A Comparison of Methods for Adjusting for the Baseline Measure

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Martin Ove Carlsson*+ and Kelly H. Zou and Ching-Ray Yu and Franklin W. Sun
• Companies: Pfizer Inc. and Pfizer Inc. and Pfizer Inc. and Pfizer Inc.
• Keywords: Clinical Trial ; Baseline Measure ; Change Score ; Percent Change from Baseline ; Scale Invariant ; Analysis of Covariance

Abstract

When analyzing randomized controlled clinical trials from baseline to the end of the study, we may encounter that within-subject baseline measures can impact inferential results once they are adjusted, compared with results from unadjusted methods. Various statistical methods may be employed to adjust for the baseline measure. For example, post-treatment score or change from baseline may be analyzed with the baseline measure as a covariate, along with additional available baseline covariates. The analysis of covariance is a common regression method for this adjustment. Alternatively, the percent change from baseline, a dimensionless quantity that is scale invariant, can be used as a covariate. The percent change is a convenient measure for examining the entire distribution of the responses. However, its exact distribution is not trivial, particularly when baseline and post-treatment scores are correlated. We propose improved approximations of the distribution of percent change from baseline using monotone nonlinear transformations. Monte-Carlo simulations are conducted to investigate and compare these adjustment methods. These methods are illustrated on multi-center trails.