Iliad and Medical HouseCall: evaluating the impact of common sense knowledge on the diagnostic accuracy of a medical expert system.

Diagnostic expert systems are gaining acceptance among physicians. Recently, a comparative study of the performance of four major commercial diagnostic programs demonstrated that the information they produce contains a certain amount of irrelevancy that the trained physician ignores. Medical HouseCall is a consumer health information expert system based on a medical expert system for physicians, Iliad. To enhance the usefulness of Medical HouseCall to health care consumers, we are interested in significantly reducing the amount of irrelevancy contained in the diagnostic differential list. Testing with over 470 'textbook' cases revealed that a large part of the irrelevancy can be eliminated by adding universal and medical 'common sense' knowledge. Using four performance measures, we compared, on a subset of cases, the differential lists from two versions of the program: the refined knowledge base (1995) and an older version (1994) 'pre-common sense'. The results suggest that the performance of a diagnostic expert system can be significantly improved with the addition of common sense knowledge.