Prospects of HIV elimination among men who have sex with men: a systematic review of modeling studies

Search strategy and selection criteria

Embase and PubMed were searched for studies published between July 1, 2016 and August 7, 2023, when the search was conducted. The starting date was chosen to coincide with the publication of the World Health Organization (WHO) consolidated guidelines on the use of antiretroviral drugs for treating and preventing HIV infection [20]. The search string was (“HIV” OR “human immunodeficiency virus”) AND (“homosexual*” OR “transgender*” OR “gay” OR “MSM” OR “men who have sex with men” OR “men having sex with men” OR “bisexual*”) AND (“model*” OR “framework” OR “simulat*”) AND (“treat*” OR “prevent*”) AND (“mathematic*” OR “transm*” OR “comput*”).

Studies were included if they (i) involved a dynamic model for HIV transmission, where the force of infection depends on the state of the population at a given time, and (ii) assessed the impact of interventions on HIV transmission among MSM. Studies for HIV transmission in a broader population involving MSM were included if they reported a direct or indirect impact of interventions on HIV outcomes among MSM specifically. Studies that involved a dynamic co-transmission model of HIV and another sexually transmitted infection (STI) were included if the primary outcome was HIV. As we were interested in the epidemiological impact of interventions assessed using dynamic transmission models, statistical, back-calculation, decision-analytic, and health economic models primarily focused on cost-effectiveness evaluations were excluded. Studies focused on methodology rather than the impact of interventions on HIV transmission were excluded. Conference abstracts, reviews, preprints, articles not published in English, or without full text were also excluded.

Data extraction and analysis

Two authors independently screened the titles and abstracts for inclusion and identified eligible studies using Rayyan software. Three authors independently conducted full-text screening and extracted data using a predefined data extraction form. Study inclusion was by consensus. Rare disagreements were resolved through detailed discussions of the studies in question. The authors regularly compared extracted data to ensure consistency in the review process. Data were extracted on article information, study population, interventions, elimination definition, model type, model structure, and calibration (43 data fields in total; Appendix 2). Data fields were summarized descriptively unless quantitative data were available.

Studies were categorized geographically by country and by UNAIDS region (Asia and the Pacific, Caribbean, Eastern and Southern Africa, Eastern Europe and Central Asia, Latin America, Middle East and North Africa, Western and Central Africa, Western and Central Europe and North America) [2]. Additionally, studies were categorized by characteristics of the modeled population, such as demographics (age and ethnicity), subgroups (MSM only or MSM and other subgroups such as the heterosexual population, female sex workers and their clients, injecting drug users, and transgender women), and geographical scale (national, regional, or urban). National scale models consider populations in the entire country. Regional scale models consider a large administrative division within a country (e.g., a state in the USA or a province in China). Urban scale models consider a city or a group of cities and their immediate metropolitan areas. Populations were regarded as stratified by these characteristics if model analyses used different model parameters to describe distinct subgroups.

Models were categorized into deterministic compartmental (DCM), stochastic compartmental (SCM), and agent-based models (ABM). Compartmental models stratify the population into compartments based on certain characteristics, such as disease stage, and track the population in each compartment over time. In contrast, agent-based models include individual heterogeneities and track the status of each individual over time. Unlike deterministic models, stochastic models account for random events. Agent-based models are inherently stochastic.

We reported primary interventions, defined as interventions for which model parameters describing different aspects of intervention engagement (e.g., uptake, retention, coverage, and adherence) and/or intervention efficacy were varied. For example, in a study investigating the increase in PrEP uptake combined with regular HIV testing, PrEP is considered the primary intervention because its uptake was varied. The definition of HIV elimination was a measurable target used in the model analyses to determine when interventions could stop HIV transmission. Elimination definitions and interventions required to achieve elimination were extracted and narratively reported. Descriptive statistics, i.e., frequencies and percentages, were used to summarize the study locations, characteristics of modeled populations, elimination definitions used, and interventions evaluated. Summary statistics were presented in tables and bar charts. The locations of the studied populations and HIV prevalence were visualized using world maps. Subgroup analyses of the summary statistics were performed to examine differences in elimination prospects, stratified by UNAIDS region, model type, elimination definition, and interventions. All results were pooled quantitatively whenever feasible.

