[arXiv]score: 0.10

DeepHHF Predicts 5-Year Heart Failure Risk from 24-Hour ECG with AUC 0.80

June 19, 2026

DeepHHF, trained on 69,663 Holter ECG recordings from 47,729 patients, achieves AUC 0.80 for predicting heart failure within five years — outperforming both 30-second segment models and clinical scores. High-risk patients flagged by the model had a two-fold increase in hospitalization or death. Explainability analysis identified arrhythmias and structural abnormalities as primary predictive signals.

HOW THIS AFFECTS YOU

●

researcherThe 24-hour temporal modeling approach and objective outcome labels on a 20-year dataset provide a strong baseline architecture for long-sequence biosignal prediction tasks.

●

healthA validated, explainable model predicting HF risk from routine Holter data could integrate into existing cardiac monitoring workflows without requiring additional diagnostics.

read original ↗arxiv.org

← back to feed