Background: Country level comparisons of HIV prevalence among men having sex with men (MSM) is challenging for a variety of reasons, including differences in the definition and measurement of the denominator group, recruitment strategies and the HIV detection methods. To assess their comparability, self-reported data on HIV diagnoses in a 2010 pan-European MSM internet survey (EMIS) were compared with pre-existing estimates of HIV prevalence in MSM from a variety of European countries.
Methods: The first pan-European survey of MSM recruited more than 180,000 men from 38 countries across Europe and included questions on the year and result of last HIV test. HIV prevalence as measured in EMIS was compared with national estimates of HIV prevalence based on studies using biological measurements or modelling approaches to explore the degree of agreement between different methods. Existing estimates were taken from Dublin Declaration Monitoring Reports or UNAIDS country fact sheets, and were verified by contacting the nominated contact points for HIV surveillance in EU/EEA countries.
Results: The EMIS self-reported measurements of HIV prevalence were strongly correlated with existing estimates based on biological measurement and modelling studies using surveillance data (R(2)=0.70 resp. 0.72). In most countries HIV positive MSM appeared disproportionately likely to participate in EMIS, and prevalences as measured in EMIS are approximately twice the estimates based on existing estimates.
Conclusions: Comparison of diagnosed HIV prevalence as measured in EMIS with pre-existing estimates based on biological measurements using varied sampling frames (e.g. Respondent Driven Sampling, Time and Location Sampling) demonstrates a high correlation and suggests similar selection biases from both types of studies. For comparison with modelled estimates the self-selection bias of the Internet survey with increased participation of men diagnosed with HIV has to be taken into account. For most countries self-reported EMIS prevalence is higher than measured prevalence, which is likely due to a combination of different time points of measurement, measurement errors for small sample sizes, different sampling methods, and an indicator-inherent overestimate of prevalence among the untested fraction of MSM.