Critical appraisal

Studies that included an elimination definition were critically appraised by evaluating the comprehensiveness of the models in addressing elimination. Given that different studies pursued different goals, ranging from conceptual analytical investigations to operational modeling, the critical appraisal was not used to exclude studies but rather to evaluate the appropriateness of model structures and methodologies for assessing elimination. In the absence of standardized tools for the critical appraisal of modeling studies focused on elimination, we developed our own scoring system. The model comprehensiveness score was calculated based on five criteria, such as whether a model accounted for adherence to interventions, sexual risk compensation, the openness of the modeled MSM population (indicating that new HIV infections could be imported or result from sexual contacts with the external population), whether uncertainty in the model outcomes was investigated and reported, and whether a model was validated. Table S4 outlines the questions we used to assess each criterion and provides clear guidance on scoring. Studies received a score between 0 and 2 for the inclusion of adherence to interventions, reporting of uncertainty in model outcomes, and model validation. A score of 0 or 1 was assigned for the inclusion of sexual risk compensation and the openness of the MSM population. The individual scores were then summed to obtain the final score for each study, ranging from 0 to 8, with a higher score indicating that more criteria were satisfied. Critical appraisal was performed independently by two authors. The scoring of studies was by consensus.

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Study selection

The initial search resulted in 2,732 records, of which 80 studies (2.93%) met the eligibility criteria and were included for data extraction (Figure 1). After the title and abstract screening, a total of 1,721 records (62.99%) were excluded. The majority of records were excluded because they did not involve a dynamic transmission model (n = 1, 427, 52.23%). Following the full-text assessment, a further 78 records (2.86%) were excluded due to being conference abstracts (n = 28, 1.02%), lack of a dynamic transmission model (n = 17, 0.62%), or for other reasons (n = 33, 1.21%). The list of 80 included studies and their characteristics is given in Table 1.

• Download figure
• Open in new tab

Figure 1. Study selection.

MSM = men who have sex with men.

Study locations and HIV prevalence

Figure 2 shows the worldwide location of the modeled populations, juxtaposed with HIV prevalence among MSM by country. Of the 80 included studies, 61 (76.25%) focused on MSM in Western and Central Europe and North America, nine (11.25%) on Asia and the Pacific, six (7.50%) on Western and Central Africa, three (3.75%) on Latin America, one (1.25%) on Eastern and Southern Africa, one (1.25%) on two UNAIDS regions [28], and one (1.25%) was not associated with a specified geographical location [27] (Figure 2 A). No studies focused on MSM in the Caribbean, Eastern Europe and Central Asia, or the Middle East and North Africa. More than half of the studies focused on MSM in the USA (n = 47, 58.75%), with the second most frequently considered location being China and Canada (n = 5, 6.25%, each) (Figure 2 C). We observed a striking discordance between the geography of the modeled populations and HIV prevalence among MSM in those populations (Figure 2 B and C). For example, South Africa, which has the highest HIV prevalence among MSM (29.7%), is represented by a single study (1.25%). Additionally, we found only one study for Senegal (1.25%) and three studies for Cameroon (4.25%), which have the next highest HIV prevalence among MSM (27.6% and 20.6%, respectively). The HIV prevalence in MSM in Western countries, such as Switzerland, France, and the USA, is similar (about 14-15%), however, most studies focus on the USA, with only a few studies considering MSM in other Western countries. Across all UNAIDS regions, 29 countries with HIV prevalence among MSM above 15% [1] were not represented in the recent studies that involve dynamic transmission models.

• Download figure
• Open in new tab

Figure 2. Worldwide location of studies and HIV prevalence among MSM.

(A) Locations of the modeled populations by UNAIDS region [2]. (B) HIV prevalence among MSM by country reported by UNAIDS [1]. In (A) and (B), countries with no data are shown in white. (C) The number of studies that did or did not define elimination and HIV prevalence among MSM by country. The study [28] that considered two countries was counted twice. The study [27] that did not apply to any specific location was not counted. MSM = men who have sex with men.

