the HIS designers’ concept of a system and the end users’ reality, the more likely the system is to fail. While this conclusion is hardly novel, and is not specific to hospitals, the authors do identify seven dimensions on which gaps may occur: Information, Technology, Processes, Objectives, Skills, Management and Resources. A more common approach taken by researchers of HIS success and failure has been to contextualise a general model. For example, England et al (England, Stewart, & Walker, 2000) apply some of Rogers’ seminal work on the diffusion of innovation within organisations (Rogers, 1995) to the diffusion of IS within hospitals. Rogers identified eight organisational factors, shown in Table 1, and five technological factors, shown in Table 2. Evaluating the organisational factors from an analysis of Australian hospitals, England et al conclude that IS innovations would be expected to diffuse slowly. Based on the technological factors, they further conclude that administrative IS, such as accounts and payroll, would diffuse more quickly than more strategic or clinical IS. Although England et al have attempted to base their arguments on previously published research, they note that there is a dearth of suitable material. Thus a number of their arguments seem to be based on their personal experience, which is not documented in any detail. In addition, their framework is relatively simple. Pettigrew et al (Pettigrew, Ferlie, & McKee, 1992) and Greco and Eisenberg (1993), for instance, have developed

830

Table 1. Organisational factors affecting the diffusion of IS (Adapted from England et al., 2000) Factors that lead to faster diffusion

Present in Australian hospital environment?

Executive sponsorship

Unclear

Decentralised IS control

No

Complex work and educated staff

Yes

Lack of formal procedures

No

Highly interconnected departments

No

Readily available resources

No

Large size

Yes

Openness to outside ideas

No

Table 2. Technological factors affecting the diffusion of IS (adapted from Rogers, 1995) New technology will diffuse faster to the extent that it… Provides relative advantage over existing technology Is compatible with existing technologies Has low complexity Has been observed to be successful for other users Can be trialled with minimal disruption to normal activities

models of innovation in hospitals that also include their general environment. An earlier model developed by Stocking (1985) also includes the characteristics of individuals. Rogers’ own work highlighted the importance of this factor. He classified individuals into five types, based on how early they adopt innovation: Innovators, Early adopters, Early majority, Late majority, and Laggards. The work of England et al does have the advantage of being the most recent of the models considered. In addition, it relates to the Australian hospitals, which are commonly used as comparisons in NZ health research (Jackson & Rea, 2007).

Organisational Factors The findings of Rogers and England et al on organisational factors are summarised in the fol-

RFID in Hospitals and Factors Restricting Adoption

lowing sections. Related information on the NZ health care sector is also presented, based on the general environmental factors identified by Pettigrew et al and Greco and Eisenberg.

Figure 1. Stakeholders in the NZ health care sector (Ministry of Health, 2008b) (NGO = Non-Government Organisation; PHO = Primary Health Organisation)

Executive Sponsorship Rogers finds that if an organisation’s leaders have a positive attitude towards change, then innovations will diffuse more quickly. England et al find that there is not enough published research on the attitude of hospital executives towards IS to evaluate this factor. In the NZ health care sector there are a number of executive layers that may impact on hospital operations. As Figure 1 illustrates the Minister of Health, the Health Ministry, and the relevant District Health Board (DHB) have direct input. The Accident Compensation Corporation (ACC), an agency of the Department of Labour, also has some indirect input. These executive layers can have different objectives. Figure 2 shows the Ministry’s current objectives for the health care sector. Figure 3 shows the current objectives of the Auckland DHB. As the Minister and the executive members of each DHB change on a regular basis, the level of sponsorship and objectives can also vary over time. Easton (2002) provides an excellent review of how changes in government and in the role of DHBs have impacted substantially on the health sector over the last twenty years.

Decentralised IS Control Rogers finds that the more control of resources is centralised, the less innovative an organisation becomes. England et al report that hospitals tend to have centralised IS control for major projects, while day-to-day IS control is less centralised. In NZ, there are levels of IS management in the Ministry, the DHBs and individual hospitals. The WAVE Advisory Board advocates greater centralisation of IS in the health care sector, in order to

reduce duplication of effort and create economies of scale (WAVE Advisory Board, 2001).

Complex Work and Educated Staff Rogers finds that more complex work presents more opportunities for innovation, but only if staff are sufficiently educated to recognise and exploit those opportunities. England et al suggest that the medical work in a hospital is complex, and that the staff who carry it out are highly educated. Medical work carried out in NZ hospitals is certainly complex. An illustration of the variety of services performed by the NZ health care sec-

831

RFID in Hospitals and Factors Restricting Adoption

Figure 2. Ministry of Health’s Health Strategy objectives (Ministry of Health, 2003b)

Figure 3. Key objectives from Auckland DHB Strategic Plan 2002-2007 (adapted from Auckland District Health Board, 2002)

nurses and general practitioners (GPs) in NZ (Engelbrecht, Hunter, & Whiddet, 2007).

Lack of Formal Procedures Rogers finds that an organisation with formal procedures is less likely to innovate. England et al suggest that hospitals do officially have many formal procedures, but in everyday work are somewhat more flexible. An example of this is provided by Heeks et al, who describe a scenario Table 3. Health care events in an average day in New Zealand (adapted from 1Ministry of Health, 2008a; 2Ministry of Health, 2006) 160 people are born1 78 people die1 83,000 prescriptions are filled1 66,000 laboratory tests are analysed2 6,000 outpatients visit hospitals for treatment1 275 people have elective surgery1 29 people are diagnosed with diabetes2 1,450 people visit an Accident and Emergency department2

tor is given in Table 3 (WAVE Advisory Board, 2001). It is less clear whether the administrative work is any more complex than that carried out in other organisations. Medical staff in NZ must be well educated. However, as Chassin points out, medical training conditions doctors to be self-reliant, and to reject support systems (Chassin, 1998). This is supported by a recent survey of administrative staff,

832

1,350 people are admitted to hospital1 94,000 people take an anti-depressant medication2 235,000 people take a cholesterol lowering drug2 25 people have a heart attack1 127 children are immunised2 2,124 children and teenagers visit the dentist2 47 asthmatics are admitted to hospital2 ACC receives 4,100 new claims2 55,000 people visit their GP1

RFID in Hospitals and Factors Restricting Adoption

where nurses followed a procedure that was morally and clinically sound, but officially prohibited.

Highly Interconnected Departments Rogers finds that highly interconnected departments lead to a faster rate of diffusion, because innovations can flow more easily between them. England et al point out that hospital departments are not generally highly interconnected. They particularly note the different perspectives of medical staff and managers, and the professional sub-cultures or ‘tribes’ that are formed by medical staff with the same specialities.

Readily Available Resources Rogers finds that innovation is more likely to happen if time and resources are available to do it. England et al suggest that public hospitals are unlikely to have such resources due to “constant funding pressure”. The NZ government funds around 77% of total health care funding. The remaining 23% is paid by individuals or private insurers for treatment in private hospital. Public and private spending on health care have both risen by an average of 5% per year over the past 10 years. According to Organisation for Economic Co-operation and Development (OECD) figures, NZ’s health care spending per capita and as a proportion of GDP are both just below the median for developed countries. Health care spending per capita is approximately 81% of that in the UK health care sector, 66% of that in Australia, and 34% of that in the US (Ministry of Health, 2007). Public hospitals in NZ are extremely limited in their ability to borrow from private sector financial institutions (Ministry of Health, 2003a). The government runs the Crown Funding Agency (CFA) to provide loans and credit to public sector bodies, including DHBs.

Large Size Rogers finds that innovation happens more quickly in larger organisations, as they have more people to identify opportunities and more resources available to exploit those opportunities. England et al have classified hospitals as large, but don’t state what metric they have based this on. Hospitals are certainly large organisations in the NZ context, where 96% of businesses employ fewer than 20 people (Ministry of Economic Development, 2007). But in comparison to hospitals in other countries, NZ’s are relatively small by most measures. Considering budget, the figures given above illustrate that NZ’s is substantially less than the UK, Australia and the US. Considering infrastructure, as approximated by the number of public hospital beds per capita, NZ has half the level of Australia and 60% that of the UK (Jackson & Rea, 2007). Considering the number of staff, NZ pales in comparison to the UK’s NHS, which is the third largest employer in the world (Wallace, 2004).

Openness to New Ideas Rogers finds that organisations that are open to new ideas are able to learn about innovations and evaluate them more easily. England et al state that medical staff tend to be open to new ideas only from within their professional sub-culture. In terms of Rogers’ classification of individuals’adoption of IS, Mathie (1997) finds that doctors generally view Innovators as “misfits”, but respect Early adopters as “opinion leaders”. Mathie suggests that the adoption of clinical IS is slowed by the principle of Evidence-Based Medicine (EBM). EBM requires that a new medical practice should not be used until there is sufficient evidence that it is safe and effective. Table 4 shows a typical scale for levels of evidence (Canadian Task Force on Preventive Health Care, 1997). In reviewing literature for this chapter, no IS evaluations were

833

RFID in Hospitals and Factors Restricting Adoption

Table 4. Levels of evidence in EBM (adapted from Canadian Task Force on Preventive Health Care, 1997) I

Evidence from at least one well-designed randomized controlled trial

II – 1

Evidence from well-designed controlled trials without randomization

II – 2

Evidence from well-designed cohort or case–control analytic studies, preferably from more than one centre or research group

II – 3

Evidence from comparisons between times and places with or without the intervention; dramatic results from uncontrolled studies (e.g., results of treatment with penicillin in 1940s)

III

Opinions of respected authorities, based on clinical experience; descriptive studies or reports of expert committees

found that might be considered ‘well-designed randomized controlled trials’. Even when an IS is adopted by some clinicians, there is no assurance of widespread diffusion. For example, Morrissey (2003) reports that Computerised Physician Order Entry (CPOE) has recently been approved as an evidence-based standard by an, albeit privately managed, medical standards organisation in the US. Yet only 24% of hospitals are planning to implement it. Of those that aren’t, 54% state that it is because doctors are resistant to using it.

Technological Factors The findings of Rogers and England et al on technological factors are summarised in the following sections.

Relative Advantage Rogers finds that innovations that offer many advantages over the status quo tend to diffuse more quickly than those that offer few or no advantages. England et al seem to suggest that IS managers in hospitals believe that IS produces relative advantage, but don’t know how to prove it. The literature clearly indicates that hospitals have problems evaluating IS. As Littlejohns et al (2003) point out, “the overall benefits and costs of hospital IS have rarely been assessed”. Ammenwerth et al (2004) discuss some of the common barriers to performing and publicizing evaluations. These reflect a number of the organi-

834

sational factors above, including a lack of formal procedures, a lack of resources, and insular departments. The authors summarise previous attempts at HIS evaluation frameworks, and conclude by drafting the Declaration of Innsbruck. Although it contains 12 valid recommendations, they are quite generic and relate to the evaluation process rather than to the evaluation details. For example, one recommendation is that an evaluation should be sufficiently funded.

Compatibility Rogers finds that innovations that are highly compatible with the status quo tend to diffuse more quickly than ones that aren’t. England et al suggest that, in the hospital context, administrative applications have the most compatibility because most people know how to use them. Larger, strategic applications have less compatibility because they generally require accompanying organisational changes. Clinical applications can also have frequent compatibility problems because of the difficulty of translating paper-based medical forms to electronic format.

Complexity Rogers finds that innovations that are less complex tend to diffuse more quickly than those that are more complex. England et al don’t appear to explicitly address complexity in their analysis.

RFID in Hospitals and Factors Restricting Adoption

Observability Rogers finds that innovations that can be observed working successfully elsewhere will diffuse more quickly than those that can’t. England et al suggest that successful HIS projects are so rare that observability is problematic. At the time of writing, the only RFID application that is observable in NZ hospitals is access control. There are a small number of vendors developing or importing RFID systems for hospitals, that are observable in other countries (Hedquist, 2007). Outside of the health care sector, a number of RFID systems have been trialled or fully implemented in NZ: Livestock identification and tracking (Stringleman, 2003); Tracking carcasses on a meat processing line (Anonymous, 2004b); Identification of dogs and other pets (Department of Internal Affairs, 2005; New Zealand Veterinary Association, 2002); Timing runners competing in road races (Timing New Zealand, 2004); Identifying and tracking books in libraries (Anonymous, 2004a); and Identifying farm and forestry equipment to support a fuel delivery service (Anonymous, 2007). There have also been trials conducted by a small number of major retailers and manufacturers, and by customs officials for tracking cargo containers and identifying passports (Bell, 2004).

Trialability A trial is defined by Rogers as testing an innovation in part before committing to it fully. He finds that innovations that can be easily trialled diffuse more quickly than those that can’t. England et al suggest that small, stand-alone operational systems are more easily trialled than strategic or clinical systems.

Research Aims At the time of writing, NZ hospitals are just beginning trials of RFID. Two pilot projects are in the

early planning stages. The first will employ RFID for patient identification and patient tracking in an ED. The second will use RFID to replace or complement barcodes in an existing BCMA system used during anaesthesia (Houliston, 2005). The researcher has been an observer of the former project, and an active participant in the latter. This research has been conducted partly to inform the further development of these two projects, and others that may follow. The analysis of England et al lead them to conclude that the organisational characteristics of Australian hospitals partly explain why they have been slow to adopt IS generally, and that they would be expected to adopt small, standalone, operational IS more quickly than strategic or clinical IS. The first aim of the current research was to investigate whether these conclusions are equally valid for NZ hospitals. The second aim of the research was to consider RFID applications for hospitals in light of these conclusions. As NZ hospitals begin to pilot RFID applications, what implications do the hospitals’ organisational characteristics, the applications’ technology factors, and the strengths and weaknesses of RFID, as discussed in previous chapters, have for the likelihood of success?

INTERVIEWS AND FINDINGS This section briefly describes the interview process and presents the findings.

Interviews A number of research methods were considered for this research project. Surveys were ruled out primarily because other recent surveys on IT in NZ hospitals had shown a very low response rate (Lau, 2003). Observation was considered to be especially appropriate for gauging ‘fuzzy’ factors such as ‘Executive sponsorship’, ‘Complex work’ and ‘Interconnected departments’. However, this

835

RFID in Hospitals and Factors Restricting Adoption

was judged to be a high-risk approach, given the researcher’s inexperience with observational methods and the hospital environment. Interviews offered the best balance. They could be performed in less time than observation, but go into more detail than surveys. Given the lack of experience with RFID, interviews also provided an opportunity to introduce the technology to interviewees. A total of nine people were interviewed. All were known to the researcher or his supervisor, or were referred by another interviewee. It had been the researcher’s intention to interview more people, but availability was an issue. This may in itself be an indication of the time pressures resulting from organisational factors such as ‘Complex work’ and ‘Readily available resources’. A brief background on each interviewee is shown in Table 5.

The interviews with M, G and C were done primarily in the context of designing a prototype for the anaesthetic RFID pilot. Therefore the technology factors were discussed more than the organisational factors. Organisational factors were mostly discussed in relation to their experiences in getting the existing BCMA system implemented. The interviews with S, E and R were more structured. All followed a similar, two-part pattern. In the first part, interviewees were asked to describe at least one IS they had been involved with. For each IS, they were then asked whether it had been a success or failure, and what factors they perceived as being most significant in that outcome. The researcher explicitly mentioned Rogers’ organisational factors only once, when E appeared to struggle with identifying factors. In the second part, the researcher briefly described

Table 5. Backgrounds of interviewees Interviewee

Background

M

M is a practicing doctor. He has recently been successful in getting a BCMA system implemented for anaesthesia in his DHB. He is the main sponsor for the anaesthetic RFID pilot. Interviews with M took place in person and by e-mail.

G

G is also a practicing doctor, and a colleague of M. Interviews with G took place in person.

C

C is doctoral researcher, primarily interested in systems that support patient safety. He has been researching alongside M. Interviews with C took place in person and by e-mail.

S

S is an experienced nurse in an Intensive Care Unit (ICU). She is an early adopter of IS, and is currently researching in the area of medical data mining. Interviews with S took place in person and by e-mail.

E

E is a nursing manager, with previous experience as a practicing nurse. Her current role includes some responsibility for the design, implementation and support of nursing IS. Interviews with E took place in person.

R

R is currently a management consultant in the private sector, but was previously the Chief Information Officer (CIO) of a DHB, and a project manager within the Ministry of Health. Interviews with R took place by e-mail.

D

D is a vendor who markets an RFID application for tracking staff and equipment to hospitals (so far with no sales in NZ) and companies in the transport industry. Interviews with D took place in person.

P

P is a vendor who does bespoke IS development for various organisations including hospitals. He is leading the vendor consortium involved in the ED RFID pilot. Interviews with P took place by telephone.

B

B is a technician for electronic medical equipment. Interviews with B took place by telephone.

836

RFID in Hospitals and Factors Restricting Adoption

RFID technology and the four types of application identified in the previous section. If the interviewee had not already done so, they were asked to indicate which application they thought would be most useful to them. They were then asked to point out obvious problems that they could foresee with any of the applications. The interviews with D and P were also structured. They were asked questions about making initial contact with a hospital, the management level of hospital staff they dealt with, how well-informed the staff were about RFID and its applications, requirements for reference sites and trials, and the most common reasons for IS implementations being delayed or abandoned. The interview with B covered only RFID technology, particularly compatibility with existing hospital equipment and applications.

she eventually stopped using it because it had no executive sponsorship. As she puts it:

Findings

There’s a pattern when you implement a new system. For the first two weeks users moan and complain because it’s different. Then they’re quiet for about a month. At six weeks there’s another peak of grumbling. Then they realise that management are committed to the system, and they just accept it…

This section presents the findings from the interviews. The findings are grouped under the eight organisational and five technology factors identified in the previous section.

Organisational Factors The organisational factors considered were: Executive sponsorship; Decentralised IS control; Complex work and Educated staff; Lack of formal procedures; Highly interconnected departments; Readily available resources; Large size; and Openness to new ideas.

Executive Sponsorship England et al suggested that hospital executives could not be considered to have a consistently positive or consistently negative attitude towards IS. The interviews supported this. S described a barcode-based costing system that she had implemented several years previously. Although the system was a success in her eyes,

Management just didn’t want to know that we spent $250,000 on this burns patient. A lack of executive sponsorship does not necessarily mean the demise of an IS. M successfully implemented an anaesthetic BCMA system apparently without executive sponsorship. However, it should be noted that M is a senior staff member in his department, and getting his system implemented was: … a long, and not altogether happy, tale. An example of positive executive sponsorship was provided by E:

R noted that even when an IS had executive sponsorship, it did not necessarily make the IS a success: We had systems on our books that should have been scrapped but weren’t, because they were a manager’s pet project. Officially they still exist, but no-one, except maybe the manager, uses them… P stated that he had never had a DHB or hospital manager present at one of his product presentations. D also stated that he rarely has contact with managers, although the ED RFID pilot is an exception. The DHB’s Chief Information Officer (CIO) been involved with the project since its inception.

837

RFID in Hospitals and Factors Restricting Adoption

E expressed a belief that executive sponsorship can be diluted by the large number of executives around hospitals. D gave an example from the ED RFID pilot. While the project has sponsorship from the DHB’s CIO, IS management in the Ministry are ambivalent. They have expressed concern that it may conflict with Ministry plans for barcode-based systems.

Complex Work and Educated Staff

Decentralised IS Control

I believe that all the medical staff I encountered, regardless of their paper qualifications, had the equivalent of a Masters’ education.

England et al suggested that hospitals tend to have centralised IS control of major projects, but not necessarily of day-to-day operations. The interviews seem to support this. As far as R was aware, the ministry and each DHB had a central IS group. However, he agreed that these IS groups exercise control mostly at the level of IS strategy and significant new projects. A large measure of this control is to satisfy the requirements of government. He noted that this is criticised in the WAVE report (WAVE Advisory Board, 2001): “There is a strong focus on spending [IS] money on governance and compliance, rather than on systems aligned to health goals”. R suggested that in day-to-day dealings with end-users, each hospital’s IS group focused on support, not on control. This is supported by S, E and G, who were not aware of any policies enforced by the IS group at their hospital. E went so far as to suggest that the IS group wouldn’t have sufficient influence with management and clinicians to enforce any major policies. She recalled an incident: IS sent out a memo saying they were no longer going to support systems developed by the departments, without IS knowledge. They were very quickly asked, or instructed rather, by management to drop that particular policy. P generally has IS staff attending his product presentations, but only from the individual hospital, never from the DHB or Ministry.

838

England et al suggest that the medical work in a hospital is complex, and that the staff who carry it out are highly educated. The interviews support both these points. All interviewees agreed with both points. R noted that:

Rogers found that educated staff are more likely to recognise and exploit opportunities for innovation. However, this is not always the case in hospitals. E pointed out that even highly educated and experienced staff are not necessarily natural innovators: We get lots of complaints when new systems go in. But hardly any suggestions on how to make things better. Normally they just ask to go back to the way the previous system did things. R suggested that senior doctors who work only part-time for the hospital might be more inclined to apply innovations in their private practice than in the hospital.

Lack of Formal Procedures England et al suggested that hospitals had formal procedures, although they were not always followed. The interviews support this point. R noted that, in his experience as a CIO: Clinicians will typically not be comfortable with being limited to a single method of operation for any given system. This can be at variance with the desire of the IS professional to develop integrated and efficient systems.

RFID in Hospitals and Factors Restricting Adoption

This is illustrated well by M and C’s anaesthetic BCMA system. It is a highly procedural system, carefully designed to reduce the likelihood of drug administration errors during surgery. Yet during simulated operation trials, half of the participants neglected the simple step of scanning the barcode before administering the drug (Merry, Webster, Weller, Henderson, & Robinson, 2002). Another example was given by both S and E. Patients in their hospital were given barcoded wristbands, but neither knew which staff or procedures actually made use of the barcodes.

Highly Interconnected Departments England et al suggest that hospital departments are not generally highly interconnected. The interviews support this. R expressed his outsiders view: The [healthcare] sector is highly tribal: There are at least 12 different practitioner groups. Within the doctors’ ambit alone there are around 15 specialist colleges. There is traditionally a high level of distrust between practitioner groups. Even within one group of practitioners, there can be a high level of mistrust [sic]. All of this was exacerbated in NZ by the endless reforms and re-structures. S expressed this ‘mistrust’ indirectly: Whenever we (the ICU) get a new patient, we always go through the admission form again. The people in admissions never do it properly. It’s not surprising – they’re very busy.

Readily Available Resources England et al suggest that public hospitals are unlikely to have readily available resources. The interviews support this point. R believed that NZ hospitals do have readily available resources. He pointed to repeated in-

creases in government funding and recent statistics on DHB deficits. The 21 DHBs had a combined deficit of NZ$58 million as at June 2004, although that was a significant decrease from the NZ$170 million deficit the previous year (Pink, 2004). But those resources were not readily available to IS: Another factor that is, I think, unique to healthcare is that there is never enough money. Every dollar that is not directly spent on the delivery of care is questioned. E agreed that investment in IS is very closely scrutinised, due to its unfortunate track record. Quoting from one of her Ministry of Health directives: “Following concern regarding the quality of health [IT] investment, the Minister of Health has directed that a stepped approval be required for information systems and communication technology” (Ministry of Health, 2003a). Any IS investment over $500,000 must be approved by the Director-General of Health. Any investment over $3 million must be approved by the Minister. P believed that the time and effort required for hospital operational staff to get IS funding was the major cause for projects being delayed, cancelled or reduced in scope. He gave an example: One hospital wanted [his product] to track their wheelchairs. I gave them a quote for a full system to cover the entire complex. They went with someone else who just put readers at the exits. They wanted something quickly and couldn’t wait for approval. S still had the barcode scanners from her abandoned costing system in a box under her desk. This raises the interesting question of what other unused IS resources from failed projects are in a similar position.

