ABSTRACT
Matching in case-control studies usually requires a matched analysis to avoid bias. Matching in randomized trials and cohort studies, however, may be free of bias if the matching is ignored in the analysis; requirements for this are discussed. When risks or rates are the outcomes, a matched analysis of trials and cohort studies may provide no gain in efficiency. Matching can be used to control confounding in some trials and cohort studies. Matching can be advantageous when a variable is hard to measure. In a matched trial or cohort study, only the sets with outcomes are needed for unbiased analyses. In some study designs, subjects can be matched to themselves. Stratified analysis of matched data. Differences between conditional logistic and conditional Poisson regression. The equivalence of Cox and Poisson methods for matched sets. Matched analyses for risk ratios, for rate ratios with a non-recurrent outcome, and for rate ratios with recurrent outcomes.
