Introduction

Since the announcement of the results of HIV pre-exposure prophylaxis (PrEP) trials and the HPTN052 early treatment for prevention trial, there have been crucial policy discussions about the use of antiretroviral (ARV) drugs to prevent HIV acquisition or transmission. With regards to PrEP, encouraging results were first reported for men and transgender women who have sex with men in the iPrEX trial [1], which showed a 44% (95% CI 15–63) reduction in HIV acquisition with a daily dose of tenofovir/emtricitabine (TDF/FTC). In two large trials, the Partners PrEP [2] and TDF2 [3] studies, PrEP was found to be effective in reducing the risk of heterosexual HIV transmission using either TDF or TDF/FTC daily (Partners PrEP) and TDF/FTC daily (TDF2). However, FEM-PrEP [4], a trial recruiting heterosexual women in South Africa, Tanzania, and Kenya for daily TDF/FTC was closed prematurely in 2011 for futility as was the oral TDF arm of the VOICE trial [5] in women in South Africa, Uganda, and Zimbabwe. Two topical PrEP trials have tested the efficacy of 1% TDF gel and a third, FACTS001 [6], is currently recruiting women in South Africa. The CAPRISA 004 trial [7] in Kwa Zulu-Natal found that pre- and post-coital vaginal TDF gel reduced women's acquisition risk by 39% (95% CI 6–60) but the VOICE trial stopped its gel arm when it became evident that daily gel use was safe but not effective [8].

Clinical guidance on oral PrEP has already been offered by the US Centers for Disease Control and Prevention, the Southern African HIV Clinicians Society, World Health Organization (WHO), and the British Association for Sexual Health and HIV [9]–[13]. An advisory panel to the US Food and Drug Administration recently recommended oral TDF/FTC for preventive use among people at higher risk of HIV exposure [14]. As PrEP emerges as an option for inclusion in the HIV prevention toolbox, it is important for national policy and decision makers to identify where PrEP may fit best within already established HIV prevention programming (and budgets) and the potential implications of introducing such policy changes. In particular, decision makers need information translating the trial results into potential population-level impact and cost-effectiveness to ensure that any additional investment will have the maximum possible effect on the epidemic.

Economic and mathematical models provide a framework to integrate information on efficacy, effectiveness, costs, and patient outcomes to support decision making and resource allocation [15]. However, due to their complexity, dependence on assumptions made, and inherent uncertainties, generalising results from these models can be difficult. In this review, we aim to assess published cost-effectiveness models that have evaluated the expected health gains and costs of PrEP interventions. Specifically, our objectives are: (1) to describe modelling approaches of cost-effectiveness analyses of PrEP; (2) to compare the effects of epidemiological and cost assumptions on cost-effectiveness results; and (3) to explore the potential impact on cost-effectiveness estimates of five issues raised by policy makers [16]–[18] when considering PrEP implementation: prioritisation, adherence, behaviour change, toxicity, and resistance.

Methods

We performed a systematic review of the published literature following the protocol available in Text S2 and adhering to the PRISMA guidelines for reporting of systematic reviews (Text S1: PRISMA checklist) [19] and guidelines for appraisal of economic evaluations [20].

Results

We screened 961 titles and abstracts retrieved from 14 electronic databases. After performing web searches and consulting experts in the field, 36 full text articles were evaluated. We also reviewed the reference lists and citations of these articles. Of these 36, 13 studies were included in the review [25]–[37]: 11 peer-reviewed publications and two peer-reviewed conference abstracts (Figure 1). Articles excluded are listed in Table S1 and a summary of conclusions of the articles included are presented in Table S2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow diagram of study selection.

Region-specific databases can be accessed as follows: African Index Medicus, http://indexmedicus.afro.who.int/; Eastern Mediterranean Literature, http://www.emro.who.int/; Index Medicus for South-East Asia Region, http://www.hellis.org/; LILACS, Latin America, http://www.bireme.br/iah2/homepagei.htm.

https://doi.org/10.1371/journal.pmed.1001401.g001

We present in Tables 1 to 4 the data extracted from the studies reviewed by study design, description of alternative programmes compared, impact, and cost assumptions. All studies were published between 2007 and 2013 and modelled the impact and cost, from a health care provider perspective, of PrEP scale-up in diverse settings. These settings included: heterosexual transmission in generalised epidemics in sub-Saharan Africa—the Southern Africa region [25], South Africa [28],[30],[31],[32],[36],[37]), and other modes of transmission in concentrated epidemics—among people who inject drugs (PWID) in Ukraine [33]; and men who have sex with men (MSM) in the USA [26],[29],[34],[35] and in Peru [27]. Timeframes varied from 5 to 20 y. All studies focused the models on high prevalence/incidence populations (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Study design.

https://doi.org/10.1371/journal.pmed.1001401.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Alternative programmes compared.

https://doi.org/10.1371/journal.pmed.1001401.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Impact: assumptions.

https://doi.org/10.1371/journal.pmed.1001401.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Cost: assumptions (US$ in publication).