Population characteristics and model types

The distribution of studies by the characteristics of the modeled population and model types across UNAIDS regions is shown in Table 2. All 27 studies (44.26%) in the Western and Central Europe and North America region that stratified MSM by ethnicity concerned MSM populations at different geographical scales in the USA. Commonly considered subgroups were non-Hispanic Black or African American, Hispanic or Latino, non-Hispanic White or other (e.g., [18, 21, 36]). Most studies that included stratification by ethnicity explored the effectiveness of interventions in achieving the dual goals of reducing the overall HIV burden and narrowing racial disparities (e.g., [36,37,45]). None of the studies in other regions used stratification by ethnicity. Twenty-nine studies (47.54%) in Western and Central Europe and North America, one study (33.33%) in Latin America [29], and five studies (83.33%) in Western and Central Africa [73, 74, 82, 86, 87] stratified MSM by age. None of the studies for Asia and the Pacific or Eastern and Southern Africa used age stratification. The age ranges mostly covered sexually active MSM, spanning from 13–18 years to 60–80 years. Two studies, both in USA populations, included adolescent sexual minority men (13 to 18 years old) to investigate the impact of interventions on HIV burden in this group specifically [38, 45]. The majority of studies (n = 64, 80.00%) considered populations consisting solely of MSM (e.g., [16, 92, 96, 97]). The remaining studies (n = 16, 20%) included other subgroups, such as the heterosexual population [18], transgender women [29], injecting drug users, female sex workers and their clients [81]. Most studies for settings in Western and Central Africa (n = 5, 83.33%) [71, 73, 74, 82, 86], and Eastern and Southern Africa (n = 1, 100%) [92] considered sexual mixing of MSM and the consequent cross-transmission of HIV with other subgroups. Studies for Western countries mostly consider MSM as a separate key population with just a few studies considering MSM mixing with other subgroups (n = 6, 9.68%) [16, 18, 26, 28, 62, 76]. Most studies developed models for an urban environment (n = 48, 60.00%), while the remaining studies developed national (n = 21, 26.25%) and regional (n = 10, 12.50%) models, or did not mention any particular geographical scale (n = 1, 1.25%) [27]. ABMs were the most frequently used model type overall (n = 38, 47.5%) but were rarely used outside Western countries (n = 2, 5.13%). DCMs were the second most used model type (n = 36, 45.00%), being the main model type in Latin America (n = 3, 100.00%), Western and Central Africa (n = 5, 83.33%), and Asia and the Pacific (n = 7, 77.78%). SCMs were rarely used (n = 6, 7.50%).

Interventions

Studies investigated classical behavioral interventions (partner reduction and condom use), biomedical interventions (PrEP, post-exposure prophylaxis (PEP), test-and-treat, voluntary medical male circumcision (VMMC), and STI treatment), and structural interventions. Classical behavioral interventions facilitated changes in the behavior of MSM pertinent to HIV transmission, such as a reduction in the number of sexual partners (e.g., [70, 72, 94]), and increased use or effectiveness of condoms (e.g., [33, 40, 73]). The term test-and-treat was used to describe interventions that accelerated HIV testing and/or ART initiation (e.g., [27,80,85]). Structural interventions involved changes in healthcare systems or support for MSM at risk, such as providing housing for homeless MSM [76]. Figure 3 shows the distribution of interventions by type and UNAIDS region. Globally, the most frequently studied interventions were biomedical, namely PrEP and test-and-treat, followed by classical HIV prevention approaches. In contrast, STI treatment, PEP, VMMC, and structural interventions were rarely included in modeling studies. PrEP was considered more often (n = 51, 51.00%) in Western and Central Europe and North America (e.g., [38, 54, 57]) than in other regions (n = 1, 33.33% in Latin America [22]; n = 4, 21.05% in Asia and the Pacific [28, 35, 81, 94]; n = 2, 15.38% in Western and Central Africa [71, 82]). The next most studied intervention was test-and-treat which was considered frequently across all included regions (n = 1, 100.00% in Eastern and Southern Africa [92]; n = 5, 38.46% in Western and Central Africa [73, 74, 82, 86, 87]; n = 1, 33.33% in Latin America [33]; n = 6, 31.58% in Asia and the Pacific [28, 69, 70, 72, 94, 97]; n = 31, 31.00% in Western and Central Europe and North America, e.g., [40, 62, 76]). Classical behavioral interventions were studied frequently outside Western and Central Europe and North America. Combinations of different interventions were more frequently investigated in Western and Central Africa (n = 5, 83.3%) [74, 82, 86, 87], Asia and the Pacific (n = 7, 77.78%) [28, 35, 69, 70, 72, 94, 97], and Latin America (n = 2, 66.67%) [22, 33] than in other regions (Figure S1).

