RENOIR is an expert system developed to assist the diagnosis of 37 dis
eases of connective tissue and inflammatory arthropathies. Precise dia
gnosis of rheumatic diseases implies great uncertainty and there is no
gold standard with which to compare the expert system output. To over
come this problem a set of clinical cases was submitted to RENOIR and
its diagnoses were compared with those of clinicians. Medical records
of 81 patients with rheumatic diseases were interpreted by RENOIR and
by 12 clinicians at three different expertise levels in rheumatology.
Distances between the likelihoods of the 37 considered diseases provid
ed by clinicians and RENOIR were computed as a disagreement measure. M
ahalanobis distance was used to correct the collinearity between the p
ossibilities of each pair of diseases. Using the resulting matrices of
distances between experts, cluster analyses were carried out to class
ify RENOIR among human experts. Greater differences between RENOIR and
clinicians than among clinicians themselves were not found. (C) 1994
Academic Press, Inc.