基于机器学习的腹膜透析患者早发性腹膜透析相关性腹膜炎风险预测模型的构建  

Construction of a risk prediction model for early-onset peritoneal dialysis-associated peritonitis in peritoneal dialysis patients based on machine learning

作　　者：杨芳 郄淑文[2] 杨丽[2] 赵健秋[2] 柏晓玲 栗欢 Yang Fang;Qie Shuwen;Yang Li;Zhao Jianqiu;Bai Xiaoling;Li Huan(School of Nursing,Guizhou University of Traditional Chinese Medicine,Guiyang 550000,China;Department of Nephrology,Guizhou Provincial People's Hospital,Guiyang 550000,China;School of Nursing,Guizhou Nursing Vocational College,Guiyang 550000,China;Department of Infection Management,Guizhou Provincial People's Hospital,Guiyang 550000,China)

机构地区：[1]贵州中医药大学护理学院,贵阳550000 [2]贵州省人民医院肾内科,贵阳550000 [3]贵州护理职业技术学院护理系,贵阳550000 [4]贵州省人民医院医院感染管理科,贵阳550000

出　　处：《中华现代护理杂志》2025年第6期778-783,共6页Chinese Journal of Modern Nursing

基　　金：2023年度贵州护理职业技术学院院级课题(gzhlyj2023-02)。

摘　　要：目的基于6种机器学习算法构建腹膜透析患者早发性腹膜透析相关性腹膜炎(PDAP)风险预测模型。方法采用回顾性研究方法。应用便利抽样法,选取2009年12月—2023年8月在贵州省人民医院肾内科规律随访的腹膜透析患者为研究对象,收集研究对象的一般资料、原发疾病和实验室指标。按7∶3比例随机分为建模集和验证集,以是否发生早发性PDAP为因变量,分别基于Logistic回归、决策树、支持向量机、随机森林、极端梯度提升和人工神经网络6种机器学习算法构建腹膜透析患者早发性PDAP风险预测模型。采用受试者工作特征曲线下面积(AUC)、准确度、F1分数评估模型性能,选出最优模型。结果最终分析890例腹膜透析患者数据,其中86例患者发生早发性PDAP,早发性PDAP的发生率为9.66%。Logistic回归、支持向量机、极端梯度提升和随机森林4种预测模型具有较高的准确性,AUC值分别为0.703、0.729、0.782和0.814,其中随机森林模型的AUC值、准确度、F1分数均较高。进一步基于随机森林模型对早发性PDAP的风险因素重要性进行排序,结果显示,排名前5位的特征变量依次为C反应蛋白、甘油三酯、血小板、铁蛋白和白细胞。结论基于随机森林模型构建的腹膜透析患者早发性PDAP风险预测模型性能最优,有助于临床医护人员早期评估和预防早发性PDAP。ObjectiveTo construct the risk prediction model for early-onset peritoneal dialysis-associated peritonitis(PDAP)in peritoneal dialysis patients based on six machine learning algorithms.MethodsThis study was retrospective.Convenience sampling was used to select peritoneal dialysis patients who were regularly followed up in the Department of Nephrology of Guizhou Provincial People's Hospital from December 2009 to August 2023 to collect general information,primary diseases,and laboratory indicators of the study population.It was randomly divided into a modeling set and validation set in the ratio of 7∶3.With the occurrence of early-onset PDAP as the dependent variable,the risk prediction model of early-onset PDAP in peritoneal dialysis patients was constructed based on six machine learning algorithms,namely,Logistic regression,decision tree,support vector machine,random forest,extreme gradient boosting,and artificial neural network,respectively.Model performance was evaluated based on the area under the receiver operating characteristic curve(AUC),accuracy,and F1 score to select the optimal model.ResultsThe final data of 890 peritoneal dialysis patients were analyzed,of which 86 patients developed early-onset PDAP,and the incidence of early-onset PDAP was 9.66%.The four prediction models,Logistic regression,support vector machine,extreme gradient boosting,and random forest,had high accuracy with AUC values of 0.703,0.729,0.782,and 0.814,respectively,with the random forest model having higher AUC value,accuracy,and F1 score.Further ranking of the importance of risk factors for early-onset PDAP based on the random forest model showed that the top five characteristic variables were C-reactive protein,triglycerides,platelet,ferritin,and leukocyte,in that order.ConclusionsThe risk prediction model for early-onset PDAP in peritoneal dialysis patients constructed based on the random forest model has optimal performance,which can help medical and nursing staff assess and prevent early-onset PDAP at an early stage.

关 键 词：腹膜透析 机器学习 腹膜炎 预测 

分 类 号：R73[医药卫生—肿瘤]