Analysis of the risk factors for catheter-related thrombosis in upper arm infusion port and construction of machine-learning prediction model
10.3969/j.issn.1008-794X.2025.03.005
- VernacularTitle:上臂输液港并发导管相关血栓形成危险因素分析及机器学习预测模型构建
- Author:
Mengsu ZHANG
1
;
Jie ZHANG
;
Guangxin JIN
;
Xiaoxia QIU
;
Xuebin ZHANG
;
Jun BU
Author Information
1. 200127 上海 上海交通大学医学院附属仁济医院心内科
- Keywords:
upper arm infusion port;
catheter-related thrombosis;
machine-learning;
prediction model
- From:
Journal of Interventional Radiology
2025;34(3):253-260
- CountryChina
- Language:Chinese
-
Abstract:
Objective To analyze the risk factors for catheter-related thrombosis(CRT)in the upper arm infusion port(UAP)and to construct a machine-learning prediction model.Methods A total of 6028 patients,who received UAP implantation at Shanghai Renji Hospital of China from February 2014 to February 2023,were enrolled in this study.The patients were divided into training set(n=4 219)and validation set(n=1 809).Six machine-learning prediction models,including Least Absolute Shrinkage and Selection Operator(LASSO)regression,random forest,decision tree,neural network,XGBoost and logistic,were constructed,and the model having best performance was selected as the optimal model.SHapely Additive exPlanations(SHAP)analysis was used to explain the neural network model,and DALEXtra package was used to explain the continuous variables.Results The neural network model was chosen as the final model.The variables,in order of the degree of importance from high to low,included sex,the diameter of catheter,catheter tip confirmation method,the length of catheter,inpatient or outpatient status,history of central venous catheter implantation,the length of subcutaneous tunnel,age,body mass index(BMI),primary tip displacement,and left or right venous approach.The learning curve,i.e.the area under curve(AUC)of the receiver operating characteristic(ROC)curve,for the training set was＞0.6,and the Delong testing and Bootstrap Methods Test showed that the neural network model performed well(P＜0.05).The Kolmogorov-Smirnov plot(KS plot)value was 0.313 5,indicating that the model had the good ability of discrimination.The clinical impact curve(CIC)assessment revealed that the model had good clinical value.Conclusion The machine-learning prediction model of upper arm infusion port with CRT has been successfully constructed.For minimizing the risk of CRT,it is recommended to prioritize the use of 5 F diameter catheters,adopt left-sided venous approach and positioning the tip of the catheter based on anatomical measurements,besides,the catheter length should be not shorter than 36.56 cm,and the subcutaneous tunnel length should not be less than 5 cm.The basic features associated with higher CRT risk include age of 50-65 years,BMI being between 18.69 kg/m2 and 20.81 kg/m2 or between 23.68 kg/m2 and 23.94 kg/m2 and male.
