Conditional Probability Program for Diagnosis of Thyroid Function

Abstract
The frequencies of occurrence of certain clinical findings (signs, symptoms, and laboratory data) were determined in 879 patients who had been classified as hypothyroid, euthyroid, or hyperthyroid on the basis of their responses to therapy during at least 6 mo of observation. A Bayesian conditional probability model was employed to estimate the probability of each of these three conditions for each possible combination of clinical findings. When the study was limited to 268 cases in which the data were considered reasonably complete, the computer program gave the correct classification (ie, successfully predicted the therapeutic result) in 258 (96%). This degree of accuracy approached the level of agreement to be expected between an expert consultant and a well-qualified diagnosing physician.