• Download figure
• Open in new tab

Figure 3. Distribution of modeled interventions by UNAIDS region.

The labels on each bar represent the number of times the intervention was included in models. A single study might have included multiple interventions. The study [27] that did not apply to any specific location was not counted. The study [28] that included both the USA and Thailand was counted for two UNAIDS regions. Pre-exposure prophylaxis = PrEP. STI = sexually transmitted infection. PEP = post-exposure prophylaxis. VMMC = voluntary medical male circumcision. The distribution of interventions by UNAIDS region for studies that defined elimination is shown in Figure S2.

Critical appraisal

Out of the 80 studies investigating the impact of interventions, only 35 studies (43.75%) defined criteria for HIV elimination. The critical appraisal results of these 35 studies are summarized in Tables S5 and S6. The model comprehensiveness scores ranged from 0 to 8, although no study received either of these extreme scores. Studies using ABMs and SCMs all scored at least 4, while three studies (14.29%) using DCMs scored 3 or lower [27, 41, 94]. The average score for ABMs (5.00) was higher than the average scores for DCMs (4.84) and SCMs (4.50). However, three studies (15.79%) using DCMs achieved a score of 7, whereas none of the ABMs did. Two of the five criteria were well satisfied across all studies. All studies presented results with some level of uncertainty, whether due to sensitivity analyses, stochastic effects, or multiple parameter sets. Only one study [27] did not account for adherence to the primary interventions. In contrast, the other three criteria were often either not included or only partially included. Fourteen studies (40.00%) (e.g., [26, 67, 80]) attempted to validate their model outputs, although six of these validations were informal or not clearly described [51, 52, 65, 66, 76, 86]. Twelve studies (34.29%) (e.g., [4, 45, 86]) modeled open MSM populations, with six studies using compartmental models and six using ABMs. Seven studies (20.00%) (e.g., [35, 39, 70]) accounted for sexual risk compensation, more often in compartmental models than in ABMs.

Elimination definitions

Table 3 provides an overview of the 35 studies that defined HIV elimination. There was no consensus on a single definition of elimination. Eight of the 35 (22.86%) studies used multiple definitions [18, 39, 51, 67, 76, 78, 86, 89, 93]. For these studies, each definition was assessed individually for achievability and feasibility.

HIV elimination criteria were defined as a threshold for incidence (n = 14, 31.11%; e.g., < 1 new infection per 1,000 person-years [4, 80, 94]), a percentage reduction in incidence (n = 12, 26.67%; e.g., 90% reduction by 2030 from 2020 [32, 39, 78]), reproduction number less than one (n = 10, 22.22%; e,g., [27, 41, 72]), elimination of racial disparities in incidence or prevalence (n = 5, 11.11% [36, 37, 45, 52, 89]), zero incidence (n = 1, 2.22% [77]), zero steady-state prevalence (n = 1, 2.22% [46]), and other (n = 2, 4.44% [89]).

Elimination definitions varied across model types and UNAIDS regions (Figure 4). A reproduction number less than one was the most frequently used definition in DCMs (n = 9, 39.13%) (e.g., [26, 27, 41] and SCMs (n = 1, 50.00%) [47], but was absent in ABMs. Conversely, incidence reduction was the most frequent definition in ABMs (n = 9, 45.00%) (e.g., [18, 39, 51]) but was seldom used in DCMs (n = 3, 13.04%) [32, 78] and absent in SCMs. Incidence threshold criteria were used across all model types (n = 14, 31.11%) (e.g., [4, 65, 80]), whereas the elimination of racial disparities was exclusively studied in ABMs (n = 5, 25.00% [36, 37, 45, 52, 89]). The preferential use of elimination definitions by UNAIDS region (Figure 4 B) resulted mainly from (i) ABMs not being used for settings outside Western and Central Europe and North America, specifically the USA, and (ii) studies often aligning their elimination definitions with the national goals of the country where elimination was assessed. For example, incidence reduction accounted for almost half of the definitions used in the USA (n = 12, 42.86%) (e.g., [32, 51, 89]), driven by the goals to end the HIV epidemic in the USA, which aim for 75% and 90% reductions in incidence by 2025 and 2030, respectively. In contrast, studies outside of the USA focused on reaching incidence thresholds (e.g., [35, 67, 86]) and reducing the reproduction number below one (e.g., [70, 72, 85]), and did not use incidence reduction as the elimination definition.

