Prediction model for postoperative atrial fibrillation in non-cardiac surgery using machine learning

Area of Science:

• Cardiology
• Medical Informatics
• Surgical Outcomes

Background:

• Postoperative atrial fibrillation (POAF) is a common complication after non-cardiac surgery.
• Identifying patients requiring treatment for POAF is crucial for clinical management.
• Existing predictive tools may lack accuracy or generalizability.

Purpose of the Study:

• To develop and validate a machine learning-based predictive model for POAF requiring treatment in patients undergoing non-cardiac surgery.
• To identify key predictors of POAF in this patient population.
• To provide a clinically useful tool for risk stratification.

Main Methods:

• Retrospective analysis of 201,864 patients undergoing non-cardiac surgery.
• Application of machine learning, specifically an extreme gradient boosting algorithm.
• Development of a predictive model using the top five identified variables and external validation.

Main Results:

• The model identified age, lung operation, operation duration, coronary artery disease history, and hypertension as key predictors.
• The model achieved an Area Under the Receiver Operating Characteristic (AUROC) curve of 0.80 in both internal and external validation.
• High accuracy (0.95) with excellent specificity (0.97) was observed.

Conclusions:

• A robust machine learning model effectively predicts POAF requiring treatment after non-cardiac surgery.
• The model demonstrates good performance and generalizability through external validation.
• The developed tool offers potential for improved clinical decision-making and patient management, though further verification is recommended.