Using machine learning algorithms to predict risk factors of heart failure after complete mesocolic excision in...

Following complete mesocolic excision (CME), heart failure (HF) emerges as a significant complication, exerting substantial impacts on both short-term and long-term patient prognoses. The primary objective of our investigation was to develop a machine learning model capable of discerning preoperative and intraoperative high-risk factors, facilitating the prediction of HF occurrence subsequent to CME. A cohort comprising 1158 patients diagnosed with colon cancer was enrolled in our study, encompassing 172 individuals who developed postoperative HF. We compiled 37 feature variables, spanning patient demographic traits, foundational medical histories, preoperative examination characteristics, surgery types, and intraoperative details. Four distinct machine learning algorithms-extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and k-nearest neighbor algorithm (KNN)-were employed to construct the model. The k-fold cross-validation method, ROC curve, calibration curve, decision curve analysis (DCA), and external validation were employed for comprehensive model evaluation. The XGBoost algorithm exhibited superior performance compared to the other three prediction models. Specifically, within the training set of the internal validation set, the XGBoost algorithm demonstrated an AUC value of 0.990 (0.983-0.996), accuracy of 0.929 (0.923-0.935), sensitivity of 0.983 (0.976-0.990), specificity of 0.918 (0.910-0.927), and F1 value of 0.799 (0.784-0.814). In the validation set of the internal validation set, the XGBoost algorithm recorded an AUC value of 0.941 (0.890-0.991), accuracy of 0.897 (0.882-0.911), sensitivity of 0.898 (0.860-0.937), specificity of 0.875 (0.845-0.905), and F1 value of 0.711 (0.656-0.766). The AUC value for the external validation set was 0.93, indicating robust extrapolative capabilities of the XGBoost prediction model. The HF prediction model post-CME, derived from the XGBoost machine learning algorithm in this study, attests to its elevated predictive accuracy and clinical utility.