Improved Prediction of Surgical-Site Infection After Colorectal Surgery Using Machine Learning.

Area of Science:

• Computational medicine
• Surgical outcomes research
• Machine learning applications in healthcare

Background:

• Surgical-site infection (SSI) is a significant cause of morbidity following colorectal surgery.
• Previous predictive models for SSI have demonstrated limited accuracy.
• Machine learning (ML) shows potential in identifying complex patterns for improved postoperative outcome prediction.

Purpose of the Study:

• To develop a more accurate predictive model for colorectal surgical-site infections using machine learning techniques.
• To compare the performance of ML models against traditional logistic regression in predicting SSI.

Main Methods:

• Utilized a national, multicenter dataset of 275,152 patients undergoing colorectal surgery (2012-2019) from the American College of Surgeons National Quality Improvement Program database.
• Employed machine learning algorithms including random forest, gradient boosting, and artificial neural networks (ANN).
• Developed a logistic regression model for comparative analysis; model performance was assessed using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The overall incidence of surgical-site infection was 10.7%.
• Artificial neural network (ANN) achieved the highest predictive accuracy with an AUC of 0.769, surpassing gradient boosting (0.766), random forest (0.764), and logistic regression (0.677).
• Key predictors identified by the ANN model included organ-space SSI at surgery, operative time, oral antibiotic bowel preparation, and surgical approach.

Conclusions:

• Machine learning techniques, particularly ANNs, offer superior accuracy in predicting colorectal surgical-site infections compared to logistic regression.
• These ML models can effectively identify patients at higher risk for SSI.
• The findings support the use of ML for targeted preventive interventions to reduce SSI incidence.