The a priori trial planning assumption of non-proportional hazards is becoming more common with the rise of immuno-oncology. However, standard analytic methods of sample size calculation are not compatible with time-to-event trials designed under this assumption.
Although the log-rank test and Cox analysis are still regularly required for these trials, the Schoenfeld and Freedman formulae for calculating required event numbers both rely upon a constant Hazard Ratio. For Restricted Mean Survival Time and Landmark Analysis, which are better suited to analysing non-proportional hazards, current analytic sample size methods also struggle to accurately account for censoring.
This talk will present novel,accurate, analytic methods of hazard ratio prediction and power calculation for all three testing strategies with minimal distributional requirements for events, censoring and recruitment. A single, fast and flexible implementation in R will also be presented, including validation of results by simulation.