Construction and performance verification of a predictive model for PICC related blood flow infection in tumor patients based on machine learning algorithms
LI Xiaoling
YAN Xiaoxia
YANG Ning
LI Xiaohong
Abstract：Objective: To explore the construction and validation of a predictive model for PICC-related bloodstream infections in cancer patients. Methods: A total of 2,608 cancer patients admitted between April 2021 and March 2023 were selected as the study subjects. All patients underwent PICC catheter placement during hospitalization and were divided into two groups based on whether they developed PICC-related bloodstream infections. Machine learning algorithms, including Support Vector Machine (SVM), XGBoost, and Logistic Regression, were used to construct predictive models for the occurrence of PICC-related bloodstream infections. The models were compared to identify the algorithm with the best predictive performance. Results: There were statistically significant differences between the two groups in platelet count, PICC dwell time, catheter dwell time, type of wound dressing, presence of catheter displacement, and the number of punctures per insertion (P < 0.05). In the XGBoost model, platelet count had the greatest impact on the model, followed by D-dimer levels. In the SVM model, the SHAP value of platelet count was the highest, indicating its most significant influence on the model. In the Logistic Regression model, five factors were finally included: PICC dwell time, catheter dwell time, wound dressing type, presence of catheter displacement, and the number of punctures per insertion. The XGBoost model showed the highest discriminative ability. Conclusion: There are differences in multiple indicators between patients who developed and those who did not develop PICC-related bloodstream infections, and the influencing factors vary across different models. The predictive models constructed using SVM and XGBoost demonstrated higher sensitivity and accuracy compared to the Logistic Regression model, enabling relatively accurate assessment and prediction of the risk of PICC-related bloodstream infections. These models can effectively reduce the incidence of PICC-related bloodstream infections in clinical practice.
Keywords：machine learning algorithmstumor patientsperipherally inserted central catheterPICCcatheter-related blood stream infectionprediction model
Publication Date：2025-04-10
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 1336-1342 )