标题：带状疱疹患者继发缺血性脑卒中的两种临床模型构建及效能分析

      作者：丁志远，陈后勤，邵凌云，何敏，左健，宋飞，杨涛，吴有丽    芜湖市第一人民医院神经内科，安徽 芜湖 241000

      卷次： 2019年30卷22期

      【摘要】 目的 构建两种带状疱疹继发缺血性脑卒中(ischemic stroke，IS)临床模型，并比对其临床评判效能。方法 连续性、回顾性收集芜湖市第一人民医院神经内科，2015年4月1日至2018年6月30日收治的 IS患者为研究队列。通过查阅病历既往史，将起病前3个月内曾患有带状疱疹者列入观察组；否则列入对照组。通过Logistic回归及人工神经网络分别构建带状疱疹继发 IS的风险模型。同时，连续性、回顾性收集该院神经内科于2018年7月1日至12月31日收治的因带状疱疹继发 IS患者21例为验证组，将验证组的相关临床资料分别代入两组临床模型，基于实际预测结果的差异性，绘制ROC曲线以比对两种模型的评判效能。结果 经纳入、排除、剔除标准，最终获取267例研究队列的临床资料，其中观察组69例，对照组198例；基于研究队列临床资料，构建两类带状疱疹继发IS的临床模型；①由以下变量构成Logistic回归模型：带状疱疹合并感染、皮损部位-头面部、VAS评分、多处神经受损、病程持续时间；②由以下变量构成人工神经网络模型：年龄、VAS评分、皮损严重度评分、病程持续时间、皮损部位-头面部；另获取21例验证组患者的临床资料，将相应临床资料分别代入前述两种预测模型予以验证，ROC曲线显示：Logistic回归和人工神经网络模型的曲线下面积分别为0.854及0.878，总体预测效果均较好，但两者的预测能力对比显示，人工神经网络模型的拟合效果相对更佳。结论 两种临床模型应用于预测带状疱疹继发 IS的风险均有良好的评判效能，相比之下人工神经网络模型更能反映出变量之间的真实关系，值得临床中推广。
      【关键词】 缺血性；脑卒中；带状疱疹；Logistic回归；人工神经网络
      【中图分类号】 R743.3 【文献标识码】 A 【文章编号】 1003—6350（2019）22—2900—05

Construction and evaluation of two clinical models of ischemic stroke in patients with herpes zoster.
DINGZhi-yuan, CHEN Hou-qin, SHAO Ling-yun, HE Min, ZUO Jian, SONG Fei, YANG Tao, WU You-li. Department ofNeurology, the First People's Hospital of Wuhu City, Wuhu 241000, Anhui, CHINA
【Abstract】 Objective To construct two clinical models of ischemic stroke by collecting clinical data from the pa-tients with ischemic stroke (IS) after herpes zoster, and to compare the clinical performance of two models. Methods TheIS patients who admitted to Department of Neurology, the First People's Hospital of Wuhu City from April 1, 2015 toJune 30, 2018 were enrolled into the subjects. By reviewing the past history of medical records, the patients who had her-pes zoster within 3 months before the onset of IS were selected as the observation group; otherwise, they were includedin the control group. Risk models of secondary herpes zoster were constructed by logistic regression and artificial neuralnetwork. At the same time, 21 cases of patients with ischemic stroke (IS) after herpes zoster, who admitted to Departmentof Neurology, the First People's Hospital of Wuhu City from July 1 to December 31, 2018, were selected as the validationgroup. The relevant clinical data of the verification group were imported into two clinical models. Based on the differencein actual prediction results, ROC curves were plotted to compare the performance of the two models. Results After inclu-sion, exclusion, and elimination of standards, the clinical data of 267 study cohorts were obtained finally. Among them,there were 69 patients in the observation group and 198 patients in the control group. Based on the clinical data of thestudy cohort, the clinical model of two types of herpes zoster after IS was constructed as follows:① Logistic regressionmodel consisting of the following variables: herpes zoster infection, lesion site-head and face, VAS score, multiple nervedamage, duration of disease; ② The artificial neural network model is composed by the following variables: age, VASscore, lesion severity score, duration of disease duration, lesion site-head and face; the clinical data of 21 patients in the