ABSTRACT
A matched pairs study design is used in medical research to minimize variability caused by extraneous variables. Many such studies result from individuals being artificially matched on a set of known covariates. When individuals in a pair are randomized to receive two different treatments, then the comparison is done between two subjects that are alike. Hence, the difference in outcome can be directly attributed to treatment effect. Paired data can also be obtained from observations with a biological link such as pairs of organs (eyes, kidneys, knees, etc.) from one person or pairs of twins or siblings. Another type of paired data arise from pairs of observations with different baseline survival or hazards for each pair, for example, pairs of patients treated at the different centers. Regardless of the pairing mechanism, outcomes between individuals might be correlated. Comparison of outcomes between treatments in such a study must account for this correlation. For positively correlated outcomes, methods for unpaired data ignoring correlation between individuals within pairs may underestimate treatment effect.
