From a regulatory perspective, it is important that the sample size recalculation is performed such that all persons involved in the study remain blinded. The proposed method is an extension of the work by Shih and Zhao (1997) to continuous endpoints. The treatment means are constructed by the convex combinations of the stratum means and then estimated by using the linear model of the stratum responses. In this article, the properties of the proposed estimators are studied. Simulation experiments are conducted to evaluate the difference between two estimators. The unblind estimators for the population mean and the population variance perform better than those of the blind estimators in terms of bias and mean square errors in the most of cases. Given a particular sample size, the accuracies of the blind means and the blind variances depend on the treatment proportions in each stratum. An example of interim analysis is given in this article to illustrate the use of sample size determination. The proposed sample size calculations are recommended in the interim analyses to meet Committee for Proprietary Medical Products requirement, retaining the blinding.
展开▼