Abstract:
|
In a time to event clinical trial, researchers may posit the possibility that survival curves may cross at some time point. Logan et al. (2008), Logan and Mo (2015) considered the analysis of censored time to event data with the objective of detecting the "better" treatment when crossing survival curves are quite plausible. We investigate the alternatives used by Logan et al's composite statistics, and use simulations to assess how their test statistics behave under the setting of non proportional hazards, both with stochastically ordered and with potentially crossing survival. We further investigate the behavior of the alternative test statistics as a function of the censoring distribution. We then examine how this may impact any sequential designs where a DMC may make an interim decision to stop the trial early or adapt trial parameters based on unanticipated differences in survival. We find little advantage to the use of Logan et al's proposed composite statistics over judicious choice among the commonly used test statistics in the sequential survival setting.
|