Effectiveness of interventions aimed at reducing HIV acquisition and transmission among gay and bisexual men who have sex with men (GBMSM) in high income settings: A systematic review

Introduction HIV transmission continues among gay and bisexual men who have sex with men (GBMSM), with those who are younger, or recent migrants, or of minority ethnicity or who are gender diverse remaining at increased risk. We aimed to identify and describe recent studies evaluating the effectiveness of HIV prevention interventions for GBMSM in high income countries. Methods We searched ten electronic databases for randomized controlled trials (RCTs), conducted in high income settings, and published since 2013 to update a previous systematic review (Stromdahl et al, 2015). We predefined four outcome measures of interest: 1) HIV incidence 2) STI incidence 3) condomless anal intercourse (CLAI) (or measure of CLAI) and 4) number of sexual partners. We used the National Institute for Health and Care Excellence (UK) Quality Appraisal of Intervention Studies tool to assess the quality of papers included in the review. As the trials contained a range of effect measures (e.g. odds ratio, risk difference) comparing the arms in the RCTs, we converted them into standardized effect sizes (SES) with 95% confidence intervals (CI). Results We identified 39 original papers reporting 37 studies. Five intervention types were identified: one-to-one counselling (15 papers), group interventions (7 papers), online interventions (9 papers), Contingency Management for substance use (2 papers) and Pre-exposure Prophylaxis (PrEP) (6 papers). The quality of the studies was mixed with over a third of studies rated as high quality and 11% rated as poor quality. There was some evidence that one-to-one counselling, group interventions (4–10 participants per group) and online (individual) interventions could be effective for reducing HIV transmission risk behaviours such as condomless anal intercourse. PrEP was the only intervention that was consistently effective at reducing HIV incidence. Conclusions Our systematic review of the recent evidence that we were able to analyse indicates that PrEP is the most effective intervention for reducing HIV acquisition among GBMSM. Targeted and culturally tailored behavioural interventions for sub-populations of GBMSM vulnerable to HIV infection and other STIs should also be considered, particularly for GBMSM who cannot access or decline to use PrEP.

rated as high quality and 11% rated as poor quality. There was some evidence that one-toone counselling, group interventions (4-10 participants per group) and online (individual) interventions could be effective for reducing HIV transmission risk behaviours such as condomless anal intercourse. PrEP was the only intervention that was consistently effective at reducing HIV incidence.

Introduction
In 2020 there were approximately 1.5 million new HIV infections globally [1], with just under half occurring in Sub Saharan Africa. However, there remains a substantial HIV epidemic in key populations in high income settings including in gay, bisexual and other men who have sex with men (GBMSM). In 2019 GBMSM accounted for 69% of new HIV diagnoses in the United States of America (USA) [2], 39% in the European Union (EU)/European Economic Area (EEA) [3], and 63% in Australia [4], although incidence in high income settings in GBMSM is now declining [3,5,6].
Rapid declines in HIV incidence in cities such as San Francisco [7,8] and London [9] have been partly attributed to Pre Exposure Prophylaxis (PrEP), the use of ART by HIV negative people before exposure to prevent HIV acquisition. Other factors contributing to declining HIV incidence in GBMSM in high income settings include an increase in the frequency of HIV testing leading to earlier HIV diagnosis, and a shorter time to ART initiation and viral suppression [10]. Nonetheless HIV transmission is still occurring among GBMSM, with specific sub populations of GBMSM in high income countries remaining at increased risk, including those of minority ethnicity, or those who have recently migrated, or those who are gender diverse [6,11]. For these and other groups of GBMSM whose HIV prevention needs are not being met by current HIV prevention interventions, a clearer understanding of effective HIV prevention interventions is required.
Several previous publications have attempted to systematically review and identify the effectiveness of HIV prevention interventions among GBMSM or specific sub populations of GBMSM [12][13][14][15][16]. A systematic review conducted in 2012 to 13 by Stromdahl et al. provided a comprehensive breakdown of the efficacy and effectiveness of a broad range of HIV prevention interventions implemented in a European setting, not restricted to RCTs [16]. Twentyfour HIV prevention interventions for GBMSM were included in Stromdahl's review, eight of which were "strongly" (condom use, universal coverage of ART or treatment as prevention, peer-led group interventions, peer-outreach) or "probably" (voluntary counselling and testing, condom-compatible lubricant use, post-exposure prophylaxis, individual counselling for GBMSM living with HIV) recommended for implementation in Europe, however Stromdahl et al's review was conducted before any data on PrEP effectiveness had been published, as the searches were run in 2012-2013 [16].
The aim of this systematic review is to identify and describe randomised controlled trials evaluating the efficacy or effectiveness of HIV prevention interventions for reducing HIV incidence among GBMSM in high income countries published since Stromdahl et al. in 2013 [16]. In contrast to Stromdahl et al's review we restricted the studies included in this review to randomised controlled trials as these are the gold standard to establish the effectiveness of an intervention, as confounding is minimised due to randomisation.

Methods
We searched ten electronic databases (PubMed, Embase, Medline, Cinahl, PsycINFO, The Cochrane Library, WHO publication database, Social Policy and Practice, EPPI Centre, Web of Science) in April 2021. Search terms limited eligible studies to those published from 1 st January 2013 up until the end of February 2021 in order to capture those not included in the Stromdahl et al. review (for which searches had been performed from December 2012 to February 2013 [16]). To find all HIV-related randomized controlled trials (RCTs) conducted among GBMSM, MeSH terms and key word synonyms for "HIV", "Homosexuality", "Bisexuality" were combined with synonyms for "randomized controlled trial", "placebo" and "drug therapy". Details of the search strategy can be found in S1 Appendix.

Selection criteria
Only peer-reviewed studies published in English, conducted in countries listed as high income on the World Bank list of economies (see S2 Appendix), and which collected data from 1996 onwards, were included in this review. Studies were eligible for inclusion if the study population included GBMSM aged � 16 years and the findings from the study sample of GBMSM were disaggregated from any other population samples included in the study. Studies were eligible for inclusion if they were randomised controlled trials, sufficiently powered (� 15 participants per trial arm) to assess the efficacy or effectiveness of interventions and initiatives directly aimed at reducing HIV incidence among GBMSM. Only studies assessing at least one of the four outcomes related to HIV transmission were included (irrespective of whether these were the primary outcomes for the trials themselves). The chosen outcomes were: 1) HIV incidence; 2) STI incidence; 3) condomless anal intercourse (CLAI) (or a measure of CLAI or condom-use) and 4) number of sexual partners (or a measure of number of partners). Data at end of the trial follow-up period was extracted for these four outcomes.

Study selection and quality appraisal
Studies were selected using a two-stage screening approach. Three reviewers independently screened the titles and abstracts of a randomly selected sample of 10% studies (JS, AH, VC). Since a high (>90%) rate of agreement was recorded for the sample, the remaining titles and abstracts were screened by one reviewer (JS).
Eligible references were selected for full paper screening and assessment by one reviewer (JS) and checked for accuracy by a second (AH/VC), using a checklist devised a priori by the author. Studies published after 2013, but based on data collected before 1996, were excluded at the full paper screening stage. We used the National Institute for Health and Care Excellence (UK) Quality Appraisal of Intervention Studies tool to assess the quality of papers included in the review. This tool (derived from Jackson et al., 2006 [17]) necessitates that each study receive a quality rating for both internal and external validity. Internal validity encompasses a range of criteria that establish whether potential sources of bias have been minimised and the extent to which a study establishes a trustworthy cause-and-effect relationship between a treatment and an outcome. External validity assesses the extent to which the study findings are generalizable to the whole 'source population' (that is, the population the participants were selected from). Each study was rated ('++', '+' or '-') to indicate its quality. Each paper included in the review was quality assessed by two reviewers (JS/AH/VC/FL). All disagreements were resolved through consensus.
Data extraction was undertaken by one reviewer (JS) and checked for accuracy by another (AH/VC). Data were extracted using forms detailing: study setting, population and objectives; data collection period; intervention and control; inclusion and exclusion criteria; recruitment dates and location, method of intervention allocation, relevant outcome measure(s) and follow-up; results (effect size; attrition); author defined strengths and limitations.

Data synthesis
After data extraction for each paper, studies were grouped post hoc according to intervention type, using the following categories: one-to-one counselling (either in-person or via telephone or text), group in-person interventions (referred to herein as group interventions), individual online interventions (referred to herein as online interventions), Contingency Management (motivational incentives) for substance use, and HIV Pre-exposure Prophylaxis (PrEP). If a study reported multiple intervention categories (i.e. online and group) it was allocated to the category which it primarily indicated as investigating. As the trials contained a range of effect measures (e.g. odds ratio, risk difference) comparing the outcomes between arms in the RCTs, we converted them into standardized effect sizes (SES) with 95% confidence intervals (CI) [18]. These are unitless measures of effect size and so allow the comparison of effect across studies using different outcomes and effect measures. We calculated effect sizes directly for studies that had a continuous outcome variable as the difference in means divided by the pooled standard deviation [19]; for studies in which the primary outcome was binary, the log odds ratios were calculated, before converting them into approximate effect sizes by dividing them by 1.81 [19]. A positive effect size indicates the outcome was more frequent in the intervention than the control group and a negative effect size indicates that the outcome was less frequent in the intervention group. We assessed the size of effect based on Cohen's suggestion that d = 0.2 (or d = -0.2) be considered a 'small' effect size, 0.5 (or -0.5) represents a 'medium' effect size and 0.8 (or -0.8) a 'large' effect size [18]. Data were analysed using Excel (Microsoft, Redmond, Wash). For a small number of studies [20][21][22][23][24], there was insufficient information to calculate an effect size, in which case we indicated as such in the Tables and text.

Results
The search process returned 10,539 records from all sources, reducing to 6,645 after excluding duplicates. Title and abstract screening excluded a further 6,346; the remaining 299 full text articles were read to assess eligibility. One study was excluded as it was a non-inferiority trial comparing two types of PrEP and the interest was in identifying interventions that were effective [25]. In total, 37 studies presented in 39 papers, fulfilled the inclusion criteria, underwent quality assessment and were included in the final review. See Fig 1 for details of the study selection process, and Tables 1 and 2 for details of the included studies.
The majority of single-country studies were conducted in the USA (n = 30) [20,21,23,24,, two in the UK (n = 2) [22,52], and one each in Canada [53], Hong Kong [54] and Taiwan [55]. Two studies reported the results of multi-country RCTs: one study took place in sites in Belgium, France, Germany, Italy, the Netherlands, Poland, Spain and England [56] and the other was undertaken in France and Canada [57][58][59] (three papers).
Twenty-one of the 39 papers in this review reported a statistically significant difference between intervention and control for one or more of the trial primary outcomes (see . There were three studies in which discrepancies were noted between our calculated SES and the original analysis due to different statistical methods such as adjustment for baseline values [28,33,36].

Impact of the interventions on the four chosen outcomes
Forest plots for each of the four chosen outcomes are shown in Figs 2-5. Results in the Figures have been grouped by intervention type as indicated by colour: red is PrEP, blue is one to one counselling, purple is online interventions, green is group interventions, grey is contingency    (2013) The mean number of UAI episodes declined in both groups at 6 months, declined further in the PCC group at 12 months, while increasing to baseline levels among controls; these differences were not statistically significant. No differences were observed in STI incidence between the two groups.

Hidalgo et al (2015)
Over the entire follow-up period, intervention participants were less likely than controls to engage in any sexual behaviour while under the influence of substances (p<0.05), and also observed in this group was a decreasing trend of unprotected anal sex while under the influence of substances (p = .08). Follow-up differences between groups on social cognitive outcomes favoured the intervention group, though these differences were nonsignificant. The rate of self-reported condomless anal intercourse (CAI) at 3-months was 32% lower in the intervention group compared to the control group, however this effect was not sustained at 12 months.

