Repeated significance tests on accumulating survival data.

The aim of the study was to compare the properties of two well-known group sequential methods and to demonstrate the effect of performing interim analyses on accumulating survival data without making appropriate adjustments of the nominal significance level. The properties of a group sequential method with fixed nominal significance level (Pocock stopping boundaries) and a method with increasing nominal level with each interim analysis (O'Brien-Fleming boundaries) were compared by stochastic simulation. Simulation experiments with survival times sampled from a breast cancer trial and from exponential distributions were performed. The true overall significance level with unplanned interim analyses increased from 5% to 14% when a maximum of five tests were performed. Both group sequential methods maintained the desired overall significance level. The O'Brien-Fleming method had higher power than Pocock's method. It also reduced the risk of early stopping based on immature data and should usually be preferred.

RCT Entities:   Population Interventions Outcomes