-----------------------------------------------------------

Statistics Seminar Series

Thursday, February 8, 2007 @ 1:00PM
Cullimore Hall, Room 611
New Jersey Institute of Technology

ROC Analysis For Longitudinal Disease Diagnostic Data Without A Gold Standard Test

Chong Wang, Ph.D

Department of Mathematics
Cornell University

We develop a Bayesian methodology based on a latent change-point model to estimate the ROC curve of a diagnostic test for longitudinal data. We consider the situation where there is no perfect reference test, i.e. no "gold standard". A change-point process with a Weibull-like survival hazard function is used to model the progression of the hidden disease status. Our model adjusts for the effects of covariate variables, which may be correlated with the disease process or with the diagnostic testing procedure, or both. Markov chain Monte Carlo methods are used to compute the posterior estimates of the model parameters that provide the basis for inference concerning the accuracy of the diagnostic procedure. Based on our Bayesian approach, the posterior probability distribution of the change-point onset time can be obtained and used as a new criterion for disease diagnosis. We discuss an application to an analysis of ELISA scores in the diagnostic testing of paratuberculosis (Johne's disease) for a longitudinal study with 1997 dairy cows.