輔仁大學
醫學資訊與創新應用學士學位學程
醫療標準及術語
臨床決策支援系統
Clinical Decision Support System
臺北市立聯合醫院仁愛院區家庭醫學科 郭冠良
Kuan-Liang Kuo, M.D., Ph.D.
2022-05-31
1
Topics
• Decision Making in Healthcare
• Clinical Decision Support System
• Knowledge Representation
• CDSS Implementation
• Evaluation of CDSS
Clinical Decision Support
System
Reference
• Greenes, Robert A., ed. Clinical decision support: the road ahead. Academic Press, 2011.
Introduction
Decision Support System
• DSS
– Combine individuals’ and computers’
capabilities
– Improve the quality of decisions
• CDSS
– DSS
– Supports physicians
– Minimize practice variation – Improve patient care
CDSS
• Support the workflow
• Improve the effectiveness of decision outcomes
• Support evidence-based practice
• Provide patient-specific assessments or recommendations
CDSS
• Examples
– Computerized Physician Order Entry (CPOE) Systems
• Provide patient-specific recommendations
• Care reminders for specific preventive care services
• Laboratory alert systems
– Page – SMS
Components of CDSS
Components of CDSS
• Knowledge base
• Clinical data
• Inference/reasoning engine
• User interface
Components of CDSS
• Knowledge base
– Guidelines – Rules
– Probabilistic models
• Clinical data
• Inference/reasoning engine
• User interface
Components of CDSS
• Knowledge base
• Clinical data
– HIS – EMR – EHR
• Inference/reasoning engine
• User interface
Components of CDSS
• Knowledge base
• Clinical data
• Inference/reasoning engine
– AI methods
• User interface
Life Cycles of CDSS
Life Cycle of Knowledge
Common Life Cycle of
Knowledge Management
CDSS Implementation and
Evaluation
Brief History of CDS
Methodologies of CDS
Information Retrieval
• Taxonomy-based search
– Ontology-based
• Text-based search
– Free text
– Web-based search engines
Information Retrieval
Evaluation of Logical Conditions
• Decision table
Evaluation of Logical Conditions
• Decision table
Venn Diagram
Logical Expression
• ECA rules (Event-Condition-Action)
– On
• Event
– If
• Condition
– Then
• Action
Embedded Conditions/Constraints
• Boolean logical expression
– Guideline
– Branch point
• Interactive dialogues for users
– Data entry form
• Validate
Probabilistic
• Applying Bayes theorem to medical diagnosis
Decision Tree
Heuristic Modeling
• Rule-based system
– If c then a
Driving Forces for CDS
• 科技進步
• 知識爆發
• 診斷與治療的新技術
• 發明與知識的同化
• Internet
• 病人與消費者的崛起
• 醫療錯誤
• 品質的變異性
• 電子病歷
• 人口老化與疾病複雜 化
• 醫療人員工作負擔增 加
• 協同照護的困難
• 防衛醫療
• 醫療成本
• 以質計價
• CDS的好處
• 由上而下的推動
Features of CDS
Principal purposes for CDS
Principal purposes for CDS
Medical Decision
Conceptual Model of CDS
Dimensions of computer-user
interaction in CDS
Case Studies
Partners Health System
Renal Dosing
Dosing
Pregnancy
Drug Interaction
Drug Allergy
Results Manager
Effective CDS
• 6. is difficult
HELP HIS
• Developed in 1980s at LDS Hospital in Salt Lake City, Utah
• HELP (Health Evaluation through Logical Processing)
• HIS (Hospital Information System)
• MLMs (Medical Logic Modules)
HELP HIS
HELP HIS
HELP: Tools for Focusing Attention
• Infectious disease monitor
• Therapeutic antibiotic monitor
• Adverse drug event monitor
• Lab alert
• Antibiotic duration monitor
• Preoperative antibiotic monitor
• High-risk alerts for hospital-acquired infections
• Drug-dose monitor
• Enhanced notification of ventilator-related events
HELP: Tools for Focusing
Attention
HELP: Tools for Focusing
Attention
HELP: Tools for Patient-specific Consultation
• Blood ordering
• Ventilator protocols
• Anti-infective agent assistance
• Patient isolation program
HELP: Tools for Patient-specific Consultation
• Ventilator protocols
HELP: Tools for Patient-specific
Consultation
Academic Prototypes of CDS
Knowledge Management
Knowledge
Knowledge Acquisition
• 由已存在的來源辨識且引出知識
– 已存在的來源
• Domain experts
• Documents
• Inferred from large datasets
– 辨識且引出知識
• Knowledge representation
– Encoding
Knowledge Acquisition
• Why?
– Knowledge preservation – Knowledge sharing
– Knowledge to form the basis for decision aids – Knowledge that reveals underlying skills
Knowledge Engineering
• Classical methods by knowledge engineer
Knowledge Engineering
• Automatic method by KA system
CTA (Cognitive Task Analysis)
• Capture cognition
– Capture the way the mind works
• Three key aspects of CTA
– Knowledge elicitation 引出 – Data analysis
– Knowledge representation
Three key aspects of CTA:
Knowledge Elicitation
• Group techniques
– Brainstorming
– Nominal group studies – Presentation discovery – Delphi studies
– Consensus decision-making
– Computer-aided group sessions
Three key aspects of CTA:
Knowledge Elicitation
• Biases in Logical and Probabilistic Reasoning
– Uncertainty in clinical medicine – Terms for managing uncertainty
• Suggest, support, consider, likely, …
Three key aspects of CTA:
Data analysis
• Protocol and discourse analysis
• Concept analysis
– Concept map
• Verification and validation of KA
– Verification (確認,查證)
• Define the design of the system
• Logical consistency, completeness, avoidance of redundancy
– Validation (驗證)
• Implementation and refinement of the system
• In real world
Three key aspects of CTA:
Data analysis
• Heuristic methods
– The most common approach to evaluate knowledge
– Reviewed by experts – Limitation
• Large data sets
• Not automatic
• Standard by human
Generation of Knowledge for
CDSS
Classification Trees
Decision Tree
Learning from Data
• Statistical and machine learning
• Data mining
– Pattern recognition techniques
• Find relationship of data
Learning from Data
• Learning model
– Unsupervised
• When knowledge is sparse
• Unveil hidden patterns
• Example
– High-throughput micro-array data – Clusters of variables
• Not been applied in clinical
– Supervised
• Predefined classes
• Example
– Range of normal values
Artificial Neural Network
• Learning stage?
• Productivity stage?
Current CDSS
• Mostly rely on the rule-based paradigm
– Data are not available or not structured for machine learning
– Techniques are not well implemented in the biomedical community
– Nonprobabilistic rules may be more clear and efficiently
