Estimating the contribution of key populations towards the spread of HIV in Dakar, Senegal

Abstract Introduction Key populations including female sex workers (FSW) and men who have sex with men (MSM) bear a disproportionate burden of HIV. However, the role of focusing prevention efforts on these groups for reducing a country's HIV epidemic is debated. We estimate the extent to which HIV transmission among FSW and MSM contributes to overall HIV transmission in Dakar, Senegal, using a dynamic assessment of the population attributable fraction (PAF). Methods A dynamic transmission model of HIV among FSW, their clients, MSM and the lower‐risk adult population was parameterized and calibrated within a Bayesian framework using setting‐specific demographic, behavioural, HIV epidemiological and antiretroviral treatment (ART) coverage data for 1985 to 2015. We used the model to estimate the 10‐year PAF of commercial sex between FSW and their clients, and sex between men, to overall HIV transmission (defined as the percentage of new infections prevented when these modes of transmission are removed). In addition, we estimated the prevention benefits associated with historical increases in condom use and ART uptake, and impact of further increases in prevention and treatment. Results The model projections suggest that unprotected sex between men contributed to 42% (2.5 to 97.5th percentile range 24 to 59%) of transmissions between 1995 and 2005, increasing to 64% (37 to 79%) from 2015 to 2025. The 10‐year PAF of commercial sex is smaller, diminishing from 21% (7 to 39%) in 1995 to 14% (5 to 35%) in 2015. Without ART, 49% (32 to 71%) more HIV infections would have occurred since 2000, when ART was initiated, whereas without condom use since 1985, 67% (27 to 179%) more HIV infections would have occurred, and the overall HIV prevalence would have been 60% (29 to 211%) greater than what it is now. Further large decreases in HIV incidence (68%) can be achieved by scaling up ART in MSM to 74% coverage and reducing their susceptibility to HIV by two‐thirds through any prevention modality. Conclusions Unprotected sex between men may be an important contributor to HIV transmission in Dakar, due to suboptimal coverage of evidence‐informed interventions. Although existing interventions have effectively reduced HIV transmission among adults, it is crucial that further strategies address the unmet need among MSM.

KPs, such that a male who becomes infected by another male could then infect a female, resulting in the initial male not only contributing to the MSM HIV epidemic but also the heterosexual epidemic. Indeed, for commercial sex, these analyses conflict with emerging dynamic modelling analyses suggesting that >90% of all HIV transmissions to date are directly or indirectly due to sex work in the low-to-moderate prevalence HIV epidemics of Burkina Faso, Cote d'Ivoire and Benin, and over 65% in the higher-prevalence HIV epidemic in Kenya [9][10][11][12][13][14]. These analyses challenge the conventional wisdom and existing epidemiological tools such as the MOT model. In contrast, two recent studies for Nigeria and Cote d'Ivoire suggest MSM contribute little (<10%) to these HIV epidemics [14,15]. This is despite MSM experiencing a high HIV burden [4,16,17] and engaging in sexual partnerships with women, so contributing bridging infections to the wider population [18]. Limitations in existing evidence emphasizes the urgent need to improve our estimates of the contribution of KP to HIV epidemics in SSA; this basic epidemiological measure is crucial for prioritizing HIV programming.
Senegal has a low HIV prevalence among adults (0.4% in 2016) [19], thought to be due to a comprehensive response to the epidemic [20]. Nevertheless, the HIV burden among KP is much higher (5.9% in FSW and 29.7% in MSM in 2016) [21][22][23][24][25][26][27]. Despite the low prevalence of HIV in the adult population and high prevalence among KP, current estimates suggest that commercial sex and sex between men contribute little (<15%) to existing HIV transmission in Senegal [3,8], although these estimations were limited as they did not incorporate the dynamic aspect of transmission.
To remedy these limitations, we undertook a dynamic model assessment of the contribution of commercial sex and sex between men to HIV transmission in Dakar, Senegal. We also estimated the impact of historical increases in the coverage of antiretroviral therapy (ART) and condom use among KP, and the potential impact of further uptake of prevention and treatment interventions.

