This is to our knowledge the first meta-analysis of HIV survey data collected from MSM participants in low- to middle-income countries. Overall, the odds of having HIV infection are markedly and consistently higher among MSM than among the general population of adults of reproductive age across Asia, Africa, the Americas, and the FSU.
There are a number of limitations to this study. MSM in many developing countries are often difficult to access and to study because of criminalization of their behavior, the social stigma associated with their behaviors and identities, participant safety concerns in some settings, and low levels of self-identification among MSM. These barriers likely limited both the number and quality of studies in the literature—only a few lower-income countries, including Mexico, Brazil, Thailand, and Peru, have systematically surveyed MSM. The majority of studies cited in this analysis are convenience samples and cross-sectional in design, and so may not be representative of MSM. To determine a corollary of risk for HIV infection among MSM in low- and middle-income countries, we used UNAIDS general population prevalence estimates for each country as the unexposed population to compute ORs. Because of the lack of controls, issues affecting internal validity could not be formally controlled for in our study. MSM tend to congregate in urban areas, at least partially explaining why the majority of reported studies are urban; again, this may limit generalizability. In very populous countries such as China and India, there may be even more marked differences between urban and rural areas in HIV prevalence and in reporting of MSM behaviors. Publication bias tends to affect the results of meta-analyses, both in the realms of clinical and public health research, and could be partly responsible for the magnitude of associations seen in this study [35
]. To minimize the effect of publication bias, the US Census Bureau HIV database and the EuroHIV surveillance report were searched to validate the sensitivity of the journal and conference search protocols. A further limitation to this study was that it was limited to English-language publications, which could serve as a source of language bias in the results. That said, using informal searches of non-English databases, the authors found no sources of primary data that had not also been reported in English journals and indexed in PubMed. Only high-level risk factors for HIV infection are assessed in this study, and these may be subject to ecologic fallacy, meaning that these measures of association may not be applicable at the individual level. Although individual drivers of HIV acquisition and transmission among MSM have been well characterized in high-income countries, the same cannot be said for the majority of countries included in this study [36
]. Only with prospective observational and evaluative studies will it become clear if the same risk factors for HIV acquisition and transmission apply to MSM in low- and middle-income settings. Finally, a portion of the difference in ORs seen between strata may be explained by a ceiling effect. That is, a bias where the magnitude of a relative association, such as an OR, decreases as the background level increases.
MSM were likely included in some samples of men in the general reproductive-age population. This is likely the case in those settings where MSM behavior is most hidden. We conducted a sensitivity analysis to assess the importance of this misclassification of MSM. Such an approach is important in assessing the validity of the assumptions made for statistical calculations in meta-analyses [38
]. Using the prevalence of MSM behavior in each setting as calculated by Caceres et al. [39
], a sensitivity analysis was conducted by removing the total (estimated) population of MSM from the population estimate of all men of reproductive age for individual countries. We then recalculated the odds of HIV infection among MSM for a hypothetical population where MSM did not contribute to the general population HIV prevalence. This modified the overall magnitude of the OR modestly, from 1.5% to 7.5%, depending on the country, and so had little impact on our interpretation of the meta-analyses. Data and methodological quality of these studies was deemed sufficient for the purposes of this analysis, due to the fact that these studies underwent peer review or were published as government reports, with high methodological standards such as that of EuroHIV and the US CDC.
Despite these limitations, this meta-analysis draws its precision strength from the combined estimates of the OR and a large aggregate sample size of MSM (n = 63,538). By calculating a measure of association, such as an OR, one can see that two regions with identical absolute measures, such as HIV prevalence among MSM, may be in very different stages of the HIV epidemic affecting the overall risk status of MSM in that region. Due to the significant heterogeneity (χ2 = 7,845.81) of the ORs of HIV infection among MSM from differing countries, one pooled OR describing the HIV risk of MSM globally is likely not valid as an accurate measure of risk. Rather, the value of these analyses is in the overall trends of the results. These trends of high HIV prevalence among MSM in the context of low-level or concentrated HIV epidemics speak to the urgent need for increased targeted prevention strategies to this at-risk population in low- and middle-income countries.
To determine if there is a differential risk status of MSM depending on the level of the HIV epidemic in given country, we stratified the pooled OR by the prevalence level of the epidemic (very low, low, and medium–high; ). There was a trend of decreasing OR with increasing general population prevalence with an OR of 58.4 in very low-prevalence countries, 14.4 in low-prevalence countries, and 9.6 in medium- to high-prevalence countries. Subgroup analysis evaluating differences in OR by income level showed an OR for HIV infection of 23.4 for middle-income countries and 7.8 for low-income countries. Given that low-income countries in this study had generally higher general population prevalence rates, these results may represent a consistent increase in odds of HIV among MSM across income levels given the potential of a ceiling effect. As more data become available, it will be important to determine to what extent poverty directly or indirectly affects epidemics of HIV among MSM. The marked differences in OR by prevalence or income level may be a function of epidemic stage: in countries with higher prevalence among adults of reproductive age, HIV transmission may be linked through sexual networks between high-prevalence general populations and MSM. In countries with very low prevalence in general populations, HIV transmission among MSM may be isolated and propagated within this group in a dislinked fashion.
