A predictive model for PICC-related thrombosis in sepsis patients using XGBoost algorithm

Percutaneous insertion of central venous catheters (PICC) is critical for the management of sepsis patients requiring prolonged intravenous therapy; however, it poses significant complications, including thrombosis. Identifying risk factors for PICC-related thrombosis can enhance clinical management and patient outcomes. This study aimed to develop a predictive model for PICC-related thrombosis in sepsis patients using the XGBoost algorithm. We analyzed data from 8,128 ICU patients diagnosed with sepsis and using PICC from the Medical Information Mart for Intensive Care IV version 3.1 (MIMIC-IV 3.1) database. Patients were divided into a training set (70%, n = 5690) and a validation set (30%, n = 2438). Variables included demographic, laboratory, and clinical factors potentially associated with PICC-related thrombosis. An XGBoost model was developed and validated, with performance assessed using the area under the receiver operating characteristic curve (AUC) and SHAP analysis for interpretability. Decision curve analysis confirmed the clinical utility of the model. The XGBoost model achieved an AUC of 0.761 (95% CI 0.734-0.787) in the training set and 0.766 (95% CI 0.731-0.801) in the validation set. The calibration curve demonstrated good calibration of the model, indicating that predicted probabilities of thrombosis closely aligned with observed outcome. Decision curve analysis confirmed clinical utility, yielding a net benefit of 0.31 at 20% risk threshold, outperforming treat-all/none strategies. Key predictors, including white blood cell count, platelet count, history of myocardial infarction, hemoglobin levels, creatinine levels, PICC indwelling time, age, presence of mild liver disease and prothrombin time (PT), and diabetes without chronic complications, were identified using SHAP analysis in the XGBoost predictive model for PICC-related thrombosis, with the top ten predictors significantly contributing to the model's performance. The XGBoost model is an effective predictor of PICC-related thrombosis among sepsis patients, indicating its potential role in guiding clinical decision-making for the management of high-risk patients.