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

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Percutaneous insertion of central venous catheters (PICCs) are essential for sepsis patients needing prolonged intravenous therapy.
• PICC use is associated with significant complications, notably thrombosis, impacting patient outcomes.
• Accurate risk identification for PICC-related thrombosis is crucial for effective clinical management.

Purpose of the Study:

• To develop and validate a predictive model for PICC-related thrombosis in sepsis patients.
• To utilize the XGBoost algorithm for enhanced predictive accuracy.
• To identify key risk factors contributing to PICC-related thrombosis.

Main Methods:

• Analysis of a large dataset (n=8,128) of sepsis patients with PICCs from the MIMIC-IV 3.1 database.
• Development of an XGBoost predictive model using demographic, laboratory, and clinical variables.
• Model validation using area under the receiver operating characteristic curve (AUC), SHAP analysis, and decision curve analysis.

Main Results:

• The XGBoost model demonstrated strong predictive performance with AUCs of 0.761 (training) and 0.766 (validation).
• SHAP analysis identified key predictors including white blood cell count, platelet count, hemoglobin, creatinine, PICC indwelling time, and age.
• Decision curve analysis confirmed the model's clinical utility, outperforming standard strategies.

Conclusions:

• The developed XGBoost model is a reliable predictor of PICC-related thrombosis in sepsis patients.
• The model's identified risk factors provide insights for targeted clinical interventions.
• This predictive tool has the potential to guide clinical decision-making and improve outcomes for high-risk patients.