Comparison of Missing Data Imputation Methods and Statistical Analysis Methods by Simulation

• Activity Number: 191
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Wenqing Li* and Yun Chon and Yi Wang
• Companies: and Amgen and Amgen
• Keywords: missing data ; imputation

Abstract

In clinical trials, subjects use an electronic diary (eDiary) to report daily disease symptoms or daily medication use, on which longitudinal efficacy endpoints will be derived. Missing data could happen in two levels, which is in daily eDiary level due to subject's missed eDiary information or in longitudinal efficacy endpoint level due to subjects' early drop out. Missing data should be appropriately handled in the derivation of efficacy endpoints as well as in the statistical analysis. In this project, an example of eDiary data is simulated when the summary efficacy endpoint is the count of symptom days. The performance of different imputation methods to impute missing daily eDiary information or to impute missing longitudinal efficacy endpoint, in combination with related statistical analysis methods (such as analysis of covariance, repeated measure etc) were compared in terms of bias and statistical power.