Issues to Consider in Selecting a Response-Adaptive Design for Dose-Finding Experiments

• Activity Number: 598
• Type: Invited
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Nancy Flournoy*+
• Companies: University of Missouri
• Keywords: sequential methods ; experimental design ; clincial trials ; phase I/II ; converngence ; efficiency

Abstract

Separately for design and analysis, one must choose between parametric and nonparametric, frequentist and Bayesian methods. Good choices depend on factors including the total sample size available and the experimental objectives. Prospects of significant toxicity bring multiple, competing objectives and dictate caution in the rules for transiting doses and/or suggest penalties be introduced. There is great heuristic appeal to using all the information obtained to date to select the next dose, but there are performance trade-offs between allocating subjects using long term memory versus short term memory procedures. Two stage designs at times are competitive with fully adaptive procedures. Procedures beckon for conditional power calculations, sample sizes recalculation, and dropping (adding) treatments. We discuss these choices and reflect on how flexible an experiment should be.