A Randomization Method to Match Multivariate Profiles of Treatment Groups

• Activity Number: 35
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Anthiyur Kannappan*+
• Companies: Cytel Inc.
• Keywords: Minimization ; multivariate profile matching ; randomization method ; frequency matching

Abstract

Minimization, in its different forms, refers to a method for treatment allocation in clinical trials that minimizes imbalances among treatment groups with respect to prognostic variables of the subjects in the groups. It can also be described as a frequency matching method, as it tries to allocate a new subject to a group in such a way the imbalances in the frequencies of different variables among the groups is minimized. In the multivariate situation, the method essentially matches the marginal univariate frequency distributions of individual variables. This gives rise to a shortcoming in the process. This method, while it matches marginal univariate profiles, it does not ensure the matching of the correlation structures in the different groups. In other words, the method does not guarantee matching of the multivariate profile of subjects in different groups. A new method is suggested, which will attempt to overcome this shortcoming by matching multivariate frequency distributions rather than univariate marginal frequency distributions. The theoretical justification of the new method and results from simulation studies are presented in the paper.