Clin Interv Aging. 2026 Jun 25;21:614064. doi: 10.2147/CIA.S614064. eCollection 2026.
ABSTRACT
PURPOSE: Hip fractures are often associated with deep vein thrombosis (DVT). This study aimed to develop machine learning models to predict preoperative DVT risk using basic clinical data from elderly patients with hip fractures.
METHODS: Clinical data were retrospectively collected from 538 elderly hip fracture patients hospitalized at Beijing Shijitan Hospital. Eighteen clinical parameters were assessed. Patients from October 2021 to September 2024 (n=405) formed the development cohort, randomly divided into training and test sets (7:3). Patients from October 2024 to September 2025 (n=133) constituted the temporal validation set. Following Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection, three models-logistic regression, light-gradient boosting machine, and support vector machine (SVM)-were developed. Performance was assessed using the area under the receiver operating characteristic curve (AUC), sensitivity, Brier score, calibration curves, and decision curve analysis (DCA). SHapley additive explanations (SHAP) were used for interpretability.
RESULTS: LASSO selected seven features: fracture type, time from injury to admission, white blood cell count, red blood cell count, C-reactive protein, D-dimer, and prothrombin time. The SVM model demonstrated the best overall performance. In the test set, it achieved an AUC of 0.8525 (95% CI: 0.7624-0.9426), sensitivity of 0.7200, and Brier score of 0.1231. Performance remained stable in the temporal validation set with an AUC of 0.8360 (95% CI: 0.7470-0.9260), sensitivity of 0.6562, and Brier score of 0.1328. Calibration curves and DCA indicated reliable probability predictions and clinical net benefit. SHAP identified prothrombin time as the most important predictor. A web-based calculator was developed based on the SVM model.
CONCLUSION: The developed SVM model shows potential as a dynamic, interpretable risk assessment tool for preoperative DVT in elderly hip fracture patients. It may provide a helpful quantitative reference to assist clinicians with perioperative DVT monitoring and early warning.
PMID:42376184 | PMC:PMC13313010 | DOI:10.2147/CIA.S614064