Forecasting a Crisis: Machine-Learning Models Predict Occurrence of Intraoperative Bradycardia Associated With...

Area of Science:

• Anesthesiology
• Medical Informatics
• Cardiovascular Physiology

Background:

• Bradycardia is a common, unpredictable intraoperative event that can lead to hypotension and require intervention.
• Predictive analytics offers a potential strategy to improve perioperative care by anticipating high-risk clinical events.
• Current methods lack robust prediction for intraoperative bradycardia, highlighting a need for advanced analytical tools.

Purpose of the Study:

• To develop and evaluate predictive models for clinically significant intraoperative bradycardia.
• To utilize preoperative electronic medical record and intraoperative anesthesia information management system data.
• To predict bradycardia events at three distinct time points during surgical procedures.

Main Methods:

• Analysis of 62,182 noncardiac procedures (2012-2017) with severe bradycardia defined as heart rate <50 bpm followed by hypotension (MAP <55 mm Hg) within 10 minutes.
• Development of machine-learning and logistic regression models using preoperative and real-time intraoperative data (patient vitals, ventilator, fluids, medications).
• Model performance evaluated using area under the ROC curve (AUC) at three time points: post-induction (TP1), procedure start (TP2), and 30 min post-procedure start (TP3).

Main Results:

• The incidence of severe bradycardia with hypotension was 5.6% (TP1), 3.9% (TP2), and 1.7% (TP3).
• Models achieved AUCs of 0.81 (TP1), 0.87 (TP2), and 0.89 (TP3), with heart rate and previous events being strong predictors.
• Key predictors included heart rate, previous events, pulse rates, and hemodynamic slopes, demonstrating model efficacy across different time intervals.

Conclusions:

• Predictive models effectively forecast unstable bradycardia using preoperative and real-time intraoperative data.
• The study demonstrates the utility of predictive models for anticipating clinical events at multiple intraoperative stages.
• Future development aims to integrate these models into real-time intraoperative decision support systems.