B1996
Title: Statistical inference for the comparison of two correlated biomarkers using the partial volume under the ROC surface
Authors: Katherine Young - University of Kansas Medical Center (United States) [presenting]
Leonidas Bantis - University of Kansas Medical Center (United States)
Abstract: Tuberculosis (TB) is the second leading cause of infectious disease worldwide. The development of biomarkers that can accurately and quickly diagnose TB in both its latent and active forms is a high priority for reducing its mortality. A framework is provided for comparing the accuracy of two biomarkers in a clinical region of interest for classifying patients into three ordinal groups: healthy, latent infection, and active infection. The partial volume under the ROC surface (pVUS) has been proposed as a measure of accuracy in such settings. We propose methods for estimating and constructing confidence intervals for the difference in pVUS's for two biomarkers collected from the same individuals. We propose both parametric and non-parametric methods and evaluate them through extensive simulations. We then apply these methods to compare candidate biomarkers in their ability to diagnose both latent and active TB.
