Explainable artificial intelligence for stroke risk stratification in atrial fibrillation

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence
• Health Informatics

Background:

• Current risk stratification tools are insufficient for emerging therapies in heart rhythm disorders.
• Stroke prevention treatments for atrial fibrillation (AF) have advanced beyond current risk stratification capabilities.

Purpose of the Study:

• To explain key artificial intelligence (AI) concepts for improved risk stratification.
• To highlight probabilistic graphical models (PGMs) for personalized risk assessment in cardiovascular medicine.

Main Methods:

• Explanation of AI concepts relevant to risk stratification.
• Development and application of a preliminary PGM for AF-related stroke risk.
• Integration of social determinants of health with clinical data.

Main Results:

• Demonstrated PGM application in a large cohort (1.6 million patients).
• Showcased PGM's ability to combine diverse data types for enhanced prediction.
• Highlighted potential for improved, personalized risk assessment.

Conclusions:

• AI, particularly PGMs, can significantly advance personalized risk stratification for heart rhythm disorders.
• PGMs enable the integration of non-traditional health data for more equitable risk assessment.
• Computational technologies combined with electronic health data promise explainable and equitable risk evaluation.