Intraoperative Features Improve Model Risk Predictions After Coronary Artery Bypass Grafting.

Area of Science:

• Cardiothoracic Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Current risk models for adverse postoperative events after coronary artery bypass grafting (CABG) do not incorporate intraoperative physiologic parameters.
• Intraoperative data holds potential predictive value for identifying patients at higher risk of complications.

Purpose of the Study:

• To evaluate if incorporating continuous intraoperative data enhances machine learning model predictions for various adverse outcomes following CABG.
• To assess improvements in predicting 30-day mortality, renal failure, reoperation, prolonged ventilation, and combined morbidity and mortality (MM).

Main Methods:

• Combined data from the Society of Thoracic Surgeons (STS) database with retrospective continuous intraoperative patient data.
• Developed logistic regression models using 5-fold cross-validation, incorporating intraoperative features and STS preoperative risk scores.

Main Results:

• Models integrating intraoperative features and STS risk scores showed improved predictive performance (AUC) for prolonged ventilation and MM compared to the STS Risk Calculator alone.
• Enhanced calibration was observed for prolonged ventilation and MM when intraoperative data was included, indicated by a lower Brier score.

Conclusions:

• The integration of time-series intraoperative data into risk models can significantly improve the prediction of adverse postoperative events.
• These enhanced models may facilitate earlier identification of high-risk patients, enabling closer postoperative monitoring and improved patient care.