Stopping clinical trials early for benefit: impact on estimation

BACKGROUND: It has been suggested in the literature that the well-known bias of treatment-effect estimators due to the possibility of early stopping for positive results is a major concern with interim monitoring. PURPOSE: To discuss approaches one might use to compare the inflation of the treatment-effect estimator when the trial is stopped early for positive results with the inflation that would be seen in a comparable set of positive trials that used fixed sample sizes with no interim monitoring, and to quantify the relative inflation of monitored trials relative to that of the corresponding subset of positive fixed sample-size trials. METHODS: Via simulation for some O'Brien-Fleming and Haybittle-Peto monitoring boundaries, the inflation of the treatment-effect estimator when the trial crossed an interim-monitoring boundary for positive results is compared with the preferred approach to estimate the inflation from a comparable set of positive fixed sample-size trials. RESULTS: Although the inflation of the treatment-effect estimator when a trial is stopped early can be considerable, only at very early interim analyses (<or=25% of information) is this inflation much larger than the inflation that would be seen for an appropriate subset of similar positive fixed sample-size trials. The treatment-effect inflation from stopping at second or later interim analyses that are not so early is relatively small and similar to that seen in the corresponding subset of fixed sample-size trials. LIMITATIONS: The results apply to adequately powered trials with well-designed prospectively specified interim-monitoring plans. CONCLUSIONS: For trials with a well-designed interim-monitoring plan, stopping at 50% or greater information has a negligible impact on estimation. Except for very early interim analyses (<or=25% of the information), concerns about the inflation of the treatment effect should be minor.