Sample size re-estimation (SSR), a popular adaptive approach to correct for incorrectly specified sample sizes based on updated knowledge. Rather than a one-stage fixed study design, SSR re-assesses the sample size at interim analysis in the mid-course of the trial. This idea of adaptive sample size increase allows us to start with a provisional sample size calculation. After recruiting a portion of the planned sample, we can choose to increase the sample size provided the interim treatment effect estimate is a little worse than expected. In this case study, its intended to assess the results of blinded and unblinded SSR on the same data.
Compare the results of blinded and unblinded SSR when there is uncertainty regarding the actual effect size.
Interim analysis data for Phase 2 Placebo controlled study to compare the change in total PANSS Score at week 4 is used. Mean and SD of the two treatment groups from previous study was used initially for sample size determination. The results were based on a small number of subjects and the duration of observation was also small compared to the study. About 60 subjects completed week 4 assessment when the interim analysis was performed.
Blinded and Unblinded SSR applied to Interim data from a study, and the results were compared.
Initially, the sample size was calculated using the classical method. Then, blinded and unblinded SSR following a comparison of the results are performed with the interim data.
Sample-size adjustment based on the effect-size ratio (Unblinded)
The effect size was re-estimated using the interim data and was used to re-estimate sample size using the classical method.
Blinded sample size re-estimation
With the treatment assignments blinded, pooled variance was re-estimated and was used to re-estimate the sample size.
Unblinded SSR was found to be more effective for sample size re-estimation when there is uncertainty regarding the effect size.
At the interim analysis there was a significant increase in the effect size along with the increase in variance as compared to the initial assumption, which resulted in the decrease of the sample size when unblinded SSR was used. On the other hand, the Blinded SSR resulted in increased sample size which was not practically feasible since only the increase in variance was used in sample size estimation.
Unblinded SSR resulted in improved power, while Blinded SSR was causing overpowering as sample size increased significantly than what was required.Though Blinded adaptations are recommended in sample size re estimation, it can be prone to bias and lead to overestimation of sample size when there is uncertainty regarding the effect size. When we have limited knowledge about the true treatment effects and decision at interim analysis is critical, it’s better to use Unblinded SSR methods as more information observed is used to determine the effective sample size.