Background:  The aim of this study was to determine the effect of methodology on HIV incidence estimates from around the world, comparing cross sectional with longitudinally derived estimates stratified by region, risk factors, and year

Methods:  A literature review obtained the incidence estimate, geographical location, year, risk group, and method used to estimate the frequency of incident HIV. Estimates using person-time were converted to percent per year. All incidence values described are in percentage per year. We considered 5 risk groups:  women attending antenatal clinics, commercial sex workers (CSW), men who have sex with men (MSM), drug users, and sexually transmitted diseases (STD) clinic attendees. We reviewed 5 methods:  Serologic Testing Algorithm for Recent HIV Seroconversions (STARHS), BED, antibody negative/antigen positive (Ab^–/Ag⁺), estimates from HIV prevalence data, and longitudinal estimation.

Results:  The review identified 389 estimates of HIV incidence within 44 countries from 1985 to 2006 from 156 peer-reviewed journal articles. Of the 389 estimates, 243 were from longitudinal cohort studies, 100 were STARHS estimates, 20 were from the BED assay, 13 were Ab^–/Ag⁺ estimates, and 13 were estimates from prevalence data. India, Thailand, Kenya, the United Kingdom, the Netherlands, Peru, and the United States had more than 1 type of estimate for each risk group. Estimates from prevalence data were consistently 74% lower than longitudinal estimates. STARHS estimates were fairly consistent with longitudinal estimates in all countries. Ab^–/Ag⁺ tests resulted in higher HIV incidence in CSW in Thailand than longitudinal estimates (25.4% in 1995 vs 20.3% in 1994) but much lower incidence in India among CSW (Ab^–/Ag⁺: 1.5% in 1997 vs longitudinal: 16.0% in 1992, 20.5% in 1995 and 26.1 % in 2003). BED estimates were consistently higher than longitudinal estimates in Thai drug-users (17.3% in 2003 vs 10.2% in 2002) and higher among U.S. STD clinic attendees (2.5% in 2004 vs 0.88% in 2002) and U.S. MSM (3.3% in 2004 vs 1.9% in 2003). 

Conclusions:  The method used affects the incidence estimate. Incidence estimates based on STARHS and those derived from longitudinal cohort studies were similar. The BED had higher estimates than longitudinal derived data, which, in turn, are higher than prevalence based estimates. The Ab^–/Ag⁺ estimates varied in both directions compared to the longitudinally derived data.