Who Needs Bayesian Phase I Trials?

• Activity Number: 378
• Type: Invited
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: International Indian Statistical Association
• Author(s): Rick Chappell*+
• Companies: University of Wisconsin-Madison
• Keywords: clinical trials ; phase I ; dose-escalation ; maximum tolerated dose ; Bayesian designs

Abstract

This talk addresses the conduct of Phase I trials in serious illnesses for which the goal is to find the maximum tolerable dose of a other treatment or combination of treatments. Tolerability is framed with respect to the frequency of explicitly defined toxicities. That the so-called "traditional" or escalation-only 3+3 design is flawed has been noted by Storer, who proposed more useful algorithmic designs in 1989, O'Quigley, et al., who published a Bayesian alternative in 1990, and many authors since.

I will consider three aspects of phase I trials in which the Bayesian paradigm may be applied: dose selection, escalation algorithm, and final analysis. I see no choice but to be Bayesian in the first and no harm in being Bayesian for the analysis. Bayesian escalation methods are flexible but at the cost of incurring ethical and logistic difficulties which limit their application.