Modeling Day-Long ECG Signals to Predict Heart Failure Risk with Explainable AI

Researchers developed DeepHHF, a deep learning model trained on 24-hour ECG recordings that predicts heart failure risk within five years with 0.80 AUC accuracy, outperforming traditional 30-second ECG analysis and clinical scoring systems. The model identified high-risk patients with a two-fold increased chance of hospitalization or death, demonstrating that continuous cardiac monitoring combined with explainable AI offers a non-invasive, cost-effective approach to preventive healthcare.

The study addresses a critical healthcare challenge: heart failure affects nearly 12% of seniors and represents a major driver of morbidity and mortality. Traditional ECG screening relies on brief 30-second snapshots that miss paroxysmal cardiac events occurring sporadically throughout the day. DeepHHF leverages 24-hour Holter monitoring data to capture these episodic abnormalities, achieving superior predictive performance by modeling temporal patterns that shorter recordings cannot detect.

This advancement reflects a broader convergence of deep learning capabilities with clinical medicine. The TLHE dataset of nearly 70,000 recordings provides the scale necessary to train robust models, while explainability analysis confirms the model focuses on clinically relevant features—arrhythmias and structural abnormalities—rather than spurious correlations. This transparency builds trust in AI-driven medical recommendations, addressing physician skepticism about black-box algorithms.

The practical implications extend beyond prediction accuracy. Holter ECGs are widely available, inexpensive compared to advanced imaging, and require no invasive procedures. Deploying such models in primary care settings could enable early identification of at-risk patients before symptomatic heart failure develops, potentially preventing hospitalizations and improving outcomes at population scale. The two-fold stratification of risk enables targeted intervention strategies.

Looking forward, validation on independent cohorts remains essential before clinical deployment. Integration with electronic health records and other biomarkers could further refine risk stratification. Regulatory pathways for AI diagnostics continue evolving, and this work demonstrates that rigorous validation combined with explainability can support FDA approval and clinical adoption.

• →DeepHHF achieved 0.80 AUC predicting heart failure risk from 24-hour ECG data, outperforming traditional 30-second analysis methods.
• →Continuous cardiac monitoring captures paroxysmal events missed by conventional screening, improving early detection of at-risk patients.
• →Explainability analysis confirmed the model focuses on clinically relevant features like arrhythmias, supporting physician trust and adoption.
• →Single-lead Holter ECG analysis offers non-invasive, low-cost population screening suitable for deployment in resource-limited healthcare settings.
• →Two-fold stratification of hospitalization and mortality risk enables targeted preventive interventions in high-risk patient cohorts.