An Evaluation of Multiple Imputation Methods for Hierarchical Data

• Activity Number: 243 - Contributed Poster Presentations: Biopharmaceutical Section
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Takayuki Abe* and Manabu Iwasaki
• Companies: Keio University School of Medicine and Seikei University
• Keywords: longitudinal clinical study ; multiple imputation ; multilevel data ; conditional models ; missing data

Abstract

In clinical studies handling missing data correctly is critical to drawing unbiased inferences on treatment effects. In our research, we focused on longitudinal response with dropout in clinical studies, in which data has hierarchical structure. We evaluated two multiple imputation (MI) strategies, one is multilevel and the other is marginal, as such research is still limited. We developed programming code for the imputation methods for longitudinal data with monotone missing pattern. Specifically, the performance of marginal imputation models was compared with that of multilevel MI for various dependence structures. An example motivated by an actual clinical study was used in the evaluations. Simulation studies were performed for continuous and categorical responses under various settings including hierarchical structures, missing mechanisms, missing fractions, and analysis models (conditional or marginal model). The details of results will be shown in the presentation. This work was supported by JSPS KAKENHI Grant Number 15K08564.