Bayesian design and analysis of two-arm cluster randomized trials using assurance.

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Cluster randomized controlled trials (cRCTs) are essential for evaluating interventions in group-level settings.
• Accurate sample size determination is critical for the statistical power and efficiency of cRCTs.
• Traditional methods for sample size calculation may be sensitive to parameter mis-specification.

Purpose of the Study:

• To propose and evaluate a Bayesian approach for sample size determination in two-arm superiority cluster randomized controlled trials (cRCTs) with continuous outcomes.
• To compare the Bayesian sample size approach with traditional methods, such as assurance based on a Wald test.
• To provide guidance on selecting prior distributions and Monte Carlo sampling schemes for accurate sample size calculations.

Main Methods:

• Utilizing Bayesian inference with a linear mixed-effects model for trial analysis.
• Developing and evaluating an assurance metric for sample size selection via a two-loop Monte Carlo simulation scheme.
• Assessing the impact of various prior distributions and Monte Carlo sample sizes on assurance and sample size determination.

Main Results:

• The Bayesian approach offers increased robustness of the chosen sample size to parameter mis-specification.
• Potential for reduced sample sizes when prior information suggests a clinically important treatment effect.
• Demonstrated application to sample size calculation for a cRCT in poststroke incontinence.

Conclusions:

• The proposed Bayesian design and analysis framework provides a robust and potentially more efficient method for sample size determination in cRCTs.
• This approach can lead to more reliable sample size estimates, especially when incorporating prior knowledge.
• The Bayesian method is particularly advantageous when prior information indicates a potentially larger treatment effect than the minimal clinically important difference.