Comparing aggregate estimates of derived thresholds for clinical decisions

Thresholds for medical decision making are the probabilities of disease at which clinicians choose to initiate testing or therapy. A descriptive analysis of clinicians' decision making can derive their test and test-treatment thresholds and has the potential to explain variations in test utilization. A previously described method summarizes thresholds for a group of clinicians by determining the range of probability which includes the maximum number of clinicians' individual thresholds. However, there is no statistical procedure to compare the summary measure of thresholds that is derived from the distribution of clinicians' thresholds. We describe two alternative methods of developing a summary measure of the thresholds for a group of clinicians. These alternative methods enable the analyst to apply standard statistical tests when analyzing the decision-making behavior of groups of clinicians. For the "Unweighted Mean of the Midpoints" method, confidence limits of means and standard t-tests can be used to compare different groups. For the "Weighted Mean of the Midpoints" method, a weighted standard error of the mean can be calculated to determine confidence intervals, and a weighted t-test or weighted regression can be used to compare weighted means of the midpoints of threshold ranges.