Pattern-matching models for the differential diagnosis of bovine spongiform encephalopathy.

This study assessed the performance of a system for making decisions about the diagnosis of bovine spongiform encephalopathy (BSE). The system consisted of four pattern-matching models. The sensitivity, specificity, likelihood ratios and accuracy of each model were determined by using clinical descriptions of 100 suspect BSE cases which had been submitted for brain histopathology by veterinary officers of the Ministry of Agriculture, Fisheries and Food, 50 of which were true positive cases (confirmed by histopathology) and 50 false positive cases (not confirmed by histopathology). The clinical description of each case consisted of 14 clinical signs, each of which was defined as either present or absent. The system compared the case descriptions with the profiles of possible differential diagnoses, each profile consisting of the frequency of occurrence of the same 14 clinical signs. The pattern-matching models used the sums of the sign frequencies to rank the differential diagnoses. Models 1 and 2 derived information only from the presence of signs; models 3 and 4 derived information from the presence and absence of signs. Models 2 and 4 excluded diagnoses which did not have in their profile a sign which was observed, and diagnoses which had a sign in their profile which should always be present according to the profile description but which was not observed. The best performances by the models were: sensitivity 96 per cent (model 1 and model 2), specificity 72 per cent (model 4), accuracy 72 per cent (model 4), likelihood ratio of a positive test 2.00 (model 4), likelihood ratio of a negative test 0.21 (model 4).