Dealing with Medical Knowledge Computers in Clinical Decision Making Dealing with Medical Knowledge Computers in Clinical Decision Making Tibor Deutsch Semmelweis University of Medicine Budapest, Hungary Ewart Carson City University London, England and Endre Ludwig P€terfy Teaching Hospital Budapest, Hungary Springer Science+Business Media, LLC Library of Congress Cataloging-in-Publication Data On file ISBN 978-1-4757-9953-8 ISBN 978-1-4757-9951-4 (eBook) DOI 10.1007/978-1-4757-9951-4 © 1994 Springer Science+Business Media New York Originally published by Plenum Press, New York in 1994 Softcover reprint of the hardcover 1st edition 1994 All rights reserved No part of this book may be reproduced s,tored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise, without written permission from the Publisher Preface "Dealing with complexity" is a phrase that neatly summarizes many of the issues associated with clinical decision making and problem solving. In fact, Dealing with Complexity was the very title of another book published by Plenum, one of the authors of which contributed to this new volume. That first book was devoted entirely to the nature of complex problems in various fields of science, manage ment, and technology and the approaches that are available for their resolution (Flood and Carson, 1993). In some respects, this book is a descendant of the earlier volume in that it focuses on problems that clinicians are facing when making clinical decisions. The individual clinician, or members of the clinical team, are required to process and interpret, in context, large quantities of complex data relating to the patient, often derived from diverse sources. In this way, data are transformed into information that is used to effect the appropriate decision, be it in relation to diagnosis, treatment, or monitoring the progress of the patient over a period of time. The complexity of the issues involved is such that the ability to make appropriate decisions and solve problems is limited to those who have the necessary expertise. However, the impact of the information revolution means that clinical decision making and problem solving can now be assisted using knowledge-based approaches; this decision-assisting capability is implemented on computers, which are providing increasingly powerful performance at ever decreasing real cost. In this way, the ability to make decisions can be made more widely available, in effect enabling the individual to operate at a higher level of expertise. This role of knowledge-based techniques in relation to clinical decision making is one facet of the growing subject of medical informatics. Medical informatics, in essence, is the application of information science and technology in clinical research and education and in clinical practice. The concepts, methods, and techniques embedded within knowledge-based approaches constitute power ful aids to clinical decision making, particularly when set within a clear control system framework, with its emphasis on the patient-clinician feedback model. v vi Preface In order to develop and disseminate these knowledge-based approaches in relation to clinical decision making and problem solving, there is the need for an interdisciplinary forum drawing together on the one hand clinicians and on the other engineers and computer scientists. There is the need for greater understand ing on the part of both groups. For the clinician, there is the need to understand the nature of these new approaches and the way in which they can be used to assist in clinical decision making when set within the control-system framework. For the engineer and computer scientist, there is the need to understand the true nature of the clinical problems that require solution using these approaches. It is these twin needs that this book addresses. This is achieved first by providing a guided tour (from a bird's-eye perspective) that highlights the general nature of the problems and issues involved and then by effectively dropping in on a number of places of interest on this tour in order to examine some representative clinical examples in greater detail. This book differs from the likes of the classic medical informatics compendia of Gremy (1987), Shortliffe and Perrault (1990), and Degoulet et al. (1992) and focuses only on that area of medical informatics that is concerned with clinical decision making. This means that topics such as signal analysis, image process ing, and biomedical engineering in general are deliberately not included. The manner in which the subject matter is presented is different from that adopted in a number of edited volumes that are available (for example, Szolovits, 1982; Carson and Cramp, 1985; Reggia and Thhrim, 1985; P. L. Miller, 1988). They convey information by presenting some of the classic papers of this field rather than providing a systematic discussion. The aim of this book differs also