Optimizing HIV pre-exposure prophylaxis implementation among men who have sex with men in a large urban centre: a dynamic modelling study

Introduction Once-daily tenofovir/emtricitabine-based pre-exposure prophylaxis (PrEP) can reduce HIV acquisition in men who have sex with men (MSM), by 44% in the iPrEx trial, and reaching up to 99% with high adherence. We examined the potential population-level impact and cost-effectiveness of different PrEP implementation strategies. Methods We developed a dynamic, stochastic compartmental model of HIV transmission among the estimated 57,400 MSM in Toronto, Canada. Parameterization was performed using local epidemiologic data. Strategies examined included (1) uniform PrEP delivery versus targeting the highest risk decile of MSM (with varying coverage proportions); (2) increasing PrEP efficacy as a surrogate of adherence (44% to 99%); and (3) varying HIV test frequency (once monthly to once yearly). Outcomes included HIV infections averted and the incremental cost ($CAD) per incremental quality-adjusted-life-year (QALY) gained over 20 years. Results Use of PrEP among all HIV-uninfected MSM at 25, 50, 75 and 100% coverage prevented 1970, 3427, 4317, and 4581 infections, respectively, with cost/QALY increasing from $500,000 to $800,000 CAD. Targeted PrEP for the highest risk MSM at 25, 50, 75 and 100% coverage prevented 1166, 2154, 2816, and 3012 infections, respectively, with cost/QALY ranging from $35,000 to $70,000 CAD. Maximizing PrEP efficacy, in a scenario of 25% coverage of high-risk MSM with PrEP, prevented 1540 infections with a cost/QALY of $15,000 CAD. HIV testing alone (Q3 months) averted 898 of infections with a cost savings of $4,000 CAD per QALY. Conclusions The optimal implementation strategy for PrEP over the next 20 years at this urban centre is to target high-risk MSM and to maximize efficacy by supporting PrEP adherence. A large health benefit of PrEP implementation could come from engaging undiagnosed HIV-infected individuals into care.


Introduction
HIV incidence among men who have sex with men (MSM) in high-income countries remains high and concentrated in large, urban settings [1Á6]. In Canada, the incidence of HIV infection among MSM ranges from 0.62 per 100 person-years to 1.14 per 100 person-years, similar to the ranges in other developed countries [2Á6]. In 2014, 837 new HIV infections were diagnosed in the province of Ontario, with half occurring in Canada's largest city, Toronto [5,6]. Of the estimated 57,400 MSM living in Toronto, nearly 20% have HIV [1]. Despite the scale-up of a combination anti-retroviral therapy (ART) and sustained investments in behavioural prevention programmes [7,8], the rates of newly diagnosed HIV infections and HIV-attributable deaths in Toronto MSM have not markedly declined over the last 10 years, and remain a major public health concern [1]. Daily use of tenofovir/emtricitabine (Truvada † ) by HIV-uninfected individuals as pre-exposure prophylaxis (PrEP) has been shown in randomized trials to reduce HIV acquisition in high-risk groups [9Á11]. The iPrEx study, a placebo-controlled trial in high-risk MSM, found that those randomized to tenofovir/emtricitabine were 44% less likely to acquire HIV compared to placebo over a median follow-up of 1.2 years [9], and pharmacokinetic analyses suggest efficacy of up to 99% if adherence is high [12]. More recent data from the iPrEx open-label extension corroborate this estimate, with dosing of four to seven times per week associated with virtually 100% (95% CI 086,100%) efficacy [13]. To inform broader PrEP implementation, additional demonstration projects are underway across North America, including the US PrEP community-based demonstration project out of New York City, an NIH-funded communitybased project out of San Francisco [14], and PREPARATORY-5 (NCT02149888), a demonstration project in Toronto evaluating PrEP acceptability, effectiveness and sexually transmitted infection rates. To contextualize the findings of these regional demonstration projects and prepare for wide-scale PrEP delivery, we need to understand how to maximize its population-level impact on HIV transmission in large, urban centres, given the added cost to the healthcare system.
Dynamic mathematical models of the population impact of PrEP on HIV spread among MSM have been developed on both national and sub-national scales in high-[15Á19] and low-income regions [20]. Five modelling studies of highincome countries suggest that PrEP is most cost-effective when restricted to the highest risk subgroups [15Á19]. Not surprisingly, these studies also showed that uptake and adherence would have large effects on predicted HIV outcomes. However, key implementation questions remain unanswered. For instance, although clinical guidelines recommend that HIV testing be done every three months in individuals using PrEP on the basis of the protocols used in clinical trials, the optimal HIV screening frequency remains unclear. It is also unclear what proportion of the total benefit of PrEP programmes accrues from the ability of PrEP-related HIV testing to diagnose infections earlier, versus from the use of PrEP itself. Furthermore, most models of PrEP populationeffectiveness and/or cost-effectiveness in high-income settings examined national-level epidemics by drawing on national-level sexual behaviour and HIV surveillance data [15,17Á19]. However, a national perspective belies the heterogeneity in HIV epidemics between locales, such as states, provinces or major cities [4]. Furthermore, HIV prevention programmes are often funded and administered at a regional level [6,21]. There is often marked heterogeneity of prevalence and rates of HIV infection/diagnosis across regions within countries, and comparing metropolitan and nonmetropolitan regions [22,23]. Regional models of PrEP may thus offer a more relevant assessment of implementation costs and outcomes for the purpose of guiding local interventions, and may be more generalizable to regions with similar epidemic characteristics. To date, there has been no modelbased evaluation of PrEP implementation in a Canadian city.
To address these gaps, we developed a mathematical model of HIV spread in Toronto MSM using the best available, regional epidemiologic data. Our aim was to evaluate the impact of different strategies of PrEP delivery in Toronto MSM on the following outcomes: reduction in the total number of diagnosed and undiagnosed HIV infections, total number of HIV-related deaths averted, incremental costs and cost per quality-adjusted life year (QALY) gained, over a 20-year period. Strategies examined included targeting PrEP at the highest risk MSM (with varying proportions of coverage) and varying the frequency of HIV screening in those on PrEP. We further assessed the sensitivity of outcomes to varying rates of adherence.

