A complete, hypermedia medical decision analysis support system

A Complete, Hypermedia Medical Decision Analysis Support System Deng Y. Chiu

Chung C. Chang Martha W. Evens Illinois Institute of Technology Chicago, Illinois 60616 (312) 567-5153

Johug C. Chern

Daniel B. Hier, M. D. Department of Neurology College of Medicine, University of Illinois at Chicago Chicago, Illinois 60612 David A. Trace Frank Naeymi-Rad UHS/Chicago Medical School North Chicago, Illinois 60064 Abstract A treatment risk analysis system was developed and presented by the MEDAS project group 2,3,6,7]. Now, a risk analysis system for medical tests has also been developed to form a complete medical decision support system. Moreover, the current system (a combined treatment and test system) has been redesigned and developed using an object-oriented hypermedia programming language, called Spinnaker Plus, running in the MS-Windows environment. The system has already been used to analyze several neurological problems. 1. Introduction Medical decisions are important to patients' health and the effective use of medical resources. Excellent medical decisions may prolong patients' lives, improve patients' health status and make better use of medical resources. In some clinical situations, the decisions are straightforward, and most clinicians would reach the same conclusion and choose the same treatments [I]. Other situations may be more ambiguous, and there is much less agreement about what to do. Even so, physicians still need to make their own decisions to treat their patients. Without medical decision analysis, physicians make medical decisions based on intuitive judgements derived from their knowledge and experience and often wonder whether science would confirm that intuition. Sometimes, medical decisions must be determined with incomplete data. Some subjective estimates must be made. In these cases, decision analysis may exhibit its greatest advantage. Decision analysis is the process of structuring the decision problem in the form of a simple tree and then inserting necessary values in a coherent way to find out the best strategy. Some other advanced analysis may be also performed. Our system provides four major types of decision analysis. They are decision trees, sensitivity analysis regarding disease (such as prevalence of disease), sensitivity analysis regarding tests (such as complications caused by tests, precision of test results), and sensitivity analysis regarding all kinds of valued outcomes (such as the valued outcome of giving a treatment to patients with/without disease). The system provides a problem creator to help users set up a new disease problem (save it and make necessary connections), a knowledge base editor used to edit pre-stored expert knowledge (such as rules used to adjust statistical data and subjective judgement according to patient's features), and other utilities.

2. The Goals of the Support System The system is designed to reach several goals. First, to provide physicians with useful information to handle ambiguous medical decision situations. Second, to provide an attractive, computer-based learning tool. Third, to increase the reliability of decision making. Fourth, to reduce unnecessary or inappropriate medical tests and treatment to protect the patient's health and facilitate efficient utilization of medical resources. 3. The Steps in Medical Test Decision Analysis The first step is to make sure that the necessary prior data for medical test analysis is available. The second step is to adjust prior data according to the patient's features. The third step is to compute posterior data that are necessary for medical test decision analysis. Possible testing errors should also be considered here. The fourth step is to construct the test decision subtree, fill the necessary data into the tree, and apply averaging-out and folding-back methods on the testing decision subtree to compute the best expected value and action for taking tests. The fifth step is to perform advanced analysis involving prior probabilities and the estimates of outcomes for the medical test. 4. Combination of Treatment Decision Analysis and Medical Test Decision Analysis The result of medical test decision analysis and the result of treatment decision analysis are combined to achieve the final medical decision analysis result. The first two steps in medical tests decision analysis and treatment decision analysis are similar. The first step is to ensure that their own necessary prior data are available. The second step is to adjust their own prior data according to the rules saved in the knowledge base and the patient's features saved in the patient record database. Also, medical test decision analysis and treatment decision analysis must both find out the best expected values and the best actions for their own subtrees and to perform advanced analysis regarding their own data that is used for analysis. But, only medical test decision analysis needs to calculate the posterior data required needed to fill in the medical test decision subtree and to carry out advanced analysis. This step is complicated. Finally, the best expected value and action emerging from the entire medical decision analysis trees and advanced analysis regarding the entire medical decision can be achieved by comparing the results produced by treatment decision analysis and medical test decision analysis if medical tests are considered.

5. The Structure of the Support System The structure of the system is shown in Figure 1. The components of the system consist of the Control Center, Problem Creator, Problem Editor, Problem Database, Knowledge Base Editor, Knowledge Base, Data Adjustment Utility, Patient Record Database, User Input All Data Utility, and Decision Analysis Utility. The Decision Analysis Utility consists of three units, the Medical Decision Analysis unit, the Treatment Decision Analysis unit, and the Medical Tests Decision Analysis unit.

Problem Creator

Control Center

Problem Database

Knowledge Base Editor

Problem Editor

Knowledge Base

User Input All Data Utility

Data Adjustment Utility

Patient Record Database

Decision Analysis Utility

Medical Decision Analysis

Treatment Decision Analysis

Medical Tests Decision Analysis

Figure 1 The Structure of the Complete Medical Decision Analysis Support System 6. Utility

