Keywords: Immuno oncology, non-proportional hazards, group sequential with delayed separation
Immunotherapy trials with time to event endpoints are characterized by delayed separation of the survival curves. In this setting the logrank test loses power because the hazard ratio under the alternative hypothesis is not constant over time. At a recent public workshop organized by the Duke-Margolis Center and the FDA (February 5, 2018), case studies were presented and new methodology was proposed, by a consortium of industry statisticians. Simulation results from a new test utilizing the maximum of several Harrington-Fleming statistics over time showed promise of improving the power compared to the logrank test. The new approach was, however, confined to a single-look design. We have extended this approach to the group sequential setting using computational tools developed by us for multi-arm multi-stage (MAMS) designs. The benefits of this extension include increased power and shorter study duration.