AI Uncovers Hidden ECG Signal That Predicts Sudden Cardiac Death Better Than Standard Tests

Every year, sudden cardiac arrest kills more than 300,000 people in the United States . The heart's electrical system suddenly stops firing, often without any warning. An effective treatment exists—an implantable defibrillator that can shock the heart back into rhythm—but doctors have had no reliable way to identify who needs one before it is too late . Now, a team at UC Berkeley led by associate professor Ziad Obermeyer has taken a significant step toward solving that problem.

The hidden signal in a routine EKG

In a study published June 24, 2026 in Nature, the researchers revealed that a deep learning model trained on over 440,000 electrocardiograms (EKGs) discovered a previously unrecognized electrical signal in routine heart scans. This signal predicts sudden cardiac death risk substantially better than current clinical standards .

The AI identified subtle waveform patterns—spikes and electrical currents produced by the heart—that neither human readers nor standard clinical tests can detect . These patterns correlate with the heart's electrical system malfunctioning before sudden cardiac arrest. The exact physiological mechanism is not yet understood, but the AI appears to have homed in on a feature related to the heart suddenly and fatally misfiring .

Training on three continents

The researchers trained their deep learning model using more than 440,000 EKGs from Sweden, linking each scan to death certificates . The system learned to recognize waveform patterns predictive of sudden cardiac death by analyzing scans from healthy people, at-risk patients, and those who later died of sudden cardiac arrest .

Crucially, the team then validated the model on thousands of additional patient files from two independent populations: the San Diego region in the United States and Taipei in Taiwan . An accompanying Nature news article confirms that the model was developed using extensive ECG data and mortality records .

Better risk detection, more lives saved

The AI system identified a high-risk group with a 7% annual rate of sudden cardiac death . For comparison, standard clinical tests—which measure how much blood the heart ejects per beat—identify a high-risk group with only a 4.6% annual rate .

The model flagged a larger high-risk pool and better predicted who would suffer sudden cardiac death. These differences translate to thousands of patients annually who currently appear low-risk by conventional measures .

Why detecting risk has been so difficult

Sudden cardiac arrest occurs when the heart's electrical system abruptly stops firing without warning. Obermeyer explains that an effective cure exists—implantable defibrillators that shock the heart back into rhythm—but doctors have been unable to determine who needs one before it is too late .

The core problem is that people die so suddenly that it is nearly impossible to know what was happening inside the heart beforehand. Autopsies reveal structural details about the heart, but they cannot show the electrical functioning at the moment immediately before death .

Next steps: from algorithm to bedside

The researchers plan to deploy the algorithm within health systems to help doctors better identify patients who would benefit from an internal defibrillator . The study also opens the door for new research into the underlying physiological mechanism behind cardiac electrical malfunctions.

"The goal is to not only make better decisions, but also start to understand what's actually going on with these patients before their heart stops," Obermeyer stated . Because EKGs are routine, low-cost, and available at medical centers worldwide, the tool could be scaled broadly to save lives .