Search

To evaluate the performance of CADIAG-II/RHEUMA as consultant in the primary evaluation of patients visiting a rheumatological outpatient clinic, a CADIAG-II/RHEUMA consultation was done for 54 patients and the list of generated diagnostic hypotheses was compared to each clinical discharge diagnosis. For 26 of a total of 126 rheumatological discharge diagnoses, no matching CADIAG-II/RHEUMA diagnosis was available. 94% of all other discharge diagnoses were found in the list of CADIAG-II/RHEUMA hypotheses, 82% among the first third of the list of hypotheses and 48% among the first five hypotheses. We identified the following factors limiting the ability of CADIAG-II/RHEUMA to generate a comprehensive and correctly ranked list of diagnostic hypotheses: (1) a large percentage of patients with early stages of not clearly identified rheumatological conditions; (2) the limited number of CADIAG-II/RHEUMA diagnoses compared to the large number of known rheumatological conditions; (3) the fact that rheumatological diseases are rarely characterized by a single pathognomonic feature but are usually diagnosed by combinations of rather unspecific findings.

In 1987, the American Rheumatism Association issued a set of criteria for the classification of rheumatoid arthritis (RA) to provide a uniform definition of RA patients. Fuzzy set theory and fuzzy logic were used to transform this set of criteria into a diagnostic tool that offers diagnoses at different levels of confidence: a definite level, which was consistent with the original criteria definition, as well as several possible and superdefinite levels. Two fuzzy models and a reference model which provided results at a definite level only were applied to 292 clinical cases from a hospital for rheumatic diseases. At the definite level, all models yielded a sensitivity rate of 72.6% and a specificity rate of 87.0%. Sensitivity and specificity rates at the possible levels ranged from 73.3% to 85.6% and from 83.6% to 87.0%. At the superdefinite levels, sensitivity rates ranged from 39.0% to 63.7% and specificity rates from 90.4% to 95.2%. Fuzzy techniques were helpful to add flexibility to preexisting diagnostic criteria in order to obtain diagnoses at the desired level of confidence.