| Abstract | The mathematical approach to diagnosis was examined in a group of 1,184 patients with congenital heart disease using a rather limited number of symptoms (46 items grouped into 25 sets of information). In each patient there was a precise ultimate diagnosis based on clearcut catheterization data and/or operative or autopsy findings. Patients were examined personally by one of the authors, who noted his diagnostic impression based upon complaints, physical findings, ECG and fluoroscopy. A FORTRAN program was prepared for an IBM 7040 computer which works out the a priori probabilities of 94 different malformations and the conditional probabilities of the 46 symptoms in the patient studied, according to Bayes' formula. The clinical diagnosis and the computer diagnosis were checked against the correct diagnosis derived from the follow-up studies, and the result was evaluated as the percentage of the cases in which the physician and computer express the correct diagnosis. Coincidence with the correct diagnosis occurred in 73.22% of the cases with the clinical approach and in 81.84% with the computer. When the cases were separated according to the incidence of diseases, the figures obtained were: clinical coincidence, 81.39%; computer coincidence 84.80% for the diseases represented by 12 or more cases; and 36.87% and 68.67% respectively for the diseases represented by less than 12 cases. |