An Analytical Variance Estimator of Partial Area Under the Empirical ROC Curve

• Activity Number: 506
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Andriy Bandos*+ and Ben Guo and Gur David
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: ROC analysis ; partial AUC ; nonparametric variance ; resampling

Abstract

Evaluation of diagnostic accuracy of medical devices and technologies is often performed based on the Receiver Operating Characteristics (ROC) curve. Because of the cost and benefits constraints, in most studies only a part of the ROC curve (e.g., in the range of low chance of false positives) describes characteristics that are of clinical interest and practice relevance. Partial area under the ROC curve (pAUC) is a conventional summary index of a clinically relevant part of the curve. We demonstrate that some of the conventional approaches for statistical inferences based on the empirical pAUC could lead to incorrect inferences, at least for the ROC curves frequently encountered in diagnostic medicine. We developed an analytical variance estimator for the empirical pAUC and we demonstrate its superiority in terms of bias, MSE-accuracy, and type I error rate of the Wald-type test over a wide range of practical scenarios. Provided example illustrates how the developed variance estimator could be used for simple inferences and for planning nonparametric statistical inferences based on pAUC.