To control for the assumption that prevalence level categories are more relevant than epidemic levels in assessing the relative increase in odds of HIV among MSM, pooled estimates were stratified using both criteria. Stratification by UNAIDS-defined epidemic level showed that the odds of being HIV positive remained high among MSM in countries with generalized epidemics (OR 10.8), and was even higher in countries with low-level epidemics (OR 24.5) or concentrated epidemics (OR 23.5) (). The UNAIDS classification of HIV epidemics was designed, in part, to provide guidance on the type of surveillance that should be conducted in a country. However, the absence of a difference in the odds of HIV infection among MSM between concentrated and low-level epidemics suggests that this classification system is currently not ideal for measuring the increased risk of specific subsets of the population. The accuracy of HIV epidemic levels may be improved as more comprehensive prevalence data of specific vulnerable populations such as MSM become available.
The direction of the measure of association among MSM appears to be quite consistent between individual countries, geographic regions, and epidemic states, highlighting the external validity of the individual studies. Eastern Europe appears to be an exception: MSM data are scarce, and the region's HIV epidemics are primarily driven by IDU exposure. No peer-reviewed published report or abstract meeting our inclusion criteria was found in Eastern Europe. The most recent EuroHIV surveillance report served as the primary source for these data. Since an unknown but potentially significant number of MSM in this region may also be IDUs, estimating the attributable risk fraction for these differing behaviors is difficult. What is clear is the need for more effort to characterize the risks for MSM in this region.
The stratification of the pooled OR estimate revealed some general differences in risk status between MSM globally. The highest OR for HIV infection was found in the Americas, at 33.3. It was lower, but still extremely high, in Asia at 18.7, lower still in Africa at 3.8, and lowest in Eastern Europe at 1.3. The relatively outlying result from Eastern Europe is likely due, as we have argued, to comparing MSM with populations where IDUs are the main driver of HIV. The very high rates in the Americas and Asia were by far the best evidenced, suggesting that these epidemics among MSM are real, and that these men are indeed at markedly greater risk than heterosexuals in these settings. Data regarding MSM in Africa were the sparsest, but are beginning to emerge. Recent reports of HIV risks (if not rates) among MSM were found from Uganda, Zambia, Sudan, and Nigeria, though not all met inclusion criteria for this analysis [40
]. These epidemics appear to be driven, in part, by marked stigma and homophobia in these settings and by a lack of specific prevention strategies. Although these data indicate that these MSM populations are in desperate need of targeted prevention campaigns, social intolerance currently limits prevention efforts. UNAIDS estimates that in 2005, fewer than one in 10 MSM globally had access to appropriate HIV prevention services [1
These results constitute a clear call to action on three fronts: surveillance, research, and prevention [39
]. The various subgroup analyses completed for this study may not necessarily explain complex differences in global HIV epidemic dynamics, but they do demonstrate that high HIV prevalence rates among MSM are not limited to any one epidemic level, prevalence category, region, or income level. HIV surveillance efforts should take into account the high burden of HIV among MSM and expand surveillance to include them in countries where they are not now included. Social science, epidemiologic, and behavioral research should use population-based sampling methods and standardized data collection tools to assess prevalence of HIV risk behaviors, knowledge about HIV, and social and sexual network interactions, and the roles individual and partner circumcision status may play in male-to-male HIV transmission and acquisition dynamics. Ethnographic assessments could further describe the cultural and behavioral nuances of MSM globally and refine data collection instruments. Human rights advocacy and cessation of discrimination against MSM could afford greater access to HIV prevention and education services and are an urgent priority in much of the world. Male-to-male sexual contact is not inherently dangerous; only in the context of an advanced stage of the epidemic and lack of preventive measures is this actually high-risk behavior for HIV infection. Notably, there exists a risk that demonstrating high HIV prevalence rates among MSM will further contribute to stigma. However, prevention expenditures are generally allocated based on need; thus, the risk of increasing stigma must be balanced by the potential benefits of successfully advocating for dedicated funding resources for MSM. In Asia, prevention expenditures targeting MSM range from nearly 0% in portions of China to a high of 4% in Thailand [43
]. This lack of governmental expenditures is notable given that two recent meta-analyses have demonstrated that prevention and harm reduction strategies targeting MSM are successful in decreasing high-risk behaviors [17
]. MSM have been largely ignored by both social and public health structures in many countries for too long, given their highly disproportionate burden of HIV. Surveillance, research, and prevention efforts should work together to begin to curb HIV transmission in this marginalized population.