尿液分析系统智能审核规则的建立与临床验证

The Establishment and Clinical Validation of Intelligent Audit Rules for Urine Analysis System

目的 利用人工智能的方法建立尿液分析系统自动审核规则,并且优化后用于临床评价。方法 收集北京首都医科大学附属朝阳医院2022年7月至2023年6月的尿液标本16 000例,阳性标本(尿干化学及尿沉渣结果任何一项出现阳性的标本)4 889例,以及阴性标本(尿干化学及尿沉渣结果均为阴性)11 111例。进行第一次方法训练,得到一组自动审核规则及方法A,评价其预测性能。所有标本均经显微镜镜检复核,结合临床患者基本信息、尿干化学、尿沉渣以及显微镜检测结果,均一致的标本标记为阴性通过,反之标记为阳性拦截。经统计,被拦截的标本为786例,将这786例标本作为重点样本组,其余的阳性标本为非重点样本组,进行第二次方法训练,得出最终自动审核的规则及方法B,并评价其预测性能。对两次预测结果进行判断(包括假阴性率、假阳性率、真阴性率、真阳性率),同时计算其诊断效能指标(包括放行率、阳性预测值、阴性预测值、总有效率、特异性、灵敏度、重点病例覆盖率等)。结果 方法B得到了71条自动规则,自动审核放行率达到51.8%,假阳性率为26.9%,假阴性率控制在3.3%,并且覆盖了99.5%的重点病例。结论 本研究建立的审核规则符合临床需求,提高工作效率、减少检验专业技术人员需求的同时保证检验质量。

Objective Artificial intelligence method was used to establish automatic review rules for urine analysis system,and was optimized for the clinical evaluation.Methods From July,2022 to June,2023,16000 urine samples were collected from Chaoyang Hospital Affiliated to Beijing Capital Medical University:including 4889 positive samples(urine dry chemistry and urine sediment results were positive)and 11111 negative samples(urine dry chemistry and urine sediment results were negative).We conducted the first method training,obtained a set of automatic audit rules and method A,and evaluated its predictive performance.All specimens were reviewed by microscope,combined with the basic information of clinical patients,urine dry chemistry,urine sediment and microscope detection results.All specimens were labeled as negative to pass,and otherwise labeled as positive to intercept.According to statistical analysis,786 specimens were intercepted.786 specimens were included as the key sample group,and the rest positive samples were included as the non-key sample group.The second method training was conducted to obtain the final automatic audit rule and method B,and was evaluated for its predictive performance.The results of the two predictions were judged(including false negative rate,false positive rate,true negative rate,true positive rate),and the diagnostic efficiency(including the application rate,positive predictive value,negative predictive value,total effective rate,specificity,sensitivity,coverage rate of key cases,etc.)were calculated.Results Method B obtained 71 automatic rules.The automatic review rate reached 51.8%,while the false positive rate was 26.9%,the false negative rate was controlled at 3.3%,and 99.5%of the key cases were covered.Conclusion The audit rules established in this study meet the clinical needs,which can improve work efficiency,reduce the need for laboratory professionals and technical personnel,and ensure the quality of testing.