| Model description
We developed a dynamic HIV transmission model to evaluate the extent to which FSW, clients of FSW (referred to as clients hereafter) and MSM contribute to the overall HIV epidemic in Dakar, Senegal. The model did not include people who inject drugs because of their low prevalence in Dakar (<0.07% of the adult population [28]). The model considers adults (15 to 49 years), and divides the population into six sub-populations: low-risk females and males, clients, FSW, young MSM (<30 years) and old MSM (≥30 years) (Figure 1). Transgender women (TGW) were not explicitly included in the model because of insufficient data to parameterize this risk population. Low-risk individuals are defined as people who are not MSM and do not report commercial sex.
Individuals enter the modelled population in the low-risk groups when they become sexually active, at a rate that balances non-HIV deaths and reflects population growth. Lowrisk males and females can become clients and FSW at specified rates, with both practising commercial sex for an average duration. Similarly, MSM transition from the low-risk male population, and age from the young to the old MSM group, but remain as MSM until death.
The model captures HIV transmission among the sub-populations through vaginal and anal sex (VI and AI, respectively) between all males and females, and AI within the MSM group ( Figure 1). The model stratifies the population with respect to HIV infection and disease progression (Figure 1b). Upon infection, susceptible individuals acquire acute HIV infection before progressing to chronic infection. Chronically-infected individuals experience HIV-related mortality, but can also be recruited onto ART, which reduces HIV-related mortality. Individuals on ART can be lost to follow up, whereupon they  return to the chronic infection stage. All sub-populations also experience non-HIV-related death. The model incorporates HIV transmission due to main, casual and commercial sexual partnerships. Commercial partnerships occur between FSW and their clients, while main and casual partnerships between men only occur among MSM. All other heterosexual main and casual partnerships occur between all groups. The risk of HIV transmission for each individual is related to the HIV prevalence of their sexual partners, with the HIV transmission risk being elevated if they have acute infection, and reduced if they are on ART. Transmission risk is also related to the frequency of sex between different partnerships and is reduced through condom use. The consistency of condom use is time dependent and varies by type of partnership. The model assumes some males are circumcised, which reduces the risk of HIV acquisition. We assume heightened transmission risk in the initial stages of an HIV epidemic to capture the effects of risk heterogeneity. The model is described in Data S1.

| Model parameterization and calibration
Recent model parameter and calibration data for FSW, clients and MSM were obtained from three integrated behavioural and biological assessment (IBBA) surveys undertaken in Dakar, Senegal, from 2015 to 2016 [25] (client data are unpublished). Sexual behaviour data related to FSW, clients and MSM came from these surveys. The survey used to parameterize the MSM component of the model also included some TGW.
In addition, older IBBA surveys were used to determine whether risk behaviour has changed over time, and how the HIV epidemic in different risk groups has evolved. Importantly, this included trends in condom use for different risk groups. However, differences in the behavioural measures used made it difficult to evaluate how behaviours changed over time, and so uncertainty was incorporated into those trends. Adult population HIV-related epidemiological and sexual behaviour data were obtained from the Demographic Health Surveys (DHS) for 2005 and 2010 [29,30]. Table S1 gives a summary of IBBA surveys used in the modelling, while the model parameterization is summarized in Table 1 and included in full in Table S2.
Based on data from numerous FSW and client IBBA surveys between 1985 and 2016 [26,[31][32][33][34][35][36], and national data on increases in condom distribution between 1988 and 1997 [37], condom use during commercial VI sex between FSW and their clients (Figure 2a) was assumed to increase from negligible levels in 1985 [33], up to a high stable level from 1998 onwards between 54 and 90% [24, 26,34]. This large range was based on the difference between client-and FSWreported condom use estimates in 2015 to 2016. Similarly, based on data from MSM IBBA surveys, condom use in last sex act for male partners of MSM was assumed to be negligible in 1985, low (10% to 30%) in 2001 [38], up to 70 to 85% by 2003 to 2007 [22,39,40] and constant thereafter ( Figure 2b) (2016 survey). The Data S1 and Figure S1 include further details on the condom use assumptions. Senegal population size estimates for 1980 to 2020 and gender-specific death rates were obtained from UNDP [41], with population growth rates being fit to this data. Population size estimates for FSW and MSM were produced as part of the 2015 IBBA surveys using the service multiplier and unique object methods [42]. The population size of clients was estimated through balancing the overall demand for commercial sex of FSW with that of clients (see Data S1).
ART data for the Senegal population came from the World Bank and UNAIDS [43,44] ( Figure S2), suggesting ART coverage increased from negligible levels in 2000 to 40% of individuals living with HIV in 2015. For FSW, these coverage trends were scaled up because the 2016 IBBA found 74% of FSW living with HIV were virally suppressed (unpublished data from 2016 FSW survey [25]). ART recruitment rates were calibrated to give the trends in ART coverage. Other HIV biological parameters came from literature (Table S2).
Uncertainty ranges were assigned to most model parameters, with most parameters being fixed over time except for the rate of ART recruitment, levels of condom use and frequency of sex for MSM with their male partners, which data suggest increased from 2007 to 2016 [21,25]. To incorporate uncertainty, 10,000 parameter sets were randomly sampled from their uncertainty ranges (Table S2 and Table 1). For each parameter set, the model was run while including temporal increases in condom use, ART coverage and the frequency of sex for male MSM partners. Any run producing HIV prevalence projections that agreed with early IBBA HIV prevalence data for FSW (1990FSW ( or 1995 and clients (1999) and recent HIV prevalence data for young MSM from 2014 to 2016 were selected as a model fit. These data suggested a HIV prevalence of 2.0 to 10.0% in 1990 and 5.0 to 15.0% in 1995 in FSW, 1.1 to 5.5% in clients (1999) and 9.7 to 37.7% in MSM, with these prevalence estimates shown in Figure 3 (data sources in Table S1). The wide range for MSM is due to contrasting estimates from two IBBA surveys in 2014 and 2016. Other HIV prevalence data for all sub-groups are also shown in Figure 3. The model was not calibrated to these data, but instead the data were used to validate the accuracy of the model projections.

