Adjusting for Missing Covariates in Bayesian Latent Class Models for Diagnostic Test Data

• Activity Number: 194
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): ZhuoYu Wang*+ and Nandini Dendukuri and Lawrence Joseph
• Companies: McGill University and McGill University and McGill University
• Keywords: Bayesian ; latent class model ; diagnostic test ; missing covariate ; conditional dependence

Abstract

Covariates that influence the sensitivity and/or specificity of different diagnostic tests can create correlation between these tests, conditional on disease status. Thus, ignoring such covariates in a latent class analysis of imperfect tests would amount to ignoring conditional dependence, leading to biased estimates of test accuracy and prevalence. For the case of a dichotomous covariate affecting two imperfect tests, we derived an expression to show that the conditional covariance is a function of the product of the differences in each test's sensitivity (or specificity) between the groups defined by the covariate. For a uniformly distributed continuous covariate, similar results were obtained numerically. Using series of simulated datasets we studied whether in the absence of covariate, unbiased estimates may be obtained by fitting an extended latent class model that allows for conditional dependence. We found that a conditional dependence model, which places no constraints on the covariance between the tests, works well in adjusting for both types of the covariates. Our methods will be illustrated by application to a problem on evaluation of tuberculosis diagnostic tests.