A machine learning approach to predicting early and late postoperative reintubation

Area of Science:

• Surgical outcomes research
• Medical informatics
• Machine learning in healthcare

Background:

• Accurate surgical risk estimation is vital for informed consent and shared decision-making.
• Postoperative reintubation (POR) is a significant complication impacting patient morbidity.
• Previous research has categorized POR into early (within 72 hours) and late (within 30 days) events.

Purpose of the Study:

• To develop and validate machine learning-based scoring systems for predicting combined, early, and late postoperative reintubation.
• To identify key pre and perioperative risk factors for POR using data from the ACS NSQIP database.
• To assess the predictive performance of logistic regression, random forest, and gradient boosting classification models for POR.

Main Methods:

• Utilized the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) database.
• Employed machine learning classification models including logistic regression, random forest, and gradient boosting.
• Developed predictive scoring systems by refining risk factors from an initial set of 37 pre and perioperative variables.

Main Results:

• Logistic regression models achieved strong performance in predicting POR outcomes, with an average Brier score of 0.172 and an average c-statistic of 0.852.
• The developed scoring systems showed predictive accuracy only marginally lower than using the full set of risk variables (average Brier score 0.145, average c-statistic 0.870).
• The models demonstrated robust discrimination across combined, early, and late POR predictions.

Conclusions:

• Machine learning models, particularly logistic regression, provide effective scoring systems for predicting postoperative reintubation.
• These validated scoring systems can be practically applied by surgeons and patients to enhance shared decision-making and improve surgical care quality.
• Further research is warranted to elucidate clinically significant distinctions between early and late POR events.