Evaluation of multi-assay algorithms for identifying individuals with recent HIV infection: HPTN 071 (PopART)

Abstract

Background Assays and multi-assay algorithms (MAAs) have been developed for population-level cross-sectional HIV incidence estimation. These algorithms use a combination of serologic and/or non-serologic biomarkers to assess the duration of infection. We evaluated the performance of four MAAs for individual-level recency assessments. Methods Samples were obtained from 220 seroconverters (infected <1 year) and 4,396 non-seroconverters (infected >1 year) enrolled in an HIV prevention trial (HPTN 071 [PopART]); 28.6% of the seroconverters and 73.4% of the non-seroconverters had HIV viral loads ≤400 copies/mL. Samples were tested with two laboratory-based assays (LAg-Avidity, JHU BioRad-Avidity) and a point-of-care assay (rapid LAg). The four MAAs included different combinations of these assays and HIV viral load. Seroconverters on antiretroviral treatment (ART) were identified using a qualitative multi-drug assay. Results The MAAs identified between 54 and 100 (25% to 46%) of the seroconverters as recently-infected. The false recent rate of the MAAs for infections >2 years duration ranged from 0.2%-1.3%. The MAAs classified different overlapping groups of individuals as recent vs. non-recent. Only 32 (15%) of the 220 seroconverters were classified as recent by all four MAAs. Viral suppression impacted the performance of the two LAg-based assays. LAg-Avidity assay values were also lower for seroconverters who were virally suppressed on ART compared to those with natural viral suppression. Conclusions The four MAAs evaluated varied in sensitivity and specificity for identifying persons infected <1 year as recently infected and classified different groups of seroconverters as recently infected. Sensitivity was low for all four MAAs. These performance issues should be considered if these methods are used for individual-level recency assessments.

Evaluation of multi-assay algorithms for cross-sectional HIV incidence estimation in settings with universal antiretroviral treatment

“…Only 24 (11.0%) of the seroconverters were classified as recently infected by all three MAAs. This is consistent with a prior report that showed that different MAAs identify different subsets of seroconverters as recently infected [ 7 ]; the different classification of individual seroconverters as recently infected was not solely due to exclusions for viral load. In this study, 80 of the seroconverters were classified as recent by the LAg + BR MAA.…”

“…Multi-assay algorithms (MAAs) have been developed for estimating population-level HIV incidence using samples collected in cross-sectional surveys [ 1 – 3 ]. These algorithms have also been used to identify individuals with recent HIV infection [ 4 – 7 ]. The most widely used MAA includes the Limiting Antigen Avidity assay (LAg-Avidity assay) and HIV viral load (LAg + VL MAA) [ 8 – 10 ].…”

“…For these reasons, accurate HIV incidence estimation may require use of MAAs that do not include viral load. Even when viral load is not included as a biomarker, development of MAAs for use in settings with universal ART is challenging, since viral suppression also impacts the performance of the LAg-Avidity assay and other serologic incidence assays [ 7 , 13 – 16 ]. This is because the level and avidity of anti-HIV antibodies often decline when persons are virally suppressed on ART [ 17 ].…”

“…We recently evaluated the impact of universal ART on the performance of the LAg + VL MAA [ 7 , 18 ] using samples and data from the HIV Prevention Trials Network (HPTN) 071 (PopART) trial [ 19 ]. HPTN 071 (PopART) was a community-randomized trial conducted in Zambia and South Africa that demonstrated that universal delivery of a combination prevention intervention that included universal testing and treatment was associated with decreased HIV incidence [ 19 ].…”

“…Two years after trial onset, ~ 75% of participants who were infected more than one year and ~ 30% of those who had acquired HIV infection during the study were virally suppressed [ 19 ]. The LAg + VL MAA provided an accurate HIV incidence estimate for this cohort [ 18 ], but performed poorly for individual-level recency assessments [ 7 ].…”