Clinical Decision Support Systems 朱爱玲 Medical Informatics Group 24 Jan,2003
Outline Introduction Medical decision making What are clinical decision support systems (CDSSs)? Historical perspective Abdominal pain system [de Dombal et al.,1972] MYCIN system [Shortliffe, 1976] HELP system [Kuperman et al, 1991; Warner, 1979] Characterizing CDSSs and challenges in building CDSSs
make a medical decision-human Patient data Human Brain Knowledge Inference mechanisms Case-specific advice
make a medical decision-computer Patient data Human Brain Inference Engine Medical knowledge base Case-specific advice
Definition: What are clinical decision support systems (CDSSs)? Active knowledge systems which use two or more items of patient data to generate case-specific advice [Musen]. Synonym: Knowledge-based system: a system with a knowledge base and an inferencing mechanism operating on a patient database. Expert system
Why CDSSs? The increasing pressure to practice cost-effective medicine; The growing distress that practitioners need to practice medicine well with the amount of information (e.g. incomplete, uncertain) The increasing recognition that computer systems can help (e.g. cost-effectiveness analysis)..
What is a good CDSS? Requirements for an excellent decision-making systems: Accurate patient data Reliable knowledge base Appropriate inferencing mechanism (problemsolving skills) User friendly interface
Historical perspective In late 1950s, the first generation of CDSSs appeared [Ledley & Lusted, 1959], and experimental prototypes appeared in [Warner et al, 1964] Three advisory systems from the 1970s: Abdominal pain system for diagnosis of abdominal pain [de Dombal et al.,1972] MYCIN system for selection of antibiotic therapy [Shortliffe, 1976] HELP system for delivery of inpatient medical alerts [Kuperman et al, 1991; Warner, 1979]
Leeds Abdominal Pain System Helps the clinician in determining the cause of acute abdominal pain. After the physician enters patient data (i.e., the symptoms of the patient), the system gives the probabilities of the possible diagnoses They emphasized the importance of deriving the conditional probabilities used in Bayesian reasoning from high-quality data An important exemplar of the clinical value of Bayesian diagnostic systems.
MYCIN A consultation system that concentrates on appropriate management of patients who have infections. straightforward algorithms or statistical approaches were inadequate for representing this clinical problem The nature of expertise was poorly understood The researchers were drawn to the field of artificial intelligence (AI) focused on manipulation of abstract symbols rather than numerical calculations. Knowledge of infectious diseases in MYCIN was represented as production rules. simply a conditional statement that relates observations to associated inferences that can be drawn.
Rule507: A production rule from the MYCIN IF: 1. the infection which requires therapy is meningitis, 2. Organisms were not seen on the stain of the culture, 3. The type of infection is bacterial, 4. The patient does not have a head injury defect, and 5. The age of the patient is between 15 years and 55 years THEN: The organisms that might be causing the infection are diplococcus-pneumoniae and neisseria-meningitidis
MYCIN-continued MYCIN is best viewed as an early exploration of methods for capturing and applying ill-structured expert knowledge to solve important medical problems It paved the way for a great deal of research and development in the 1980s The development of knowledge-based systems, and the commercialization of the rule-based approach during the early 1980s, evolved from MYCIN and related systems developed during the 1970s.
HELP (Health Evaluation through Logical Processing) Developed by Warner and colleagues at the Latter Day Saints (LDS) Hospital in Salt Lake City, Utah; Notable in that it was among the first to incorporate decision-support logic. Modules of specialized decision logic permit the system to react to patient data and to generate patient-specific warnings, alerts, diagnostic suggestions, and limited management advice
HELP-2 PAL-a specialized language for writing medical knowledge in HELP sectors (decision logics). In 1992, Arden syntax was created based largely on the encoding scheme used in HELP (PAL). Arden syntax: A standard language for writing situation-action rules that can trigger alerts based on abnormal clinical events detected by a clinical information system.
