Nonmaleficence and the Diagnostic Cutpoints

Abstract

The choice of cutpoints in clinical epidemiology involves trading off the utility gain of the true-positives with the utility loss of the false-positives. For tests with a continuous scale of values, this paper extends the ROC framework to include the a priori prevalence and the test's ratio of utility gain and utility loss, the so called 'relative benefit factor'. It formalizes the cutpoint choice by recurring to the principle of nonmaleficence and analyzes the effect of changes in prevalence, treatment outcome and exogenous improvements of test accuracy on the cutpoint.