Machine learning-based prediction of postoperative venous thromboembolism in orthopedic surgery using the MIMIC-IV database and external validation

Abstract

Objective A risk prediction model for venous thromboembolism (VTE) after orthopedic surgery was constructed and validated using machine learning algorithms, providing a reference for the early identification of high-risk individuals in clinical settings. Method Patients who underwent orthopedic surgery were retrospectively collected from the Medical Information Mart for Intensive Care (MIMIC-IV) database as the modeling group. The patients were randomly divided into training and internal test sets in a 7:3 ratio. A total of 156 patients who underwent orthopedic surgery in our hospital from January 2022 to July 2025 were selected for the external validation set. Predictive variables were selected using the least absolute shrinkage and selection operator (LASSO) regression. Six machine learning models were constructed: logistic regression (LR), decision tree (DT), random forest (RF), extreme gradient boosting (XGB), light gradient boosting machine (LGBM), and naive Bayes (NBM). The model performance was evaluated by using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, and G-mean. The interpretability of the optimal model was analyzed using the calibration curves, decision curve analysis (DCA), and SHAP methods. Results The MIMIC-IV database included 461 patients (VTE incidence rate, 15.8%). LASSO regression identified two predictive variables: fibrinogen and D-dimer levels. The AUC of the LGBM model in the internal test set was 0.893 (95% CI: 0.820–0.966), and in the external validation, the AUC of the LGBM model was 0.788. The predictive performance of this model was superior to that of the other five models tested. The calibration curve indicated that the model was well calibrated, and the decision curve suggested a clinical net benefit within the 0.1–0.6 threshold range. SHAP analysis showed that elevated fibrinogen and D-dimer levels were positively correlated with the risk of VTE. Conclusion The LGBM model constructed based on the MIMIC-IV database has excellent predictive efficacy for VTE after orthopedic surgery and certain external generalization abilities. Fibrinogen and D-dimer levels are key predictive factors. This model can serve as a preliminary reference tool for risk stratification of VTE after orthopedic surgery.