| Contribution of commercial sex and sex between men to HIV transmission
To estimate the contribution of commercial sex between FSW and clients and sex between men to the overall HIV epidemic (referred to as population attributable fraction or PAF), model fits were used to estimate the proportion of new HIV infections that would be prevented by setting the transmission probability for commercial sex or sex between men to zero over a specific time period. This was estimated for 1 or 10 years from 1995, 2005 and 2015.

| Impact of existing interventions
Model fits were used to explore the likely impact of historical increases in ART coverage and condom use on the evolution of the HIV epidemic for different population sub-groups. This was determined by re-running the model fits, but with no ART and/or condom use.

| Impact of scaling up interventions
We then assessed the impact of increasing the coverage of ART among MSM from 2017 to 2030, such that the proportion of MSM living with HIV who are virally suppressed increases to the same coverage as FSW (74%) by 2020. This increase in ART coverage was also considered among low-risk individuals. To capture the possible effect of introducing pre-exposure prophylaxis for HIV (PrEP) and/or further increases in condom use, we also investigated the impact of an intervention that reduces an individual's average risk of becoming HIV positive. This was estimated by reducing by a third the susceptibility to HIV transmission among (i) MSM, (ii) FSW or (iii) all low-risk individuals. This is roughly equivalent to putting 40% of MSM on PrEP (assuming 85% effectiveness [46,47]), 50% of FSW on PrEP (assuming 66% effectiveness [47,48]) or reducing the number of unprotected sex acts by 40% through increased condom use (assuming 82.5% effectiveness per sex act). We also assumed a scenario where both were achieved for MSM (PrEP and condom use), reducing susceptibility by two-thirds.

| Uncertainty analysis
We performed linear regression analyses of covariance (ANCOVA) [49] to determine which parameters contribute most to the variability in the 10-year PAF estimates for commercial sex and sex between men for 2015. 3 | RESULTS

| Existing epidemiological insights
Two hundred model runs agreed with the HIV prevalence calibration data from FSW, clients and MSM. Despite only being calibrated to one-third of the prevalence estimates (5/17 data points), these model fits agreed closely with observed HIV prevalence trends for all groups (Figure 3). These model fits suggest the HIV prevalence among FSW, clients and the adult population has been in steady decline since the mid-90s, but may have increased recently among MSM.

