Efficient statistical tests to compare Youden index: accounting for contingency correlation

摘要: Youden index is widely utilized in studies evaluating accuracy of diagnostic tests and performance predictive, prognostic, or risk models. However, both one two independent sample on have been derived ignoring the dependence (association) between sensitivity specificity, resulting potentially misleading findings. Besides, paired test currently unavailable. This article develops efficient statistical inference procedures for sample, independent, by accounting contingency correlation, namely associations specificity samples typically represented tables. For tests, variances are estimated Delta method, based central limit theory, which then verified bootstrap estimates. test, we show that covariance sensitivities specificities can be as a function kappa statistic so readily carried out. We remarkable variance using constrained optimization approach. Simulation performed to evaluate properties tests. The proposed approaches yield more stable type I errors at nominal level substantially higher power (efficiency) than does original Youden's Therefore, simple explicit large solution performs very well. Because implement asymptotic exact computation with common software like R, method broadly applicable evaluation model performance.