https://doi.org/10.1371/journal.pmed.1001401.t004

In all models, the comparison scenario did not include PrEP and assumptions varied regarding current treatment scale-up: from no ART programme included [25],[26],[32] to ART coverage remaining stable at a current level [27],[29],[31],[34],[35] or an ongoing ART programme coverage expansion [28],[30],[33],[36],[37]. While most of the studies looked at systemic PrEP (daily oral dosing), two studies in South Africa looked at vaginal gels [31],[32]. Coverage assumptions were stated as scenarios. The criteria used to characterise priority populations varied among the studies, including high risk of acquisition (defined by sexual activity, condom use, or HIV incidence) [25]–[27],[35],[37], age [25],[30],[31],[34],[36], and timing of PrEP use [28] in relation to users' life events (Table 2).

All models were transmission models, except for two Markov simulations [31],[34]. Efficacy and effectiveness estimates varied from estimated ranges that were assumed prior to the results from clinical trials and had wide ranges (from 10% to 90%) [25],[26],[30],[33],[34] to estimates available directly from clinical trials [27]–[29],[31],[32],[35]–[37]. Several authors interpreted effectiveness as being dependent on the product efficacy and the individual-level adherence, specifically modelling this interaction [25]–[29],[32],[36]. Adherence assumptions varied from random [25] to profiles based on observations from published trials [27]–[29],[32]. Potential behaviour change following the introduction of a PrEP programme was included in the models as an increase in the number of sexual partners [25],[26], changes in condom use [27],[30], or both [35]. Drug resistance associated with PrEP use was explicitly modelled in one study [25], while in two further studies it was represented as a decrease in the rate of virologic suppression while on subsequent treatment [31],[34]. The studies did not assume any reduction in the quality of life or disability weights due to PrEP use, with the exception of three studies where toxicity to PrEP was addressed through a reduction in the quality of life and/or an excess fatality rate among PrEP users (Table 3) [31],[34],[35].

The majority of studies presented costs for PrEP programmes including drugs and monitoring costs, except for two studies that included drug costs only [25],[32]. Costs above service level were only included in two studies, as overheads [31] or a mark-up percentage of 5% [27]. Overall PrEP programme costs were consistent among studies by setting and ranged from high in the USA (between US$8,000 and US$12,000 per person-year) to low in South Africa (between US$80 to US$250). All cost estimates were driven by the cost of drugs. Three studies in the USA [26],[29],[35] and six in South Africa [28],[30]–[32],[36],[37] included averted ART costs. Ranges of estimated cost of ART were consistent among studies and context-specific (<US$1,000 per person-year in South Africa to >US$15,000 per person-year in the USA) (Table 4).

We present all cost-effectiveness estimates in Table 5 by epidemiological context and scenario modelled.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Cost-effectiveness estimates by scenario.

https://doi.org/10.1371/journal.pmed.1001401.t005

Discussion

This systematic review included 13 modelling studies estimating the cost and potential population-level impact of introducing a PrEP programme in generalised and concentrated epidemic settings. Our findings show that PrEP is estimated to have the potential to be a cost-effective addition to HIV prevention programmes in some settings. However, the cost-effectiveness of PrEP is likely to depend on considerations such as cost, the epidemic context, PrEP programme coverage and prioritisation strategies, as well as individual adherence levels and PrEP efficacy estimates.

To prevent heterosexual transmission in the generalised epidemics of southern Africa, PrEP is potentially a cost-effective intervention. Topical PrEP, in particular, could have a significant impact in South Africa, providing a much-needed female-initiated prevention option. However, it should be noted that funding PrEP while other potentially more cost-effective HIV prevention interventions remain under-funded may have high opportunity costs. In concentrated epidemics, such as MSM-driven epidemics both in Peru and in the USA, PrEP could have a substantial impact on the epidemic but may not be affordable at current drug prices. In Ukraine, expansion of ART coverage and methadone maintenance treatment programmes for PWID should be a first priority, with PrEP potentially added on within a combination prevention framework. However, evidence to date shows PrEP might not be cost-effective in this setting at current drug prices. Nevertheless, findings from the phase III Bangkok Tenofovir Study of PrEP among PWID will shed light on the efficacy estimates of PrEP in this population and inform future model estimates in similar epidemic contexts [38].

In all settings, the price of drugs is a limiting factor in terms of affordability of PrEP programmes as has been previously suggested [39],[40], and is key to determining cost-effectiveness. Moreover, the findings above predominately exclude important service and above service costs of providing PrEP (i.e., regular HIV testing and blood chemistry panels; the costs of possible adverse outcomes, including PrEP-related toxicity and potential drug resistance attributable to PrEP; and system-wide costs of implementing a PrEP programme). All of these should be considered to improve the validity and utility of estimates. Another key limitation among the studies is that the majority did not include savings in treatment and hospitalisation due to secondary infections averted. Although carrying significant uncertainties, the inclusion of these benefits allows a more informed consideration of potential PrEP benefits within broader programmatic planning for HIV prevention and care.

