摘要
目的:基于不同机器学习(ML)算法构建腹膜透析相关性腹膜炎(PDAP)风险预测模型,为识别高风险病人提供参考依据。方法:回顾性选取2009年12月—2024年5月在贵州省人民医院行规律腹膜透析的病人作为研究对象,按7∶3的比例随机分为训练集和验证集,在训练集中经LASSO回归筛选自变量,基于Logistic回归(LR)、决策树、支持向量机、随机森林(RF)、极端梯度提升和人工神经网络6种ML算法构建PDAP风险预测模型。基于受试者工作特征曲线下面积(AUC)、准确度、精确率、召回率、F1分数评估模型性能,选出最优模型。结果:共纳入982例腹膜透析病人,221例病人发生PDAP,发生率为22.51%。基于十折交叉验证的LASSO回归筛选出5个自变量后构建6种ML模型,在训练集中LR模型(AUC=0.800)相较于其他模型表现更好,在验证集表现最佳者为RF模型(AUC=0.772),LR模型在训练集上的AUC值较高,可能存在过拟合的情况。进一步基于RF模型对特征变量进行重要性排序,依次为透析龄、白细胞计数、接触性污染、导管出口处感染和/或隧道感染、便秘或腹泻。结论:基于RF算法构建的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病人PDAP的发生。
Objective:To construct risk prediction models for peritoneal dialysis⁃associated peritonitis(PDAP)based on different machine learning(ML)algorithms,providing a reference for identifying high⁃risk patients.Methods:Retrospectively collect patients who underwent peritoneal dialysis in Guizhou Provincial People's Hospital from December 2009 to May 2024.Randomly divide them into a training set and a validation set at a ratio of 7∶3.In the training set,independent variables were screened through Lasso regression.Risk prediction models for PDAP were constructed based on six machine⁃learning algorithms,namely Logistic regression(LR),decision tree,support vector machines,random forest(RF),extreme gradient boosting,and artificial neural network.The performance of the models was evaluated based on the area under the receiver operating characteristic curve(AUC),accuracy,precision,recall,and F1⁃score,and the optimal model was selected.Results:A total of 982 peritoneal dialysis patients were included,among whom 221 patients developed PDAP,with an incidence rate of 22.51%.After five independent variables were screened out by LASSO regression based on ten⁃fold cross⁃validation,six ML models were constructed.In the training set,LR(AUC=0.800)performed the best compared with other models.In the validation set,RF(AUC=0.772)had the best performance.The LR model had a relatively high AUC value in the training set,which might indicate over⁃fitting.Further,based on the RF model,the feature variables were ranked in terms of importance,in the order of dialysis vintage,white blood cell count,contact⁃line contamination,catheter exit⁃site infection and/or tunnel infection,and constipation or diarrhea.Conclusion:The PDAP risk prediction model constructed based on the RF algorithm has the optimal performance,which can assist clinical medical staff in early assessment and prevention of PDAP.
作者
杨芳
赵健秋
郄淑文
杨丽
YANG Fang;ZHAO Jianqiu;QIE Shuwen;YANG Li(Department of Nephrology,Guizhou Provincial People's Hospital,Guizhou 550002 China)
出处
《全科护理》
2025年第24期4626-4631,共6页
Chinese General Practice Nursing
基金
贵州省护理学会立项科研课题一般项目,编号:GZHLKY202405。
关键词
机器学习
腹膜透析相关性腹膜炎
风险预测模型
machine learning
peritoneal dialysis⁃associated peritonitis
risk prediction model
