UC Berkeley scientists say a new artificial intelligence tool can spot a clear, previously overlooked signal on a routine electrocardiogram that singles out a small group of patients at much higher risk of sudden cardiac death. In the study, that high-risk subset faced about a 7% chance of sudden death within a year, a rate that topped the risk levels flagged by the standard ejection-fraction screening used in cardiology today. The work, published this month in Nature, could eventually change how doctors decide who needs close monitoring or an implantable defibrillator, if future testing backs it up.
How the study was done
The peer-reviewed paper describes a deep-learning model trained on more than 440,000 ECGs from a region in Sweden, linked to death certificates and electronic health records. Researchers then checked the algorithm on separate datasets from a United States health system and a hospital registry in Taiwan. According to Nature, the model identified a high-risk group that made up about 2.2% of the Swedish sample and had a one-year sudden-death rate of about 7.0%. By comparison, people flagged by reduced left-ventricular ejection fraction had a sudden-death rate of roughly 4.6%.
A new, visible ECG clue
Instead of spitting out a mysterious black-box score, the team paired its classifier with a generative model so they could see what the algorithm was actually keying in on. That approach highlighted a previously undescribed change in the QRS complex on lead aVL of the ECG. The paper reports that this specific shape, or morphology, was a strong predictor of arrhythmic death and had not appeared in earlier medical literature. “Medical decisions are really hard, and I think that’s why AI is so exciting for me,” said Ziad Obermeyer, the study’s lead author, in an interview, per Berkeley News.
What the numbers mean
The Nature paper notes that most of the people the model labeled as high risk would have sailed past current screening tools. About 86.1% of those flagged by the algorithm did not have reduced left-ventricular ejection fraction. The study also reports that high-risk patients who already had implantable defibrillators were about 54% less likely to die than expected. The authors take that as suggestive evidence that the patients identified by the AI could benefit from treatments that are already available.
Next steps and local testing
For now, the researchers are working with health systems in Sweden, Taiwan and the United States to roll out and test the algorithm on hospital ECG databases, then follow the flagged patients with wearable heart monitors. The team, part of UC Berkeley’s Computational Precision Health network and affiliated start-ups, stresses that the tool is still in the research phase, not something doctors can routinely order and not something the public can download. Coverage from Berkeley News also notes a website where people can sign up to hear about future analyses, while the study authors emphasize that any broader clinical rollout will need more validation and regulatory oversight.
Why experts say proceed cautiously
Outside commentators have praised the sheer size of the study and the attempt to tie the algorithm’s output to a visible ECG pattern, rather than accept a purely opaque score. At the same time, they warn against rushing this into practice. Definitions of sudden cardiac death, differences between datasets and privacy concerns can all affect how well a model like this works in the real world. Reporting in STAT notes that the signal may be linked to cardiac fibrosis, which would be a biologically plausible explanation, but also underscores that prospective testing is needed before guidelines change. The study’s authors and independent experts point to the tradeoffs of wider screening: more intensive monitoring might catch serious risks early, yet it could also increase follow-up procedures for many patients.
What patients should know
Clinicians say this is not something patients should try to apply on their own. The algorithm is not available for home use, and medical decisions still hinge on established tools like echocardiography combined with a full clinical assessment. Anyone dealing with fainting, unexplained dizziness, palpitations or a family history of sudden cardiac death is urged to talk with a doctor about symptoms and testing options. For accessible rundowns of the study and what might come next, readers can look to coverage from FOX5 Atlanta.
