Integrative clinical decision support systems: foundations and implementations

The book is concerned with theoretical and methodological foundations of clinical decision support systems (CDSSs), i.e., systems designed to help healthcare professionals make clinical decisions. It is focused on patient-specific CDSSs that help physicians make diagnostic and treatment decisions directly at the point of care and it addresses theoretical and methodological aspects of their design and implementation. The book introduces the notion of integrative clinical decision support that covers all tasks from the clinical decision making process and proposes a theoretical architecture for integrative CDSSs (ICDSSs), called the MET3 architecture. The architecture combines the principles of multi-agent systems with ontology-driven design in order to distribute the complex support functionality among independent entities and to represent different types of clinical knowledge in form of ontological models. The MET3 architecture is a theoretical solution that provides a common platform for other methodologies, algorithms and frameworks proposed in this book.
Responding to the requirement for comprehensive and accurate diagnostic decision models, modifications of rule induction algorithms rooted in rough set theory have been proposed. These algorithms allow generating robust decision rules from inconsistent and incomplete data. The book also describes their combination that successfully deals with imbalanced distribution of decision classes that is typical for clinical data. Addressing the need for accurate and efficient retrieval of relevant clinical evidence, a framework for enhanced indexing and automatic retrieval of evidence-based documents (systematic reviews of clinical trials) has been introduced. The framework allows for retrieval of documents that are relevant in the current context of a patient-physician encounter. Retrieved evidence augments diagnostic and treatment suggestions and enables physicians to make more informed decisions. Finally, principles guiding the design of interactions in CDSSs and ICDSSs have been provided. These principles have been derived from task-oriented and user-oriented design approaches, and when formulating them, special attention was paid to modern mobile computing platforms, including handhelds devices. The MET3 architecture and the remaining proposed methodological solutions were successfully validated in practice. Proof-of-concept ICDSSs for two clinical problems (pediatric scrotal pain and asthma exacerbations) were developed, diagnostic decision models for these problems were constructed, and finally the indexing and retrieval framework for asthma exacerbations was implemented. Results of this verification, as well as experience with designing and running a clinical trial of a CDSS, is described in details in this book.

Acronyms and abbreviations 5 Abstract 7 1. Scope and objectives 9 1.1. Scope 9 1.2. Motivations 10 1.3. Objectives 12 1.4. Book organization 14 1.5. Acknowledgments 14 2. An architecture for an ICDSS and its theoretical foundations 15 2.1. Introduction 15 2.2. Conceptual model of decision making implying the MET3 architecture 16 2.3. Related research on patient-specific CDSSs 17 2.4. Methods for developing the MET3 architecture 19 2.5. The MET3 architecture 21 2.5.1. Requirements 21 2.5.2. Goal model 23 2.5.3. Domain ontology 24 2.5.4. Agent model 26 2.6. Discussion 28 3. Methodologies for constructing diagnostic decision models 33 3.1. Introduction 33 3.2. Rough set theory 35 3.2.1. Handling inconsistent and incomplete objects 35 3.2.2. Identifying important condition attributes and assessing their information value 38 3.2.3. Discovering rules in decision tables 40 3.2.4. Using decision rules to classify new objects 45 3.3. Proposed methods for induction of robust decision rules 47 3.3.1. Modifications of existing rule induction algorithms 47 3.3.2. Combination of LEM2* and EXPLORE* to handle imbalanced data 50 3.4. Experimental evaluation of the proposed methods on clinical data 52 3.5. Discussion 55 4. A framework for indexing and retrieval of context-specific evidence 57 4.1. Introduction 57 4.2. Related research 59 4.3. The Cochrane Library 61 4.4. Proposed methods 62 4.4.1. Design of the indexing and retrieval framework.62 4.4.2. Phase 1: identification of core index concepts 65 4.4.3. Phase 2: indexing of evidence-based documents 66 4.4.4. Phase 3: retrieval of evidence-based documents 67 4.5. Discussion 69 5. Principles of designing interactions for an ICDSS 73 5.1. Introduction 73 5.2. Related research on interactions in CDSSs 74 5.3. Methodological foundations of interaction design 75 5.3.1. Framework of interactions 75 5.3.2. Scenario-based design methodology 76 5.3.3. Object-Action-Interface model 76 5.3.4. Eight Golden Rules of Interface Design 77 5.4. Proposed principles for interactions in ICDSSs 77 5.5. Discussion 82 6. Practical applications of the proposed methodologies 85 6.1. Introduction 85 6.2. Implementation of the MET3 architecture 85 6.3. Generic process of constructing diagnostic decision models 87 6.4. Construction of the diagnostic model for pediatric scrotal pain 89 6.4.1. Description of the clinical problem 89 6.4.2. Collected retrospective data 90 6.4.3. Development of possible decision models 91 6.4.4. Evaluation of selected decision models 93 6.5. Construction of the diagnostic model for pediatric asthma exacerbations 94 6.5.1. Description of the clinical problem 94 6.5.2. Collected retrospective data 95 6.5.3. Development of possible decision models 98 6.5.4. Evaluation of selected decision models 99 6.6. Implementation of the evidence indexing and retrieval framework for pediatric asthma exacerbations 100 6.6.1. Identification of core index concepts 100 6.6.2. Indexing evidence-based documents 102 6.6.3. Retrieving evidence-based documents 104 6.6.4. Evaluation of the automatic indexing 105 6.7. Discussion 107 7. Practical evaluation of a CDSS 109 7.1. Introduction 109 7.2. Methods 111 7.2.1. Trial design and implementation 111 7.2.2. Patient enrollment and follow-up 113 7.2.3. Outcome measures and analysis 114 7.3. Results 115 7.3.1. Enrolled patients 115 7.3.2. Accuracies of MET2-AP and physicians 116 7.3.3. Inter-observer agreement 116 7.4. Discussion 118 8. Conclusions 121 8.1. Results and contributions 121 8.2. Future research directions 122 References 123 Streszczenie 137