A Tutorial for Handling Suspected Missing Not at Random Data in Longitudinal Clinical Trials

Missing data in longitudinal randomized clinical trials, even if assumed to be missing at random (MAR), can result in biased parameter estimates and incorrect treatment conclusions. If missing data are suspected to be missing not at random (MNAR, i.e., missing data due to the unobserved values themselves), accepted missing data handling techniques are inadequate and the problem is compounded due to the introduction of additional bias. The goal of this paper is to provide trialists the analytic tools and methodological steps needed to assess treatment effect veracity if MNAR is suspected, enabling researchers to reach reasonable and defensible treatment effect conclusions. The explanations, steps, and conclusions are demonstrated using trial data involving binge drinking behavior among college students. All analysis model diagrams are presented, and both linear model equations and Mplus syntax scripts for all analyses are included in a supplemental Appendix to further provide trialists the methodological tools needed to further test treatment effect estimates when MNAR is suspected.