AI discovers a hidden heart signal that could predict sudden death

Now, researchers at UC Berkeley have identified a previously overlooked warning signal in routine heart scans that could make it easier to recognize people at risk before a fatal event occurs.

The discovery was made using artificial intelligence trained on more than 440,000 electrocardiograms, or EKGs, collected in Sweden. By comparing the electrical patterns recorded in these scans with information from death certificates, the researchers taught an AI system to recognize subtle differences between healthy individuals, patients with known heart risks, and people who eventually died from sudden cardiac causes.

The team then evaluated the technology using thousands of additional patient records from the United States and Taiwan, testing whether the patterns could reliably identify people at heightened risk in different populations.

AI Detects Hidden Heart Risks That Standard Tests Miss

The results suggest that AI can identify vulnerable patients more effectively than the conventional tests doctors typically rely on. Current screening methods largely focus on how much blood the heart pumps with each contraction, a measurement used to assess whether someone might benefit from an implanted defibrillator.

Using these traditional measurements, doctors can identify a group of patients with an annual sudden cardiac death rate of 4.6%. The AI system, however, identified a group with an annual death rate of 7%, demonstrating a greater ability to distinguish patients facing serious danger.

Importantly, the algorithm also recognized a broader population of people at elevated risk, including thousands who would not normally raise concerns under existing medical guidelines. Most of these individuals would have been classified as relatively low risk using conventional methods.

Another major advantage is that the AI system works with ordinary EKG images, which are already routinely collected in hospitals and medical centers around the world. This means the technology could potentially improve screening without requiring an entirely new type of medical test.

The findings, published in Nature, could eventually help physicians make more informed decisions about who should receive an implanted defibrillator. They also raise an intriguing scientific question about what changes inside the heart produce the electrical patterns that AI has learned to recognize.

"Medical decisions are really hard, and I think that's why AI is so exciting for me," said Ziad Obermeyer, an associate professor at UC Berkeley's School of Public Health and the study's lead author. "We can not only make better decisions, but also start to understand what's actually going on with these patients before their heart stops."

Why Sudden Cardiac Death Is So Difficult to Predict

Sudden cardiac arrest is different from a heart attack. A heart attack happens when blood flow to part of the heart becomes blocked or severely restricted. Cardiac arrest, by contrast, results from a failure of the electrical signals that coordinate the heartbeat.

Immediate CPR and an electrical shock from an automated external defibrillator can sometimes restore the heart's rhythm. However, roughly 90% of people who experience sudden cardiac arrest outside a hospital die within minutes, making early recognition of those at risk especially important.

For Obermeyer, an emergency physician whose research combines artificial intelligence, medicine, and health policy, sudden cardiac death remains one of medicine's most frustrating unsolved problems.

Because victims often die without warning, researchers have limited opportunities to observe what happens inside the heart immediately before it stops. An autopsy may uncover blocked blood vessels, hardened tissue, or other structural abnormalities, but it cannot fully reveal how the heart was functioning in the moments leading up to death.

"One thing that makes the problem very tragic, but also very well suited for AI, is that we have the cure for this problem," Obermeyer said. "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."

The Problem With Current Heart Risk Screening

Doctors currently rely heavily on a measurement of how effectively the heart pushes blood out with each beat. Patients whose hearts pump below a specified threshold may be considered candidates for an implantable cardioverter defibrillator, a device that can deliver an electrical shock when a dangerous heart rhythm develops.

But obtaining this measurement generally requires a more extensive medical evaluation. Many people who eventually experience sudden cardiac arrest never receive such testing because they have no obvious reason to suspect they are in danger.

The approach also has limitations for patients who are identified as high risk. About two-thirds of implanted defibrillators in these patients never need to deliver a shock. As a result, many individuals undergo expensive, invasive procedures for potentially fatal emergencies that may never occur.

At the same time, thousands of people who are not recognized as being at risk die unexpectedly each year.

"In some fraction of those people, we could have prevented those deaths if we had just known it in time," Obermeyer said. "There are a lot of lives being lost from people who are dropping dead of sudden cardiac death that are preventable if we just had better AI tools to find these things."

The researchers hope that identifying more accurate warning signs could help doctors direct preventive treatments toward the patients most likely to benefit.

How Researchers Trained AI to Recognize Deadly Heart Patterns

Developing the algorithm required three independent collections of medical data gathered from different parts of the world.

The researchers began with six years of EKG recordings from Sweden's unified healthcare system. By linking these scans to death certificates, they could examine the electrical patterns of patients who later experienced sudden cardiac death and compare them with those of other patients.

The AI learned to identify combinations of spikes, waves, and other characteristics in the recordings that appeared to be associated with increased risk. Once the system had been trained, the team tested it against two additional datasets: two years of anonymized EKG records from a San Diego hospital system and a separate collection of patient data from Taipei.

This international testing helped researchers evaluate whether the algorithm's predictive abilities extended beyond the Swedish population used to develop it.

"Good AI starts with good data," said Obermeyer, who is also part of the joint UCSF-UC Berkeley Computational Precision Health program. "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."

Obtaining and organizing the information was itself a massive undertaking. According to Obermeyer, assembling the datasets took approximately 10 years and involved contributions from two organizations he helped establish, Dandelion Health and Nightingale Open Science. Both organizations work to advance medical research through better access to data and artificial intelligence.

"It's one of the tremendous luxuries of being at Berkeley," Obermeyer said. "You can pursue these projects that just are a slow burn and then have -- hopefully -- a big return when they get over the finish line."

From Hidden Heart Signals to Lifesaving Treatments

The researchers have already begun exploring how the system might be used in actual healthcare settings.

Obermeyer is collaborating with medical systems in Sweden, Taiwan, and the United States to introduce the algorithm into hospital EKG databases. The goal is to examine existing heart scans for signs of elevated risk and identify patients who may benefit from additional monitoring.

Under the proposed approach, physicians would contact patients whose EKGs receive a high-risk assessment and offer them the opportunity to wear a patch that continuously records their heart's electrical activity.

These longer recordings could provide valuable clues about the biological processes responsible for the warning signals detected by the AI. Understanding those processes could help researchers explain why certain hearts are more vulnerable to suddenly developing fatal electrical disturbances.

For some patients, the additional information might ultimately support a recommendation to receive an implanted defibrillator, potentially preventing a sudden and otherwise unexpected death.

Obermeyer has also created a website where people interested in learning about their own risk can provide basic personal information and an email address. The research team can then contact them about EKG analysis when the technology becomes more widely available.

Could AI Reveal New Secrets About the Human Heart?

Beyond its potential to improve the prevention of sudden cardiac death, the discovery highlights how artificial intelligence could help scientists investigate medical mysteries that have resisted conventional approaches.

Rather than simply identifying patients who might need treatment, the algorithm has uncovered patterns in heart activity that researchers do not yet fully understand. Studying those signals could lead to new insights into the electrical disturbances responsible for sudden cardiac death and potentially reveal additional opportunities for prevention.

Obermeyer believes this ability to uncover previously unrecognized biological patterns could become one of AI's most important contributions to medicine, opening new directions for research while helping physicians make better decisions.

"There is also going to be a new way of doing science that comes out of these tools," Obermeyer said, "and it's fun to think about how that starts happening."