+ Not Estimable
Chiou et al (2020) Compared to the control group, the experimental group had significantly higher mean score of safe behaviour knowledge, motivation, and skills; percentage of condom use during anal intercourse; frequency of searching for testing resources and getting HIV and syphilis tests. compared to no or delayed PrEP. The other four papers in which PrEP was the intervention reported outcomes of CLAI [48,49], or STI incidence [58,59]. Fig 3 presents the results of the six papers (one paper reported separate results for HSV1 and HSV2 [59], and another paper reported separate results for chlamydia and gonorrhoea [34]) that looked at the intervention effects on STI incidence. Only one intervention that included  The occurrence of a first STI in participants taking PEP was lower than in those not taking PEP (p = 0�008). Similar results were observed for the occurrence of a first episode of chlamydia (p = 0�006) and of syphilis (p = 0�047); for a first episode of gonorrhoea the results did not differ significantly (p = 0�52).

PLOS ONE
PrEP plus a single oral dose of 200 mg doxycycline post exposure prophylaxis (PEP) within 24 hours after sex compared to no prophylactic dose of doxycycline showed a significant reduction in STI incidence in the intervention arm compared to no intervention [58] (Fig 3).  (Fig 4). However two studies that used one-to-one counselling as an intervention [28,33] and one that used PrEP [52] as an intervention actually demonstrated a small but significant increase in measures of CLAI after standardised effect sizes were calculated [28] (Fig 4).   Fig 5 presents the results of the ten papers that examined the different intervention effects on partner numbers. Only one study that used one-to-one counselling as an intervention demonstrated a significant reduction in partner numbers [26] whilst the other interventions did not reduce partner numbers significantly [27, 28, 36-38, 41, 47, 52], and one online intervention actually demonstrated a significant increase in partner numbers [50] (Fig 5).

Intervention assessment
One-to-one counselling. Fifteen studies evaluated the efficacy of one-to-one counselling interventions (Table 1 ( [20,21,24,[26][27][28][29][30][31][32][33][34][35], one in the UK [22] and one had multiple sites across Europe [56]. Only four studies were rated as high quality [22,26,27,30] (Table 2). There was inconsistent evidence about the efficacy of one-to-one counselling as a behavioural intervention. In terms of outcomes, none of them reported on HIV incidence, two on STI incidence [30, 34], ten on measures of condom use or CLAI [26-33, 35, 56] and three on partner numbers [26][27][28]. Of the two studies that reported on STI incidence, one was rated as high quality [30] and the other mixed quality [34], but neither found that the intervention had a statistically significant effect on the outcome. Of the ten papers that investigated the effect of one to one counselling on measures of condom use of CLAI, six demonstrated

