Activity Number:
|
294
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract - #309491 |
Title:
|
A Bayesian Nonparametric Multivariate Model for Evaluation of Correlated Diagnostic Tests
|
Author(s):
|
Adam Branscum*+ and Timothy E. Hanson and Ian Gardner
|
Companies:
|
University of Kentucky and The University of Minnesota and University of California, Davis
|
Address:
|
College of Public Health, Lexington, KY, 40536,
|
Keywords:
|
Bayesian nonparametrics ; Polya trees ; Diagnostic tests
|
Abstract:
|
We develop a flexible nonparametric model for the statistical analysis of multivariate serologic data. In this setting, sampled individuals are tested for a condition of interest using several imperfect continuous diagnostic tests. To provide for a degree of robustness to the structure imposed by multivariate normal models, the data for diseased and non-diseased individuals are modeled with independent multivariate mixtures of Polya trees. Data-driven estimates are available for ROC curves and areas under them for all diagnostic tests under consideration. We consider a tractable two-stage empirical Bayesian estimation procedure that is applied to the evaluation of two ELISA tests for Johne's disease.
|