Interoperability of phenome-wide multimorbidity patterns: a comparative study of two large-scale EHR systems

Area of Science:

• Biomedical Informatics
• Computational Biology
• Network Science

Background:

• Electronic health records (EHR) are increasingly utilized for multimorbidity research.
• Concerns exist regarding EHR data accuracy, completeness, and suitability for research due to administrative design.
• Consistency and reproducibility of EHR-based multimorbidity studies are questioned.

Purpose of the Study:

• To assess the consistency and reproducibility of multimorbidity networks derived from two major EHR systems.
• To explore the utility of EHR data for phenome-wide multimorbidity analysis.
• To uncover clinically interpretable disease relationships and patterns.

Main Methods:

• Utilized phecodes to represent diseases and analyzed pairwise comorbidity strengths.
• Constructed multimorbidity as weighted undirected graphs using dual logistic regression.
• Assessed network consistency at local, meso, and global scales across EHR systems.
• Provided an interactive web tool and knowledge base for online analysis.

Main Results:

• Observed strong correlations in disease frequencies (Kendall's τ = 0.643) and comorbidity strengths (Pearson ρ = 0.79) between EHR systems.
• Found consistent network statistics and structures across different scales and EHRs.
• Demonstrated alignment of multimorbidity patterns with genetic correlations.
• Identified a consistent core-periphery network structure and uncovered clinically relevant disease relationships through case studies.

Conclusions:

• Large-scale EHR data are robust for phenome-wide multimorbidity studies.
• Multimorbidity patterns align with genetic data, suggesting shared disease biology.
• Consistent network structures offer insights into complex disease interactions and support precision medicine.