Bias-Corrected Partial AUC Estimation in Exponential ROC Analysis with Clinical Case Studies

Partial AUC Estimation

Résumé

In diagnostic research, ROC-based performance evaluation is widely used to quantify the discriminative ability of clinical markers. In many practical situations, investigators are interested in a limited false positive range rather than the full curve, which makes the partial area under the ROC curve (pAUC) a more meaningful measure. Existing literature indicates that most pAUC-based developments assume normality, whereas several clinical measurements exhibit non-normal distributions, particularly exponential-type distributions. Under such conditions, conventional pAUC estimates tend to produce biased results and reduce the reliability of interpretation. In this work, we develop a new bias-corrected estimator for partial AUC under the exponential ROC framework. A closed-form expression for the corrected estimator is established, and its statistical behaviour is studied through simulation experiments. Performance is evaluated in terms of bias and mean squared error (MSE). The results demonstrate noticeable improvement over the standard estimator, particularly under non-normal exponential scenarios. The usefulness of the approach is further demonstrated using real clinical case studies. This contribution supports more accurate pAUC computation in exponential ROC settings and can enhance evidence-based medical decision-making.