Matched case-control designs and overmatched analyses.

Abstract

In a 1:1 matched-pairs case-control study, several pairs may have the same matching criteria. The objective of this paper is to consider various analytic approaches for odds ratio estimation in this situation. One approach is to perform the usual matched-pairs analysis; the alternative is to combine pairs into strata and compute the conditional maximum likelihood or Mantel-Haenszel estimate. Matched-pairs analysis based on theoretic and simulation results may give misleading inferences, although on average it performs nearly as well as the stratified analysis. The occasional poor performance of the matched-pairs analysis results from the possibility of obtaining an extreme number of discordant pairs due to the random pairing within a stratum. It is concluded that the prudent policy is to perform a stratified rather than a paired analysis, even at the expense of additional computation.