A hybrid approach to sample size re-estimation in cluster randomized trials with continuous outcomes

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Statistical Inference

Background:

• Cluster randomized trials (CRTs) are susceptible to misspecification of the intra-cluster correlation (ICC) at the design stage.
• Sample size re-estimation (SSRE) aims to adjust trial sizes based on interim data, but traditional methods can exhibit high variability.
• Accurate estimation of the ICC is crucial for efficient and reliable SSRE in CRTs.

Purpose of the Study:

• To present a novel hybrid (Bayesian-frequentist) approach for SSRE in CRTs with continuous outcomes.
• To evaluate the performance of the hybrid approach in controlling type I error and reducing sample size variability compared to a frequentist approach.
• To investigate the impact of prior information informativeness on the hybrid SSRE method.

Main Methods:

• Development of a hybrid framework incorporating a Truncated Normal prior for the ICC.
• Interim analysis to update the ICC prior using study data for expected power control.
• Comparison of the hybrid approach with a standard frequentist SSRE approach using simulations or theoretical analysis.

Main Results:

• Both hybrid and frequentist approaches mitigate ICC misspecification issues and control type I error rates.
• The hybrid approach significantly reduces the variability of the re-estimated sample size compared to the frequentist approach, particularly with informative priors.
• Misspecification of a highly informative prior in the hybrid approach can lead to substantial power loss.

Conclusions:

• A hybrid SSRE approach offers advantages for CRTs, especially when moderate or low-quality prior information on the ICC is available.
• The hybrid method provides a more stable sample size re-estimation process than purely frequentist methods.
• The utility of the hybrid approach is greatest when prior evidence for the ICC is not definitive, balancing stability and accuracy.