基于机器学习算法构建老年糖尿病住院患者衰弱预测模型  

Construction of a frailty prediction model for elderly diabetic inpatients based on machine learning algorithms

作　　者：郑雪梅 张静 郑金龙 Zheng Xuemei;Zhang Jing;Zheng Jinlong(Health Science Center,Yangtze University,Jingzhou 434023,China)

机构地区：[1]长江大学医学部,荆州434023

出　　处：《中华现代护理杂志》2025年第3期340-346,共7页Chinese Journal of Modern Nursing

摘　　要：目的基于机器学习算法构建老年糖尿病住院患者衰弱预测模型,并评价模型预测效能,为早期识别并预防老年糖尿病患者衰弱的发生提供依据。方法采用便利抽样法,选取2023年3—10月入住荆州市2家三甲医院内分泌科、老年医学科的380例老年糖尿病患者为研究对象。采用二项Logistic回归分析老年糖尿病患者发生衰弱的影响因素。使用Python 3.8.2软件sklearn库中的函数构建随机森林、支持向量机和K邻近算法的预测模型,得出各个模型的准确率、精确率、召回率、F1值及受试者工作特征曲线下面积(AUC)。结果多病共存、多重用药、糖尿病自我管理能力、营养状况、25-羟基维生素D 3、糖尿病病程、日常生活活动能力是老年糖尿病患者发生衰弱的危险因素(P<0.05);随机森林、支持向量机和K邻近算法预测模型的AUC分别为0.85、0.83和0.79。结论构建的随机森林模型为最优模型,能够有效预测老年糖尿病住院患者衰弱发生的风险,有利于医护人员早期筛选出发生衰弱的高风险人群。Objective To construct a frailty prediction model for elderly diabetic inpatients based on machine learning algorithms and evaluate the predictive performance of the model,providing a basis for the early identification and prevention of frailty in elderly diabetic patients.Methods A convenience sampling method was used to select 380 elderly diabetic inpatients from the Endocrinology Department and Geriatrics Department of two ClassⅢGrade A hospitals in Jingzhou,admitted from March to October 2023.Binary Logistic regression analysis was used to identify the factors influencing frailty in elderly diabetic patients.The prediction models,including random forest,support vector machine,and K-nearest neighbors algorithms,were constructed using Python 3.8.2 and sklearn library functions.The accuracy,precision,recall,F1 score,and area under the receiver operating characteristic curve(AUC)for each model were evaluated.Results Factors such as comorbidity,polypharmacy,self-management ability of diabetes,nutritional status,25-hydroxyvitamin D 3,duration of diabetes,and activities of daily living were identified as risk factors for frailty in elderly diabetic patients(P<0.05).The AUC for the random forest,support vector machine,and K-nearest neighbors prediction models were 0.85,0.83,and 0.79,respectively.Conclusions The constructed random forest model is the optimal model,capable of effectively predicting the risk of frailty in elderly diabetic inpatients,which is beneficial for healthcare professionals to early screen high-risk populations for frailty.

关 键 词：糖尿病 老年人 衰弱 机器学习算法 预测模型 

分 类 号：R473.5[医药卫生—护理学]