A clinical trial design applying Bayesian adaptive randomization for targeted therapy development in lung cancer: A step toward personalized medicine

Abstract

7697 Background: Multiple axes of signaling pathways are associated with lung carcinogenesis. These signaling axes are different in pts (pts) and their cancers. Utilization of molecularly targeted agents may inhibit these specific aberrant pathways and lead to clinical efficacy. Biomarkers expressions can be used as indicators for the aberrant signaling to identify effective targeted therapy. Methods: The program, “Biomarker-integrated Approaches of Targeted Therapy of Lung Cancer Elimination (BATTLE),” consists of an umbrella screening trial and 4 parallel phase II targeted therapies trials (with erlotinib, sorafenib, vandetanib, and the combination of erlotinib and bexarotene) in advanced non-small cell lung cancer pts with prior chemotherapy. All pts will have biopsy samples taken for biomarker profile assessment prior to the randomization. A ‘surrogate response‘ to treatment is defined as progression free at 8 weeks after randomization. The Bayesian ordinal probit model is used to characterize the response rate. Pts with certain biomarker profile will be adaptively randomized (AR) to one of the 4 treatment arms with the randomization rate based on the updated response rate based on accumulated data in the trial. For each biomarker profile, better performing arms will have higher randomization rates and vise versa. Early stopping rules are set so that low-performing arms may be suspended for new patient entry. Results: Based on extensive simulation studies, the proposed design with a total of 200 pts has desirable operating characteristics to: (1) identify effective agents with high probability; (2) suspend ineffective agents; and (3) treat more pts with effective agents according to their biomarker profiles. The Bayesian design incorporates prior data and findings from the current pts to form better estimates of the treatment efficacy for pts with different biomarker profiles. The design continues to “learn” and improve the estimates as the trial moves along. Conclusion: The Bayesian AR design is a smart and ethical design and ideally suitable for the development of targeted therapy. It may help in identifying effective agents based on pts’ tumor biomarker profile and thus treat more pts with effective therapies. No significant financial relationships to disclose.