ABSTRACT
Objective To assess the population impact of long-acting injectable pre-exposure prophylaxis (LAI-PrEP) when available concurrently with daily oral (DO) PrEP among men who have sex with men (MSM).
Design Dynamic network model of HIV transmission among MSM.
Methods We calibrated our model to the current uptake of DO-PrEP (15% among indicated) and HIV prevalence (25.3%) among MSM in the southeastern US. Parameters for LAI-PrEP pharmacokinetics and efficacy were estimated from Phase IIa safety trials in human and macaque models. Primary analyses explored varied conditions of PrEP uptake and proportion choosing LAI-PrEP given initiation. Secondary analyses evaluated uncertainty in pharmacokinetic efficacy and in persistence to LAI-PrEP relative to DO-PrEP.
Results If 50% of MSM currently accessing DO-PrEP choose LAI-PrEP when it becomes available, an additional 14.3% of infections could be averted over 10 years. If the availability of LAI-PrEP increased overall uptake of PrEP to 50%, the expected infections averted would increase to 23.9%. The benefit of LAI-PrEP was weakly sensitive to pharmacokinetics with 9.9% of infections averted even if LAI-PrEP efficacy were half of current empirical estimates. The superiority of LAI-PrEP was strongly attributable to better adherence and persistence with only 4.3% of infections averted if persistence were the same as DO-PrEP.
Conclusions LAI-PrEP availability is projected to decrease HIV incidence. However, increased benefits depend upon improvements to adherence and persistence above current DO-PrEP levels. As LAI-PrEP becomes available, uptake and persistence should be monitored to ensure optimal implementation.
INTRODUCTION
Men who have sex with men (MSM) remain at increased risk for HIV infection in the United States, accounting two-thirds of all new infections in 2016 [1]. The introduction of pre-exposure prophylaxis (PrEP), currently formulated as a once-daily oral tablet of emtricitabine and tenofovir (FTC/TDF or FTC/TAF), has allowed some MSM to effectively reduce their risk of infection [2, 3]. However, implementation challenges undermine the potential for PrEP to reduce incidence at the population level. For example, uptake of PrEP remains low [4-7]. The US Centers for Disease Control and Prevention (CDC) estimate that 1.1 million US adults were behaviorally indicated to use PrEP in 2017 [8], but only 100,000 accessed PrEP [4]. Efforts to increase PrEP use may be undermined by poor persistence, resulting in early discontinuation among individuals with ongoing HIV risk and low clinical retention [9-12]. Previous PrEP models have predicted that current coverage levels are insufficient to reach many HIV prevention targets [13, 14]. Among MSM accessing PrEP, inadequate adherence further reduces its population-level prevention benefits [10, 15].
Alternative PrEP products that are more acceptable and easier to use may address these implementation challenges. There is precedent from other medications, such as contraceptives, that increased choice results in greater uptake and more effective use [16]. Daily oral PrEP (DO-PrEP) is currently the only dosing strategy recommended by the CDC and the US Food and Drug Administration [17]. However, alternative dosing regimens [18, 19] have proven efficacious; novel formulations are in development, including long-acting injectable [20] and topical formulations [21], implants, and other devices [22].
Cabotegravir is a strand-transfer integrase inhibitor currently in Phase III clinical trials for PrEP [20]. When formulated as a long-acting injectable suspension, cabotegravir is maintained with periodic dosing at 8-week intervals [20]. Long-acting injectable PrEP (LAI-PrEP) would address adherence challenges associated with DO-PrEP because the burden of taking a pill would be eliminated [15, 20]. Persistence may also increase with LAI-PrEP if the 8-week interval between dosages is easier for individuals to manage and maintain (compared to daily dosage of DO-PrEP). LAI-PrEP may be more acceptable for many unwilling or unable to take a daily tablet [23-27]. Estimates of preference for LAI-PrEP range from 30% to 67% [23, 25, 27-30], but real-world selection of LAI-PrEP over DO-PrEP is unknown. Translating LAI-PrEP potential to real-world impact will ultimately depend on unknown uptake and persistence levels.
