Applications of Simulation for Missing Data Issues in Longitudinal Clinical Trials

Abstract

Missing data handling in longitudinal clinical trials has gained considerable interest in recent years. Although a lot of research has been devoted to statistical methods for missing data, there is no universally best approach for analysis. It is often recommended to perform sensitivity analyses under different assumptions to assess the robustness of the analysis results from a clinical trial. To evaluate and implement statistical analysis models for missing data, Monte-Carlo simulations are often used. In this chapter, we present a few simulation-based approaches related to missing data issues in longitudinal clinical trials. First, a simulation-based approach is developed for generating monotone missing data under a variety of missing data mechanism, which allows users to specify the expected proportion of missing data at each longitudinal time point. Secondly, we consider a few simulation-based approaches to implement some recently proposed sensitivity analysis methods such as control-based imputation and tipping point analysis. Specifically, we apply a delta-adjustment approach to account for the potential difference in the estimated treatment effects between the mixed model (typically used as the primary model in clinical trials) and the multiple imputation model used to facilitate the tipping point analysis. We also present a Bayesian Markov chain Monte-Carlo method for control-based imputation which provides a more appropriate variance estimate than conventional multiple imputation. Computation programs for these methods are implemented and available in SAS.