Machine Learning-Based Model for Predicting Preoperative Deep Vein Thrombosis in Patients with Peri-ankle Fractures

Objective To investigate the risk factors for preoperative deep vein thrombosis (DVT) in patients with peri-ankle fractures and to develop a machine learning-based prediction model for preoperative DVT risk. Methods A retrospective study was conducted involving 1000 patients with peri-ankle fractures. Predictors were selected using Lasso regression, and the dataset was randomly split into a training set and a validation set at a 7:3 ratio. Four models—Logistic Regression, Decision Tree, Random Forest, and XGBoost—were constructed and internally validated using the Bootstrap method. Model performance was compared using metrics including the area under the ROC curve (AUC), calibration curves, decision curve analysis, and the F1 score. The optimal model was interpreted using the SHAP method. Results Four predictors were identified: age, BMI, preoperative waiting time, and D-dimer level. The XGBoost model demonstrated the best performance, with an AUC of 0.827 in the validation set. SHAP analysis confirmed that higher values of these four features contributed positively to the model's predictions, aligning with clinical knowledge and ensuring a transparent and credible decision-making process. Conclusion The XGBoost-based prediction model exhibits favorable performance and interpretability. It delineates core risk factors to guide targeted nursing interventions and holds potential for translation into a clinical decision-support tool, offering a novel strategy for the precise prevention of preoperative DVT.