The cost-effectiveness of oral HIV pre-exposure prophylaxis and early antiretroviral therapy in the presence of drug resistance among men who have sex with men in San Francisco

In July 2012, the US Food and Drug Administration (FDA) approved tenofovir/emtricitabine for use as oral pre-exposure prophylaxis (PrEP) [1]. Estimates of PrEP efficacy in preventing HIV infection range from 39% to 86% in randomized controlled trials [2‐6]. PrEP consumer demand has accelerated since mid-2013 [7]. PrEP coverage among men who have sex with men (MSM) in San Francisco was estimated at 9.6% in 2014 [8]. In May 2016, more than 6000 MSM in San Francisco were reported to receive PrEP [9], suggesting a coverage of approximately 12% given that the HIV-negative MSM population is estimated at 50,000 [7, 10, 11]. In 2010, prior to this increased PrEP uptake, San Francisco was one of the first cities to institute guidelines to initiate antiretroviral therapy (ART) as early as possible post-infection rather than waiting for signs of disease progression, such as clinical symptoms or low CD4+ cell counts [12], given findings that early ART initiation improves survival while reducing the risks of transmission to others [13]. However, it has been hypothesized that early ART initiation might provide more time for the evolution of drug resistance and that subsequent transmission of drug-resistant HIV might reduce PrEP effectiveness [14‐16].

Herein, we address the long-term population-level costs and health benefits of expanding PrEP coverage in combination with increasingly early ART initiation among the MSM population in San Francisco. We assess the effectiveness and cost-effectiveness of expanding PrEP coverage, with and without earlier ART guidelines, using an infection–age-structured model [17, 18]. The model tracks the transmission rate and life expectancy of individuals at each infection age, while accounting for ART-mediated drug resistance and the costs of treatment failure and second-line regimens. We fit this model to epidemiological data on the annual incidence of newly diagnosed AIDS cases and deaths amongst MSM in San Francisco (simultaneously matching prevalence, resistance, and demographic data [18]) and estimate the impact of various intervention scenarios over the next 20 years.

Several modeling studies have addressed the impact of PrEP in San Francisco, including estimates of coverage required to curtail transmission [7], an assessment of whether PrEP might increase the drug-resistant transmission [14], and a cost-effectiveness analysis of PrEP in combination with a partially effective HIV vaccine [19]. Our study differs from these prior analyses and others of PrEP cost-effectiveness [20‐26] in the United States in several aspects. First, we analyze the expansion in PrEP coverage in conjunction with changing ART guidelines, rather than comparing initiation of PrEP to no PrEP [7, 14, 19‐24, 27, 28]. Second, we model the acquisition and transmission of ART-mediated resistance and the indirect population-level effects of PrEP (i.e., by reducing the number of infected individuals, PrEP indirectly benefits other individuals in the population who could have otherwise been infected by individuals on PrEP), which have not been addressed in previous studies [7, 19‐25]. While two prior analyses of PrEP [16, 26] considered ART-mediated resistance and subsequent treatment failure, one [16] did not address the impact of earlier ART initiation on PrEP cost-effectiveness, and the other [26] did not account for increased costs and effectiveness of second-line drugs. Third, we consider increasingly earlier initiation of ART and the consequent increases in survival. In contrast, many prior studies modeled early ART by assuming individuals initiate ART whilst their CD4+ T cell counts remain above 350 cells/mm³ [29], by assuming ART eligibility occurs once their CD4+ T cell count drops below 500 cells/mm³ [30, 31], or by assuming ART initiation occurs within 2 years of infection [32]. Two studies that modeled ART initiation at 1 year post-infection [33, 34] did not incorporate the increased survival times, which may cause underestimation of intervention cost-effectiveness.

We extended our previously developed infection–age-structured model [18] to include a PrEP class (see the schematic diagram of the model structure in Fig. 1a, Additional file 1: Figure S1 and Table S1). We modeled the MSM population in San Francisco aged 18–65 years old [10, 18], assuming that first- and second-line ART reduced infectivity by 96% [13] and 80% [18], respectively. Here, the effectiveness of second-line drugs for drug-resistant cases is assumed to be lower than that of first-line drugs for drug-sensitive cases due to lower adherence [35]. We assume that earlier ART initiation confers longer life expectancy, as indicated previously [18, 36], based on clinical data [37, 38] and that treated individuals with drug resistance have life expectancies that are 11 years (varying from 0 to 20 years in sensitivity analyses) shorter than drug-sensitive patients [18, 37, 38]. We estimate that the average period from infection to ART initiation in San Francisco has been 1.6 years, on average [18] (‘early ART’, see Additional file 1: Supplementary Material for details). We compared early ART to ‘earlier’ ART, i.e., treatment initiation at 1 year post-infection, on average. Of the 33% of treated MSM who are virally unsuppressed [10], 76% exhibit drug resistance [39]; we thus assume that 25% of treated MSM in San Francisco [18] have acquired drug resistance as our base case (varied from 0 to 100% in a sensitivity analysis). Resistance here is assumed to be resistant to one or more first-line ART drugs. Given local monitoring protocols, we assume that drug-resistant cases are quickly switched to second-line drugs. We do not differentiate drug combinations beyond the categorizations of first- or second-line for two reasons. First, data to estimate drug-specific efficacies and adherence levels is not readily available. Second, drug-specific parameters are unnecessary to achieve our goal of estimating intervention cost-effectiveness when using these assumptions regarding the average relative effectiveness of first-line and second-line regimens.

