Constrained Longitudinal Data Analysis as an Alternative to Multiple Imputation for Handling Missing Data in Randomized Clinical Trials

• Activity Number: 308
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Jin Xu*+
• Companies: Merck
• Keywords: ANCOVA ; cLDA ; missing data ; multiple imputation

Abstract

In randomized trials with a pre-post design, responses are measured at pre-treatment (baseline value) and at post-treatment times. The baseline is often adjusted in the analysis models such as ANCOVA. Despite the best effort, some measurements are inevitably missing for various reasons. When the amount of missing values is substantil the ANCOVA is suboptimal because the method includes only the complete cases. How to handle the missing data then becomes an important issue in estimating the treatment difference. Two major approaches are commonly used: multiple imputation (MI) and maximum likelihood (ML). Assuming data are missing at random, both MI and ML have good statisticial perporties in terms of bias in the estimate, Type I error control and power. The MI methods are more commonly selected compared with the ML methods because of its simplicity and its ease to understand. We compared the performance of cLDA (an ML method) with several MI methods, and show that cLDA can be an attractive alternative to MI methods in analyzing randomized clinical trials with a pre-post design. Simulations and real trial applications will be used for the comparisons and discussions.