Answers are grounded in this paper and cite their sources.
Accurate evaluation of the discriminatory capability of diagnostic biomarkers is essential for disease detection and clinical decision making. The receiver operating characteristic (ROC) curve illustrates the trade-off between sensitivity and specificity, and the area under this curve (AUC) is the most widely used measure of diagnostic accuracy. However, AUC may conceal informative structure when a biomarker has a non-monotone relationship with disease and the ROC curve crosses the diagonal. The length of the ROC curve provides a direction-invariant geometric summary that is related to distributional divergence and the optimal discriminatory capacity of a biomarker. We propose a piecewise linear estimator constructed from survival probabilities evaluated on a threshold grid. We first develop the method under a Gaussian parametric model and then consider its nonparametric implementation using empirical survival probabilities. We investigate consistency and asymptotic normality and compare the proposed methods with the kernel-density estimators using several bandwidth selectors. Simulation studies evaluate the finite-sample performance of the proposed estimators under both proper and improper ROC settings. Applications to two medical data sets illustrate how ROC length can complement AUC when the score distributions differ in ways that are not adequately represented by a single threshold direction.
No takes yet. Share an insight, caveat, or question.
Gang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: