Men who have sex with men (MSM) continue to bear the largest burden of the HIV epidemic. In 2017 MSM in the United States (US) accounted for 82% of incident HIV cases among men [1]. Black MSM (BMSM) and Latino MSM experience disproportionately high rates of HIV, making up 37% and 29% of incident cases among MSM, respectively [1]. The disparity is even greater for young BMSM, ages 13–24, who made up 51% of the new HIV infections among young MSM in 2017 [1].

HIV prevention efforts have largely focused on reducing risk behaviors in populations affected by the disease. Research has shown, however, that these behavioral factors do not adequately explain racial disparities in HIV infection. A seminal paper from Millett et al. [2] found that BMSM report lower rates of substance abuse and fewer sexual partners than White MSM, and report similar rates of condomless anal intercourse (CAI), commercial sex work, sex with a partner who is known to be HIV positive, and lifetime HIV testing. A meta-analysis of HIV risk factors in MSM [3] found that BMSM were less likely to have a history of substance abuse and more likely to use condoms. Despite these risk-reducing behaviors, BMSM were found to have higher rates of HIV and were less likely to have initiated medical treatment. Importantly, this study found BMSM experienced structural barriers, including poverty, incarceration and unemployment, at twice the rate of White MSM. These findings suggest that behavioral factors are not a primary contributor to the racial disparities in HIV incidence in MSM, and that higher-level distal factors (e.g. network factors or structural factors) likely play a more prominent role [4].

One legacy of the Millett et al. [2] paper is the implication of sexual networks as a possible driver of racial disparities in HIV. This suggestion sparked a wave of inquiry into how the dynamics of social and sexual networks relate to HIV transmission risk [5,6,7,8]. It has been hypothesized that racial disparities in HIV are perpetuated and exacerbated by increased prevalence of HIV within Black communities and high racial homophily (i.e., the degree to which members of a network have the same racial identity) in the sexual networks of BMSM [9]. Social support within networks has also been identified as an influential factor in determining sexual risk behavior, viral load suppression, and HIV testing [10,11,12]. Recently, researchers have proposed network viral load, which refers to the total HIV viral activity within a connected network, as a metric of HIV risk [13].

The persistent racial disparities, particularly between BMSM and non-Hispanic White MSM, in HIV incidence represents an urgent public health issue. BMSM experience simultaneous marginalization related to racial identity and sexual identity and are disproportionately burdened by a highly stigmatized disease. Network factors could play an important role in understanding how racial disparities in HIV have developed and persisted. The present review aims to (1) collect and summarize the existing body of literature on networks and HIV risk and disparities, (2) place the findings on network factors in the context of research on individual and structural factors, (3) explore implications for intervention and policy development, and (4) propose directions for future research.

Methods

Search Strategy

In collaboration with an academic librarian and an established HIV researcher, we collected scientific literature relevant to network factors in HIV among MSM from four databases representing the most comprehensive collections of behavioral HIV research: Pubmed, Scopus, PsycINFO, and Sociological Abstracts. Search logic was constructed by combining terms related to networks, social support, HIV and other STIs, sexual risk behavior, racial disparities, and MSM (Fig. 1). The search was limited to peer-reviewed articles published between January 1, 2008 and July 1, 2018, representing all literature published since the release of the seminal review from Millett et al. in 2007 that identified network factors as possible drivers of racial disparities in HIV.

Fig. 1
figure 1

Search logic used for Pubmed query, as an example

Full size image

Inclusion/Exclusion Criteria

We included articles if they described an empiric study that: (1) quantitatively measured either the structure (i.e., size, density, concurrency) or function (i.e., social support) of a social, sexual, or drug network, (2) measured one or more relevant HIV outcomes (i.e., HIV prevalence, seroconversion, sexual risk behavior), (3) sampled primarily racial/ethnic minority men, with more than 50% of the study population being racial/ethnic minorities, (4) sampled primarily MSM, with more than 50% of the study population being MSM, (5) took place in the US, (6) was published in a peer-reviewed journal, and (7) was published in English or Spanish. The first author screened the titles and abstracts of all articles identified from the search query using the inclusion and exclusion criteria and reviewed the full texts of the remaining articles in detail. Those marked for exclusion were annotated with specific reasons for exclusion. Any article that was not clearly a candidate for inclusion or exclusion was discussed with a second reviewer, and consensus was reached regarding inclusion/exclusion.

Data Extraction

Data were extracted from the reviewed articles and compiled into a table of evidence. For each included article, we extracted the goal of the study, the data source (site of recruitment for primary data, name of original study for secondary data), the setting, the sample size, demographic characteristics of the sample, type of network measured, HIV outcome measures, and key findings.

