Prognostic implications of machine learning-derived echocardiographic phenotypes in community hypertensive patients

Echocardiogram is commonly used to evaluate cardiac remodeling in hypertension (HTN). However, study on echocardiographic phenotypes and their prognostic implications in HTN is limited.

We aimed to evaluate the prognostic implications echocardiographic phenotypes in community hypertensive patients.

A total of 1881 community hypertensive patients without overt cardiovascular disease and severe renal disease (mean age 62.8 years, women 57.9%) were included. Using Two-Step cluster analysis with four conventional echocardiographic variables, two clusters with distinct echocardiographic phenotypes were identified.

The Cluster 1 (namely "mild-remodeling" HTN; n = 1492) had low prevalence of enlarged left atrium (LA; 0.9%) and left ventricular hypertrophy (LVH; 16.2%) and better LV diastolic function. They were younger and more likely to be men and had lower comorbid burden. The Cluster 2 (namely "severe-remodeling" HTN; n = 389) had higher prevalence of enlarged LA (26.0%) and LVH (83.0%) and worse LV diastolic function. They were older and more likely to be women and had higher comorbid burden. After a median follow-up of 4.2 years, compared to the Cluster 1, the Cluster 2 had higher incidence of cardiovascular (4.1% vs 1.7%; P = .006) and all-cause (9.8% vs 4.8%; P < .001) death, with adjusted hazard ratio of 2.80 (95% CI 1.39-5.62; P = .004) and 2.04 (95% CI 1.32-3.14; P < .001) respectively.

These findings indicate that the conventional echocardiographic variables-based algorithm could help identify asymptomatic community hypertensive patients at risk for cardiovascular and all-cause death. Further studies are needed to develop and validate phenotype-specific prevention and intervention strategies in HTN.