| Contribution of commercial sex and sex between men to HIV transmission
For the 10 years from 1995, a median of 42.1% (2.5 to 97.5%, percentile range 23.7 to 59.0%) of new HIV infections could have been prevented (10-year PAF) if the risk due to sex between men had been removed over this period, respectively, with this increasing to 64.1% (37.4 to 79.4%) by 2015 (Figure 4). The increase in the 10-year PAF for sex between men in 2015 is due to the increase in the frequency of AI sex acts between MSM over this time period. Much of this effect of MSM is due to their heterosexual partnerships with females, such that removing this risk prevents 37.1% (18.7 to 51.3%) of HIV infections over 10 years from 2015 ( Figure 4). In contrast, the 10-year PAF for commercial sex was 20.6% (7.3 to 38.7%) and 13.6% (4.8 to 35.0%) for 1995 and 2015 respectively (Figure 4). For sex between men, one-year PAFs were lower than 10-year PAFs, but were similar for commercial sex ( Table 2).
The analysis of covariance showed that many parameters contributed to the uncertainty in the PAF estimates (Data S1).

| Impact of existing interventions
Model projections ( Figure 5)

| Impact of scaling up interventions
Scaling up ART among MSM such that 74% are virally suppressed by 2020 (35.4% of MSM living with HIV are currently virally suppressed) would decrease the overall HIV incidence in Dakar by 14.7% (4.9 to 47.1%) by 2030 ( Figure 6). If the HIV susceptibility in MSM is also reduced by a third then incidence would decrease by 43.9% (14.9 to 76.0%), while it decreases by 68.3% (42.0 to 88.6%) if susceptibility reduces by 66%. This captures what could be achieved from putting 40% of HIV-negative MSM on PrEP and reducing the number of non-condom protected sex acts by 40%. However, further decreases in FSW or general population susceptibility (by a third) or increases in ART coverage in the general population (so 74% are virally suppressed) will not add much more impact, with these interventions in combination decreasing the overall incidence by 82.4% (62.4 to 93.3%). This is due to the small contribution of these groups to HIV transmission in Dakar.    [46,[50][51][52][53], cost-effective [54,55] and feasible among MSM, and so their expansion should not be delayed. As same-sex practices remain criminalized in Senegal, and MSM face discrimination [56], it is also important that effective stigma mitigation interventions are combined with these interventions to ensure their effectiveness [25]. Our findings suggest little impact is achieved from reducing risk or increasing ART coverage among other population subgroups, including FSW, further emphasizing the need to focus new interventions on MSM. However, this is contingent on sustained high levels of condom use and ART uptake among FSW. That is, our results do not suggest that efforts should be transferred from existing interventions to refocus on MSM interventions, but rather that additional efforts are focused on MSM. Indeed, our projections emphasize existing interventions have had considerable impact, halving HIV prevalence compared to what it could have been in 2017.

