Covariance Adjustment of Relative-Risk Estimates in Matched Studies

The matched-sample case-control study is widely used by epidemiologists to estimate the relative incidence of disease among persons exposed to different levels of one or more risk factors. An apparent limitation of such designs has been the necessity to control the effects of all potential confounding variables in the matching process. Conditional likelihood analyses based on the linear logistic equation enable one to model the effects of covariates while preserving the original matching. This paper reviews the logical basis for this methodology and illustrates its value by application to a case-control study of esophageal cancer.