Abstract:
|
Variable length time to event clinical trials are based on observing predetermined total numbers of events. The primary events of interest are e.g., death, relapse, adverse drug reaction, or new disease development. We propose strategies for determining the timing of the final events in blinded settings. Parametric naïve models for determining the timing of final events: (i) ignore issues involving staggered entry; (ii) do not take account of the evolution of hazard rates; and (iii) fail to provide model flexibility. Nonparametric Bayesian survival models and the Bayesian bootstrap address these concerns and are shown to be highly accurate predictors. Methodologies are tested using a randomized, double-blind, placebo-controlled trial Schizoaffective trial where the primary endpoint was time to relapse.
|