839

RFID in Hospitals and Factors Restricting Adoption

Large Size England et al have classified hospitals as large. The issue of size was not raised in any of the interviews. Rogers finds that innovation happens more quickly in larger organisations, as they have more staff to identify opportunities and more resources to exploit those opportunities. Given that ‘Readily available resources’ and ‘Openness to new ideas’ already appear as separate factors, it may be that size is redundant in the case of hospitals.

Openness to New Ideas England et al suggested that medical staff were generally not open to new ideas. The interviews support this point. R suggested that being closed to new ideas was the result of medical training: Healthcare workers take a very long term view. It takes ten years for a doctor to become fully qualified, so there is an inherent conservatism. M and C provide an illustration of the challenges posed by the requirements of evidencebased medicine. They carried out two sets of simulated operation trials of their barcode-based ID (Merry et al., 2002; Webster, Merry, Gander, & Mann, 2004). The trials were level II-1 on the evidence-based medicine scale shown in Table 4. The results were positive, yet there are still only a limited number of doctors using the system. P and D both agreed that initial interest in RFID nearly always came from operational staff, and never from the medical staff. As P put it: My first contact always comes from the guy who looks after, say, the wheelchairs. He finds that a lot are going missing. He searches the ‘net and finds us.

840

P believes that operational staff are generally open to new ideas. Part of his standard product presentation is discussing how the staff could employ RFID beyond their initial problem.

Technological Factors The technological factors considered were: Relative advantage over existing technology; Compatibility with existing technology; Low complexity; Trialability; and Observability.

Relative Advantage M, G, C, S, E and R were asked which of the RFID applications described in the previous section would be the most useful to them. The responses illustrate that relative advantage is in the eye of the beholder. M and G selected drugs identification. In particular, M was interested in using RFID to replace barcodes in their anaesthetic BCMA system, leading to the initiation of the anaesthetic RFID pilot. This would remove the need for anaesthetists to scan barcodes, which has proven to be a major usability issue for the BCMA system. C was most interested in RFID ‘smart’ cabinets. He expressed particular interest in having alerts when drugs were about to reach their expiry date, and when quantities reached minimum levels. S nominated data storage. If every person had an RFID chip containing their medical history embedded under their skin, this would mitigate the problem of having inaccurate or incomplete details on patient admission forms. She went so far as to suggest a ‘Top 10’ list of data to store. E believed that nurses would be interested in anything that made their medication rounds easier. She thought that a medication administration would be most useful. R suggested RTLS. He noted a problem that many hospitals have with theft of laptop computers, personal digital assistants (PDAs) and similar items. The ability to raise an alert when a piece of

RFID in Hospitals and Factors Restricting Adoption

equipment leaves a particular area, and to track it, may help to prevent stolen items from leaving the building. The issue of evaluating IS was raised with E and R, as the only two interviewees with any official responsibility for evaluation. R had a formal evaluation process, albeit inherited from a previous role in the private sector. E had no formal evaluation process, and tended to rely heavily on user feedback. She gave the example of an IS being selected by the doctors in one particular department: They’ve been looking for nearly four years, and they’re finally about to make a choice. Basically, we’re so desperate to get something in place, we’ll go along with whatever they decide. D and P were asked how often the systems were replacing existing systems, and what benefits their clients expected. The ED RFID pilot, in which D was involved, was not replacing any existing system. P stated that his RTLS is often intended as a replacement for some form of manual ‘equipment booking’ system that hospital staff are too busy to use. The advantage is having a system automatically detect where a piece of equipment is, rather than staff having to write that on a board or enter it into a PC.

Compatibility The interviewees raised a number of compatibility issues that they believed might impact on the adoption of particular RFID applications. They are discussed here grouped under the dimensions identified by Heeks et al. Information: Two major issues regarding compatibility of information were raised by S. First, the medical data stored in embedded RFIDs must be compatible with the needs of end users. As noted above, the major reason for the failure of S’s costing IS was that management did not want the information it produced. She also cited

the National Health Index system (NHI) (NZ Health Information Service, 2004). Although an estimated 95% of the population are recorded in the system, the information kept is too basic to be of use. Second, the data must be consistent with that held in paper-based records. Technology: B confirmed that there were potential compatibility issues with RFID technology and existing electronic medical devices, such as pacemakers. However, he suggested that a more likely problem would be other devices interfering with communication between RFID readers and tags. Surgical and medical diathermy machines are two examples of devices that emit much stronger electromagnetic fields than RFID. They are widely recognised as major sources of interference (Bassen, 2002). D highlighted the importance of technology compatibility in his RFID pilot: We had a few hospitals express an interest in running the trial. We chose [the pilot DHB] because they have the wireless network infrastructure that makes it easiest for us. The other interviewees didn’t mention any issues related to technology compatibility. Although RFID is novel in the hospital environment, all interviewees were familiar with related technologies such as barcodes and proximity cards. Processes: All interviewees made the point that adopting RFID for a particular process should not prevent the process from being carried out as it is now. For example, M, G and C required that, if RFID labels were to replace barcode labels in their drug identification application, then the barcode must be printed on them. Then, if the RFID application failed, the existing barcode application could be reverted to. E and S both suggested that staff would be reluctant to wear RFID tags. Both accepted that there were valid reasons for wearing them, such as authenticating access to smart cabinets. However, they expressed concern that the data gathered

841

RFID in Hospitals and Factors Restricting Adoption

might be used for other purposes, such as covert tracking. As E put it: There are managers who would love to know exactly how much time doctors and nurses spend on their feet. E, S and R all believed that management would have to make the adoption of any RFID applications that involve tagging staff as transparent as possible. Skills: D and P both reported that the hospital staff they interacted with knew little or nothing about RFID technology when they first gave product presentations. But all interviewees seemed to agree that medical staff already had the skills to deal with the RFID applications discussed. When it was found that few of the staff in S and E’s departments used barcode scanners, the researcher pointed out that patient and drug identification applications would probably require the use of handheld devices, such as PDAs or tablet PCs. Even though the staff had little or no experience with these either, S and E believed they would have no difficulty learning to use them. Some RFID applications may also require new skills to be learned by patients. S noted that if medical data were to be stored on embedded RFID chips, then the general public may have to be educated in protecting that data. Resources: M, G and C are the only interviewees with whom the costs of RFID were discussed in any detail. The tag and reader prices mentioned did not appear to be a major concern. The researcher suggested that with the interest shown in RFID by pharmaceutical manufacturers, and the possible influence of DHBs and PHARMAC over pharmaceutical suppliers, the cost involved in tagging drugs may largely be met at those levels of the supply chain. M considered this unlikely at the current time.

Complexity Two interviewees also raised complexity issues that they believed might impact on the adoption of particular RFID applications. E suggested that getting all parties involved in the health care sector to agree on what medical data should be stored on an embedded RFID chip would be: …next to impossible. That’s why data in the NHI is so basic. It’s mostly stuff that’s available publicly from other sources. B believed that designing networks of RFID readers to support an asset tracking application would be complex. While some IS groups have experience with RF engineering through the implementation of wireless networks, such networks consist largely of homogeneous components. In contrast, RFID tags are more heterogeneous. Patient identification applications, for example, are likely to use short-range, passive tags while RTLSs are likely to use longer-range, active tags. He also expected that keeping up to date with the RFID standards would be a challenge.

Trialability The importance of trialability varied between the interviewees. It was vital for M, G and C. They regard the principles of evidence-based medicine as very important, particularly for the clinical systems that they use in theatre. This is illustrated by the simulated operation trials they performed on their own IS. D and P both stated that they always recommend trials of their systems before full implementation. E provided a contrasting view. When asked specifically about trialability, her reply indicates that it isn’t essential: We don’t normally have time for trials. We just put systems in. If they don’t work we stop using them.

842

RFID in Hospitals and Factors Restricting Adoption

Observability

Diffusion of IS in NZ Hospitals

Observability appeared to be important for most of the interviewees. It was one of R’s evaluation criteria, and he noted that it regularly appeared in proposals for major new IS projects. For instance, a recent proposal for a clinical HIS by one DHB is based largely on the success of the same system at a neighbouring DHB (Wright, 2003). Both D and P stated that their clients always asked for reference sites. D recalled a meeting he attended where a few DHB CIOs were hosting a visiting Canadian CIO:

The interview findings seem to suggest that there are three organisational factors where England et al’s evaluation of Australian hospitals may differ for NZ hospitals. These are: Decentralised IS control, Lack of formal procedures, and Large size. Assigning No to ‘Decentralised IS control’ conceals the fact that day-to-day IS control is largely decentralised. Likewise, assigning No to ‘Lack of formal procedures’ conceals the fact that such procedures exist primarily at high levels. It is therefore suggested that these factors be assigned a value of Unclear. It is suggested that the ‘Large size’ factor also be assigned a value of Unclear. The basis on which England et al classified Australian hospitals as large is not clear. However, on measures such as number of employees, spending per capita, and public hospital beds per capita, NZ hospitals are significantly smaller. Finally, the factor ‘Large size’ may be redundant, given the presence of the ‘Readily available resources’ factor. Table 6 shows a comparison of the values suggested by England et al and the values suggested by this research, with differences highlighted. If each of these organisational factors had equal weighting in determining the rate of diffusion, then removing two Nos and one Yes should result in a higher rate. Thus it might be expected to observe NZ hospitals diffusing IS, including RFID, more quickly than Australian hospitals. Yet that does not appear to be the case. This anomaly may be explained in a number of ways. One possibility is that the organisational factors do not have equal weightings in determining the rate of diffusion, perhaps specifically in hospitals. The factors ‘Highly interconnected departments’, ‘Readily available resources’, and ‘Openness to new ideas’ appear to outweigh the factors ‘Decentralised IS control’, ‘Complex work and educated staff’, ‘Lack of formal procedures’, and ‘Large size’. Based simply on the number of times that these factors were mentioned in the interviews,

He asked what their innovation strategy was. [A NZ CIO] replied ‘We look at what other hospitals have been doing for at least three years. A further example of the importance of observability was provided at a demonstration of the prototype anaesthetic RFID system. In previous demonstrations the application had been standalone, simply displaying the RFID tag numbers of the items it was reading. In the last demonstration it was integrated with the existing BCMA system, providing the same functionality as using barcodes. As G commented: It’s one thing to see the tag numbers come and go on the screen. It’s something quite different to see the drug names come up in [the existing IS].

DISCUSSION This section proposes some refinements to the conclusions of England et al in light of the interview findings, and considers the implications for RFID applications. Some possible research areas for further grounded theory iterations are also highlighted.

843

RFID in Hospitals and Factors Restricting Adoption

Table 6. Organisational factors affecting the diffusion of IS in NZ hospitals (adapted from England et al., 2000) Factors that lead to faster diffusion

Present in Australian hospital environment ?

Executive sponsorship

Unclear

Decentralised IS control

No

Complex work and educated staff

Yes

Lack of formal procedures

No

Highly interconnected departments

No

Readily available resources

No

Large size

Yes

Openness to outside ideas

No

Present in NZ hospital environment ? Unclear Unclear Yes Unclear No No Unclear No

this seems plausible. However, that is based on the researcher’s interpretations of interviewee responses. A more explicit, quantitative research method, such as a questionnaire with all the factors listed, should be used to confirm this. Another possibility is that there are additional organisational factors, perhaps specific to hospitals, that are not included in Rogers’ work and have not been considered by England et al. In analysing the interviews for this research, statements were associated with the established factors. Analysing the interviews without these a priori classifications may highlight new factors. A third possibility is that RFID technology is sufficiently different from the generic innovation considered by Rogers and the general IS considered by England et al, that some of the organisational factors have a different effect on its diffusion. Specifically, the factor ‘Large size’ may have an exaggerated effect on the diffusion of RFID, or the factors ‘Decentralised IS control’ and ‘Lack of formal procedures’ may have a minimal or even negative effect.

844

Decentralised IS control may in fact decrease the diffusion rate of RFID. RFID is an infrastructure technology, similar to other networking technologies such as mobile phones or wireless networks. The pattern seen with other infrastructure technologies is that diffusion is slow until a critical mass of end users is reached, at which point diffusion increases sharply (Rogers, 1995). Reaching the critical mass of end users is likely to be easier for an organisation with centralised IS control than for an organisation with decentralised IS control. Centralised IS control should ensure that all end users have compatible technology. With decentralised IS control, end users may independently choose incompatible technologies. This is likely to result in a longer wait for one technology to reach a critical mass of end users. Even when this does occur, there may be resistance from end users of other technologies to adopt a new one. This resistance may be exacerbated in the health care sector by the presence of non-interconnected departments. A lack of formal procedures may also decrease the diffusion rate of RFID. A formal, documented procedure lends itself to automation more easily than an informal, unwritten procedure. This is illustrated by the fact that physical security, a domain with many formal procedures, is currently the most widespread application for RFID technology. Walker et al suggest that the greatest value of RFID will come from using it for “the real work of organising data to help trigger transactions and business rules” (Walker, Spivey Overby, Mendelsohn, & Wilson, 2003). It is less obvious how large size might directly affect the diffusion of RFID. Large size implies more staff who may identify opportunities to apply RFID technology. But it has already been established that clinicians tend not to be open to outside ideas. Large size suggests more resources to exploit opportunities. But in the health care sector, resources are not readily available for investment in IS. Large size in terms of physical area or services provided may increase diffusion,

RFID in Hospitals and Factors Restricting Adoption

as it offers improved trialability. RFID applications are frequently trialled in distinct areas of the hospital, such as the ED, or for specific services, such as blood transfusion. But diffusing an RFID application over a large physical area would be impeded by the lack of readily available resources to create a reader network. Diffusing across a range of services would be complicated by noninterconnected departments and a lack of openness to outside ideas.

Diffusion of RFID Applications The interviews seem to suggest that all of Rogers’ technological factors will impact on the diffusion of particular RFID applications, but not to the same extent. For instance, all interviewees raised issues about compatibility, but complexity issues were noted by only two. The interviews also seem to support England et al’s conclusion that small, administrative or operational IS should diffuse more quickly than large, strategic or clinical IS. The three largest, most strategic applications discussed were data storage - RFID chips holding a person’s medical data embedded under their skin - and hospital-wide staff tracking, and RTLS. More concerns were raised about the first two of these applications than any others. Four classes of RFID application were identified in the section on RFID and Hospital IS. A discussion of the factors impacting on each class is presented in the following sections.

Identification Applications Patient identification supports the major health care concern of easily sharing medical information. Drug identification directly supports the major concern of patient safety. In addition, drug identification was nominated by two interviewees as the applications that would be of the most benefit to them. Overall this suggests a high level of relative advantage. Thus it is no surprise to see that patient and drug identification are the major

components of the first two RFID pilots in NZ hospitals. As the first experience with RFID technology, there are likely to be issues around technology standards, integration with existing IS, training and other aspects of compatibility and complexity. But there are many patient identification systems already deployed overseas, so it should be more observable than other types of application. Trialability is possible by limiting the application to a single area of the hospital, as is being done in the NZ anaesthetic drug and ED patient RFID trials. The cost of handheld RFID readers suitable for identification applications is relatively small. The cost of RFID wristbands for patient identification is minimal, since they can be re-used. However, the cost of tagging individual drug doses is likely to be the major impediment to the adoption of drug identification applications. The Ministry of Health gives cost as the primary reason for recommending the use of barcodes in its proposed BCMA system. Cost may become less of an issue if more pharmaceutical manufacturers and distributors choose, or are required by law, to adopt RFID for drug e-pedigrees.

Location-Based Applications A number of location-based applications directly support the major health care concern of patient safety. Medication administration, prevention of retained surgical items, and quarantine management are clear examples. Smart cabinets offer indirect support by ensuring that sufficient quantities of drugs and supplies are in stock, and that they are not expired. Medication administration and smart cabinets were each nominated by one interviewee as the application that would be of the most benefit to them. Overall this suggests a reasonably high level of relative advantage. These systems are typically automating standard operational practices, so should not be complex. But the interviewees did raise some compatibility issues, notably that staff may feel

845

RFID in Hospitals and Factors Restricting Adoption

they are being monitored. Trialability of smart cabinets, for example, may be difficult if staff are able to avoid using them by obtaining drugs or supplies from other storage areas. The number of such systems implemented around the world is relatively low, which may make observability an issue. As noted above, the cost of tagging individual drug doses and surgical items will be significant. Smart cabinets themselves are also expensive. There are currently no companies in NZ providing them as a managed service, as there are in other countries.

Tracking Applications Patient tracking applications directly support the major health care concern of patient safety. Staff tracking and RTLSs offer indirect support by ensuring that staff and equipment can be found quickly if required for an emergency. RTLS was nominated by one interviewee as the application that would be of the most benefit to them. Both vendors interviewed also deal mainly in patient tracking and RTLS, although not with any sales in NZ. Overall this suggests a medium level of relative advantage. Tracking applications are the most complex of the four types discussed here. Some idea of the complexity is given by Sokol and Shah’s detailed cost-benefit analysis for an RTLS (Sokol & Shah, 2004). Designing an effective RFID reader network requires expertise in RF engineering. Making use of existing wireless networks is one way to reduce complexity, but that creates potential compatibility issues with existing IS and requires more expensive tags. Tracking applications also have the most potential compatibility issues identified by interviewees. Most significant are the privacy of staff, and potential electromagnetic interference between RFID readers and common medical electronic devices. Despite these issues there have been a relatively high number of RTLSs piloted overseas, so observability should not be difficult.

846

Trialability can be made easier by limiting the tracking to a specific area, such as a single ward. The cost profile of tracking applications differs from the other four types discussed. They require more RFID readers, but significantly fewer tags than, for example, drug identification. The cost can be controlled by varying the degree to which existing wireless networks are used, the area to be covered, the number of people or items to be tracked, and the accuracy required. There are currently no companies in NZ providing RTLSs as a managed service, as there are in other countries.

Data Storage Applications Data storage applications offer one solution to the major concern of easily sharing medical records. Storing patient medical data in an RFID tag was nominated by one interviewee as the application that would be of the most use to them. Overall this suggests a reasonably low level of relative advantage. Compatibility of information is the most significant issue identified by interviewees. There must be agreement among potential users on what information is stored. Assuming that the information is encoded for security, and to compress it into the small storage space available on an RFID tag, there must also be agreement on encoding standards. Writable RFID tags are currently more expensive than the read-only versions, but the cost difference is likely to decrease over time.

Future Research This research would benefit from further interviews with a more representative range of people. The current group of interviewees is dominated by clinicians, who could be considered to be ‘Innovators’ or ‘Early Adopters’ under Rogers’ classification, and whose experience is in public hospitals in the same DHB. Further interviews should include staff from other DHBs, and health

RFID in Hospitals and Factors Restricting Adoption

care organisations other than public hospitals, more managers, more ‘Late Adopters’, and possibly some patients. There is a danger in interviews that people give ‘correct’ responses rather than truthful responses. While the researcher didn’t sense such behaviour during interviews conducted in person, it is more difficult to discern in an e-mail or telephone interview. As already noted, observation may have been a more suitable methodology for judging the true state of factors such as ‘Executive sponsorship’, ‘Complex work and educated staff’ and ‘Highly interconnected departments’. Privacy has emerged as a clear concern for RFID applications that involve staff tracking and smart cabinets. Further investigation seems warranted. What exactly are the privacy concerns of staff? What measures, if any, could be taken to reduce these concerns? Security of RFID tags has been identified as a key issue for data storage applications. A number of approaches have been suggested in the research (Juels, Rivest, & Szydlo, 2003) but these mostly relate to the retail scenario. Are they suitable for securing tags containing medical information? Chao, Hsu and Miaou (2002) propose a scheme intended specifically for keeping medical data on an RFID chip confidential. There remain a number of aspects still to be investigated. For instance, how will the source of data written to a tag be authenticated? If it is done by a medical data certification authority, how might that function? It is clear that most RFID trials and implementations are taking place in US hospitals. This may simply reflect the fact that the recent development of RFID technology has taken place largely in the US. It might also indicate that some aspect of US hospitals increases the diffusion of RFID. The major feature distinguishing US hospitals from those of other countries is the level of competition. Might operating in a competitive market affect the diffusion of RFID? Is the ‘Need to compete’ reflected in Rogers’ organisational factors?

Executive sponsorship is widely regarded as a critical success factor for most types of IS. Further research on the attitude of hospital executives to IS, and RFID in particular, would be valuable.

CONCLUSION This chapter has investigated the factors that have restricted the adoption of IS by hospitals, and specifically of RFID technology in NZ hospitals. The health care sector has historically invested relatively little in IS, in comparison to the manufacturing sector, the financial sector and most others. But this is to be expected. Hospitals and other health care organisations exhibit characteristics associated by Rogers with slow diffusion of innovation: No clear executive sponsorship; Centralised IS control; Formal procedures; Shortage of readily available resources; Little or no connection between departments; and Lack of openness to new ideas. When hospitals have implemented IS, there has been a high rate of failure. In recent years governments, regulators, insurers, and consumers have been calling for the greater use IS to, among other things, make it easier to improve patient safety and more easily share patient medical records. The NZ government has played its part, supporting a National Health Index and Health Information Strategy, and planning to invest in CPOE and BCMA systems. Innovative hospitals overseas, notably in the US, have been applying RFID technology to these problems. A range of applications, from simple patient identification to medication administration to real-time tracking of patients, staff and equipment have been successfully piloted. Despite this, and successful applications of RFID in other industries in NZ, local hospitals are just beginning RFID pilots. Interviews conducted with practicing clinicians, a nurse manager, a CIO, an RF technician, and RFID vendors suggest that NZ hospitals may operate day-to-day with less centralised IS con-

847

RFID in Hospitals and Factors Restricting Adoption

trol and fewer formal procedures than hospitals in Australia, and perhaps other countries. While this might be expected to hasten the adoption of IS generally, it may explain why an infrastructural technology such as RFID has been slower to diffuse. The two RFID pilots that are about to begin – anaesthetic drug identification and patient tracking in ED - will be closely observed by other NZ hospitals. The pilots have technological profiles that give them a good chance of success. They don’t raise issues of privacy, security, or any of the other major concerns highlighted in the interviews. However both will be expensive to implement beyond the trial, and will have to demonstrate real benefits to the inherently cautious health care sector.

