“A novel alternative model to MMRM as the primary model for sporadic AD clinical trials: the proportional MMRM”
Background – The standard statistical model used in the primary efficacy analysis for phase III Alzheimer’s disease (AD) trials is the mixed effects model for repeated measures (MMRM). The MMRM is a large parametric model without any assumption on the underling longitudinal trajectory and is well accepted by regulatory agencies. The primary efficacy analysis, however, hinges only on the baseline and end-of-study assessments and other intermediate assessments provide little information on the test of efficacy hypothesis, leading to inefficient use of study data and lower power. We developed a novel proportional MMRM (PMMRM) which assumes the efficacy effect from the active treatment is proportional to the change in the placebo group at each post-baseline visit.
Methods – We simulate clinical trials with different sample sizes and durations, and allow extended follow-up for early enrollees while awaiting the late ones to complete the study. Individual data are simulated from MMRM or proportional MMRM based on multivariate normal distribution. We then compared the power and type I error between MMRM and PMMRM models.
Results – Type I error for PMMRM is well controlled. Compared with MMRM, the PMMRM leads to large increase in power, and utilizes data from extended follow-up for early enrollees.
Conclusion – The use of PMMRM can largely reduce the sample size for AD trials. The PMMRM enables all study assessments to contribute to the test of null hypothesis and can offers flexibility by allowing a consistent proportion or multiple proportions (e.g. easily take into account of practice effect by using a different proportion for the 1st visit post baseline). The PMMRM can be implemented using standard SAS procedures such as Nlmixed or MCMC, thus has great potential to serve as an alternative primary analysis model to the MMRM.
Division of Biostatistics seminars
For inquiries contact Chengjie Xiong.