• Download figure
• Open in new tab

Figure 4. HIV elimination definitions.

Distribution of definitions by (A) model type and (B) UNAIDS region. The labels on each bar represent the number of times the definition was included in studies. A single study might include multiple definitions. In (B), the study [27] that did not reference any specific location was not counted. SCM = stochastic compartmental model. DCM = deterministic compartmental model. ABM = agent-based model.

Elimination prospects

In the following, we distinguished between elimination being (i) achievable, if it was technically possible within modeled scenarios, and (ii) feasible, if, based on the authors’ judgment and discussion, the modeled scenarios where elimination was achievable were practical in a real-world context (Table 4). Feasibility was considered (un)likely if the authors judged that the intervention scenario required to achieve elimination in the model was (im)plausible. In 32 out of 44 modeled scenarios (72.73%), elimination was achievable. Using UNAIDS regional stratification, 26 scenarios in Western and Central Europe and North America (e.g., [16, 61, 80]), one scenario in Western and Central Africa [74], and four scenarios in Asia and the Pacific [35, 70, 81, 94]) could achieve elimination in the model. However, only six (18.75%) of these 32 modeled scenarios were deemed likely to be feasible in practice, half of which were in Western and Central Europe and North America [4, 16, 47] and half in Asia and the Pacific [35, 81, 94]. Sixteen (50.00%) elimination scenarios were deemed unlikely to be feasible in practice, all for settings in Western and Central Europe and North America (e.g., [46, 76, 85]) or without a specific location [27]. The feasibility of ten (31.25%) scenarios was not discussed by the authors (e.g., [18, 66, 74]).

Achievability of elimination differed with respect to interventions by UNAIDS region and country (Tables S2 and S3). The majority of scenarios in Western and Central Europe and North America that included PrEP and test-and-treat achieved elimination (n = 22, 68.75%, e.g., [4, 41, 55] and n = 21, 75.00%, e.g., [18, 39, 67], respectively), while classical behavioral and structural interventions were included less frequently. However, only three elimination scenarios were discussed to be feasible, all for models that consider populations consisting solely of MSM in Europe [4, 16, 47]. Feasible elimination scenarios included an increase of ART coverage and introduction of oral PrEP in the Netherlands [16] and Denmark [4], and a further reduction in time to diagnosis with an increase in condom use in Sweden [47]. In the USA, elimination using a similar composition of interventions was considered unlikely despite the majority of modeled scenarios predicting elimination (e.g., [26, 32, 41]). In Western and Central Africa, test-and-treat and condom use could achieve elimination in half of the modeled scenarios [74], but their feasibility was not discussed. In Asia and the Pacific, modeled elimination scenarios included test-and-treat (n = 4, 36.36%), PrEP (n = 3, 27.27%), condom use (n = 2, 18.18%), and other behavioral (n = 2, 18.18%) interventions. Except one scenario that involved test-and-treat only [72], all of them were successful, and their feasibility was also reported as high for India [81] and Japan [35, 94]. The feasible elimination scenarios involved the introduction of oral PrEP and an increase in test-and-treat rates, which could be complemented with condom use and other behavioral interventions. Notably, of the six feasible elimination scenarios, five used a combination of interventions [4, 16, 35, 47, 94], while only one used PrEP as an individual intervention [81]. Elimination prospects stratified by the use of combination and individual interventions are shown in Table S1.

Underrepresented populations

HIV among MSM is a global problem that transcends geographical borders [1]. To our knowledge, this study is the first to systematically review mathematical modeling studies that assess HIV elimination prospects in this key population worldwide. Our findings show that across all UNAIDS regions, many countries with high HIV burden among MSM were not represented in the recent studies that involve dynamic transmission models. In particular, we did not identify any studies that assess the epidemiological impacts of interventions on MSM in the Caribbean, Eastern Europe and Central Asia, and the Middle East and North Africa. Several factors could contribute to this gap in knowledge, including the lack of high quality sexual behavior and epidemiological data needed for model parameterization, the shortage of local expertise in HIV modeling [98], criminalization of HIV and sodomy, insufficient interest from public health systems in countries where the HIV epidemic in non-MSM populations is more severe than among MSM, poor surveillance, underfunding, discrimination, and stigma.

