Bayesian estimation of variance partition coefficients adjusted for imperfect test sensitivity and specificity.

Abstract

The variance partition coefficient (VPC) measures the clustering of infection/disease among individuals with a specific covariate pattern. Covariate-pattern-specific VPCs provide insight to the groups of individuals that exhibit great heterogeneity and should be targeted for intervention. VPCs should be taken into consideration when planning study designs, modeling data and estimating sample sizes. We present a Bayesian discrete mixed model for the estimation of covariate-pattern-specific VPCs when measurement of the infection/disease is based on an imperfect test. The utility of the presented model is demonstrated with three applications. In all cases, imperfect tests biased VPC estimates towards the null but corrected estimates could be obtained by modeling the sensitivity and specificity of the test procedure with beta distributions. The comparison of adjusted VPCs between the intercept only and the fitted models with higher level covariates explained the portion of heterogeneity in the data that was accounted for by the covariates.