Model design
We developed a dynamic, deterministic-stochastic hybrid compartmental model of HIV spread among MSM in Toronto, Canada. The model was represented by a combined deterministic-stochastic model [24] of state-transitions based on the probability of moving from one compartment to another. The transitions were stochastic for compartment population sizes fewer than 5000 individuals, and were deterministic for compartments exceeding 5000 individuals. The motivation for using a stochastic approach was to better accommodate relatively small population sizes within compartments that would be encountered in a regional model. A deterministic function was used to accommodate the few compartments that would have a consistently higher population, allowing improved computational efficiency. For this analysis, we chose to present mean values, corresponding to repeated realizations of the model. Figure 1 shows the schematic of the model structure, while Table 1 shows the parameter values. Model compartments reflected the natural history (or ''states'') of untreated and treated HIV, and populations were further stratified by known/unknown HIV serostatus and sexual behaviour (high/ low risk). The natural history of untreated HIV was divided into four progressive stages reflecting data from untreated HIV-infected cohorts [25]. Each stage was associated with a different probability of infectiousness and HIV-attributable mortality (Table 1), and defined by time to specific CD4 cell counts [25]: (1)  The sexual behaviour of high-risk MSM was drawn from the annual number of sexual partners taken from regional empiric data (using the highest decile of the reported number of sexual partners in the last year). These data were collected from MSM undergoing anonymous HIV testing at a downtown Toronto sexually transmitted infection clinic in Spring 2013 and suggested that the highest decile of MSM had a mean of 36 partnerships in the last year [26]. The remainder had a mean of five partnerships in the last year. For this aforementioned study, research ethics board approval was obtained from the University of Toronto, protocol #30129. Consent for Figure 1. A schematic of the model compartments and intercompartmental flow, organized by serostatus (known/unknown). *Not denoted are the entry of individuals into the population as susceptible, and the exit of individuals out of the population. **Compartment numbers are listed in brackets for simplicity of reference with model equations. Odd numbers refer to high sexual activity/risk compartments, and even numbers refer to low sexual activity/risk compartment. the use of these data was obtained at the time of entry into the study. Data analysis performed for this study was done using anonymous and deidentified data.
HIV serostatus was based on an individual's awareness of his diagnosis of HIV. We assumed that HIV infectivity decreased after a diagnosis of HIV infection, based on behaviour change Unless otherwise specified, all units are per year. Subscripts (i and j) denote compartment of reference (See Figure 1); b Annual testing rate derived based on number of HIV tests performed annually in MSM within the City of Toronto and the size of the susceptible MSM population. observed in population-based studies [27]. The reduction in infectivity was operationalized as a reduced probability of transmission between HIV-infected and uninfected individuals, reflecting either increased use of condoms or seropositioning favouring lower transmission, but not a change in the number of partners. HIV transmission between compartments was a function of the following variables (Table 1): (1) the number of sexual partners per individual per annum; (2) condom use, which was combined with condom efficacy; (3) HIV transmission probability based on HIV stage, awareness of HIV serostatus and HIV treatment status and (4) PrEP use, which decreased HIV susceptibility (per sex act) but did not impact on the number of partners.
Individuals who were diagnosed with HIV could initiate ART at a rate that varied depending upon CD4 stage, as described below. Similarly, individuals who were on ART could discontinue treatment and return to the diagnosed, infected, but untreated population with CD4 !500. In the intervention scenarios, PrEP could be initiated among HIV-uninfected individuals. If taking PrEP, the likelihood of acquiring HIV infection per act was reduced by 44% (unless otherwise evaluated). If individuals became infected with HIV while on PrEP, they were infectious while undiagnosed (with no effect of PrEP on HIV infectivity). Once diagnosed, they moved into the diagnosed HIV category. We did not incorporate the potential impact of PrEP use on emergence of antiretroviral drug resistance.
Epidemiological and behavioural data for model parameterization Toronto-specific epidemiologic data was used wherever possible for parameters and initial conditions (Table 1,  Supplementary Tables 1 and 2). Baseline (non-HIV) mortality rates were based on average North American life expectancies [28], and adjusted based on expected increased mortality in late-stage (CD4 B200) HIV infection [15]. The probability of HIV transmission, as a function of the stage of CD4 count (transmissions per partnership per year) was drawn from the literature [15,29Á31]. The annual HIV testing rate among MSM in Toronto was estimated by dividing the mean annual number of HIV tests performed in Toronto MSM by the estimated population of undiagnosed MSM [1]. We used an annual 1% population growth rate of susceptible MSM, in keeping with Canadian population growth rates [32]. Individuals on ART were assumed to be fully adherent and non-infectious [33]. Estimated rates of condom use and efficacy, used for calibration, were based on published values and local data [1,34,35].

