Missing Data Handling for Composite Endpoints

• Activity Number: 324
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Sabrina Wan* and G. Frank Liu and Weili He
• Companies: Merck and Merck and Merck
• Keywords: composite endpoint ; missing data ; multiple imputation

Abstract

This talk is on missing data handling in clinical trials with composite endpoints derived from underlying component variables. This topic is motivated by a study with data collected longitudinally. Missing data may occur at any time point for a given component which results in missing data of the composite endpoint. One common approache is to impute missing data by a single imputation method such as using the mean score and to derive the component variable based on the imputed values. Then some model-based missing data handling method is applied to the composite endpoint. This two-step method does not incorporate the uncertainty associated with the imputation of the component variable. In addition, the information from the adjacent time point is not used in the imputation of the component variables, which decreases the accuracy of the final estimates. We propose to perform multiple imputations over time and in component level with more appropriate imputation models, such as cLDA for underlying continuous variables. Simulations and application to a real dataset are performed to assess and compare different methods.