| STRENGTHS AND LIMITATIONS
Strengths of our analysis include the use of detailed setting-specific data from numerous bio-behavioural surveys among FSW, their clients and MSM from 1985 to 2015, as well as two general population surveys from 2005 and 2010. In addition, the inclusion of MSM in our modelling improves on previous models that have frequently ignored MSM based on assumptions of decreased relevance across Sub-Saharan Africa [57], and rarely included them in dynamic PAF estimates [9,58]. Lastly, another key strength in our analyses is the accuracy of our model projections compared to data that were not fitted, which included two-thirds of available HIV prevalence estimates.
Despite using best available data, this model has several limitations. It did not incorporate any commercial sex among MSM due to limited data. Fortunately, including commercial sex among MSM would not have changed our general results since the HIV infections would still be attributed to same-sex practices. The model also did not explicitly include TGW because of insufficient data to do so. However, the 2016 MSM survey used to parameterize and calibrate the model included some TGW, as did the MSM size estimation data used by the model, and so their contribution was captured to an extent by the model. Moving forward, these analyses and associated bio-behavioural surveys need to better assess gender identity to enable a better assessment of their role in HIV transmission. The modelling was also limited by uncertainty in many model parameters. In addition, differences in the condom use and behavioural measures used by studies made it hard to evaluate temporal changes in risk. To account for these uncertainties, we associated wide ranges to all uncertain parameters and trends in condom use and risk behaviours, and used Bayesian fitting methods to account for and constrain this uncertainty through calibrating to HIV prevalence data. Importantly, our findings were robust to this uncertainty.
Another limitation of our analysis was the relative simplicity of the model with respect to HIV natural history, the portrayal of ART and heterogeneity in sexual behaviours among each sub-population. Although greater detail has been included in other models, there is no consensus on the appropriate level  Figures show baseline (median with 95% credibility intervals) trends, and median trends with no effect of ART (median-blue) or no condom use (median-orange).
of complexity for specific models. One specific simplification is not stratifying MSM by whether they normally have insertive or receptive anal sex. This was done to avoid overly complicating the model, which should not have affected our results because they have the same sexual behaviour with females; the main indirect mechanism for MSM contributing to overall HIV transmission. Lastly, the model only considered Dakar. Although this limits the generalizability of the findings, it does increase the likely precision of the modelling because we did not make generalizing assumptions to produce an average portrayal of the Senegal epidemic. Despite the limited scope of the analysis, it is still likely that the results are relevant to the whole of Senegal because the HIV prevalence in MSM is generally high and increasing, while it is lower and decreasing among FSW. In addition, our analyses are also likely relevant to other West African settings with growing epidemics in MSM, similar or lower HIV prevalence in FSW [3,4], and similar reporting of heterosexual sex among MSM [59][60][61].

| CONCLUSION AND IMPLICATIONS
Previous analyses have shown a large role for commercial sex in some sub-Saharan HIV epidemics [10][11][12][13][14], but this is the first to suggest that MSM, including individuals who identify as cis-male and TGW, could also be having an important role in some lowerand middle-income country settings, where other analyses have suggested a small role [14,15,62]. This is likely due to the different behavioural and epidemiological situation among MSM in Dakar compared to these other settings. Although existing interventions have been successful in reducing HIV transmission, including early increases in condom use and recent scale-up of ART among FSW, public health efforts are now needed to address the ongoing unmet need among MSM. This unmet need is resulting in MSM in Senegal and other settings in SSA experiencing uncontrolled HIV epidemics with high prevalence [57], which is driving HIV transmission in this setting and could be elsewhere. Scaling up prevention and treatment interventions for MSM should now be a high priority, with these initiatives needing to be sensitive to the legal context, and associated stigmas or discrimination that MSM experience. Without this policy shift, the HIV epidemics among MSM in Senegal and elsewhere in SSA are unlikely to decrease.

C O M P E T I N G I N T E R E S T S
The authors have no competing interests.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article: Data S1. Supplementary materials. Table S1. Details of surveys used to parameterise and calibrate the model for different risk groups in Dakar  Table S2. (a) Demographic, sexual and behavioural parameters of female sex workers, clients, MSM and low risk populations.
(b) HIV epidemiological parameters Figure S1. Modelled condom use trends for female sex workers (FSW) and men who have sex with men (MSM). (a) FSW condom use with commercial partners for vaginal intercourse. Assume condom use for anal intercourse (AI) is half that of vaginal intercourse (VI) for all the years. (b) FSW condom use with main partners for VI. Condom use for anal intercourse (AI) with main partners is assumed to be 0.6 to 1.0 times that for VI. FSW condom use with casual partners for VI is 1 to 1.5 times that for main partners VI; AI with casual partners is 0.6 to 1.0 times that for VI with casual partners. (c) MSM condom use with male regular and casual partners. Assume some bias in reporting so all rates are multiplied by a bias factor of 0.7 to 1.0lower bound of 0.7 chosen to give overall lower bound of 0.5. (d) MSM condom use with female main and casual partners. Condom use is assumed to be the same for VI and AI. Figure S2. ART coverage in Senegal from 2000 from UNAIDS AIDS info [44] and World Bank [43] Figure S3. Analysis of covariance results of parameters that contribute more than 4% variability to the 2015 (a) 10-year commercial sex PAF and (b) 10-year MSM PAF.