Heart HealthCommuniqué de presse

AI finds a hidden heart signal that predicts sudden cardiac death

UC Berkeley researchers trained AI on 440,000 EKGs to flag people at risk of sudden cardiac death, outperforming the standard ejection fraction screening.

mardi 17 novembre 2026 0 vue
Publié dans ScienceDaily Heart
Article visualization: AI finds a hidden heart signal that predicts sudden cardiac death

Résumé

Sudden cardiac arrest kills more than 300,000 people in the United States each year, and doctors struggle to tell who needs an implanted defibrillator. UC Berkeley researchers trained an AI system on more than 440,000 electrocardiograms from Sweden, linking the electrical patterns to death certificate data. They then tested it on records from the United States and Taiwan. Standard screening, which measures how much blood the heart pumps per beat, identified a group with a 4.6% annual sudden cardiac death rate. The AI identified a group with a 7% annual rate and also flagged many people that current guidelines would class as low risk. Because it works on ordinary EKGs already collected worldwide, it could improve screening without a new test. The work was published in Nature. Researchers still want to understand what changes in the heart create the signal the AI reads.

Résumé détaillé

Sudden cardiac arrest claims more than 300,000 lives a year in the United States. It happens when the heart's electrical system fails and the heart stops pumping blood. It can strike older adults with known heart disease, and it can also strike seemingly healthy young athletes. Implanted defibrillators can prevent some of these deaths, but doctors have long struggled to decide who should get one.

Researchers at UC Berkeley trained an artificial intelligence system on more than 440,000 electrocardiograms collected in Sweden. They compared the electrical patterns with death certificate information. The AI learned to separate healthy people, patients with known heart risks, and people who later died of sudden cardiac causes. The team then tested it on thousands of additional records from the United States and Taiwan to see whether it held up in different populations.

The AI beat the conventional approach, which relies mainly on ejection fraction, the amount of blood pumped with each contraction. Ejection fraction identified a group with a 4.6% annual sudden cardiac death rate. The AI identified a group with a 7% annual rate. It also found a broader set of at-risk people, including thousands who would usually be classed as low risk under current guidelines.

The system works with ordinary EKG images that hospitals and clinics already collect. That means it could be added to routine care without a new type of test. In time, it could help physicians make better decisions about implanted defibrillators and catch risk earlier in people who would otherwise be missed. The findings appeared in Nature.

There are caveats. The article does not say whether the tool has been tested prospectively in clinics, or whether using it improves patient outcomes. The biological meaning of the signal remains unknown, and researchers are now trying to learn which changes in the heart produce these electrical patterns. Death certificate data can also be imprecise about cause of death. Readers should check the full Nature paper for the validation details and for how the tool performs across subgroups.

Principales conclusions

  • AI trained on over 440,000 Swedish EKGs predicted sudden cardiac death using electrical patterns linked to death certificates.
  • Standard ejection fraction screening flagged a group with a 4.6% annual sudden cardiac death rate.
  • The AI flagged a higher-risk group with a 7% annual sudden cardiac death rate.
  • The model also found many at-risk people who would be classed as low risk under current guidelines.
  • It works on routine EKG images and was tested on records from the United States and Taiwan.

Méthodologie

This is a news report summarizing a peer-reviewed study published in Nature, relayed via a UC Berkeley press release on ScienceDaily. The evidence is retrospective AI modeling on EKG and death certificate data, with external testing in United States and Taiwan cohorts. The source is credible, but the article gives limited detail on methods.

Limites de l'étude

The report does not describe prospective trials or evidence that using the tool improves outcomes. The biological source of the signal is unexplained, and death certificate cause-of-death data can be imprecise. Verify the performance metrics, false-positive rates, and subgroup results in the original Nature paper.

Ce résumé vous a plu ?

Recevez les dernières recherches sur la longévité dans votre boîte de réception chaque semaine.

Saisissez votre e-mail pour vous abonner :