Medical Logic Module (MLM) in Arden syntax DATA: POTAS-STORAGE := event {serum potassium} POTAS := LAST {serum potassium} THIAZIDE-US E := {current prescription for thiazides} EVOKE: [defines a situation that causes the rule to be triggered] POTAS-STORAGE LOGIC: [encodes the decision logic of the rule] IF POTAS < 3 THEN CONCLUDE TRUE ELSE CONCLUDE FALSE ACTION: [defines the procedure to follow ] SEND "Patient is hypokalemic. This condition could be caused by thiazides."
How to characterize CDSSs 1. The system s intended function What is true about a patient (correct diagnosis) e.g. Leeds abdominal pain program What to do for the patient (what test, what therapy, cost and risk) e.g. MYCIN 2. The mode for giving advice Passive (the physicians use the system when they need help) Active( providing advice as a byproduct of monitoring or of data-management activities) e.g HELP
How to characterize CDSSscontinued 3. The style of communication Consulting model ( program serves as an advisor) MYCIN Critiquing model (computer acts as a sounding board for the user s own ideas) HELP 4. The underlying decision-making process Problem-specific flowchart (too simplistic for routine use) Bayesian modeling, decision analysis, artificial neural networks, and AI 5. Human-computer interaction User friendly interface, easy to learn, use
Tasks for constructing CDSSs Acquisition and validation of patient data Development, validation and maintenance of medical knowledge base Integration of decision-support tools Modeling of medical knowledge Elicitation of medical knowledge Representation of and Reasoning about medical knowledge Validation of system performance
Challenges in Constructing CDSSs Acquisition and Validation of patient data we lack standardized ways of expressing most clinical situations in a form that computers can interpret (e.g. the details of physicians progress notes) Modeling of medical knowledge A model of both the required problem-solving behavior and the clinical knowledge Developing computer-based tools that can assist in the modeling of clinical knowledge remains an active area of investigation [Eriksson et al, 1995; Musen et al, 1995]
Challenges in Constructing CDSSs-continued Elicitation of medical knowledge The rapid development of medical knowledge makes knowledge-base maintenance a particularly important problem. Knowledge-acquistion tool: E.g. ONCOCIN[Shortliffe,1986], OPAL [Musen et al, 1987], Protégé [Musen, 1995] Representation of and Reasoning about medical knowledge how people store and use their knowledge is unknown. Well-established techniques such as frame or rules exist for storing factual or inferential knowledge, but complex challenges remain. The need to refine the computational techniques for encoding the wide range of knowledge used in problemsolving
Challenges in Constructing CDSSscontinued Validation of system performance The currency of large clinical knowledge bases Performance of the decision-support tools Diagnostic tools: compare the program s advice with that accepted standard of correctness ; Therapy-advice systems: the gold standard is difficult to define Integration of decision-support tools Challenges are inherently tied to issues of networking and system interfaces.
Legal and regulatory questions Formal legal precedents for dealing with CDSSs are lacking Questions regarding the validation of CDSSs before their release. This is often no such thing as the correct answer to a clinical question it is unrealistic to require that CDSSs make correct assessments under all circumstances. Issues arise when considering to enable global access to electronic patient records through internet, e.g. privacy, security
Reference Handbook of Medical Informatics J.H. van Bemmel & M.A. Musen Medical Informatics:Computer Applications in Health Care Shortliffe & Perrault
Recommended literature & websites Links to sites featuring clinical decision support systems Useful stuff Recommended literature : Clinical Decision Support Systems- Theory and Practice Eta S. Berner Handbook of Medical Informatics J.H. van Bemmel & M.A. Musen Medical Informatics:Computer Applications in Health Care Shortliffe & Perrault Medical Information on the Internet, Robert Kiley An Introduction to Artificial Intelligence Janet Finley & Alan Dix Artificial Intelligence-A Modern Approach, Stuart Russel & Peter Novig