Semiparametric approaches for matched case–control studies with error-in-covariates

The matched case–control study is a popular design in public health, biomedical, and epidemiological research for human, animal, and other subjects for clustered binary outcomes. Often covariates in such studies are measured with error. Not accounting for this error can lead to incorrect inference for all covariates in the model. The methods for assessing and characterizing error-in-covariates in matched case–control studies are quite limited. In this article we propose several approaches for handling error-in-covariates that detect both parametric and nonparametric relationships between the covariates and the binary outcome. We propose a Bayesian approach and two approximate-Bayesian approaches for addressing error-in-covariates that is additive and Gaussian, where the variable measured with error has an unknown, nonlinear relationship with the response. The Bayesian approaches use an approximate latent variable probit model. All methods are developed using the nonparametric method of low-rank thin-plate splines. We assess the performance of each method in terms of mean squared error and mean bias in both simulations and a perturbed example of 1–4 matched case-crossover study.