Interim sample size re-estimation (SSR) often affects the Type I and II error rates. We propose and investigate a method based on resampling the whole study design at the interim analysis. This resampling starts with SSR at the observed interim analysis values of nuisance parameters and finishes with the decision to accept or reject the null hypothesis. The proposed approach finds a new critical value and an updated sample size. As shown in a Monte-Carlo simulation study this resampling method shows superior performance for logistic regression when compared with the naive SSR. Another set of simulation studies shows comparable performance of the resampling and several previously published procedures for a 2-sample t-test with random allocation. An illustrative example highlights the benefits of our approach for logistic regression analysis.
August 19, 2016