Optimal dichotomization of screening test variables

In the application of a screening test to ascertaining the status of a characteristic on individuals from a certain population, the accuracy of the screening test often depends on the dichotomization of the test outcome variable. There is then the need to find a most favorable dichotomizer for which optimal performance of the test is obtained. In this paper, determination of optimal dichotomizers is considered in a decision-theoretic Bayesian approach. Parametric models are introduced for continuous, numerical discrete, and ordinal categorical screening variables. Optimal dichotomization adjusted for individual-dependent covariates also is discussed. When the within-model parameters are unknown, predictive inference of optimal dichotomizers is emphasized.