All Times EDT
Keywords: Clinical trial, Cancer immunotherapy, Dichotomized response, Immuno-oncology trial, Proportional hazards assumption, Non-proportional hazards pattern, Sample size and power calculation.
A typical challenge facing the design and analysis of immmuno-oncology (IO) trials is the prevalence of non-proportional hazards (NPH) patterns manifested in Kaplan-Meier curves under time-to-event endpoints. The NPH patterns would violate the proportional hazards assumption and yet, conventional design and analysis strategies often ignore such violation, resulting in underpowered or even falsely negative IO studies. In this paper, we show, both empirically and analytically, that treating nonresponders in IO studies of inadequate size would give rise to a variety of NPH patterns and then present a novel strategy, P%-Responder Information Embedded (PRIME), to incorporate the dichotomized response incurred from treating nonresponders into the design and analysis strategies. Empirical studies demonstrate that the proposed strategy can achieve desirable power whereas the conventional alternative leads to a severe power loss. The PRIME strategy allows us to quantify the impact of treating nonresponders on study efficiency and thereby enables a proper design of IO trials with adequate power. More importantly, it pinpoints a solution to enhance the study efficiency and alleviates the occurrence of NPH patterns by enrolling more prospective responders. An R package (Immunotherapy.Design) is developed for implementation.