摘要
目的 基于Logistic回归(LR)和XGBoost算法构建并验证3~5期慢性肾脏病(CKD)非透析患者营养不良风险模型,对比两者预测效能。方法 选取506例CKD患者为研究对象,按时间先后顺序以8∶2比例分为训练集404例和测试集102例。训练集以是否存在营养不良划分为病例组和对照组,每组202例。建立LR与XGBoost模型,通过受试者工作特征(ROC)曲线的曲线下面积(AUC)、灵敏度、特异度、GiViTI校准曲线带和临床决策曲线评价模型效能。结果 LR模型显示,年龄≥60岁、疾病5期、食欲减退、低白蛋白、低前白蛋白、低上臂肌围、高感知压力是3~5期非透析患者营养不良的独立危险因素,体力活动为保护因素(P<0.05)。XGBoost模型中,前5位影响变量为白蛋白、食欲、体力活动、前白蛋白与上臂肌围。训练集LR与XGBoost模型的AUC分别为0.930和0.947,测试集分别为0.925和0.933,后者预测能力略高(P>0.05)。GiViTI校准曲线带均显示良好的校准能力。结论 XGBoost模型联合沙普利加法解释在识别营养不良患者及指导精准护理方面表现更优。
Objective To construct and validate a risk prediction model for malnutrition in non-di-alysis patients with stages 3 to 5 chronic kidney disease(CKD)based on Logistic regression(LR)and XGBoost algorithms,and to compare the predictive performance between the two models.Methods A total of 506 CKD patients were enrolled as study subjects.According to chronological order,they were divided into training set(n=404)and test set(n=102)at the ratio of 8 to 2.The training set was divided into case group and control group based on whether they were malnourished,with 202 cases in each group.The LR and XGBoost models were established,and the model efficacy was evaluated through the area under the receiver operating characteristic(ROC)curve(AUC),sensitivity,speci-ficity,GiViTI calibration curve band and clinical decision curve.Results The LR model identified age≥60 years,disease of stage 5,reduced appetite,hypoalbuminemia,low prealbumin,low mid-arm muscle circumference and high perceived stress as independent risk factors for malnutrition among non-dialysis CKD patients,while physical activity was identified as a protective factor(P<0.05).In the XGBoost model,the top five influential variables were serum albumin,appetite,physical activity,prealbumin and mid-arm muscle circumference.The AUC of the LR and XGBoost models in the train-ing set were 0.930 and 0.947 respectively,and those in the test set were 0.925 and 0.933.The pre-dictive ability of the latter was slightly higher(P>0.05).The GiViTI calibration curve bands all showed good calibration capability.Conclusion The XGBoost model combined with shapley additive explanation performs better in identifying malnourished patients and guiding precise care.
作者
郑玉婷
郑泽璇
孙亚波
万雅菲
高占红
李淑杏
ZHENG Yuting;ZHENG Zexuan;SUN Yabo;WAN Yafei;GAO Zhanhong;Li Shuxing(School of Nursing and Rehabilitation,North China University of Science and Technology,Tangshan,Hebei,063000;Department of Nephrology,Hebei General Hospital,Shijiazhuang,Hebei,050000)
出处
《实用临床医药杂志》
2025年第11期110-117,共8页
Journal of Clinical Medicine in Practice
基金
华北理工大学研究生创新项目(2024S43)。
关键词
慢性肾脏病
营养不良
影响因素
曲线下面积
XGBoost模型
沙普利加法值
上臂肌围
食欲
chronic kidney disease
malnutrition
influencing factors
area under the curve
XGBoost model
shapley additive explanation value
mid-arm muscle circumference
appetite
作者简介
通信作者:李淑杏。
