Computer study of thyroid diseases. III. Statistical models

A binary-variable multiple-regression model is proposed for the computer diagnosis of thyroid disease. The procedure employs a multiple-regression technique in which each variable is assigned a value of either unity or zero. Unlike other current approaches, the proposed model uses only information about the patients themselves; no prior knowledge of the incidence of the disease in the population under study is required. The model was applied to data on 906 patients with suspected thyroid disorders referred to endocrine clinics in Israel. On the basis of simple physical findings and clinical signs and symptoms, the correct diagnosis was obtained in 95% of the cases for males and in 85% for females.