Stratified Two-Sample Tests on the Change Scores When the Outcomes Are Correlated Between Baseline and Post-Treatment: A Simulation Study Using Bivariate Data

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ching-Ray Yu and Martin Ove Carlsson*+ and Kelly H. Zou
• Companies: Pfizer Inc. and Pfizer Inc. and Pfizer Inc.
• Keywords: change score ; two-sample tests ; van Elteren's test ; receiver operating characteristic analysis ; Monte-Carlo ; bivariate data

Abstract

Multi-center randomized placebo control trials typically involve measuring a continuous outcome at baseline and post-treatment in the presence of strata. When the baseline and post-treatment scores are correlated, the change score may be computed. With the presence of strata, adjustment methods using different weights across strata may be applied. We investigate and compare the performances of nonparametric and parametric two-sample tests using change from baseline between the active medication and placebo control groups, after adjustment for strata. Nonparametrically, the van Elteren test using ranks of the data are conducted, using either local best or design free weights. Parametrically by assuming a normal distribution of the change scores, the McClish inverse-variance weight is applied to compute the area under the curve based on receiver operating characteristic analysis. Monte-Carlo simulations are conducted to compare and examine these tests using different weighting schemes by simulated bivariate normally or non-normally distributed data between baseline and post-treatment. The effect due to the correlation between the measurements at the two time points is evaluated