Results

A total of 1184 articles were found over the four databases queried. After removing duplicates, 656 unique articles remained for review. Of these, 581 were excluded in the title and abstract screening phase. An additional 50 articles were excluded in the full text screening phase: 22 studies were excluded because they did not include a network measure, 18 because they did not measure a relevant outcome, nine because they were not empiric papers, and one because it was not peer reviewed. The remaining 25 articles met criteria and were included in this review (Fig. 2).

Fig. 2
figure 2

PRISMA flow diagram for systematic review. From Ref. Moher et al. [53]

Full size image

A summary of the data extracted for this review can be found in Table 1. Three of the included studies used a prospective cohort design, while the remaining studies used a cross-sectional design (n = 22). All but one of the included studies (n = 24) sampled from large and medium metro areas, as defined by the 2013 NCHS urban–rural classification scheme [14]. One study sampled 13 distinct communities and included several small metro areas and neighborhoods in large central cities [15]. Seven studies focused on young MSM, and the remaining studies included adult or older MSM (n = 18). Sample sizes in the studies ranged from 121 to 10,295. Twelve studies enrolled only BMSM and Black transwomen. Nine studies measured social networks only, six measured sexual networks only, and ten studies measured both sexual and social networks. Of the studies reviewed, 22 measured egocentric networks (i.e. information is collected only from the participant) and three measured sociometric networks (i.e. information is obtained both from participants and from their network members). HIV outcomes measured included sexual risk behaviors, HIV prevalence, and seroconversion during study period.

Table 1 Summary of studies on networks and HIV risk and disparities for MSM in the U.S
Full size table

Structural Network Determinants

Size and Density

The size and density of both social and sexual networks are possible predictors of HIV risk. Three studies found that the size of one’s social network had an effect on sexual risk behavior, although the effect was different for BMSM than for Latino MSM. Larger social networks were associated with increased CAI in BMSM, while smaller social networks were associated with increased CAI in Latino MSM [16,17,18]. This difference by race was seen both in older MSM and young MSM (YMSM) and was consistent across studies in various US cities. In one study it was found that greater density amongst members of a social network was associated with lower rates of serodiscordant condomless intercourse (SDCI) with a non-primary partner, an effect that did not differ by race/ethnicity [19]. Another study found no connection between social network density and sexual risk behavior [20].

The studies that focused on size and density of sexual networks were not conclusive. One large multi-city study of BMSM found that larger sexual networks had a bivariate association with higher rates of CAI [21], however a separate analysis of the same dataset that used SDCI as an outcome rather than any CAI found no association between sexual network size or density and SDCI in multivariate models [22]. Similarly, for YMSM in Chicago, no association was found between size or density of sexual networks and CAI [23]. Structural characteristics of social and sexual networks are inconsistent predictors of sexual risk behavior. The structure of social networks is more salient to behavior than the structure of sexual networks, and network factors manifest differently across racial/ethnic groups.

Functional Network Determinants

Social Support

The role of support networks and social support was investigated in seven of the reviewed studies. Studies varied in how they measured social support, although most either measured general social support, or measured specific subdomains of social support (i.e. emotional, instrumental, informational, appraisal) [24]. Two studies included measures of “HIV-related” social support [15, 20]. Most studies measured perceived social support (i.e. a personal belief that social support is available and adequate) [25], although one study measured enacted social support [26]. The majority of the studies measured network-wide social support, but three studies used an actor-specific measure of social support where support from individual network members was assessed [20, 26, 27]. General perceived social support within a network was associated with lower rates of CAI [17, 28, 29] as was a measure of actor-specific HIV-related social support [20]. Lauby et al. [29] found that the relationship between general perceived social support within a network and HIV infection in their study was partially explained by changes in CAI. The direct impact of perceived social support on HIV incidence was explored in one notable prospective cohort study of BMSM in six major US cities, which found that having more people in one’s emotional support network, medical support network, and social participation support network were all protective against seroconversion during the study period [27].

Two studies found no connection between social support and sexual risk behavior. One small study of homeless youth found no connection between actor-specific tangible enacted social support and sexual risk behavior [26]. A large multicity study of YMSM found that social support was unrelated to CAI [15], although this study employed a measure of HIV-related social support within a network that had not previously been used or validated. Social support is an important determinant of sexual risk behavior and HIV risk, although differences in measurement and study population might modify this relationship.

Compositional Network Determinants

Stability

To our knowledge, only one study has investigated the effects of network stability (i.e., the extent to which the members of an individual’s network stay constant over time) on sexual risk behavior [30]. This study found that stability in sexual networks was associated with decreased odds of engaging in condomless sex, as well as with decreases in other sexual risk behaviors. Stability in social networks was found to be protective against sex drug use, but was not protective against condomless sex. Network stability remains a potentially important factor in racial disparities in HIV and additional research is warranted.

