Berkeley AI uncovers hidden EKG signal tied to sudden cardiac death risk
More than 300,000 people in the United States die each year from sudden cardiac arrest, when the heart’s electrical system fails and pumping stops. The emergency can hit older adults with known heart disease and seemingly healthy young athletes. Implanted defibrillators can prevent some deaths, but doctors have struggled to decide who needs them. Researchers at UC Berkeley, led by Ziad Obermeyer, an associate professor at the School of Public Health and an emergency physician, used artificial intelligence on more than 440,000 Swedish electrocardiograms (EKGs), matching electrical patterns with death-certificate data to tell apart healthy people, patients with known risk, and those who later died of sudden cardiac causes. The work was then checked on thousands of additional records from the United States and Taiwan.
Conventional screening mainly measures how much blood the heart pumps with each beat, a figure used to judge defibrillator eligibility. That approach flags a group with an annual sudden cardiac death rate of 4.6 percent. The AI identified a group with a 7 percent annual death rate and a wider set of people at elevated risk—thousands who would look relatively low-risk under existing guidelines. It reads ordinary EKG images already collected in hospitals worldwide, so screening could improve without a new test. The findings, published in Nature, may help decide who should receive an implanted defibrillator and raise the question of what cardiac changes produce the patterns the model detects.
Sudden cardiac arrest is not a heart attack: the latter is blocked blood flow; arrest is a failure of the signals that coordinate the heartbeat. CPR and a shock from an automated external defibrillator can restore rhythm, but roughly 90 percent of people who arrest outside hospital die within minutes. Because death often comes without warning, autopsies can show structural damage but not how the heart was functioning just before it stopped—the gap the Berkeley team is trying to close.
Continue in the app — vote & join in ➔