Because clinical trials of LAI-PrEP will only establish non-inferiority to DO-PrEP at the individual level [31], modeling studies are an important tool to understand the mechanisms by which LAI-PrEP may differ from DO-PrEP at the population level. A recent modeling study [32] simulated the population impact of single formulations of PrEP by themselves. If LAI-PrEP was allocated to 35% of MSM, HIV incidence would decrease by 44%; if DO-PrEP was used instead of LAI-PrEP, incidence would decrease by 33%. Based on this, LAI-PrEP could improve upon the population benefits of DO-PrEP, but the scenario of only one formulation available at a time is unrealistic as having an option for both formulations is expected in practice. It was also unclear from the model how realistic challenges to PrEP persistence will limit the population impact of LAI-PrEP which, unlike DO-PrEP, provides continued but waning protection even after discontinuation [20]. Since publication of this model, pharmacokinetic data from a Phase II clinical trial of LAI-PrEP has become available [20], which can be used to specify model parameters.
To estimate the potential population-level impact of LAI-PrEP, we used a sexual network model of HIV transmission dynamics among MSM in the southeastern US. To understand how concurrent availability of LAI-PrEP and DO-PrEP could maximize HIV prevention, we simulated scenarios varying PrEP uptake and the proportion choosing LAI-PrEP compared to DO-PrEP. Second, we conducted sensitivity analyses of the LAI-PrEP pharmacokinetic and efficacy parameters to understand how uncertainty in these parameters could impact projections of impact. Finally, to investigate the mechanisms by which LAI-PrEP could provide additional HIV prevention benefits over an equivalent uptake level of DO-PrEP, we explored varying levels of persistence among LAI-PrEP users.
METHODS
We used a m,athematical model of HIV transmission dynamics among MSM in the Atlanta area aged 18–40 years over a 10-year period. The model was simulated using EpiModel [33], a software platform for simulating epidemics in sexual networks using temporal exponential random graph models [34]. Our prior modeling work with this platform included a study of racial disparities in the PrEP continuum of care, including initiation, adherence, and persistence [35]. In the present study, we build on this work by developing the model framework and parameterization of LAI-PrEP and by incorporating the PrEP continuum of care to explore differences between LAI-PrEP and DO-PrEP. Full methodological details can be found in the Web Appendix.
Sexual Network Model of MSM
The model simulates one-off, casual, and main partnerships. Partnership formation was determined stochastically by partnership type, age mixing, number of existing partnerships, and sorting by receptive and insertive positioning. Relational dissolution was modeled with a constant hazard based on the median duration for partnership type.
Behavior within sexual partnerships was determined stochastically. The probability of HIV transmission was modified based on condom use [36], sexual positioning [37], circumcision status of the insertive partner [38], and non-HIV sexually transmitted infections [39]. For MSM with HIV, rates were assigned for diagnosis, initiation of antiretroviral therapy, and viral suppression [40, 41]. Antiretroviral therapy decreased HIV viral loads, with a corresponding decrease in the probability of HIV transmission [42, 43].
PrEP Uptake, Adherence, and Persistence
Uptake of PrEP was determined using a two-step process. Parameters for LAI-PrEP and DO-PrEP are shown in Table 1. First, we specified the probability of initiating PrEP of either kind. Second, we specified the probability of choosing LAI-PrEP versus DO-PrEP. By simulating uptake using probabilities on two domains, we were able to explore counterfactual scenarios in which the decisions interact. Eligibility to initiate PrEP was determined based on CDC guidelines for DO-PrEP indications [13, 17]. For MSM using DO-PrEP, we assigned a probability distribution for medication adherence corresponding to low (< 2 doses/week, 8.9%), medium (2–3 doses/week, 12.7%), and high adherence (≥ 4 doses/week, 78.4%) [44], which modified the level of protection against HIV acquisition (hazard ratios: 0.69, 0.19, and 0.02, respectively) [15].
Pre-exposure prophylaxis parameters in all models and values varied in experimental scenarios.
