Multicentre validation of a machine learning model for predicting respiratory failure after noncardiac surgery

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Surgical Outcomes Research

Background:

• Postoperative respiratory failure is a significant complication following surgery.
• Early identification of patients at high risk is crucial for mitigating adverse outcomes.
• Prolonged mechanical ventilation or reintubation defines postoperative respiratory failure.

Purpose of the Study:

• To develop and validate a machine learning model for predicting postoperative respiratory failure.
• To identify easily extractable electronic health record (EHR) variables for risk prediction.
• To enable personalized risk stratification for surgical patients.

Main Methods:

• A gradient boosting machine learning algorithm was trained on EHR data from 99,025 noncardiac surgical cases.
• Easily extractable EHR variables, not requiring subjective clinical assessment, were utilized.
• The model was validated externally on three independent cohorts totaling 208,308 cases.

Main Results:

• The predictive model incorporated eight variables: serum albumin, age, anesthesia duration, serum glucose, prothrombin time, serum creatinine, white blood cell count, and BMI.
• Internal validation demonstrated high performance with an AUROC of 0.912 and AUPRC of 0.113.
• External validation across three cohorts showed robust performance with AUROCs ranging from 0.872 to 0.931.

Conclusions:

• A machine learning model using eight readily available EHR variables effectively predicts postoperative respiratory failure.
• The model exhibits strong predictive performance in both internal and external validation settings.
• This tool supports data-driven clinical decisions and personalized risk assessment for surgical patients.