|
Abstract:
|
The log-rank test and the Cox proportional hazards model are commonly used in design and analysis in modern oncology trials with time-to-event endpoints. When non-proportionality, i.e. the hazard ratio is not constant over the entire follow-up period, is observed due to the delayed treatment effect, the proportional hazard assumption is violated, which may lead to loss of study power. Therefore, design adaptations such as follow-up time extension, sample size modification, etc., may be applied to mitigate the power loss. In this paper, we propose an adaptation method when non-proportional hazards are observed. This method only takes the information up to the interim look to model the survival outcomes using Bayesian piecewise exponential model. Posterior samples are generated and probability of success conditioning on the observed information can be achieved to drive the adaptive decisions. The power of this proposed method is evaluated by extensive simulations and compared to existing methods to show that the type I error is controlled at the pre-specified level ?.
|
