Sample size determination for GEE analyses of stepped wedge cluster randomized trials

Area of Science:

• Biostatistics
• Health Services Research
• Clinical Trials

Background:

• Stepped wedge cluster randomized trials are increasingly used in health services research.
• These trials involve clusters of individuals switching from control to intervention over time.
• Longitudinal follow-up of closed cohorts requires careful sample size considerations.

Purpose of the Study:

• To propose sample size procedures for stepped wedge cluster randomized trials with longitudinal follow-up.
• To address the complexities of intraclass correlations in these designs.
• To provide methods for both continuous and binary responses.

Main Methods:

• Utilizing generalized estimating equations (GEE) with a block exchangeable within-cluster correlation structure.
• Accounting for three distinct intraclass correlations: within-period, inter-period, and within-individual.
• Developing bias-corrected estimating equations and a sandwich variance estimator.

Main Results:

• Intraclass correlations impact power through two eigenvalues of the correlation matrix for continuous responses.
• Analytical power calculations closely align with simulated power.
• Effective sample size calculations are demonstrated with as few as eight clusters.

Conclusions:

• The proposed GEE-based methods provide accurate sample size calculations for stepped wedge trials.
• These methods are applicable to both continuous and binary outcomes.
• The findings support the use of these procedures in health services research for efficient trial design.