We assumed PrEP effectiveness against drug-sensitive strains was 53% based on a meta-analysis [40], where this value reflects both biomedical efficacy and adherence. We assumed relative PrEP effectiveness against resistant strains was 50% (the ratio of PrEP effectiveness against drug-resistant versus drug-sensitive strains) [14‐16]. In a sensitivity analysis, we varied PrEP effectiveness against drug-sensitive strains from 10% to 90% [21, 27] and relative effectiveness against drug-resistant strains from 0 to 1. We assumed an 8% annual rate of PrEP attrition, based on a cohort study in San Francisco [41], and varied it from 1% to 30% per year in sensitivity analyses. Based on low empirically observed dropout proportions [10], we assume that all individuals who initiate ART remain on ART until the end of life and do not drop out of care. We considered only ART-mediated and not PrEP-mediated resistance, because both clinical data [42, 43] and mathematical models [44, 45] suggest that PrEP contributes less than 5% to the total burden of resistance since PrEP-selected resistant phenotypes decay below detection by 6 months after drug cessation and remain undetectable for at least 2 years thereafter [46].

We found that expanding PrEP while initiating ART earlier (at 1 year post-infection versus the status quo of 1.6 years post-infection) could provide substantial, cost-effective health benefits to the San Francisco MSM population. However, the most cost-effective intervention is earlier ART without further PrEP expansion. This would prevent 9700 (22%) total new infections and 1457 (8%) new drug-resistant infections, and add 43,649 QALYs over the next 20 years; this is very cost-effective at $4745 per QALY gained. Expanding PrEP coverage in addition to earlier ART would further reduce incidence, but only add 1.7–2.4 times more QALYs while costing 28–58 times more than earlier ART alone (Table 1). Still, high PrEP coverage plus earlier ART ($115,320 per QALY gained) is cost-effective for San Francisco even considering the acquisition and transmission of ART-mediated drug resistance. For a budget under $470 million annually, the best strategy is to keep current strategies of PrEP enrollment while initiating ART earlier. This may be advisable for San Francisco, based on the city’s 2015–2016 annual HIV/AIDS budget [58] (though this excludes private insurance and may therefore underestimate the total funding available). A higher budget would suggest that PrEP expansion in addition to earlier ART is advisable.

Our estimated ICER of expanding PrEP without earlier initiation of ART of $117,130–132,520 per QALY gained (Table 1) is far greater than estimates for targeted high-risk MSM interventions in New York City ($31,970 [20]), Los Angeles County ($27,863–37,181 [26]), and nationwide US ($52,443 [23]), but lower than estimates for the general MSM population in New York City ($353,739 [22]) and nationwide US ($172,091–216,480 [23]). These discrepancies may stem from the different settings, PrEP targeting strategies, or model assumptions. For example, San Francisco has higher testing and treatment rates than New York City and the entire US [10, 18, 22, 23]. We assume two-fold higher life expectancy gains following treatment than Juusola et al. [23], based on more recent life expectancy estimates [37, 38]. Recent reviews discussed variation amongst ICER calculations in greater detail [28, 59].

Drabo et al. [26] conducted a similar study for high-risk MSM in Los Angeles, but used a different model structure. Our infection–age-structured model allows us to track the life expectancy at different ART initiation times and among individuals with drug sensitive or resistant strains (the primary innovation of our model). In addition, their model initiated ART treatment at a CD4+ cell count ≤ 500 cells/mm³, whereas we considered an even earlier ART strategy. Finally, we accounted for the switch to second-line drug treatment after treatment failure and its associated costs. Despite these differences, we concur that all combinations of earlier ART and PrEP coverage are cost-effective to varying degrees, and that the optimal strategy depends on budgetary constraints.

Our study has several limitations. First, to explore the cost-effectiveness of the above scenarios, we fit our model to multiple sources of data, including incidence of AIDS diagnoses and deaths, HIV prevalence, the fraction of drug-resistance amongst incident cases, and other demographic data [18]. Fitting to so many sources data has the advantage of integrating much of what is known about this epidemic but is ambitious in the sense that it is difficult to closely fit all data sets simultaneously. One possibility to solving this would have been to fit only a subset of these data such that model fits would have better reflected those datasets but not been informed by others. Second, we did not model PrEP-mediated acquired drug resistance. Our results should be robust to this exclusion, as recommended screening practices are thought to limit PrEP-acquired resistance via quick identification of breakthrough infections amongst individuals on PrEP [21, 43, 60]. Third, we assumed that sexual partners were assorted independent of PrEP or ART usage. If PrEP were targeted towards individuals whose partners, if infected, were less likely to be on ART, then PrEP would be expected to be more effective and cost-effective. Should individuals on PrEP be more likely to engage in sexual partnerships with infected individuals on ART than with those not on ART, then this would reduce the estimated cost-effectiveness of PrEP. Fourth, we assumed a homogenous PrEP coverage for the entire population irrespective of risk and age. We made this assumption for simplicity and due to the lack of information about age- and risk-group mixing contact patterns. If PrEP coverage is higher in individuals within the MSM population who are at higher risk of infection than the average (i.e., because of PrEP targeting to high-risk groups or high-risk individuals seeking out PrEP), then PrEP would be more cost-effective. In contrast, if high-risk groups also constitute groups that are hard to reach with PrEP, then PrEP cost-effectiveness would be lower than estimated. Similarly, the indirect effects of PrEP on the population would depend on the propensity to transmit of individuals on PrEP, should they have been infected if they were not on PrEP. Finally, while our results were qualitatively robust to parameter uncertainty, future changes in intervention costs or efficacy may affect our estimates.

In summary, expanding PrEP coverage and shifting ART initiation even earlier in San Francisco would reduce the total and drug resistant incidence, and add over 40,000 QALYs to the population over 20 years. Combining higher PrEP coverage and earlier ART is predicted to be cost-effective at three times the per capita GDP threshold. These results are robust across a wide range of assumptions regarding drug resistance and the effectiveness and cost of second-line drug regimens.