Title: Model-assisted adaptive designs in oncology
Authors: J Jack Lee - University of Texas MD Anderson Cancer Center (United States) [presenting]
Abstract: Oncology drug development is a long, arduous, and expensive process. Recent report shows that the cost of developing a drug is at \textdollar2.6 billion. The overall success rate from Phase I trials to FDA approval was 9.6\% with oncology drug at 5.1\%. Bayesian adaptive designs have been shown to increase efficiency, allow more flexible trial conduct, and treat more patients with more effective treatments. However, model-based Bayesian adaptive designs have not gained much traction largely because computation burdens. Although software tools are available to ease the design and implementation, many such designs are still being viewed as too complicated and too difficult to do. Conversely, algorithm-based methods, such as the 3+3 design for Phase I studies, remain popular in spite of inferior operating characteristics. In viewing of this quandary, several model-assisted designs have been developed to attain superior statistical properties while provide simple rules for trial conduct. Easy-to-use tools for trial design and conduct have also been introduced. These include the Bayesian optimal interval designs for single and combination agents (Phase I) and the Bayesian optimal Phase II design for simple and complex endpoints. The model-assisted design fulfilled the new Keep it Simple and Smart (KISS) principle. This new class of model-assisted design can be easily and widely applied to treat patients better and to improve the success rate of drug development.