Associations between Intimate Partner Violence and Health among Men Who Have Sex with Men: A Systematic Review and Meta-Analysis

Ana Maria Buller and colleagues review 19 studies and estimate the associations between the experience and perpetration of intimate partner violence and various health conditions and sexual risk behaviors among men who have sex with men. Please see later in the article for the Editors' Summary


Introduction
Violence by an intimate partner affects nearly one in three women globally, ranging from 16.3% in East Asia to 65.6% in central sub-Saharan Africa [1][2][3]. The associated adverse health consequences for female victims include depression, anxiety, posttraumatic stress disorder, eating disorders [4,5], and sexual and reproductive health problems [6,7]. However, intimate partner violence (IPV) is not exclusive to opposite-sex relationships, and there is a growing body of research highlighting the prevalence of IPV in same-sex relationships [8][9][10][11][12][13][14]. Recent reviews suggest that the prevalence in same-sex couples, in particular male-male couples, is as high as or higher than it is for women in opposite-sex relationships [15][16][17][18]. The reported lifetime experience of IPV in gay male relationships lies between 15.4% and 51% [9,11,15,18], depending on the population studied [11,17], the definition of ''partner'' or ''relationship'' [11], and the type of measures used [19]. Most reviews addressing of IPV in the LGBT (lesbian, gay, bisexual, and transgender) population focus on the prevalence of IPV, with limited research on the health associations of IPV [20]. The two existing reviews that explored this association [11,17] used a narrative approach, which summarises and explains results in words rather than pooling quantitative results, and concluded that further research was needed to understand the range of health conditions associated with IPV among MSM.
A UK study on safety planning and advocacy services for men affected by IPV, found that gay and heterosexual men need different services and approaches to accessing support [21]. Insight into IPV health associations for this population will both inform the identification of opportunities for intervention, and guide the design of tailored interventions. To our knowledge this systematic review and meta-analysis is the first to quantitatively synthesise evidence on the association between exposure to IPV as a victim and a perpetrator with health conditions and sexual risk behaviours among men who have sex with men (MSM). We aimed to assess whether MSM who experience or perpetrate IPV have increased odds of common mental health disorder symptoms, eating disorder symptoms, substance misuse, sexually transmitted infections (STIs), HIV positive status, or sexual risk-taking behaviours compared with MSM who are not experiencing or perpetrating IPV.

Methods
The review followed PRISMA [22] and MOOSE [23] guidelines (see Table S1 for the PRISMA checklist). Thirteen bibliographic databases ( Terms for IPV and MSM were adapted from Cochrane protocols and peer-reviewed systematic reviews [24][25][26][27]. Terms for mental disorders were adapted from a review by Trevillion and colleagues [5]. No language restrictions were applied. Reference lists of all included studies were also searched and backward and forward citation tracking used to identify additional potentially relevant studies. Three LGBT specialised journals that featured in our searches (Journal of Homosexuality, Journal of Gay & Lesbian Social Services, and Journal of LGBT Issues in Counseling) were hand searched. Furthermore, experts were asked to identify additional studies. An example search strategy is provided in Text S1, and the protocol of the review can be found in Text S2.

Inclusion and Exclusion Criteria
The study selection process is summarised in the flowchart in Figure 1. Studies were eligible for inclusion if they (a) measured IPV in any of its forms (physical, emotional, and/or sexual); (b) reported results for men who defined themselves as gay or bisexual and/or reported having had sex with men, and who were 18 y or older; (c) were cohort studies, case-control studies, or crosssectional studies; and (d) measured the prevalence and odds or risk of the health outcomes selected for inclusion in this review, or reported data from which these statistics could be calculated. Studies were excluded if they (a) reported on adult sexual assault or non-consensual sex outside of an intimate relationship; (b) reported IPV in a specific group that made it difficult to generalise the results to the wider population, such as IPV in the armed forces, among prison inmates, or in HIV positive individuals. Health conditions included in the review were selected based upon the most frequent outcomes associated with IPV in the broader literature [11,20,[28][29][30]. They were classified in two groups: health conditions (depression, anxiety, and post-traumatic stress symptoms; suicide ideation and suicide attempt; substance use; eating disorders; STIs; and HIV) and sexual risk behaviours (unprotected receptive or insertive anal intercourse and number of sexual partners).

Screening and Data Extraction
AM. B. and L. J. B. screened the abstracts and full texts of potentially eligible studies. Details of the 1,848 excluded papers and reasons for exclusion were documented. Data from included papers were extracted by AM. B. and L. J. B. into a customised Excel spreadsheet. Extracted data included the following: study design; sample characteristics; definitions and measures of IPV; the health conditions and sexual risk behaviours measured; and their effect estimates and measures of uncertainty. Details about confounders controlled for were also recorded.

