The cohort of 278 children were at a median age of four months (IQR: 2.0 – 7.7 months) at the time of DBS collection. Samples were stored for a median of 14 months (IQR: 11.4 – 15.7 months; range 7–26 months) before testing with the Up24 assay. The HIV- infection prevalence was 14.7% by HIV DNA PCR testing. The median age of the 41 HIV- infected infants was four months (IQR: 2.2 – 7.4 months).
shows the distribution of adjusted OD values for HIV-exposed children who tested positive or negative by HIV-DNA PCR, and for the HIV-seronegative controls. The median OD for children identified as infected by HIV-DNA PCR was significantly higher (p < 0.001) as compared to those who tested negative by HIV-DNA PCR (2.24 [IQR, 0.58–3.16] versus 0.05 [IQR, −0.023–0.11]), respectively (). One HIV-DNA PCR negative sample seemed to be an outlier, but repeat testing based on Up24 results were not planned.
Fig. 1 Optical Density (OD) values measured at 490nm with reference at 630nm adjusted by subtracting OD of the average of the negative controls by Ultrasensitive p24 (Up24) antigen testing on dried blood spots (DBS) from HIV-exposed children with HIV-DNA PCR (more ...)
Previous studies on the use of Up24 antigen testing to diagnose HIV infection in children have used various approaches to discriminate between positive and negative samples; a cut-point OD value equal to or greater than five times the NCSD is often used.2,4,7,8,10,12,13
Using this criterion, the overall sensitivity of Up24 antigen assay was 87.8% and the specificity was 86.5% () with 86.7% of 278 HIV-exposed infants correctly classified. Five of 41 (12%) infected children were misclassified as uninfected, and 32 of 237 uninfected children (13.5%) were misclassified as infected.
Sensitivity and Specificity of Up24 for Infant HIV diagnosis using Receiver Operator Characteristics Curves (N=278)
The sensitivity and specificity of Up24 for identifying HIV-infection in children was examined using ROC curves over a range of cut-point OD values (). With increasing multiples of the NCSD, sensitivity decreased and specificity increased. The Youden index16,17
, the cut-point at which sensitivity (87.8%) and specificity (92%); were optimized was identified at 8-fold the NCSD; with 91.4% of children correctly classified. A cut-point OD that was 15 times the NCSD resulted in correct classification of 95.7% of the children’s infection status.
Another approach to evaluate the discriminatory capabilities of Up24 is the AUC derived from the ROC curve (, ).18
The AUC, a summary statistic for the ROC curve, is the probability that the measure used to classify samples will rank a randomly chosen positive sample higher than a randomly chosen negative sample.18
The AUC was lowest (0.69) at zero times the NCSD and highest between 7 to 15 times NCSD (range: 0.88–0.90); these comparisons were statistically different across the 15 NCSD multiples (p < 0.001). Using logistic regression analysis, the OD cut-point that optimized sensitivity (87.8%) and specificity (92%) was similarly identified as eight times the NCSD with 91.4% of the samples correctly classified.
Fig. 2 Receiver Operator Characteristics curves of the Up24 antigen data generated at the different cut-point optical density tested between 0–8-fold NCSD [Panel A] and 8–15-fold NCSD [Panel B]. The diagonal or chance line shown in grey is the (more ...)
Since age of testing and duration of storage of DBS may also affect the performance characteristics of Up24 antigen detection in DBS9,13
, we examined, using a logistic regression model with 8-fold NCSD as the cut-point, their effect on the sensitivity and specificity of Up24 antigen assay. No significant difference in the median age (p = 0.87) or duration of storage (p=0.246) between the samples with negative and positive PCR results were found.
summarizes the positive and negative predictive values for Up24 antigen assay at different HIV prevalence levels and when multiples of 5, 8 and 15-fold the NCSD are used for the cut-point. In areas where the expected prevalence of true infection is approximately 5%, the positive and negative predictive values of Up24 at 8- and 15-fold the NCSD were 36.6 and 99.3%, and 75.9 and 98.8%, respectively.
Effect of Varying Population Prevalence of HIV Infection on Test Performance Characteristics (PPVa and NPVa) at Different ODa Cut-Point Values