Comparing modelled HIV incidence estimates with empirical HIV incidence observations in high-burden HIV African epidemic settings: systematic review and meta-regression

BackgroundHIV incidence in sub-Saharan Africa has declined substantially since 2000 according to epidemic estimates published by UNAIDS. These estimates, derived by fitting mathematical models to national surveillance data, guide HIV programmes and epidemic response strategies. We assessed whether the level and age distribution of HIV incidence from modelled estimates were consistent with empirical HIV incidence observations, and whether incidence levels and trends were systematically different between study types, populations, and age groups. MethodsWe conducted an updated systematic review of adult HIV incidence data from sub-Saharan Africa published July 2019-February 2024 by searching Scopus, PubMed, Embase, and OVID databases, and combined with earlier systematic review data. We matched empirical incidence measurements between 1990-2023 to UNAIDS HIV incidence estimates by study area, sex, age group, and year. We used Bayesian mixed-effect Poisson regression to estimate (1) incidence rate ratios (IRR) between empirical observations and matched modelled incidence estimates adjusted for sex, year and study type/population; and (2) time trends in age-specific incidence from population-based cohort studies and household surveys. Results3560 HIV empirical incidence measurements were included from 179 studies conducted in 21 countries, comprising 23,000 new infections and 3.1 million person-years. Incidence observations from nationally-representative household surveys (IRR 1.07 95%CI 0.68, 1.67) and population-representative study populations (IRR 0.98 95%CI 0.51, 1.89) were not significantly different from matched modelled estimates, and declined at the same rate as modelled estimates (annual aRR 0.99 95%CI 0.98, 1.01). Studies among pregnant women (IRR 2.60 95%CI 1.58, 4.28), control arms of clinical trials (IRR 3.01 95%CI 1.90, 4.77) and key populations (FSW IRR: 6.46 95%CI 4.18, 10.00; MSM 44.02 95%CI 27.35, 70.87) had significantly higher incidence than modelled total population incidence estimates. Across population cohorts in Eastern and Southern Africa, HIV incidence among adults aged 15-49 declined by 75-90% between 2010-2023, and declined 7% (95%CI 4-10%) faster per year among young adults 15-24 compared to age 25+ years. Modelled incidence declined similarly to cohort data, but did not reflect the aging of the epidemic. ConclusionObserved incidence in population-representative studies in sub-Saharan Africa has declined steeply. Mathematical models that infer incidence from cross-sectional HIV surveillance data estimated the same incidence level and decline over time as population-representative studies. Studies with non-representative inclusion criteria had significantly higher incidence, including those among pregnant women and most HIV prevention/vaccine efficacy trials. The age pattern of incidence in modelled estimates should be reconsidered to capture the aging of the epidemic indicated by cohort studies.