Exponential Failure-Time Mixture Model Approach for Validating KRAS as a Predictive Biomarker for Panitumumab Monotherapy in the Treatment of Metastatic Colorectal Cancer

Abstract

Progression-free survival (PFS) results from the Amgen Study 20020408 were presented and discussed in the December 2008 Oncologic Drugs Advisory Committee. Using the retrospective “sorted” subgroup analyses it was claimed that Kirsten rat Sarcoma (KRAS) status should be considered when selecting metastatic colorectal cancer (mCRC) patients as candidates for panitumumab monotherapy. A similar conclusion was reached by Amado et al. (2008). Results based on sorted subgroup analyses may not be conclusive. We propose an exponential failure-time mixture model that helps to both validate a marker and demonstrate drug efficacy. This analysis includes all randomized KRAS-evaluable subjects and considers the KRAS status as a random outcome. The model also facilitates a comparison of PFS between the wild-type and mutant KRAS subgroups of subjects on the panitumumab arm. We conclude that KRAS is a predictive biomarker for the panitumumab therapy as measured by PFS in the treatment of mCRC patients. The objective of this article is to provide details and to discuss the scope and limitation of the proposed methodology.

Keywords: :

• Accelerated failure-time model
• Identifiability
• Intent-to-treat analysis
• Interaction
• Likelihood ratio test