Contrastive Machine Learning to Quantify Hypertensive Multiorgan Damage and Identify New Disease Phenotypes: A...

Area of Science:

• Cardiovascular Medicine
• Medical Imaging
• Machine Learning

Background:

• Hypertension causes significant organ damage, increasing vascular event and mortality risk.
• Subclinical organ damage is a key predictor but challenging to detect clinically.
• Novel methods are needed to quantify and track hypertension-induced multiorgan damage.

Purpose of the Study:

• To develop a machine learning framework for quantifying multiorgan damage in hypertension.
• To map disease progression and predict organ-specific trajectories.
• To identify distinct hypertension-associated organ damage phenotypes.

Main Methods:

• Utilized a semisupervised contrastive trajectory inference (cTI) framework on UK Biobank data (27,099 participants).
• Analyzed 566 multimodal imaging and nonimaging variables across multiple organs (heart, brain, kidneys, etc.).
• Validated externally using the Atherosclerosis Risk in Communities (ARIC) study (5,507 participants).

Main Results:

• Developed HyperScore, a global organ damage score with AUC of 0.964 for severe disease identification in UK Biobank.
• HyperScore stages significantly predicted survival, outperforming blood pressure stratification.
• Identified 6 distinct hypertensive disease phenotypes (HyperTrajectory) with external validation in ARIC.

Conclusions:

• Machine learning-derived organ damage scores are feasible for assessing hypertension.
• HyperScore enables identification of distinct hypertension-associated organ damage phenotypes.
• Potential for personalized risk assessment and phenotype-specific interventions using imaging data.