False-positive HIV screening tests in pregnancy may lead to unnecessary interventions in labor. In 2014, the Centers for Disease Control and Prevention released a new algorithm for HIV diagnosis using a fourth-generation screening test, which detects antibodies to HIV as well as p24 antigen and has a shorter window period compared with prior generations. A reactive screen requires a differentiation assay, and supplemental qualitative RNA testing is necessary for nonreactive differentiation assay. One screening test, the ARCHITECT Ag/Ab Combo assay, is described to have 100% sensitivity and >99% specificity in nonpregnant populations; however, its clinical performance in pregnancy has not been well described.

The objective of the study was to determine the performance of the ARCHITECT assay among pregnant women at a large county hospital and to assess whether the relative signal-to-cutoff ratio can be used to differentiate between false-positive vs confirmed HIV infections in women with a nonreactive differentiation assay.

This is a retrospective review of fourth-generation HIV testing in pregnant women at Parkland Hospital between June 1, 2015, and Jan. 31, 2017. We identified gravidas screened using the ARCHITECT Ag/Ab Combo assay (index test), with reflex to differentiation assay. Women with reactive ARCHITECT and nonreactive differentiation assay were evaluated with a qualitative RNA assay (reference standard). We calculated sensitivity, specificity, predictive value, and false-positive rate of the ARCHITECT screening assay in our population and described characteristics of women with false-positive HIV testing vs confirmed infection. Among women with a nonreactive differentiation assay, we compared interventions among women with and without a qualitative RNA assay result available at delivery and examined relative signal-to-cutoff ratios of the ARCHITECT assay in women with false-positive vs confirmed HIV infection.

A total of 21,163 pregnant women were screened using the ARCHITECT assay, and 190 tested positive. Of these, 33 of 190 (17%) women had false-positive HIV screening tests (28 deliveries available for analysis), and 157 of 190 (83%) had confirmed HIV-1 infection (140 available for analysis). Diagnostic accuracy of the ARCHITECT HIV Ag/Ab Combo assay in our prenatal population (with 95% confidence interval) was as follows: sensitivity, 100% (97.7–100%); specificity, 99.8% (99.8–99.9%); positive likelihood ratio, 636 (453–895); negative likelihood ratio, 0.0 (NA); positive predictive value, 83% (77–88%); and false positive rate, 0.16% (0.11–0.22%), with a prevalence of 7 per 1000. Women with false-positive HIV testing were younger and more likely of Hispanic ethnicity. A qualitative RNA assay (reference standard) was performed prenatally in 24 (86%) and quantitative viral load in 22 (92%). Interventions occurred more frequently in women without a qualitative RNA assay result available at delivery, including intrapartum zidovudine (75% vs 4%, P = .002), breastfeeding delay (75% vs 8%, P = .001), and neonatal zidovudine initiation (75% vs 4%, P = .002). The ARCHITECT signal-to-cutoff ratio was significantly lower for women with false-positive HIV tests compared with those with established HIV infection (1.89 [1.27, 2.73] vs 533.65 [391.12, 737.22], respectively, P < .001).

While the performance of the fourth-generation ARCHITECT HIV Ag/Ab Combo assay among pregnant women is comparable with that reported in nonpregnant populations, clinical implications of using a screening test with a positive predictive value of 83% in pregnancy are significant. When the qualitative RNA assay result is unavailable, absence of risk factors in combination with an ARCHITECT HIV Ag/Ab assay S/Co ratio <5 and nonreactive differentiation assay provide sufficient evidence to support deferral of unnecessary intrapartum interventions while awaiting qualitative RNA results.