Keywords: MMRM, Mixed effects model
Clinical trials in diabetes, schizophrenia, and Alzheimer’s Disease measure effects of treatments on each patient repeatedly at multiple time points. Such data are typically analyzed using a mixed effects model. Insight into the variance-covariance of the Repeated Measures is given by deriving and examining its structure for a simple Random coefficients response model. To isolate the consequence of specifying a variance-covariance structure far from the truth, we cast our investigation in a setting where the fixed treatment effect is orthogonal to the random effect and the interactions. By analyzing data sets satisfying such orthogonality condition under different variance-covariance types, we find the type specification itself has a significant impact on testing for treatment efficacy at multiple time points.