Impact of Stratified Randomization on Estimating Treatment Effect in ANCOVA Models

• Activity Number: 35
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Peter G Mesenbrink*+ and Jeff Maca
• Companies: Novartis and Novartis
• Keywords: stratification ; randomization ; statistical power ; ANCOVA

Abstract

A common consideration when designing a clinical trial is to stratify over known factors. These factors are chosen because they are believed to have a strong effect on the primary efficacy outcomes. These factors are incorporated over the concern that without inclusion the treatment groups would be unbalanced and lead to loss in study efficiency or power. We will examine different numbers of stratification factors and different magnitudes of effect for the stratification factors relative to the magnitude of the treatment effect in an ANCOVA model. In simulation studies conducted, even when the magnitude of the effect of the stratification factor is equal to or greater than the treatment effect, the loss of power is minimal even for small sample sizes (n~100)and there is minimal difference in estimation of treatment effect when adjusting the model using the factor(s) as continuous covariate(s) vs. stratifying the model with factors at multiple levels.