Quality Appraisal
The quality of each association estimate was appraised by AM. B. and L. J. B. using criteria adapted from validated tools [31] and STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) [32]. An appraisal checklist for included studies was developed. The checklist provided explanatory notes for extracting each data element to ensure consistency between reviewers (see Text S3). The inter-rater agreement rate between reviewers for the quality appraisal was high (K = 0.86), and discrepancies were resolved through discussion with a third reviewer (K. M. D.).

Data Analyses
All analyses were conducted by K. M. D. and AM. B using Stata 12.0 [33]. Descriptive statistics on study characteristics and quality are presented. We computed an overall quality score (Table 1) for descriptive purposes, but in accordance with recommended practice, this was not used to weight studies in the meta-analysis [34]. Random effects meta-analysis was used to calculate the pooled prevalence of different forms of violence and to calculate pooled odds ratios (ORs) representing associations between exposure to IPV and various health outcomes and sexual risk behaviours. Where studies did not report prevalence estimates or ORs, these were calculated from raw data where possible. One study [35] reported a risk ratio, which was converted to an OR using the formula of Zhang and Kai [36]. Higgins's I 2 statistic, which describes the percentage of variability in point estimates that is due to heterogeneity rather than sampling error [37], was calculated. We also calculated confidence intervals for each I 2 statistic using the heterogi command in Stata [38]. Following other reviews 25%, 50%, and 75% were taken to indicate low, moderate, and high levels of heterogeneity, respectively [39,40]. Meta-regression analyses were run in order to establish predictors of heterogeneity.
Some studies reported results in multiple publications or had multiple estimates using overlapping measures of either violence or health outcomes. To avoid double-counting participants, which can lead to falsely precise pooled estimates, we selected only the least biased estimate per study. Decisions were based on the quality criteria described above and the following algorithm: preference was given to multivariate estimates over bivariate, to where the reference group was unexposed to any violence, and to where the estimate was most precise (i.e., smallest confidence interval). To investigate small study effects and possible publication bias, we constructed contour funnel plots in Stata [41] ( Figure S1). For this we used all available estimates for a given outcome (rather than choosing one per study).

Study Characteristics
Nineteen studies describing 18 datasets (Greenwood et al. [47] and Stall et al. [51] are based on the same dataset) with 13,797 participants, and reporting 87 estimates, met the inclusion criteria; 17 studies were included in the meta-analysis. Sixteen studies were conducted in the US, one in Canada [42], one in South Africa [43], and one in China [44] (study characteristics are shown in Tables 2 and S2). The mean age of participants ranged from 19 [45] to 47 [46] y old. Average participation rate was 63%, with 11 studies [35,42,[46][47][48][49][50][51][52]57,58] reporting it. Table 1 summarises estimates' quality. Most studies used convenience, non-probabilistic samples, with the exception of one study that used a population-based sample [48] and three studies that used random samples of particular cities in the US and Canada [42,47,51]. Response bias was addressed in two studies [42,50]. Similarly, two studies [54,55] reported their missing data policy or performed a sensitivity analysis. All the estimates reported or calculated from the raw data had uncertainty measures, and four studies [47,51,52,56] did not report their non-significant results. Four studies [48][49][50]52] did not control for confounders within the analyses, or we calculated the estimates from raw data and did not control for confounders. Confounders included a wide range of variables, with most studies controlling for age, ethnicity, level of education, substance use, and HIV status.
The recall times for which IPV was measured included ''lifetime prevalence'', ''past 5 years'', ''in the last 12 months'', ''in the past three months'', and ''in your current relationship''. Five studies [42,47,49,51,58] reported using an adapted version of the Conflict Tactics Scale (CTS), and one [53] used an adapted version of a scale by Smith and colleagues [59]. Four studies used unique IPV measures developed by the authors that included multiple behaviours to measure all the types of IPV included in the study [35,43,45,60], five studies used multiple behaviour measures for physical IPV but not for emotional or sexual violence [44,50,52,54,57], and four studies used single behaviour items to measure all types of IPV included in the study [46,48,55,56]. Physical violence was measured in all 19 studies (13,812 participants). Emotional violence included verbal and psychological violence, and it was measured in 14 studies [35,42,44,46,47,49-
Estimates of the association between IPV victimisation and substance use were provided in nine studies [46,[48][49][50][51]53,54,57,60] that included 9,607 MSM who had been exposed to any kind of IPV and had consumed different kinds of substances ( Figure 2). We found that exposure to IPV was associated with increased odds of substance use, with an OR for pooled estimates of 1.88, (95% CI OR 1.59-2.22), but with moderate heterogeneity between studies (I 2 = 46.9%, 95% CI I 2 0%-78%) ( Table 4). In order to investigate sources of heterogeneity for this meta-analysis we ran meta-regressions to test the impact of IPV modality and type of drug consumed on heterogeneity. Our results showed that none of these variables explained the heterogeneity (Table 5).
Estimates for the odds of HIV were provided in ten studies [43,46,47,[50][51][52]54,55,57,58] that included 8,835 MSM who had been exposed to any kind of IPV ( Figure 2). Exposure to IPV was associated with a positive HIV status, with a pooled OR of 1.46 (95% CI OR 1.26-1.69) and high consistency between pooled estimates (I 2 = 0.0%, 95% CI I 2 0%-62%) ( Table 4). MSM exposed to any kind of violence also had increased odds of reporting depression symptoms. Depression symptoms were measured in three studies including 3,999 MSM using the Center for Epidemiologic Studies Depression Scale (CESD) [46,51,54], with a pooled OR of 1.52 (95% CI OR 1.24-1.86, I 2 = 9.9%, 95% CI I 2 0%-91%) ( Figure 2; Table 4). Estimates for unprotected anal sex (UAS) were calculated in nine studies [35,[44][45][46]50,52,54,56,58]. After running the meta-analysis and removing an outlier study [56], we kept eight estimates with a total population of 4,447 MSM ( Figure 2). We found that exposure to IPV was associated with   increased odds of engaging in UAS (pooled OR = 1.72, 95% CI OR 1.44-2.05, I 2 = 0.0%, 95% CI I 2 0%-68%) ( Table 4). Contourenhanced funnel plots showed no evidence of publication bias for substance use, HIV, and UAS results ( Figure S1). In these plots studies are missing from the left side of the graph, but across areas of both statistical significance and non-significance. This suggests that publication bias is not responsible for funnel plot asymmetry [41]. The small number of studies that could be pooled for the depression symptoms meta-analysis precluded the use of such plots [46,61,62] for this health correlate.

