基于机器学习的机械通气患者ICU获得性衰弱风险预测模型的构建  

Construction of a risk predictive model for ICU-acquired weakness in patients with mechanical ventilation based on machine learning

作　　者：姜金霞[1] 刘树炀 孙晓[2] 田梅梅 刘艺[4] 许金玲 Jiang Jinxia;Liu Shuyang;Sun Xiao;Tian Meimei;Liu Yi;Xu Jinling(Nursing Department,Tenth People's Hospital of Tongji University,Shanghai 200072,China;Nursing Department,Shanghai Fourth People's Hospital Affiliated to Tongji University,Shanghai 200072,China;Nursing Department,Renji Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 200072,China;Department of Rehabilitation,Tenth People's Hospital of Tongji University,Shanghai 200072,China)

机构地区：[1]同济大学附属第十人民医院护理部,上海200072 [2]同济大学附属上海市第四人民医院护理部,上海200072 [3]上海交通大学医学院附属仁济医院护理部,上海200072 [4]同济大学附属第十人民医院康复科,上海200072

出　　处：《中华现代护理杂志》2025年第8期1059-1065,共7页Chinese Journal of Modern Nursing

基　　金：上海申康医院发展中心医企融合创新支撑技能培训专项课题(SHDC2023CRS002)。

摘　　要：目的筛选机械通气患者ICU获得性衰弱的风险因子并构建预测模型,为机械通气患者的健康管理提供依据。方法采用便利抽样法,选取2019年10月—2020年8月同济大学附属第十人民医院收治的312例ICU机械通气患者为研究对象,将其按照7∶3比例划分为训练集(n=220)和测试集(n=92)。基于机器学习算法,分别采用决策随机森林(DRF)、极度随机树(XRT)和广义线性模型(GLM)构建3种机械通气患者ICU获得性衰弱风险预测模型,并采用受试者工作特征曲线下面积(AUC)、精确率-召回率曲线下面积(AUPRC)、对数损失和均方根误差(RMSE)评价预测模型的效能。结果机械通气患者ICU获得性衰弱风险预测因子共有7个,分别为年龄、性别、制动、机械通气时间、血糖水平、乳酸水平以及肠外营养。测试集和训练集验证显示,GLM预测模型的AUC、AUPRC均大于DRF、XRT预测模型;测试集验证显示,GLM预测模型的RMSE、对数损失均小于DRF、XRT预测模型。结论基于机器学习算法GLM预测模型具有较好的预测性能,医护人员可在制动、机械通气时间及血糖管理等方面构建基于循证证据的干预策略。ObjectiveTo screen risk factors for ICU-acquired weakness in patients with mechanical ventilation and construct a predictive model,so as to provide a basis for the health management of patients with mechanical ventilation.MethodsConvenience sampling was used to select 312 ICU patients with mechanical ventilation admitted to the Tenth People's Hospital of Tongji University from October 2019 to August 2020 for the study.Patients were divided into training set(n=220)and test set(n=92)in a 7∶3 ratio.Based on machine learning algorithms,decision random forest(DRF),extremely-randomized trees(XRT)and generalized linear model(GLM)were used to construct three ICU-acquired weakness risk prediction models for patients with mechanical ventilation,respectively.The performance of the prediction model was evaluated using the area under the receiver operating characteristic curve(AUC),the area under the precision-recall curve(AUPRC),and the root mean square error(RMSE).ResultsThere were 7 predictors of risk of ICU-acquired weakness in patients with mechanical ventilation,including age,gender,braking,duration of mechanical ventilation,blood glucose,lactic acid,and parenteral nutrition.Test set and training set validation showed that AUC and AUPRC of GLM prediction model were greater than those of DRF,XRT prediction model.Test set validation indicated that the RMSE,logarithmic loss of GLM prediction model was less than those of DRF,XRT prediction model.ConclusionsMachine learning algorithm based GLM prediction model has good prediction performance.Healthcare professionals can construct evidence-based decisions for interventions in areas such as braking,duration of mechanical ventilation,and blood glucose management.

关 键 词：机械通气 ICU获得性衰弱 预测模型 机器学习 

分 类 号：R47[医药卫生—护理学]