REFERENCES Ammenwerth, E., Brender, J., Nykanen, P., Prokosch, H. U., Rigby, M., & Talmon, J. (2004). Visions and strategies to improve evaluation of health information systems: Reflections and lessons based on the HS-EVAL workshop in Innsbruck. International Journal of Medical Informatics, 73, 479–491. doi:10.1016/j.ijmedinf.2004.04.004 Anonymous. (2004a). Radio Frequency ID. Retrieved November 1, 2004, from wiki.lianza.org.nz

Bacheldor, B. (2007a). ASD Healthcare Deploys RFID Refrigerated Drug Cabinets. Retrieved December 20, 2007, from www.rfidjournal.com Bacheldor, B. (2007b). AT&T Debuts Managed RTLS for Health Care Organisations. Retrieved December 20, 2007, from www.rfidjournal.com Bacheldor, B. (2007c). Local Hospital Spearheads Mexico’s Digital-Hospital Initiative. Retrieved February 12, 2008, from www.rfidjournal.com Bacheldor, B. (2007d). N.J. Medical Center Uses LF Tags to Protect Patient Records. Retrieved February 12, 2008, from www.rfidjournal.com Bacheldor, B. (2007e). RFID Documents Surgery at Huntsville Hospital. Retrieved February 12, 2008, from www.rfidjournal.com Bacheldor, B. (2007f). UMass Med Centre Finds Big Savings Through Tagging. Retrieved December 20, 2007, from www.rfidjournal.com Bacheldor, B. (2008). RFID Take Root in Bangladesh. Retrieved February 12, 2008, from www. rfidjournal.com Bassen, H. (2002). Electromagnetic interference of medical devices and implications for patient safety. International Journal of Bioelectromagnetism, 4(2), 169–172.

Anonymous. (2004b). RFID - Tracking every step you take. Retrieved February 12, 2008, from www.istart.co.nz

Bates, D. W., Cullen, D. J., & Laird, N. (1995). Incidence of adverse drug events and potential adverse drug events. Journal of the American Medical Association, (274): 29–34. doi:10.1001/ jama.274.1.29

Anonymous. (2007). Tracient Technologies helps Mini Tankers NZ deliver to its customers. Retrieved February 12, 2008, from www.istart.co.nz

Bell, S. (2004). Kiwi firms radio in RFID progress. Retrieved April 12, 2004, from www.computerworld.co.nz

Auckland District Health Board. (2002). Proposed Strategic Plan for the Auckland District Health Board 2002-07. Auckland.

Broder, C. (2004). Hospitals Wade into AssetTracking Technology. Retrieved October 30, 2004, from www.ihealthbeat.com

848

RFID in Hospitals and Factors Restricting Adoption

Canadian Task Force on Preventive Health Care. (1997). Levels of Evidence - Research Design Rating. Retrieved October 1, 2004, from www. ctfphc.org

Engelbrecht, J., Hunter, I., & Whiddet, R. (2007). Further Evidence of How Technology Availability Doesn’t Guarantee Usage. Health Care and Informatics Review.

Chao, H. M., Hsu, C. M., & Miaou, S. G. (2002). A data-hiding technique with authentication, integration, and confidentiality for electronic patient records. IEEE Transactions on Information Technology in Biomedicine, 6(1), 46–53. doi:10.1109/4233.992161

England, I., Stewart, D., & Walker, S. (2000). Information technology adoption in health care: when organisations and technology collide. Australian Health Review, 23(3), 176–185.

Chassin, M. R. (1998). Is healthcare ready for sigma six quality. The Milbank Quarterly, 76(4). doi:10.1111/1468-0009.00106 Classen, D. C., Pestotnik, S. L., Evans, R. S., Lloyd, J. F., & Burke, J. P. (1997). Adverse drug events in hospitalized patients: excess length of stay, extra costs, and attributable mortality. Journal of the American Medical Association, 277, 301–306. doi:10.1001/jama.277.4.301 Collins, J. (2004). Healthy RFID Rivalry for Hospitals. Retrieved September 9, 2004, from www.rfidjournal.com Davis, P., Lay-Yee, R., Briant, R., Ali, W., Scott, A., & Schug, S. (2003). Adverse events in New Zealand public hospitals II: preventability and clinical context. The New Zealand Medical Journal, 116(1183). Deare, S. (2004). Hospitals Ga-Ga Over RFID. Retrieved October 30, 2004, from www.pcworld. idg.com.au Department of Internal Affairs, Dog Control (Microchip Transponder) Regulations 2005, (2005). Drucker, P. (1989). The New Realities. New York: Harper & Row. Easton, B. (2002). The New Zealand health reforms of the 1990s in context. Applied Health Economics and Health Policy, 1(2), 107–112.

Greco, P. J., & Eisenberg, J. M. (1993). Changing physicians practices. The New England Journal of Medicine, 329, 1271–1274. doi:10.1056/ NEJM199310213291714 Greengard, S. (2004). A Healthy Dose of RFID. Retrieved September 8, 2004, from www.rfidjournal.com Health Information Strategy Steering Committee. (2005). Health Information Strategy for New Zealand 2005. Wellington: Ministry of Health. Hedquist, U. (2007). Orion Health teams with Oracle and Intel in Spain. Retrieved February 12, 2008, from www.computerworld.co.nz Heeks, R., Mundy, D., & Salazar, A. (1999). Why Health Care Information Systems Succeed or Fail. In A. Armoni (Ed.), Health Care Information Systems: Challenges of the Next Millenium: Idea Group Publishing. Herper, M. (2004). Tiny Chips Could Combat Counterfeit Pills. Retrieved August 1, 2004, from www.forbes.com Houliston, B. (2005). Integrating RFID Technology into a Drug Administration System. Paper presented at the Health Informatics NZ Conference, Auckland, New Zealand. Institute of Medicine. (2001). Crossing the Quality Chasm: A New Health System for the 21st Century. Washington DC: National Academies Press.

849

RFID in Hospitals and Factors Restricting Adoption

Jackson, G., & Rea, H. (2007). Future hospital trends in New Zealand. The New Zealand Medical Journal, 120(1264).

Mathie, A. Z. (1997). Doctors and change. Journal of Management in Medicine, 11(6), 342–356. doi:10.1108/02689239710195233

Japsen, B. (2008, 2 January 2008). Technology cuts risk of surgical sponges. Retrieved February 12, 2008, from www.chicagotribune.com

Merry, A., & Webster, C. (2004). Bar codes and the reduction of drug administration error in anesthesia. Seminars in anesthesia. Perioperative Medicine and Pain, 23, 260–270. doi:10.1053/j. sane.2004.05.004

Johnson, J. A., & Bootman, J. L. (1995). Drugrelated morbidity and mortality: a cost of illness model. Archives of Internal Medicine, 155, 1949–1956. doi:10.1001/archinte.155.18.1949 Johnston, M. (2007). Wired for saving lives. Weekend Herald, August 25, p. B4. Jonietz, E. (2004). Making Medicine Modern. Retrieved August 1, 2004, from www.technologyreview.com Juels, A., Rivest, R. L., & Szydlo, M. (2003). The Blocker Tag: Selective Blocking of RFID Tags for Consumer Privacy. Paper presented at the Tenth International Conference on Computer and Communication Security, Washington, DC. Kim, K. K., & Michelman, J. E. (1990). An examination of factors for the strategic use of information systems in the healthcare industry. MIS Quarterly, 14(2), 201–215. doi:10.2307/248778 Lau, L. (2003). Pen Based Computers in Health Care. Unitec New Zealand, Auckland. Leape, T., Bates, D. W., & Cullen, D. J. (1995). Systems analysis of adverse drug events. Journal of the American Medical Association, 274, 35–43. doi:10.1001/jama.274.1.35 Littlejohns, P., Wyatt, J. C., & Garvican, L. (2003). Evaluating computerised health information systems: hard lessons still to be learnt. British Medical Journal, 326, 860–863. doi:10.1136/ bmj.326.7394.860 Malkary, G. (2005). Healthcare without Bounds: Trends in RFID. Menlo Park, CA: Spyglass Consulting Group.

850

Merry, A., Webster, C., Weller, J., Henderson, S., & Robinson, B. (2002). Evaluation in an anaesthetic simulator of a prototype of a new drug administration system designed to reduce error. Anaesthesia, 57, 256–263. doi:10.1046/j.00032409.2001.02397.x Ministry of Economic Development. (2007). SMEs in New Zealand: Structure and Dynamics. Wellington: Ministry of Economic Development. Ministry of Health. (2003a). Guidelines for Capital Investment. Wellington: Ministry of Health. Ministry of Health. (2003b). New Zealand Health and Disability Sector Overview. Wellington: Ministry of Health. Ministry of Health. (2006). A Day in the Life statistics. In DayInTheLife.xls (Ed.). Wellington, New Zealand. Ministry of Health. (2007). Health Expenditure Trends in New Zealand 1994-2004. Wellington: Ministry of Health. Ministry of Health. (2008a). Every Day in New Zealand: DVD. Wellington: Ministry of Health. Ministry of Health. (2008b). Statement of Intent 2008-11. Retrieved September 26, 2008, from www.moh.govt.nz Morrissey, J. (2003). An info-tech disconnect. Modern Healthcare, 33, 6. Morrissey, J. (2004). Capital crunch eats away at IT. Modern Healthcare, 34, 32.

RFID in Hospitals and Factors Restricting Adoption

New Zealand Veterinary Association. (2002). Annual Report. Retrieved November 1, 2004, from www.vets.org.nz NZ Health Information Service. (2004). National Health Index FAQ. Retrieved August 31, 2004, from www.nzhis.govt.nz Pettigrew, A., Ferlie, E., & McKee, L. (1992). Shaping Strategic Change. London: Sage. Pink, B. (2004). District Health Board Deficit Decreases. Retrieved November 1, 2004, from www.stats.govt.nz Reddy, M., & Dourish, P. (2002). A Finger on the Pulse: Temporal Rhythms and Information Seeking in Medical Work. Paper presented at the ACM Conference on Computer-Supported Co-operative Work, New Orleans, Louisiana. Roberti, M. (2003). Singapore Fights SARS with RFID. Retrieved August 1, 2004, from www. rfidjournal.com Rogers, E. M. (1995). Diffusion of Innovations. New York: The Free Press. Rosenal, T., Paterson, R., Wakefield, S., Zuege, D., & Lloyd-Smith, G. (1995). Physician involvement in hospital information system selection: a success story. Paper presented at the Eighth World Conference on Medical Informatics, Vancouver, Canada. Ross, D. A., & Blasch, B. B. (2002). Development of a wearable computer orientation system. ACM Personal and Ubiquitous Computing, 6(1), 49–63. doi:10.1007/s007790200005 Schuerenberg, B. K. (2007). Bar Codes vs RFID: A Battle Just Beginning. Retrieved October 2, 2007, from www.healthdatamanagement.com Sokol, B., & Shah, S. (2004). RFID in Healthcare. Retrieved October 30, 2004, from www. rfidjournal.com

Stocking, B. (1985). Initiative and Inertia. London: Nuffield Provincial Hospital Trust. Stringleman, H. (2003). Electronic identification comes a step closer. Retrieved November 1, 2004, from www.country-wide.co.nz Swedberg, C. (2008). Italian Hospital Uses RFID to Document Patient Location, Treatment. Retrieved February 12, 2008, from www.rfidjournal.com Timing New Zealand. (2004). Winning Time Timing System. Retrieved November 1, 2004, from www.poprun.co.nz Walker, J., Spivey Overby, C., Mendelsohn, T., & Wilson, C. P. (2003). What You Need to Know About RFID in 2004. Retrieved February 9, 2004, from www.forrester.com Wallace, P. (2004). The health of nations. The Economist, 372, 1–18. Wasserman, E. (2007). A Healthy ROI. Retrieved October 12, 2007, from www.rfidjournal.com WAVE Advisory Board. (2001). From Strategy to Reality: The WAVE Project. Webster, C., Merry, A., Gander, P. H., & Mann, N. K. (2004). A prospective, randomised clinical evaluation of a new safety-orientated injectable drug administration system in comparison with conventional methods. Anaesthesia, 59, 80–87. doi:10.1111/j.1365-2044.2004.03457.x Wessel, R. (2007a). German Hospital Expands Bed-Tagging Project. Retrieved February 12, 2008, from www.rfidjournal.com Wessel, R. (2007b). RFID Synergy at a Netherlands Hospital. Retrieved October 31, 2007, from www.rfidjournal.com Wilson, R. M., Runciman, W. B., Gibberd, R. W., Harrison, B. T., Newby, L., & Hamilton, J. (1995). The quality in Australian health care study. The Medical Journal of Australia, 163, 458–471.

851

RFID in Hospitals and Factors Restricting Adoption

Wright, D. (2003). Business Case for Clinical Information System - Phase 1. Auckland: Waitemata District Health Board.

KEY TERMS AND DEFINITIONS Clinical HIS: IS that supports day-to-day clinical activity in radiology, pathology, pharmacy, and other specialist areas. Diffusion of Innovation: The process by which adoption of an innovation spreads through an organisation. Diffusion may be informally, through social peer networks, and/or formally, through organisational hierarchies. Electronic Patient Record (EPR): Also known as Electronic Health Record (EHR) or Electronic Medical Record (EMR). An electronic longitudinal collection of personal health information relating to an individual entered or accepted by health care providers, and organised primarily to support ongoing, efficient and effective health care.

Evidence Based Medicine (EBM): The use of current best evidence in making decisions about the care of patients. Evidence comes from both the individual physician’s clinical expertise, to accurately diagnose a patient, and external research, to identify the safest and most effective treatment for a given diagnosis. Operational Hospital Information System (HIS): IS that supports day-to-day non-clinical activity, such as accounting, payroll, inventory, and patient (customer) management. Patient Safety: The minimisation of unintended injury to a patient as a result of health care practices (as opposed to injury resulting from the patient’s underlying disease). Injury includes death, permanent and temporary disability, prolonged hospital stay, and/or financial loss to the patient. Strategic HIS: IS that supports analysis of day-to-day activity in order to identify potential improvements, such as traffic patterns in ED, equipment utilisation, and workflow.

This work was previously published in Auto-Identification and Ubiquitous Computing Applications, edited by Judith Symonds, John Ayoade and David Parry, pp. 91-118, copyright 2009 by Information Science Reference (an imprint of IGI Global).

852

853

Chapter 3.14

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease Pablo Arias University of A Coruña, Spain Nelson Espinosa University of A Coruña, Spain Javier Cudeiro University of A Coruña, Spain

ABSTRACT Transcranial Magnetic Stimulation (TMS) is a stimulation technique introduced to clinical practise by Anthony Baker in 1985. TMS has become very valuable either for neurophysiological examination as for research. When use in a repetitive way it has shown to have a therapeutic role for treatment of neurological and psychiatric disorders. This chapter summarizes the basic prinDOI: 10.4018/978-1-60960-561-2.ch314

ciples of the technique focusing on its interaction with the neural tissue along with the analysis of different stimulation protocols, potential risks, and its effectiveness on Parkinson’s disease.

INTRODUCTION TMS is based on the generation of magnetic fields to induce currents in the cortical tissue (or any excitable tissue) interfering, therefore, with the electrical activity of neurons. TMS was introduced

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

by Anthony Barker in 1985 as a painless and noninvasive technique that, based on the principles of electromagnetic induction, is able to generate currents in the brain and modulate the function of the cortex. The prototype presented by Barker was capable of generating current pulses of 110μsec at a frequency of 0.3Hz and an intensity of 4kA. This device (developed by Magstim and under an exclusive license to the University of Sheffield) was the first magnetic stimulator built for commercial purposes. Today there are new versions of magnetic stimulators manufactured by different companies specially designed for clinical and research applications and able to work with higher current and frequency ranges. The following is a review of the technique that includes the general aspects of the device, the physics behind TMS and its main applications in both, clinical and experimental protocols, focused on Parkinson’s disease.

Basic Setup of TMS: A TMS apparatus is composed by (see Figure 1): A: A capacitive systems of charge-discharge current designed to store an electric charge Figure 1. Transcranial magnetic stimulator Magstim model Rapid. A: two capacitive systems to accumulate electric charge above 2.8 kV. B: control system to adjust stimulation parameters and safeguards. C: circular coil and two figure-eight coils with different diameters, each of them induces a magnetic field with different characteristics

854

at high potential (in the kV range) and to supply a high current (in the kA range). B: A control system which allows to determine the current level, according to the user needs, and generates a magnetic stimulus either manually or by an external trigger circuit. In addition, it governs all electric and thermal protections for a safe operation. C: A stimulating coil whose design is crucial because it determines the profile of the induced magnetic field and because it is the only component in direct contact with the subject to study.

Operating Principles Whenever the capacitor bank is discharged by the action of the control system, a pulse of current I flows through the stimulating coil and, according to Ampère’s Law (eq. [1]), induces a magnetic  field H surrounding the coil. Figure 2A shows a simplified circuit equivalent to a figure of eight (or butterfly) coil (Figure 1C, center & right). Note that the induced magnetic field is greater (2H) where the more current flows.  Because I varies with time (pulse) so does H and, by virtue of the Faraday’s law (eq. [2] in Figure 3), an electric field is induced in the air Figure 2 A. a circulating current I induces a magnetic field H. The magnetic field is greater (2H) at points with greater current flow and is zero where the net current is zero. B: The right-hand rule is a mnemonic rule to determine the spatial relationship between current and magnetic field. The Ampère’s law (eq. [1]) expresses the relation between the current and the magnetic field

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Figure 3. Electric field A: 3D and B: 2D induced by a figure of eight coil (Fig 1C, right) at 100% of the capacity of the Magstim Rapid. The maximum value is located at the center of the coil (Jalinous, 1996). Equation [2] shows the relation between the electric field E and the magnetic field H. ∇× is the curl operator and 𝜇 is the permeability of the medium

(see Figure 3) and all tissues located on the coil vicinity. Then, if the coil is placed on the scalp, the magnetic field crosses the scalp, skull, cerebrospinal fluid and brain. As a result, the induced electric field gives rise to eddy currents whose magnitude depends on the conductivity of each medium (Wagner et al., 2004). TMS stimulus intensity can be inferred from the equation [2] in such a way that the bigger the slope of the current pulse, the higher the electric field intensity. Therefore, both the capacitive charge-discharge system design and the electrical parameters of the stimulation coil determinate the stimulus strength. The characteristics of the spatial distribution of the induced electric field (i.e. the precision and penetration depth of the stimulus) depend on the region where the eddy currents are induced (Starzynski et al., 2002, Miranda, Hallett and Basser, 2003, Roth et al., 1991, Nadeem et al., 2003) as well as on the geometric characteristics of the stimulation coil.

Coil Configurations and Companies Different coil shapes (see Figure 1C) are available from different companies and the choice depends on the objective of the study. “Figure of eight” or “round” coils are those more often used in clini-

cal practise, though some other shapes are also being developed. Circular coil: Circular coils (Figure 1C, left) are developed in different sizes and produce a wide electric field, allowing bihemispheric stimulation. Large 90mm mean diameter coil is most effective in stimulating the motor cortex and the spinal nerve roots. Using this kind of coil the stimulation is maximum under the winding of the coil, but tends to zero at the center. It is important to point out the role of current direction in the case of applying stimulation with a circular coil; given both sides of the coils have stimulating properties the appropriate direction must be chosen in order to preferentially activate each hemisphere. When the coil powered from a monophasic stimulator is placed centrally over the vertex of the skull and the current induced in the tissue flows clockwise, the stimulated hemisphere is chiefly the left one, and viceversa; given the motor cortex is more sensitive when current flows from posterior to anterior (Hovey and Janilous, 2008). Figure of eight coil: eight-shaped coils (see Figure 1C, center & right) are built from two windings place side by side, so that the induced tissue current reaches its maximum under the centre with a lower electric field in the remaining area, allowing a very good definition of the stimulated area. At this point, coil orientation critically influences the way stimulation interferes with the stimulated area, if a monophasic pulse is applied with this coil with latero-medial orientation corticospinal activity emerges from activation of either activation of the cortico-spinal tract directly, as well as from trans-synaptic activation of the piramidal cells, whereas stimulation with a posterior-anterior orientation elicits only the latter (Patton and Amassian, 1954; Kaneko et al., 1996; Nakamura et al., 1996; Di Lazzaro et al. 2004) Some other coil shapes: Apart from these classic coil shapes some other special configurations have been devised, allowing a more focused stimulation on the lower limb areas; some dedicated to

855

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

sham stimulation or a recently developed coil to deeper stimulation. Double Cone Coil: it has two large cup shaped windings side by side that allows a very good coupling to the skull, so that the induced current at the central fissure is very high, and the areas controlling lower limbs and torso are better stimulated (Hovey and Janilous, 2008). Sham Coil: it looks like a standard coil but just a very small portion of the energy is delivered to the head of the coil, since the coil is routed via a special box. Inside the box most of the energy is discharged but the characteristic ‘click’ is heard when the stimulator is fired (Hovey and Janilous, 2008). Sham coils have been a matter of debate, given a perfect sham coil should match a number of features to perfectly mimic real stimulation (i.e. identical coil positions on the scalp, acoustic artefacts, and identical sensations evoked on the scalp). Traditionally sham stimulation has been performed by tilting the coil 90º, however this technique is known not to fulfil the conditions mentioned beforehand. In addition some degree of real stimulation might also be present. To overcome the problems some works have proposed to use two superimposed coils (Okabe et al., 2003), and new sham coils are being developed such as the one composed by joining two coils, one real coil, and one sham coil, place upside-down so that the real coil provides some key elements to achieve realistic sham stimulation, like noise artefacts, and the sham coil some others, like coil’s position on the skull. Despite these steps forward the development of sham stimulators keeps on being a target given the devices developed so far do no match sideeffect scalp stimulation provided by real coils. Some researches (Okabe et al. 2003) have dealt with this issue by applying cutaneous electrical stimulation of the scalp. For instance, Rossi et al., (2007) have recently presented an adaptation of the traditional figure of eight coil which is built in compact wood. This is attached to the real

856

coil by Velcro strips, so that the first one acts as a physical spacer between the real coil and the scalp; also the latter houses a bipolar electrical stimulator within the surface which is in contact with the scalp to reproduce cutaneous sensations felt by the subject when is really stimulated. H-coil: recently a way to obtain deeper TMS stimulation of the brain has been proposed, it is based on a new configuration of the coil. This coil allows to generate a summation of the evoked electric fields in a given brain region by means of placing several coil elements around this area; all this elements induce a electric field in the same direction (Roth, et al., 2002; Zangen et al. 2005) Coil companies: From a theoretical point of view one coil-shape will generate a given magnetic field, and will induce a characteristic electric field. However based on everyday practice things are not so easy and a new source of variability appears given several companies have developed differently the same kind of coil. For instance, figure of eight coils have been developed with both windings in the same plane, side by side, or in different plane and overlapping. The difference in configuration has been proved to have different neurophysiological effects when generating a pulse to interact with the cortex. Figure 4 illustrates how the output intensity from the stimulator at which motor threshold is determine varies depending on the coil manufacturer, regardless pulse waveform or the current direction (Kammer et al., 2001). This is a critical point to take into account when interpreting data, given motor thresholds are usually the key point in order to determine stimulation intensity in experimental or clinical procedures.

STIMULATION PARAMETERS Pulse Waveform Pulse waveform: this feature is also a basic element to take into account during TMS. According

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Figure 4. Variation in motor threshold in 8 subjects expressed as percentage of the stimulator as a function of the manufacturer, the pulse waveform and the current direction (anterior-posterior or postero-anterior) (modified from Kammer et al., 2001)

to the technical specifications of the equipments, single pulse may be monophasic, biphasic and polyphasic. Each type leads to different neurophysiological interactions with the tissue, and leads to different applications (Di Lazzaro et al., 2001). Monophasic pulse is more accurate, and produces lower noise and heat, but bilateral responses are not easy achievable (Hovey and Janilous, 2008). Neurophysilogical studies demonstrated that monophasic and biphasic pulses require different output intensities in order to determinate the motor threshold (MT), either active (AMT), or at rest (RMT) (Sommer et al., 2006).