Racial Homophily

Four studies explored aspects of the hypothesis that racial disparities in HIV are amplified by high HIV prevalence among BMSM and high racial homophily in the sexual networks of BMSM. The networks of BMSM in Atlanta had higher HIV prevalence than those of White MSM [31]. BMSM in New York City had higher rates of HIV and higher levels of racial homophily when compared to White MSM [32]. Birkett et al. [23] found that racial homophily in egocentric networks was associated with participant HIV status among BMSM, but not among other racial ethnic groups. One notable exception to this pattern of findings was a multi-city study with young BMSM that observed no connection between egocentric network HIV prevalence and participant HIV status [33].

Network Viral Load

Morgan et al. [34] investigated the effects that differences in HIV transmissibility among network members plays in HIV risk, a concept known as network viral load. This prospective cohort study exploring network factors and HIV found that for young BMSM in Chicago, each additional recently infected HIV positive partner (high transmissibility) in one’s network was associated with 13 times greater odds of seroconversion during the study period. There was no relationship between number of long-term infected partners (low transmissibility) and seroconversion, and number of HIV negative partners and number of partners using PrEP (both have negligible transmissibility) were each significant protective factors against seroconversion. These results are consistent with network viral load being a key HIV risk factor.

Non-Network Correlates

In the reviewed literature, constructs at multiple ecological levels were implicated consistently across many studies. A majority of the studies reviewed (n = 18) examined factors related to HIV risk and disparities at multiple levels in additional to network factors; the findings related to these non-network factors are summarized here.

Nine studies reported results related to psychosocial factors including six that investigated how attitudes and norms influence sexual risk behavior and three that explored the role that communication within relationships plays in HIV risk. Norms and attitudes were found to be consistently connected to sexual risk behavior [15, 16, 19, 20, 26, 35], and partner communication and power dynamics emerged as key relationship factors that could influence HIV risk and disparities [22, 36, 37].

Nine studies included structural factors in their main analysis. Socioeconomic status was the most commonly measured structural factor (n = 5), but a small number of studies explored incarceration (n = 2) or discrimination (n = 2) as a determinant of HIV risk. Higher educational achievement, current employment, and stable housing were all found to be related to lower HIV risk [22, 28, 31]. Incarceration was related to participation in exchange sex and instability in social and sexual networks, but was not tied directly to HIV outcomes [30, 38]. There was not clear consensus regarding the effects of discrimination, either related to racism or to homophobia, on HIV risk and additional research in this area is warranted [21, 39].

Discussion

We sought to conduct a systematic review of the literature on networks and racial disparities in HIV to describe the state of the science and identify potential directions for future research. The findings suggest that size and density of sexual networks may be less influential than the size and density of social networks, that social support plays a protective role regarding HIV risk, and that racially homophilous networks with high HIV prevalence are linked to disparities in HIV. In addition to network factors, psychosocial factors and structural factors were also found to be involved in determining HIV risk and were often studied alongside network factors. This review highlights the importance of studying network characteristics in understanding racial disparities in the HIV epidemic. Where individual behavioral factors fail to explain disparities, network factors and their antecedent structural factors emerge as key drivers of sexual risk behavior and racial disparities in HIV incidence.

Networks have emerged as a central factor in HIV risk and disparities. While a great deal of research has measured structural characteristics of networks such as size and density, the relationship between these structural network factors and HIV risk and disparities is not clear. Little evidence was found to suggest the structure of sexual networks was highly predictive of HIV risk, and the relationship between the structure of social networks and risk was complicated. Of note, we discovered that social network size had different effects on CAI in BMSM and Latino MSM samples. We believe this may be attributed to the different relationships that Black and Latino men might have to the mainstream gay community. Patterns of racism in the MSM community, including racialized sexual stereotypes [40], may lead to Latino MSM finding greater support within this community, while BMSM may experience alienation or discrimination. Thus, as we consider the role of social networks in HIV risk among MSM, it is important to acknowledge varying effects that may be experienced among racial/ethnic minority groups. While basic structural characteristics such as size and density should still be measured, future research that measures more complex measures of network position, such as bridging [41], could offer additional insight into how networks affect HIV risk.

Social support within networks was well studied in the literature surrounding HIV risk and disparities. Social support was seen to be protective against HIV infection among BMSM. Additionally, there was evidence that this relationship was partially mediated by the effect of social support on sexual risk behavior. One of the few prospective cohort studies reviewed found a direct relationship between social support and decreased seroconversion [34]. These findings are compelling because they reveal the influence of social support on HIV risk, as well as suggesting sexual risk behavior as a potential mechanism by which this effect operates. Several evidence-based interventions for racial/ethnic minority MSM have been described that leverage social networks to influence behavior change (e.g. Many Men, Many Voices and d-up: Defend Yourself! [42, 43]), but little is known about the effects of these interventions at the network level. Research and intervention development should focus on evaluating their impact on network-level characteristics (e.g., changes in size, density, heterogeneity) and its mechanisms of change (e.g., increased availability of social support), alongside HIV outcomes, in an effort to inform strategies that address racial disparities in HIV.

