Unbiased Characterization of Atrial Fibrillation Phenotypic Architecture Provides Insight Into Genetic Liability and...

Area of Science:

• Cardiology
• Genetics
• Computational Biology

Background:

• Atrial fibrillation (AF) is a common arrhythmia with unclear underlying mechanisms and clinical heterogeneity.
• Machine learning can identify data-driven disease subtypes, but their clinical relevance in AF is not well-established.

Purpose of the Study:

• To identify distinct atrial fibrillation (AF) phenotypic clusters using machine learning.
• To investigate the association of these AF clusters with polygenic liability, inflammation, and clinical outcomes.

Main Methods:

• Applied unsupervised coclustering machine learning to 35 clinical features in 23,271 individuals with AF.
• Assessed clinical inflammation using biomarkers and genetic liability using polygenic risk scores for AF and cytokine levels.
• Analyzed associations with mortality, stroke, bleeding, and device use.

Main Results:

• Identified 3 AF phenotypic clusters characterized by increasing comorbidity burden (renal disease, coronary artery disease).
• The low comorbidity cluster had the highest polygenic liability to AF, while the high comorbidity cluster showed elevated clinical inflammation.
• Cluster assignment was significantly associated with mortality, stroke, bleeding, and cardiac device implantation.

Conclusions:

• AF patient subgroups identified by clustering differ in comorbidity burden, clinical inflammation, and polygenic liability to AF.
• These AF clusters are associated with distinct risks for clinical outcomes.
• Comorbidity and genetic liability are key drivers of AF phenotypic variability.