Machine Learning-Derived Cardiovascular Aging Phenotypes From Cardiac Function and Stroke Risk in the UK Biobank:...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Cardiovascular magnetic resonance (CMR) is essential for assessing cardiac structure and function.
• Machine learning (ML) offers powerful tools for predicting clinical outcomes and analyzing complex health data.

Purpose of the Study:

• To derive CMR-derived phenotypes associated with cardiovascular aging.
• To evaluate the relationship between these phenotypes and long-term stroke risk.
• To refine phenotype identification using supervised ML.

Main Methods:

• Utilized UK Biobank data from 36,467 participants without prior stroke.
• Employed generative topographic mapping and agglomerative hierarchical clustering to identify phenotypes.
• Applied supervised ML (random forest) for phenotype prediction and Cox models for stroke risk analysis.

Main Results:

• Identified two phenotypes: one reflecting cardiovascular aging with adverse cardiac function, and another associated with reduced stroke risk (HR 0.695).
• The reduced stroke risk phenotype remained significant after adjusting for competing mortality (HR 0.578).
• The random forest model demonstrated high accuracy (AUC 0.914 training, 0.867 validation) and calibration.

Conclusions:

• Integrated unsupervised and supervised ML to identify cardiovascular aging phenotypes.
• These phenotypes show strong predictive power for incident stroke.
• Findings may enhance preventive and therapeutic strategies for populations at high risk of stroke.