T-Statistic-Based Up and Down Design for Dose-Finding Competes Favorably with Bayesian 4-Parameter Logistic Design

• Activity Number: 446
• Type: Invited
• Date/Time: Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Statistics in Biopharmaceutical Research Journal
• Author(s): James A. Bolognese*+ and Nitin R. Patel and Yevgen Tymofyeyef and Inna Perevozskaya and Jeffrey Palmer
• Companies: Cytel, Inc. and Cytel, Inc. and Merck & Co., Inc. and Merck & Co., Inc. and Cytel, Inc.
• Keywords: dose-finding ; adaptive design ; t-statistic design ; Bayesian adaptive design ; 4-parameter logistic model ; up&down design

Abstract

Two adaptive dose-finding designs are studied via simulation: a Bayesian method using variance reduction criterion with a 4-parameter logistic dose-response model by Scott Berry; a T-Statistic method (Ivanova, et al 2008). They use accumulating data to optimize observations to estimate dose-response curve features. Performance was comparable for true underlying 4-parameter logistic models (Patel et al 2008), with slight advantages for the Bayesian design. For NON-4-parameter logistic models (linear, umbrella, Emax) we found the 2 methods show useful performance criteria; Bayesian method better if model is closer to S-shaped or linear, and to model dose-response; T-Stat design easier to implement, has better properties at dose-range extremes and at doses with targeted responses. We recommend choice of design be driven by objectives and potential true underlying dose-response curves.