Stimulation Frequency Depending on the frequency of the pulses there are three possible stimulation protocols: •

Simple stimulation: TMS is applied with a frequency below 1Hz (Barker, 2002). This is suitable for studies of peripheral nerves, being able to depolarize neurons in the motor cortex and to induce a motor evoked potential (MEP) that can be measured in a given muscle of the contralateral hemibody (usually in the hand). This type of stimulation is applied, for example, to evaluate the

•

•

motor conduction time, which is altered in cases of multiple sclerosis or amyotrophic lateral sclerosis (de Noordhout et al., 1998). Paired pulse stimulation: It consist of two TMS pulses applied in the same or different brain areas. Pulses are usually of different intensities and presented within an interval of milliseconds (Kujirai et al., 1993). When applied to motor cortex and varying the interval or intensity of the first pulse, the response to the second pulse can be inhibited or facilitated (Ziemann et al., 1998, Di Lazzaro et al., 1999, Ilic et al., 2002). These two phenomena are known respectively as intracortical inhibition and facilitation since, supposedly, it would be mediated by intracortical connections in pyramidal neurons, being inhibitory or excitatory depending on the interval between stimuli (Kujirai et al., 1993, Ziemann, Rothwell and Ridding, 1996). For other applications, further studies have proposed a range of protocols, varying the number of pulses, intensity and the interval between stimuli (Ziemann et al., 1998, Bestmann et al., 2004, Tokimura et al., 1996). Repetitive stimulation: Repetitive TMS (rTMS) consists of pulse trains delivered with a maximum frequency of 50 Hz during tens, hundreds or thousands of milliseconds. It is useful to modify the excitability of brain areas at the site of stimulation or in other cortical areas due to cortico-cortical connections (Jalinous, 1991), lasting even several minutes after the end of the rTMS protocol (Robertson, Theoret and PascualLeone, 2003), “off-line effect” or “after effect” (Siebner and Rothwell, 2003)). Depending on the frequency of stimulation, rTMS have been classified as of low and high frequency, being the separation between types around 5 Hz. This division is not random and is due mainly to results

857

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Other Parameters

obtained in experiments devoted to analyze this frequency range. They concluded that frequencies above 5 Hz tended to increase corticospinal excitability (Berardelli et al., 1998, Maeda et al., 2000a, Pascual-Leone et al., 1994, Pascual-Leone et al., 1998). In contrast, frequencies below 5 Hz tended to depress the excitability (Pascual-Leone et al., 1998, Chen et al., 1997, Muellbacher et al., 2000). One special type of repetitive stimulation is called “theta burst” and consists of trains of pulses of short duration applied with a repetition rate of 5 Hz (it has also been tested in humans with 10 and 20 Hz, see De Ridder et al., 2007). Each train consists normally of 3 to 5 pulses with an intra-train frequency of 100 Hz; although human safety restrictions (Wassermann, 1998, Rossi et al. 2009, Table 1) limits the stimulation trains to 3 pulses with an intra-train frequency of 50 Hz (Huang and Rothwell, 2004). Its effect can be controlled, being facilitator in the “intermittent mode” (using groups of trains of 2 sec duration every 10 sec) or inhibitor in the “continuous mode” (Huang et al., 2005, Di Lazzaro et al., 2005). Its main advantage is that to achieve a specific period of delayed effect it requires a comparatively smaller period of stimulation than conventional rTMS (Huang and Rothwell, 2004).

Other parameters to be determined are the current that flows through the stimulation coil (i.e. the magnitude of the magnetic field, see eq. [1]) and the period of stimulation. Several studies have shown that frequency, intensity and duration are closely dependent variables. For example, a facilitatory effect similar to that obtained using 10 TMS pulses at 20 Hz and 150% of motor threshold (Pascual-Leone et al., 1994) is obtained by applying a train of 240 pulses at the same frequency but with an intensity of 90% (Maeda et al., 2000a, Maeda et al., 2000b). On the other hand, interdependence between frequency and duration of the stimulus is revealed in a second study devoted to analyze visual evoked potentials after rTMS applied on occipital cortex of the cat at low and high frequencies during periods ranging from 1 to 20 minutes (Aydin-Abidin et al., 2006). Trains of low frequency (1 to 3 Hz) require a long period of application (20 min) to significantly decrease the amplitude of the evoked potential; in contrast, high frequency trains (10 Hz) require less time (1 to 5 min) to increase the amplitude of evoked potentials and appear to be more effective than longer trains (20 min).

Table 1. Maximal rTMS train duration (secs) as a function of stimulation frequency to be bound to safety guidelines. Values have been calculated from data obtained from motor cortex stimulation; therefore it is worth remarking cortical excitability varies from area to area (Wassermann, 1998) Frequency (Hz)

Stimulation Intensity (% RMT)

1

>1800

>1800

360

>50

>50

>50

>50

27

11

11

8

7

6

5

>10

>10

>10

>10

7.6

5.2

3.6

2.6

2.4

1.6

1.4

1.6

1.2

10

>5

>5

4.2

2.9

1.3

0.8

0.9

0.8

0.5

0.6

0.4

0.3

0.3

20

2.05

1.6

1.0

0.55

0.35

0.25

0.25

0.15

0.2

0.25

0.2

0.1

0.1

25

1.28

0.84

0.4

0.24

0.2

0.24

0.2

0.12

0.08

0.12

0.12

0.08

0.08

858

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

CLINICAL APPLICATIONS OF TMS The use of TMS as a tool for clinical diagnosis is widely recognized. It has become a regular technique in neurology and neurophysiology, with application in the study of central conduction in multiple sclerosis, diagnosis of psychogenic disorders, study of facial paralysis, diagnosis and monitoring of spine injury, or as an operating room monitoring system. Likewise its application for basic research has grown exponentially in the last decade, given it is a suitable tool for exploring visual system physiology, language, learning, memory, and of course, motor system, for which it was originally developed. However, the effectiveness of TMS as a therapeutic tool is an issue of debate. Some researchers have stated that no treatment based on rTMS has consistently improved any disorder in the last decade. Despite, it is clear the technique is very attractive, given the possibility of stimulating cerebral circuitry through a non-invasive way with proved effect on functional re-organization, which can become of clinical utility. The technique has been used in several neurological and psychiatric diseases, like depression, stroke, migraine, or tinnitus. Results have been variable. Part of scepticism around the therapeutic effectiveness of TMS might be due to a wrong choice of the disorders to be treated; probably these are still to be determined. Wassermann and Lisanby (2001) reviewed the therapeutic use of rTMS, Pascual-Leone et al. (2002) and Hallet and Chokroverty (2005) have edited excellent monographs also. Some disorders have been the subject of research applying rTMS at low frequency with the aim of depressing neuronal activity of abnormally hyperexcitable cerebral circuits. This is the case, for instance, of focal dystonia, epilepsy, or auditory hallucinations present in schizophrenia, which might be good targets to be treated with rTMS (Lefaucheur et al., 2004, Poulet et al., 2005; Allam et al., 2007; Joo et al., 2007). On the other hand,

in some other disorders which affect (diffusely) larger areas of the brain, the hope of succeeding should be more modest. An intermediate example is Parkinson’s disease, where histological and neurochemical deficits are subcortical (beyond the direct TMS pulses influence), but with concomitant cortical dysfunctions well documented. Some of the affected cortical areas are the dorsolateral prefrontal cortex, the supplementary motor area, the lateral premotor cortex, the parietal cortex, the sensorimotor cortex, and also the motor cortex. Therefore, it seems conceivable that some of these areas can represent a target for rTMS application with the aim to relieve parkinsonian symptoms.

TMS over the Motor Cortex (Area 4) This area was the first target for TMS application in PD and some other movement disorders. Several of the parameters that can be evaluated with this technique are impaired in PD: •

•

Motor threshold: The minimum stimulation intensity that can produce a motor output of a given amplitude from a muscle at rest (RMT) or during a muscle contraction (AMT). It reflects excitability within the corticospinal system (Rossini et al. 1994). It is usually registered by electromyography recordings of the Motor Evoked Potential (MEP). Some works report normal threshold in PD; though some other reports RMT to be reduced in PD (Cantello et al. 2002). MEP amplitude: size (peak-to-peak) of MEP which increases in direct proportion to stimulus intensity. Most of the published works have reported that MEP is enlarged at rest; this may reflect an excitation-inhibition unbalance in the corticospinal system and desinhibition would be increased (Cantello et al. 2002).

859

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Figure 5. Cortical silent period obtained after stimulation of M1. After the MEP, a period without electrical activity is observed

• •

•

860

Cortical silent period: this parameter has become a key element of the use of TMS in PD. If a TMS pulse is delivered over the motor cortex whilst a muscle is voluntary contracted, a period of silent EMG activity is displayed following the muscular evoked potential (Figure 5) (Cantello et al. 2002). Its duration reflects the functionality of the cortical inhibitory mechanisms; it is shortened in PD, and increased by antiparkinsonian drugs. Neurophysiological evaluation of cortical physiology by mean of paired pulse stimulation: Stimulation with two distinct stimuli through the same coil over the cerebral cortex at a range of different intervals; the intensities can be varied independently. In this technique, a subthreshold conditioning stimulus (CS) is applied prior to a suprathreshold test stimulus (TS). The effect of the conditioning stimulus depends critically on the interstimulus interval (ISI) between CS and TS but also on CS strength. The motor-evoked response (MEP) elicited by the test stimulus is reduced if CS precedes TS by approximately 1–4 ms, while facilitation of the MEP results in ISIs of 7–20 ms (Ziemann et al., 1996). This phenomenon is commonly referred to as intracortical inhibition (and facilitation, since it is thought that either inhibitory or excitatory intracortical connections to the pyramidal tract neurones are activated in an

ISI-dependent way. More recently, this kind of inhibition has been termed shortinterval intracortical inhibition to distinguish it from another kind of cortically mediated inhibition observed with longer (50–200 ms) interstimulus intervals (VallsSolé et al., 1992) both altered in PD (Cantello et al. 2002). Repetitive TMS (rTMS) in PD: Alvaro Pascual Leone and collaborators in 1994 were the first to describe the effects of rTMS on PD signs. They showed how stimulation the motor cortex at 5 Hz, with an intensity of 90% of the RMT sped up movements in a hand dexterity test (Grooved Pegboard). However, other authors using similar protocols have not been able to reproduce those results (Ghabra et al. 1999).

Some other stimulation paradigms, more complex from a methodological point of view, have yielded promising results. In a placebo controlled trial by Siebner et al., (1999), rTMS with an intensity of 90% RMT resulted in significant improvement of hand movements. Stimulation was applied by means of 15 trains of 30 seconds each at 5 Hz, for a total of 2250 pulses over the motor cortex contralateral to the evaluated hand. In the task, patients had to perform balistic movements to a target in front of them. Movement time, accuracy, and velocity variability improved after real stimulation, but not in the case of those patients receiving sham stimulation. Though the experiment was carefully designed, it has been criticised given real stimulation facilitates ongoing movement, a fact that has not been achieved in the case of sham stimulation. Based on studies showing long lasting effects on parkinsonian sings obtained with just one session of rTMS, it seems conceivable to look for cumulative effects after different sessions. Recently, data obtained in a well controlled study have shown an improvement in clinical signs in PD

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

after several sessions of rTMS at high frequency applied bilaterally over the motor cortex. The effects lasted for several days (Kherd et al., 2006).

Stimulation over Some Other Areas Involved in Motor Control Some other areas than the motor cortex have been also stimulated in order to ameliorate motor and non-motor sings in PD. •

•

Supplementary motor area (SMA): This area is involved in sequencing complex movements along with motor planning. When SMA was stimulated at 5 Hz (intensity 110% AMT) delivered by a figure of eight coil with a slight improvement in clinical signs was found (Hamada et al., 2008). Dorsolateral prefrontal cortex (DLPFC): This region mediates cognitive processes, like decision making or working memory. It is also related to movement given it connection to the basal ganglia-SMA loop. DLPFC has become a target for stimulation in PD given that stimulation of this area induces dopamine release in the caudate nucleus. Also some reports have shown improvement in symptoms of depression, which are common in PD. Although early reports indicate that motor sings might be improved after stimulation of DLPFC, a later placebo controlled study has pointed out that placebo effect could be responsible for the observed improvement in several motor task (arm movements, grip strength, and gait) after DLPFC stimulation of the contralateral hemisphere for several days (del Olmo et al. 2007).

DLPFC and Motor Cortex rTMS stimulation delivered over these areas may be beneficial and with long-lasting effects.

Stimulation (25 Hz at 100% RMT) was applied twice per week, for 4 weeks, resulting in a significant improvement in gait, and bradykinesia in the PD receiving real stimulation, but not in the sham group. This was a double-blind study with a positive outcome which reinforces the view that TMS might be a promising tool to treat PD (Lomarev et al. 2006).

Cerebellum Recently the use of rTMS has been explored with promising results in order to control some side effects of antiparkinsonian medication. Kock et al., 2009 have shown how rTMS applied over the cerebellum under continuous theta-burst protocol reduced levodopa induced diskynesias in the patients, as well as modified some altered cortical excitability parameters in the PD.

CONCLUSION Any kind of therapeutic approach must fulfil at least three main requirements in order to be consider useful from a clinical perspective: (i) the effect must last, hours or days, (ii) the improvement must be measurable and, finally (iii) benefit/risk ratio must be clearly positive. rTMS (even with the existing discrepancies) seems to match all requirements to be taken into account as a therapeutic tool in order to treat PD, the topic keeps on being matter of research. Some reports indicate improvements in clinical sings lasting several days after the last rTMS session with minor side effects, which are chiefly restricted to headache just after stimulation, which lasts a few minutes and appears seldom. In spite of the promising results, some questions remain open. What is it mechanism of action? What is the best protocol of stimulation? Is it suitable for all PD or is it phenotype dependent? Is the effect observed really different to that obtained by placebo? So far, the main conclusions to be drawn

861

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

are: the technique allows a better knowledge of the disease from a neurophysiological point of view and it gives hope to find a new way to ameliorate parkinsonian symptoms.

REFERENCES Barker, A. T. (1999). The history and basic principles of magnetic nerve stimulation. Electroencephalography and Clinical Neurophysiology. Supplement, 51, 3–21. Bartres-Faz, D., Junque, C., Tormos, J. M., & Pascual-Leone, A. (2000). Aplicación de la estimulación magnética transcraneal a la investigación neuropsicológica. Revista de Neurologia, 30(12), 1169–1174. Boylan, L. S., Pullman, S. L., Lisanby, S. H., Spicknall, K. E., & Sackeim, H. A. (2001). Repetitive transcranial magnetic stimulation to SMA worsens complex movements in Parkinson’s disease. Clinical Neurophysiology, 112(2), 259–264. doi:10.1016/S1388-2457(00)00519-8 Calvo-Merino, B., & Haggard, P. (2004). Estimulación magnética transcraneal. Aplicaciones en neurociencia cognitiva. Revista de Neurologia, 38(4), 374–380. Cantello, R., Tarletti, R., & Civardi, C. (2002). Transcranial magnetic stimulation and Parkinson’s disease. Brain Research. Brain Research Reviews, 38(3), 309–327. doi:10.1016/S01650173(01)00158-8 Del Olmo, M. F., Bello, O., & Cudeiro, J. (2007). Transcranial magnetic stimulation over dorsolateral prefrontal cortex in Parkinson’s disease. Clinical Neurophysiology, 118(1), 131–139. doi:10.1016/j.clinph.2006.09.002

862

Di Lazzaro, V., Oliviero, A., Mazzone, P., Insola, A., Pilato, F., & Saturno, E. (2001). Comparison of descending volleys evoked by monophasic and biphasic magnetic stimulation of the motor cortex in conscious humans. Experimental Brain Research, 141(1), 121–127. doi:10.1007/ s002210100863 Di Lazzaro, V., Oliviero, A., Pilato, F., Saturno, E., Dileone, M., & Mazzone, P. (2004). The physiological basis of transcranial motor cortex stimulation in conscious humans. Clinical Neurophysiology, 115, 255–266. doi:10.1016/j. clinph.2003.10.009 Espinosa, N., de Labra, C., Rivadulla, C., Mariño, J., Grieve, K. L., & Cudeiro, J. (2007). Effects on EEG of Low (1Hz) and High (15Hz) Frequency Repetitive Transcranial Magnetic Stimulation of the Visual Cortex: A Study in the Anesthetized Cat. The Open Neurosci Journal, 1, 20–25. Fitzgerald, P. B., Fountain, S., & Daskalakis, Z. J. (2006). A comprehensive review of the effects of rTMS on motor cortical excitability and inhibition. Clinical Neurophysiology, 117(12), 2584–2596. doi:10.1016/j.clinph.2006.06.712 Fregni, F., Simon, D. K., Wu, A., & PascualLeone, A. (2005). Non-invasive brain stimulation for Parkinson’s disease: a systematic review and meta-analysis of the literature. Journal of Neurology, Neurosurgery, and Psychiatry, 76(12), 1614–1623. doi:10.1136/jnnp.2005.069849 Ghabra, M. B., Hallett, M., & Wassermann, E. M. (1999). Simultaneous repetitive transcranial magnetic stimulation does not speed fine movement in PD. Neurology, 52(4), 768–770. Hamada, M., Ugawa, Y., & Tsuji, S. (2008). Effectiveness of rTMS on Parkinson’s Disease Study Group, Japan. High-frequency rTMS over the supplementary motor area for treatment of Parkinson’s disease. Movement Disorders, 23(11), 1524–1531. doi:10.1002/mds.22168

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Hovey, C., & Janilous, R. (2008). The guide to magnetic stimulation. The Magstims Company UK. Jalinous, R. (1991). Technical and practical aspects of magnetic nerve stimulation. Journal of Clinical Neurophysiology, 8(1), 10–25. doi:10.1097/00004691-199101000-00004 Joo, E. Y., Han, S. J., Chung, S. H., Cho, J. W., Seo, D. W., & Hong, S. B. (2007). Antiepileptic effects of low-frequency repetitive transcranial magnetic stimulation by different stimulation durations and locations. Clinical Neurophysiology, 118(3), 702–708. doi:10.1016/j.clinph.2006.11.008 Kammer, T., Beck, S., Thielscher, A., Laubis-Herrmann, U., & Topka, H. (2001). Motor thresholds in humans: a transcranial magnetic stimulation study comparing different pulse waveforms, current directions and stimulator types. Clinical Neurophysiology, 112(2), 250–258. doi:10.1016/ S1388-2457(00)00513-7 Kaneko, K., Kawai, S., Fuchigami, Y., Morita, H., & Ofuji, A. (1996). The effect of current direction induced by transcranial magnetic stimulation on the corticospinal excitability in human brain. Electroencephalography and Clinical Neurophysiology, 101, 478–482. doi:10.1016/ S0013-4694(96)96021-X

Koch, G., Brusa, L., Carrillo, F., Lo Gerfo, E., Torriero, S., & Oliveri, M. (2009). Cerebellar magnetic stimulation decreases levodopa-induced dyskinesias in Parkinson disease. Neurology, 73(2), 113–119. doi:10.1212/WNL.0b013e3181ad5387 Lefaucheur, J. P., Fénelon, G., Ménard-Lefaucheur, I., Wendling, S., & Nguyen, J. P. (2004). Low-frequency repetitive TMS of premotor cortex can reduce painful axial spasms in generalized secondary dystonia: a pilot study of three patients. Neurophysiologie Clinique, 34(3-4), 141–145. doi:10.1016/j.neucli.2004.07.003 Lomarev, M. P., Kanchana, S., Bara-Jimenez, W., Iyer, M., Wassermann, E. M., & Hallett, M. (2006). Placebo controlled study of rTMS for the treatment of Parkinson’s disease. Movement Disorders, 21(3), 325–331. doi:10.1002/mds.20713 Machii, K., Cohen, D., Ramos-Estebanez, C., & Pascual-Leone, A. (2006). Safety of rTMS to non-motor cortical areas in healthy participants and patients. Clinical Neurophysiology, 117(2), 455–471. doi:10.1016/j.clinph.2005.10.014 Mally, J., & Stone, T. W. (1999). Improvement in Parkinsonian symptoms after repetitive transcranial magnetic stimulation. Journal of the Neurological Sciences, 162(2), 179–184. doi:10.1016/ S0022-510X(98)00318-9

Khedr, E. M., Rothwell, J. C., Shawky, O. A., Ahmed, M. A., & Hamdy, A. (2006). Effect of daily repetitive transcranial magnetic stimulation on motor performance in Parkinson’s disease. Movement Disorders, 21(12), 2201–2205. doi:10.1002/mds.21089

Nakamura, H., Kitagawa, H., Kawaguchi, Y., & Tsuji, H. (1996). Direct and indirect activation of human corticospinal neurons by transcranial magnetic and electrical stimulation. Neuroscience Letters, 210, 45–48. doi:10.1016/03043940(96)12659-8

Kobayashi, M., & Pascual-Leone, A. (2003). Transcranial magnetic stimulation in neurology. The Lancet Neurology, 2(3), 145–156. doi:10.1016/ S1474-4422(03)00321-1

Pascual-Leone, A., Valls-Sole, J., Brasil-Neto, J. P., Cammarota, A., Grafman, J., & Hallett, M. (1994). Akinesia in Parkinson’s disease. II. Effects of subthreshold repetitive transcranial motor cortex stimulation. Neurology, 44(2), 892–898.