Cost and quality of life data
Cost-effectiveness analysis was conducted from the healthsystems perspective and included costs associated with inpatient and outpatient clinical care and diagnostics. PrEP, ART and HIV/AIDS related care costs were drawn from previous studies based on Canadian estimates, and included physician-visit and nursing costs, acute hospitalization costs, diagnostic testing and prescription costs [36,37]. Specific diagnostic testing costs in susceptible individuals were determined based on provincial (Ontario) cost listings for HIV serology as of July 2013 [37]. Estimates of stage-specific QALYs lost/gained among persons with HIV/AIDS and those on ART were determined from the literature [38]. We assumed that PrEP did not affect quality of life. In evaluating cost-effectiveness, we determined the total health care related costs, as well as costs associated with PrEP implementation, HIV treatment related costs, untreated HIVassociated costs, HIV testing costs among the susceptible population and QALYs. QALYs were calculated at each time step of the model. Costs and QALYs were discounted at a rate of 3% per annum.

Model calibration
We calibrated the model at an equilibrium state against the following historical estimates (10-year period in Toronto, 2001 to 2011) using acceptance-rejection sampling: (1) a stable annual HIV diagnosis rate (330 to 400 cases per annum) and (2) a stable annual HIV-attributable mortality (58 to 72 cases annually) [1,39]. Parameter values were selected using Monte-Carlo sampling. The calibrated parameters are listed in Table 1, and included those about which we were most uncertain and those expected to influence study outcomes the most: ART initiation and cessation rates, reduction in risky sexual behaviour after HIV diagnosis, AIDS mortality, condom use and efficacy. We cross-validated the model-generated HIV prevalence ratio (high-risk versus low-risk population), against local data on HIV prevalence among the 10% of MSM with the highest risk of HIV (using syphilis co-infection as a risk factor) compared to the lowest-risk 90%, which was a ratio of 3 [35].

