Development and multinational validation of an ensemble deep learning algorithm for detecting and predicting...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Devices

Background:

• 12-lead ECGs enhanced by AI can identify structural heart diseases (SHDs), but their use in community screening is limited.
• A noise-resilient single-lead AI-ECG algorithm was developed to detect SHDs and predict their development risk using wearable devices.

Purpose of the Study:

• To develop and validate a novel AI-ECG algorithm for detecting and predicting structural heart diseases (SHDs) using portable devices.
• To assess the algorithm's performance in diverse populations and its utility as a predictive biomarker for future SHD risk.

Main Methods:

• Developed ADAPT-HEART, a deep learning algorithm, using 266,740 ECGs from 99,205 patients with paired echocardiographic data.
• Validated ADAPT-HEART in four US community hospitals, the ELSA-Brasil cohort, and the UK Biobank.
• Defined SHD as LVEF < 40%, moderate/severe valvular disease, or severe LVH.

Main Results:

• ADAPT-HEART demonstrated strong performance in detecting SHD with an AUROC of 0.879 in the test set, consistent across external validation sites (AUROCs 0.852-0.891).
• Among individuals without baseline SHD, elevated ADAPT-HEART probability significantly increased the risk of future SHD (2.8- to 5.7-fold increase).

Conclusions:

• A novel AI model effectively detects and predicts SHDs from single-lead ECGs on portable devices.
• This technology offers a scalable strategy for community-based screening and risk stratification of SHD.