Estimating the utility of matching in case-control studies

Abstract

The estimation of the utility of matching in completed case-control studies is examined. The utility is defined in terms of providing an estimate of a more valid odds ratio than the crude odds ratio, and in providing a more efficient study than an unmatched study with a stratified analysis. Methods are given to estimate the utility of frequency-matching in a completed study where the cases and controls are matched into broad strata such as age or decade of birth. An example from the literature demonstrates the technique. Since the proposed analysis requires knowledge of the proportions of individuals in the study population associated with each stratum, it is recommended that estimates of these proportions be given in the results of frequency-matched case-control studies. For pair-matching where each pair is regarded as defining its own stratum (such as with neighborhood or friend controls) it is shown to be impossible to estimate the utility of the matching from a completed study. In particular, it is incorrect for either frequency- or pair-matching to try to estimate the utility of matching by performing an unmatched analysis on a completed case-control study.