Bivariate correlation coefficients in family-type clustered studies.

Area of Science:

• Biostatistics
• Statistical Genetics
• Quantitative Psychology

Background:

• Estimating bivariate correlation coefficients (BCCs) in clustered family data presents challenges.
• Existing methods struggle with single-subject families and missing paired measurements.

Purpose of the Study:

• To develop a unified approach for estimating multiple BCCs and their variances in family-type clustered data.
• To address analytic challenges posed by family-based study designs, including missing data and imbalanced family sizes.

Main Methods:

• Utilized a bivariate linear mixed-effects model for robust estimation.
• Implemented likelihood-based inferences and provided SAS software solutions.
• Defined BCCs at cluster and subject levels to capture different relational aspects.

Main Results:

• The proposed model effectively handles missing data and imbalanced family sizes, outperforming existing estimators.
• Extensive simulations confirmed the superiority of the new estimators.
• The Wald test demonstrated good size and power for hypothesis testing.

Conclusions:

• The bivariate linear mixed-effects model offers a superior, unified approach for analyzing correlations in clustered family data.
• This method maximizes data usage, even with missing values and varying family sizes.
• Applied to Alzheimer's disease data, it revealed significant correlations across different biomarker modalities.