VTE risk prediction model for ICU patients with CVC placement combined with albumin use - XGBoost machine learning model based on SHAP interpretation

Venous thromboembolism (VTE) is a common and life-threatening complication in ICU patients, particularly those undergoing central venous catheterization (CVC) and receiving albumin therapy. Accurate assessment of VTE risk is essential for individualized treatment. This study aimed to develop and validate a VTE risk prediction model for ICU patients requiring CVC and albumin therapy, using the XGBoost machine learning algorithm. A total of 610 ICU patients were included, of whom 174 developed VTE (incidence 28.5%). LASSO regression was used for feature selection, and the XGBoost model demonstrated strong predictive performance in both training and testing sets, with AUC values of 0.97 and 0.94, sensitivities of 0.92 and 0.94, specificities of 0.91 and 0.82, accuracies of 0.91 and 0.85, and F1 scores of 0.84 and 0.76, respectively. SHAP analysis identified the most important predictors, ranked as follows: PCT, CVC retention time, FIB, CRP, age, length of stay, and Padua score. The model's high sensitivity (94%) and specificity (82%) support its clinical utility for proactive VTE prevention. SHAP visualizations enhanced interpretability by quantifying feature impacts. In conclusion, we developed an interpretable, high-performing XGBoost-based VTE risk prediction model for ICU patients undergoing CVC and albumin therapy. Rigorous validation confirmed robust discrimination, and the model offers a practical tool for individualized risk stratification. Further multicenter validation is warranted to generalize these findings.