Researchers at UC Berkeley have used artificial intelligence trained on more than 440,000 electrocardiograms from Sweden to identify a previously hidden electrical pattern in heart scans that predicts sudden cardiac death risk.

The AI system compared EKG recordings with death certificates to learn which electrical patterns distinguished healthy individuals from those who eventually died from sudden cardiac causes. The team then validated the technology using patient records from San Diego hospitals and Taipei.

The algorithm identified patients with a 7% annual sudden cardiac death rate, compared to 4.6% using conventional screening methods that measure how much blood the heart pumps with each contraction. The AI also recognized a broader population at elevated risk who would not normally trigger concern under existing medical guidelines.

Standard screening relies heavily on a single measurement of heart function. About two-thirds of implanted defibrillators in high-risk patients never deliver a shock, meaning many people undergo expensive procedures unnecessarily. Meanwhile, thousands of people not identified as at-risk die unexpectedly each year.

The major advantage of the new approach is that it works with ordinary EKG images already routinely collected in hospitals. This means the technology could improve screening without requiring new medical tests.

"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."

The findings were published in Nature. Obermeyer noted that assembling the datasets took approximately 10 years, involving contributions from Dandelion Health and Nightingale Open Science, organizations he helped establish.