Background: Several approaches have been used for measuring HIV incidence in large areas, yet each presents specific challenges in incidence estimation.
Methodology/principal findings: We present a comparison of incidence estimates for Kenya and Uganda using multiple methods: 1) Epidemic Projections Package (EPP) and Spectrum models fitted to HIV prevalence from antenatal clinics (ANC) and national population-based surveys (NPS) in Kenya (2003, 2007) and Uganda (2004/2005); 2) a survey-derived model to infer age-specific incidence between two sequential NPS; 3) an assay-derived measurement in NPS using the BED IgG capture enzyme immunoassay, adjusted for misclassification using a locally derived false-recent rate (FRR) for the assay; (4) community cohorts in Uganda; (5) prevalence trends in young ANC attendees. EPP/Spectrum-derived and survey-derived modeled estimates were similar: 0.67 [uncertainty range: 0.60, 0.74] and 0.6 [confidence interval: (CI) 0.4, 0.9], respectively, for Uganda (2005) and 0.72 [uncertainty range: 0.70, 0.74] and 0.7 [CI 0.3, 1.1], respectively, for Kenya (2007). Using a local FRR, assay-derived incidence estimates were 0.3 [CI 0.0, 0.9] for Uganda (2004/2005) and 0.6 [CI 0, 1.3] for Kenya (2007). Incidence trends were similar for all methods for both Uganda and Kenya.
Conclusions/significance: Triangulation of methods is recommended to determine best-supported estimates of incidence to guide programs. Assay-derived incidence estimates are sensitive to the level of the assay's FRR, and uncertainty around high FRRs can significantly impact the validity of the estimate. Systematic evaluations of new and existing incidence assays are needed to the study the level, distribution, and determinants of the FRR to guide whether incidence assays can produce reliable estimates of national HIV incidence.