In the models reviewed, several prioritisation strategies were explored. Prioritisation by sexual activity characteristics to deliver PrEP to those populations at highest risk of HIV exposure improved the cost-effectiveness estimates. However, the extent to which prioritising populations at higher risk improves cost-effectiveness results in the models depends largely on the assumptions made about sexual mixing and risk heterogeneity. Extra costs related to the identification and engagement of priority populations were not included in any of the studies, neither were considerations in terms of economies of scale. Furthermore, as results from the enrolment phase of iPrEX Ole show (65% of trial participants decided to continue taking PrEP), not all individuals at higher risk are willing to use PrEP [41]. Identifying and meaningfully engaging those at highest risk in tailored HIV prevention strategies represents a significant challenge for decision makers, health care providers, and prevention advocates.

Prioritisation by age was a strategy advanced in several studies. In the studies reviewed, prioritisation by age group resulted in a lower cost-effectiveness benefit compared to prioritisation strategies based on self-reported risk behaviour [25],[30],[31],[34]. However, the former has the advantage of being straightforward to implement compared to a selection of potential PrEP users based on self-reported risk behaviour. In contexts such as South Africa and the USA, age prioritisation clearly would focus on those populations at higher risk of HIV acquisition. Another prioritisation strategy analysed in one study was the delivery of PrEP depending on the stage in users' lives. In reality, PrEP use will not need to be sustained throughout an individual's lifetime but may vary as his or her risk situation changes over time. People may opt to use PrEP during specific higher risk life periods, such as during periods of active sex work or when serodiscordant couples are trying to conceive a child [42],[43]. Understanding potential scenarios of PrEP use over the life cycle is essential for decision makers to be able to evaluate the possible impact of PrEP programmes in their local contexts. An additional consideration concerns intermittent PrEP. The first report of safety and adherence to an intermittent PrEP regimen in Kenya showed that among MSM and female sex workers adherence was lower than for daily dosing [44]. Results from two phase II trials underway in France [45] and the USA [46] will help inform adherence requirements and, should intermittent pre- and post-exposure dosing be proven effective, help tailor PrEP programmes to consumer demand [47].

Behavioural change due to PrEP use and adherence to PrEP were estimated to have potentially significant impacts on programme effectiveness. While the emergence of drug resistance due to PrEP programme scale-up and PrEP-related toxicity assumptions did not significantly affect cost-effectiveness estimates, improvement of drug resistance surveillance systems as well as effective adherence counselling will be essential components of PrEP programme implementation, in addition to behavioural counselling.

This review has several limitations. The geographical coverage of the studies reviewed is partial and both the impact and cost evaluations are highly setting-specific, limiting the generalisability of the findings. We were unable to perform a meta-analysis due to the variability across studies in reporting outcomes. Nevertheless, in order to compare cost-effectiveness estimates across settings, we used the thresholds proposed by the WHO-CHOICE Project and the Commission on Macroeconomics as a benchmark [22]. These standards are based on the GDP per capita, assuming that a society is willing to pay the equivalent of up to one GDP per capita (for highly cost-effective interventions) or between one and three times the GDP per capita (for a cost-effective intervention for a DALY averted, QALY saved, or LYS). This is a normative selection of cost-effectiveness thresholds, albeit regarded as useful from a decision analytic perspective [24].

It is worth noting that, with the exception of four studies in South Africa [28],[30],[36],[37], research comparing the potential trade-offs of earlier treatment for prevention versus PrEP remains an important gap in the literature that should be addressed, especially in concentrated epidemics. Cost-effectiveness studies that demonstrate where resources applied can have the greatest impact will help inform this complicated decision-making, but these are not the only considerations. The decision to include a PrEP option within the combination prevention package requires input from all strata of society. For instance, in contexts where universal access to ART for patients in need has not been achieved, PrEP programme planning processes will be challenged by concerns about social justice, equity, and affordability. This is in addition to the hurdles of overcoming the marginalisation, stigmatisation, and criminalisation of many of the populations that would most benefit from tailored HIV prevention programming that includes the choice of PrEP. Disentangling these issues will be critical for effective decision-making, as will the consideration of potential synergies between an expanded testing and treatment programme and a PrEP programme.

While the interest of donors for modelling studies that compare the cost-effectiveness of different HIV prevention methods is expected to increase [48], current evidence is already available to aid policy makers in assessing PrEP as a new prevention option. In this context, our review sheds light on the main considerations that decision makers need to address when judging the relevance of cost-effectiveness estimates of a potential PrEP programme and the potential gaps in the modelling evidence. Given that our review shows that setting and target population are critical drivers of cost-effectiveness, the next step is to conduct context-specific demonstration studies, including comprehensive cost analyses, of different prioritisation and adherence promotion strategies to ensure that the maximum benefit from the introduction of PrEP is realised within combination HIV prevention programmes.

Supporting Information

Acknowledgments

The authors would like to thank Sabina Alistar for her assistance with providing additional information on her studies.