A Dose-Finding Design for Combination Therapy with Bayesian Bivariate Ordinal Probit Model

• Activity Number: 414
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Rui Qin*+ and Jianchang Lin and Sumithra J. Mandrekar and Daniel J. Sargent
• Companies: Mayo Clinic and Florida State University and Mayo Clinic and Mayo Clinic
• Keywords: dose finding ; bivariate ordinal probit model ; continual reassessment method ; combination therapy

Abstract

Combination therapies are becoming increasingly popular in oncology for potential synergistic efficacy. While previous knowledge about each individual drug may suggest a proper dosing range for combination therapy, a separate dose-finding trial is generally required to determine the optimal dose combination. We propose a dose-finding design with Bayesian bivariate ordinal probit model incorporating toxicity and efficacy as ordinal variables. The contribution of each agent in the combination therapy is modeled in the linear predictors of ordinal outcomes. Priors are used to capture historical information of single agents and the Monte Carlo Markov Chain approach is used for parameter estimation and guiding dose-escalation. Simulation studies are conducted to understand and assess the operating characteristics of this design under multiple scenarios. Considering 50% as a benchmark for the recommendation rate of the optimal dose region (within 15% of the pre-specified optimal dose level), our design performs well under most scenarios considered (most between 50.2%-65.8%, except two scenarios with 32.4%, 43.2%).