Use of Novel Sequential Multiple Imputation Analysis for Extrapolation of Missing Data Points as Part of In Vivo Tumor Growth Assays

• Activity Number: 244
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xiaoli Hou*+ and George N. Naumov
• Companies: Merck and Merck Research Laboratories
• Keywords: censored data ; missing data ; xenograft assay ; in vivo studies ; multiple imputation

Abstract

In vivo tumor models, such as xenografts, play an important role in cancer research. In these models, animals are typically inoculated with cancer cells, treated with anticancer agents, and monitored for tumor growth. Often censored data evolve due to animal health deterioration. These events result in missing data. Ignoring or inappropriately handling missing data may lead to incorrect conclusions, especially for missing not at random (MNAR). Standard statistics for complete-case analyses usually are biased or misleading. Unbiased results can be obtained by truncating the data set prior to the censored event, but loss useful information. A common practice for missing data is using multiple imputations (MI). Most of the MIs are designed for clinical trials or surveys. We propose a sequential multiple imputations (SMI) for quantitative estimation of censored data NMAR, in which missing data mechanism doesn't need to be modeled. SMI predicts missing data, draw values from multiple predictions, and then determines single imputations sequentially by a pre-specified role. Simulation studies demonstrate the strength of SMI and gives unbiased estimates.