Artificial Intelligence ECG Model Identifies Patients at Higher Risk for Sudden Cardiac Death

A deep learning model trained on electrocardiograms (ECGs) identified patients at elevated risk for sudden cardiac death and outperformed left ventricular ejection fraction (LVEF), the most widely used clinical biomarker for determining defibrillator eligibility, according to findings published in Nature.¹

The model isolated a high-risk group representing 2.2% of the Swedish study sample with a 7.0% annual sudden cardiac death rate. By comparison, patients with reduced LVEF represented 1.9% of the sample and had a 4.6% annual sudden cardiac death rate. Notably, 86.1% of patients flagged as high risk by the ECG model were not identified by reduced LVEF.¹

Why is sudden cardiac death risk difficult to predict?

Sudden cardiac death is considered preventable in some patients through implantable cardioverter defibrillators, but current tools do not reliably identify who will benefit. LVEF is widely used in clinical practice, but the study authors noted that it misses many patients who later experience sudden cardiac death and also identifies many patients who receive defibrillators that never deliver a lifesaving shock.¹

Each year, sudden cardiac arrest causes more than 300 000 deaths in the United States, according to the Berkeley News announcement. Ziad Obermeyer, MD, associate professor at the University of California, Berkeley School of Public Health and lead author of the study, said the central challenge is that “doctors can’t figure out who needs one before it’s too late.”²

How was the AI model developed?

Investigators trained the deep learning model using ECG waveforms from a Swedish region linked to death certificates and electronic health records. The primary analysis was conducted in a locked holdout dataset of 119 541 ECGs with 1-year follow-up from 35 885 patients younger than 80 years. The model’s area under the receiver operating characteristic curve for sudden cardiac death was 0.872.¹

The model also was externally validated in 2 additional datasets. In a large US health system, the model predicted ventricular fibrillation or ventricular tachycardia with an AUC of 0.822. In that cohort, the riskiest 2.2% of ECGs had a 29.1% annual incidence of ventricular fibrillation or ventricular tachycardia, compared with a base rate of 3.8%. In a Taiwan hospital registry, the model distinguished future arrhythmic arrests from controls with an AUC of 0.767.¹

How did the model compare with LVEF?

The ECG model appeared to identify a group of patients not captured by current LVEF-based risk assessment. Only 13.9% of the high-risk ECG group had known reduced LVEF. Patients in the model-defined high-risk group had a higher sudden cardiac death rate than those with reduced LVEF, at 7.0% vs 4.6%.¹

Even among patients with measured and normal LVEF, the ECG model identified a group with a 6.4% sudden cardiac death rate, higher than the rate among patients with reduced LVEF.¹

Did defibrillators appear to help high-risk patients?

The study was observational and was not designed to prove that defibrillators reduce mortality in patients flagged by the ECG model. However, researchers found that high-risk patients who already had defibrillators implanted were 54.4% less likely to die than expected from sudden cardiac death.¹

The authors wrote that randomized trials will be needed to determine whether patients identified by the ECG biomarker benefit from defibrillator implantation.¹

What did the AI model find in the ECG?

To better understand what the model was detecting, investigators paired the predictive model with a generative model of ECG waveforms. This approach revealed a visible ECG morphology associated with elevated risk, including axis deviation and a previously undescribed slurring pattern in the terminal portion of the R wave in lead aVL.¹

For primary care clinicians, the findings are not yet practice-changing. However, they suggest that routinely available ECG data may eventually help identify patients who require more detailed cardiac evaluation, ambulatory monitoring, or electrophysiology referral. Future prospective validation and randomized trials will be critical before the model can be incorporated into clinical decision-making.¹

References

1. Obermeyer Z, Schubert A, Ross J, Mullainathan S, Lingman M. An ECG biomarker for sudden cardiac death discovered with deep learning. Nature. Published online June 24, 2026. doi:10.1038/s41586-026-10674-6
2. Pohl J. With AI, researchers discover new way to detect sudden cardiac death risk. Berkeley News. Published June 24, 2026. https://news.berkeley.edu/2026/06/24/with-ai-researchers-discover-new-way-to-detect-sudden-cardiac-death-risk/