The LightGBM machine learning model demonstrated the highest accuracy in predicting preoperative deep vein thrombosis in elderly hypertensive patients with hip fractures, achieving an AUC of 0.910.
Observational (n=637)
No
Can machine learning models accurately predict preoperative DVT incidence in elderly hypertensive patients with hip fractures?
A LightGBM machine learning model accurately predicts preoperative DVT risk in elderly hypertensive patients with hip fractures, offering a practical tool for clinical risk assessment.
Hip fractures in the elderly present a significant public health challenge globally, especially among patients with hypertension, who are at an increased risk of developing preoperative deep vein thrombosis (DVT). DVT not only heightens surgical risks but also severely impacts the rehabilitation and quality of life of patients. Early risk assessment and management in this population are therefore critically important. This study aimed to develop and validate a machine learning-based predictive model to enhance the accuracy of predicting preoperative DVT in elderly patients with hypertension undergoing hip fracture surgery, thereby optimizing preoperative assessment and management. A retrospective study design was employed, selecting patients with hypertension and hip fractures treated at the First Hospital of Qinhuangdao from January 2018 to December 2022. Key predictive factors were identified using LASSO regression, and logistic regression was utilized to construct both a nomogram and an online interactive nomogram. Various machine learning algorithms were also employed to build predictive models. The contribution of variables in the models was explained using SHAP values, and model performance was evaluated through ROC curves, AUC values, and other statistical methods. The study included 637 patients, with LASSO regression selecting key variables that were further used to develop a logistic regression-based nomogram and its online version, providing intuitive tools for assessing DVT incidence. Among the multiple machine learning predictive models, the LightGBM model exhibited the best performance, achieving an AUC of 0.910. The model's effectiveness and reliability were confirmed through decision curves, calibration plots, and precision-recall curves. SHAP value analysis highlighted the significance of factors such as age, time from injury to hospital admission, atrial fibrillation, C-reactive protein, hypoalbuminemia, and D-dimer levels in the predictions, enhancing the model's transparency and interpretability. This study successfully developed a logistic regression-based nomogram and multiple machine learning algorithms to predict the risk of preoperative DVT in elderly hypertensive patients with hip fractures. The nomogram provides clinicians with a practical tool for rapid risk assessment, thus optimizing patient management and prognosis. The LightGBM model, recommended for its high predictive accuracy, along with SHAP value analysis, enhanced the transparency and clinical applicability of the models. These findings not only deepen our understanding of DVT incidence factors but also demonstrate the potential of machine learning technologies in enhancing medical decision-making and advancing precision medicine.
Ge et al. (Wed,) conducted a observational in Preoperative deep vein thrombosis in elderly hypertensive patients with hip fractures (n=637). LightGBM machine learning model vs. Other machine learning models (e.g., XGBoost, Random Forest, Logistic Regression) was evaluated on Area Under the Receiver Operating Characteristic Curve (AUC) for predicting preoperative DVT. The LightGBM machine learning model demonstrated the highest accuracy in predicting preoperative deep vein thrombosis in elderly hypertensive patients with hip fractures, achieving an AUC of 0.910.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: