Keywords: sample size re-estimation, conditional variance, mixture model, crossover studies
The blinded sample size re-estimation is widely used in clinical studies. This adaptation has a negligible effect on the Type I error rate. Researchers have estimated the common within-group variance for a continuous variable based on the sample variance (Gould & Shih, 1992) or the expectation of the sample variance (Kieser & Friede, 2003). This paper proposes the conditional variance approach based on the mixture model to estimate the common within-group variance. This study shows that the common within-group variance can be expressed in term of the pooled variance without unblinding the treatments and their treatment difference. While maintaining the Type I error rate, the estimation of the common within-group variance using this new approach is similar to the results presented by Gould & Shih (1992), and Kieser & Friede (2003). Simulations were conducted to support these results. This mixture model approach is also used for crossover studies and extended to studies with multiple treatments. The use of this method in clinical studies will be discussed.