Use of probability estimates in medical communications and decisions

Quantitatively sophisticated physicians in general internal medicine gave a series of probability estimates about each of several cases, which consisted of a written background and radiological test report. First, they estimated the three quantities of Bayes' formula: the pretest probability based on the background, the likelihood ratio based on the test report, and the posttest probability based on both. They made two further estimates of the posttest, probability, one after being informed of the consensus of expert radiologists on the likelihood ratio of the test and another after being told a consensus pretest probability formed by experts in the medical speciality relevant to the case backgrounds. They also stated how high (low) the posttest probability would have to be in order to confirm (rule out) the suspected disease and estimated the chances that the test report would serve to move the posttest probability across one of those thresholds. On average, the clinician subjects agreed quite well with the experts' estimates and with Bayes' formula. Information about experts' consensuses influenced the subjects' estimates in the appropriate direction and decreased the variation among subjects. These results are taken to indicate the feasibility and potential effectiveness of certain quantitative aids to clinical decision making. On the other hand, the variation among the subjects' estimates, and among experts' estimates, was substantial - enough to warrant attempts to understand it and to reduce it. Diagnostic criteria that the subjects reported using for case backgrounds and for tests reports were analyzed as a way of accounting for some of the variation.