Random Delayed Effects and the Power of the Unweighted Log-Rank Test in Survival Trials

• Activity Number: 345
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jonathan Siegel*+
• Companies: Bristol-Myers Squibb
• Keywords: Survival Analysis ; Non-Proportional Hazards ; Delayed Effect ; Random Delay ; Log-Rank Test

Abstract

Clinical survival trials traditionally assume piece-wise constant hazards with fixed change points. Novel classes of immunotherapy cancer treatments have biological mechanisms of action suggesting a variable delayed effect. This paper focuses on the effects of random delay on the power of the unweighted log-rank test. It proposes a random time-to-repair approach, analogous to system-component failure bivariate interaction mixture models, in which drug effect delay is modeled as a Poisson signal whose onset reduces the hazard. The model is shown to be more powerful for finite observation periods than a two-compartment fixed delayed-effect model with the same true mean delay. The effects of random delay on power vis-a-vis constant hazards, fixed delay, and Weibull models are simulated. Problems of delayed effects in survival trial design and in regulatory use are discussed.