Stopping at Nothing? Some Dilemmas of Data Monitoring in Clinical Trials

1. 

   View larger version:

   • In this window
   • In a new window

   Figure. Distribution of observed effects in trials with and without stopping rules.

   The trials were designed to have 90% power to detect a 10% mortality benefit (for example, 50% vs. 40%). Each panel corresponds to a different underlying true difference: no difference (top), 10% difference (middle), and 20% difference (bottom). The distribution of results is shown for trials of 2 designs: 1 using a 4-look O'Brien–Fleming stopping rule (“stopping”) and 1 using a fixed sample size (“no stopping”). Median effect size and 2.5% and 97.5% percentiles of each estimate are reported in parentheses. The mean sample size is reported for the “stopping” trial only: n = 1040 for the fixed sample size design.