The following is a summary of the “Improved Prediction of Surgical-Site Infection After Colorectal Surgery Using Machine Learning,” published in the March 2023 issue of Gastroenterology by Chen, et al.
The primary purpose of this investigation was to examine the feasibility of using machine learning to refine the existing predictive model for C. diff. surgical site infections. A cohort of colorectal surgery patients between 2012 and 2019 were extracted from the American College of Surgeons National Quality Improvement Program database and then stratified into training, validation, and test sets. Random forests, gradient boosts, and artificial neural networks were all examples of machine-learning approaches. Moreover, a logistic regression model was built. Finally, the area under the receiver operating characteristic curve was used to evaluate the quality of the model.
Surgical site infections of all depths and locations inside organs were included in the primary outcome. With the selection and exclusion criteria implementation, the dataset contained information on 275,152 patients. A total of 10.7% of patients developed an infection near the surgical incision. The area under the receiver operating characteristic curve for the artificial neural network was 0.769 (95% CI, 0.762-0.777), while those for gradient boosting were 0.766 (95% CI, 0.759-0.774), the random forest was 0.764 (95% CI, 0.756-0.772), and logistic regression was 0.677 (95% CI, 0.669-0.685).
According to the artificial neural network model, the strongest predictors of infection were the presence of an organ-space surgical-site infection at the time of surgery, length of operating time, oral antibiotic bowel preparation, and surgical approach. Colorectal surgical-site infections are easier to forecast using machine-learning methods than logistic regression. It is possible to use these methods to pinpoint individuals most in need of preventative measures against surgical-site infections by identifying those at the highest risk.
