Overview
- The study published June 24 in Nature trained a deep‑learning model on more than 440,000 Swedish ECGs linked to death records and validated the tool on independent datasets from a San Diego hospital system and Taipei.
- The algorithm detected a previously unrecognized ECG signal and isolated a high‑risk group with about a 7% annual sudden cardiac death rate compared with 4.6% identified by standard clinical screening.
- Researchers have begun running the model in health systems in Sweden, Taiwan and the United States and plan to follow flagged patients with continuous wearable monitoring to confirm the signal and gather real‑time data.
- If prospectively validated, the tool could expand who is offered implantable defibrillators or targeted monitoring because current methods miss many people who later suffer sudden cardiac death and many implants never fire.
- Building the datasets took roughly a decade and the team now plans physiological study, prospective trials and regulatory review while also managing data access and privacy as the work moves toward clinical use.