Abstract 13901: Deep Learning-Based Digital Biomarker to Diagnose Hypertension and Stratify Cardiovascular Risk From the Electrocardiogram

Abstract

Background/Aim: Hypertension affects nearly half of US adults and is a major modifiable risk factor for cardiovascular disease (CVD). Guidelines recommend serial blood pressure (BP) measurement under controlled conditions or ambulatory BP monitoring, but BP is often measured intermittently under suboptimal conditions, leading to underdiagnosis and undertreatment. We developed a deep learning model (HTN-AI) to identify hypertension and stratify risk of hypertension-associated CVD from 12-lead ECGs. Methods: HTN-AI is a convolutional neural network trained to detect prevalent hypertension or elevated baseline BP (systolic ≥ 140 mmHg or diastolic ≥ 90 mmHg) using 747,992 ECGs from 106,879 adults receiving primary or cardiology care at Massachusetts General Hospital (MGH). Associations between HTN-AI-predicted risk of hypertension and CVD (all-cause mortality, heart failure [HF], myocardial infarction [MI], stroke, and aortic dissection) were assessed in 56,730 primary care patients at Brigham and Women’s Hospital (BWH) with cumulative incidence analysis and cause-specific hazard regression. Results: In an MGH test sample and the BWH sample, HTN-AI discriminated prevalent hypertension or elevated baseline BP (AUC 0.791 and 0.762, respectively). Quintiles of HTN-AI risk stratified cumulative incidence of mortality (Figure), HF, MI, stroke, and aortic dissection (p < 0.001 for all). HTN-AI risk was significantly associated with age- and sex-adjusted risk of incident mortality (HR per SD: 1.85 [95% CI: 1.77-1.92], p<0.001), HF (3.55 [3.25-3.88], p<0.001), MI (2.33 [2.20-2.47], p<0.001), stroke (1.74 [1.64-1.86], p<0.001), and aortic dissection (1.80 [1.41-2.31], p<0.001). Conclusion: A deep learning model, HTN-AI, discriminates patients with hypertension and stratifies risk of incident CVD using 12-lead ECGs. HTN-AI has the potential to facilitate diagnosis of hypertension and serve as a novel digital biomarker for hypertension-associated CVD.