Analyses
We used the accepted parameter sets to simulate the baseline scenario (without PrEP), and different strategies of PrEP delivery. PrEP was introduced at an HIV-equilibrium state and outcomes were measured at 20 years. Because we were interested in the mean outcome per parameter set (i.e., the average of a large number of stochastic realizations), we performed 150 stochastic realizations per parameter set to ensure the mean incidence of diagnosed and undiagnosed HIV infections varied less than 1% with each additional simulation.
Outcomes included the impact of daily PrEP on cumulative number of newly diagnosed and undiagnosed HIV infections, the cumulative number of HIV-associated deaths, annual costs and cost per QALY. Newly diagnosed infections included those individuals who were already infected at the start of the simulation, and then were subsequently diagnosed. We compared the following PrEP implementation strategies: (1) targeting all MSM (high-and low-risk sexual activity) versus only high-risk activity MSM alone, with varying proportions of coverage; (2) varying the frequency of HIV testing in those taking PrEP including the scenario in which screening is performed without PrEP use and (3) increasing PrEP efficacy (as a surrogate for adherence) based on pharmacokinetic analyses from the iPrEx trial, from 44% risk reduction (corresponding to fewer than two doses per week), to 76% (two doses per week), 96% (four doses per week), and 99% risk reduction (seven doses per week) [12]. We then conducted a sensitivity analysis to explore the influence of changing coverage in an ideal adherence scenario (99% efficacy).

Clinical implementation parameters
To evaluate the impact of HIV testing frequency and PrEP efficacy among PrEP users, we assumed a baseline scenario in which 25% of high-risk MSM used PrEP. When keeping efficacy fixed at 44%, changing the frequency of HIV testing in those on PrEP had minimal influence on the number of HIV infections prevented over 20 years (1209 to 1113 HIV infections prevented with Q1 month versus Q12 month testing, respectively). Similarly, costs and cost-effectiveness did not significantly change with variation in testing frequency (Table  3). However, when keeping test frequency fixed at Q3 months, increasing PrEP efficacy from 44% to 99% resulted in a marked increase in infections prevented, from 1166 to 1540. Costeffectiveness improved with cost per QALY gained, decreasing from $34,999 to $15,275 CAD (Table 3). Finally, in the scenario of 25% coverage of high-risk HIV-uninfected MSM, we considered the option of HIV screening only and no PrEP administration. This resulted 898 infections prevented, 1164 QALYs gained and a cost savings per QALY gained of $4,332 CAD.

Discussion
In this regional model of HIV transmission, we found that PrEP implementation in all MSM, irrespective of risk profile, provides important reductions in HIV infections and mortality. Increasing the frequency of HIV testing in susceptible individuals on PrEP resulted in minimal improvement in the number of infections averted, even when testing monthly. When compared to established cost-effectiveness thresholds ($20,000 to $100,000 CAD per QALY) [41], general PrEP delivery to high-and low-risk MSM was not cost-effective.
Focusing PrEP on the highest risk HIV-uninfected MSM was a more efficient strategy than targeting all high and low risk at comparable levels. Specifically targeting a fraction of highest risk MSM (25% coverage) was the most cost-effective option, with a cost per QALY of approximately $35,000 CAD, similar to values previously described in the literature for comparable populations [15,16]. This occurs because even with 25% PrEP coverage, few high-risk MSM (in absolute numbers) remain susceptible to HIV. In contrast, improving PrEP efficacy in high-risk individuals, via better medication adherence [12,13], increased cost-effectiveness. Recent empirical studies suggest the potential for even better efficacy than the 44% efficacy reported in iPrEX [42]. With 99% efficacy, increasing PrEP coverage in high-risk MSM showed even less costs per QALY that could be achieved. Understanding a community's high-risk population (its relative size and relative HIV risk) will be important in guiding endpoints for PrEP implementation programmes, such that coverage targets can be set that optimize cost-effectiveness.
Increasing the frequency of HIV testing in susceptible individuals on PrEP resulted in minimal improvement in the number of infections averted, even when testing monthly. This is because most new infections still occur in non-users when PrEP efficacy is moderate to high. Decreasing the  frequency of testing (outside of guideline recommendations of every three months) suggests minimal cost savings and modestly more infections. There has also been documented drug resistance in breakthrough infections [43]. Together, these suggest that it may be more efficient to focus efforts on PrEP adherence rather than testing frequency if there are little marginal benefits. Given the cost savings of approximately $4,000 CAD per QALY of HIV testing alone in high-risk populations, our findings suggest a potentially large collateral health benefit of PrEP programmes via engaging more high-risk MSM in routine HIV testing, even if they do not ultimately take PrEP. This is likely occurring due to both behavioural changes in those who are diagnosed with HIV and early initiation of ART, both of which will lead to reduced transmission. Some of the barriers to HIV screening among MSM include a belief that they are not at risk for infection, fear of a positive test and fear that other people might find out they are infected [44Á47]. Particularly because of its novelty, PrEP programmes offer a new opportunity to engage these individuals in care by providing a means through which they may reduce their likelihood of infection, and potentially feel more comfortable with routine screening. Retention in PrEP programmes, regardless of PrEP initiation, may thus be an important programmatic objective, given the significant cost benefit of having regular screening in high-risk MSM.
Of note, with increased HIV testing in a PrEP programme, the fraction of undiagnosed men with HIV declines, and the number of new HIV diagnoses can outpace incident HIV infections. Since HIV diagnoses are used for surveillance and assessing response to interventions, this could misinterpret the population-effectiveness of PrEP programmes. There may be increased HIV cases diagnosed due to increased HIV testing, especially early in the PrEP programme. Similarly, observed declines in new diagnoses may underestimate true declines in new HIV infections.

