Test selection in jaundice: a comparison between physician behavior and a diagnostic model

The results of an observational study aimed at a formal assessment of clinicians' test-selection behavior are presented. We first make a proposal for diagnostic test usage in the latter phases of jaundice diagnosis. Next we compute a probabilistic estimate of the patient diagnosis, based on the COMIK algorithm. From the proposal and the probabilistic estimate we can predict the "test-selection behavior" of clinicians. The assessment follows from a tabulation of the predicted behavior against the tests selected by clinicians. It is shown that for most tests, the predictions are consistent with the observed test-selection behavior at a statistically significant level. Discussions of discrepancies between prediction and observation, and reasons for deviations from general guidelines, provide new dimensions for medical education. The methodology applied is a useful tool to improve medical care for the jaundiced patient.