New AI Tool May Effectively Predict Sudden Cardiac Arrest Risk in Otherwise Healthy Patients

A new artificial intelligence (AI) tool developed by UC Berkeley routinely outperformed standard clinical tests in identifying patients at risk of sudden cardiac arrest.¹

Clinicians have long understood that sudden cardiac arrest is theoretically preventable with defibrillators. However, without an efficient way of identifying which patients may be at risk, this strategy is extremely difficult to implement reliably. The only predictive biomarker for this condition, cardiac left ventricular ejection fraction (LVEF), notoriously misses most sudden cardiac deaths. Currently, cardiac arrhythmias cause >100,000 sudden deaths in the US every year, despite the availability of these implanted defibrillators, which can find and terminate these arrhythmias before they can result in mortality.²

“One thing that makes the problem very tragic, but also very well suited for AI, is that we have the cure for this problem,” Ziad Obermeyer, MD, an associate professor at the School of Public Health at UC Berkeley and the lead author of the study, said in a statement. “If you knew you were one of the people who was going to drop dead, you would go to a cardiologist, and you’d get a defibrillator implanted. The problem is that doctors can’t figure out who needs one before it’s too late.”¹

Obermeyer and colleagues collected >440,000 electrocardiogram (ECG) records from patients in Sweden, linking them to data from death certificates, and fed these data into a deep learning program. They also included scans from healthy patients, aiming to train the model to recognize cardiac waveform patterns for those who later suffered sudden cardiac arrest.²

The program isolated the highest-risk group from the records provided, accounting for roughly 2.2% of the sample. These patients exhibited a 7% annual rate of sudden cardiac arrest, which was higher than patients with reduced LVEF, which made up 1.9% of the sample with a 4.6% annual rate. Roughly 86.1% of the high-risk patients were not flagged by LVEF, indicating the insufficiency of this biomarker. Additionally, patients who had high-risk ECG records who had been implanted with defibrillators were 54.4% less likely to die than expected, according to the model.²

A second phase of the study is currently ongoing, in which Obermeyer and colleagues will validate the model externally, working with a US health system and a Taiwanese hospital registry. The former will focus on predicting ventricular arrhythmias that may cause sudden cardiac arrest, while the latter will predict future arrhythmic cardiac arrests. For those scans flagged as high-risk, clinicians will then notify patients, offering them a patch to continually monitor their heart. This data could, according to the press release, allow clinicians to better understand the physiological mechanism within the heart.¹

The team has assembled a website where individuals who want to assess their own risk can submit basic information and their email address, which will allow the team to contact them for analysis when the AI becomes more widely available. The site can be reached here. The information from this site, along with the ongoing second phase of the study, will afford Obermeyer and colleagues additional testing data in a specialty with a lack of accessible patient records.¹

“Good AI starts with good data,” Obermeyer said in a statement. “Unfortunately, data like the ones we used for this study are incredibly hard to access. It’s a big part of why there’s so little clinical AI in use today.”¹

References

1. Pohl J. With AI, researchers discover new way to detect sudden cardiac death risk. UC Berkeley News. June 24, 2026. Accessed June 29, 2026. https://news.berkeley.edu/2026/06/24/with-ai-researchers-discover-new-way-to-detect-sudden-cardiac-death-risk/

2. 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