Constructing a risk prediction model for post-TKA DVT formation based on machine learning methods

To construct and compare the performance of risk prediction models for deep vein thrombosis (DVT) in the lower extremities after total knee arthroplasty (TKA) using logistic regression, XGBoost, Random Forest, AdaBoost, gradient boosting decision tree, and KNN models. The study also aims to explore the risk factors for DVT after TKA, providing a reference for evaluating and preventing DVT post-TKA. A retrospective collection of medical records from 3058 patients who underwent TKA at the Second Hospital of Shanxi Medical University from July 2020 to July 2023 was conducted. Based on inclusion and exclusion criteria, 1238 patient cases were selected for the study. Lasso regression and Boruta feature selection methods were used to identify risk factors associated with post-TKA DVT. Using R 4.2.3 statistical software, the dataset was randomly divided into training and validation sets. Six machine learning models were established, and clinical decision curves, calibration curves, and receiver operating characteristic curve curve metrics were used to evaluate model performance. The models were then compared to select the best predictive model, which was further visualized using the SHAP method. Eleven independent variables, including anemia, blood transfusion volume, blood loss, HCT, D- dimer, thrombin time (TT), CL, anesthesia duration, surgery duration, activated partial thromboplastin time (APTT), and postoperative pain scores, were identified as characteristic factors and included in the 6 machine learning models. The optimal predictive model was found to be the Logistic model. According to SHAP values, shorter APTT, longer anesthesia duration, elevated TT, higher postoperative pain scores, reduced D-dimer, decreased CL, increased blood transfusion volume, increased blood loss, shorter surgery duration, and anemia increase the risk of DVT in the lower extremities post-TKA. In constructing a model to predict the risk of deep vein thrombosis in the lower extremities after artificial knee joint replacement surgery, the Logistic model demonstrated outstanding performance and excellent generalization ability. This model can identify risk factors for deep vein thrombosis complications after total knee arthroplasty, enabling clinicians to manage patients with these risk factors more comprehensively and meticulously, thereby potentially reducing the incidence of deep vein thrombosis in the lower extremities after total knee arthroplasty.