Modeling Association Among Multivariate Exposures In Case -Control Studies

The paper considers the problem of modeling association among multiple exposures in a matched case-control study where some of the exposures may be partially missing. The exposure variables could all be categorical or continuous or could be a mixed set of some categorical and some continuous variables. Association models using completely observed covariates are proposed for each of the three scenarios. The models account for varying stratum heterogeneity in different matched sets. Three real data examples accompany the proposed models. The examples as well as a small scale simulation study indicate that in presence of missingness and association, modeling the association between the exposures rather than ignoring it, leads to better estimates of the relative risk parameters with smaller standard errors. Estimation of the model parameters is carried out in a Bayesian framework and the estimates are compared with classical conditional logistic regression estimates.