THE ANALYSIS OF CASE-CONTROL STUDIES UNDER VALIDATION SUBSAMPLING

In case-control studies, one often finds that covariates are missing o r measured with error in the entire sample, whereas complete or exact covariate information is available only in a subsample. This paper dis cusses the analysis of case-control studies under a double-sampling sc heme, where at the first stage covariates are measured with error or m issing, and at the second stage are validated in a subsample. The meth od proposed combines the risk information from both samples by assumin g that (1) the disease incidence model is logistic, (2) the partial or proxy information takes on finitely many values, and (3) the error is non-differential. The estimator is obtained by jointly fitting logist ic models to the first and second stage data, a variance formula is pr esented. Parameters can be estimated by use of standard packages for d ose-response data. Data from an ongoing case-control study on lung can cer serve as an example.