A Comparison of Clustered Knowledge Structures in Iliad and in Quick Medical Reference

Abstract Iliad is a medical expert system whose medical knowledge is organized by disease into “frames” that each contain multiple findings that may be expected in that disease. These findings are processed sequentially, using Bayes' Theorem, when knowledge about the patient becomes available. Iliad incorporates newly designed knowledge frames called “clusters”. Clusters are Boolean decision frames containing conditionally dependent findings that often describe important pathophysiologic concepts. Pathophysiologic concepts are so pervasive in medical teaching that supposedly non-clustered expert systems might contain implicit pathophysiologic clusters. Quick Medical Reference (QMR) is a non-clustered expert system. QMR assigns each patient finding a score called an “evoking strength”. We performed a cluster analysis upon these evoking strengths for findings in pulmonary disease and discovered definite clusters of findings. The clusters discovered corresponded closely with Iliad clusters. Clustered knowledge structures are natural human mental models. We believe QMR's knowledge engineers unconsciously imposed natural, clustered knowledge models onto QMR. Clustered knowledge models could improve both the performance and teaching qualities of medical expert systems.