Clinical Phenotypes in Hypertension: A Data-Driven Approach to Risk Stratification

Area of Science:

• Cardiology
• Medical Informatics
• Genetics

Background:

• Hypertension is a leading cause of cardiovascular disease, but its diverse nature complicates risk assessment.
• Unsupervised machine learning offers a way to identify distinct patient subgroups for better risk stratification and targeted prevention.
• This study aimed to use a data-driven approach to define hypertension phenotypes and their links to cardiovascular outcomes.

Purpose of the Study:

• To identify distinct clinical phenotypes of hypertension using unsupervised machine learning.
• To investigate the association of these phenotypes with cardiovascular imaging characteristics and adverse clinical outcomes.
• To evaluate the mediating role of cardiac imaging features in the relationship between hypertension phenotypes and outcomes.

Main Methods:

• Analysis of 14,840 UK Biobank participants with diagnosed hypertension and cardiovascular magnetic resonance imaging data.
• Application of k-means clustering to 77 clinical variables to identify distinct hypertension clusters.
• Examination of associations between clusters and outcomes (heart failure, atrial fibrillation, atherosclerotic events, mortality) adjusted for risk factors, with mediation analyses for imaging features.

Main Results:

• Three distinct hypertension clusters were identified.
• Cluster 2, predominantly male with high atherosclerosis, exhibited the highest risk for all adverse cardiovascular events, linked to severe cardiac remodeling and dysfunction.
• Cluster 3, resembling metabolic syndrome, showed moderate risks for atrial fibrillation and mortality, with risk partially mediated by left ventricular hypertrophy and left atrial dysfunction.

Conclusions:

• Clustering analysis successfully identified distinct hypertension phenotypes with unique risk profiles.
• These findings suggest that a data-driven approach can improve hypertension risk stratification.
• Tailored treatment strategies may be developed based on these identified phenotypes.