Bayesian Model Averaging Continual Reassessment Method for Bivariate Binary Efficacy and Toxicity Outcomes in Phase I Oncology Trials

Abstract

Many dose-finding approaches that could evaluate bivariate binary efficacy and toxicity outcomes have been proposed in recent years. In such designs, the operating characteristics with finite sample size can be greatly affected by the assumed dose–toxicity and/or dose–efficacy relationship. However, we do not have much information about a new agent we investigated at the planning stage of Phase I trials and so always face to the risk of misspecifying the true dose–toxicity and/or dose–efficacy relationship by arbitrarily and subjectively choosing skeletons. In this article, we proposed the Bayesian model averaging bivariate continual reassessment method to cope with above risk.

Key Words:

• Bayesian model averaging
• Continual reassessment method
• Dose-finding
• Oncology
• Phase I

ACKNOWLEDGMENT

The authors thank the associate editor and two referees for very helpful suggestions that substantially improved the article.

Notes

Note. (p _ej , p _tj ) means efficacy and toxicity probabilities in percentage, respectively, for jth dose level in each working model (WM).