Design and Implementation of Group Sequential Trials for Censoring Robust Hazard Ratio Estimators with Nonlinear Information Growth Under Nonproportional Hazards

• Activity Number: 138
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Jeffrey Lambert* and Adam Boyd and John Kittelson
• Companies: University of Colorado Denver and Array Biopharma and University of Colorado Denver
• Keywords: hazard ; sequential ; trials ; non-proportional ; proportional

Abstract

Under non-proportional hazards, the usual Cox model hazard ratio estimator is a function of the censoring distribution. When monitoring a trial the censoring distribution must change across interim analyses, and therefore under non-proportional hazards the usual Cox estimator is not consistently estimating the same parameter. Several estimators that correct this problem have been described, but unlike the Cox estimator the information growth with these censoring-robust estimators is not linear in the number of events. Incorrect estimates of the information fraction potentially lead to incorrect early rejection boundaries in group sequential trials. This paper describes methods for designing the trial monitoring plan using non-linear information growth from the estimator proposed by Boyd et al. This pre-trial design is based on the best pre-trial projection for the censoring distribution at the interim analyses. We also describe methods, akin to alpha-spending functions, that maintain desired operating characteristics when implementing the trial monitoring plan using the observed censoring distributions and non-linear pattern of information growth.