PLOS ONE
statistically significant reductions in CLAI [26,29,30,31,32,35], of which two were judged to be high quality by reviewers [26,30]. One study used a single-session risk decision intervention to highlight misbeliefs about sexual behaviour among HIV negative GBMSM and had a large overall standardised effect size of -1.49 (95%CI -1.57, -1.41) (mean number of condomless sex acts(last 3 months) in intervention group 1.9 at 12 months vs 3.1 in control group) [26] (Fig  4). The other study used a single session, tailored, interactive programme to educate and promote condom use in young, black GBMSM, and reported a moderate standardised effect size of 0.42 (95%CI 0.31, 0.53) (likelihood of consistent condom use for receptive anal sex among HIV negative individuals) [30] (Fig 4). Although there were two other RCTs of a one-to-one counselling intervention that were assessed as high quality [22,27], neither demonstrated a statistically significant result and one provided limited results. The first used a targeted HIV/STI risk reduction intervention (targeting condom use) over three sessions and examined consistent condom use among African American GBMSM (regardless of HIV status) [27] and found no significant difference between experimental and control arm at 12 month follow up for a reduction in either CLAI (SES: 0.01 95%CI-0.19, 0.20) (Fig 4) or partner numbers (SES -0.02 95%CI -0.19,0.14) (Fig 5). The second study used two telephone sessions of augmented motivational interviewing (MI) to reduce risky sexual behaviour in GBMSM prescribed PEP. The results reported in the paper demonstrated no significant impacts on sexual risk behaviour or any of the psychological measures, and no discernible reduction in requests for repeat PEP or rates of STIs within a year [22]. Lack of information provided in this latter paper meant that we could not produce a SES (and so this is not presented in the Figure). Of the three studies that reported on number of partners, two were rated as high quality [26,27] whilst the other was rated as mixed quality [28]. One of the high quality studies demonstrated that a single one to one counselling session (45 minutes) that highlighted misbeliefs using graphic novel and sexual network diagrams, demonstrated a small but significant effect on partner numbers at 12 months (from 1.75 (past 3 months) in the control arm to 1.65 in the intervention arm) (SES -0.32 95%CI:-0.4,-0.24), despite not having any overall effect on STI outcomes [26].
Group interventions. Seven group intervention trials were included in this review: six were conducted in the USA (Table 1 (colour green in Figs 2-5)) [36-41] and one in Canada [53]. Five studies were aimed at HIV negative or undiagnosed GBMSM and two were aimed at HIV positive men [41,53]. The interventions delivered theory-driven, interactive, behaviour change group sessions (ranging from one to eight sessions) in-person designed to reduce CLAI or increase condom use. The majority of the studies recruited sub-groups of GBMSM; two recruited black bisexual men [36, 41], two were targeted at Latino GBMSM [39, 40], one intervention was aimed at young HIV negative GBMSM [37], and one at substance using GBMSM [38]. None of them reported on HIV incidence or STI incidence, all seven reported on measures of CLAI or condom use [36-41, 53] and three found a statistically significant and beneficial effect on condom use [39,40] or reduction in CLAI [38]. Four reported on partner numbers [36-38, 41], however only one found the intervention to be borderline effective at reducing partner numbers (SES -0.12, 95%CI: -0.23, 0.00) (from 2.49 partners at baseline in the intervention group to 1.04 at 6 month follow up, and 1.91 partners at baseline in the control group to 1.50 partner at 6 month follow-up) [36]. Two of these RCTs were rated as high quality [40,53], four were of mixed quality [36, 38, 39, 41] and one was rated as poor quality [37] ( Table 1). The first group study that demonstrated a significant increase in condom use used a four-session, 16-hour, Spanish language group intervention that showed four-fold difference in odds of consistent condom use between intervention and control [40] (SES 0.76 95%CI 0.37, 1.15) at 6 months follow up and was rated as high quality (Fig 4). Another RCT, rated mixed quality, was aimed at Latino GBMSM and found that a single-session group intervention increased condom use at last sex compared with non-group activity control at 3 months follow-up (SES 0.29 95%CI 0.01, 0.57) (Fig 4) [39]. The remaining interventions did not demonstrate any significant efficacy for reducing CLAI among GBMSM.
Online interventions. The majority (seven) of the nine studies that used online interventions were conducted in the USA [23, 42-45, 50, 51] and the other two were conducted in Taiwan [55] and Hong Kong [54] (Table 1 (colour purple in Figs 2-5)). The studies delivered cognitive or behavioural interventions online using video-games [42,51], interactive modules, forums or apps [23,44,45,55], short videos [54], Information-Motivation-Behavioural Skills (IMB) [50] or online chat with a live facilitator [43]. The majority (six) of studies aimed to reduce CLAI among HIV negative or undiagnosed GBMSM [23, 42-44, 54, 55] using selfreported outcomes, and three studies aimed to reduce HIV transmission risk behaviours in HIV diagnosed GBMSM [45,50,51]. Of the three studies that recruited HIV diagnosed GBMSM, one used cumulative STI incidence as the primary outcome measure [45] and the other two used reduced sexual risk behaviour (i.e. CLAI) [50,51]. Two studies that recruited HIV diagnosed GBMSM were of high-quality [45,50] and one was rated as mixed quality [51].
