Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 211 - Disease Prediction
Type: Contributed
Date/Time: Tuesday, August 10, 2021 : 1:30 PM to 3:20 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #318497
Title: Flexible Lehman Family ROC Curves
Author(s): Soutik Ghosal* and Zhen Chen
Companies: NICHD/NIH and NICHD/NIH
Keywords: diagnostic accuracy; proportional hazard; proportional reverse hazard; cox regression; ROC curve; AUC

The binormal ROC curve is a widely used statistical tool in diagnostic accuracy and works by assuming normality on test scores from a healthy and a disease population. Lehmann family ROC model provides an attractive alternative to its binormal counterpart, as it is distribution free and focuses on the relation of diseased and healthy scores. It has been shown that the Lehmann family ROC model can be easily fit by a proportional hazards model and covariate effects can be directly assessed on the ROC curve. However, a Lehmann family ROC curve can be restrictive, as it rises quickly at low false positive rate region. We propose to extend the Lehmann family ROC model to overcome this pitfall. We first consider an ROC model based on Lehmann's alternative. We show that this new proposal is equivalent to a proportional reversed hazards model and that it can be fit easily using existing survival analysis software. We further consider a more flexible framework that is a mixture of Lehmann family and Lehmann's alternative. An EM algorithm is developed to carry out the model estimation. We conduct both simulation and real data analysis to illustrate how these extensions work.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program