Notably, a relatively small number of studies targeted MSM in the UNAIDS regions of Western and Central Africa, and Eastern and Southern Africa, collectively known as sub-Saharan Africa [71, 73, 74, 82, 86, 87, 92], where modeling HIV transmission in the general population has traditionally received a lot of attention [99, 100]. This region is known for generalized heterosexual epidemics and a high HIV burden in the general population, but there is also strong evidence of epidemics among MSM [10, 101, 102]. Finally, within Western and Central Europe and North America UNAIDS region, a relatively small number of studies concerned MSM in Europe [4, 16, 46, 47, 55, 63, 83–85] compared to the USA. The likely explanation for this is that modeling methods outside the scope of our review are used to investigate HIV elimination in Europe. For example, back-calculation models have been developed for the Netherlands [6], the UK [5, 103], and Denmark [104] but are not included in our review focused on dynamic transmission models.

Elimination scenarios

Elimination was achieved in models far more often than authors deemed feasible for real-world implementation (six out of 32 scenarios). Several reasons could contribute to this discrepancy. Firstly, intervention parameters in models (e.g., PrEP, ART, condom use coverage and adherence, testing rates) can be selected from the maximum possible range, potentially resulting in values that are not achievable in practice (e.g., [39, 76, 78]). Secondly, the feasibility of about one-third of modeled elimination scenarios was not discussed by the authors (e.g., [70]), possibly due to the lack of authors with relevant real-world implementation expertise. Thirdly, the six feasible scenarios [4, 16, 35, 47, 81, 94] were obtained in either DCMs or SCMs that were mostly among the least complex models, as described by their comprehensiveness scores. This implies that the results reported by these six models may be overly optimistic.

Our findings show that significant gains in HIV control among MSM have been made in some settings. Elimination is likely in certain Western European countries due to the scale-up of test-and-treat programs, and PrEP emerges as one of the key interventions that can help to reach elimination faster [4]. This aligns with evidence that HIV incidence in countries like the Netherlands and the UK had started to decline with the expansion of test-and-treat programs [5, 6] but declined much further after the introduction of national PrEP programs [105]. Notably, none of the studies in the USA settings considered elimination feasible. This outcome could be partly explained by the use of ABMs and the complexity of the subepidemics in the USA, characterized by strong heterogeneity in transmission among different ethnic and geographical subgroups that require culturally and regionally tailored interventions.

Our findings also highlight inequitable responses in HIV control worldwide. Unlike many studies in Western countries [4, 16, 32, 46, 55, 66, 67, 77, 78] and some studies in Asia [35, 81, 94] that consider PrEP as an essential intervention for faster approach to elimination, studies in Africa still focus on the expansion of treatment and condom use [74, 86]. However, the real-world effectiveness of condoms is undermined by adherence issues. No studies considered elimination scenarios with PrEP introduction among MSM in Africa, potentially due to delays and structural barriers in implementing this intervention [106].

Lastly, our review underscores the importance of combination interventions in increasing the feasibility of elimination. Although the sample of feasible elimination scenarios was small and their conclusions may be overly optimistic, 5 out of 6 considered a combination of interventions [4, 16, 35, 47, 94]. This is consistent with other studies suggesting that a combination of intervention strategies is necessary to control and eliminate HIV [13, 107, 108], and that transmission models should incorporate multiple, simultaneously acting interventions [15].

Relatedness of elimination definitions

Various definitions of elimination used by different studies reflect their perspectives in terms of modeling and public health. While each elimination definition focuses on a specific aspect of the epidemic dynamics of HIV, these aspects are epidemiologically related. In theory, when the effective reproduction number is below one, the incidence will decrease and eventually elimination will be reached. The smaller the effective reproduction number, the faster the decline in incidence, and the sooner a given incidence threshold will be reached. From a modeling perspective, using the effective reproduction number is attractive (e.g., [47, 55, 94]), because it summarizes the qualitative behavior of the epidemic, and elimination is a consequence of reducing the reproduction number below the threshold of one. From an estimate of the reproduction number, one could, in principle, compute the incidence and the time it takes for the incidence to fall below a threshold. The converse does not hold, i.e., knowing that the incidence is below a certain threshold does not necessarily mean that elimination will be reached in the long term. The incidence could still stabilize at a new lower endemic prevalence if the reproduction number is above one. Therefore, from a modeling perspective, calculating the effective reproduction number has clear advantages over simply calculating incidence. However, an explicit calculation of the reproduction number is only possible for DCMs which explains the frequent use of this definition for these models in our review (e.g., [16,41,67]). For SCMs and ABMs, approximations for a reproduction number can be computed numerically by calculating the average number of secondary cases per infected individual. The involved numerical computations probably explain why none of the ABMs in our review used this definition.