863

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Patton, H. D., & Amassian, V. E. (1954). Single and multiple-unit analysis of cortical stage of pyramidal tract activation. Journal of Neurophysiology, 17, 345–363. Poulet, E., Brunelin, J., Bediou, B., Bation, R., Forgeard, L., & Dalery, J. (2005). Slow transcranial magnetic stimulation can rapidly reduce resistant auditory hallucinations in schizophrenia. Biological Psychiatry, 57(2), 188–191. doi:10.1016/j. biopsych.2004.10.007 Ridding, M. C., & Rothwell, J. C. (2007). Is there a future for therapeutic use of transcranial magnetic stimulation? Nature Reviews. Neuroscience, 8(7), 559–567. doi:10.1038/nrn2169 Rossi, S., Ferro, M., Cincotta, M., Ulivelli, M., Bartalini, S., & Miniussi, C. (2007). A real electro-magnetic placebo (REMP) device for sham transcranial magnetic stimulation (TMS). Clinical Neurophysiology, 118(3), 709–716. doi:10.1016/j. clinph.2006.11.005 Rossi, S., Hallett, M., Rossini, P. M., & PascualLeone, A. (2009). The Safety of TMS Consensus Group. Safety, ethical considerations, and application guidelines for the use of transcranial magnetic stimulation in clinical practice and research. Clinical Neurophysiology, 120(12), 2008–2039. doi:10.1016/j.clinph.2009.08.016 Rossini, P. M., Barker, A. T., Berardelli, A., Caramia, M. D., Caruso, G., & Cracco, R. Q. (1994). Non-invasive electrical and magnetic stimulation of the brain, spinal cord and roots: basic principles and procedures for routine clinical application. Report of an IFCN committee. Electroencephalography and Clinical Neurophysiology, 91(2), 79–92. doi:10.1016/0013-4694(94)90029-9 Roth, Y., Amir, A., Levkovitz, Y., & Zangen, A. (2007). Three-dimensional distribution of the electric field induced in the brain by transcranial magnetic stimulation using figure-8 and deep Hcoils. Journal of Clinical Neurophysiology, 24(1), 31–38. doi:10.1097/WNP.0b013e31802fa393

864

Rothwell, J. C. (1997). Techniques and mechanism of action of transcranial stimulation of the human motor cortex. Journal of Neuroscience Methods, 74(2), 113–122. doi:10.1016/S01650270(97)02242-5 Rothwell, J. C., Thompsom, P. D., Day, B. L., Boyd, S., & Marsden, C. D. (1991). Stimulation of the human motor cortex through the scalp. Experimental Physiology, 76(2), 159–200. Shimamoto, H., Morimitsu, H., Sugita, S., Nakahara, K., & Shigemori, M. (1999). Therapeutic effect of repetitive transcranial magnetic stimulation in Parkinson’s disease. Rinsho Shinkeigaku. Clinical Neurology, 39(12), 1264–1267. Siebner, H. R., Mentschel, C., Auer, C., & Conrad, B. (1999). Repetitive transcranial magnetic stimulation has a beneficial effect on bradykinesia in Parkinson’s disease. Neuroreport, 10(3), 589–594. doi:10.1097/00001756-199902250-00027 Siebner, H. R., Rossmeier, C., Mentschel, C., Peinemann, A., & Conrad, B. (2000). Short-term motor improvement after sub-threshold 5-Hz repetitive transcranial magnetic stimulation of the primary motor hand area in Parkinson’s disease. Journal of the Neurological Sciences, 178(2), 91–94. doi:10.1016/S0022-510X(00)00370-1 Sommer, M., Alfaro, A., Rummel, M., Speck, S., Lang, N., Tings, T., & Paulus, W. (2006). Half sine, monophasic and biphasic transcranial magnetic stimulation of the human motor cortex. Clinical Neurophysiology, 117(4), 838–844. doi:10.1016/j. clinph.2005.10.029 Tormos, J. M., Catalá, M. D., & Pascual-Leone, A. (1999). Estimulación magnética transcraneal. Revista de Neurologia, 29(2), 165–171. Valls-Solé, J., Pascual-Leone, A., Wassermann, E. M., & Hallett, M. (1992). Human motor evoked responses to paired transcranial magnetic stimuli. Electroencephalography and Clinical Neurophysiology, 85(6), 355–364. doi:10.1016/01685597(92)90048-G

Transcranial Magnetic Stimulation (TMS) as a Tool for Neurorehabilitation in Parkinson’s Disease

Wassermann, E. M. (1998). Risk and safety of repetitive transcranial magnetic stimulation: report and suggested guidelines from the international Workshop on the Safety of Repetitive Transcranial Magnetic Stimulation. Electroencephalogr Clin Neurophysiol, 108(1), 1-16. Wassermann, E. M., & Lisanby, S. H. (2001). Therapeutic application of repetitive transcranial magnetic stimulation: A review. Clinical Neurophysiology, 112(8), 1367–1377. doi:10.1016/ S1388-2457(01)00585-5 Zangen, A., Roth, Y., Voller, B., & Hallett, M. (2005). Transcranial magnetic stimulation of deep brain regions: evidence for efficacy of the H-coil. Clinical Neurophysiology, 116(4), 775–779. doi:10.1016/j.clinph.2004.11.008 Ziemann, U., Rothwell, J. C., & Ridding, M. C. (1996). Interaction between intracortical inhibition and facilitation in human motor cortex. The Journal of Physiology, 496, 873–881. Zimmermann, K. P., & Simpson, R. K. (1996). ‘Slinky’ coils for neuromagnetic stimulation. Electroencephalography and Clinical Neurophysiology, 101(2), 145–152. doi:10.1016/0924980X(95)00227-C

KEY TERMS AND DEFINITIONS AMT: Motor threshold (MT) during a muscle contraction. Dorsolateral Prefrontal Cortex (DLPFC): Part of the cerebral cortex involved in cogni-

tive processes, like decision making or working memory. It is also related to movement given it connection to the basal ganglia-SMA loop. Interstimulus Interval (ISI): Time between consecutive stimulus. Motor Cortex (Area 4): Part of the cerebral cortex involved in controlling muscle contraction. Motor Evoked Potential (MEP): Electrical activity recorded from muscles following direct stimulation of motor cortex. Motor Threshold (MT): The minimum stimulation intensity that can produce a motor output of a given amplitude from a muscle at rest (RMT) or during a muscle contraction (AMT). PD: Parkinson’s disease. Repetitive TMS (rTMS): Modality of TMS in which pulse trains are delivered with a maximum frequency of 50 Hz during tens, hundreds or thousands of milliseconds. RMT: Motor threshold at rest. Subthreshold Conditioning Stimulus (CS): Stimulus with an intensity below the motor threshold. Supplementary Motor Area (SMA): In sequencing complex movements along with motor planning. Supra-Threshold Test Stimulus (TS): Stimulus with an intensity above the motor threshold. Transcranial Magnetic Stimulation (TMS): Stimulation technique based on the generation of magnetic fields to induce currents in the cortical tissue (or any excitable tissue) interfering, therefore, with the electrical activity of neurons.

This work was previously published in Handbook of Research on Personal Autonomy Technologies and Disability Informatics, edited by Javier Pereira, pp. 131-143, copyright 2011 by Medical Information Science Publishing (an imprint of IGI Global).

865

866

Chapter 3.15

Modeling of Porphyrin Metabolism with PyBioS Andriani Daskalaki Max Planck Institute for Molecular Genetics, Germany

ABSTRACT Photodynamic Therapy (PDT) involves administration of a photosensitizer (PS) either systemically or locally, followed by illumination of the lesion with visible light. PDT of cancer is now evolving from experimental treatment to a therapeutic alternative. Clinical results have shown that PDT is at least as efficacious as standard treatments of malignancies of the skin and Barrett’s esophagus. Hemes and heme proteins are vital components of essentially every cell in virtually all eukaryote organisms. Protoporphyrin IX (PpIX) is produced in cells via the heme synthesis pathway from the DOI: 10.4018/978-1-60960-561-2.ch315

substrate aminolevulinic acid (ALA). Exogenous administration of ALA induces accumulation of (PpIX), which can be used as a photosensitiser for tumor detection or photodynamic therapy. Although the basis of the selectivity of ALA-based PDT or photodiagnosis is not fully understood, it has sometimes been correlated with the metabolic rate of the cells, or with the differential enzyme expressions along the heme biosynthetic pathway in cancer cells. An in silico analysis by modeling may be performed in order to determine the functional roles of genes coding enzymes of the heme biosynthetic pathway like ferrochelatase. Modeling and simulation systems are a valuable tool for the understanding of complex biological systems. With PyBioS, an object-oriented mod-

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Modeling of Porphyrin Metabolism with PyBioS

elling software for biological processes, we can analyse porphyrin metabolism pathways.

INTRODUCTION The use of PDT for curative treatments of superficial tumors of the skin and for palliative treatments of disseminated tumors of skin and oral mucosa is well known (Daskalaki 2002). PDT also is efficacious as treatment of malignancies of Barrett’s oesophagus (Foroulis and Thorpe 2006). PDT is based on a photochemical process, where photosensitizers (PS) act cytotoxic by generation of 1O2 after laser irradiation. The use of fluorescence measurements as quantitative indicators for PpIX accumulation after exogenous ALA administration is suitable for differentiating neoplastic, necrotic and inflammatory tissues from normal tissues. The modulation of ALA-induced PpIX accumulation and expression will provide more diagnostic information and more accuracy for the diagnosis of unhealthy tissue, especially in border-line cases. The modulation of fluorescence characteristics of ALA-induced PpIX with NAD has been used for differentiation between fibroblast and fibrosarcoma (Ismail et al. 1997). The flow of substrates into the porphyrin pathway is controlled by tsynthesis of d–aminolevulinic acid (ALA), the first committed precursor in the porphyrin pathway. Although light is required to trigger ALA synthesis and differentiation of chloroplasts (Reinbothe and Reinbothe, 1996), a feedback inhibition of ALA synthesis by an end product of the porphyrin pathway is thought to be involved in the regulation of influx into the pathway (Wettstein et al., 1995; Reinbothe and Reinbothe, 1996). Both the nature of the product and the mechanism involved in effecting feedback inhibition remain unknown, probably because there have been no porphyrin pathway mutants identified so far that affect both chlorophyll and heme biosyntheses. Thus, modelling of porphyrin

pathway may fill this gap and allow researchers to address these questions. Downey (2002) tried to show how the porphyrin pathway may be an integral part of all disease processes through a model. Analytical techniques capable of measuring porphyrins in all cells are needed. Data gathered from plant and animal studies need to be adapted to humans where possible. An inexpensive, accurate and rapid analysis needs to be developed so porphyrins can be measured more routinely. The committed step for porphyrin synthesis is the formation of 5-aminolevulinate (ALA) by condensation of glycine (from the general amino acid pool) and succinyl-CoA (from the TCA cycle), in the mitochondrial matrix. This reaction is catalyzed by two different ALA synthases, one expressed ubiquitously (ALAS1) and the other one only expressed in erythroid precursors (ALAS2) (Ajioka, 2006). Heme inhibits the activity of ALA synthetase, the first and rate-limiting enzyme of the biosynthetic pathway, thereby preventing normal cells from drowning in excess production of its own porphyrins. This negative feedback control can be bypassed in certain malignant cells exposed to an excess amount of ALA, which is metabolised leading to overproduction of PpIX. Excess accumulation of PpIX occurs because of the enzyme configuration in malignant cells (Kirby 2001). The enzyme ferrochelatase (FECH) catalyzes insertion of an iron atom into PpIX forming heme which is not photoreactive. However, cancer cells have a relatively low activity of ferrochelatase which leads to an excess accumulation of PpIX (Schoenfeld 1988). Another factor leading to augmented PpIX synthesis is an increased activity of the rate-limiting enzyme porphobilinogen deaminase in various malignant tissues (Wilson 1991). Kemmner W et al. (2008) recently showed that in malignant tissue a transcriptional downregulation of FECH occurs causeing endogenous PpIX accumulation. Furthermore, accumulation of intracellular PpIX because of FECH small

867

Modeling of Porphyrin Metabolism with PyBioS

interfering RNA (siRNA) silencing provides a small-molecule-based approach to molecular imaging and molecular therapy. Kemmner W et al (2008) demonstrated accumulation of the heme precursor PpIX in gastrointestinal tumor tissues. To elucidate the mechanisms of PpIX accumulation expression of the relevant enzymes in the heme synthetic pathway has been studied. Kemmner W et al (2008) described a significant down-regulation of FECH mRNA expression in gastric, colonic, and rectal carcinomas. Accordingly, in an in vitro model of several carcinoma cell lines, ferrochelatase down-regulation and loss of enzymatic activity corresponded with an enhanced PpIX-dependent fluorescence. Silencing of FECH using siRNA technology led to a maximum 50-fold increased PpIX accumulation. Bhasin G et al. (1999) investigated the hypothesis that inhibition of ferrochelatase will cause in situ build up of high PpIX concentrations which may act as a putative agent for photodestruction of cancer cells. The parenteral administration of lead acetate, a known inhibitor of ferrochelatase, to mice bearing cutaneous tumors (papillomas and carcinomas) caused a six-fold enhancement in the concentration of PpIX in tumors within a period of one month. A significant reduction in tumor size was observed starting as early as day one following the treatment. 5-Aminolevulinate synthase (ALAS) is a mitochondrial enzyme that catalyzes the first step of the heme biosynthetic pathway. Mitochondrial import as well as synthesis of the nonspecific ALAS isoform (ALAS1) is regulated by heme through a feedback mechanism (Munakata 2004).

Mutations A deficiency of FECH activity underlies the excess accumulation of protoporphyrin that occurs in erythropoietic protoporphyria (EPP). In some patients, protoporphyrin accumulation causes liver damage that necessitates liver transplanta-

868

tion. Mutations of the codons for 2 of [2Fe-2S] cluster ligands in patients with EPP support the importance of the iron-sulfur center for proper functioning of mammalian FECH and, at least in humans, its absence has a direct clinical impact (Schneider-Yin X. et al. 2000). Fu-Ping Chen et al.(2002) studied patients who developed liver disease with mutations in the FECH gene. Recent attempts to increase the efficacy of ALA-mediated PDT include the use of iron chelators to decrease the amount of PPIX converted to heme by FECH in removing the free iron that is necessary for the enzyme to work (Curnow, 1998. Ferreira et al., 1999). X-linked sideroblastic anemias (XLSAs), a group of severe disorders in humans characterized by inadequate formation of heme in erythroblast mitochondria, are caused by mutations in the gene for erythroid eALAS, one of two human genes for ALAS (Astner et al., 2005). Cloning and expression of the defective gene for deltaaminolevulinate dehydratase (ALAD) from patients with ALAD deficiency porphyria (ADP) were performed by Akagi et al.(2000). Mitchel et al.(2001), studied Escherichia coli and human porphobilinogen synthase (PBGS) mutants.

MODELING OF PORPHYRIN METABOLISM Quantitative modeling studies of pathways have been successfully applied to understand complex cellular processes (Schoeberl 2002, Klipp 2005). Particular attention has been paid to the way, in which PpIX is distributed and accumulated in cells under the effect of ALA (Gaullier et al. 1995). For induction of a clinical effect it is important to recognize the kinetics of PpIX accumulation in cells, as influenced by the applied ALA dose. Cellular content of tis photosensitizer precursor should be optimal for induction of the photodestructive effect, following light exposure of targeted neoplas-

Modeling of Porphyrin Metabolism with PyBioS

tic lesions. The kinetics of PpIX formation under the effect of exogenous ALA is thought to result from circumvented bottle-neck linked to synthesis of endogenous ALA, the level of which remains under control of free heme (Kennedy and Pottier (1992). Considering that these problems may not only be of theoretical significance, but also have a practical value for establishing conditions of a photodynamic therapy,we have to define kinetics of PpIX accumulation in different cells under the effect of various concentrations of ALA.

Pathway Databases Pathway databases can act as a rich source for such graphs, because a reaction graph is simply a pathway. The reactome pathway database (Vastrik et al., 2007) has been used as a key starting point for kinetic modelling since it entails detailed models for reaction graphs, which describe the series of biochemical events involved in the models and their relationships. The graphs establish a framework for the models and suggest the kinetic coefficients that should be obtained experimentally. Kamburov et al. (2006) developed ConsensusPathDB (Figure 1), a database that helps users to summarize and verify pathway information and to

enrich a priori information in the process of model annotation. The database model allows integration of information on metabolic, signal transduction and gene regulatory networks. Cellular reaction networks are stored in a PostgreSQL database and can be accessed under http://pybios.molgen.mpg.de/CPDB. By forward modelling we integrate all interactive properties of molecular components to understand systems behavior (Westerhoff et al. 2008). The forward-modeling approach supports the formulation of hypothesezing for e.g. in silico knock-out experiments. Thus, to construct a model of the porphyrin metabolism pathway, one should consider one enzymatic or transport step at a time, should comb the literature for information about this enzyme, its cofactors and modulators, and should translate this information into a mathematical rate law which could be a Michaelis-Menten, among a wide variety of possibilities. The collection of all rate laws governs the dynamics of this model. Comparisons of model responses with biological observations support the validity of this appointed model or suggest adjustments in assumptions or parameter values. This forward process may lead to model representations of the pathway exhibiting the same features as reality, at least qualitatively, if not quantitatively.

Figure. 1. Schematic illustration of PpIX biosynthesis. One part of the synthesis is localized in mitochondrium the other part in cytoplasm. In this scheme not all biosynthesis steps are to be seen.

869

Modeling of Porphyrin Metabolism with PyBioS

The porphyrin metabolism model was assembled, simulated and analysed by PybioS. PyBioS is an object-oriented tool for modeling and simulation of cellular processes. This tool has been established for the modelling of biological processes using metabolic pathways from databases like KEGG and the Reactome database. Modeling and simulation techniques are valuable tools for the understanding of complex biological systems. The platform is implemented as a Python-product for the Zope web application server environment. PyBioS acts as a model repository and supports the generation of large models based on publicly available information like the metabolic data of the KEGG database. An ODE-system of this model may be generated automatically based on pre- or user-defined kinetic laws and used for subsequent simulation of time course series and further analyses of the dynamic behavior of the underlying system.

Modeling with PyBioS A model of a disease-relevant pathway, such as porphyrin metabolism, has been employed to study the relationship between basic enzymes and products in the biosynthetic pathway. Visualization of the porphyrin metabolism interaction network (Figure 2) was enabled by automatically generated graphs that include information about the objects, reactions and mass- and information-flow. The model includes a total of 16 reactions, and 42 objects. It is composed of an ordinary differential equations system with 14 state variables and 16 parameters. The law of mass-action has been applied to describe the rate of porphyrin metabolism.Time-dependent changes of the concentration of participating proteins and protein complexes are determined by a system of differential equations.

Figure 2. CPDB database. ConsensusPathDB assists the development, expansion and refinement of computational models of biological systems and the context-specific visualization of models provided in SBML.

870

Modeling of Porphyrin Metabolism with PyBioS

Figure 3. A part of the porphyrin metabolism pathway is illustrated as a network diagram in PyBioS. Catalysis of heme formation by the enzyme ferrochelatase is illustrated in the network graphic in PyBioS.

Mutations related to genes FECH and ALAS have been analyzed by simulating knockouts of these genes by using a mathematical model in order to study mutation effects in the concentration of heme.

CONCLUSION The modeling and simulation platform PyBioS has been used for the in silico analysis of porphyrin metabolism pathway. This model of a porphyrin metabolim pathway should be used for hypotheses generation by forward modeling. Also the model should be

Figure 4. Ferrochelatase (FECH) catalyzes the terminal step of the heme biosynthetic pathway. Graphical illustration of the time course of the PpIX concentration, heme and ferrochelatase (FECH). FECH catalyzes the production of heme by PpIX.

871

Modeling of Porphyrin Metabolism with PyBioS

Table 1. Parameters in the model Parameter symbol

Biological meaning

massi_k (FECH homodimer)

Heme production rate constant

deg_degradation_k (heme [mitochondrial matrix])

Heme degradation rate constant

massi_k (ALAS homodimer)

5-aminolevulinate production rate constant

massi_k (PPO homodimer (FAD cofactor)

Protoporphyrin IX production rate constant

massi_k (CPO homodimer)

Protoporphyrinogen IX production rate constant

ComplexFormationReversible_kf (ALAD homooctamer)

Formation of complex rate constant

massi_k (ALAD homooctamer (Zinc cofactor)

Porphobilinogen production rate constant

massi_k (Porphobilinogen deaminase)

Hydroxymethylbilane (HMB) production rate constant

massi_k (UROD homodimer)

Coproporphyrinogen III production rate constant

Degradation_degradation_k (ALAD homooctamer (Pb and Zn bound)

Degradation rate constant

Degradation_degradation_k (Coproporphyrinogen I

Degradation rate constant

massi_k (Uroporphyrinogen-III synthase)

Uroporphyrinogen III production rate constant

massi_k (Protoporphyrin IX)

Export rate constant for Protoporphyrin IX

massi_k (Coproporphyrinogen)

Export rate constant for Coproporphyrinogen III

massi_k (ALAD)

Formation of ALA Dehydratase inactive complex

massi_k Hydroxymethylbilane

Dissociation rate constant for Hydroxymethylbilane

Table 2. State variables (proteins) in the model Parameter symbol

Biological meaning

5-aminolevulinate [cytosol]

5-aminolevulinate in the cytosol

5-aminolevulinate [mitochondrial matrix]

5-aminolevulinate in the mitochondrial matrix

ALAD homooctamer (Pb and Zn bound) [cytosol]

ALA Dehydratase inactive complex in the cytosol

ALAD homooctamer (Zinc cofactor) [cytosol]

ALA Dehydratase in the cytosol

ALAS homodimer [mitochondrial matrix]

ALA Synthetase in the mitochondrial matrix

Coproporphyrinogen I [cytosol]

Coproporphyrinogen I in the cytosol

Coproporphyrinogen III [cytosol]

Coproporphyrinogen III in the cytosol

Coproporphyrinogen III [mitochondrial intermembrane space]

Coproporphyrinogen III in the mitochondrial intermembrane space

CPO homodimer [mitochondrial intermembrane space]

Coproporphyrinogen oxidase in the mitochondrial intermembrane space

FECH homodimer (2Fe-2S cluster) [mitochondrial matrix]

Ferrochelatase in the mitochondrial matrix

heme [mitochondrial matrix]

Heme in the mitochondrial matrix

PPO homodimer (FAD cofactor) [mitochondrial intermembrane space]

Protoporphyrinogen oxidase in the mitochondrial intermembrane space

UROD homodimer [cytosol]

Uroporphyrinogen III-Decarboxylase in the cytosol

Uroporphyrinogen-III synthase [cytosol]

Uroporphyrinogen-III synthase in the cytosol

872

Modeling of Porphyrin Metabolism with PyBioS

Figure 5. A diagram summarizing the heme production. Illustration of (A) heme production and (B) PPIX time course in case of mutation of ferrochelatase. FECH inhibition is indicated by a blunted line. The simulation analysis of the model indicates that FECH inhibition caused a (B) decrease of heme.

Figure 6. Illustration of 5-aminolevulinate production time course of ALAS. ALAS inhibition is indicated by a blunted line. The simulation analysis of the model indicates that ALAS inhibition because of ALAS mutation caused a decrease of 5-aminolevulinate.

873

Modeling of Porphyrin Metabolism with PyBioS

disturbed to test tinfluences of gene knock-outs, mutations and performance of this model system. Knock-out experiments can be performed in order to determine the functional roles of genes coding enzymes of the heme biosynthetic pathway like ferrochelatase by studying the defects caused by the resulting mutation. A next step should be integration of experimental data into the kinetic model of this pathway. The results of the in silico experiments have to be compared with the experimental data to decide, which kind of perturbation caused the phenotype of the investigated system. Thus, we should be able to test mutations of enzymes playing an important role in the heme biosynthetic pathway.

REFERENCES Ajioka, R. S., Phillips, J. D., & Kushner, J. P. (2006). Biosynthesis of heme in mammals, biochimica et biophysica acta (BBA). Molecular Cell Research, 1763(7), 723–736. Akagi, R., Shimizu, R., Furuyama, K., Doss, M. O., & Sassa, S. (2000, March). Novel molecular defects of the delta-aminolevulinate dehydratase gene in a patient with inherited acute hepatic porphyria. Hepatology (Baltimore, Md.), 31(3), 704–708. doi:10.1002/hep.510310321 Astner, I., Schulze, J. O., van den Heuvel, J., Jahn, D., Schubert, W.-D., & Heinz, D. W. (2005). Crystal structure of 5-aminolevulinate synthase, the first enzyme of heme biosynthesis, and its link to XLSA in humans. The EMBO Journal, 24, 3166–3177. doi:10.1038/sj.emboj.7600792 Bhasin, G., Kausar, H., & Athar, M. (1999, November). December). Ferrochelatase, a novel target for photodynamic therapy of cancer. Oncology Reports, 6(6), 1439–1442.

