Relative efficiency of longitudinal, endpoint, and change score analyses in randomized clinical trials

In the last two decades, the design of longitudinal studies, especially the sample size determination, has received extensive attention (Overall and Doyle 1994; Hedeker et al. 1999; Roy et al. 2007). However, there is little discussion on the relative efficiency of three strategies widely used to analyze randomized clinical trial data: a full longitudinal analysis using all data measured over time, an endpoint analysis using data measured at the endpoint (the primary time point for outcome evaluation), and a change score analysis using data measured at the baseline and the endpoint. When designing randomized clinical trials, investigators usually need to decide whether they would collect the interim data and if so, which type of analysis among the three would be the primary analysis. In this work, we compare the relative efficiency of detecting an intervention effect in randomized clinical trials using longitudinal, endpoint, and change score analysis, assuming linearity of the outcome trajectory and several commonly used within-individual correlation structures. Our analysis reveals an important and ∗Corresponding author