Stopping randomized trials early for futility: error in estimating the treatment effect

Abstract

In the past few decades, interim analyses in long-term clinical trials have become increasingly popular. If a new treatment or intervention is truly ineffective or have lower effect than expected, it is better that the trial be stopped early for futility so as not to expose further patients to clinically inefficient or harmful treatment, and thus save time, resources and money by avoiding recruitment of more patients. Conditional power is one of the typical methods of futility monitoring. In this paper, I develop expressions for the error in estimating the treatment effect in studies stopped early for futility, the over-estimations in completed studies and the overall bias for a study with a stopping rule for futility. I also discuss the probability of stopping and the weight of such stopped studies in a meta-analysis. Then I evaluate the expressions using commonly used parameters. The numerical results show that the error in estimating the treatment effect in studies stopped early for futility is usually substantial but fortunately the overall bias in single study is quite small. Corresponding biases are a little bit greater if two interim analyses are planned instead of only one interim analysis, thus it is expected that the under-estimation of treatment effect in truncated studies and the overall bias of a single study would increase with the number of planned interim analyses.