尿液常规智能审核规则验证与改进的多中心研究  

A multicenter research on validation and improvement of the intelligent verification criteria for routine urinalysis

作　　者：王力 郝晓柯 杨大干[3] 蒋黎[4] 孙成铭[5] 史伟峰[6] 伍勇[7] 吴卫[8] 刘家云 徐卫益[3] 张娟[4] 杨丽萍[5] 蒋丽娟[6] 袁金玲[7] 金晶[8] 王刚强 俞倩[3] 熊志刚[4] 王臣玉[5] 蒋舒娜[6] 廖金凤 何贝 崔巍 Wang Li;Hao Xiaoke;Yang Dagaan;Jiang Li;Sun Chengming;Shi Weifeng;Wu Yong;Wu Wei;Liu Jiayun;Xu Weiyi;Zhang Juan;Yang Liping;Jiang Lijuan;Yuan Jinling;Jin Jing;Wang Gangqiang;Yu Qian;Xiong Zhigang;Wang Chenyu;Jiang Shuna;Liao Jinfeng;He Bei;Cui Wei(Department of Clinical Laboratory,National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100021,China;Department of Clinical iMboratory,Xijing Hospital of Air Force Medical University,Xi'an 710000,China;Center of Clinical Laboratory,the First Affiliated Hospital,Zhejiang University School of Medicine,Hangzhou 310000,China;ClinicaI laboratory,Sichuan Provincial Peopled Hospital,Chengdu 610072,China;Medicine Ldboratory,Yantai Yuhuangding Hospital,Yantai 264000,China;Department of Clinical Ixiboratory.the Third Affiliated Hospital of Suzhou University,changzhou 213000,China;Department of Clinical Laboratory,the Third Xiangya Hospital of Central South University,Changsha 410013,China;Department of,Mboratory Medicine,Chinese Academy of Medical Sciences and Peking Union Medical College Hospital,Beijing 100730,China)

机构地区：[1]国家癌症中心/国家肿瘤临床医学研究中心/中国医学科学院北京协和医学院肿瘤医院检验科,北京100021 [2]空军军医大学西京医院检验科,西安710000 [3]浙江大学医学院附属第一医院检验科,杭州310000 [4]四川省人民医院检验科,成都610072 [5]烟台毓璜顶医院检验科,烟台264000 [6]苏州大学附属第三医院检验科,常州213000 [7]中南大学湘雅三医院检验科,长沙410013 [8]中国医学科学院北京协和医院检验科,北京100730

出　　处：《中华检验医学杂志》2020年第8期794-801,共8页Chinese Journal of Laboratory Medicine

基　　金：中国医学科学院医学与健康科技创新工程(2017-I2M-3-005)。

摘　　要：目的对前期建立的尿液常规智能审核规则(含复检规则和人工审核规则,简称智能规则)进行多中心大样本的验证及改进研究,以提高智能规则的临床适用性。方法智能规则的验证与改进,共收集2019年3月至9月浙江大学医学院附属第一医院、空军军医大学西京医院、烟台毓璜顶医院、苏州大学附属第三医院、四川省人民医院以及中南大学湘雅三医院等6家医院门诊或住院患者尿液标本31456份,其中3105份标本以镜检结果及人工审核为参考用于智能规则的验证与调整,28351份标本以100%人工审核为参考用于改进后的智能规则临床适用性的验证。结果原发布智能规则中的人工审核规则存在8.59%(202/2352)的无效审核率以及8.84%(208/2352)的无效拦截率,复检规则的假阴性率和复检率与前期结果相似。依据大数据深度分析结果和8家医院(含中国医学科学院肿瘤医院、中国医学科学院北京协和医院)技术骨干的讨论,增加了1条复检规则和4条人工审核规则、删除了2条人工审核规则,并对人工审核标准进行了统一,形成了改进后的智能规则。其中,复检规则的假阳性率、假阴性率(漏诊率)及复检率没有明显变化,分别为14.72%(457/3105)、4.06%(126/3105)及24.73%(768/3105);人工审核规则的无效审核率、无效拦截率均降到0;智能规则的人工审核率、自动审核通过率分别为50.89%(1580/3105)、49.11%(1525/3105)。智能规则的大样本验证结果与之相一致。结论多中心大样本验证显示改进后的智能规则具有良好的临床适用性。Objective A multi-center and large sample volume study was conducted on the verification and improvement of the early established criteria for intelligent routine urinalysis validation(including the microscopic review rules and manual validation rules,referred to as intelligent criteria for short),in order to improve the clinical application of this intelligent criteria.Methods A total of 31456 urine specimens were collected from the inpatients and outpatients in six hospitals in China,from March to September 2019.Firstly,3105 specimens were analyzed for preliminary verification and improvement of the intelligent criteria based on the results of the microscopic examination and manual validation.Secondly,28351 specimens were used to verify the clinical application of the improved intelligent criteria.All samples were manually validated as reference.Results The approval inconsistency rate of the manual validation rules in the original intelligent criteria was 8.59%(202/2352),and the interception inconsistency rate was 8.84%(208/2352).The false negative rate and the microscopic review rate of the microscopic review rules were similar to the previous results.Based on an in-depth analysis of big data and the discussions by senior technicians from eight hospitals,one microscopic review rules and four manual validation rules were added,meanwhile two manual validation rule was deleted.The manual validation standards were unified.Finally,the intelligent criteria was improved.Based on the improved intelligent criteria,for microscopic review rules,the false positive rate,false negative rate(misdiagnosis rate),and microscopic review rate did not change significantly,which were 14.72%(457/3105),4.06%(126/3105),and 24.73%(768/3105),respectively.The approval inconsistency rate and the interception inconsistency rate of manual validation rules were both reduced to 0;the total manual validation rate of the intelligent criteria was 50.89%(1580/3105),and the auto-validation rate was 49.11%(1525/3105).The large sample volume verific

关 键 词：尿分析 诊断试验 常规 人工智能 质量改进 多中心研究 

分 类 号：R446.12[医药卫生—诊断学]