Background:  Operating characteristics of STARHS assays, including their combined use, are not well described in Sub-Saharan Africa. We assessed the sensitivity and specificity of the BED-CEIA and avidity index assays in female sex workers in Kigali, Rwanda.

Methods:  In a cross-sectional survey, 800 female sex workers of unknown HIV status were HIV tested. HIV⁺ women were tested by BED-CEIA and avidity index (using AxSYM HIV-1/2gO ELISA) at baseline and ≥12 months later to estimate assay false-recent rates. HIV^– women (N=396) were enrolled in a prospective HIV seroconversion study, where BED-CEIA and avidity index sensitivity and specificity were calculated using 52 serial specimens from 19 seroconverters. HIV incidence was estimated for the cross-sectional sample from STARHS results (based on recent HIV infection classification by BED-CEIA, avidity index, or BED-CEIA and avidity index), and in the prospective cohort. STARHS incidence estimates were adjusted with assay false-recent rates and with CD4 results (excluding <500 cells/µL as long-term infection).

Results:  The overall observed prospective HIV incidence rate was 3.0 infections per 100 person-years (95%CI 1.2 to 4.7); incidence in the first and second 6 months of the cohort were 4.7 (1.6 to 7.7) and 1.1 (0 to 2.7), respectively. Assay sensitivity and specificity were 100% (95%CI 88 to 100%) and 43% (27 to 61) for BED-CEIA; 92% (74 to 98) and 43% (27 to 61) for avidity index; and 91% (72 to 97) and 55% (33 to 67) for a combined BED-CEIA/avidity index algorithm. In the cross-sectional survey, 190 women tested HIV⁺; 23 (12%) were classified as recent HIV infection. Assay false-recent rates, representing longer-term specificity, were 6.4% for BED-CEIA; 16.3% for avidity index; and 2.8% for BED-CEIA/avidity index combined. After false-recent rates-adjustment, incidence estimates by BED-CEIA, avidity index, and BED-CEIA/avidity index combined were 9.8 per 100 person-years (95%CI 6.6 to 13.0); 11.1 per 100 person-years (8.2 to 14.0); and 5.9 per 100 person-years (3.5 to 8.3), respectively. After CD4-adjustment, incidence estimates by BED-CEIA, avidity index, and BED-CEIA/avidity index were 7.2 per 100 person-years (4.0 to 10.3), 8.2 per 100 person-years (5.1 to 11.4), and 5.1 per 100 person-years (2.8 to 7.5), respectively.

Conclusions:  In this sample of Rwandan female sex workers, adjusted incidence estimates based on a combined BED-CEIA/avidity index algorithm were similar to the observed HIV incidence rate in the first 6 months of the cohort, when incidence was highest. Furthermore, assay specificity in panel specimens around the time of seroconversion was low, but improved markedly with time since seroconversion. Given this, specificity would be expected to be substantially higher in population-based testing where a larger proportion of individuals have long-term infection.