Sources of Bias in Adaptive Seamless Phase II/III Designs

• Activity Number: 36
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jing Zhang*+ and Kun He
• Companies: FDA and FDA
• Keywords: seamless phase II/III trial ; adpative design ; bias ; time-to-event ; correlated endpoints

Abstract

There is an increasing interest in adaptive, seamless phase II/III designs in oncology clinical trials from both a time and cost savings perspective. A typical objective of such a design is to select one treatment arm from several competing arms in the phase II portion (stage 1) to carry forward into the confirmatory phase III portion (stage 2), without interrupting study accrual. Only patients accrued in stage 1 on the selected phase III treatment arm and the control arm are included in the phase III primary analysis. Criteria for treatment arm selection in stage 1 are usually based hierarchically on safety, pharmacokinetic (PK), and efficacy (e.g. overall survival [OS]) endpoints. Such a trial design possesses several inherent sources of bias, which may result in type I error inflation if not appropriately handled. We discuss these sources of bias when using time-to-event endpoints, with particular focus on the potential type I error inflation from making design adaptations using information correlated with the outcome of patients still at risk at the end of stage 1 (e.g. PK and OS endpoints).