An Analytic Road Map for Incomplete Longitudinal Clinical Trial Data
*Craig H Mallinckrodt, Eli Lilly 

Keywords: Clinical trials, Missing data, Longitudinal data, Maximum likelihood, Multiple Imputation

Common practice for longitudinal clinical trials has shifted away from last- and baseline-observation carried forward (LOCF, BOCF) methods that are based on the restrictive missing completely at random (MCAR) assumption. More principled methods based on the less restrictive assumption of missing at random (MAR) are now routinely implemented and methods assuming data are missing not at random (MNAR) have been given extensive attention in the literature. However, ambiguity remains regarding which methods are most appropriate for various scenarios. Results from simulation studies and actual clinical trial data will be used to support the position that LOCF and BOCF are not appropriate choices for the primary analysis and that in many areas of drug development the primary analysis can be based on an MAR approach, with MNAR methods and other tools used as sensitivity analyses.