Precision medicine demands diagnostic tests evaluated within clinically relevant operating ranges. In cardiovascular practice, controlling the False Positive Rate is essential to avoid unnecessary invasive procedures. This dissertation addresses two regional accuracy measures, the partial Youden Index and the partial area under the ROC curve (pAUC),under verification bias, which arises when gold-standard disease confirmation is available only for a selected patient subset under the missing at random assumption.

We make three contributions. First, we extend four bias-correction estimators, full Imputation, mean score imputation, inverse probability weighting, and the semiparametric efficient estimator, to the partial Youden Index, yielding 16 hybrid estimators. SPE-based hybrids maintain low bias and mean squared error across correct and misspecified models in the settings considered, leveraging the double robustness of the SPE component.

Second, we develop and compare confidence interval methods for the partial Youden Index, spanning MOVER-based and bootstrap-based families. Bootstrap-based intervals combined with the SPE estimator achieve near-nominal coverage across all misspecification scenarios.

Third, we introduce a Hybrid Empirical Likelihood framework for pAUC confidence intervals under verification bias. Traditional Wald-type intervals show under-coverage across the scenarios, while the proposed HEL intervals demonstrate a self-adjusting property that maintains more reliable coverage. Simulation studies and real data analyses support the use of SPE-based and HEL-based methods when full disease verification is infeasible.