目的 

意大利米兰会议最新确定的性能规范设定模式有3种。本研究依照其中2种，即基于生物学变异（biological variation, BV）导出性能规范和基于当前技术水平导出性能规范，并利用室间质量评价（external quality assessment, EQA）和室内质量控制（internal quality control, IQC）数据，为我国68个临床检验常规定量项目推荐允许不精密度（允许CV）、允许偏倚（允许bias）和允许总误差（allowable total error, TEa）。

方法 

采用基于Web方式的EQA软件，收集68个研究项目从有EQA数据起至今，各年份每次EQA活动全部实验室的数据，以及各研究项目从有IQC数据起，各年份每次EQA活动前一个月全部实验室回报的IQC数据。利用Excel 2016、SPSS 19.0以及卫生部临床检验中心（National Center for Clinical Laboratories, NCCL）与北京科临易检信息技术有限公司合作研发的后台分析软件，分别统计各研究项目的EQA数据，各年份均按照批号计算每个实验室的“百分差值”以及各批号所有实验室“百分差值”的P₁₀（第10百分位数）和P₉₀（第90百分位数）；分别统计各研究项目的IQC数据，各年份均按照批号计算所有实验室“当月累积在控CV”的P₈₀（第80百分位数）。对于有BV数据的项目，将各研究项目的EQA“百分差值”分别与基于BV导出的三种水平TEa、我国EQA评价标准、澳大利亚皇家病理学家学会（RCPA）质量保证计划（RCPA Quality Assurance Programs, RCPAQAP）评价标准作比较，计算各年份所有批号的百分差值通过率，通过比较多个年份百分差值通过率的差异，反映我国临床实验室检测各项目的总误差是否有变化；并且，若在基于BV导出的某水平TEa下，2016年所有批号的百分差值通过率均大于80%，则该水平TEa满足作为该项目推荐TEa的要求；若多个水平TEa均满足作为推荐TEa的要求，则选择最严格（即最小）的TEa作为推荐TEa。将各研究项目的IQC“当月累积在控CV” 分别与基于BV导出的三种水平允许CV、1/3TEa和1/4TEa（TEa是各项目在中国EQA中的评价标准）作比较，计算各年份所有批号的CV通过率，通过比较多个年份CV通过率的差异，反映我国临床实验室检测各项目的不精密度是否有变化；并且，若在基于BV导出的某水平允许CV下，2016年所有批号的CV通过率均大于80%，则该水平允许CV满足作为该项目推荐允许CV的要求；若多个水平允许CV均满足作为推荐允许CV的要求，则选择最严格（即最小）的允许CV作为推荐允许CV。对于无BV数据的项目或者基于BV导出的三种水平性能规范均无法作为推荐标准的项目，则基于当前技术水平来导出推荐TEa和允许CV。各项目在获得推荐TEa和允许CV后，利用公式|bias|≤TEa-z⨯CV(z通常为1.65)导出推荐允许bias。

结果 

68个研究项目的EQA数据分析结果显示，40个项目2016年各批号的百分差值通过率在基于BV导出的一个或多个TEa下均达80%以上，因此使用基于BV导出的TEa作为推荐TEa；28个项目使用基于当前技术水平导出的TEa作为推荐TEa。68个研究项目中，2个项目的推荐TEa等于目前我国的EQA评价标准，51个项目的推荐TEa小于目前我国的EQA评价标准，7个项目的推荐TEa大于目前我国的EQA评价标准。68个研究项目的IQC数据分析结果显示，25个项目2016年各批号的CV通过率在基于BV导出的一个或多个允许CV下均达80%以上，因此使用基于BV导出的允许CV作为推荐允许CV；42个项目使用基于当前技术水平导出的允许CV作为推荐允许CV；1个项目由于不精密度大于总误差无法导出合理的推荐允许CV。68个研究项目中，可以导出允许bias和无法导出允许bias的项目均是34个。

结论 

本研究是国内首次对大量（68个）临床检验常规定量项目的允许CV、允许bias和TEa进行研究，反映了68个研究项目在目前能够达到的性能水平，并为68个研究项目推荐了符合我国实验室当前检测水平的允许CV、允许bias和TEa。

Objective

There are 3 models of setting performance specifications that were defined in the Milan Conference. This study used 2 of them—setting performance specifications based on biological variation (BV) and setting performance specifications based on the state-of-the-art, as well as the data of External Quality Assessment (EQA) and Internal Quality Control (IQC), to discuss and recommend allowable imprecision (allowable CV), allowable bias and allowable total error (TEa) for 68 routine quantitative tests in clinical laboratories in China.

