The proposed research has the broad long-term objective of providing useful computer-based consultative assistance to clinicians faced with complex choices among diagnostic and therapeutic options. This advice will be provided by an intelligent decision system, a computer program employing artificial intelligence techniques to automate the generation of decision models and analysis of those models to produce a recommendation. The advantages of incorporating decision analytic principles are explicit consideration of uncertainty and patients' preferences, and an axiomatic approach which links the decision elements to a recommendation. The system will permit users to control the elements of a decision model and will thus constitute a decision analysis workbench rather than a traditional expert system. The project will focus on the evaluation of pulmonary infiltrates in patients with the Acquired Immunodeficiency Syndrome (AIDS) or suspected AIDS. This is an important and increasingly common problem which involves high stakes for individual patients and a bewildering array of diagnostic and therapeutic options with complex trade-offs for clinicians. In this medical area, knowledge of disease prognoses and efficacy of therapy is rapidly accumulating. Thus, diagnostic and therapeutic strategies must continuously evolve in response to new data. The system is to be implemented in the Common LISP programming language on an Intel 80386-based microcomputer. The system will employ separate knowledge bases for decision analytic knowledge and medical domain knowledge. The modularity inherent in this organization will facilitate expansion and refinement of the knowledge base in response to new research findings and the availability of new techniques. The system will use a frame-based representation of diseases, diagnostic tests and treatments. These frames and their relations will determine the alternatives and outcomes modeled by the system. A network representation of probabilistic dependencies will ensure that the consistent updating of probabilities is performed in each decision tree context. Generation of a decision model will be guided by context-dependent rules which will determine at any given point in the tree which events to consider and how deeply to expand the decision model. The system will also contain facilities to tailor preference functions and probabilities to individual patients. Abstracted cases from the medical records of patients seen in our institution who have pulmonary infiltrates and AIDS or suspected AIDS will be used for system evaluation.

National Institute of Health (NIH)
National Library of Medicine (NLM)
First Independent Research Support & Transition (FIRST) Awards (R29)
Project #
Application #
Study Section
Biomedical Library and Informatics Review Committee (BLR)
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Medicine & Dentistry of NJ
Schools of Medicine
United States
Zip Code
Sonnenberg, F A; Hagerty, C G; Kulikowski, C A (1994) An architecture for knowledge-based construction of decision models. Med Decis Making 14:27-39
Sonnenberg, F A; Roberts, M S; Tsevat, J et al. (1994) Toward a peer review process for medical decision analysis models. Med Care 32:JS52-64
McCarthy, B D; Wong, J B; Munoz, A et al. (1993) Who should be screened for HIV infection? A cost-effectiveness analysis. Arch Intern Med 153:1107-16
Jubran, A; Gross, N; Ramsdell, J et al. (1993) Comparative cost-effectiveness analysis of theophylline and ipratropium bromide in chronic obstructive pulmonary disease. A three-center study. Chest 103:678-84
Sonnenberg, F A; Beck, J R (1993) Markov models in medical decision making: a practical guide. Med Decis Making 13:322-38