The Control Center is connected to six components, (1) the Problem Creator, used to add new problems; (2) the Problem Editor, used to edit existing problems; (3) the Knowledge Base Editor, used to edit rules that are used to adjust prior data of problems according to the patient features stored in the patient record database; (4) the User Input All Data Utility, used to collect all necessary data for decision analysis from users; (5) the Data Adjustment utility, used to adjust prior data; (6) the Decision Analysis Utility, used to perform ail kinds of decision analysis. The Problem Creator can be activated by the Control Center. All the data for the newly created problem is stored in the Problem Database. The Problem Editor can be activated by Control Center to edit existing data stored in the Problem Database. The Problem Database stores data created by the Problem Creator, provides the data for the Problem Editor to modify and restore, and provides data for the Knowledge Base Editor to create rules for problems. The Knowledge Base Editor can be activated by the Control Center. It uses data stored in the Problem Database and the Patient Record Database to edit rules used to adjust prior data. The resulting rules are stored in the Knowledge Base. The Knowledge Base stores rules created by the Knowledge Base Editor and provides rules for the Data Adjustment Utility that will use the rules to adjust prior data. The Patient Record Database stores all patient records input by the data entry module. It provides patient data for the Knowledge Base Editor that creates rules using the patient features, and for the Data Adjustment Utility. The Data Adjustment Utility can be activated by the Control Center. It uses the rules stored in the Knowledge Base and the patient features stored in the Patient Record Database to adjust the prior data. It finally sends the adjusted prior data to the Decision Analysis Utility for use in the decision analysis. The User Input All Data Utility can be activated by the Control Center. It provides a tool to collect all necessary data for decision analysis from users. In this case, no prestored data will be used. Users can enter their own estimated prior data to run the decision analysis utility. This User-input-All-Data Utility also provides users with a good environment to learn medical decision analysis. The data entered by users will be sent to the Decision Analysis Utility. The Decision Analysis Utility can be activated in three ways. First, it can be activated directly by the Control Center; in this case, the decision analysis utility will perform decision analysis not for any specific patient, but only for typical patients. Second, it can be activated by the Data Adjustment Utility; in this case, the decision analysis utility will perform decision analysis for a specific patient. Third, it can be activated by the User-Input-All-Data Utility; in this case, the decision analysis utility will perform decision analysis according to the data entered by users only. The Decision Analysis Utility includes three units, the Medical Decision Analysis Unit, the Treatment Decision Analysis Unit, and the Medical Test Decision Analysis unit. The decision analysis performed by the Medical Decision Analysis unit includes medical decision tree analysis, sensitivity analysis for the prevalence of disease, sensitivity analysis for test complications, sensitivity analysis for the precision of tests, and sensitivity analysis for ail kinds of valued outcomes. The precision of tests involves P[T+ D], the probability of having a positive test result, given that the patient has disease, and P[T- No D], the probability of having a negative test

result, given that the patient has no disease. The valued outcomes include among others the valued outcomes of giving treatment A to patient with/without disease. The decision analysis performed by the Treatment Decision Analysis unit includes the treatment decision tree analysis, sensitivity analysis for the prevalence of disease, and sensitivity for all kinds of valued outcomes. The decision analysis performed by the Medical Test Decision Analysis includes the medical test decision tree analysis, sensitivity analysis for the prevalence of disease, sensitivity analysis for test complications, sensitivity analysis for the precision of tests, and sensitivity analysis for ail kinds of valued outcomes. 7. Functions There are several other functions provided by the system. The colors of branches and nodes of the decision trees are changed to indicate how the final decision and final expected value are derived. For all sensitivity analyses, the analysis table first shows the analysis for eleven values, 0.0, 0.1.., 1.0, for the variable in question and the sensitive points are pointed out. If the user clicks on a value (such as 0.3) in question, it will calculate a new analysis table for eleven values in a smaller range (0.30, 0.31..0.40 for 0.3). A explanatory window is attached to almost every value in decision trees and analysis tables. It will pop out to explain how the value is calculated. Many other functions are also provided. 8. Conclusion The system has been used to analyze several neurological problems [4,51, carotid endarterectomy for symptomatic carotid stenosis, warfarin for non-rheumatic atrial fibrillation and surgery for unruptured AVM. In this research, this system analyzes not only the key values but the values that are easy to ignore but may influence the final decision. The system provides multi-aspect, deep, valuable and reliable decision analyses in a very convenient, quick, and attractive way. References [1]

Weinstein, M. C. and Fineberg, H. V. 1980. Clinical Decision Analysis. W. B. Saunders Company, Philadelphia, PA.

[2]

Chang, C. C. Chiu, D. Y., Wan, T. L., Evens, M., Trace, D., Naeymi-Rad, F., and Carmony, L. 1993. A Quantitative, Multi-Function Risk Analysis System for Treatment Decision Support. Proceedings of the Sixth Annual IEEE Computer-Based Medical Systems Symposium, Ann Arbor, Ml, pp. 200-205.

[3]

Chang, C. C. 1993. Design and Modeling of a Computer-based Medical Decision Analysis for the MEDAS Project. Ph.D. Thesis, Illinois Institute of Technology, IL.

[4]

Aminoff, M. 1987. Treatment of Unruptured Cerebral Arteriovenous Malformations. Neurology, Vol 37, pp. 815-819.

[5]

Barnett, H., North American Symptomatic Carotid Endarterectomy Trial Collaborators. 1991. Beneficial Effect of Carotid Endarterectomy in Symptomatic Patients with High-Grade Carotid Stenosis. The New England Journal of Medicine, Vol 325, pp. 445453.

[6]

Chang, C. C. Evens, M., Chiu, D. Y., Wan, T.L., Trace, D., Naeymi-Rad, F., and Carmony, L. 1993. Using a Rule-Based Approach and a Case-Based Reasoning Approach in an Automated Treatment Risk Analysis System. Proceedings of the Fifth Midwest Artificial Intelligence and Cognitive Science Conference, Chesterton, IN, pp. 88-92.

[7]

Trace, D., Evens, M., Naeymi-Rad, F., and Carmony, L. 1990. Medical Information Management: The MEDAS Approach. Symposium on Computer Applications in Medical Care, Washington, DC, pp. 635-639.