Model-based approach and signal detection theory to evaluate the performance of recruitment centers in clinical trials with antidepressant drugs.

The objective of this study was to characterize the performance of each recruitment center in multicenter clinical trials and to provide criteria to discriminate between informative and noninformative centers using the signal detection approach. Data were derived from the GlaxoSmithKline Clinical Trial database on paroxetine and bupropion, totaling 4,016 subjects with major depressive disorders (MDDs) across nine trials. The probability of observing clinically relevant difference of active treatment from placebo was estimated in each center as a function of the placebo Hamilton Depression Rating Scale (HAMD-17) scores at baseline and at week 8. The center's performance was defined using the posterior probability (PP) of detecting a signal of a treatment effect. Only 60% of the centers were classified as informative. In these centers, the signal of treatment effect increased by approximately 80%. The signal detection approach appears to be a useful methodology to rank the performance of recruitment centers and to classify each center as informative or not in respect of detection of clinically relevant signals of efficacy. A further analysis indicated that a minimal sample of four subjects is required in order to predict the typical placebo response in each center.