Application of Statistical-Decision Theory to Strategic Trial Design

• Activity Number: 220
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): A. Lawrence Gould*+ and Julie G. Ma and Michael Nessly
• Companies: Merck & Co., Inc. and Merck & Co., Inc. and Merck & Co., Inc.
• Keywords: group sequential ; Bayesian decision

Abstract

A proof-of-concept trial will evaluate the potential of a new drug with a novel mechanism of action and potentially novel clinical effect. Interim analyses will be done to decide whether to terminate the trial early for futility, continue the current trial to the next scheduled evaluation and then decide on how to proceed, or design and implement further trials on an at-risk basis and continue the current trial to obtain pivotal trial design information. We describe the application of decision science principles to select among specific decision rule alternatives, evaluate their statistical properties, and choose among them on the basis of anticipated payoffs under assumptions of costs and benefits and various States of Nature. Evaluating the consequences of differing decision rules (actions) on payoffs under reasonable assumptions of States of Nature and assumed cost/benefit structures provides guidance about how aggressive one should be about go/no go decision rules. Assumptions about costs turn out to have substantial effects on the decision rule payoffs. Moderately aggressive go rules and conservative no go rules tend to produce nearly optimal payoffs for the cost structure.