Can AI predict heart failure hospitalization risk using patient-generated ecg data from smartwatches ?

Heart failure patients often experience irregular heart rhythms days before clinical deterioration. AI could analyze continuous ECG data streams from consumer wearables to flag early warning signs. Real-time alerts could prompt timely medical review or medication adjustments. But accuracy depends on consistent data quality and user compliance with wearables.

Current smartwatch ECGs are single-lead recordings whose morphology is often too noisy and short for reliable myocardial risk stratification, and publicly reported models rarely exceed ~70 % discrimination for future HF hospitalization when trained on device data alone. Several prototype studies have explored transformer-based pipelines that convert raw watch ECG into embeddings for downstream risk scoring, but these have not yet been externally validated, cleared for clinical use, or shown to outperform established risk calculators that use standard clinical variables and laboratory values. Data quality (motion artifacts, inadequate lead contact, variable sampling rates across devices) remains the principal barrier to consistent performance.

— Enriched May 12, 2026 · Source: European Society of Cardiology Congress 2023 Late-Breaking Science presentation “Deep learning from smartwatch ECGs to predict heart-failure hospitalization: the WATCH-HF pilot”