AI discovers a new ECG signal that predicts sudden cardiac death better than standard tests

How the model was trained and validated

The research team took a rigorous approach to model development. They paired over 440,000 EKGs from Swedish healthcare records with death certificate data so the AI could learn which waveform patterns preceded sudden cardiac death . The deep learning architecture analyzed the full 12-lead signal, not just summary measurements — letting it find subtle, nonlinear patterns invisible to human readers.

To ensure the findings weren't specific to Sweden, the model was externally validated on thousands of independent patient records from the United States and Taiwan. The predictions held across different populations and healthcare systems, providing strong evidence of generalizability .

Why detecting sudden cardiac death has been so difficult

Sudden cardiac arrest is fundamentally different from a heart attack. A heart attack involves a blocked artery starving the heart muscle of oxygen; sudden cardiac arrest is an electrical malfunction — the heart's electrical current stops firing without warning .

People die so quickly that studying what the heart was doing moments before is nearly impossible. Autopsies can show structural problems (blocked vessels, scarred tissue), but as the researchers noted, "the actual functioning before death remains something of a black box" .

The current gold-standard risk test — measuring left ventricular ejection fraction (LVEF), the percentage of blood the heart pumps per beat — is a blunt instrument. Many people who die from sudden cardiac arrest have a normal LVEF, and many with low LVEF never experience an arrest . The standard approach misses most of the people who need help.

Practical results compared to standard tests

The AI identified a high-risk group comprising about 2.2% of the screened population. The 7.0% annual sudden cardiac death rate in this group is comparable to or better than the risk threshold used in clinical trials for implantable defibrillators (ICDs) . This means many patients who would be missed by current guidelines could be candidates for life-saving devices.

What comes next

The research points to three clear next steps:

1. Clinical deployment for defibrillator decisions: EKGs are cheap, noninvasive, and available in nearly every clinic worldwide. The AI model could help doctors decide who needs an implantable cardioverter-defibrillator (ICD). As Obermeyer put it, "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" .

2. New physiological understanding: The novel waveform the AI discovered — without being told what to look for — opens a new research direction. Understanding the exact electrical mechanism behind the slurred R wave in lead aVL could reveal why some hearts suddenly misfire. "We can not only make better decisions, but also start to understand what's actually going on with these patients before their heart stops," Obermeyer said .

3. Prospective trials before widespread adoption: While the external validation across three countries is strong evidence, the model needs to be tested in prospective clinical trials before it enters routine clinical practice. The research team's work demonstrates the kind of rigorous, cross-population validation that makes this finding especially promising .