Based on CDC guidelines, behavioral indications were reassessed annually; if indications had lapsed, PrEP would be discontinued for both DO-PrEP and LAI-PrEP [17]. We also allowed for spontaneous discontinuation of DO-PrEP to occur based on an observational study showing 57% of PrEP patients remained in care 6 months after initiation. We translated this into a constant probability of discontinuation assuming a geometric distribution to calculate the median time to event (224 days) [45]. It is unknown whether MSM using LAI-PrEP would have the same rate of spontaneous discontinuation; the clinical burden is different than DO-PrEP, and data from clinical trials are insufficient to predict persistence in the real-world. Therefore, in the base model, we assumed that 21.5% of men would have poor LAI-PrEP persistence (hereafter referred to as the “low-persistent” group), matching the total proportion of DO-PrEP users with low or medium adherence. The rate of discontinuation among the low-persistent matched the spontaneous discontinuation rate among those using DO-PrEP, while the remainder of MSM using LAI-PrEP were not eligible for spontaneous discontinuation. We varied the proportion low-persistent and the rate of discontinuation in counterfactual models.
Injectable PrEP Pharmacokinetics and Efficacy
Based on studies of nonhuman primates, the target cabotegravir plasma concentration was set to 4-times the protein-adjusted 90% inhibitory concentration (PA-IC90) [46-48]. A Phase IIa clinical trial showed that 600 mg intramuscular injections at 8-week intervals were sufficient to maintain the target plasma concentration [20]. Prior to the maintenance phase, patients first complete a 4-week drug safety phase of oral cabotegravir. Next, two 600 mg injections are administered with a 4-week interval, which is necessary to initially achieve the target peak plasma concentration. For simplicity, we assumed that men would achieve the target peak plasma concentration following the first injection and without an initial oral safety phase.
Using data from the Phase IIa trial to model the pharmacokinetics of injectable cabotegravir [20], we estimated the median peak plasma concentration (3.59 µg/ml) following each injection after and including the second injection, as well as the rate of decay in longitudinal follow-up. We then created a continuous half-life function (35 days) of drug elimination based on these two parameters. Our model shows that plasma concentrations above PA-IC90 and 4-times PA-IC90 are maintained for 22 and 12 weeks, respectively, with limited drug still present 48 weeks after the final injection. Except where specified otherwise, we held constant the base model parameters for LAI-PrEP for each of the experimental models.
A Phase III study to determine the efficacy of cabotegravir as LAI-PrEP is ongoing. Thus, we used the results of a macaque study to model the relative reduction in the per-exposure probability of HIV infection [46]. We used logistic regression to estimate the probability of infection given the plasma concentration of cabotegravir at the time of exposure. We then transformed the results to the human equivalent of the PA-IC90 to obtain a single parameter for the relative reduction in infection probability given the plasma concentration of cabotegravir. Our model predicts that LAI-PrEP is more than 99% efficacious at 4-times PA-IC90 (HR = 0.002).
Base Model and Experimental Scenarios
The base model represents current uptake of DO-PrEP, in which approximately 15% of indicated MSM in the southeastern US use PrEP [49, 50]. We calibrated the model to maintain a steady state prevalence of DO-PrEP use among indicated MSM. To reach 15%, the probability of initiating PrEP given indications was set to 21%, while LAI-PrEP selection was set to 0%.
We examined a range of experimental scenarios. First, in our primary analysis, we explored scenarios of overall PrEP uptake by simultaneously varying the probability of initiating PrEP and the probability of selecting LAI-PrEP, each from 0% to 100%. The probability of selecting LAI-PrEP was set to 50% for all secondary analyses, while the probability of initiating PrEP was maintained at the base model value (21%). Second, to better understand the pharmacological properties of LAI-PrEP, we varied the pharmacokinetic and efficacy parameters. We varied the LAI-PrEP pharmacokinetics by simultaneously modifying the peak plasma concentration after injection (1.5 to 6.5 µg/ml) and the half-life of drug elimination (15 to 60 days). Separately, we varied the efficacy of LAI-PrEP by scaling the per-exposure hazard ratio for transmission (5%, 10%, 25%, and 50% increase or decrease in efficacy compared to the base parameter). Finally, to understand how the effectiveness of LAI-PrEP could depend on longitudinal use, we varied the proportion of MSM using LAI-PrEP that are considered low-persistent and the rate of spontaneous discontinuation among the low-persistent. To do this, we modified the median time using LAI-PrEP by 50% (112 days), 100% (224 days), 150% (336 days), and 200% (448 days) and calculated new spontaneous discontinuation rates. We simultaneously varied the proportion considered low-persistent (21.5%, 25%, 50%, 75%, and 100%).
