Machine Learning Predicts Anastomotic Leakage in Colorectal Surgery

A Swiss pilot study tested machine learning models to predict anastomotic leakage (AL) in colorectal surgery using preoperative data from four hospitals. The Random Forest model achieved an AUC-ROC of 0.78 internally and 0.60 externally, with an accuracy of 0.82 and 0.87, respectively. Logistic Regression showed consistent accuracy (0.81 internal, 0.88 external) but lower AUC-ROC. Key predictors included surgical procedure, emergency surgery, renal function, and smoking. The study underscores the necessity for multicenter data and external validation for broader clinical application.

For more details, visit: https://doi.org/10.1007/s00464-024-10926-4