The burden of HIV infection among men who purchase sex in low- and middle-income countries – a systematic review and meta-analysis

Background Since the start of the HIV epidemic, transactional sexual relationships have been considered to present a high risk of HIV transmission to both the client and the person offering the sexual service. However, prevention research and programs have focused predominantly on sex workers rather than on their clients, who are generally men. To support effective and targeted interventions, we undertook a systematic review and meta-analysis of the evidence of the prevalence of HIV infection among men who purchase sex (MWPS) in low- and middle-income countries (LMICs), and the association between HIV infection and purchase of sex. Methods We included articles that reported from LMICs on the prevalence of HIV in MWPS and those that reported on HIV prevalence among both MWPS and non-MWPS in the same study, or any information which allowed calculation of the prevalence. We defined MWPS as heterosexual males (not men who purchase sex or individuals of other sexual orientation) who purchased sex mostly from women (and not men), or who have had sexual contact with female sex workers (FSWs). We searched Medline, Global Health, Scopus, Embase and Cinahl for articles published up until 1 March 2020. Meta-analysis was conducted using a random effects model to estimate the pooled HIV prevalence and the relative risk (RR) of HIV infection associated with purchasing sex. Results Of 34862 studies screened, we included 44 studies (59515 men, 47753 MWPS) from 21 countries. The pooled HIV prevalence among MWPS was 5% (95%CI: 4%-6%; I2 = 95.9%, p < 0.001). The pooled HIV prevalence calculated from studies that reported data collected pre-2001 was highest, i.e. 10% (95% CI: 6%-14%; I2 = 91.2%, p < 0.001), compared to studies whose data was collected between 2001–2010, i.e. 4% (95%CI: 2%-6%; I2 = 96.6%, p < 0.001), and from 2011 and beyond, i.e. 3% (95% CI: 2%-5%; I2 = 94.3%, p < 0.001). For studies which included comparisons of HIV infection among MWPS and non-MWPS, the relative risk of HIV infection was consistently higher among MWPS than among non-MWPS within the same study, with the overall pooled relative risk of 1.95 (95%CI: 1.56–2.44; I2 = 84.3%, p < 0.001), and 2.85 (95%CI: 1.04–7.76; I2 = 86.5%, p < 0.001) for more recent studies. Conclusions This review represents the first comprehensive assessment of the burden of HIV among MWPS in LMICs. We found that HIV prevalence was elevated compared to the population as a whole, and that there was a strong association between purchasing sex and HIV prevalence. Despite a reduction over time in prevalence, these data highlight that MWPS need better access to HIV preventive interventions.


Introduction
Transactional sex, defined as the exchange of sexual services for money or goods [1], has been considered to be a major risk factor for HIV transmission since the epidemic began in the early 1980s. However, HIV control programs have targeted predominantly people who provide transactional sex, particularly women, as a "risk group" and more recently as one of several "key populations" to be prioritized for prevention and treatment services [2]. Although focusing HIV interventions on these groups, mainly females sex workers (FSWs), has proven its effectiveness in curbing HIV transmission during transactional sex activities [3], it is also important to reduce HIV infections in men who purchase sex (MWPS), who often have more power in decision making about condom use with FSWs [4]. In addition to potential HIV exposure from purchased sexual contact, a number of studies have documented other risk behaviours among MWPS, such as having multiple sexual partners [5][6][7][8][9], including with men [10], high rates of STIs [5][6][7][8][9]11], and injecting drug use [12,13]. Improving our understanding of the HIV burden among MWPS in order to tailor effective interventions targeting this group is therefore important, particularly in light of the recent UNAIDS commitment to reaching out to men in the HIV responses [14], and recent initiatives to expand access to HIV diagnosis and treatment in this group [15].
To provide a global picture of HIV prevalence, systematic reviews have been conducted among men who have sex with men (MSM) in low-and middle-income countries (LMICs) [16,17], trans and gender diverse people [18], people who inject drugs [19], and prisoners and detainees [20]. Reviews have also summarized the HIV burden among providers of transactional sex, both women in LMICs [21] and men [1]. However, little attention has been focused on the HIV burden among MWPS. The question even remains as to whether it is appropriate to consider this group a "key population" in HIV programming.
While there have been regional systematic reviews of HIV prevalence among MWPS in China, the Middle East and North Africa, and West and Central Africa [22-24], a broaderscope systematic review is needed to assess the current evidence for HIV risk among MWPS in LMICs. To respond to this need, we conducted the first systematic review and meta-analysis of HIV among MWPS in LMICs, focusing on HIV prevalence and relative risk. A further objective was to provide an updated estimate of the extent to which purchasing sex remains a risk factor for HIV infection. Another goal of this systematic review was also to identify interventions that aimed to reduce HIV infection among MWPS; those results have been presented in a separate paper [25].

