AI finds hidden ECG signal that predicts sudden cardiac death

Researchers at UC Berkeley trained an AI on 440,000 Swedish EKGs and identified subtle electrical patterns linked to sudden cardiac death. The model flagged a group with a 7% annual death rate versus 4.6% for patients deemed high risk by standard ejection-fraction screening, according to a Nature study.

• Model trained on 440,000 EKGs collected over six years in Sweden
• AI-flagged group had 7% annual sudden death rate vs 4.6% for standard screening
• Many flagged patients would not qualify as high risk under current criteria
• About 90% of out-of-hospital cardiac arrests are fatal