CASEY, developed as a doctoral thesis by Phyllis Koton[10], adds a case-based diagnostic reasoning operator to the Heart Failure Program. This is an alternative diagnosis generator, allowing a comparison of different methodologies. It also allows the user to find similar cases for comparison. The CASEY operator uses the values from the input, the causal diagnostic model, and the case base. The first step in case-based diagnosis is using the input findings to find similar cases in the case base. CASEY has no predetermined list of important findings, so all of the findings are used at this stage to search for matches. From the causal diagnostic model, CASEY has a mechanism for assessing the similarity between findings. For example, rales on the physical exam and vascular redistribution on a chest X-ray are both evidence for the same physiologic state, cardiac pulmonary congestion. Thus, CASEY can find and justify the matches, not because the findings exactly match, but because the findings can provide evidence for the same physiologic states. Once a partial match has been found, the next step is to try to adapt the stored case to the new findings. There may be parts of the old case for which there is no support in the findings or findings that require additions to the old causal model. CASEY has a set of operators that take care of these situations. The simplest situation is new findings which add support for states already in the causal hypothesis. If the finding can not be explained by an existing state, CASEY makes use of the mechanism in the diagnostic reasoner to look for a good pathway from some part of the hypothesis to the new finding. If this will require new diagnostic states or primary causes, the search is abandoned and the match is rejected. If when all of the new findings have been added, there are parts of the hypothesis that are unsupported, CASEY has a pruning procedure that eliminates states until the whole structure is supported. The result of this process is a hypothesis that completely explains the findings and can be treated in the same way as a hypothesis from the causal diagnostic reasoner.
A case-based reasoner also needs to learn from the cases that are handled. Thus, CASEY needs a source of correct answers. After diagnosis has taken place, using CASEY, the causal reasoner, or a user guided diagnostic procedure, the user can assert that a diagnosis is correct and allow CASEY to add it to the case base. The structure and entry process for the case base is described in section 3.3.
CASEY is quite conservative in its approach to revision, but even with 50 cases in the initial database (including about ten different primary diseases) it was able to find sufficiently close matches to satisfactorily diagnose cases 80%of the time. Usually, the diagnoses generated with the probabilistic reasoner are better, but there are a few instances where CASEY has found a better hypothesis. The interaction between the case-based reasoner and the causal model reasoner gives CASEY the leverage needed to handle cases when the superficial similarity of the findings may be quite small and add the potential efficiency of the associational reasoning to the system. Indeed, CASEY is the first effective combination of case-based reasoning with an extensive causal model.