Intimate Partner Violence Perpetration: Prevalence and Health Associations
Prevalence rates of IPV perpetration among MSM were reported in seven studies [42,43,49,52,53,58,60] with a total population of 4,336 MSM. These studies reported prevalence for different types of violence and different recall periods, precluding the possibility of meta-analysis. Estimates for the odds of substance use among perpetrators were provided in two studies [49,60] that included 1,910 MSM who disclosed perpetration of IPV. Perpetration of IPV was associated with increased odds of substance use (pooled OR = 1.99, 95% CI OR 1.33-2.99) ( Table 4). The results showed moderate heterogeneity (I 2 = 73.1%). Due to the small number of studies, CI I 2 could not be calculated for this result. Likewise, it was not possible to assess publication bias. No other significant associations between IPV perpetration and health outcomes or sexual risk behaviours were found when conducting the metaanalyses (Table 4).

Main Findings
Our findings suggest that IPV victimisation is prevalent among MSM. We also found evidence that exposure to IPV as a victim is associated with increased odds of substance use, depressive symptoms, being HIV positive, and UAS among MSM. Relatively few studies measured perpetration of IPV, but these studies suggested an association between IPV perpetration and substance use. However, these results should be interpreted with caution given the methodological limitations in the included studies.
The associations of IPV with both UAS and being HIV positive suggested in this review are consistent with the existence of a causal pathway. Emotional, physical, or sexual violence (or a combination of all three) could increase risk for UAS [28,63], which in turn could increase risk for HIV infection. Further research is needed to investigate this link, including the specific associations between different types of IPV and HIV status. In particular, the intersection of IPV and HIV seropositivity might be problematic. MSM may fear partner violence following disclosure of HIV status and may experience difficulties adhering to treatment regimes and accessing health services. Sexual health and HIV services are opportune points of intervention for MSM affected by IPV [64], and these findings underscore the need for these services to provide comprehensive sexual health care and awareness of repeat STI or positive HIV status as indicators for IPV and vice versa among the MSM population. Our findings on the association of substance use and depression symptoms with IPV confirm previous findings [48,65] and also highlight the need for mental health and substance misuse services to address IPV issues. Men who are depressed or engage in heavy substance use and are also part of a violent relationship might find it hard to seek help, to end the relationship, or adhere to prescribed treatment.

