A machine learning based approach to identify carotid subclinical atherosclerosis endotypes

Area of Science:

• Cardiovascular Medicine
• Biomarkers
• Medical Imaging

Background:

• Subclinical atherosclerosis in the carotid arteries is a key indicator of future cardiovascular events.
• Identifying distinct biological subtypes (endotypes) of carotid atherosclerosis can refine risk prediction.
• Current risk assessment models may not fully capture the heterogeneity of atherosclerosis progression.

Purpose of the Study:

• To define and characterize distinct endotypes of carotid subclinical atherosclerosis.
• To investigate the association of these endotypes with atherosclerosis progression and atherosclerotic cardiovascular disease (ASCVD) risk.
• To evaluate the utility of endotype-based models for improving ASCVD risk prediction.

Main Methods:

• Integrated demographic, clinical, and molecular data with carotid ultrasonography in the IMPROVE cohort (n=3340).
• Applied neural network and hierarchical clustering to identify endotypes, using SHapley Additive exPlanations (SHAP) for feature analysis.
• Validated endotype prediction in the independent PIVUS cohort (n=1061).

Main Results:

• Identified four distinct carotid atherosclerosis endotypes, ranging from mild (endotype 1) to severe (endotype 4).
• Endotype 4 showed significantly thicker carotid intima-media thickness (c-IMT), more plaques, faster progression, and higher ASCVD risk compared to endotype 1.
• Endotype-based models consistently improved ASCVD risk discrimination and reclassification in both cohorts.

Conclusions:

• Four replicable endotypes of subclinical carotid atherosclerosis were identified, linked to disease progression and ASCVD risk.
• The endotype-based approach offers a promising strategy for precision medicine in ASCVD prevention.
• These findings can guide targeted interventions for individuals at different atherosclerosis risk levels.