Mixed model of repeated measures versus slope models in Alzheimer's disease clinical trials

Randomized clinical trials of Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) typically assess intervention efficacy with measures of cognitive or functional assessments repeated every six months for one to two years. The Mixed Model of Repeated Measures (MMRM), which assumes an "unstructured mean" by treating time as categorical, is attractive because it makes no assumptions about the shape of the mean trajectory of the outcome over time. However, categorical time models may be over-parameterized and inefficient in detecting treatment effects relative to continuous time models of, say, the linear trend of the outcome over time. Mixed effects models can also be extended to model quadratic time effects, although it is questionable whether the duration and interval of observations in AD and MCI studies is sufficient to support such models. Furthermore, it is unknown which of these models are most robust to missing data, which plagues AD and MCI studies. We review the literature and compare estimates of treatment effects from four potential models fit to data from five AD Cooperative Study (ADCS) trials in MCI and AD.