From a public health perspective, it is essential to consider how the path to elimination can be achieved and monitored [109]. This implies that we need elimination definitions based on measurable quantities [110]. Although incidence is not directly observable, it can be estimated from the number of diagnosed cases. More concretely, definitions based on an epidemiological goal such as a threshold (e.g., [4, 60, 80]) or a reduction in incidence (e.g., [18, 32, 78]) can be used in practice to validate model predictions. Both can be measured and monitored over time, and serve as a basis for defining standardized indicators for public health policy evaluation [110]. Therefore, for modeling to contribute effectively to public health policy, it is necessary and valuable to report intervention coverage and incidence, preferably in a format that can easily be compared to standard indicators [109,110]. A clear advantage of using the elimination definition based on an incidence threshold is that incidence can be computed in all model types. Besides offering meaningful guidance to policymakers, this definition would also enhance study comparability within and across different MSM settings.

The elimination of disparities, often discussed in the context of HIV, is a different concept and does not necessarily lead to the elimination of HIV from a population. The goal is to eliminate large differences in HIV incidence between population groups, such as ethnic groups (e.g., [36,37,45]), rather than to eliminate HIV entirely. However, in many epidemiological contexts, it is an important milestone to overall HIV elimination once we consider that groups with high incidence are often characterized by a high risk of HIV acquisition, resulting in effective reproduction numbers in these groups remaining above one the longest. In the final stages before elimination, the groups with the highest reproduction numbers will be the remaining drivers of transmission. Therefore, targeting these groups will be crucial to achieving the final goal of HIV elimination.

Future elimination modeling

Based on our findings, we identify several areas where HIV elimination modeling in MSM needs to advance. In countries where HIV incidence has dropped considerably in recent years [4–6] and elimination may be possible, modeling should focus not only on interventions that achieve elimination but also on those that sustain it. This conclusion was also supported by [111]. Elimination of transmission among MSM in specific settings may be interpreted as HIV no longer being a public health problem, which could lead to cutbacks in national prevention programs in these communities. For example, scenarios of future changes in the capacity of the national program for oral PrEP are being investigated in the Netherlands [112]. A decrease in PrEP uptake and condom use, combined with a potential increase in sexual risk behavior among MSM [113–115], may lead to a rebound in HIV incidence. A new priority for countries nearing elimination and aiming to sustain it could be interventions targeting HIV infections acquired abroad, either through immigration or travel. Examples of such interventions include offering voluntary HIV testing to incoming migrants or providing PrEP to MSM who travel abroad [4]. Additionally, transmission modeling will need to be complemented with health economic evaluations to guide future interventions that balance the cost of maintaining them and keeping HIV transmission at bay.

Our review highlighted a lack of modeling studies in some regions that have a high HIV prevalence among MSM. Within locations that warrant such attention, a special place is held by so-called island countries. To wit, the Caribbean region, which has the second highest HIV prevalence among MSM across UNAIDS regions [116] and a high volume of migration both within the region and internationally, will require the development of meta-population models. These models including population mobility were absent in our review. Understanding HIV transmission networks and the impact of migration [117] including from Latin America will be crucial for achieving HIV elimination in this region. Meta-population models could also be relevant for MSM populations in island countries in Asia and the Pacific, which have so far received little attention. Movement and migration [117] will generally become increasingly important as they could aid in importing HIV infections from high-prevalence regions to regions nearing elimination, including in the context of Western countries [4].

