Adjusting for Partially Missing Baseline Measurements with Nonlinear Models in Randomized Trials

• Activity Number: 353
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Chunyao Feng*+ and Chunlei Ke
• Companies: Amgen Inc. and Amgen
• Keywords: Missing Baseline Covariates ; nonlinear models ; imputations

Abstract

In randomized clinical trials, the primary concern is with the estimate of treatment effect which usually is adjusts for some baseline covariates. Missing baseline covariates, if handled inappropriately, will impede the construction of valid and reliable models and will potentially introduce bias in the treatment effect estimate. For normally distributed outcome, White and Thompson(2005) showed a simple mean imputation for the missing baseline covariates would provide an excellent alternative method to handle partly missing baseline covariates in estimating the treatment effect . We will evaluate the performance of the simple mean imputation method for non-linear models such as Cox proportional hazards models and logistic regression models and compared with other common methods to handling missing data. Simulation results and applications will be presented.

Ian R. White and Simon G. Thompson, Adjusting for partially missing baseline measurements in randomized trials, Statistics in Medicine 2005; 24:993-1007)