Method

For each research item, the EQA data from the beginning of EQA plan and the IQC data from the beginning of IQC plan were collected by the web-based EQA software. The Excel 2016, SPSS 19.0 and the analytical software developed together by National Center for Clinical Laboratories (NCCL) and Beijing Clinet Information Technology Limited Company (www.clinet.com.cn) were applied to analyze and calculate the ‘percentage difference’ of all lots in each year and P₁₀ (the 10th percentile) and P₉₀ (the 90th percentile) of ‘percentage difference’ of all laboratories in each lot by EQA data, as well as P ₈₀ (the 80th percentile) of ‘in-control coefficient of variation (CV)’ of all lots in each year by IQC data for all items. For each item having BV data, ‘percentage difference’ were respectively compared with three level TEa derived from BV, evaluation criteria in Chinese EQA and RCPAQAP, then the pass rates of ‘percentage difference’ were calculated for all lots in each year. By comparing the difference of the pass rates of ‘percentage difference’ among many years, the change of total error in the testing of each item would be reflected. In addition, for each item, if all lots in 2016 had the pass rates of ‘percentage difference’ more than 80% when compared with one TEa derived from BV, this TEa would satisfy the requirement to be the recommendatory TEa of this item. If many TEa derived from BV all satisfied the requirement, the most stringent (smallest) one would be the recommendatory TEa. Meanwhile, for each item having BV data, the ‘in-control CV’ were respectively compared with three level allowable CV derived from BV, 1/3TEa and 1/4TEa (TEa is the evaluation criterion of each item in Chinese EQA), then the pass rates of ‘in-control CV’ were calculated for all lots in each year. By comparing the difference of the pass rates of ‘in-control CV’ among many years, the change of imprecision in the testing of each item would be reflected. In addition, for each item, if all lots in 2016 had the pass rates of ‘in-control CV’ more than 80% when compared with one allowable CV derived from BV, this allowable CV would satisfy the requirement to be the recommendatory allowable CV of this item. If many allowable CV derived from BV all satisfied the requirement, the most stringent (smallest) one would be the recommendatory allowable CV. For items not having BV data or items which could not use performance specifications derived from BV as recommendatory criteria, the performance specifications based on the state-of-the-art would be used as their recommendatory TEa and allowable CV. For each item, after acquiring the recommendatory TEa and allowable CV, the formula |bias|≤TEa-z⨯CV (the value of z is always 1.65) could be applied to obtain the recommendatory allowable bias.

Result

The results of EQA data of all items showed that 40 items had pass rates of percentage difference more than 80% for all lots in 2016 in the comparison to one or several TEa derived from BV, therefore they could use the TEa derived from BV as their recommendatory TEa, but other 28 items used the TEa based on the state-of-the-art as their recommendatory TEa. The comparison between recommendatory TEa and the evaluation criterion in Chinese EQA for each item suggested that 2, 51 and 7 items had recommendatory TEa respectively equal to, less than and larger than their evaluation criteria in Chinese EQA. The results of IQC data of all items indicated that 25 items had pass rate of CV more than 80% for all lots in 2016 in comparison to one or several allowable CV derived from BV, therefore they could use the allowable CV derived from BV as their recommendatory allowable CV, other 42 items exploited the allowable CV based on state-of-the-art as their recommendatory allowable CV, and the remained 1 item did not have the recommendatory allowable CV because its imprecision was larger than total error. Among all items, items which had allowable bias and not had allowable bias are both 34.

Conclusion

This was the first study on allowable CV, allowable bias and TEa for such many (68) routine quantitative tests in China. In this study, the performance level which could be achieved for 68 items were reflected and the allowable CV, allowable bias and TEa conformed to the Chinese current situation for all items were recommended.