In our review, almost all studies considered MSM in Africa as part of a wider population consisting of the heterosexual population and key populations other than MSM [73, 74, 82, 86, 87, 92]. In contrast to concentrated epidemics among MSM in Western and Central Europe and North America (e,g. [49, 63, 91]), HIV epidemics among MSM and the heterosexual population in Africa are substantially mixed due to men having sex with men and women [118,119]. However, while the numbers of new infections in the overall adult population, sex workers, and their clients in Africa have been falling at the same rate, no such progress has been observed for MSM [3, 10]. If MSM are left behind in the HIV response in Africa, they may continue to be a source of infection for women in the general population in the future, potentially undermining the progress made in eliminating HIV within this population. The inclusion of MSM in transmission models for African countries is needed to inform tailored interventions that ensure equal rates of progress towards HIV elimination for different key populations and the general population [120]. Given the proven benefits of PrEP in Western countries and Asia, exploring its impact on HIV elimination among MSM in Africa is desirable. Additionally, while public health literature highlights the benefits of PEP, the population-level impact of this intervention remains largely unexplored in Africa and globally [121], as does the impact of long-acting injectable PrEP compared to oral PrEP. One study in our review [78] demonstrated that long-acting injectable PrEP programs can achieve HIV elimination at much lower coverage compared to oral PrEP.

Another modeling direction involves understanding the elimination of disparities in HIV burden among diverse subgroups of MSM. As overall incidence decreases, disparities may be exacerbated [120]. Therefore, eliminating disparities is a parallel goal to achieving the overall target for HIV elimination. In our review, this topic has received much attention in the context of epidemics in the USA [36, 37, 45, 52], where pronounced disparities in HIV burden have been observed due to assortative mixing of racial and ethnic subgroups and differences in their interaction with the healthcare system (e.g., testing rates, ART, and PrEP coverage). However, the issue of disparities is not limited to the USA and is relevant to characteristics other than ethnicity in many countries. For example, substantial age differences in HIV prevalence are observed among MSM in Africa [122] and the Caribbean [9]. On the road to elimination, disparities may increase between native and migrant populations [117], as well as between rural and urban communities. In our review, simpler models with fewer stratifications such as DCMs and SCMs predominated outside Western and Central Europe and North America. More complex ABMs will be required to simulate elimination interventions that are sufficiently nuanced to address the vulnerabilities of diverse subgroups of MSM. The need for improved incorporation of characteristics such as ethnicity in future models was also discussed in other literature [15, 99].

Finally, modeling could be useful for understanding how new HIV technologies and biomedical prevention tools might affect transmission dynamics and either strengthen or undermine the feasibility of elimination in different regions. Research efforts are increasingly dedicated to improving the quality of life for people already living with HIV. Biomedical advancements may lead to a potential HIV cure in the future, with a target product profile already formulated [123]. Notably, the target product profile considers the possibility of re-infection and viral rebound after ART-free suppression acceptable. These characteristics imply that introducing an HIV cure may result in new infections among MSM. Investigating future scenarios involving such potential interventions will be necessary to guide their implementation in the context of current HIV elimination efforts.

Limitations

In the absence of prior systematic reviews on the possibility of global elimination among MSM, we focused our study on mathematical transmission models. These models have the advantage of providing a mechanistic understanding of the transmission process and are frequently used to inform policymakers. The synthesis of elimination definitions and interventions that may lead to elimination in a broader class of models available in the literature (e.g., statistical, back-calculation, decision-analytic, health economic) should be a subject of future research. Additionally, we compared the worldwide location of the studied populations with the HIV prevalence among MSM as reported by the UNAIDS key population Atlas [1]. These UNAIDS data are compiled from public sources and reviewed for quality, though the quality may vary. Other recent estimates of HIV prevalence among MSM cover only specific UNAIDS regions (e.g., [9, 101, 102]) and their use would not alter our overall conclusions about the underrepresentation of MSM populations. However, it is important to note that our decision to exclude studies not published in English may have contributed to the finding that some UNAIDS regions are underrepresented in modeling studies. Furthermore, the presence of various elimination definitions and model types complicated the comparison of efficient intervention scenarios across settings. Different studies modeled similar interventions using a range of modeling paradigms, making cross-comparison challenging. Although we extracted some details on model structures and methodologies, our review primarily focused on elimination definitions and the potential for elimination through various interventions rather than on the structural composition of the models. Therefore, we did not comment on the suitability of the model structure for projecting elimination prospects beyond the criteria included in the comprehensiveness score and used for critical appraisal of studies. In the future, more research should focus on systematic comparative analyses of different models using consistent elimination definitions, epidemiological outcomes. Although it does not specifically define elimination, the study by Eaton et al. [100] serves as an example of such an analysis.