179 – Advances in Longitudinal Data Analysis
Mixed Model with Repeated Measures: How Many Observations to Include in the Model?
Qing Li
Merck
Ziliang Li
Merck
For clinical trials where repeated measurements from a subject are obtained over the course of study, the Mixed Model with Repeated Measure (MMRM) approach is frequently used to analyze the aggregated data and explore the longitudinal profile of the investigational medication. A common practice when applying the MMRM is to model all non-missing measurements as response. However, due to the complexity of the estimation algorithm and the lack of closed form solution of the estimates, the effect of intermediate measurements on the efficacy estimates is not transparent. The understanding of such effect is of practical importance for study planning and results interpretation. This work focuses on some hypothetical experiments where the MMRM is employed to analyze the change from baseline treatment effect based on repeated observations. Theoretical and numerical properties of the treatment effect estimates, with or without the intermediate observation(s) as response will be discussed. Technical considerations when pre-specifying the analysis method will be highlighted.