Material and methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [26] was used to guide the development of the protocol and reporting the study.
In this systematic review and meta-analysis, we sought articles that reported on the prevalence of HIV in MWPS, and also those which reported HIV prevalence among both MWPS and men who did not purchase sex (non-MWPS) in the same study; or articles with any information which allowed us to calculate the prevalence. We defined MWPS as heterosexual males (not MSM or individuals with other sexual orientation) who purchase sex mostly from women (and not males), or those who have had sexual contact with FSWs. Studies published up until 1 March 2020 were eligible.
A search was performed in Medline, Global Health, Scopus, Embase and Cinahl. The search strategy used three concepts of MESH and non-MESH key terms to cover FSWs, heterosexual men, and HIV infection. In addition, terms were used to identify male-dominated occupational groups associated with travel or extended periods of work-related separation from family, and often exposed to transactional sex [27] ( Table 1).
Articles were excluded if they were letters, case series, commentaries, or editorials, the full text was not available, the full text was not in English, the study was a review or modelling study, categorisation of transactional sex contact failed to distinguish between purchasing and selling sex, the study did not break down participants by sex, HIV prevalence was not reported separately from other STIs, insufficient information to calculate the prevalence and risk, not peer-reviewed, or conducted in high-income countries. The exclusion criteria based its definition of a high income country on the World Bank's 2019-2020 classification [28].
All articles found through the search were exported into EndNote. Duplicates within and between databases were removed in EndNote, followed by visual inspection. One author (LPLW) independently screened the titles and abstracts of articles, and the full texts of potentially eligible articles were obtained and further assessed for final inclusion. Finally, the full texts of the remaining articles were used to extract the main information needed. RG and JK checked the data extracted and discrepancies were discussed among the authors with disagreements resolved by consensus. The study was conducted between September 2015 and March 2020.

Summary of studies identified
Overall, 34862 articles were retrieved and 7862 duplicates removed, resulting in 27000 articles to be screened by title or abstract. Of these, 128 were considered relevant to our research questions and screened for a full text review. Among these 128 articles, 84 were further excluded due to: reporting the same studies (7 studies); no full text (24 studies); the full text was not in English (21 studies); they failed to report participants by gender (11 studies); categorise whether the men had purchased or sold sex (5 studies) or separate HIV and other STI prevalence (5 studies); had incomplete information to calculate the prevalence and relative risk (5 studies); or conducted in high income countries (6 studies). After these exclusions, 44 articles were retained for analysis. Of the 44 studies, there were 21 in which MWPS comprised the whole study sample, and 23 which included a comparison group of MWPS and non-MPWS (Fig 1).   [78]. Most studies used cross-sectional designs (40 studies; 90.9%). Studies mostly recruited participants from sex work venues (n = 18; 40.9%), sexually transmitted infection (STI)/voluntary counselling and testing (VCT) clinics (n = 13; 29.5%), and from work places (n = 10; 22.7%). Eleven studies (25%) used probability sampling techniques. Purchase of sex was mostly defined based on self-report history of purchasing sex (n = 36; 81.8%), with "ever" being the most common recall period of purchasing sex used (n = 15; 34.1%), and some (n = 13; 29.5%) based on recent history (current and the last 1-3 months). A smaller number of studies (n = 8; 18.2%) used presentation at the sex work venues to define MWPS ( Table 2).
The male sample sizes ranged from 41 to 7068. A total of 59515 men were included in the studies, of whom 47753 (80%) were MWPS (Figs 2 and 4).

HIV prevalence among MWPS
Prevalence estimates for MWPS were obtained from 21 studies (47.7%) (Figs 2 and 3), and ranged from 1% to 17%. Among the studies measuring HIV prevalence among MWPS for which UNAIDS estimates of HIV prevalence in the general adult male population in the same country were available [29], the reported HIV prevalence among MWPS was higher in all studies ( Table 3).

