Abstract:
|
Standard designs for testing the mean, such as mean treatment effect or mean treatment difference in a clinical trial, do not allow for an increase in the sample size based on the observed mean from an interim analysis. This is due to inflation of the overall Type I error rate. Fisher's self-designing method (1998, Statistics in Medicine 17) permits sample size extension based on the observed mean while preserving the overall type I error rate. The method uses the variance-spending statistic, which allows the sample size and the proportion of the test statistic's variance to be determined sequentially for each stage of a trial based on ALL prior observed data. Thus, if the treatment effect is less than expected, the sample size for the next stage of a group sequential trial can be chosen to increase the power of the trial to detect the true treatment effect. We present group sequential designs for clinical trials, based on Fisher's self-designing method, satisfying an optimality criterion. Examples will be shown for designs satisfying the criteria of minimal expected cost and minimal expected sample size. Optimal designs are determined from Bayes sequential decision theory.
|