Effects of simulated sample sizes on the mortality effect estimatesinthreerandomised intensive care unittrials.

BACKGROUND: Randomised clinical trials (RCTs) are occasionallystopped prematurely before reaching theirplanned sample sizes. It has been suggested that early stopped RCTs are associated with under- and overestimation of the effect estimates. We simulated the effect of hypothetical premature stopping of three large RCTs done in the intensive care unit (ICU) setting. METHODS: In thispost-hoc study, we simulated the impact of stopping trials early by calculating mortality effect estimates continuously after the inclusion of each individual patient in three large RCTs, i.e. the 6S trial on hydroxyethyl starch versus Ringer's acetate in sepsis in ICU, the TRISS trial on lower versus higher hemoglobin threshold for transfusion in septic shock in ICU and the SUP-ICU trial on pantoprazole in patients at risk for gastrointestinal bleeding in the ICU. RESULTS: The three trialsincluded a total of 5087 patients; 798 from the 6S trial, 998 from the TRISS trial and 3291 patients from the SUP-ICU trial. Thepremature mortality effect estimates showed considerable fluctuationsuntil at least 20-30% of the sample size was included.The premature estimates became stable after inclusion of 205patients (26% of the final sample size) in the 6S trial, 133patients(13%) in the TRISS trialand1926patients(59%) in the SUP-ICU trial. CONCLUSIONS: In this post-hoc study of three international RCTswithin intensive care, we found that the simulated interim mortality effect estimatesshowed considerable fluctuations until at least 20-30% of the sample size was included, but remained instable until the final sample sizes had been included. Thus, this study illustrates the necessity for cautious interpretations of prematurely stopped trials.