Machine learning-based risk prediction of postoperative deep vein thrombosis in Chinese patients undergoing gastrointestinal surgery.

Background: Deep vein thrombosis (DVT) is a common and potentially life-threatening complication after gastrointestinal surgery. Traditional risk assessment tools rely on static variables and may not effectively capture dynamic perioperative changes. Methods: Clinical data from 596 Chinese patients undergoing gastrointestinal surgery were retrospectively collected. Patients were randomly divided into training and validation sets (7:3 ratio). Five machine learning algorithms-logistic regression (LR), Extreme Gradient Boosting (XGBoost), multilayer perceptron (MLP), random forest (RF), and elastic net (ENet)-were applied to identify key predictive features and build risk prediction models. The optimal model was visualized using a nomogram and validated through calibration curves, receiver operating characteristic (ROC) curves, and decision curve analysis (DCA). Results: Among the five models, the RF model achieved the best predictive performance. Postoperative Day-7 D-dimer, Day-1 D-dimer, and Day-5 D-dimer were identified as the most important predictive features. The calibration curve and DCA further confirmed the nomogram's predictive accuracy and clinical utility. Conclusion: We developed a novel machine learning-based model for predicting postoperative DVT in Chinese patients after gastrointestinal surgery. Integrating dynamic biomarkers and nonlinear modeling, the tool enhances early identification of high-risk individuals. Multicenter validation is warranted to further strengthen the model's applicability.