Impact of correlation structure on sample size requirements of statistical methods for multiple binary outcomes: A...

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Methods

Background:

• Randomized clinical trials increasingly use complex methods for multiple binary outcomes.
• Outcome correlations impact sample size requirements, but their effect on statistical power and sample size needs further investigation.

Purpose of the Study:

• To compare the power and sample sizes of different statistical methods for analyzing multiple binary outcomes under varying correlation structures.
• To evaluate co-primary endpoints, composite endpoints, and prioritized outcome approaches.

Main Methods:

• Simulations were used to assess statistical power and sample size requirements.
• Methods were evaluated across different correlations, marginal proportions, treatment effects, and numbers of outcomes.
• A case study involving a migraine treatment trial was conducted for sample size analysis.

Main Results:

• Correlations significantly influenced the power and sample size of composite endpoints.
• Co-primary endpoints showed stable power across correlations but declined with opposing treatment effects or >2 components.
• Prioritized outcome approaches generally yielded higher power and smaller sample sizes when correlations were similar between arms.

Conclusions:

• Anticipating and accounting for correlations is vital when choosing statistical methods for multiple binary outcomes.
• Co-primary endpoints are reliable for demonstrating superiority but not for balancing opposing treatment effects.
• Generalized pairwise comparisons provide an effective alternative for prioritized outcomes, often minimizing sample size when correlations are shared.