Calibration, Simulation, and Analysis
The network was initialized with 10,000 MSM and calibrated to the observed HIV prevalence (25.3%) at baseline in a cohort of MSM in Atlanta, Georgia [51]. Each model scenario was simulated over 10 years and for 250 times. The outcomes reported were HIV prevalence and incidence per 100 person-years at risk at the end of 10 years, and the percent of infections averted (PIA). The PIA was calculated by comparing the cumulative incidence in each experimental scenario to the base model. We summarized the results using the median values of all simulations and 95% simulation intervals (SI).
RESULTS
In the base model scenario, 15.1% of indicated MSM are using DO-PrEP (0% LAI-PrEP). At the end of the 10-year simulation, the HIV incidence was 2.77 (95% SI: 2.27, 3.23) per 100 person-years and prevalence was 21.8% (95% SI: 20.4, 23.4). Figure 1 shows that increasing both the probability of initiating PrEP and the probability of selecting LAI-PrEP would result in fewer infections over 10 years, compared to the base model scenario. The benefit of introducing LAI-PrEP is maximized in scenarios where both probabilities increase. Substantial improvements, however, would still be realized at lower proportions of MSM initiating PrEP and selecting LAI-PrEP.
Percent of infections averted (PIA) over 10 years, compared to the base model scenario of 21% probability of initiating pre-exposure prophylaxis (PrEP) among behaviorally indicated MSM and 0% choosing long-acting injectable pre-exposure prophylaxis (LAI-PrEP), across varying PrEP-initiation and formulation-selection probabilities (LAI-PrEP versus daily-ora PrEP). Dashed line indicates 21% probability of initiating PrEP (0% LAI-PrEP) used in the base model.
Table 2 shows that increasing both the initiation and the LAI-PrEP selection probabilities will increase the proportion of indicated MSM using PrEP of any kind, which in turn causes HIV incidence to decline. While keeping the model limited to DO-PrEP, increasing the probability of initiating PrEP from 21% to 50% resulted in twice as many indicated MSM using DO-PrEP (15.1% to 30%, respectively) and 13.8 PIA (95% SI: 3.3, 22.0). Alternatively, total PrEP use increased to 30.5% when the probability of selecting LAI-PrEP was increased to 50%, even though the probability of initiating PrEP was held constant at 21%, causing 14.3 PIA (95% SI: 5.4, 22.6). Due to similar overall PrEP use, these two scenarios yielded comparable values for PIA despite different proportions using LAI-PrEP. If 100% of the MSM accessing PrEP were to choose LAI-PrEP and initiation remained at 21%, PrEP use would further increase to 41.0% among indicated MSM with 23.9 PIA (95% SI: 15.3, 31.0). Finally, if the probability of initiating PrEP and the proportion choosing LAI-PrEP were each set to 50%, then the population impact would increase to 34.3 PIA (95% SI: 27.5, 41.2).
Proportion of indicated MSM using daily-oral or long-acting injectable pre-exposure prophylaxis, HIV prevalence, HIV incidence per 100 person-years at risk, and percent of infections averted over 10 years compared to the base model scenario, across varying PrEP-initiation and formulation-selection probabilities.
Figure 2 shows that the underlying estimates for LAI-PrEP pharmacokinetics are robust to uncertain parameter values. If the peak plasma concentration were reduced to 2 µg/ml and the half-life were reduced to 15 days (an extreme scenario, well below the a priori clinical trial targets), the impact on the population benefit of LAI-PrEP would be small. If the parameters are modified within a smaller range around our estimates, the impact is negligible. Similarly, Table 3 shows that the population impact of LAI-PrEP would decrease only modestly if our estimate of the drug efficacy were overestimated. For example, a 50% decrease in LAI-PrEP efficacy still prevented 9.9% (95% SI: 0.2, 19.0) of infections, compared to 14.3 PIA (95% SI: 5.4, 22.6) if the base model parameter for efficacy is used.
