Marginal Analysis for Cluster - Based Case - Control Studies

Cluster-based case-control design refers to a design where the sampling unit is a cluster and the sampling probability depends on the responses from individuals within the cluster. Data from a cluster-based case-control design arise in many practical applications. For example, in some epidemiologic genetic studies, due to the low prevalence of the disease of interest, families with more members having the disease are sampled with a higher probability. Current approaches for analyzing this type of data rely mainly on parametrically modeling the joint distribution of the responses within a cluster. In this paper, we develop a marginal approach to analyze data from cluster-based casecontrol studies when the main interest is the mean structure of the association between exposures and outcomes and the correlation within cluster is considered nuisance. We specify a marginal regression model for an individual response given covariates and leave the correlation within the cluster unspecified. We establish the statistical properties for the proposed estimator and investigate its finite sample performance through simulation studies. We apply the proposed method to a data set from the Baltimore Eye Survey.