Machine learning-based prediction of gastroparesis risk following complete mesocolic excision

Gastroparesis is a major complication following complete mesocolic excision (CME) and significantly impacts patient outcomes. This study aimed to create a machine learning model to pinpoint key risk factors before, during, and after surgery, effectively predicting the risk of gastroparesis after CME.

The study involved 1146 patients with colon cancer, out of which 95 developed gastroparesis. Data were collected on 34 variables, including demographics, chronic conditions, pre-surgery test results, types of surgery, and intraoperative details. Four machine learning techniques were employed: extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and k-nearest neighbor (KNN). The evaluation involved k-fold cross-validation, receiver operating characteristic (ROC) analysis, calibration curves, decision curve analysis (DCA), and external validation.

XGBoost excelled in its performance for predictive models. ROC analysis showed high accuracy for XGBoost, with area under the curve (AUC) scores of 0.976 for the training set and 0.906 for the validation set. K-fold cross-validation confirmed the model's stability, and calibration curves indicated high predictive accuracy. Additionally, DCA highlighted XGBoost's superior patient benefits for intervention treatments. An AUC of 0.77 in external validation demonstrated XGBoost's strong generalization ability.

The XGBoost-fueled predictive model for post-surgery colon cancer patients proved highly effective. It underlined gastroparesis as a significant post-operative issue, associated with advanced age, prolonged surgeries, extensive intraoperative blood loss, surgical techniques, low serum protein levels, anemia, diabetes, and hypothyroidism.