SEARCH

SEARCH BY CITATION

Keywords:

  • biomarker;
  • treatment-effect modifier;
  • interaction;
  • survival;
  • Weibull;
  • permutation test

Abstract

The recent revolution in genomics and the advent of targeted therapies have increased interest in biomarker-defined subgroups of patients who respond to therapy or exhibit specific toxicities. Such biomarker-defined subgroups are also being investigated for non-targeted therapies (e.g. chemotherapy and statins). However, even when the targeting pathway has been identified, a broadly available test to identify the appropriate subgroup will rarely exist prior to the launch of the pivotal phase III trial.

Our aim in this paper is to provide guidance for the analysis of a phase III clinical trial with a survival endpoint, in order to ascertain whether a therapy is more effective in the biomarker-positive patients as compared with biomarker-negative patients, when the trial is conducted on the entire population and when there are multiple candidate biomarkers. We studied treatment-by-biomarker interactions in a Weibull regression model. Different permutation procedures, using single-biomarker statistics and novel composite statistics, are proposed in order to control the family-wise error rate accounting for dependence structures among the biomarkers. A simulation study was performed to compare the operational characteristics of the permutation tests under different scenarios. The tests were applied to a phase III trial of adjuvant chemotherapy in early breast cancer, for which 10 biomarkers were measured in tumor samples from 798 patients.

These permutation tests can be applied to retrospective biomarker studies and to prospective phase III trials of new drugs for which a few clues are known about the targeting pathway at the start of the trial. Copyright © 2011 John Wiley & Sons, Ltd.