Sensitivity and specificity of a monitoring test

Abstract

The usefulness of a diagnostic test is generally assessed by calculating the sensitivity and specificity, or the predictive value positive and predictive value negative of the test. When subjects are monitored periodically for evidence of disease, these calculations must incorporate the varying amounts of information per individual. If in addition, the test results lie on a continuous scale, these quantities vary with the cutoff value (cutpoint) used to define a positive test. They are usually calculated for a spectrum of potential cutpoints in order to produce receiver-operator characteristic curves. In this paper we use a partial likelihood solution to the discrete logistic model in order to obtain estimates of the diagnostic test indices and to provide a significance test when the diagnostic test is administered repeatedly to individuals.