Breaking

AI Detects Sudden Cardiac Death Risk from Routine ECG, Nature Reports

June 30, 2026 at 09:14 EDT

Healthcare
Research & Papers

A research team led by UC Berkeley has used deep learning to discover a new biomarker that detects sudden cardiac death (SCD) risk from a routine electrocardiogram (ECG), publishing the findings in Nature on June 24, 2026. The work shows it may be possible to identify high-risk patients—missed by the standard metric of left ventricular ejection fraction (LVEF)—using a test that is cheap and widely available.

June 24, 2026 · UC Berkeley · Nature

AI Reads a Routine ECG to Spot Hidden Sudden-Cardiac-Death Risk

A deep-learning model trained on 440,000+ ECGs found a new waveform biomarker in the aVL lead — flagging high-risk patients that the standard LVEF metric misses, using a test that is cheap and everywhere.

440K+

ECGs used for training (Sweden)

0.872

Patient-level AUC

54.4%

Suggested ICD mortality benefit in high-risk patients

Annual SCD rate: AI high-risk vs reduced-LVEF group

Same unit — % of group dying suddenly per year. Taller column = higher risk concentration.

7.0%

AI high-risk
2.2% of sample

4.6%

Reduced-LVEF
1.9% of sample

86.1%

of AI-flagged high-risk patients were NOT flagged by LVEF

These are the missed cases — patients with normal or never-measured ejection fraction who slip past the conventional standard.

300,000+

US deaths per year from sudden cardiac arrest — a leading cause of death

~2 in 3

ICD implants never fire — fueling concern over overuse alongside missed cases

From discovery to the clinic

Train

440K Swedish ECGs + new aVL biomarker

→

Validate

External US + Taiwan registries

→

Roll out

Hospital ECG databases; wearable patches ahead

PROMISE

Risk detection from a cheap, ubiquitous ECG
Catches cases LVEF misses; AI biomarker is visible
Could sharpen who truly needs an ICD

CAUTION

Death-certificate labels carry uncertainty
Large linked datasets are hard to access
Needs further validation before clinical use

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.