A review of statistical methods on testing time-to-event data
- Biometrics & Biostatistics International Journal
-
Tu Xu,1 Danting Zhu2
Agios Pharmaceuticals, Cambridge, USA
Abstract
The proportional hazard (PH) is commonly assumed for claiming efficacy and planning sample size in randomized clinical trials with time-to-event (TTE) type of endpoints. It is well known that the log-rank test is the most powerful testing method when the PH assumption holds. In recent years, with the advancement of immuno-oncology therapies, the non-PH scenarios, such as the delayed treatment effect and the diminished treatment effect, are frequently observed. A variety of alternative methods have been proposed for testing the time-to-event data while there is no uniformly most powerful method under the non-PH setting. In this paper, six popularly used methods for testing the TTE data are reviewed followed by a numerical comparison.
Keywords
time-to-event, proportional hazard, log-rank test, restricted mean survival time