Bayesian Two-Parameter Logistic Regression Model--Based Method for Single Dose-Finding Trial

• Activity Number: 636
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Olivia Liao*+ and Vlad Dragalin
• Companies: Onyx Pharmaceuticals and Aptiv Solutions
• Keywords: Bayesian logistic regression model ; continual reassessment method ; Phase I trial design

Abstract

Model-based approach for Phase I trials has gained popularity since the continual reassessment method (CRM) proposed by O'Quigley et al (1990), but concerns about potentially aggressive dose escalation, decreased flexibility and computational difficulties of the modeling paradigm still prevail. In this talk, we address and clarify some of the myths. In particular, we present a case study using the Bayesian two parameter logistic regression model developed by Neuenschwander et al. (2008) which can incorporate flexible clinical decision making in more complex settings. The advantages of this design over the conventional 3+3, including better identification of the MTD, are demonstrated by the simulation result. The derivation of the prior distribution that can take into account prior clinical knowledge, yet not too informative, is discussed since it is of particularly importance for a Bayesian regimen. We also address some practical considerations in determining the cohort size and trial stopping rule.