874

Chen, F.-P., Risheg, H., Liu, Y., & Bloomer, J. (2002, February). Ferrochelatase gene mutations in erythropoietic protoporphyria: Focus on liver disease. Cell Mol Biol (Noisy-le-grand), 48(1), 83–89. Daskalaki, A. (2002). The use of photodynamic therapy in dentistry. Clinical and experimental studies. Diss. Berlin: FU. Downey, D. C. (2002). The porphyrin pathway: The final common pathway? Medical Hypotheses, 59(6), 615–621. doi:10.1016/S03069877(02)00115-9 Ferreira, G. C., Franco, R., & José, J. (1999). Ferrochelatase: A new iron-sulfur center containing enzyme. 3.3 Steady - State kinetic properties of ferrochelatase. Iron metabolism. Wiley-VCH. Foroulis, C. N., & Thorpe, J. A. C. (2006). Photodynamic therapy (PDT) in Barrett’s esophagus with dysplasia or early cancer. European Journal of Cardio-Thoracic Surgery, 29, 30–34. doi:10.1016/j.ejcts.2005.10.033 Gaullier, J. M., Geze, M., Santus, R., Sa, M. T., Maziere, J. C., & Bazin, M. (1995). Subcellular localization of and photosensitization by protoporphyrin IX in human keratinocytes and fibroblasts cultivated with 5-aminolevulinic acid. Photochemistry and Photobiology, 62, 114–122. doi:10.1111/j.1751-1097.1995.tb05247.x Ismail, M. S., Dressler, C., Strobele, S., Daskalaki, A., Philipp, C., & Berlien, H.-P. (1997). Modulation of 5-ALA-induced PplX xenofluorescence intensities of a murine tumour and non-tumour tissue cultivated on the chorio-allantoic membrane. Lasers in Medical Science, 12, 218–225. doi:10.1007/BF02765102

Modeling of Porphyrin Metabolism with PyBioS

Kamburov, A., Wierling, C., Lehrach, H., & Herwig, R. (2006, December 1-2). ConsensusPathDB - Database for matching pathway annotation. Systems Biology, Proceedings of Computational Proteomics Joint Annual RECOMB 2005 Satellite Workshops on Systems Biology and on Regulatory Genomics, San Diego, CA, USA. Kemmner, W., Wan K., Rüttinger S., Ebert B., Macdonald R., Klamm U., & Moesta K.T. (2008, February). Silencing of human ferrochelatase causes abundant protoporphyrin-IX accumulation in colon cancer. FASEB J., (2), 500-9. Epub 2007, Sep 17. Kennedy, J. C., & Pottier, R. H. (1992). Endogenous protoporphyrin IX, a clinical useful photosensitizer for photodynamic therapy. J Photoresponse Chem Photobiol, 14, 275–292. doi:10.1016/1011-1344(92)85108-7 Kirby, I., Bland, J., & Daly, M. Constantine, & Karakousis, P. (2001). Surgical oncology: Contemporary principles and practice. New York: McGraw-Hill. Klipp, E., Herwig, R., Kowald, A., Wierling, C., & Lehrach, H. (2005). Systems biology in practice. Concepts, implementation and application. WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. Klipp, E., Nordlander, B., Kruger, R., Gennemark, P., & Hohmann, S. (2005). Integrative model of the response of yeast to osmotic schock. Nature Biotechnology, 23, 975–982. doi:10.1038/nbt1114 Mitchell, L. W., Volin, M., Martins, J., & Jaffe, E. K. (2001) Mechanistic implications of mutations to the active site lysine of porphobilinogen synthase. J Biol Chem, 12, 276(2), 1538-44. Munakata, H., Sun, J-Y., Yoshida, K., Nakatan, T., Honda, E., Hayakawa, S., Furuyama, K., & Hayashi, N. (2004). Role of the heme regulatory motif in the heme-mediated inhibition of mitochondrial import of 5-aminolevulinate synthase. J Biochem, 136, 233-238, 136(2), 233.

Reinbothe, S., & Reinbothe, C. (1996). The regulation of enzymes involved in chlorophyll biosynthesis. European Journal of Biochemistry, 237, 323–343. doi:10.1111/j.1432-1033.1996.00323.x Schneider-Yin, X., Gouya, L., Dorsey, M., Rüfenacht, U., & Deybach, J.-C. (2000). Mutations in the iron-sulfur cluster ligands of the human ferrochelatase. Blood, 96, 1545–1549. Schoeberl, B., Eichler-Jonsson, G., Gilles, E. D., & Muller, G. (2002). Computational modeling of the dynamics of the MAP kinase cascade activated by surface and internalized EGF receptors. Nature Biotechnology, 20, 370–375. doi:10.1038/ nbt0402-370 Schoenfeld, N., Epstein, O., Lahav, M., Mamet, R., Shaklai, M., & Atsmon, A. (1988). The heme biosynthetic pathway in lymphocytes of patients with malignant lymphoproliferative disorders. Cancer Letters, 43, 43–48. doi:10.1016/03043835(88)90211-X Vastrik, I., D‘Eustachio, P., Schmidt, E., JoshiTope, G., Gopinath, G., & Croft, D. (2007). Reactome: A knowledge base of biologic pathways and processes. Genome Biology, 8, R39. doi:10.1186/ gb-2007-8-3-r39 Von Wettstein, D., Gough, S., & Kannangara, C. G. (1995). Chlorophyll biosynthesis. The Plant Cell, 7, 1039–1057. Westerhoff, H. V. et al. (2008). Systems biology towards life in silico: Mathematics of the control of living cells. J Math Biol. Wierling, C., Herwig, R., & Lehrach, H. (2007). Resources, standards and tools for systems biology. Briefings in Functional Genomics and Proteomics, 10, 1093/bfgp/elm027. Wilson, J.H.P., van Hillegersberg, R., van den Berg, J.W.O., Kort, W.J., & Terpsta, O.T. Photodynamic therapy for gastrointestinal tumors. Scand J Gastroenterol, 26(Suppl), 188: 20-25.

875

Modeling of Porphyrin Metabolism with PyBioS

KEY TERMS AND DEFINITIONS ALA: The first committed precursor of the porphyrin pathway Ferrochelatase: Also known as FECH is an enzyme involved in porphyrin metabolism, converting protoporphyrin IX into heme. Forward Modeling: Modeling approach to study the interaction of processes to produce a response. Heme: Iron-containing porphyrin (heme-containing protein or hemoprotein) that is extensively found in nature ie. hemoglobin. Knockout-Experiment: A experiment, where an organism is engineered to lack the expression and activity of one or more genes.

Mutation: Changes to the nucleotide sequence of the genetic material of an organism. Mutations can be caused by copying errors in the genetic material during cell division, by exposure to ultraviolet or ionizing radiation, chemical mutagens, or viruses. Porphyrin: Porphyrins are heterocyclic macrocycles, consisting of four pyrrole subunits (tetrapyrrole) linked by four methine (=CH-) bridges. The extensive conjugated porphyrin macrocycle is chromatic and the name itself, porphyrin, is derived from the Greek word for purple. PyBioS: PyBioS is a system for the modeling and simulation of cellular processes. It is developed at the Max-Planck-Institute for Molecular Genetics in the department of Prof. Lehrach.

This work was previously published in Handbook of Research on Systems Biology Applications in Medicine, edited by Andriani Daskalaki, pp. 643-654, copyright 2009 by Medical Information Science Publishing (an imprint of IGI Global).

876

877

Chapter 3.16

AI Methods for Analyzing Microarray Data Amira Djebbari National Research Council Canada, Canada Aedín C. Culhane Harvard School of Public Health, USA Alice J. Armstrong The George Washington University, USA John Quackenbush Harvard School of Public Health, USA

INTRODUCTION Biological systems can be viewed as information management systems, with a basic instruction set stored in each cell’s DNA as “genes.” For most genes, their information is enabled when they are transcribed into RNA which is subsequently translated into the proteins that form much of a cell’s machinery. Although details of the process for individual genes are known, more complex interactions between elements are yet to be discovered. What we do know is that diseases can result DOI: 10.4018/978-1-60960-561-2.ch316

if there are changes in the genes themselves, in the proteins they encode, or if RNAs or proteins are made at the wrong time or in the wrong quantities. Recent advances in biotechnology led to the development of DNA microarrays, which quantitatively measure the expression of thousands of genes simultaneously and provide a snapshot of a cell’s response to a particular condition. Finding patterns of gene expression that provide insight into biological endpoints offers great opportunities for revolutionizing diagnostic and prognostic medicine and providing mechanistic insight in data-driven research in the life sciences, an area with a great need for advances, given the urgency

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

AI Methods for Analyzing Microarray Data

associated with diseases. However, microarray data analysis presents a number of challenges, from noisy data to the curse of dimensionality (large number of features, small number of instances) to problems with no clear solutions (e.g. real world mappings of genes to traits or diseases that are not yet known). Finding patterns of gene expression in microarray data poses problems of class discovery, comparison, prediction, and network analysis which are often approached with AI methods. Many of these methods have been successfully applied to microarray data analysis in a variety of applications ranging from clustering of yeast gene expression patterns (Eisen et al., 1998) to classification of different types of leukemia (Golub et al., 1999). Unsupervised learning methods (e.g. hierarchical clustering) explore clusters in data and have been used for class discovery of distinct forms of diffuse large B-cell lymphoma (Alizadeh et al., 2000). Supervised learning methods (e.g. artificial neural networks) utilize a previously determined mapping between biological samples and classes (i.e. labels) to generate models for class prediction. A k-nearest neighbor (k-NN) approach was used to train a gene expression classifier of different forms of brain tumors and its predictions were able to distinguish biopsy samples with different prognosis suggesting that microarray profiles can predict clinical outcome and direct treatment (Nutt et al., 2003). Bayesian networks constructed from microarray data hold promise for elucidating the underlying biological mechanisms of disease (Friedman et al., 2000).

BACKGROUND Cells dynamically respond to their environment by changing the set and concentrations of active genes by altering the associated RNA expression. Thus “gene expression” is one of the main determinants of a cell’s state, or phenotype. For example, we can investigate the differences between a normal cell and a cancer cell by examining their relative gene expression profiles. Microarrays quantify gene expression levels in various conditions (such as disease vs. normal) or across time points. For n genes and m instances (biological samples), microarray measurements are stored in an n by m matrix where each row is a gene, each column is a sample and each element in the matrix is the expression level of a gene in a biological sample, where samples are instances and genes are features describing those instances. Microarray data is available through many public online repositories (Table 1). In addition, the Kent-Ridge repository (http://sdmc.i2r.a-star.edu. sg/rp/) contains pre-formatted data ready to use with the well-known machine learning tool Weka (Witten & Frank, 2000). Microarray data presents some unique challenges for AI such as a severe case of the curse of dimensionality due to the scarcity of biological samples (instances). Microarray studies typically measure tens of thousands of genes in only tens of samples. This low case to variable ratio increases the risk of detecting spurious relationships. This problem is exacerbated because microarray data contains multiple sources of within-class variability, both technical and biological. The high

Table 1. Some public online repositories of microarray data Name of the repository

URL

ArrayExpress at the European Bioinformatics Institute

http://www.ebi.ac.uk/arrayexpress/

Gene Expression Omnibus at the National Institutes of Health

http://www.ncbi.nlm.nih.gov/geo/

Stanford microarray database

http://smd.stanford.edu/

Oncomine

http://www.oncomine.org/main/index.jsp

878

AI Methods for Analyzing Microarray Data

levels of variance and low sample size make feature selection difficult. Testing thousands of genes creates a multiple testing problem, which can result in underestimating the number of false positives. Given data with these limitations, constructing models becomes under-determined and therefore prone to over-fitting. From biology, it is also clear that genes do not act independently. Genes interact in the form of pathways or gene regulatory networks. For this reason, we need models that can be interpreted in the context of pathways. Researchers have successfully applied AI methods to microarray data preprocessing, clustering, feature selection, classification, and network analysis.

MINING MICROARRAY DATA: CURRENT TECHNIQUES, CHALLENGES AND OPPORTUNITIES FOR AI Data Preprocessing After obtaining microarray data, normalization is performed to account for systematic measurement biases and to facilitate between-sample comparisons (Quackenbush, 2002). Microarray data may contain missing values that may be replaced by mean replacement or k-NN imputation (Troyanskaya et al., 2001).

Feature Selection The goal of feature selection is to find genes (features) that best distinguish groups of instances (e.g. disease vs. normal) to reduce the dimensionality of the dataset. Several statistical methods including t-test, significance analysis of microarrays (SAM) (Tusher et al., 2001), and analysis of variance (ANOVA) have been applied to select features from microarray data. In classification experiments, feature selection methods generally aim to identify relevant

gene subsets to construct a classifier with good performance (Inza et al., 2004). Features are considered to be relevant when they can affect the class; the strongly relevant are indispensable to prediction and the weakly relevant may only sometimes contribute to prediction. Filter methods evaluate feature subsets regardless of the specific learning algorithm used. The statistical methods for feature selection discussed above as well as rankers like information gain rankers are filters for the features to be included. These methods ignore the fact that there may be redundant features (features that are highly correlated with each other and as such one can be used to replace the other) and so do not seek to find a set of features which could perform similarly with fewer variables while retaining the same predictive power (Guyon & Elisseeff, 2003). For this reason multivariate methods are more appropriate. As an alternative, wrappers consider the learning algorithm as a black-box and use prediction accuracy to evaluate feature subsets (Kohavi & John, 1997). Wrappers are more direct than filter methods but depend on the particular learning algorithm used. The computational complexity associated with wrappers is prohibitive due to curse of dimensionality, so typically filters are used with forward selection (starting with an empty set and adding features one by one) instead of backward elimination (starting with all features and removing them one by one). Dimension reduction approaches are also used for multivariate feature selection.

Dimension Reduction Approaches Principal component analysis (PCA) is widely used for dimension reduction in machine learning (Wall et al., 2003). The idea behind PCA is quite intuitive: correlated objects can be combined to reduce data “dimensionality”. Relationships between gene expression profiles in a data matrix can be expressed as a linear combination such

879

AI Methods for Analyzing Microarray Data

that colinear variables are regressed onto a new set of coordinates. PCA, its underlying method Single Value Decomposition (SVD), related approaches such as correspondence analysis (COA), and multidimensional scaling (MDS) have been applied to microarray data and are reviewed by Brazma & Culhane (2005). Studies have reported that COA or other dual scaling dimension reduction approaches such as spectral map analysis may be more appropriate than PCA for decomposition of microarray data (Wouters et al., 2003). While PCA considers the variance of the whole dataset, clustering approaches examine the pairwise distance between instances or features. Therefore, these methods are complementary and are often both used in exploratory data analysis. However, difficulties in interpreting the results in terms of discrete genes limit the application of these methods.

Clustering What we see as one disease is often a collection of disease subtypes. Class discovery aims to discover these subtypes by finding groups of instances with similar expression patterns. Hierarchical clustering is an agglomerative method which starts with a singleton and groups similar data points using some distance measure such that two data points that are most similar are grouped together in a cluster by making them children of a parent node in the tree. This process is repeated in a bottomup fashion until all data points belong to a single cluster (corresponding to the root of the tree). Hierarchical and other clustering approaches, including K-means, have been applied to microarray data (Causton et al., 2003). Hierarchical clustering was applied to study gene expression in samples from patients with diffuse large B-cell lymphoma (DLBCL) resulting in the discovery of two subtypes of the disease. These groups were found by analyzing microarray data from biopsy samples of patients who had not been previously treated. These patients continued to be studied after

880

chemotherapy, and researchers found that the two newly discovered disease subtypes had different survival rates, confirming the hypothesis that the subtypes had significantly different pathologies (Alizadeh et al., 2000). While clustering simply groups the given data based on pair-wise distances, when information is known a priori about some or all of the data i.e. labels, a supervised approach can be used to obtain a classifier that can predict the label of new instances.

Classification (Supervised Learning) The large dimensionality of microarray data means that all classification methods are susceptible to over-fitting. Several supervised approaches have been applied to microarray data including Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), and k-NNs among others (Hastie et al., 2001). A very challenging and clinically relevant problem is the accurate diagnosis of the primary origin of metastatic tumors. Bloom et al. (2004) applied ANNs to the microarray data of 21 tumor types with 88% accuracy to predict the primary site of origin of metastatic cancers with unknown origin. A classification of 84% was obtained on an independent test set with important implications for diagnosing cancer origin and directing therapy. In a comparison of different SVM approaches, multicategory SVMs were reported to outperform other popular machine learning algorithms such as k-NNs and ANNs (Statnikov et al., 2005) when applied to 11 publicly available microarray datasets related to cancer. It is worth noting that feature selection can significantly improve classification performance.

Cross-Validation Cross-validation (CV) is appropriate in microarray studies which are often limited by the number of instances (e.g. patient samples). In k-fold CV,

AI Methods for Analyzing Microarray Data

the training set is divided into k subsets of equal size. In each iteration k-1 subsets are used for training and one subset is used for testing. This process is repeated k times and the mean accuracy is reported. Unfortunately, some published studies have applied CV only partially, by applying CV on the creation of the prediction rule while excluding feature selection. This introduces a bias in the estimated error rates and over-estimates the classification accuracy (Simon et al., 2003). As a consequence, results from many studies are controversial due to methodological flaws (Dupuy & Simon, 2007). Therefore, models must be evaluated carefully to prevent selection bias (Ambroise & McLachlan, 2002). Nested CV is recommended, with an inner CV loop to perform the tuning of the parameters and an outer CV to compute an estimate of the error (Varma & Simon, 2006). Several studies which have examined similar biological problems have reported poor overlap in gene expression signatures. Brenton et al. (2005) compared two gene lists predictive of breast cancer prognosis and found only 3 genes in common. Even though the intersection of specific gene lists is poor, the highly correlated nature of microarray data means that many gene lists may have similar prediction accuracy (Ein-Dor et al., 2004). Gene signatures identified from different breast cancer studies with few genes in common were shown to have comparable success in predicting patient survival (Buyse et al., 2006). Commonly used supervised learning algorithms yield black box models prompting the need for interpretable models that provide insights about the underlying biological mechanism that produced the data.

Network Analysis Bayesian networks (BNs), derived from an alliance between graph theory and probability theory, can capture dependencies among many variables (Pearl, 1988, Heckerman, 1996).

Friedman et al. (2000) introduced a multinomial model framework for BNs to reverseengineer networks and showed that this method differs from clustering in that it can discover gene interactions other than correlation when applied to yeast gene expression data. Spirtes et al. (2002) highlight some of the difficulties of applying this approach to microarray data. Nevertheless, many extensions of this research direction have been explored. Correlation is not necessarily a good predictor of interactions, and weak interactions are essential to understand disease progression. Identifying the biologically meaningful interactions from the spurious ones is challenging, and BNs are particularly well-suited for modeling stochastic biological processes. The exponential growth of data produced by microarray technology as well as other highthroughput data (e.g. protein-protein interactions) call for novel AI approaches as the paradigm shifts from a reductionist to a mechanistic systems view in the life sciences.

FUTURE TRENDS Uncovering the underlying biological mechanisms that generate these data is harder than prediction and has the potential to have far reaching implications for understanding disease etiologies. Time series analysis (Bar-Joseph, 2004) is a first step to understanding the dynamics of gene regulation, but, eventually, we need to use the technology not only to observe gene expression data but also to direct intervention experiments (Pe’er et al., 2001, Yoo et al., 2002) and develop methods to investigate the fundamental problem of distinguishing correlation from causation.

CONCLUSION We have reviewed AI methods for pre-processing, clustering, feature selection, classification and

881

AI Methods for Analyzing Microarray Data

mechanistic analysis of microarray data. The clusters, gene lists, molecular fingerprints and network hypotheses produced by these approaches have already shown impact; from discovering new disease subtypes and biological markers, predicting clinical outcome for directing treatment as well as unraveling gene networks. From the AI perspective, this field offers challenging problems and may have a tremendous impact on biology and medicine.

Brenton, J. D., Carey, L. A., Ahmed, A. A., & Caldas, C. (2005). Molecular classification and molecular forecasting of breast cancer: ready for clinical application? Journal of Clinical Oncology, 23(29), 7350–7360. doi:10.1200/ JCO.2005.03.3845

REFERENCES

Causton, H. C., Quackenbush, J., & Brazma, A. (2003) Microarray Gene Expression Data Analysis: A Beginner’s Guide. Oxford: Blackwell Science Limited.

Alizadeh, A. A., Eisen, M. B., Davis, R. E., Ma, C., Lossos, I. S., & Rosenwald, A. (2000). Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature, 403(6769), 503–511. doi:10.1038/35000501 Ambroise, C., & McLachlan, G. J. (2002). Selection bias in gene extraction on the basis of microarray gene-expression data. Proceedings of the National Academy of Sciences of the United States of America, 99(10), 6562–6566. doi:10.1073/ pnas.102102699 Bar-Joseph, Z. (2004). Analyzing time series gene expression data. Bioinformatics (Oxford, England), 20(16), 2493–2503. doi:10.1093/bioinformatics/bth283 Bloom, G., Yang, I. V., Boulware, D., Kwong, K. Y., Coppola, D., & Eschrich, S. (2004). Multiplatform, multi-site, microarray-based human tumor classification. American Journal of Pathology, 164(1), 9–16. Brazma, A., & Culhane, A. C. (2005). Algorithms for gene expression analysis. In Jorde LB., Little PFR, Dunn MJ., Subramaniam S. (Eds.) Encyclopedia of Genetics, Genomics, Proteomics and Bioinformatics., (3148 -3159) London: John Wiley & Sons.

882

Buyse, M., Loi, S., Van’t Veer, L., Viale, G., Delorenzi, M., & Glas, A. M. (2006). Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer. Journal of the National Cancer Institute, 98, 1183–1192.

Dupuy, A., & Simon, R. M. (2007). Critical review of published microarray studies for cancer outcome and guidelines on statistical analysis and reporting. Journal of the National Cancer Institute, 99(2), 147–157. doi:10.1093/jnci/djk018 Ein-Dor, L., Kela, I., Getz, G., Givol, D., & Domany, E. (2004). Outcome signature genes in breast cancer: is there a unique set? Bioinformatics (Oxford, England), 21(2), 171–178. doi:10.1093/ bioinformatics/bth469 Eisen, M. B., Spellman, P. T., Brown, P. O., & Botstein, D. (1998). Cluster analysis and display of genome-wide expression patterns. Proceedings of the National Academy of Sciences of the United States of America, 95, 14863–14868. doi:10.1073/ pnas.95.25.14863 Friedman, N., Linial, M., Nachman, I., & Pe’er, D. (2000). Using Bayesian networks to analyze expression data. Journal of Computational Biology, 7(34), 601–620. doi:10.1089/106652700750050961 Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., & Mesirov, J. P. (1999). Molecular Classification of Cancer: Class Discovery and Class Prediction by Gene Expression Monitoring. Science, 286(5439), 531. doi:10.1126/ science.286.5439.531

AI Methods for Analyzing Microarray Data

Guyon, I., & Elisseff, A. (2003). An introduction to variable and feature selection. Journal of Machine Learning Research, 3, 1157–1182. doi:10.1162/153244303322753616

Quackenbush, J. (2006). Microarray Analysis and Tumor Classification. The New England Journal of Medicine, 354(23), 2463–2472. doi:10.1056/ NEJMra042342

Hastie, T., Tibshirani, R., & Friedman, J. (2001). The Elements of Statistical Learning: Data Mining, Inference, and Prediction. New York: Springer Series in Statistics.

