Health
AI model detects hidden heart signal that predicts sudden cardiac death, study finds
Coverage
2 reports · over 22 hours
Researchers trained a deep-learning model on 440,000 ECG recordings from Sweden and tested it on about 140,000 US patients, the study in Nature reports.
- The model flagged 2.2% of the Swedish sample as high risk, a group that recorded a 7% sudden cardiac death rate over the following year.
- The researchers said the standard left ventricular ejection fraction measure failed to identify 86.1% of the patients the AI flagged as high risk.
- The team reported a previously undescribed waveform feature, a disguised downstroke, that independently predicted sudden cardiac death and may reflect heart muscle scarring.
- Among AI-flagged high-risk patients who received implanted defibrillators, sudden cardiac death ran 54.4% below the statistically expected rate, the researchers reported.

