首页|基于人工智能投票算法建立识别血清钠离子随机误差的实时质量控制法

基于人工智能投票算法建立识别血清钠离子随机误差的实时质量控制法

扫码查看
原文链接
  • 万方数据
  • 维普
目的 利用人工智能投票(voting)算法,建立一种快速识别血清钠离子随机误差的实时质量控制新方法,并评价在此基础上构建模型的相关效能.方法 采用回顾性调查研究方法,通过南京鼓楼医院医学检验科实验室信息系统导出2021年1月至5月在Beckman AU5400生化分析仪上检测的住院患者的血清钠离子结果,共计144 754条,作为本研究的无偏数据.人为引入随机误差,生成相应有偏数据.随后,根据投票算法的原理建立质量控制方法(ViQC)模型.针对每种偏差,用ViQC模型与5种传统PBRTQC算法进行测试,利用分类模型评估指标评价ViQC模型的分析性能.绘制偏差检测曲线,采用误差检出所需对称修剪平均样本数(tANPed)来评价模型的临床检测效能,并与5种传统PBRTQC算法进行比较.结果 ViQC模型对所有偏差检测的假阳性率均小于0.002,准确度大于0.951.当误差因子为1.5、2.5和3.0时,ViQC模型假阳性率均为0;当误差因子为2.5时,该模型的准确度高达0.979.与5种传统PBRTQC算法相比,ViQC模型对所有偏差检测的平均tANPed最多下降34％,误差检测敏感度更高.此外,ViQC模型在测试环节TEa定值偏差下的ROC曲线下面积高达0.989,tANPed仅为5.结论 成功建立了基于人工智能算法的患者数据实时质量控制模型,其临床检测效能优于传统PBRTQC算法.
Establishment of a real-time quality control method for identifying random error in serum sodium ion based on artificial intel-ligence voting algorithm
Objective To establish a novel real-time quality control method for rapidly identifying the random error of sodium ion con-centration in serum using an artificial intelligence voting algorithm,and evaluate the relevant effectiveness of the model established on this basis.Methods A total of 144 754 test results of serum sodium ion rom the inpatients measured by Beckman AU5400 biochemis-try analyzer from January to May 2021 were obtained retrospectively from laboratory information system of the Department of Clinical La-boratory,Nanjing Drum Tower Hospital,and all the data were used as unbiased data for the current study.The random errors were arti-ficially introduced to generate the corresponding biased data set.Subsequently,the voting algorithm-based internal quality control model(ViQC)was established using the principles of the voting algorithm.The ViQC model and five classical PBRTQC(patient-based real-time quality control)algorithms were performed direct to each biased data.The analytical performance of the ViQC model was evaluated by using classification model criteria.The trimmed average number of patient samples until error detection(tANPed)was used to com-pare the clinical detection efficacy of the ViQC model with those of the five classical algorithms,and the error detection curves were plotted.Results Compare with all the classical algorithms,the ViQC model showed a false positive rate below 0.002 and achieved ac-curacy above 0.951 in detecting all the deviations.When the error factors were 1.5,2.5,and 3.0,the false positive rate of the ViQC model was zero.When the error factor was 2.5,its accuracy reached 0.979.Compared to the five classical PBRTQC algorithms,the ViQC model reduced the overall average tANPed by up to 34％and showed higher sensitivity for error detection.In addition,the ViQC model demonstrated the area under the ROC curve was as high as 0.989 at TEa on the test set,but the value of tANPed wasonly five.Conclusion We successfully established a real-time quality control model for the data of patients based on artificial intelligence algo-rithms,and its efficacy of clinical detection was superior to the traditional PBRTQC algorithms.

quality controlpatient datareal-time quality controlrandom errorartificial intelligence

刘园、郑和翔、徐志晔、陈文琴、宋宏岩、陈雨欣

展开 >

南京大学医学院附属鼓楼医院检验科,南京 210008

上海理工大学管理学院,上海 200093

质量控制 患者数据 实时质控 随机误差 人工智能

国家卫生健康委"检验检查结果智能互认应用研究"项目江苏省医院协会医院管理创新研究课题南京大学现代医院管理与发展研究所课题项目南京鼓楼医院医学发展医疗救助基金会资助项目

JYHRJG2024B33JSYGY-3-2024-650NDYG2021014

2024

10.13602/j.cnki.jcls.2024.10.11

临床检验杂志
江苏省医学会

临床检验杂志

CSTPCD
影响因子：0.746
ISSN：1001-764X
年,卷(期)：2024.42(10)