Strengths and Limitations of This Review
To our knowledge this is the first systematic review to quantitatively summarise the associations between IPV victimisation and perpetration with various health conditions in MSM. Our review entailed extensive searches of the international research literature without any time or language restrictions, and most of our results were consistent and showed no publication bias. Despite this, the review has some limitations. First, prevalence results are only for studies included in the review and should be interpreted with caution. Second, we did not contact authors for additional data. Third, for studies that did not provide ORs, it was not possible to control for confounders when ORs were calculated from raw data [48][49][50]52].

Limitations of Included Studies
The included studies had a number of methodological weaknesses, and the search produced few published studies addressing the association between IPV and health among MSM. Similar to other meta-analyses with a low number of studies, confidence intervals for the heterogeneity estimates were very wide [66]. This means that if further studies were to be conducted on the associations between IPV and health outcomes among MSM the results could be more homogenous or more heterogeneous. Heterogeneity in our review could have been due to different factors such as different recall periods, types of IPV included, the use of convenience samples, different definitions of sexual identity, and the quality of tools used to assess IPV and health outcomes. Unfortunately, tests for sources of heterogeneity have low power with small numbers of studies.
Furthermore, all studies included were cross-sectional, and there was a lack of uniformity regarding the recall periods for IPV and key outcomes, which made it impossible to assess the temporality of the associations. Most studies used convenience, non-probabilistic samples, which makes it difficult to generalise the results to the wider population. Few studies used validated tools to measure IPV, health outcomes, or sexual risk behaviours. HIV measurement relied mainly on self-report, which can provide a biased measure of HIV prevalence. There was a lack of consistency in the types of IPV measured in the studies; however, most authors used either a modified version of CTS or specific behavioural acts to measure IPV. Studies used scales that have been validated for heterosexual samples, which did not necessarily capture MSM's experience of IPV, and most studies used an author's IPV measure without discussing issues of validity or psychometric characteristics. Measures of sexual violence were limited and usually included a single behaviour item asking about forced or unwanted sex, thus potentially resulting in an underestimation of this type of abuse. Additionally, most of the IPV was measured without considering contextual factors such as the frequency with which the abuse occurred or its perceived impact [67]. A step towards a validated tool for IPV among MSM in the US has been taken by Stephenson and Finneran [19], which may be useful in future studies.
Another limitation arises from the fact that we included studies that used different kinds of sexual identities (i.e., self-recognised as gay or bisexual, reporting same-sex sexual encounters, or both). Men's experiences of IPV and the impact it has on their health may be affected by their identification as gay or not, which in turn might be related to their levels of internalised homophobia. Further studies including a differentiated analysis by sexual identity and sexual behaviour might help to understand these relationships better. One such study found that men who identified themselves as gay had a higher prevalence of IPV compared to men who disclosed same-sex sexual behaviour, but did not identify themselves as gay or bisexual [18]. The term ''partner'' was not clearly operationalised in many of the studies. Studies measuring IPV often define ''partner'' as someone with whom the respondent has had sex. Therefore, some studies may have captured violence and abuse experienced in the context of casual encounters. This may have diluted the association between health outcomes and IPV because it is less likely that these relationships are characterised by a pattern of abuse and coercive, controlling behaviour that escalates over time [68,69]. Moreover, the role of sexual agreements, which can be monogamous or non-monogamous [70] and are common among gay couples [70,71], was missing from the studies identified. Lastly, there was a dearth of research relating to MSM in low-and middle-income countries [72], where MSM are at elevated risk of HIV infection and in some countries might face serious discrimination and multiple barriers to accessing health care [73].

Implications for Research
Future research should include national studies using probability-based samples to establish more reliable estimates of the prevalence of different types of IPV experience and perpetration amongst MSM. Further studies should try to establish the impact of particular (more common) forms of IPV among MSM in order to clarify the connections between specific forms of violence and health conditions. Additionally, the sexual orientation and identity of participants should be clearly established, and analysis should differentiate between the two. This will help determine whether the factors that drive IPV and the consequences of IPV are different across men who identify as gay or bisexual compared with men who have same-sex sexual behaviours but do not identify themselves as gay or bisexual.
Longitudinal studies are needed to establish the causal pathway between IPV and adverse health outcomes and sexual risk behaviours. The concepts of bi-directional violence or reciprocity of violence, common in the literature on IPV in MSM, should be taken into consideration in future research. Future studies should also use measurement tools validated for MSM populations and  prospective biological measures of STIs and HIV. Other health outcomes and health risk behaviours for which we did not find any studies, such as eating disorders, repeated STIs, sexual compulsivity [74], and concurrent sexual partnerships [70,71,75], may also be important in understanding the impact that IPV has for this particular population. Furthermore, in view of the increasing evidence that the chronic stress of both childhood and adulthood abuse is associated with inflammatory markers that increase the risk of long term diseases including cardiovascular disease and diabetes [76,77], longer term adverse health outcomes are also possible and should be measured in future studies.

