Sample size and robust marginal methods for cluster-randomized trials with censored event times

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Cluster-randomized trials (CRTs) often use marginal models with robust variance estimates to handle correlated responses within clusters.
• Accurate sample size determination is crucial for the validity and power of CRTs.

Purpose of the Study:

• To develop and validate sample size criteria for CRTs using semiparametric Cox regression models.
• To provide a framework for sample size calculation that accounts for within-cluster dependence in event times.

Main Methods:

• Utilized copula models to derive the asymptotic variance of marginal Cox regression estimators.
• Developed sample size formulas based on these derivations for right-censored event times.
• Conducted simulation studies to assess the performance of the sample size formula under various conditions (cluster size, censoring, dependence).

Main Results:

• The proposed sample size formula is valid in finite samples across different scenarios.
• Demonstrated the impact of copula misspecification and within-cluster dependence in censoring times on power and efficiency.
• Extended sample size considerations to interval-censored data.

Conclusions:

• The developed sample size criteria provide a robust approach for designing CRTs with correlated event times.
• The findings offer practical guidance for researchers to ensure adequate power in studies with complex data structures.
• The study addresses important design issues for both right-censored and interval-censored data in CRTs.