The compositional characteristics of networks offered important insights into the connection between networks and HIV disparities. The evidence largely supported the hypothesis [7] that BMSM tend to have racially homophilous networks and networks with greater HIV prevalence than White MSM. It was found that viral load within a network emerged as a highly predictive correlate of seroconversion. This suggests that network viral load may be a more salient measure of HIV risk within a network than network HIV prevalence. Assessing network viral load also has several advantages over community viral load, another common measurement of aggregated viral activity [44]. Community viral load captures transmissibility within a geographic area. Since sexual networks often span wide geographic areas, network viral load offers a more direct measurement of transmissibility within socially defined networks.

While compositional characteristics of networks are an important factor in HIV risk and disparities, addressing network HIV prevalence and racial homophily through policy and intervention could prove problematic. The elevated rates of HIV within the networks of BMSM are both the cause and the consequence of racial disparities, and offers little direction for intervention. In fact, seeking to address racial homophily without mitigating the structural forces that have historically shaped these networks (e.g. racism, segregation) will be insufficient [45, 46]. Additionally, focusing on racial homophily as a risk factor for HIV risks pathologizing love and connection among communities of color [47]. Intervention and policy should instead focus on testing and treatment, as well as improving access and adherence to PrEP and routine HIV care, to reduce viral load within the networks of BMSM. Increasing regular testing and improving partner communication skills could also help to narrow the knowledge gap about partner HIV status.

Confirming a large body of research connecting individual level factors to HIV risk, psychosocial processes were found to have effects on sexual risk behavior. Theory-driven measurements should be incorporated into intervention design to better target the most salient psychological pathways leading to behavior change. Interventions that are theoretically based have been proven to reduce HIV risk behaviors and lower HIV incidence among MSM, including racial/ethnic minority MSM [48, 49].

Structural factors are key determinants of racial disparities in HIV. The mechanisms by which structural factors affect HIV disparities are not well understood and warrant investigation. The role of networks as mediators between structural factors and HIV disparities have largely been unstudied [50]. Research that measures how network stability might mediate a relationship between incarceration and HIV risk behavior and disparities would help illuminate this possible connection. Additionally, investigations into how structural discrimination such as homophobia and racism might manifest in social and sexual networks could be a fruitful direction for further research. The two studies in this review that explored the effects of discrimination used validated measures of perceived discrimination [21, 39], and future research should adopt or adapt these measures to insure comparability across studies.

Change in structural factors remains the foundation of an effective response to the persistent racial disparities in the HIV epidemic. Studies included in this review demonstrated that incarceration affects network stability and contributes to the interplay among sexual risk behaviors, sex work, and drug use. A limited body of research suggests that the criminal justice system could be a point of intervention to reduce these disparities [4]. Reforming policies and policing tactics that target sex workers, injection drugs users, and people of color could help to reduce network instability and HIV transmission [51]. Standardizing post-release support, including connection to testing and treatment, may serve to mitigate the effects of incarceration on sexual networks. Finally, policies that improve access to housing, jobs, and education for racial/ethnic minority MSM will be fundamental to efforts to reduce racial disparities in HIV.

Limitations

This review had several limitations. First, by limiting our review to studies that used quantitative measurement of network factors, we may miss important insights from foundational qualitative work. Second, our review was limited in scope. We only examined research that had an explicit focus on racial/ethnic disparities with individual-level outcomes in order to address our research question efficiently. As a result, our review excludes some network research (e.g., stochastic network modeling studies; see Fujimoto et al. for exemplar study) [52] that informs racial/ethnic HIV disparities. Additionally, the majority of the reviewed studies employed a cross-sectional study design, limiting our ability to draw conclusions about causation. Finally, there are inherent limitations to the egocentric (information derived solely from participant) network designs used by most studies in this review. Sociometric designs (information derived from all members of network), while more costly and burdensome, offer a more robust measurement of social and sexual networks. Future research should aim to measure complete networks over time in order to better understand how the dynamics of social and sexual networks effect HIV risk and disparities.

Conclusions

Racial disparities in HIV remain a pressing public health issue and a top priority to be addressed by future research, intervention development, and policy change. The existing literature supports a multilevel model of HIV risk that implicates factors at the individual, network, and structural levels in creating and propagating racial disparities. Structural, functional, and compositional characteristics of networks have all been linked to HIV risk behaviors or racial disparities, suggesting that networks play a key role in shaping these outcomes. Development of intervention and policy should look beyond individual level behavior change. Research and intervention that focuses on factors at the network and structural levels to effectively address racial disparities in HIV are needed. These higher-level factors, which require cultural and economic shifts, may address the fundamental causes of these disparities and hold the potential to affect lasting change.