恢复期脑卒中患者发生认知障碍的风险预测模型构建与评价  

Construction and validation of the risk prediction model for developing cognitive impairment in convalescent stroke patients

作　　者：王倩雯 詹乐昌[1,2] 欧阳雨婷 徐钒锋 陈红霞 詹杰[1,2,3] WANG Qianwen;ZHAN Lechang;OUYANG Yuting(Guangzhou University of Chinese Medicine,Guangzhou,510120)

机构地区：[1]广州中医药大学,广东省广州市510120 [2]广东省中医院、广州中医药大学第二临床医学院康复科 [3]广东省中医药防治难治性慢病重点实验室

出　　处：《中国康复医学杂志》2024年第12期1810-1817,共8页Chinese Journal of Rehabilitation Medicine

基　　金：广东省基础与应用基础研究基金省企联合基金面上项目(2022A1515220025);广东省卫生健康委项目(A2022239);广东省中医院中医药科学技术研究专项(YN2020QN23);沈宝藩学术经验传承工作室建设项目(中医二院[2018]101号);广东省中医院与香港中文大学医学院生物医学学院基础临床协同创新计划项目(YN2018HK03)。

摘　　要：目的:认知障碍是脑卒中常见的并发症状之一,会影响患者的康复及其生活质量,但目前临床仍缺乏恢复期脑卒中后认知障碍(post-stroke cognitive impairment,PSCI)发生的风险预测模型,构建可靠的风险预测模型来尽早发现PSCI对卒中患者康复极为重要。因此本研究探讨恢复期PSCI的影响因素,并构建风险预测列线图模型。方法:采用回顾性研究设计,选取2019年12月—2022年12月在广东省中医院住院的恢复期脑卒中患者为研究对象,收集其人口学特征及临床相关资料。按照7∶3随机将全数据集划分为训练集与验证集。训练集用于构建恢复期脑卒中患者PSCI风险预测列线图模型,验证集用于评估模型性能。采用单因素和多因素Logistic回归分析恢复期脑卒中患者发生PSCI的影响因素,并应用R软件构建恢复期脑卒中患者发生PSCI的风险预测列线图模型。采用受试者工作特征曲线下面积(area under the curve,AUC)、灵敏度、特异度、校准曲线、决策曲线等指标评估模型性能。结果:年龄、右侧偏瘫、高血压病、冠心病、高同型半胱氨酸血症、Fugl-Meyer运动功能评定量表(simplified FuglMeyer assessment scale,FMA)积分、改良Barthel指数(modified Barthel index,MBI)、平均红细胞血红蛋白量是恢复期脑卒中患者发生PSCI的重要影响因素;以上述8个变量构建恢复期脑卒中患者发生PSCI的风险预测列线图模型,训练集模型AUC、灵敏度、特异度分别为0.804、75.5%、73.7%,验证集模型AUC、灵敏度、特异度分别为0.737、82.9%、62.8%;校准曲线显示该模型预测概率与实际一致性良好,决策曲线显示该模型具有良好的临床净获益。结论:本研究构建的PSCI风险预测列线图模型可个性化预测恢复期脑卒中患者发生认知障碍的概率,有助于医务人员尽早发现及诊治PSCI,并改善患者的预后。Objective:Cognitive impairment is one of the common complications of stroke,which can affect the rehabilitation and their quality of life.It is very important to build reliable risk prediction model tools to detect poststroke cognitive impairment(PSCI)in advance,but there is still no clinical risk prediction model for PSCI.Our aim was to identify the influencing factors of PSCI in convalescent stroke patients and construct a nomogram model for predicting the risk of PSCI based on these factors.Method:We retrospectively collected the demographic characteristics and clinically relevant data of convalescent stroke patients hospitalized in Guangdong Provincial Hospital of Chinese Medicine from December 2019 to December 2022.Then we randomly divided the whole data set into the training set and the validation set according to 7:3,the former data was used to construct a nomogram model for predicting the risk of PSCI,and the latter data was used to evaluate the model performance.Univariate and multivariate logistic regression were used to analyze the factors affecting PSCI in convalescent stroke patients.Based on these factors,we used the R software to construct a PSCI risk prediction model who was visualized through a nomogram.The model performance was evaluated using the area under the curve(AUC),sensitivity,specificity,calibration curve,and decision curve analysis(DCA).Result:Our prediction model indicated that age,right hemiparesis,hypertension,coronary heart disease,hyperhomocysteinemia,Fugl-Meyer assessment scale(FMA)score,modified Barthel index(MBI)score and,mean corpuscular hemoglobin were independent factors influencing the occurrence of PSCI in convalescent stroke patients.The AUC,sensitivity and specificity of the model were 0.804,75.5%and 73.7%in the training set,and 0.737,82.9%and 62.8%in the validation set,suggesting that the model had a good discrimination.The calibration curve of the training and validation sets indicated a good consistency between the prediction and the real observation.The decision curv

关 键 词：脑卒中 认知障碍 风险因素 列线图 预测模型 LOGISTIC模型 

分 类 号：R493[医药卫生—康复医学] R743.2[医药卫生—临床医学]