Online Program Home
  My Program

Abstract Details

Activity Number: 666 - Prediction and Calibration
Type: Contributed
Date/Time: Thursday, August 3, 2017 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #322812 View Presentation
Title: Beta Regression for Modeling the ROC as a Function of Continuous Covariates
Author(s): Sarah Stanley* and Jack Tubbs
Companies: Baylor University and Baylor University
Keywords: Beta regression ; ROC analysis ; placement values ; diagnostic tests
Abstract:

The receiver operating characteristic (ROC) curve is a well-accepted measure of accuracy for diagnostic tests. In many applications, test performance is affected by covariates which should be accounted for in analysis. Several regression methods have been developed to model the ROC as a function of covariates within a generalized linear model framework. Two such methods, a parametric and semiparametric approach, estimate the ROC using binary indicators based on placement values which quantify the probability that a test result from a non-diseased subject exceeds that of a diseased subject with the same covariate values. A consequence of using binary indicators in this way is added correlation. As an alternative, we propose a new method that models the distribution of the placement values directly through beta regression. Given that the placement values are independent probabilities, a direct beta model is easily implemented and avoids the added correlation in the previous methods. We compare our beta method with the pre-existing parametric and semiparametric approaches via simulation and show that the new method yields comparable ROC estimates without adding correlation.


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

Back to the full JSM 2017 program

 
 
Copyright © American Statistical Association