Limitations
While a strength of this study is its regional focus, using a city with epidemic features similar to other urban MSM HIV epidemics [22,48], the findings must be interpreted in this context and are conditional on the assumptions and data inputs. First, we did not include female sexual partners of MSM. Based on available data, the sexual structure of the model did not detail a more nuanced sexual network and behavioural heterogeneity. Second, we tested the PrEP interventions under an endemic equilibrium, which was supported by local stability in the rates of HIV diagnoses and mortality in the last decade [1,4]. It is possible that this equilibrium could be disrupted, and the cost-effectiveness of PrEP may vary during periods of growing or declining epidemics [49]. A third limitation is our assumption that the cost of PrEP medication remains stable, even if adherence is less than daily. Dosing PrEP as infrequently as four times weekly has recently been estimated to still provide nearly 100% risk reduction among MSM [13]. Less frequent dosing and the possibility of future reductions in PrEP drug costs suggest the potential for further cost savings [12]. Fourth, the cost-effectiveness could be overestimated by the low fre-quency of HIV testing used for this population compared to guidelines. However, the baseline HIV testing rate drew on numbers of performed HIV tests and the susceptible population size [1] and is corroborated by other studies of the same population [35]. Fifth, cost-effectiveness ratios may be over-or underestimated if the quality of life ratio for individuals on PrEP is not equal to one Á the value we used in the absence of data to suggest otherwise. Lastly, we assumed ART adherence in those with diagnosed HIV to be excellent and highly effective, which is an increasing reality in the era of welltolerated once-daily dosing regimens in high-income settings [50]. However, the assumption of perfect ART adherence for those with HIV may overestimate the benefit of a testing-only strategy, and underestimate the potential benefit of PrEP in the susceptible population, whereby individuals with poor ART adherence could actively transmit HIV, increasing the force of infection. Important next steps include an evaluation of the impact of PrEP use on rates of ART resistance accumulation among those who become infected while taking PrEP, as has preliminarily been explored in some models [51]. Further, given the results of the placebo-controlled IPERGAY trial, in which ''on-demand'' PrEP (averaging roughly three to four tablets weekly) was associated with an 86% risk reduction [52], future work should assess how best to utilize daily and intermittent PrEP strategies from both a preventive efficacy and costeffectiveness perspective. Lastly, although clinical data has not suggested significant increases in risk behaviour on PrEP (''risk compensation'' or ''behavioural disinhibition'') [13,14], the clinical and cost-effectiveness impact of PrEP-related reductions in condom use and increased rates of other STIs warrant further study [53].

Conclusions
Providing once-daily PrEP for HIV prevention among MSM in a large, urban city could be associated with important reductions in HIV infections and deaths from the use of PrEP itself, and from newly engaging MSM in HIV care through PrEP-related screening. The most efficient approach to PrEP delivery would involve identifying high-risk MSM and aiming for modest PrEP coverage, with a focus on maximizing PrEP adherence over more frequent HIV testing. Competing interests DT has received investigator-driven research grants form Gilead Sciences and Viiv Healthcare. DT has also received honoraria for delivering educational lectures from Abbvie, Bristol Myers-Squibb, Gilead Sciences, Janssen, Merck and Viiv Healthcare. DT is a site principal investigator for a GSK-sponsored clinical trial. The authors have no other competing interests to declare.
Authors' contributions DM conceived of and designed the study and developed and analyzed the model. SM reviewed the model. DT and SM helped design the study and provided methodological and analytical inputs. DM wrote the manuscript. DT and SM critically edited and revised the manuscript and helped finalize the manuscript with DM.