Simon, R., Radmacher, M. D., Dobbin, K., & McShane, L. M. (2003). Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification. Journal of the National Cancer Institute, 95(1), 14–18.

Heckerman, D. (1996). A Tutorial on Learning with Bayesian Networks. Technical Report MSRTR-95-06. Microsoft Research. Inza I., Larrañaga P., Blanco R., & Cerrolaza A.J. (2004). Filter versus wrapper gene selection approaches in DNA microarray domains. Artificial Intelligence in Medicine, special issue in “Data mining in genomics and proteomics”, 31(2), 91-103. Kohavi, R., & John, G. H. (1997). Wrappers for feature subset selection. Artificial Intelligence, 97(1-2), 273–324. doi:10.1016/S00043702(97)00043-X Nutt, C. L., Mani, D. R., Betensky, R. A., Tamayo, P., Cairncross, J. G., & Ladd, C. (2003). Gene Expression-based Classification of Malignant Gliomas Correlates Better with Survival than Histological Classification. Cancer Research, 63, 1602–1607. Pe’er, D., Regev, A., Elidan, G., & Friedman, N. (2001). Inferring subnetworks from perturbed expression profiles. Bioinformatics (Oxford, England), 17(S1), S215–S224. Pearl, J. (1988). Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, San Mateo: Morgan Kaufmann Publishers. Quackenbush, J. (2002). Microarray data normalization and transformation. Nature Genetics, 32, 496–501. doi:10.1038/ng1032

Spirtes, P., Glymour, C., & Scheines, R. Kauffman, S., Aimale, V., & Wimberly, F. (2001). Constructing Bayesian Network Models of Gene Expression Networks from Microarray Data. Proceedings of the Atlantic Symposium on Computational Biology, Genome Information Systems and Technology. Statnikov, A., Aliferis, C. F., Tsamardinos, I., Hardin, D., & Levy, S. (2005). A comprehensive evaluation of multicategory classification methodsfor microarray gene expression cancer diagnosis. Bioinformatics (Oxford, England), 21(5), 631–643. doi:10.1093/bioinformatics/bti033 Troyanskaya, O., Cantor, M., Sherlock, G., Brown, P., Hastie, T., & Tibshirani, R. (2001). Missing value estimation methods for DNA microarrays. Bioinformatics (Oxford, England), 17(6), 520–525. doi:10.1093/bioinformatics/17.6.520 Tusher, V. G., Tibshirani, R., & Chu, G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proceedings of the National Academy of Sciences of the United States of America, 98(9), 5116–5121. doi:10.1073/ pnas.091062498 Varma, S., & Simon, R. (2006). Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics, 7, 91. doi:10.1186/1471-2105-7-91 Wall, M., Rechtsteiner, A., & Rocha, L. (2003). Singular value decomposition and principal component analysis. In D.P. Berrar, W. Dubitzky, M. Granzow (Eds.) A Practical Approach to Microarray Data Analysis. (91-109). Norwell: Kluwer. 883

AI Methods for Analyzing Microarray Data

Witten, I. H., & Frank, E. (2000). Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations. Morgan Kaufmann Publishers Inc. Wouters, L., Gohlmann, H. W., Bijnens, L., Kass, S. U., Molenberghs, G., & Lewi, P. J. (2003). Graphical exploration of gene expression data: a comparative study of three multivariate methods. Biometrics, 59, 1131–1139. doi:10.1111/j.0006341X.2003.00130.x Yoo, C., Thorsson, V., & Cooper, G. F. (2002). Discovery of causal relationships in a gene-regulation pathway from a mixture of experimental and observational DNA microarray data. Biocomputing: Proceedings of the Pacific Symposium, 7, 498-509

KEY TERMS AND DEFINITIONS Curse of Dimensionality: A situation where the number of features (genes) is much larger than the number of instances (biological samples) which is known in statistics as p >> n problem.

Feature Selection: A problem of finding a subset (or subsets) of features so as to improve the performance of learning algorithms. Microarray: A microarray is an experimental assay which measures the abundances of mRNA (intermediary between DNA and proteins) corresponding to gene expression levels in biological samples. Multiple Testing Problem: A problem that occurs when a large number of hypotheses are tested simultaneously using a user-defined α cut off p-value which may lead to rejecting a nonnegligible number of null hypotheses by chance. Over-Fitting: A situation where a model learns spurious relationships and as a result can predict training data labels but not generalize to predict future data. Supervised Learning: A learning algorithm that is given a training set consisting of feature vectors associated with class labels and whose goal is to learn a classifier that can predict the class labels of future instances. Unsupervised Learning: A learning algorithm that tries to identify clusters based on similarity between features or between instances or both but without taking into account any prior knowledge.

This work was previously published in Encyclopedia of Artificial Intelligence, edited by Juan Ramón Rabuñal Dopico, Julian Dorado and Alejandro Pazos, pp. 65-70, copyright 2009 by Information Science Reference (an imprint of IGI Global).

884

885

Chapter 3.17

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization Guang Li National Cancer Institute, USA

Boris Mueller Memorial Sloan-Kettering Cancer Center, USA

Deborah Citrin National Cancer Institute, USA

Borys Mychalczak Memorial Sloan-Kettering Cancer Center, USA

Robert W. Miller National Cancer Institute, USA

Yulin Song Memorial Sloan-Kettering Cancer Center, USA

Kevin Camphausen National Cancer Institute, USA

INTRODUCTION Image registration, segmentation, and visualization are three major components of medical image processing. Three-dimensional (3D) digital medical images are three dimensionally reconstructed, often with minor artifacts, and with limited spatial resolution and gray scale, unlike common digital pictures. Because of these limitations, image filtering is often performed before the images DOI: 10.4018/978-1-60960-561-2.ch317

are viewed and further processed (Behrenbruch, Petroudi, Bond, et al., 2004). Different 3D imaging modalities usually provide complementary medical information about patient anatomy or physiology. Four-dimensional (4D) medical imaging is an emerging technology that aims to represent patient motions over time. Image registration has become increasingly important in combining these 3D/4D images and providing comprehensive patient information for radiological diagnosis and treatment.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

3D images have been utilized clinically since computed tomography (CT) was invented (Hounsfield, 1973). Later on, magnetic resonance imaging (MRI), positron emission tomography (PET), and single photon emission computed tomography (SPECT) have been developed, providing 3D imaging modalities that complement CT. Among the most recent advances in clinical imaging, helical multislice CT provides improved image resolution and capacity of 4D imaging (Pan, Lee, Rietzel, & Chen, 2004; Ueda, Mori, Minami et al., 2006). Other advances include mega-voltage CT (MVCT), cone-beam CT (CBCT), functional MRI, open field MRI, time-of-flight PET, motioncorrected PET, various angiography, and combined modality imaging, such as PET/CT (Beyer, Townsend, Brun et al., 2000), and SPECT/CT (O’Connor & Kemp, 2006). Some preclinical imaging techniques have also been developed, including parallel multichannel MRI (Bohurka, 2004), Overhauser enhanced MRI (Krishna, English, Yamada et al., 2002), and electron paramagnetic resonance imaging (EPRI) (Matsunoto, Subramanian, Devasahayam et al., 2006). Postimaging analysis (image processing) is required in many clinical applications. Image processing includes image filtering, segmentation, registration, and visualization, which play a crucial role in medical diagnosis/treatment, especially in the presence of patient motion and/ or physical changes. In this article, we will provide a state-of-the-art review on 3D/4D image registration, combined with image segmentation and visualization, and its role in image-guided radiotherapy (Xing, Thorndyke, Schreibmann et al., 2006).

ing anatomic or physiologic information in 3D space. The smallest element of a 3D image is a cubic volume called voxel. A 4D medical image contains a temporal series of 3D images. With a subsecond time resolution, it can be used for monitoring respiratory/cardiac motion (Keall, Mageras, Malter et al., 2006). Patient motion is always expected: faster motion relative to imaging speed causes a blurring artifact; whereas slower motion may not affect image quality. A multislice CT scanner provides improved spatial and temporal resolution (Ueda et al., 2006), which can be employed for 4D imaging (Pan et al., 2004). Progresses in MRI imaging have also been reported, including parallel multichannel MRI (Bodurka, Ledden, van Gelderen et al., 2004). Because PET resolution and speed are limited by the physics and biology behind the imaging technique, some motion suppression techniques have been developed clinically, including patient immobilization (Beyer, Tellmann, Nickel, & Pietrzyk, 2005), respiratory gating (Hehmeh, Erdi, Pan et al., 2004), and motion tracking (Montgomery, Thielemans, Mehta et al., 2006). Motion tracking data can be used to filter the imaging signals prior to PET image reconstruction for reliable motion correction. Motion blurring, if uncorrected, can reduce registration accuracy. Visual-based volumetric registration technique provides blurring correction (filtering) before registration, by defining the PET volume with reference to the CT volume, causing blurred PET surface voxels to be rendered invisible (Li, Xie, Ning et al., 2007).

BACKGROUND

Medical image segmentation defines regions of interest used to adapt image changes, study image deformation, and assist image registration. Many methods for segmentation have been developed including thresholding, region growing, clustering, as well as atlas-guided and level sets

3D/4D Medical Imaging A 3D medical image contains a sequence of parallel two-dimensional (2D) images represent-

886

Image Segmentation and Visualization

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

(Pham, Xu, & Prince, 2000; Suri, Liu, & Singh et al., 2002). Atlas-guided methods are based on a standard anatomical atlas, which serves as an initial point for adapting to any specific image. Level sets, also called active contours, are geometrically deformable models, used for fast shape recovery. Atlas-based level sets have been applied clinically for treatment planning (Lu, Olivera, Chen et al., 2006a; Ragan, Starkschall, McNutt et al., 2005;) and are closely related to image registration (Vemuri, Ye, & Chen, et al., 2003). Figure 1 shows automatic contours. Depending on how the 3D image is segmented, it can be either 2D-based or 3D-based (Suri, Liu, Reden, & Laxminarayan, 2002). 3D medical image visualization has been increasingly applied in diagnosis and treatment (Salgado, Mulkens, Bellinck, & Termote, 2003), whereas 2D-based visualization is predominantly applied clinically. Because of the demand on computing power, real-time 3D image visualFigure 1. Orthogonal 2D-views of CT images and comparison of automatic recontours (solid-lines) and manual contours (dash-lines) in different phases (A&B) of a radiotherapeutic treatment (courtesy of Dr. Weiguo Lu)

ization is supported by specialized graphics hardware (Terarecon, Inc.) (Xie, Li, Ning et al., 2004) or high-end consumer graphics processors (Levin, Aladl, Germanos, & Slomak, 2005). 3D image visualization has been applied to registration of four imaging modalities with improved spatial accuracy (Li, Xie, Ning et al., 2005; Li et al., 2007). Figure 2 shows 3D volumetric image registration using external and internal anatomical landmarks.

Rigid Image Registration Rigid image registration assumes a motionless patient such that the underlying anatomy is identical in different imaging modalities for alignment. Three approaches to rigid registration are: coordinate-based, extrinsic-based, and intrinsicbased (Maintz & Viergever, 1998). Coordinatebased registration is performed by calibrating the coordinate system to produce “co-registered” images. Multimodality scanners, such as PET/CT and SPECT/CT, are typical examples. Extrinsic-based image registration relies on the alignment of extrinsic objects placed in/on a patient invasively/noninvasively. Such objects can be fiducials or frames that are visible in all imaging modalities and serve as local coordinate markers (sets of points) for rigid registration. Examples are gold seeds for prostate localization in radiotherapy and head frame for stereotactic radiosurgery. Intrinsic-based image registration uses a patient’s anatomy (anatomic landmarks, segmented geometries, or intact voxels) as the registration reference. Alignment of visual landmarks or segmented geometries requires user interaction, so the registration is manual or semi-automatic. The statistical similarity of the intact voxels (grayscale) of two images, such as mutual information (Viola & Wells, 1995), has been widely used for fully automated registration (Pluim, Maintz, & Viergever, 2003).

887

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

Figure 2. 3D-views of CT (A, B, and C: head) and MR (D, E, and F: segmented brain) phantom images. The homogeneity of color distributed on an anatomic landmark is used as the registration criterion. From (A) to (C), three images (red, green, and blue) are approaching registration with shifts from 5.0mm to 0.5mm and 0.0mm, respectively. From (D) to (E and F), four images (grey, blue, green, and red) are 5.0mm apart from each other (D) and registered in front (E) and side (F) views.

Automatic image registration requires three key elements: a metric function, a transformation, and an optimization process. One common voxelbased image registration uses mutual information as the metric, a rigid or nonrigid transformation and a maximization algorithm. Recently, the homogeneity of color distributed on a volumetric landmark has been used as quantitative metric, assisted by the ray-casting algorithm in 3D visualization (Li et al., 2007). Figures 3 and 4 show clinical examples using the 3D visualization-based registration technique.

Deformable Image Registration Deformable image registration contains a nonrigid transformation model that specifies the way to deform one image to match another. A rigid image registration is almost always performed to determine an initial position using rigid transformation with six variables (3 translations and 3 rotations). For nonrigid transformation, the

888

Figure 3. 3D-views of before (A, B, and C) and after (D, E, and F) 3D volumetric image registration, using homogeneity color distribution as registration criterion. Voluntary patient head movement is corrected in three MR images, T1 (green), FLAIR (red), and T2 (blue), which are acquired in the same scanner with time interval of 3 and 20 minutes.

Figure 4. 3D-views of before (A, B, and C) and after (D, E, and F) rigid volumetric image registration of PET/CT images, correcting patient movement.

number of variables will increase dramatically, up to three times the number of voxels. Common deformable transformations are spline-based with control points, the elastic model driven by image similarity, the viscous fluid model with region growth, the finite element model using rigidity classification, the optical flow with motion estimation, and free-form deformation (Chi, Liang, & Yan, 2006; Crum, Hartkens, & Hill, 2004; Lu, Olivera, Chen et al., 2006b).

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

The image similarity measures are ultimately the most important criteria for determining the quality of registration. They can be feature-based and voxel-based (Maintz & Viergever, 1998). The former is usually segmentation/classification based, adapting changes in shape of anatomical landmarks, while the latter is based on statistical criteria for intensity pattern matching, including mutual information. Most deformable registrations are automated. Combining the two methods can improve registration accuracy, reliability, and/ or performance (Hellier & Barillot, 2003; Liu, Shen, & Davatizikos, 2004; Wyatt & Noble, 2003). Figure 5 shows one example of deformable registration.

Figure 5. Orthogonal 2D-views of before (A) and after (B) deformable image registration of two 3D images (red and green) in a 4D CT image (courtesy of Dr. Weiguo Lu)

A Challenge from 3D/4D Conformal Radiotherapy: Deformable Image Registration Broadened Concept of 4D Medical Imaging The 4D imaging concept has been broadened to cover various time resolutions. The common 4D image has subsecond temporal resolution (Pan et al., 2004), while a series of 3D images, reflecting patient changes over a longer time span, should be also qualified as a 4D image with sufficient resolution to assess slower changes, including tumor growth/shrinkage and weight gain/loss during a course of treatment. Registration of a 3D image to a 4D image involves a series of deformable registration. Because patient motion/change is inevitable, deformable image registration is the key to combining these images for clinical use. Clinically, MVCT images can be acquired daily and used for patient daily setup via rigid registration to the reference planning image, assuming minimal patient changes. Within a treatment, 4D CT imaging has shown dramatic anatomical changes during respiration (Keall et al., 2006). Image-guided frameless cranial and extra-cranial stereotactic radiosurgery has been

performed clinically (Gibbs, 2006). Rigid image registration provides the best current solution to these clinical applications. Ultimately, deformable image registration will improve the registration accuracy substantially (Lu et al., 2006b), permitting highly conformal 3D/4D radiotherapy (Barbiere, Hanley, Song et al., 2007; Mackie, Kapatoes, Ruchala et al., 2003).

Challenges in Deformable Image Registration For deformable image registration, the underlying anatomy changes, therefore voxel mapping among images is a challenge. First, the deformable transformation handles a large number of positioning variables that must be determined for every voxels within the anatomy. This can be abstracted as a multiple variable optimization problem in mathematics, limiting the performance of deformable image registration for many years (Crum, 2004). Second, the deformable registration is extremely difficult to be validated as there is

889

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

lack of absolutes with respect to the location of corresponding voxels. Therefore, the accuracy and reliability of deformable registration should be evaluated on a case specific basis (Sanchez Castro, Pollo, Meuli et al., 2006; Wang, Dong, O’Daniel et al., 2005). Regardless the limitations above, progress has been made by combining image registration and segmentation/classification to provide intrinsic simplification and cross verification. It remains a challenge, however, to develop a fully automated deformable registration algorithm because image segmentation often requires human interaction. Deformable image registration is generally a “passive” mapping process. It does not anticipate how patient anatomy might deform. An example is whether superficial 3D contour information detected by a real-time infrared camera can be used to predict the motion of internal organs (Rietzel, Rosenthal, Gierga et al., 2004). Anatomically, the correlation between superficial and internal organ motion should exist, although as a complex relationship. Therefore, an anatomic model based image registration with motion estimation can provide an “active” mapping process, but is beyond the current scope of deformable image registration.

Gaps between Frontier Research and Clinical Practice Despite of the advances in 3D and 4D imaging and image registration, 2D-based rigid registration techniques are predominantly used in the clinic; although automatic rigid registration methods exist in most commercial treatment planning software. Two reasons are primarily responsible for this disconnect: First, the user must visually verify the final registration using the 2D-based visualization tools available for image fusion in most commercial software. Second, most clinical images have some degree of pre-existing defor-

890

mation so that automatic rigid registration can prove unreliable but manual methods allow the user to perform local organ registration. Some recent commercial software has recognized this problem and provides the option of selecting the region-of-interest to ease the deformation problem. This method, however, is only a partial solution to cope with image changes. The gap between the clinical research and routine practice can be reduced by translational research and development. Recently, open-source medical image processing and visualization tool kits have become available for public use. Many recently published algorithms in medical image processing are implemented in a generic, objectoriented programming style, which permits reusability of such toolkits.

FUTURE TRENDS 3D rigid image registration will dominate clinical practice and will remain essential as more specialized complementary 3D imaging modalities become clinically relevant. Although the simplicity of automatic image registration is more attractive, manual image registration with 2D/3D visualization is irreplaceable because it permits incorporation of medical knowledge for verification and adjustment of the automatic registration results. As awareness of the problems of the patient motion and anatomic changes increases, further research on 4D imaging and deformable registration will be stimulated to meet the clinical demands. Motion correction in the PET/CT and SPECT/CT will continue to improve the “coregistration” of these images. Interdisciplinary approaches are expected to offer further improvements for the difficult registration problem. With advances in hybrid registration algorithms and parallel computing, more progresses are expected, resulting in improved accuracy and performance.

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

CONCLUSION Higher dimensional deformable image registration has become a focus of clinical research. The accuracy, reliability, and performance of 3D/4D image registration have been improved with assistance of image segmentation and visualization.

REFERENCES Barbiere, J., Hanley, J., & Song, Y. (2007). Concise matrix analysis of point-based prostate targeting for intensity modulated radiation therapy. Technology in Cancer Research & Treatment, 6(1), 1–10. Behrenbruch, C. P., Petroudi, S., & Bond, S. (2004). Image filtering techniques for medical image post-processing: An overview. The British Journal of Radiology, 77, S126–S132. doi:10.1259/bjr/17464219 Beyer, T., Tellmann, L., Nickel, I., & Pietrzyk, U. (2005). On the use of positioning aids to reduce misregistration in the head and neck in whole-body PET/CT studies. Journal of Nuclear Medicine, 46(4), 596–602. Beyer, T., Townsend, D. W., & Brun, T. (2000). A combined PET/CT scanner for clinical oncology. Journal of Nuclear Medicine, 41(8), 1369–1379. Bodurka, J., Ledden, P. J., & van Gelderen, P. (2004). Scalable multichannel MRI data acquisition system. Magnetic Resonance in Medicine, 51, 165–171. doi:10.1002/mrm.10693 Chi, Y., Liang, J., & Yan, D. (2006). A material sensitivity study on the accuracy of deformable organ registration using linear biomechanical models. Medical Physics, 33(2), 421–433. doi:10.1118/1.2163838 Crum, W. R., Hartkens, T., & Hill, D. L. G. (2004). Non-rigid image registration: Theory and practice. The British Journal of Radiology, 77, S140–S153. doi:10.1259/bjr/25329214

Gibbs, I. C. (2006). Frameless image-guided intracranial and extracranial radiosurgery using the Cyberknife robotic system. Cancer Radiotherapie, 10(5), 283–287. doi:10.1016/j. canrad.2006.05.013 Hehmeh, S. A., Erdi, Y. E., & Pan, T. (2004). Four-dimensional (4D) PET/CT imaging of the thorax. Medical Physics, 31(12), 3179–3186. doi:10.1118/1.1809778 Hellier, P., & Barillot, C. (2003). Coupling dense and landmark-based approaches for nonrigid registration. IEEE Transactions on Medical Imaging, 22, 217–227. doi:10.1109/TMI.2002.808365 Hounsfield, G. (1973). Computerized transverse axial scanning (tomography): Part I. Description of system. The British Journal of Radiology, 46, 1016. doi:10.1259/0007-1285-46-552-1016 Keall, P. J., Mageras, G. S., & Malter, J. M. (2006). The management of respiratory motion in radiation oncology report of AAPM Task Group 76. Medical Physics, 33(10), 3874–3900. doi:10.1118/1.2349696 Krishna, M. C., English, S., & Yamada, K. (2002). Overhauser enhanced magnetic resonance imaging for tumor oximetry: Coregistration of tumor anatomy and tissue oxygen concentration. Proceedings of the National Academy of Sciences of the United States of America, 99(4), 2216–2221. doi:10.1073/pnas.042671399 Levin, D., Aladl, U., Germano, G., & Slomak, P. (2005). Techniques for efficient, real-time, 3D visualization of multi-modality cardiac data using consumer graphics hardware. Computerized Medical Imaging and Graphics, 29, 463–475. doi:10.1016/j.compmedimag.2005.02.007 Li, G., Xie, H., & Ning, H. (2005). A novel 3D volumetric voxel registration technique for volume-view-guided image registration of multiple imaging modalities. International Journal of Radiation Oncology, Biology, Physics, 63(1), 261–273. doi:10.1016/j.ijrobp.2005.05.008 891

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

Li, G., Xie, H., Ning, H., et al. (2007). Registering molecular imaging information into anatomic images with improved spatial accuracy. In Proc. Int. Symp. Biomed. Imaging, IEEE Signal Processing Society (pp.1144-1147).

