诊断试剂系统评价中性能评价指标的使用现状与指标池构建

The current status and construction of evaluation index list for the performance of diagnostic reagents in systematic reviews

目的分析诊断试剂系统评价中相较于金标准的性能评价指标使用现状,并梳理、形成诊断试剂系统评价的性能评价指标体系清单。方法计算机检索PubMed、Embase(OVID)、Cochrane Library(OVID)、CBM、VIP、WanFang Data和CNKI数据库,搜集有关诊断试剂性能评价的系统评价,检索时限为建库至2023年4月28日。由2名研究者独立筛选文献、提取资料,采用频数(构成比)描述评价指标使用现状,采用定性合成对评价指标进行归纳、汇总。结果纳入诊断试剂系统评价133篇,使用频率超过50%的评价指标为灵敏度(133/133,100.0%)、特异度(131/133,98.5%)和SROC曲线下面积(80/133,60.2%),使用频率不足5%的评价指标为Q指数(6/133,4.5%)、假阳性率(3/133,2.3%)、Kappa值(2/133,1.5%)、假阴性率(1/133,0.8%)和约登指数(1/133,0.8%)。为全面评价诊断试剂相较于金标准的性能,可从真实性、预测性两个角度开展评价,供使用的评价指标共14个。结论已发表的诊断试剂系统评价使用的评价指标差异性较大且较局限,亟需基于专家共识形成标准化评价指标。

Objective To describe the current status of the evaluation index for the performance of diagnostic reagents compared with gold standards in systematic reviews and develop the list of evaluation indexes.Methods PubMed,Embase(OVID),Cochrane Library(OVID),CBM,WanFang Data and CNKI databases were searched for systematic reviews about the performance of diagnostic reagents compared with gold standards from inception to 28th April,2023.Two reviewers independently screened literature and extracted data.The frequency and ratio were used to describe the current status,while the qualitative synthesis was used to develop the list.Results A total of 133 systematic reviews were included.Sensitivity(133/133,100.0%),specificity(131/133,98.5%)and AUC(80/133,60.2%)were used more frequently than 50%.Q index(6/133,4.5%),false positive rate(3/133,2.3%),Kappa value(2/133,1.5%),false negative rate(1/133,5%)and Youden's index were used less frequently than 5%.In order to evaluate the performance of diagnostic reagents compared with gold standards in systematic reviews comprehensively,a total of 14 index related to validity and predictability could be considered.Conclusion The evaluation index for the performance of diagnostic reagents in systematic reviews are inconsistent and limited,so there is an urgent need to develop standardized evaluation indicators based on expert consensus.