Knowledge based solution strategies in medical reasoning

Abstract

The techniques of propositionol analysis are used to examine the protocols of seven cardiologists in a task involving the diagnosis of a case of acute bacterial endocarditis and an explanation of its underlying pathophysiology. It is shown that the explanations of physicians making an accurate diagnosis can be accounted for in terms of a model consisting of pure forward reasoning through a network of causal rules, actuated by relevant propositions embedded in the stimulus text. These rules appear to derive from the physician's underlying knowledge base rather than any information in the text itself. In contrast, subjects with inacurate diagnoses tend to make use of a mixture of forward and backward reasoning, beginning with a high level hypothesis and proceeding in a top-down fashion to the propositions embedded in stimulus text, or to the generation of irrelevant rules.