Implications for Practice
The findings of this review underscore the need for health professionals to be aware that IPV is a problem for the MSM population and to carefully assess for IPV and refer affected individuals to appropriate support services in the community. Therefore, same-sex IPV awareness training for health professionals caring for this population is needed. This training should include developing skills for asking about IPV [78,79], dealing with partner notification and safety issues, and making referrals to specialist IPV support services [80]. However, the availability of local and national support services for MSM experiencing or perpetrating IPV varies considerably internationally, and there are very limited data available on their effectiveness [81]. From a health services perspective, adequate protocols for dealing with IPV should be in place, ensuring that all primary and secondary health care services provide holistic care. Addressing IPV is a global health priority, and rigorous evaluations of the feasibility and effectiveness of health service interventions for IPV amongst MSM are needed to inform future policy and practice.   Editors' Summary Background. Intimate partner violence (IPV, also called domestic violence) is a common and widespread problem. Globally, nearly a third of women are affected by IPV at some time in their life, but the prevalence of IPV (the proportion of the population affected by IPV) varies widely between countries. In central sub-Saharan Africa, for example, nearly two-thirds of women experience IPV during their lifetime, whereas in East Asia only one-sixth of women are affected. IPV is defined as physical, sexual, or emotional harm that is perpetrated on an individual by a current or former partner or spouse. Physical violence includes hitting, kicking, and other types of physical force; sexual violence means forcing a partner to take part in a sex act when the partner does not consent; and emotional abuse includes threatening a partner by, for example, stalking them or preventing them from seeing their family. The adverse effects of IPV for women include physical injury, depression and suicidal behaviour, and sexual and reproductive health problems such as HIV infection and unwanted pregnancies.

Supporting Information
Why Was This Study Done? IPV affects men as well as women. Men can be subjected to IPV either by a female partner or by a male partner in the case of men who have sex with men (MSM, a term that encompasses homosexual, bisexual, and transgender men, and heterosexual men who sometimes have sex with men). Recent reviews suggest that the prevalence of IPV in same-sex couples is as high as the prevalence of IPV for women in opposite-sex relationships: reported lifetime prevalences of IPV in homosexual male relationships range between 15.4% and 51%. Little is known, however, about the adverse health effects of IPV on MSM. It is important to understand how IPV affects the health of MSM so that appropriate services and interventions can be provided to support MSM who experience IPV. In this systematic review (a study that identifies all the research on a given topic using predefined criteria) and meta-analysis (a study that combines the results of several studies using statistical methods), the researchers investigate the associations between the experience and perpetration of IPV and various health conditions and sexual risk behaviours among MSM.
What Did the Researchers Do and Find? The researchers identified 19 studies that investigated associations between IPV and various health conditions or sexual risk behaviours (for example, unprotected anal sex, a risk factor for HIV infection) among MSM. The associations were expressed as odds ratios (ORs); an OR represents the odds (chances) that an outcome will occur given a particular exposure, compared to the odds of the outcome occurring in the absence of that exposure. The researchers estimated pooled ORs from the data in the individual studies using meta-analysis. The pooled lifetime prevalence of experiencing any IPV (which was measured in six studies) was 48%. Exposure to IPV as a victim was associated with an increased risk of substance (alcohol or drug) use (OR = 1.88, data from nine studies), reporting depressive symptoms (OR = 1.52, data from three studies), being HIV positive (OR = 1.46, data from ten studies), and engagement in unprotected sex (OR = 1.72, data from eight studies). Perpetration of IPV was associated with an increased risk of substance abuse (OR = 1.99, data from six studies).
What Do These Findings Mean? These findings suggest that MSM frequently experience IPV and that exposure to IPV is associated with several adverse health conditions and sexual risk behaviours. There were insufficient data to estimate the lifetime prevalence of IPV perpetration among MSM, but these findings also reveal an association between IPV perpetration and substance use. The accuracy of these findings is limited by heterogeneity (variability) between the studies included in the meta-analyses, by the design of these studies, and by the small number of studies. Despite these and other limitations, these findings highlight the need to undertake research to identify interventions to prevent IPV among MSM and to learn more about the health effects of IPV among MSM. They highlight the importance of health care professionals being aware that IPV is a problem for MSM and of training these professionals to assess MSM for IPV. Finally, these results highlight the need to improve the availability and effectiveness of support services to which health care professionals can refer MSM experiencing or perpetrating IPV.