The first high quality study compared enhanced online internet-delivered tailored messages about safer-sex, disclosure and ART intervention against monthly sexual behaviour surveys, but did not find a difference in 12-month STI incidence (SES 0.17 95%CI -0.21, 0.55) (Fig 3) [45]. The second compared a four-session online HIV sexual risk reduction intervention (HINTS) using an IMB model with a time-matched 'healthy living' comparison [50], however the results after six months follow-up did not demonstrate a reduction in total number of CLAI acts (SES -0.15 95%CI -0.3, 0.001) (IRR: 0.91 95%CI 0.63, 1.30) and in fact showed a small but significant increase in the number of partners reported (SES 0.25 95%CI 0.09, 0.4) (IRR: 1.81 95% CI: 1.23, 2.68) [50]. The only other high quality RCT that used an online intervention tested the effect of a bespoke seven module intervention taking two hours, based on the information-motivation-behavioural skills model of HIV risk behaviour change [44]. Although the follow-up period for this intervention was very short (12 weeks), those in the intervention arm reported 44% lower prevalence of CLAI compared with controls (SES -0.27 95%CI -0.46 to -0.07) (Fig 4) [44]. The remaining RCTs were rated as mixed [23,42,43,51,54] or poor quality [55] (Table 2) and only one showed any intervention effect of increasing consistent condom use among GBMSM [55] (Fig 4).
Contingency management for substance use. The two Contingency Management (CM) RCTs were conducted among methamphetamine-using, GBMSM in the USA [46,47] (Table 1 (colour grey in Figs 2-5)). One study was aimed at reducing substance use as well as CLAI among HIV negative men and also assessed increasing PEP initiation and course completion [46]. This study was rated as mixed quality and found no significant difference in CLAI at six months (SES 0.15 95%CI -0.03, 0.34) (Fig 4) [46].
The other study recruited substance using homeless GBMSM and assessed the impact of two culturally sensitive intervention programs on reduction of drug use and sexual partner numbers [47]. This study was rated poor quality for several reasons (see Table 2) including inappropriate control conditions and poor statistical analysis [47] and the intervention had no impact on reducing partner numbers (SES -0.14 95%CI -0.48, 0.19) (Fig 5).
HIV pre-exposure prophylaxis. This review included three high quality RCTs (five papers) among GBMSM that included PrEP in the experimental arm [49,52,[57][58][59] and one mixed quality RCT (Table 2) [48]. Two studies were conducted in the USA [48,49], one in the UK [52] and one in France and Canada [57] (Table 1 (colour red in ). Two of the five papers reported on HIV incidence [52,57], three on STI incidence [52,58,59], three on measures of condom use or CLAI [48,49,52] and one on partner numbers [52], one study reported on all four outcomes [52].
Two studies assessed the effect of PrEP on incident HIV infection and were both scored as high quality [52,57]. The UK based open-label RCT, PROUD study (Pre-exposure prophylaxis to prevent the acquisition of HIV-1 infection) randomised participants to either immediate daily PrEP initiation or initiation after a one-year deferral and was the only study that provided data on all four chosen outcomes. An 86% (90% CI 64-96) proportionate reduction in HIV incidence was demonstrated (SES -1.06 95%CI -2.02, -0.41) (Fig 2) [52]. There was no significant difference in STI diagnosis between the two groups after 2 years of follow-up (SES 0.09 95%CI -0.03, 0.21) (Fig 3) or in partner numbers (10 or more Vs less than 10) (SES 0.11 95%CI -0.1, 0.33) at 2 years follow-up (Fig 4), although the intervention was not aimed at achieving these outcomes. However, a significantly larger proportion of men in the immediate PrEP group reported receptive CLAI with ten or more partners at 2 years follow-up (21% vs 12% p = 0.03) (SES 0.35 95%CI 0.05, 0.64) (Fig 5).
The multicentre RCT of high-risk GBMSM in Canada and France known as IPERGAY (Intervention Preventive de l'Exposition aux Risques avec et pour les Gays) reported a relative reduction in HIV incidence in the intervention group who received on demand PrEP of 86% (95%CI 40-98 p = 0.002) (SES -1.07 95%CI -2.33, -0.28) (Fig 2). Unlike previous PrEP trials where participants took daily PrEP, those in the IPERGAY trial took a loading dose of two pills 2-24 hours before sex, followed by daily pills for 48 hours after sex [57]. A separate study of participants from the ANRS IPERGAY study evaluated the impact of on-demand PrEP on herpes simplex virus (HSV)-1/2 incidence and demonstrated that oral PrEP failed to reduce incidence of HSV 1/2 in this population (SES (for HSV1) -0.26 95%CI:-1.05, 0.55. SES (for HSV2) -0.6 95%CI:-0.06, 0.14) (Fig 3) [59]. A sub-study of the ANRS IPERGAY study randomised consented participants to take, in addition to PrEP, either a single oral dose of 200 mg doxycycline PEP within 24 h after sex or no prophylaxis [58]. This was the only study to demonstrate a reduction in incident STI and reported that the occurrence of a first STI in participants taking PEP was lower than in those not taking PEP (HR 0�53; 95% CI 0�33-0�85; p = 0�008) (SES 0.5 95%CI 0.29, 0.88) (Fig 3) [58].
The mixed quality US-based RCT examined the effect of a nurse delivered cognitive behavioural intervention, compared to time and session matched nurse-led counselling control, on CLAI and PrEP adherence [48]. The study demonstrated no significant difference in adherence compared to information and support counselling control and no impact on CLAI (SES -0.03 95%CI: -0.21, 0.14) ( Table 2) (Fig 4).

