An Application of Multiple Imputation in Analyses of Longitudinal Clinical Trials with Left-Truncated Data

• Activity Number: 239
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Guanghan Liu*+
• Companies: Merck Research Laboratories
• Keywords: Multiple imputation ; Longitudinal analysis ; Left truncation

Abstract

For analysis of lab measurements collected in clinical trials, the outcomes may suffer from left truncation at the detection limit of the lab assay. In the conventional methods, the left-truncated values are often replaced with the detection limit or half of the detection limit. This simple single imputation could be biased and under estimating the variance although the impact may be limited when the proportion of truncation is small. Likelihood based methods were proposed in the literature. However, the methods can be complicated and suffered convergence problems because of the numerical integration. In this talk, we investigate a multiple imputation approach. Because the truncated values are known to be less than the detection limit, special considerations are needed to implement the imputation process. We evaluate this approach and compare it with the simple imputation and the full maximum likelihood approach with simulations. An application to a vaccine trial will also be presented.