from those that focus either on a specific area of medical decision making, such as medical diagnosis (for inst~ce, Warner, 1978; Sox et al., 1988) or on using a specific method, such as statistical decision making (Grenier, 1990), decision trees (Weinstein and Fineberg, 1980), fuzzy logic (Fieschi, 1990), or artificial intel ligence (de Lotto and Stefanelli, 1985; Keravnou, 1992). In contrast to these other works with their own particular approaches to presenting the subject, this volume is intended to provide a systematic and unified presentation of problems associated with clinical decision making and methods that can be used in assisting their solution. It should be noted, however, that it is not a technical treatise on decision-making methods. Rather, it aims to convey the flavor of clinical decision-making issues in terms of frameworks and ap proaches that are available, accessible, and relevant. Central to the theme of the book is the synergy that results from the bringing together of the classic control-system approach, which views the interaction between clinician and patient in terms of a feedback model, with the symbolic approaches that have been central to developments in artificial intelligence. Stress has been placed on concepts and examples. This means that only a very limited mathematical treatment is included. Equally, the examples considered essentially Preface vii represent classes of clinical problems rather than provide complete coverage of the clinical domain. Again, within the chosen examples, it is the informatics "mes sage" that is highlighted rather than the use of the "medium," which is the computer. There is no lengthy computer code in the text. Medical knowledge can be likened to a "living" system in which the concepts (diseases, symptoms, drugs, etc.) constitute its anatomy while the means of dealing with such concepts, that is, information processing and decision making, correspond to different "metabolic" and "physiological" processes (Schoolman, 1982). A particular feature of the style of presentation adopted is to regard clinical decision making as an information-processing "living system" that has its anatomy (structure of concepts that are used), physiology (how it is to be manipulated), and pathophysiology (what happens if it is manipulated inappro priately). Problem solving also requires intensive knowledge "metabolism" whereby pieces of concepts and relations (as catabolized and stored products) are resynthesized (anabolized) when more complex constructs and reasoning are needed. The techniques described in this volume in some respects resemble the "enzymes" that facilitate this knowledge synthesis needed for solving medical problems. Chapter 1 introduces a number of the features of the control-system approach through the description of a clinical case report. This serves to motivate the reader for whom such an approach might be unfamiliar. Chapter 2 focuses on the fundamentals of systems and control, introducing concepts that are highly relevant as aids to understanding the complex nature of clinical situations. This is followed, in Chapter 3, by a general discussion of data and knowledge representation and manipulation, outlining general methodologies that can be used when tackling clinical information-processing and problem-solving tasks. The tools that emerge from these two chapters are then employed in Chapter 4 in analyzing clinical activity (the therapeutic process). The anatomy of medical knowledge is the focus of Chapter 5, providing a formalism in terms of which changes in patient state and/or function can be analyzed and of which schemes for intervention (control) can be assessed. This leads naturally on to Chapter 6, which discusses methods for computer-assisted clinical decision making. Chapters 7-9 focus on clinical application. The several types of clinical activity are represented, including diagnosis and assessment of the patient's state, therapy planning, and the monitoring of ongoing therapeutic regimes. Chapter 7 describes methods for medical diagnosis. Chapter 8 focuses on the planning of therapeutic action, including the computer-assisted planning of drug dosage, Finally, Chapter 9, as a "firework," provides examples of clinical problem solving in a particular medical field, diabetes management, that serve to illustrate how wide is the range over which doctors may find decision-support tools to be useful assistants. viii Preface In preparing this book for publication, we are indebted to many of our colleagues who, in their various ways, have helped to bring it to reality. First, we are grateful to members of our own respective institutions for having encouraged and supported an endeavor in applying knowledge-based approaches, within a control-system framework, to a range of clinical problems. We also wish to thank those colleagues who have worked with us in many projects over the past seven years. Much