Comparison of the prevalence rates of HIV infection between men who have sex with men (MSM) and men in the general population in sub-Saharan Africa: a systematic review and meta-analysis

The inclusion criteria for the studies were as follows: (1) the study population comprised MSM populations in sub-Saharan Africa aged 18 years and over; (2) exposure was defined as having voluntary and consensual sexual intercourse with a man at least once in the 12 months prior to the study; (3) the outcome sought was an HIV infection rate whose frequency measurement was expressed as prevalence and was based on actual HIV tests conducted in the course of the study; (4) they took place in a country where Demographic Health Surveys (DHS) provided information on the prevalence of HIV infection for men from the general population within a few years of the MSM study; and (5) the study design was cross-sectional or had a cross-sectional component of recruitment in a longitudinal study. General population HIV prevalence estimates for a given country were all based on contemporary DHS surveys in the same country. The latter included all men, and therefore could include some MSM. There were no period restrictions or year limitations. All studies with a sample size of less than 50 subjects and/or self-reported HIV infection were excluded. We found that studies with sample sizes of less than 50 subjects were not sufficiently precise to be included and we excluded those with self-reported HIV status to avoid information bias, be more specific and facilitate comparisons.

The search was carried out using the following electronic databases: PubMed, EMBASE, Cochrane, Web of Science, Scopus and Google Scholar. The PICOS (Population Intervention Comparison Outcomes Study design) approach was used in this search strategy (Additional file 1). The following keywords were employed in this search: “HIV,” “prevalence,” “men,” “men who have sex with men,” and “sub-Saharan Africa.” The published studies included did not deal with both the MSM and general male population simultaneously (Additionnel file 2). They provided the HIV infection prevalence among MSM. In an approach similar to that of Baral et al. [6], we consulted the Macro International database for Demographic and Health Surveys in each country in sub-Saharan Africa (http://​dhsprogram.​com). This led to estimates of the prevalence of HIV infection among men in the general population. For each study included, the DHS from its corresponding country was selected: this DHS was contemporary to the study on the prevalence of HIV among MSM (most often the same year or within one to 3 years of the selected MSM study). Otherwise, the grey literature was consulted, mainly UNAIDS and WHO periodic reports, but also monitoring reports of HIV interventions at country level, periodic progress reports on the fight against HIV in different countries, and reports on the national census of populations at a countrywide level to complete the information necessary to meet the objectives of this systematic review.

Two independent reviewers selected the articles based on the inclusion criteria; this was carried out in two stages. The first selection was made from the titles and abstracts of the articles. For a mutual and reproducible understanding between the two reviewers, a pilot test was carried out on approximately fifteen samples before the first selection (5% of the 299 articles randomly drawn after the elimination of duplicates). The inter-reviewer agreement represented by the Kappa coefficient was 74%. At the end of the first selection, 104 articles were selected out of the 299 initially pre-selected: the Kappa coefficient for the two reviewers was 92%. The second selection was carried out by reading the full text of the pre-selected 104 articles. The only articles selected were those that fully and completely met the inclusion criteria. This second selection was also preceded by a pilot test: it was carried out on 5 articles (5% of 104). The Kappa coefficient obtained was 85%. A summary of the key points was made to validate the final selection of articles by the two reviewers. The actual selection made it possible to definitively retain 17 articles; the inter-reviewer agreement was 71%. At each stage of the selection process, disagreements were resolved through discussion and consensus between the two reviewers (Fig. 1).

Data collection was carried out using a previously designed tool (Additional file 3). It was used to collect the following information: the identification of studies, target population, method used, software employed for data analysis, and results. This extraction was carried out independently by two (2) reviewers. Disagreements between the two were usually resolved through discussion based on the information to be retrieved as found in the data collection guide. The assessment of the risks of bias was performed using the Mixed Methods Appraisal Tool (MMAT) [8].

For each MSM study included, we used HIV prevalence as reported for this population. HIV prevalence estimates in the general male population were taken from the results of contemporary DHS surveys in the same country (Table 1). To estimate the uncertainty associated with these measurements, estimates of the prevalence of HIV among MSM and men in the general population were used for each country with 95% confidence intervals (95% CI). The prevalence ratio (PR) was calculated by dividing HIV prevalence among MSM by that of men in the general population. Using meta-analysis, the pooled PR estimates of all the countries included were made using random-effect models. The heterogeneity between studies was assessed by examining the forest plot and Higgins I² statistic. An I² greater than 50% suggests high heterogeneity. Subgroup analysis was performed using PR estimates pooled according to the different geographic regions of sub-Saharan Africa (western, central, eastern and southern). Another subgroup analysis was performed according to the UNAIDS classification of the HIV epidemic level in the countries (mixed epidemic = prevalence < 1%; generalized epidemic = prevalence 1–5%; hyper-endemic = prevalence > 5%) [2, 42‐44]. It is likely that MSM was included in some samples of men of childbearing age in the general population. Sensitivity analysis was performed to assess the impact of misclassification of exposure. This sensitivity analysis was carried out according to the approach used by Baral and al [6].. For each country, the total population of MSM (estimated from the sample) was subtracted from the general population of men of childbearing age. The prevalence of HIV infection among MSM was recalculated for a hypothetical population where MSM did not contribute to the prevalence of HIV in the general population. Descriptive analysis and meta-analysis were performed using the statistical software SAS version 9.1 and Revman 5.

In general, access to MSM in many African countries has been difficult, particularly in terms of participation in studies because of discrimination and/or criminalization of their sexual orientation. The problems of participants‘safety and security in certain contexts would result in low levels of self-identification among MSM [6, 15, 55]. These barriers have likely limited the number and quality of studies and the availability of data on MSM in many countries in sub-Saharan Africa. Normally, the correct and efficient use of the RDS technique as a sampling method would help to partially solve this problem. Nevertheless, the literature also shows that RDS can give biased estimates when the recruitment does not completely follow the RDS principles in practice. This could be another limitation of our analysis [49‐53].

Some of the studies included in this analysis used convenience sampling or a cross-sectional design and, therefore, may not be representative of MSM. To determine the risk of HIV infection among MSM in countries in sub-Saharan Africa, we used the DHS estimates of the prevalence in the general male population for each country as the unexposed population to calculate PR. However, this approach does not make the populations comparable and may have affected the validity of our results. MSM tend to congregate in urban areas, explaining, at least in part, why most of the reported studies were urban. This may limit the generalizability of the studies. Although 82.35% of the authors used the RDS method for sampling subjects, only half of them produced weighted data that considered the overall contribution of each referral chain [44, 56, 57]. This situation may affect the representativeness of the MSM in the studies. Otherwise, MSM are included in men of the general population. This could have increased estimates of HIV infection prevalence in the general population of men.

Selection bias is also possible because our research did not extend to unpublished articles and conference abstracts, which could affect the results of our systematic review. This could be partly responsible for the degree of associations observed in our study. To minimize this bias, DHS data and annual progress reports were sought to complement and refine the information provided by conventional databases.

The extent to which MSM are included, excluded or unidentified in these national estimates affects both their validity and the ability of our study to compare the prevalence of HIV among MSM with that among men in the general population. This may result in a classification bias or even an underestimation of the final result. However, the sensitivity analysis carried out for this purpose showed similar results in both groups.