Sample size and power calculation for testing treatment effect heterogeneity in cluster randomized crossover designs

Area of Science:

• Biostatistics
• Clinical Trial Design
• Epidemiology

Background:

• Cluster randomized crossover designs offer improved efficiency over parallel-arm designs.
• Existing statistical methods primarily focus on average treatment effects, neglecting subgroup variations.
• Understanding treatment effect heterogeneity across patient subpopulations is increasingly important.

Purpose of the Study:

• To develop sample size formulas for detecting treatment effect heterogeneity in two-treatment, two-period cluster randomized crossover trials.
• To incorporate interaction tests for pre-specified patient subpopulations.
• To address both cross-sectional and closed-cohort sampling schemes.

Main Methods:

• Derivation of sample size formulas using linear mixed models.
• Assumption of a patterned correlation structure for covariates and outcomes.
• Inclusion of methods to handle unequal cluster sizes.

Main Results:

• New formulas for sample size calculation for interaction tests in cluster randomized crossover designs.
• Analytical assessment of the impact of unequal cluster sizes on statistical power.
• Validation of methods through simulations and real-world trial applications.

Conclusions:

• The proposed methods provide a robust framework for designing cluster randomized crossover trials focused on treatment effect heterogeneity.
• Accurate sample size determination is crucial for detecting differential treatment effects in subpopulations.
• The formulas facilitate a better understanding of how unequal cluster sizes influence study power.