An explainable artificial intelligence approach for predicting cardiovascular outcomes using electronic health...

Area of Science:

• Computational biology
• Medical informatics
• Cardiovascular research

Background:

• Precision medicine requires understanding complex clinical variables affecting cardiovascular health.
• Electronic Health Records (EHRs) contain vast data but are challenging to analyze for comorbidities.

Purpose of the Study:

• To apply a novel comorbidity discovery method to large-scale EHR data.
• To identify and analyze the impact of comorbidities and demographics on cardiovascular outcomes.
• To develop portable, privacy-preserving tools for EHR-based research.

Main Methods:

• Utilized Poisson Binomial based Comorbidity discovery (PBC), a scalable method for comorbidity analysis.
• Analyzed EHR data from over 1.6 million patients across 77 million visits.
• Employed explainable Artificial Intelligence (AI) to dissect conditional dependencies between comorbidities, demographics, and cardiovascular health.

Main Results:

• Discovered complex multimorbidity networks associated with cardiovascular conditions.
• Identified key comorbid conditions and demographic factors influencing heart transplant, sinoatrial node dysfunction, and congenital heart disease.
• Developed web-based tools for exploring comorbid and demographic landscapes of cardiovascular outcomes.

Conclusions:

• The PBC method effectively transforms large EHR datasets into accessible, privacy-preserving research tools.
• This approach facilitates deeper understanding of multimorbidity and its impact on cardiovascular health.
• The developed tools support community-based outcomes research and advance precision medicine.