Confidence interval construction for the Youden index based on partially validated series

Confidence interval construction for the Youden index of a diagnostic test based on partially validated series with dichotomous response is considered in this article. Using the Wald and Agresti-Coull, the Wilson score, logit-transformation and the method of variance estimates recovery, eight methods for constructing confidence intervals for a single Youden index and nine methods for the difference between two independent Youden indices are developed. Comparisons among the various methods with respect to their empirical coverage probabilities and confidence interval widths are conducted through simulation studies over a variety of parameter settings. Based on the simulation results, recommendations for practical use under different conditions are provided. A real aplastic anemia data set is used to illustrate the proposed methods.