Multi-Level Latent Variable Models for Coheritability Analysis in Electronic Health Records

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Electronic health records (EHRs) with familial data offer insights into complex trait genetics.
• Existing methods struggle with familial structures, phenotype heterogeneity, and scalability.
• A robust framework is needed for large-scale genetic analysis in EHRs.

Purpose of the Study:

• To develop a flexible statistical framework for estimating heritability and genetic correlation in EHR-based family studies.
• To address limitations of existing methods regarding familial correlations, phenotype types, and computational efficiency.
• To enable joint analysis of continuous and binary phenotypes within EHR data.

Main Methods:

• Utilizing multi-level latent variable models to partition phenotypic covariance into genetic and environmental factors.
• Incorporating both within- and between-family variations for comprehensive analysis.
• Employing generalized estimating equations (GEE) for robust estimation and inference.

Main Results:

• Simulation studies confirm the consistency and validity of the proposed estimators across diverse settings.
• Application to real-world EHR data identified significant genetic correlations between mental health and endocrine/metabolic phenotypes.
• Demonstrated shared etiological pathways between seemingly distinct complex conditions.

Conclusions:

• The developed framework offers a scalable and rigorous approach for coheritability analysis in high-dimensional EHR data.
• Facilitates the identification of shared genetic influences within complex disease networks.
• Advances the understanding of the genetic architecture underlying common diseases.