Self-designing group sequential trials to minimize expected sample size

Sample size estimation is an essential step in the design of clinical trials. For normal data, the calculation is based on estimates of the mean and variance of the treatment effect. These estimates are often unreliable because they are based on smaller studies, possibly looking at a different population than the one of interest. If the true treatment effect is overestimated and/or the variance is underestimated, a trial may be underpowered to detect a clinically meaningful treatment effect. If the true treatment effect is underestimated and/or the variance is overestimated, the trial would be unnecessarily large. In addition to ethical concerns for the patients, both time and money would be wasted in this situation. One solution to the uncertainty in the sample size estimate is to be conservative and estimate the maximum sample size needed to detect the minimal clinically meaningful treatment effect. Using an appropriate group sequential trial design, the trial can be stopped early when a worthy treatment effect is detected. In practice however, the sample size calculated for a study is usually not the maximum sample size needed to detect the minimal clinically meaningful treatment effect. With respect to funding, a study requiring a smaller sample size is usually preferred over one requiring a larger sample size. Furthermore, the expected treatment effect used for sample size estimation is usually an overestimate because the researchers are overly optimistic. Finally, numerous compounds are studied and only the most promis-