An ECG biomarker for sudden cardiac death, discovered with deep learning
Trained on every ECG in a Swedish region linked to death certificates, a deep learning model isolated 2.2% of patients with a 7.0% annual rate of sudden cardiac death, above the 4.6% of those flagged by reduced ejection fraction, and 86.1% of them were missed by that standard screen; a generative model then exposed the waveform feature it used, which the authors say had not been described.
- Model
- A deep learning model on 12-lead ECG waveforms, paired with a generative model of the waveform; not named
- Field
- Medicine
- Date
- 2026-06-24
- Human collaborators
- Ziad Obermeyer, Alexander Schubert, James Ross, Sendhil Mullainathan, Markus Lingman
Sources
Original work
Media coverage
Independent commentary
What was found
Defibrillators can prevent sudden cardiac death, but the only predictive marker in wide use, left ventricular ejection fraction, misses most such deaths and flags many low-risk patients for devices that never fire. The model's high-risk group (2.2% of the sample) had a 7.0% annual rate of sudden cardiac death against 4.6% for patients with reduced ejection fraction (1.9% of the sample), and high-risk ECG patients who had defibrillators implanted were 54.4% less likely to die than expected. The model was externally validated in a US health system, where it predicts the ventricular arrhythmias that cause sudden death, and in a Taiwanese hospital registry, where it predicts arrhythmic cardiac arrest. Pairing it with a generative model of the ECG waveform revealed the morphology it had learned, which the authors describe as easily visible, and they tie its shape to electrophysiological first principles to propose and preliminarily test a mechanism.
Novelty check
Europe PMC, 2026-11-19: the paper's DOI returns the Nature record, first published 24 June 2026, and a title search for sudden cardiac death and electrocardiogram from October to November 2026 returned only a European Heart Journal commentary on it. This registry carries several AI-ECG studies, such as 2021-05-06-eagle-ai-ecg-low-ejection-fraction and 2026-11-07-latent-ai-ecg-af-rct, but none for sudden cardiac death or a new waveform biomarker. It reached this registry through November 2026 ScienceDaily coverage, five months after publication, and is entered under its publication date.
Caveats and known objections
Retrospective: the model was trained and tested on records linked to death certificates, and the defibrillator benefit is an observational comparison against expected mortality, not a trial, so it may reflect who received devices. The external validations used other outcomes (ventricular arrhythmia in the US, arrhythmic arrest in Taiwan), not sudden cardiac death itself, and as far as the abstract shows they were run by the same author group on other institutions' data, so they are not counted as independent replications in this registry's sense. The European Heart Journal commentary asks whether the marker is ready for clinical use. The mechanism is described as preliminarily tested. Funding and data access were not read for this entry beyond the abstract. Autonomy is search-scaffold: the discovery came from a supervised model and a generative visualization inside a design the researchers built, matching how this registry grades neural-network screening finds such as 2020-02-20-halicin.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Peer reviewed and Search scaffold.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded peer reviewed for verification and search scaffold for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). An ECG biomarker for sudden cardiac death, discovered with deep learning. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-06-24-ecg-biomarker-sudden-cardiac-death
BibTeX
@misc{whataifound-ucberkeleywiththeuniversityofchicagomitandhalmstaduniversityandregionhalland-2026-death,
title = {An ECG biomarker for sudden cardiac death, discovered with deep learning},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by UC Berkeley, with the University of Chicago, MIT and Halmstad University and Region Halland. Verification: Peer reviewed. Autonomy: Search scaffold.},
url = {https://whataifound.org/finding/2026-06-24-ecg-biomarker-sudden-cardiac-death}
}