Development of Clinical Decision Support Systems Using Bayesian Networks : With an Example of a Multi-Disciplinary Treatment Decision for Laryngeal Cancer

For the development of clinical decision support systems based on Bayesian networks, Mario A. Cypko investigates comprehensive expert models of multidisciplinary clinical treatment decisions and solves challenges in their modeling. The presented methods, models and tools are developed in close and intensive cooperation between knowledge engineers and clinicians. In the course of this study, laryngeal cancer serves as an exemplary treatment decision. The reader is guided through a development process and new opportunities for research and development are opened up: in modeling and validation of workflows, guided modeling, semi-automated modeling, advanced Bayesian networks, model-user interaction, inter-institutional modeling and quality management. Contents Patient-specific Bayesian Network in a Clinical Environment TreLynCa: A Tumor Board Decision Model for Laryngeal Cancer Model Validation and Tools for Guided BN Modeling GUI for PSBN-based decision verification Target Groups Scientists and students in the field of medical informatics, computer science, medicine and psychology About the Author Dr.-Ing. Mario A. Cypko completed his PhD at the Computer Science department of the University of Leipzig, Germany. He was a postdoctoral research fellow in the Human Research Office of the European Space Agency in the Netherlands. He is currently a postdoctoral research assistant at the German Heart Center Berlin, Germany Acknowledgments Contents List of Figures List of Tables List of Abbreviations List of Equations Part I Basic Consideration Chapter 1 Introduction 1.1 Motivation 1.2 Objective 1.3 Method 1.4 Thesis Outline Chapter 2 Tumor Board Decision for Larynx Cancer Patients 2.1 Laryngeal Cancer 2.1.1 Epidemiology 2.1.2 Anatomy and Physiology 2.1.3 Cancer and Diagnostic 2.1.4 Therapy Decision of Laryngeal Cancer 2.2 Best Practice of Clinical Judgment 2.3 Head and Neck Tumor Board 2.3.1 Aims of a Tumor Board 2.3.2 Head and Neck TPU at Leipzig University Hospital Chapter 3 Development of a Clinical Decision Support System Introduction 3.1 Machine Intelligence in CDSSs 3.1.1 Non-knowledge-based System 3.1.2 Knowledge-based System 3.2 Impact of BN-based CDSS 3.3 Best Practice in Developing a CDSS 3.4 Integrated (model-based) Patient Care with MIMMS Chapter 4 Clinical Decision Model using Bayesian Networks 4.1 Bayesian Network 4.1.1 Bayesian Network Model 4.1.2 Inferencing 4.1.3 Methods for Model Analysis 4.1.4 Complexity 4.2 Model Design 4.2.1 Selection of Variables, States and their Identifiers 4.2.2 Building Dependencies and Causalities 4.2.3 Reducing the Effort in Modeling Parameters with Noisy-ORgates 4.3 Model Development 4.3.1 Machine Learning 4.3.2 Expert Modeling 4.4 Bayesian Network Extensions 4.4.1 Multi-Entity Bayesian Network 4.4.2 Influence Diagrams 4.5 Validation methods 4.5.1 Quantitative Validation 4.5.2 Qualitative Validation 4.6 Visualization of Probabilistic Graphical Models 4.7 Tools and Methods for Modeling, Validation and Visualization 4.7.1 GeNIe and SMILE 4.7.2 Domain-specific Modeling Tools Part II Therapy Decision Support System using Bayesian Networks Chapter 5 Patient-specific Bayesian Network in a Clinical Environment 5.1 Introduction 5.2 Decisions in an Oncological Therapy Workflow 5.3 Concept of a BN-based TDSS 5.4 Concept for Semi-automatic Quality Management of BN based TDSS Chapter 6 TreLynCa: A Tumor Board Decision Model for Laryngeal Cancer 6.1 Introduction 6.2 Teamwork and Knowledge Acquisition 6.3 Software and Tools for Modeling and Validation 6.4 Model Development 6.4.1 Model Structure 6.4.2 Modeling Variables and States 6.4.3 Modeling Probabilistic Parameters 6.5 Scope of the TreLynCa Model 6.6 Discussion 6.6.1 TreLynCa’s Development 6.6.2 TreLynCa’s Applicability Chapter 7 Model Validation 7.1 Introduction 7.2 Prediction in a Clinical Treatment Decision Context 7.3 Validation and Modification Workflow 7.4 TNM Model and Study Set-up 7.5 Quantitative Validation 7.5.1 Accuracy 7.5.2 ROC 7.5.3 Confusion Matrix 7.5.4 Calibration Curve 7.6 Qualitative Validation 7.6.1 Patient Record-based Model Validation 7.6.2 Subnetwork Validation 7.7 Results and Modifications 7.8 Discussion and Conclusion Chapter 8 Tools for Guided BN Modeling 8.1 Introduction 8.2 Concept for Guided Modeling and Validation 8.2.1 Requirements for an Autonomous Expert Modeling and Validation 8.2.2 Concept of a Web-based Expert Modeling System 8.3 Graph Modeling Tool 8.3.1 Conceptual Framework of a Graph Modeling Tool 8.3.2 Guided Graph Modeling 8.3.3 Study Set-up 8.3.4 Study Results 8.3.5 Discussion 8.3.6 Conclusion 8.4 CPT Modeling Tool 8.4.1 Questionnaire to Find Dominant Parent States 8.4.2 Web-based Framework for CPT Assessment 8.4.3 Study Designs 8.4.4 Results 8.4.5 Discussion 8.5 Conclusion Chapter 9 GUI for PSBN-based decision verification 9.1 Introduction 9.2 Motivation for TNM Verification 9.3 Requirements for a TNM Verification GUI 9.4 Verification Framework 9.5 Graph View Design 9.6 User Workflow 9.7 Study Set-up of TNM Verification 9.8 Results 9.9 Discussion 9.9.1 Study results and limitations 9.9.2 Layout 9.9.3 Functionality 9.9.4 Applicability 9.10 Conclusion Part III Conclusion Chapter 10 Summary Chapter 11 Appraisal and Future Work 11.1 Applicability of Methodologies and Tools 11.2 Compliance of Requirements 11.3 Impact of Decision Graphs 11.4 Storage and Cross-Sharing of BN and PSBN 11.5 Reasoned PSBN Justification 11.6 Appropriate Patient Data Quality 11.7 Extension to Multi-Entity Decision Graphs Bibliography List of Relevant Publications Appendix A TNM staging by the NCCN guidelines