Development and Validation of Deep-Learning Algorithm for Electrocardiography-Based Heart Failure Identification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early heart failure (HF) detection is crucial but faces challenges with conventional methods.
• Electrocardiography (ECG) offers a potential avenue for non-invasive HF screening.
• Limitations in current diagnostic tools necessitate novel approaches for HF identification.

Purpose of the Study:

• To develop and validate a deep-learning algorithm for ECG-based heart failure identification (DEHF).
• To assess the algorithm's accuracy in detecting HF with reduced ejection fraction (HFrEF) and mid-range to reduced EF.
• To compare the DEHF algorithm's performance against established machine learning models.

Main Methods:

• Utilized 55,163 ECGs from 22,765 patients across two hospitals.
• Developed the DEHF algorithm using derivation ECG data, incorporating demographic and ECG features.
• Validated the DEHF algorithm against logistic regression (LR) and random forest (RF) models using separate validation data.

Main Results:

• DEHF achieved high accuracy in identifying HFrEF, with AUROCs of 0.843 (internal) and 0.889 (external validation).
• The algorithm significantly outperformed LR and RF models in both internal and external validation sets.
• DEHF also demonstrated superior performance in identifying the secondary endpoint (EF ≤50%) compared to LR and RF.

Conclusions:

• The developed deep-learning algorithm (DEHF) accurately identifies heart failure using ECG features.
• DEHF demonstrates superior diagnostic performance compared to traditional machine learning methods like LR and RF.
• This AI-driven approach holds promise for improved early screening and diagnosis of heart failure.