Comparing intraclass correlation coefficient estimators for binary outcomes in sample size calculations in twin...

Area of Science:

• Biostatistics
• Epidemiology
• Reproductive Health

Background:

• Intraclass correlation coefficients (ICCs) are crucial for sample size calculations in clustered study designs.
• Estimates for ICCs can vary significantly depending on the chosen statistical method, particularly for binary outcomes.
• Accurate sample size determination is essential for achieving adequate statistical power in research studies.

Purpose of the Study:

• To evaluate the performance of five common intraclass correlation coefficient (ICC) estimators in sample size calculations for twin pregnancy studies.
• To determine which ICC estimation methods reliably achieve desired statistical power (80%) with a 5% Type I error rate.
• To compare sample size calculations based on ICCs derived from logistic GEE, ANOVA, LMM, and logistic GLMM.

Main Methods:

• Simulated twin pregnancy studies with diverse clustering levels and outcome prevalence.
• Calculated sample sizes using a standard formula based on ICC estimates from logistic GEE, ANOVA, LMM, and logistic GLMM.
• Assessed empirical power through simulation to validate the accuracy of calculated sample sizes.

Main Results:

• ICC estimates from GEE, ANOVA, and LMM were consistent and produced accurate sample sizes across varying outcome prevalence.
• Logistic GLMM-derived ICC estimates showed variability with outcome prevalence.
• Logistic GLMM resulted in overly large sample size calculations, particularly with high clustering or low outcome prevalence, leading to inflated power.

Conclusions:

• For sample size calculations in twin pregnancy research, logistic GLMM-based ICC estimates should be used with caution.
• GEE, ANOVA, and LMM provide more reliable ICC estimates for sample size determination in these studies.
• Choosing appropriate ICC estimators is critical for efficient and accurate sample size planning in clustered study designs.