Sample Size Consideration for Mixed Models

• Activity Number: 458
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Richard McNally*+ and Yuki Matsushima
• Companies: Covance and Otsuka Pharmaceutical Co.
• Keywords: Mixed models ; Growth curve models ; clinical trials ; sample size ; power ; missing data

Abstract

We discuss methodology for computing the sample size for a clinical trial where the primary outcome is measured repeatedly and analyzed using mixed models. We focus on two types of models: in the first the outcome is averaged over multiple visits (i.e., time-averaged difference model; TADM), and in the second a regression line is fit for each treatment (i.e., growth curve model; GCM). The discussion for the TADM focuses on a mixed model with a combined random-intercept/first-order autoregressive (AR(1)) covariance structure. This type of covariance structure is useful even when the planned analysis calls for an unstructured covariance matrix. Special cases are the pure AR(1) and the compound symmetry covariance structures. For the GCM, it is assumed the model has both random intercepts and slopes for each subject. We have derived sample size formulas for both models which can be applied to special cases such as the GCM with no random slope. We discuss how to adjust the sample size for dropout and show simulation results for both models with and without dropout. We also discuss how to extend the TADM formula to more general covariance structures.