Using Causal Knowledge to Create Simulated Patient Cases: CPCS Project as an Extension of INTERNIST-1

The authors propose that this ability to generate artificial data can be much like a “Turing Test” for judging the adequacy of any given knowledge base.

Abstract

Expert systems for medical diagnosis require associated knowledge bases. A dichotomy, based on depth of causal representation, has been made by various researchers between the types of medical expert system knowledge bases. "Shallow" knowledge bases, such as that of the INTERNIST-1 program,1 do not contain detailed representations of causality (i.e., they do not represent mechanisms by which a disease process causes clinical abnormalities in a given patient), whereas "deep," or "second-generation" expert systems, such as the ABEL program developed by Patil et al.,2 contain sufficient causal information to make qualitative and quantitative predictions about pathophysiologic abnormalities in a given patient case.