Efficient Analyses of Phase II Dose-Range Trials to Optimize Phase III Trial Design

• Activity Number: 386
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Committee on Applied Statisticians
• Author(s): Devan Mehrotra*+ and Robin Mogg and Jianchang Lin
• Companies: Merck Research Laboratories and Merck Research Laboratories and Millennium: The Takeda Oncology Company
• Keywords: dose-response ; Emax model ; modeling and simulation ; non-normality ; posterior distribution ; probability of success

Abstract

In a typical phase II clinical trial, patients are randomized to receive either placebo or one of several (typically 3-4) dose levels of an experimental treatment. A common goal of the phase II data analysis is to identify doses that appear to be generally safe and well tolerated, and, from that subset, select the dose that appears to deliver a desired level of clinical efficacy. If such an "optimal" dose (say D*) can be identified, then the treatment is advanced to a phase III trial to conclusively demonstrate safety and efficacy based on pre-specified criteria. Focusing on the efficacy component, we contrast different utilizations of the phase II data to help assess the phase III probability of success (POS). Specifically, we compare a variety of approaches to quantify the distribution of plausible values of the unknown treatment effect at dose D*, which is a key input for the phase III POS determination. We report simulation results that (i) highlight the risk of a common approach found in the PK/PD literature, and (ii) justify alternate Bayesian and bootstrap-based analyses that leverage the dose-response association via suitable nonlinear models.