O’Connor, M. K., & Kemp, B. J. (2006). Singlephoton emission computed tomography/computed tomography: Basic instrumentation and innovations. Seminars in Nuclear Medicine, 36(4), 258–266. doi:10.1053/j.semnuclmed.2006.05.005

Liu, T. M., Shen, D. G., & Davatizikos, C. (2004). Deformable registration of cortical structures via hybrid volumetric and surface warping. NeuroImage, 22, 1790–1801. doi:10.1016/j.neuroimage.2004.04.020

Pan, T., Lee, T.-Y., Rietzel, E., & Chen, G. T. (2004). 4D-CT imaging of a volume influenced by respiratory motion on multi-slice CT. Medical Physics, 31(2), 333–340. doi:10.1118/1.1639993

Lu, W., Olivera, G. H., & Chen, Q. (2006a). Automatic re-contouring in 4D radiotherapy. Physics in Medicine and Biology, 51, 1077–1099. doi:10.1088/0031-9155/51/5/002 Lu, W., Olivera, G. H., & Chen, Q. (2006b). Deformable registration of the planning image (kVCT) and the daily images (MVCT) for adaptive radiation therapy. Physics in Medicine and Biology, 51, 4357–4374. doi:10.1088/00319155/51/17/015 Mackie, T. R., Kapatoes, J., & Ruchala, K. (2003). Image guided for precise conformal radiotherapy. International Journal of Radiation Oncology, Biology, Physics, 56(1), 89–105. doi:10.1016/ S0360-3016(03)00090-7 Maintz, J. B. A., & Viergever, M. A. (1998). A survey of medical image registration. Medical Image Analysis, 2(1), 1–36. doi:10.1016/S13618415(01)80026-8 Matsunoto, K., Subramanian, S., & Devasahayam, N. (2006). Electron paramagnetic resonance imaging of tumor hypoxia: Enhanced spatial and temporal resolution for in vivo pO2 determination. Magnetic Resonance in Medicine, 55, 1157–1163. doi:10.1002/mrm.20872 Montgomery, A. J., Thielemans, K., & Mehta, M. A. (2006). Correction of head movement on PET studies: Comparison of methods. Journal of Nuclear Medicine, 47, 1936–1944.

892

Pham, D. L., Xu, C., & Prince, J. L. (2000). Current methods in medical image segmentation. Annual Review of Biomedical Engineering, 2, 315–237. doi:10.1146/annurev.bioeng.2.1.315 Pluim, J. P. W., Maintz, J. B. A., & Viergever, M. A. (2003). Mutual-information-based registration of medical images: A survey. IEEE Transactions on Medical Imaging, 22(8), 986–1004. doi:10.1109/ TMI.2003.815867 Ragan, D., Starkschall, G., & McNutt, T. (2005). Semiautomated four-dimensional computed tomography segmentation using deformable models. Medical Physics, 32(7), 2254–2261. doi:10.1118/1.1929207 Rietzel, E., Rosenthal, S. J., & Gierga, D. P. (2004). Moving targets: Detection and tracking of internal organ motion for treatment planning and patient set-up. Radiotherapy and Oncology, 73(S2), S68–S72. doi:10.1016/S0167-8140(04)80018-5 Salgado, R., Mulkens, T., Bellinck, P., & Termote, J.-L. (2003). Volume rendering in clinical practice: A pictorial review. JBR-BTR; Organe de la Societe Royale Belge de Radiologie (SRBR), 86(4), 215–220. Sanchez Castro, F. J., Pollo, C., & Meuli, R. (2006). A cross validation study of deep brain stimulation targeting: From experts to atlas-based, segmentation-based and automatic registration algorithms. IEEE Transactions on Medical Imaging, 25(11), 1440–1450. doi:10.1109/TMI.2006.882129

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

Suri, J. S., Liu, K., Reden, L., & Laxminarayan, S. (2002). A review on MR vascular image processing: Skeleton versus nonskeleton approaches: Part II. IEEE Transactions on Information Technology in Biomedicine, 6(4), 338–350. doi:10.1109/ TITB.2002.804136

Xing, L., Thorndyke, B., & Schreibmann, E. (2006). Overview of image-guided radiation therapy. Medical Dosimetry, 31(2), 91–112. doi:10.1016/j.meddos.2005.12.004

Suri, J. S., Liu, K., & Singh, S. (2002). Shape recovery algorithms using level sets in 2-D/3-D medical imagery: A state-of-the-art review. IEEE Transactions on Information Technology in Biomedicine, 6(1), 8–28. doi:10.1109/4233.992158

KEY TERMS AND DEFINITIONS

Ueda, T., Mori, K., & Minami, M. (2006). Trends in oncological CT imaging: Clinical application of multidetector-row CT and 3D-CT imaging. International Journal of Clinical Oncology, 11, 268–277. doi:10.1007/s10147-006-0586-1 Vemuri, B. C., Ye, J., Chen, Y., & Leonard, C. M. (2003). Image registration via level-set motion: Applications to atlas-based segmentation. Medical Image Analysis, 7, 1–20. doi:10.1016/ S1361-8415(02)00063-4 Viola, P., & Wells, W. M., III. (1995). Alignment by maximization of mutual information. In Int. Conf. Computer Vision, IEEE Computer Society Press, Los Alamitos, CA (pp. 16-23). Wang, H., Dong, L., & O’Daniel, J. (2005). Validation of an accelerated demons algorithm for deformable image registration in radiation therapy. Physics in Medicine and Biology, 50, 2887–2905. doi:10.1088/0031-9155/50/12/011 Wyatt, P. P., & Noble, J. A. (2003). MAP MRF joint segmentation and registration of medical images. Medical Image Analysis, 7, 539–552. doi:10.1016/S1361-8415(03)00067-7 Xie, H., Li, G., Ning, H., et al. (2004). 3D voxel fusion of multi-modality medical images in a clinical treatment planning system. In R. Long, S. Antani, D.J. Lee, B. Nutter, M. Zhang (Eds.), Proc. Computer-Based Med. Sys., IEEE Computer Science Society (pp. 48-53).

3D Medical Imaging: A process of obtaining a 3D volumetric image composed of multiple 2D images, which are computer reconstructed using a mathematical “back-projection” operation to retrieve pixel data from projected image signals through a patient, detected via multichannel detector arrays around the patient. 4D Medical Imaging: A process of acquiring multiple 3D images over time prospectively or retrospectively, so that patient motions and changes can be monitored and studied. Imaging Modality: A type of medical imaging technique that utilizes a certain physical mechanism to detect patient internal signals that reflect either anatomical structures or physiological events. Image Processing: A computing technique in which various mathematical operations are applied to images for image enhancement, recognition, or interpretation, facilitating human efforts. Image Registration: A process of transforming a set of patient images acquired at different times and/or with different modality into the same coordinate system, mapping corresponding voxels of these images in 3D space, based on the underlying anatomy or fiducial markers. Image Segmentation: A process in which an image is partitioned into multiple regions (sets of pixels/voxels in 2D/3D) based on a given criterion. These regions are nonoverlapping, homogeneous with respect to some characteristics such as intensity or texture. If the boundary constraint of the region is removed, the process is defined as classification.

893

3D and 4D Medical Image Registration Combined with Image Segmentation and Visualization

Image Visualization: A process of converting (rendering) image pixel/voxel into 2D/3D graphical representation. Most computers support 8-bit (256) grayscale display, sufficient to human vision that can only resolve 32-64 grayscale. A common 12/16-bit (4096/65536 grayscales) medical image can be selectively displayed based on grayscale classification. Window width (display range in grayscale) and linear level function (center of the window width) are frequently used in adjusting display content.

This work was previously published in Encyclopedia of Healthcare Information Systems, edited by Nilmini Wickramasinghe and Eliezer Geisler, pp. 1-9, copyright 2008 by Medical Information Science Publishing (an imprint of IGI Global).

894

895

Chapter 3.18

Time-Sequencing and ForceMapping with Integrated Electromyography to Measure Occlusal Parameters Robert B. Kerstein1 Tufts University School of Dental Medicine, USA

ABSTRACT Computerized Occlusal Analysis technology records, and quickly displays for clinical interpretation, tooth contact timing sequences of .003 second increments, and each tooth contacts’ fluctuating force levels which occur during functional jaw movements. These measurements are recorded intraorally with an ultra-thin, mylar-encased sensor that is connected to a computer via a USB interface. This sensor is placed between a patients’ teeth during functional jaw movements to record DOI: 10.4018/978-1-60960-561-2.ch318

changing tooth-tooth interactions. The displayed occlusal data aids in the examination and treatment of occlusal abnormalities on natural teeth, dental prostheses, and dental implant prostheses. The software can be linked to an Electromyography software program that simultaneously records the electromyographic potential of 8 head and neck muscles. This combination of dynamic tooth contact force and time data, and functional muscular data, affords a dentist detailed, precise, and unparalleled diagnostic and treatment information, with which to address many differing clinical occlusal pathologies.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

INTRODUCTION Computerized Occlusal Analysis technology (TScan III for Windows®, Tekscan Inc., S. Boston, MA, USA) records, and quickly displays for clinical interpretation, tooth contact Time-Sequences while simultaneously Force-Mapping each tooth contacts’ fluctuating relative occlusal force levels which occur during functional jaw movements (known as Occlusal Events) (Maness 1993, Montgomery and Kerstein 2000, Kerstein 2001, Kerstein and Wilkerson 2001). These occlusal event measurements are recorded intraorally, with an ultra-thin, electronically charged, mylarencased sensor that is connected to a computer via a USB interface. By measuring Relative Force, the T-Scan III system can detect whether an occlusal force on one set of contacting opposing teeth is greater, equal to, or less than, the occlusal forces occurring on other contacting teeth all throughout the dental arches. Determining relative force is important to the user-Dentist, as relative force illustrates measured differences of varying applied loads within all contacting tooth locations. The T-Scan III software displays the relative occlusal force, at any instant within a recorded occlusal event on all contacting teeth, as a percentage of the maximum occlusal force obtained within the recording. Detected relative force variances can be employed clinically to: •

•

Precisely balance an unbalanced occlusal force distribution with targeted time-based and force-based occlusal adjustments Diagnose excessively high occlusal load present in one area of the occlusion, while simultaneously diagnosing where there is little, moderate, or no occlusal load in other areas of the same occlusion

After various occlusal events are recorded, the retrieved occlusal contact time-sequence data and

896

relative force data are mapped for analysis in 4 different quantifiable ways: •

•

• •

By real-time increments that measure the sequential order of each individual tooth contacts’ loading occurrence. The realtime sequence can be separated into increments as short as .003 seconds duration By force location on the contacting surfaces of many teeth individually and collectively within the 2 dental arches By relative changing force percentage values on all individual contacting teeth By a changing, moveable occlusal force summation icon and trajectory, that describes the changing concentrated force position within the 2 dental arches of all the collective tooth contacts’ relative occlusal forces

The mapped contact time-sequence and relative occlusal force data, when analyzed by a user-Dentist, aids in the clinical determination of, and occlusal treatment of, many time premature contacts, and occlusal force abnormalities, which occur during occlusal events on natural teeth, dental prostheses, and dental implant prostheses (Kerstein Chapman and Klein 1997, Kerstein 1997, Kerstein and Wilkerson 2001, Kerstein and Grundset 2001,). This technology’s ability to isolate timepremature tooth contacts, and excessive occlusal contact forces, is vastly superior to the commonly utilized, non-technology based occlusal indicators which dentists routinely employ (articulating paper, wax imprints, silicone imprints, and articulated stone dental casts). None of these dental materials have any scientifically proven capability to time-sequence tooth contacts, or force-map occlusal forces. Additionally, they all necessitate the user-Dentist to “subjectively interpret” their occlusal representations. The most commonly employed occlusal indicator is dental Articulating Paper. It is comprised of

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

ink-impregnated strips of either Mylar, or various thicknesses and forms of papers. To date, no study has shown Articulating Paper to be reliably accurate when it is used to describe occlusal forces (Reiber Fuhr and Hartmann 1991, Millstein and Maya 2001, Carossa Lojacono Schierano and Pera 2001, Saraçoğlu and Ozpinar 2001, Carey Craig Kerstein and Radke 2007, Saad Weiner and Ehrenberg 2008). Therefore, user-Dentist’s subjective interpretation of the size and shape, and color intensity of various Articulating Paper markings, can lead to incorrect clinical determinations of where in the dental arches occlusal problems exist, and where occlusal treatment should be performed or not performed. These incorrect clinical determinations occur because the user-Dentist is taught (incorrectly), to perceive widespread and uniform shaped markings’ appearance on the teeth as representative of “time simultaneous” tooth contact sequences with “normal and balanced” occlusal forces present within the dental arches.

EVOLUTION OF TIME SEQUENCING AND RELATIVE OCCLUSAL FORCE-MAPPING BY COMPUTERIZED TECHNOLOGY The evolution of pressure sensitive ink-Mylar encased sensor technology that could time-sequence and force-map relative occlusal forces, was introduced as the T-Scan® I computerized occlusal analysis system (Maness Benjamin Podoloff and Bobick and Golden 1987). Occlusal data was obtained by instructing patients to occlude through an intraoral recording sensor that was connected to a stand-alone computer. The computer screen displayed a Force Movie (Maness 1988), which illustrated a dynamic columnar time and force display of the recorded occlusal contacts for playback and analysis. The original T-Scan I Recording Sensor was comprised of a flexible laminated epoxy matrix surrounding a pressure sensitive ink grid, which

was formed in the shape of a dental arch. When inserted intraorally, and occluded into and through by a patients teeth, the sensor relayed occlusal contact real-time and relative force-mapping information to compatible software that was capable of interpreting 16 levels of intraoral relative force in approximately .01 second time increments. The resultant occlusal analysis was displayed in two or three dimensions, as a Snapshot (Maness 1988), or as a continuous Movie (Maness 1988) of the entire occlusal contact event the patient made. The data could be played forwards, backwards, or in individual .01 second time increments. In 1998, the entire T-Scan I system was redeveloped as the T-Scan II Occlusal Analysis System for Windows®. The changes included hardware, software, and further sensor advances. T-Scan II was a Microsoft Windows® (Microsoft Corp, Seattle WA, USA) compliant system that was integrated into a clinical diagnostic computer workstation. An IBM compatible PC with a Pentium processor, and a minimum of 4-8 megabytes of RAM, were required to properly operate the system (Figure 1 and Figure 2). The graphical interface used familiar Windows® toolbar icons, to display the software features that analyzed recorded occlusal contact information. In 2006, the T-Scan II hardware was altered to interface with the PC via a USB plug that replaced a serial parallel port interface. Then, in 2008, the T-Scan III USB with Turbo Mode Recording was introduced. A recording handle hardware improvement resultant from increased speeds with which computer chips process commands, made it possible to significantly increase the recording speed of a given occlusal event. The Evolution Handle of the T-Scan III USB System, when operated in Turbo Mode, can record timesequences and map relative occlusal force data in increments of .003 seconds, thereby capturing 3 times more occlusal data for analysis than was possible with the T-Scan III USB. Turbo Mode data acquisition provides a user-Dentist with an increased ability over the conventional non-

897

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

Figure 1. T-Scan III Evolution Handle with sensor

Turbo Mode recording, to locate more non-simultaneous tooth contact sequences and aberrant occlusal force concentrations within a patient’s occlusion.

T-SCAN III SENSOR DESIGN The most current available recording sensor for the T-Scan III system represents the 4th generation of T-Scan system sensors. It is known as the High Definition sensor (HD). The HD sensor is dental arch-shaped such that it fits between a patient’s maxillary and mandibular teeth during the recording of occlusal events. When correctly positioned between a patients 2 maxillary Central Incisor teeth, and held stable by a hard plastic Sensor Support (Figure 1), the sensor is then clenched upon, and chewed over by a patient, so as to record and precisely map changing toothto-tooth occlusal force interactions in real-time. A number of different occlusal events can be captured by the recording sensor:

898

Figure 2. T-Scan III Evolution Handle connected to a personal computer

• • •

• •

Single or multiple patient-self closures into complete tooth intercuspation Dentist-assisted patient closure into the Centric Relation mandibular position Patient self-closure into complete tooth intercuspation followed by a mandibular excursion to the left, right, or forward, out of the complete intercuspated position Patient clenching and grinding motions in and around complete tooth intercuspation Habitual Occlusal Force Patterns

The HD sensor’s structural design consists of two layers of Mylar encasing a grid of resistive ink rows and columns printed between them (Figure 3 and Figure 4). The high definition design has both larger Sensels (the Tekscan proprietary force recording element within each sensor; a sensel is 1.61 mm2) and less space between the sensels than all of the previous T-Scan I and II sensor designs. Within the HD sensor, the sensels are packed closer together within the recording grid, resulting in

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

Figure 3. Sensor schematic

Figure 4. HD sensor

mean force variances were significantly larger than for the same 6 occlusal contact locations measured with the HD sensor. Additionally, the HD sensor exhibited consistent force reproduction in at least 20 in-laboratory crushing cycles (Kerstein Lowe Harty and Radke 2006).

SENSOR RECORDING DYNAMICS

increased recording surface area available for teeth to occlude upon. When compared to all pervious sensor designs, the HD configuration specifically limits the amount of inactive, non-recording grid areas between the sensels. A published performance study (Kerstein Lowe Harty and Radke 2006) comparing Generation-3 (G3) sensors against HD sensors revealed that the HD sensor reproduced more consistent force levels than the older G-3 sensor. The G-3 sensor

When tooth contact is made with a T-Scan III sensor, the applied force on individual sensels changes their resistance. Larger applied forces produce larger resistance changes and lower occlusal forces produce lesser resistance changes. The electronics in the Recording Handle of the T-Scan III system excites the sensor by applying voltage to each column of recording sensels in succession, Then, as a T-Scan III sensor is occluded upon, a change in the applied force at various tooth contacts results in a change in the resistance of the resistive ink at each of the contacted sensels. This resistance change is measured by the system’s hardware electronics as a change in Digital Output Voltage (DO). Higher forces on contacting teeth result in larger decreases in the resistance of the loaded sensels, and therefore, a higher measured output voltage from the loaded sensels. The greater the applied force to the contacting teeth then the higher the output voltage that is measured.

899

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

The electronics in the recording handle reads the output voltage of each row of sensels in a process called Scanning. The output signal is conditioned and converted to an 8-bit digital value so that the measured resistance change is proportional to force applied to each sensel with a range between 0 and 255 raw counts. The raw count for an individual sensel can be conceptualized as the minimum pressure that is on that sensel. And, because each sensel has an area of 1.61 mm2, the raw count can also be conceptualized as a force. A map in the software receives from the recording handle, the stream of changing sensel digital voltage output values, and organizes them for display to the user-Dentist into the same orientation pattern that the sensels have on the sensor. Within the software, the user-Dentist can raise or lower the Recording Sensitivity across 16 differing levels to compensate for varying occlusal strengths of individual patients. This allows the user-Dentist to modify the sensor recording behavior to select a pressure range which will record optimum individual patient occlusal data for analyses, regardless of a patients’ absolute occlusal strength.

RELATIVE FORCE MEASUREMENT VS. ABSOLUTE FORCE MEASURED IN ENGINEERING UNITS As stated in the Introduction, the T-Scan III system records relative force. It therefore, does not provide absolute occlusal force in engineering units (calibrated force numbers) such as in Newtons (n) or pounds per square inch (lb/in2). To obtain accurate numeric force values in engineering units, a T-Scan sensor would need to undergo pre-use Equilibration, some surface preparationand thenCalibration: •

900

Sensor Equilibration is accomplished by applying a uniform load all over a sensor at three levels (low, near the load of interest,

•

•

and at a high load) to determine that each sensel has a uniform digital output from a uniform applied load. This can be accomplished by using an air bladder with the sensor being compressed on a flat surface. Next, the recording sensor would need to be covered with a shim material – The shim would prevent sensor crinkling when loaded, while at the same time, distribute the load amongst a greater sensor surface area than the surface areas of individual tooth contact points. The shim helps avoid individual sensor saturation. This shim material increases the sensor thickness from approximately 0.1 mm (no shim) to about 3 mm (with shim). This is a “clinical use problem” for the user-Dentist because a modified sensor thickness increase does not allow a patient’s teeth to completely and fully intercuspate against each other. Lastly, the sensor must undergo sensor with shim Calibration - This would be performed on a set of articulated dental casts made from a patient’s dental arches, by applying 1 or 2 representative occlusal loads to a modified sensor that are within the patient’s occlusal force range. Calibration is performed to correlate the digital output of the sensels to engineering units (force or pressure) and confirms and improves the uniformity of output from sensel to sensel throughout the entire recording grid.

Because individual patients have variable occlusal strengths, and the dental community continually resists the use of a thickened sensor which interferes with complete tooth contact intercuspation, and there are no commercially available dental calibration devices with which to apply known test loads to a modified T-Scan sensor, it is not possible to use the T-Scan III system to obtain force values in engineering units. Therefore, throughout the remainder of this chapter, all references made forward from

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

here regarding occlusal force are to be describing relative occlusal force. For any measurement of force in engineering units, the I-Scan system is commercially available (I-Scan for Windows®, Tekscan Inc. S Boston, MA. USA). I-Scan is a general purpose research product designed for industrial applications that lacks many software features which aid in the understanding of dental occlusion (such as those related to time-sequencing analyses). Alternatively, the I-Scan’s software features include equilibration, calibration, general purpose analysis graphs, and the ability to export arrays of X–Y sensel pressure data into ASCII format into other applications. These features make the I-Scan a more versatile absolute force analysis system than is the T-Scan III system.

RECORDED T-SCAN III MOVIE ANALYSES A T-Scan III recording of an occlusal event is captured in a dynamic Movie (Maness 1988) which reveals all the varying occlusal contact forces, as time evolves within a given occlusal event. Once recorded, the Movie is retrieved for review in a 2-Dimensional View window, a 3-Dimensional View window, and two Force vs. Time Graph windows (Figure 5). The mapped occlusal force data is converted into colors so each frame of the Movie is an image of the pressure at varying tooth contact locations at that instant of time. This pressure results from the tooth contact forces that dissipate between opposing teeth as they interact during a recorded occlusal event. By advancing the Movie or going back in time, all force changes that occur between sets of interacting teeth are observed in the exact order with which they occurred throughout the entire recorded occlusal event. A Legend with Color Spectrum (Figure 5) depicts a scale of relative occlusal forces. Darker, cooler colors, such as dark blue through light blue,

Figure 5. Legend with color spectrum

represent lower forces. Brighter, warmer colors such as yellow, orange and red, indicate higher occlusal contact forces. Pink indicates that the sensel has saturated, or reached its maximum digital output. Pink sensels may have an unknown amount of greater force than red sensels. In 2 or 3 dimensions, the occlusal contact timing sequence can be played forwards or backwards continuously, or in either .01-second increments when recorded non-Turbo Mode, or in .003 second increments when recorded in Turbo Mode. In the 3 dimensional (3-Dimensional ) playback window the force columns change both their height and color designation. In the 2 dimensional (2-D) contour view, the color-coded force concentration zones change size, shape, and color, as occlusal forces change. Below the 2-Dimensional view and 3-Dimensional view windows are 2 Force vs. Time Graph windows (Figure 6). It is here where TimeSequencing is mapped incrementally, through changing Red and Green colored lines that represent the changing force percentages found in both dental arch halves over the entire recorded occlusal event. 4 additional moveable, vertical, hyphenated lines denote key Time-Regions within a recorded occlusal event. Lastly, a Data Window to the far

901

Time-Sequencing and Force-Mapping with Integrated Electromyography to Measure Occlusal Parameters

Figure 6. T Scan III playback windows

right describes the occlusal Time Parameters that can be calculated from the recorded occlusal event. The Occlusal Time Parameters that can be displayed for analyses within the graph Data Window are: •

•

902

Occlusion Time (OT): The elapsed time from 1st tooth contact to the complete intercuspation of all occluding teeth.