“不确定”结果对ROC预测和ELISA-HCV/HIV临界值设定的探讨

Indeterminacy results for ROC prediction and ELISA-HCV/HIV threshold setting

目的探讨确证试验不确定结果(IR)纳入受试者工作曲线(ROC)分析对酶联免疫吸附法(ELISA)抗-HCV和HIV Ag抗-HIV临界值设定的影响。方法收集本实验室3种ELISA抗-HCV试剂(A、B和C)和3种HIVAg抗-HIV试剂(D、E和F)结果(S/CO)及其确证结果,共1 184人次。构建3种预测模型(模型1:舍弃IR;模型2:IR纳入阳性结果;模型3:IR纳入阴性结果)的ROC曲线,计算各模型曲线下最大面积(AUC)、最大Youden指数、临界值,探讨最佳的预测阳性献血者ELISA临界值。结果抗-HCV共检测486人次,其中试剂A、B和C分别检测:199例、191例和96例,A、B和C在模型1的临界S/CO值分别为4.261、6.159、8.172;AUC分别为:0.985、0.959、0.999;A、B和C在模型2的临界S/CO值(分别为1.821,1.565,1.320)小于模型3的临界S/CO值(分别为:9.048,8.700,4.918);HIV Ag抗-HIV共检测698人次,其中D、E和F分别检测:281例、264例和153例,D、E和F在模型1的AUC分别为:0.983、0.994、0.997,临界S/CO值分别为11.890、11.540、13.640;D、E和F在模型2和3的临界S/CO值,除E变化较大(模型2为5.187,模型3为11.540),D变化微小(11.760-12.040),F均无变化(13.640)。结论不同试剂的"不确定"结果纳入ROC预测的临界S/CO值不同;基于现有数据,抗-HCV和HIV Ag抗-HIV检测的S/CO值分别大于8.0和13.0时预测为阳性献血者的概率较大;模型2计算的临界值低于模式1,但依然高于试剂盒的推荐临界值,该值对于预测潜在阳性献血者及制定归队策略有一定参考价值。

Objective To investigate the influence of indeterminacy results(IRs) of the confirmation test in the receiver operating curve(ROC) analysis on the setting of anti-HCV and HIV Ag/Ab thresholds by enzyme-linked immunosorbent assay(ELISA). Methods The results(S/CO) of three ELISA anti-HCV reagents(A, B and C) and three HIV Ag/Ab reagents(D, E, and F) and their confirmation results were collected in the laboratory, with a total of 1184 samples. ROC curves were constructed for three prediction models(model 1: IR were not included, model 2: IR were included in positive results, model 3: IR were included in negative results). The maximum area(AUC), the maximum Youden index and the critical value were calculated in each model curves to explore the optimal ELISA critical values for predicting positive blood donors.Results A total of 486 anti-HCV samples were tested, including 199, 191, and 96 cases by reagents A, B, and C, respectively. The critical S/CO values of reagents A, B, and C in model 1 were 4.261, 6.159, and 8.172, and the AUCs were 0.985, 0.959, and 0.999, respectively. The critical S/CO values of reagents A, B, and C in model 2(1.821, 1.565, and 1.320, respectively) were lower than the critical S/CO values in model 3(9.048, 8.700, and 4.918, respectively). A total of 698 HIV Ag/Ab samples were tested, and 281, 264, and 153 cases were tested by reagents D, E, and F, respectively. The AUCs of D, E, and F in model 1 were 0.983, 0.994, 0.997, and the critical S/CO values were 11.890, 11.540, and 13.640, respectively. For critical S/CO values of D, E, and F in model 2 and 3, except for E, which changed significantly(model 2 was 5.187, model 3 was 11.540), and D changed slightly(11.760—12.040), there was no change in F(13.640). Conclusion The critical S/CO value for IR included in the ROC prediction of different reagents were different. Based on available data, when the S/CO value of anti-HCV and HIV Ag/Ab tests was greater than 8.0 and 13.0, respectively, the probability of blood donors predicted positive were greater. The critical value calculated in model 2 was lower than it in model 1, but still above the recommended critical value of the kit. The critical value calculated by model 2 had certain reference value for predicting potential positive blood donors and formulating the reentry strategy.