An ECG biomarker for sudden cardiac death discovered with deep learning
4 hours ago
- #defibrillator risk prediction
- #deep learning ECG
- #sudden cardiac death
- Deep learning model using ECGs predicts sudden cardiac death with AUC of 0.872, outperforming LVEF which misses most cases.
- High-risk group (2.2% of sample) identified by model has 7.0% annual sudden cardiac death rate, higher than reduced LVEF group (4.6%).
- 86.1% of high-risk patients were not flagged by LVEF, indicating model discovers new at-risk individuals overlooked by current methods.
- Model validated in US and Taiwanese datasets, showing generalization to diverse populations and predicting ventricular arrhythmias.
- Generative model reveals new ECG biomarker, including slurred R-wave in lead aVL, linked to sudden cardiac death and possibly fibrosis.
- Observational data suggests defibrillators reduce mortality by 54.4% in high-risk patients, supporting potential clinical benefit.
- Model's practical advantages include using ubiquitous, standardized ECGs, enabling cost-effective screening without human expertise.