Estimation of the combined response to treatment in multicenter trials

Analyses of multicenter trials consider the estimated treatment effect differences of the individual centers and combine them into an estimate of the overall treatment effect. There has been much debate in the literature concerning the best way to combine these treatment effect differences. We emphasize that first of all one should define the combined response to treatment (CRT), the object that has to be estimated from the results of a multicenter clinical trial. It is shown that the choice of CRT determines not only the best estimator, but also the allocation of patients among the centers that minimizes the mean squared error. A new estimator of the CRT is proposed that is based on a preliminary clustering of the centers and the use of a weighted average of the Type I estimators obtained from within each cluster. The clustering aims to minimize the bias of the combined estimator. We show via a simulation study that the simple clustering procedure provides a reasonably improved estimator. The clustering can be done on blinded data, as long as the numbers of patients on each treatment arm in each center are known. The methodology is illustrated by analyzing a multicountry, multicenter trial to compare an active treatment with placebo for the treatment of a psychiatric disorder.