Particle.news
Download on the App Store

Johns Hopkins AI Uses Preop ECGs to Predict 30-Day Surgical Complications With 85% Accuracy

A retrospective study in the British Journal of Anaesthesia found a fusion model outperformed conventional risk scores.

Overview

  • Researchers analyzed 37,000 preoperative ECGs from patients who underwent surgery at Beth Israel Deaconess Medical Center in Boston.
  • Two models were developed, with the best performance coming from a fusion approach that combined ECG data with basic medical-record details such as age, sex, and existing conditions.
  • The models predicted which patients would have a heart attack, a stroke, or die within 30 days after surgery, exceeding the roughly 60% accuracy of current risk scores.
  • The team introduced an explainability method to highlight ECG features associated with adverse postoperative cardiovascular events.
  • Authors plan to test the model on additional datasets and in prospective studies, and reporting indicates the research received federal funding.