of the material in this book is the result of collaboration between systems scientists, engineers and computer scientists, and members of the relevant clinical professions. To all those who have catalyzed and supported this fruitful interdisciplinary activity we extend our sincere gratitude. Special thanks must be expressed to a number of friends and colleagues, Derek Cramp, Ivan Futo, Istvan Hermanyi, Roman Hovorka, Eldon Lehmann, Emma Nicolosi, Abdol Roudsari, Peter Sonksen, Ron Summers, and Gyula Tamas, who have worked with one or more of us and whose stimulating ideas and enthusiasm have helped to maintain the momentum of our progress. The preparation of this book would not have been possible without regular travel between Budapest and London. Funding for these visits was provided by the British Council, the Science and Engineering Research Council of the United Kingdom, and the Wellcome Trust. We are most grateful to all of them for their generous support. The authors are indebted to Csaba Stanka for his technical assistance in preparing the manuscript and the figures. TIBOR DEUTSCH EWART R. CARSON ENDRE LuDWIG Budapest and London, February 1994 Contents 1. Introduction 1. 1. Case Report ........................................... . 1.1.1. Patient History and Clinical Data ................... 1 1.1.2. Decisions, Actions, and Patient Response. . . . . . . . . . . . . 2 1.1.3. Commentary .................................... 4 1.2. Role of Computers in the Therapeutic Process ................ 5 2. Systems and Control 2.1. Introduction............................................ 11 2.2. Basic Concepts in Systems Dynamics ................... . . . . 11 2.3. Systems and Modeling Methodology ....................... 14 2.4. Control Loops.......................................... 17 2.5. Clinical Context ........................................ 19 2.6. Summary .............................................. 20 3. Knowledge Representation and Manipulation 3.1. Introduction............................................ 23 3.2. Representing Uncertainty ................................. 24 3.3. Sets .................................................. 26 3.4. Representing Objects and Relationships ..................... 29 3.4.1. Frames......................................... 29 3.4.2. Databases ...................................... 31 3.4.3. Graphs......................................... 32 3.4.4. Logic .......................................... 36 3.5. Representing Dynamic Knowledge ......................... 39 3.6. Manipulating Knowledge ................................. 43 3.6.1. Manipulating Probabilistic Knowledge (Bayes's Rule) ... 44 ix x Contents 3.6.2. Rule-Based Reasoning ............................ 47 3.6.3. Manipulating Knowledge Stored in Models ........... 52 3.7. Problem Solving ........................................ 54 3.7.1. Knowledge-Based versus Algorithmic Problem Solving 56 3.7.2. Architectures for Knowledge-Based Problem Solving ... 58 3.8. Summary.............................................. 63 4. Analysis of the Therapeutic Process 4.1. Introduction............................................ 65 4.2. Basic Medical Concepts .................................. 67 4.3. Describing the Patient's Illness ............................ 69 4.4. Clinical Problem Solving ................................. 71 4.5. Control-System Model of Therapeutics. . . . . . . .. . . . . . . . . . . . . . 74 4.6. The Therapeutic Process Seen from a Systems Perspective ..... 79 4.7. Summary.............................................. 81 s. Anatomy of Medical Knowledge 5.1. Introduction............................................ 83 5.2. Declarative Medical Knowledge ........................... 84 5.2.1. Coding Schemes ................................. 85 5.3. Representing Preclinical Knowledge ........................ 87 5.3.1. Anatomical Classification. . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.3.2. Describing Physiological Systems. . . . . . . . . . . . . . . . . . . 88 5.4. Representing Clinical Knowledge .......................... 95 5.4.1. Pathophysiology-Based Organizations of Clinical Knowledge ..................................... 96 5.4.2. Findings and Pathophysiological States. . . . . . . . . . . . . . . 96 5.4.3. Diseases........................................ 101 5.4.4. Tests and Therapies .............................. 105 5.4.5. Diseases as System FauIts ......................... 106 5.4.6. Representing Case-Based (Associational) Clinical Knowledge ..................................... 110 5.4.7. Characterizing Diagnostic Tests . . . . . . . . . . . . . . . . . . . . . 113 5.5. Suminary.............................................. 118 6. Methods for Computer-Assisted Clinical Decisions 6.1. Introduction............................................ 119 6.2. The Need for Assistance in Decision Making. . . . . . . . . . . . . . . . . 120