A preference-based measure for test performance with an application to prenatal diagnostics.

Abstract

Clinical epidemiology generally uses the receiver operating characteristic curve to summarize the accuracy of a diagnostic test and to compare the relative performance of different tests. This paper extends this concept to include the utility gains and losses of true and false test outcomes over the range of a priori risk. A utility index is developed first in situations where test accuracy is exogenously given, second where the test cutpoint can be chosen by the clinician according to the patient's a priori risk and preferences. By integrating over the a priori risk range, we derive an overall measure for a test's performance weighted by utility gains and losses. An example in prenatal diagnostics finally illustrates the clinical uses of the novel approach. Integrating patients' preference into clinical decision making will lead to different cutpoints and different assessments of test performance compared to unweighted policies.