Initial evaluation of a subjective Bayesian diagnostic system.

A computer-aided diagnostic system using subjectively estimated probabilities for symptom—disease relationships is described and applied to a sample of 200 cases divided among hypothyroid, euthyroid, and hyperthyroid diagnoses. The subjective system is evaluated by comparing it with one using actuarial probabilities developed in standard fashion and one using separate actuarial probabilities for suspected hypothyroid and suspected hyperthyroid categories. Analysis of the data indicates that the subjective model's developmental cost and time requirement is much lower, while it performs as well as either actuarial model.