All Times EDT
Keywords: SSR, Non-Proportional Hazards, Bayesian Predictive Power, Conditional Power
In time-to-event studies with an adaptive sample size re-estimation (SSR) design, the interim conditional power (CP) is used to make the decision if the targeted number of events should be increased in order to regain the originally planned power in case the true hazard-ratio is larger than the design assumption but still clinically relevant. The analytical computation of the interim conditional power for time-to-event studies relies on the assumption of proportional hazards (PH) which may not be valid in many applications.
We consider two scenarios where the PH assumption may not be valid: (i) Delayed treatment effect where the two survival curves have a late separation; and (ii) Early short-term benefit only, where the two survival curves may separate out early but then tend to converge. For these situations we propose using the Bayesian predictive power (PP) instead of the CP in order to decide on the adaptive increase in number of targeted events. For studies with a planned minimum follow-up time, the use of the PP also allows for achieving the increased target number of events by choosing a combination of increases in sample size and the minimum follow-up such that the overall study duration can be minimised. We present a numerical study to highlight the advantages and establish the operating characteristics of the proposed SSR design based on the Bayesian predictive power.