Is bigger better for depression trials?

Objective: A key assumption underlying the principle that power increases with sample size is that the standardized effect size is fixed over time. In therapeutic areas where it may be difficult to continually recruit from a homogeneous population, this assumption may not be valid; patients randomized toward the end of enrollment may derive from a more heterogeneous population and negatively impact the power of a study. Post hoc analyses were performed on clinical data from four phase III depression trials with paroxetine to evaluate this possibility. Methods: Each study used a randomized, double-blind, placebo-controlled design and enrolled approximately 150 patients per treatment arm. Plots of observed p-values for the treatment difference between paroxetine and placebo (on the HAM-D-17 change from baseline score at week 8) by cumulative enrollment were made for each study. Results: As previously reported, three of the four studies showed an overall significant treatment effect and one did not. In each study, a significant treatment effect was observed before approximately 100 patients had been enrolled per treatment arm. Continuing to enroll additional patients did not maintain the achieved level of significance in most instances, and in one case appeared to alter a potentially positive study into a failed study. Plots of p-values versus cumulative enrollment by patient quarters using combined data from all four studies suggested that late-enrolling patients were more likely to be placebo responders than early-enrolling patients. Hypothesized explanations for this finding include a depleted pool of depressed patients and the rush for patient recruitment at the end of a study in order to meet completion timelines. However, no corroborative evidence could be found to support either possibility. Conclusions: This analysis demonstrates that bigger is not necessarily better for depression trials. (C) 2007 Elsevier Ltd. All rights reserved.