Extending the skill test for disease diagnosis.

For diagnostic tests, we present an extension to the skill plot introduced by Briggs and Zaretski (Biometrics 2008; 64:250-261). The method is motivated by diagnostic measures for osteopetrosis in a study summarized by Hans et al. (The Lancet 1996; 348:511-514). Diagnostic test accuracy is typically defined using the area (or partial area) under the receiver operator characteristic (ROC) curve. If partial area is used, the resulting statistic can be highly subjective because the focus region of the ROC curve corresponds to a set of low false-positive rates that are chosen by the experimenter. This paper introduces a more objective diagnostic test for which the focus region depends on a skill score, which in turn depends on the loss functions associated with misdiagnosis. More specifically, the skill-based diagnostic test serves as a more objective version of the nonparametric test introduced by Dodd and Pepe (Biometrics 2003; 59:614-623).