Relative risk of HIV infection among MWPS compared with non-MWPS
The estimated relative risk of HIV infection among MWPS compared with non-MWPS within the same population was obtained from 23 studies (52.3%) (Figs 4 and 5). The relative risk of HIV infection was consistently higher among MWPS than among non-MWPS within the same study, with relative risks associated with purchase of sex ranging from 1.02 to 13.75. The overall pooled relative risk was 1.95 (95%CI: 1.56-2.44; I 2 = 84.3%, p < 0.001). When stratified by data collection period, the pooled relative risk was 1.79 (95%CI: 1.41-2.28; I 2 = 86.1%, p < 0.001) in studies which collected their data before 2001, increasing to 2.01 (95%CI: 1.12-3.61; I 2 = 34.3%, p = 0.206) in those which collected data between 2001 and 2010, and to 2.85 (95%CI: 1.04-7.76; I 2 = 86.5%, p = 0.001) in studies from 2011 and beyond (Fig 4).
The results of sensitivitiy analysis showed that neither studies with extremely high nor extremely low HIV prevalence and RR skewed the overall summary estimate, meaning that overall summary was not particularly influenced by any single study.

Quality appraisal of studies
The quality appraisal of studies is summarised in Table 4. Challenges in appraisal included the representativeness of the target population being unclear, and that MWPS are a hidden population. Due to practicality, exposure to sex workers was only identified through observation at a sex work venue or in interviews. Study and participant characteristics ( Table 2) and the reporting characteristics of other factors that might influence the association between HIV infection and purchasing sex (Table 5) indicate the considerable diversity of methods, completeness of the study reports, and gaps in information provided by the publications.

Discussion
In the first systematic review and meta-analysis of HIV burden among MWPS in LMICs, we identified 44 studies across regions of the world, and found the pooled HIV prevalence among
We found that reported HIV prevalence was highest in studies conducted before 2001 compared to those conducted more recently. The decreasing HIV prevalence among MWPS since  [79], and a further 23% decline since 2010 [80].
The declining prevalence of HIV may also reflect the benefits of improved access to antiretroviral therapy over the past two decades since the Doha Declaration on TRIPS and Public Health adopted in November 2001 [81,82]. Although this effect might have been unevenly distributed, being contingent on country politics, local and national testing and treatment policies, and cultural differences between and within countries.
The reduction in new HIV infections may also be owed to various other initiatives and momentous changes in global HIV responses such as the introduction of policies recommending antiretroviral treatment for all people living with HIV regardless of CD4 count and preexposure prophylaxis [83] and the '90-90-90' Fast Track initiatives [84]. Numerous recent studies have recorded the benefits at both individual and population levels of improved access to antiretroviral therapy [85][86][87].
The various interventions made may have also influenced condom use among this group, which has, in turn, influenced this declining trend. Several recent studies, such as that in China [88], Indonesia [89] and Benin [38] have reported high consistent condom use, or condom use at last sex with a sex worker. In some other settings, however, condom use is apparently still quite low, as shown in a study from India [90].
Overall, the relative risk of HIV infection was consistently higher among MWPS than among non-MWPS within the same study, with a summary relative risk of 1.95 (95%CI: 1.56-2.44), and 2.85 (95%CI: 1.04-7.76) in studies conducted in more recent years. This might be due to various HIV exposure modes among this group. In addition to potential exposure to HIV from unsafe transactional sex, MWPS are more likely to have multiple partners [10,91,92], including with men, to have sex under the influence of a drug which [92] which may reduce the likelihood of condom use during transactional sex, high rates of STIs [5][6][7][8][9]11], and a history of injecting drug use [12,13]. These findings are further evidence that MWPS should be designated as a key population in many countries, and that interventions to reduce HIV risk among these men should be prioritized.
A global report has noted that there are around 8.1 million people living with HIV/AIDS (PLWH) who are not aware of their HIV status [93], many of them are men [94]; lower testing coverage among men, compared to women, has also been emphasized by the WHO [95].

PLOS ONE
Several studies reveal that the low HIV testing rates among this group [89,96] is due to many reasons, including resistance to being seen at the VCT clinic [97], logistical issues [96], low levels of knowledge about HIV and HIV testing [98], self-perception of being at low HIV risk [99], and concerns about confidentiality in the event of a positive test result [99]. In light of the low rates of HIV testing and treatment access among men [96,100]; the lack public health interventions to improve HIV testing and treatment among MWPS [25]; the 2020 global target of ensuring 90% of people living with HIV are aware of their status [84]; the UNAIDS commitment to reaching out to men to fill the gap in HIV responses [14]; and WHO's recent advocacy of supporting HIV service uptake among men [101], strategies to better reach MWPS for HIV testing and referral should be strengthened, and include targeted strategies for introducing HIV self-testing which have been recently piloted [102,103].
Caution should be taken when interpreting the results of this study. The high degree of heterogeneity found in this study was likely due to the various measures used to identify MWPS, the recall period of purchasing sex, and other differences in risk behavioural characteristics among participants. No gold standard exists for identifying MWPS, or for the recall period of history of purchasing sex, and studies included in this review used various parameters for

