Identification and validation of a multi-assay algorithm for cross-sectional HIV incidence estimation in populations with subtype C infection.

INTRODUCTION: Cross-sectional methods can be used to estimate HIV incidence for surveillance and prevention studies. We evaluated assays and multi-assay algorithms (MAAs) for incidence estimation in subtype C settings.

METHODS: We analysed samples from individuals with subtype C infection with known duration of infection (2442 samples from 278 adults; 0.1 to 9.9 years after seroconversion). MAAs included 1-4 of the following assays: Limiting Antigen Avidity assay (LAg-Avidity), BioRad-Avidity assay, CD4 cell count and viral load (VL). We evaluated 23,400 MAAs with different assays and assay cutoffs. We identified the MAA with the largest mean window period, where the upper 95% confidence interval (CI) of the shadow was <1 year. This MAA was compared to the LAg-Avidity and BioRad-Avidity assays alone, a widely used LAg algorithm (LAg-Avidity <1.5 OD-n + VL >1000 copies/mL), and two MAAs previously optimized for subtype B settings. We compared these cross-sectional incidence estimates to observed incidence in an independent longitudinal cohort.

RESULTS: The optimal MAA was LAg-Avidity <2.8 OD-n  +  BioRad-Avidity <95% + VL >400 copies/mL. This MAA had a mean window period of 248 days (95% CI: 218, 284), a shadow of 306 days (95% CI: 255, 359), and provided the most accurate and precise incidence estimate for the independent cohort. The widely used LAg algorithm had a shorter mean window period (142 days, 95% CI: 118, 167), a longer shadow (410 days, 95% CI; 318, 491), and a less accurate and precise incidence estimate for the independent cohort.

CONCLUSIONS: An optimal MAA was identified for cross-sectional HIV incidence in subtype C settings. The performance of this MAA is superior to a testing algorithm currently used for global HIV surveillance.