基于机器学习算法的PICC相关性血栓风险预测模型比较研究  

Comparison of PICC-associated thrombosis risk prediction models based on machine learning algorithm

作　　者：汪淑华 程博 朱丽群[3] 曹松梅[3] 梁怡青 Wang Shuhua;Cheng Bo;Zhu Liqun;Cao Songmei;Liang Yiqing(Medical School,Jiangsu University,Zhenjiang 212000,China;School of Mathematical Sciences,Jiangsu University,Zhenjiang 212000,China;Nursing Department,Affiliated Hospital of Jiangsu University,Zhenjiang 212000,China)

机构地区：[1]江苏大学医学院,镇江212000 [2]江苏大学数学科学学院,镇江212000 [3]江苏大学附属医院护理部,镇江212000

出　　处：《中华现代护理杂志》2022年第16期2144-2151,共8页Chinese Journal of Modern Nursing

基　　金：江苏省医院协会医院管理创新研究课题(JSYGY-3-2019-360);镇江市软科学研究计划项目(RK2019029)。

摘　　要：目的基于机器学习算法构建3种不同的经外周静脉置入中心静脉导管(PICC)相关性血栓风险预测模型,并比较模型性能,为评估及预防PICC相关性血栓提供依据。方法基于最佳证据和专家函询形成PICC相关性血栓风险因素调查表。采取便利抽样法,选取2016年1月—2020年10月在江苏大学附属医院行PICC置管的626例患者为研究对象收集临床资料,基于机器学习算法,分别采用支持向量机(SVM)、XGBoost和Logistic回归方法构建3种不同的PICC相关性血栓风险预测模型,并进行评价和比较。模型评价指标包括马修斯相关系数(MCC)、F1分数、受试者工作特征曲线下面积(AUC)及Brier得分。结果共30个变量纳入研究,预测因子包括患者的人口学资料、患者病情、治疗因素、导管相关性因素4个方面。测试集上验证后的模型,在MCC、F1分数上,Logistic回归预测模型得分低于XGBoost、SVM预测模型;在AUC上,Logistic回归预测模型得分等于SVM,小于XGBoost;在Brier得分上,Logistic回归预测模型得分高于XGBoost、SVM预测模型。结论基于机器学习算法XGBoost、SVM预测模型性能在敏感性及准确性上优于传统Logistic回归模型。血栓预测因子有助于指导临床医务人员识别高风险患者,降低PICC相关性血栓发生率。Objective To build the three different risk prediction models for peripherally inserted central catheter(PICC)-associated thrombosis based on machine learning algorithm,and compare the performance of the models,so as to provide a basis for evaluating and preventing PICC-associated thrombosis.Methods The PICC-associated Thrombasis Risk Factor Questionnaire was developed based on the best evidence and expert consultation.From January 2016 to October 2020,convenience sampling was used to select 626 patients with PICC in the Affiliated Hospital of Jiangsu University as the research object to collect clinical data.Based on machine learning algorithms,Support Vector Machine(SVM),XGBoost and Logistic regression methods were used to construct three different PICC-associated thrombosis risk prediction models,which were evaluated and compared..Model evaluation indicators included Matthews correlation coefficient(MCC),F1 value,area under the receiver operating characteristic curve(AUC)and Brier score.Results A total of 30 variables were included,and the predictors included four aspects,namely,demographic data of patients,patient condition,treatment factors,and catheter-related factors.For the model verified on the test set,the Logistic regression prediction model had lower scores than the XGBoost and SVM prediction models in terms of MCC and F1 values.On AUC,the Logistic regression prediction model score was equal to SVM and smaller than XGBoost.On Brier,the Logistic regression prediction model scored higher than the XGBoost and SVM prediction models.Conclusions The performance of the prediction model based on the machine learning algorithm XGBoost and SVM is superior to the traditional Logistic regression model in terms of sensitivity and accuracy.Thrombotic predictors can help guide medical and nursing staff to identify high-risk patients and reduce the incidence of PICC-associated thrombosis.

关 键 词：导管插入术 中心静脉 XGBoost 支持向量机 LOGISTIC回归 机器学习 PICC相关性血栓 预测模型 

分 类 号：R473[医药卫生—护理学]