This is to our knowledge the first meta-analysis of HIV prevalence from three highly vulnerable populations carried out in Brazil: FSW, MSM and IDU/NIDU. Overall, the odds of having HIV infection are markedly and consistently higher among those populations than in the general population of adults of reproductive age. This latter has been stabilized around 0.6% since 2000 in Brazil [16, 18, 86].
There are a number of limitations to our study. FSW, MSM and drug users are often difficult to access and to enroll in surveys because of the social stigma associated with their behaviors and criminalization of drug use. These barriers may limited both the number and quality of studies identified, particularly those assessing FSW (n = 8) and MSM (n = 10). Drug users have been more systematically surveyed in Brazil, particularly due to surveys carried out in collaboration with international institutions, such as the World Health Organization (WHO) and the National Institute on Drug Abuse (NIDA/NIH). The majority of studies cited in this analysis used small convenience samples and were cross-sectional.
Gaps identified amongst the data and the resulting lack of accuracy of the pooled HIV/AIDS prevalence for MSM and FSW seem to be associated with the fragmentary character of sero- and behavioral surveillance targeting those populations in Brazil. This scenario is likely to be modified in the near future, with the implementation by the Brazilian Ministry of Health of biannual seroepidemiological studies targeting MSM, FSW and IDU/NIDU. The first round of such studies was conducted in 2009, but no peer-reviewed paper was published so far on such studies.
MSM, FSW and drug users tend to congregate in urban areas, at least partially explaining why the vast majority of reported studies are urban; again, this may limit the generalizability of our study findings, further complicated by the fact that most studies were carried out in large metropolitan areas of the most industrialized regions in Brazil. Publication bias tends to affect the results of meta-analyses, both in terms of clinical and public health research, and may compromise the accuracy of pooled measures .
To minimize the effect of publication bias, gray literature (e.g. national and international conferences and local reports) were included and researchers were directly contacted to obtain upcoming publications. We conducted a sensitivity analysis to assess each study impact on the pooled HIV-prevalence for each population group. Such an approach is important in assessing the validity of the assumptions made for the statistical calculations in meta-analyses [37, 88]. Studies that may have an influence on the pooled analyses were removed. Results were then compared and a few studies were included only in the systematic review in order to control for potential biases. Unfortunately information on methodological quality of selected studies was insufficient to allow a detailed analysis of their quality.
Despite these limitations, this meta-analysis draws its strength from the pooled estimates for a large aggregate sample size of drug users (N = 13,063), while the aggregate sample size for FSW and MSM were not very large (N = 3,625 and 6,475, respectively). The small number of studies assessing FSW (N = 8) and MSM (N = 10), basically analyzing data from small-scale samples precluded further analysis and lead us to conclude that our meta-analysis for those two populations would not be conclusive .
Due to the significant heterogeneity identified in our pooled HIV prevalence for FSW (I2 = 81.9), MSM (I2 = 97.1%), and drug users (I2 = 98.6%), our pooled HIV-prevalence is likely not to be valid as an accurate measure of risk. Subgroup analysis were not conducted for studies addressing FSW and MSM, due to the small number of studies addressing those populations and the lack of key variables that could guide such analyses. The subgroup analysis of studies on drug users highlighted high between-study heterogeneity. These trends of high HIV prevalence among FSW, MSM and IDU/NIDU speak in favor of the urgent need to improve targeted prevention strategies to those at-risk populations, particularly in the context of a developing country providing a comprehensive array of prevention interventions and HIV-treatment free of charge to any eligible individual living with AIDS.
The critical factor in making adequate HIV estimates in countries with low-level or concentrated epidemics is the availability of data. Countries that have serological and behavioral data for most at-risk groups may profit from better and more comprehensive estimates to inform policymaking and monitoring . Exception made to the US, a few countries from Western Europe, and African countries where many different cooperative studies have been carried out assessing different populations and geographic areas such as Uganda, similar limitations as the ones identified by our group have been observed in most countries and have been associated with gross under and overestimates, such as the ones respecting India, later corrected by more recent studies .
Comprehensive information about vulnerable populations, behaviors that place people at risk, and knowledge of the current status and trends of infection rates among those populations are an essential component of sound programmatic decisions. In this sense, our pooled analyses may help to inform public policies but also to highlight the limitations of available data in Brazil.
The small number of studies on FSW and MSM, and the lack of key information that could help to better understand between-studies heterogeneity precluded meta-regression analyses for those subpopulations . For studies addressing the drug using population, meta-regression analysis identified that injection drug users had a seven-fold increase in their risk to be HIV-infected, when compared to non-injection drug users. The measures of association among drug users seem to be consistent across injection and non-injection drug users, incarcerated and non-incarcerated participants, irrespectively of the geographic regions where they have been surveyed, the study period and the recruitment site. Those findings speak in favor of the external validity of many individual studies on drug using populations carried out in Brazil.