Discussion
This systematic review of the literature examines recent evidence about the effectiveness of interventions to reduce HIV incidence among GBMSM in high income countries, and adds to the review conducted by Stromdahl et al in the period before 2013 [16]. Thirty-seven randomised controlled trials of five intervention types reported by thirty-nine papers were included in this review. Overall, PrEP was the only intervention that demonstrated a significant reduction in HIV incidence; in fact both high quality RCTs reported a reduction of 86% in HIV incidence [52,57]. The PROUD study was conducted in the UK [52] and the ANRS IPERGAY study in France and Canada [57]. Prior to these two studies, iPREX, a large international study conducted in 2007 to 2009 in Peru, Ecuador, Thailand, Brazil, USA and South Africa (included in Stromdahl's review [16] but not included in this review due to data being collected before 2008 and no disaggregation of data from low and high countries) also demonstrated that PrEP provides significant protection against HIV acquisition [60]. As a result, PrEP has become a central part of HIV prevention interventions around the world and PrEP initiatives are currently offered in 78 countries [61].
The other interventions included in this review yielded mixed results and used other measures reflecting HIV risk such as self-reported CLAI or acquisition of STIs or partner numbers rather than HIV incidence. In terms of reducing risk of HIV transmission through STI management [62], PrEP combined with the antibiotic doxycycline prophylaxis was shown to reduce the occurrence of a first episode of bacterial STI compared to PrEP alone [58], however none of the other interventions had any impact on reducing STI incidence and the majority did not have any impact on partner numbers. In fact, only two interventions, one-to-one counselling [ . A previous systematic review of behavioural interventions for Latino GBMSM also identified some successful interventions for this particular group but highlighted the need to better incorporate and describe cultural features if such interventions are to be successful [63]. There was some evidence that online interventions effectively reduced CLAI in both HIV diagnosed [44] and HIV negative MSM [42,43] however the majority of follow-up times for these studies were relatively short (</ = 6 months).
It is important that the results of this systematic review are interpreted in the context of the restrictions placed upon it by only including randomised controlled trials, and the calendar years of included studies (2013-2021). This review retrieved far fewer intervention types than previous reviews [16,64], possibly because the selection criteria restricted the study type to randomised controlled trials, however randomized controlled trials, when feasible, do provide the best evidence to assess the efficacy of interventions. In their review that did not restrict to RCTs, Stromdahl et al, (2015) strongly recommended four interventions for implementation in Europe: condom use, peer out-reach (providing information and peer support), peer-led group interventions (interactive group activities where a trained peer facilitates promotion of precautionary behaviours for HIV) and using universal coverage of ART and treatment as prevention (TasP). Our search did not retrieve any studies that investigated universal coverage of ART or TasP, probably because overwhelming earlier evidence of the efficacy of these interventions has reduced the need for studies examining the interventions' effectiveness over the past decade. Additionally, results from the PARTNER2 study demonstrated that the risk of sexual transmission of HIV in the context of virally suppressive ART in serodifferent gay partnerships is zero [65,66] and the resulting Undetectable = Untransmittable campaign is championed by all major global health organisations (including WHO) and over one thousand community partners in over one hundred countries [67].
Despite the increasing number of countries adopting PrEP as a prevention intervention, ongoing HIV transmission remains in sub-populations of GBMSM, particularly those who are younger or from minority ethnic backgrounds, or who are recent migrants or are gender diverse [68,69]. It has become increasingly clear that combination prevention that match the needs of a country or community, is necessary to end HIV transmission [70,71]. Whilst the results from this systematic review (focusing on the evidence published between 2013 and 2021) suggest that PrEP as a biomedical intervention provides the strongest evidence for reducing HIV incidence, other interventions, outside the restrictions of this review, such as Treatment as Prevention, and rapid linkage to care following diagnosis and support to attain viral suppression, have had a significant impact on HIV incidence [72]. Our results further demonstrate that targeted interventions such as online and group interventions, which can be tailored for individual communities, could also impact on sexual risk behaviours. However more high quality, culturally tailored and robust trials are needed. It is also increasingly understood that individual health behaviours are shaped by cultural contexts and social interactions, and that drivers of HIV transmission, as with many infections, are based on unmet need and social inequality which must be addressed as a cohesive approach to HIV prevention. Given the overwhelming evidence of the effectiveness of PrEP, more research is needed into the access and uptake of PrEP among populations that are not accessing it.

