Artificial Intelligence

Study: AI ECG Model Shown to Predict Sudden Cardiac Death Risk

Is this a breakthrough for predicting sudden cardiac death or a warning sign of unchecked medical data use?

Published JUN 29

Study: AI ECG Model Shown to Predict Sudden Cardiac Death Risk

Story

Above: Close-up of an ECG heart monitor displaying cardiac rhythm. Image credit: Randy Faris/Getty Images

The Facts

A UC Berkeley-led study published in Nature last Wednesday trained an AI model on more than 440,000 ECGs from Sweden, paired with death certificates and health records, to identify waveform patterns linked to sudden cardiac death.
Berkeley News
Nature
The AI model isolated a high-risk group with a 7% annual rate of sudden cardiac death, compared to a 4.6% annual rate identified by the standard clinical measure of left ventricular ejection fraction, with more than 80% of AI-flagged patients missed by the existing test.
FOX News
The Daily Californian
Researchers discovered a previously unrecognized ECG signal in a region called aVL, within the QRS complex, that strongly predicted sudden cardiac death and had not been previously described in medical literature.
Nature
FOX News
The study, led by UC Berkeley associate professor Ziad Obermeyer, also found that cardiac fibrosis — scar tissue in the heart — is commonly present in patients with the highest risk of sudden cardiac death.
Berkeley News
STAT
The model was validated on patient data from a San Diego health system and a Taiwanese hospital registry. Patients identified as high-risk are offered a wearable ECG patch to continuously monitor their heart activity.
FOX News
The Daily Californian
Obermeyer noted that compiling the study's data took roughly a decade, citing limited access to clinical records as a key barrier to AI development in medicine. Researchers are now deploying the algorithm on hospital ECG databases in Sweden, Taiwan and the U.S.
Berkeley News
FOX News

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Techno-optimist narrative

An AI model trained on over 440,000 ECGs can identify sudden cardiac death risk far better than standard clinical tests, flagging a high-risk group with a 7% annual death rate versus 4.6% under current methods. That gap represents thousands of preventable deaths every year among people who look perfectly healthy by today's standards. This breakthrough could finally tell doctors who actually needs an implantable defibrillator before it's too late.

Berkeley News

Techno-skeptic narrative

Powerful medical AI means nothing if patients can't trust how their data gets used — and it took a decade to compile the records behind this study, which shows just how murky the data pipeline really is. Hospitals, researchers and AI companies need clear guardrails before more health records get swept into training models. Better prediction can save lives, but trust will determine whether people ever accept these tools in the first place.

FOX News