PLOS ONE
these. Higgins argues that heterogeneity is indeed unavoidable in any meta-analysis study given individual studies are conducted and performed by different research teams in different countries, employ different methodologies, and whose participants have varying characteristics [104]. In fact, high levels of heterogeneity have also been found in several broad-scope meta-analysis studies, for example, those conducted to estimate the HIV burden among women who engage in sex work in LMICs [21], prisoners and detainees worldwide [20], and transgender women globally [18]. Recall and social desirability biases are possible among the studies in this review since most used self-report to identify MWPS. Because of the stigma attached to purchasing sex, and because some sexual contacts might not be regarded as purchased sex, men might under-report their behaviour in this respect. Future consideration MWPS are a hidden population and thus it is difficult to assess the extent to which the participants involved in the study were representative of all MWPS in the population. For ease of access to potential participants, 38 studies recruited participants from venues frequently visited by MWPS such as sex venues and STI/VCT clinics, or recruited men from male-dominated occupational groups associated with travel or extended periods of work-related separation from family, and often exposed to transactional sex. Among these, 11 studies using a probability sampling method, improving the representativeness of the study participants. 2 Were study participants recruited in an appropriate way? 38 0 6 As above, MWPS are a clandestine population. To gain access to research participants, 38 studies recruited participants from venues frequently visited by MWPS such as sex venues and STI/VCT clinics, or recruited men from male-dominated occupational groups often exposed to transactional sex. 3 Was the sample size adequate? 21 0 23 More than half of the studies included in the analysis did not specifically aim to identify the prevalence of HIV among MWPS, or the risk associated with purchasing sex, thus were not powered specifically for these analyses. Only 21 studies were specifically conducted to identify the prevalence of HIV among MWPS.

4
Were the study subjects and the setting described in detail?
44 0 0 To satisfy this criterion, we required the study to report on sociodemographic characteristics such as age or education. Only two studies did not report participants' ages, but both of those described participants' education levels. The study setting, with respect to where the study participants were recruited from, was reported for all studies.

5
Was the data analysis conducted with sufficient coverage of the identified sample?
44 0 0 All studies tested more than 60% of men agreeing to participate. 6 Were objective, standard criteria used for the measurement of the condition?
42 0 2 Two studies did not clearly describe the HIV testing procedure they had used.

7
Was the condition measured reliably? 42 0 2 As above, two studies did not clearly describe the HIV testing procedure they had used. Among the 44 studies that described the testing procedure, all HIV status was confirmed with rapid test algorithm or laboratory test.

8
Was there appropriate statistical analysis? 44 0 0 As above, some of the studies included in the analysis were not specifically conducted to identify either the prevalence of HIV among MWPS, or the risk associated with purchasing sex, thus were not powered specifically for these analyses. However, all studies included the number or percentages of HIV positive results among those who were tested or sampled.

9
Are all important confounding factors/ subgroups/differences identified and accounted for?

0 11
To satisfy this criterion, we required a study to report HIV infection stratified by the main potential confounders: STI risk behaviours (number of partners or condom use or STI history/infection), history of having sex with men, and history of injecting drug use. Only four studies reported HIV positivity by all of the three behaviours, 29 at least one of them, and 11 none of them.

10
Were subpopulations identified using objective criteria?
42 0 2 Self and peer reporting were used to identify MWPS versus non-MWPS in 42 studies; although it might incur social desirability bias and recall bias, it is the only practical way to identify these groups. In two studies, men were assumed to be MWPS since they attended sex work venues. should be given to how best to define MWPS so as to increase comparability among studies. In addition, restricting inclusion to only those articles with full English text availability might have resulted in a language bias [31]. The possibility that some relevant studies were missed during the search is also acknowledged. Also, the missing value on I 2 and p value in some figures was likely due to the small number of studies included in the analysis. It is important as well to note that the countries from which data on HIV prevalence in this population are available may not be representative of LMICs more generally. It is therefore not possible to extrapolate the findings to LMICs as a whole. However, it is likely that the relative risks observed for purchasing sex comparing to men who did not purchase sex in the same populations are more generalizable.

Conclusions
This review represents the first comprehensive assessment of the burden of HIV among MWPS in LMICs. The strength of our meta-analysis is its combining of estimates of prevalence and relative risk from a large aggregate of MWPS and non-MWPS populations. We found that HIV prevalence was elevated compared to the male population as a whole, and that there was a strong association between purchasing sex and HIV prevalence. Despite a reduction over time in prevalence, these data highlight that MWPS need better access to preventive interventions, including HIV testing and treatment.