Limitations of the data
Firstly, only studies that met the inclusion criteria for this systematic review were analysed. In particular we restricted to RCTs as the strongest study design for providing evidence on the effectiveness of an intervention. We excluded observational and non-randomised experimental studies, and therefore may have excluded potential prevention interventions supported by a lower level of evidence. Secondly, most of the studies relied on self-report data about changes in sexual behaviour, which are subject to social desirability and other biases, rather than collecting data on STI or HIV incidence. Studies which included both a self-reported outcome and an outcome of serological testing for STI reported mixed results [30,34,45]. In these cases, the self-report sexual behaviour data suggested the intervention had an effect but there was no difference in incident STIs between intervention and control arms. It is possible that studies that use self-report cannot reliably inform the evidence-base about the efficacy of a given intervention, particularly in the context of unblinded trials, where a social desirability effect on the endpoint in the intervention group may be especially relevant. Additionally, follow-up times for some of the studies was limited (three months) and almost all the studies were conducted in the USA, making generalisability to non-USA populations uncertain. Finally, the intervention groupings are broad and, within the various groups, none of the interventions were exactly the same. Many of the interventions were bespoke and tailored to specific sub-groups of GBMSM and readers should assess the overall effectiveness of an intervention type with that in mind.

Limitations of the review
Our review was limited to the English language. While it is likely that relevant peer-reviewed studies were published in English language journals, we are aware that some potentially relevant papers may have been excluded. By restricting the review to studies with data collected after 2008 and published after 2013, we have limited our ability to make a greater case for the strength of evidence of particular interventions. However, the HIV epidemic among GBMSM in high income countries is continually changing, and factors such as migration, ethnicity, socio-economic status and health policy also have an impact on patterns of HIV transmission. Whilst this review has demonstrated that certain interventions were effective in specific populations, it is important that interventions are culturally appropriate in their implementation if they are to be accessible for all. Finally, as noted above, the review was limited by study type and several observational studies that may have added to the evidence base were excluded. It should also be noted that just because an RCT has failed to find an intervention effective it does not mean it is not effective, just that it has not been demonstrated to be effective in an RCT. However, the benefit of including only RCTs is that this review has limited the impact of bias in the overall conclusions.

Conclusion
Our systematic review of randomised controlled trials from 2013 to 2021 evaluated five intervention types, of which PrEP was the only intervention that was consistently reported to be effective in reducing HIV incidence. Other interventions such as one-to-one counselling, online and group interventions had some impact on reducing high risk sexual behaviour such as CLAI for sub-populations of GBMSM. A systematic review focusing on calendar years before 2013 demonstrated the importance of interventions such as condom use, universal coverage of ART or treatment as prevention and PEP. Our results highlight the role of PrEP in combination HIV prevention but also emphasise the importance of culturally competent, targeted interventions that are designed and tested robustly.