HIV prevalence, HIV incidence (per 100 person-years at risk), and percent of infections averted over 10 years compared to the base model scenario, across varying relative changes in long-acting injectable pre-exposure prophylaxis efficacy.
Percent of infections averted (PIA) over 10 years, compared to the base model scenario of 21% probability of initiating pre-exposure prophylaxis (PrEP) among behaviorally indicated MSM and 0% choosing long-acting injectable pre-exposure prophylaxis (LAI-PrEP), after LAI-PrEP is introduced (50% LAI-PrEP) across varying parameters for LAI-PrEP pharmacokinetics. Dashed lines indicate peak plasma concentration (3.59 µg/ml) and half-life (5 weeks) used in the base model.
Table 4 shows the effect of varying longitudinal persistence for LAI-PrEP. As persistence decreases, the prevalence of LAI-PrEP use among indicated MSM also decreases. When the percent low-persistent increased from 21% to 100%, the prevalence of LAI-PrEP use decreased from 24.1% to 8.2%. This caused the PIA to also decline from 14.3% (95% SI: 5.4, 22.6) to 4.3% (95% SI: -4.8, 14.4). Modifying the rate of discontinuation among the low-persistent had a modest impact on the overall prevalence of LAI-PrEP if the percent low-persisent was maintained at the base level (21.5%). If the median time to discontinuation was reduced by half (from 224 days to 112 days) the prevalence of total LAI-PrEP use would only decrease to 23.4% (from 24.1%). However, if the percent low-persistent also increased to 100%, the decreased time to discontinuation would have a greater effect, with only 5.1% of indicated MSM using LAI-PrEP and 2.1 PIA (95% SI: -8.2, 11.2). In scenarios with 100% of MSM using LAI-PrEP considered low-persistent, doubling the median time to discontinuation (from 224 days to 448 days) was insufficient to compensate for the increased proportion experiencing spontaneous discontinuation.
Proportion of indicated MSM using daily-oral pre-exposure prophylaxis or long-acting injectable pre-exposure prophylaxis, HIV prevalence, HIV incidence per 100 person-years at risk, and percent of infections averted over 10 years, compared to the base model scenario, across varying spontaneous discontinuation rates and proportion of MSM using LAI-PrEP considered low-persistent.
DISCUSSION
In this study, we demonstrated that the introduction of LAI-PrEP as a formulation concurrently available to DO-PrEP could substantially reduce HIV incidence among MSM compared to offering only DO-PrEP. The largest improvements to HIV prevention would occur when the availability of LAI-PrEP coincides with overall increased PrEP initiation. However, LAI-PrEP could provide important gains in HIV prevention, even if overall PrEP uptake remains at current levels. This is primarily due to improvements in adherence and persistence expected with LAI-PrEP. To our knowledge, this is the first modeling study to predict the population impact of LAI-PrEP when available concurrently with DO-PrEP.
Low uptake has been a limiting factor to the population benefit of DO-PrEP since it first became available [4-6]. We designed our model to explore PrEP uptake on two domains, overall uptake of PrEP and proportion using LAI-PrEP, in order to represent a range of possible scenarios. Similar to other studies [13, 14], we found that large scale reductions in HIV incidence will require greater uptake of PrEP among MSM than has currently been observed in the US [4-6]. Our model shows that this can be achieved if the availability of LAI-PrEP increases the proportion of at-risk individuals accessing PrEP; and that further reductions in HIV incidence could occur if those currently accessing DO-PrEP transition to LAI-PrEP.
The benefit of LAI-PrEP compared to DO-PrEP at fixed levels of PrEP uptake is contingent upon increased persistence which results in longer sustained person-time using PrEP in our model. Partial protection provided during the long metabolic half-life after discontinuation further increases the person-time of protection in the population. Our finding is in direct contrast to the earlier LAI-PrEP modeling study, which reported greater population impact of LAI-PrEP as persistence decreased [32]. In their model, PrEP was allocated to maintain a constant prevalence of active users in the population, so that a new individual was forced to replace each discontinuation. As a result, the total person-time of LAI-PrEP would counterintuitively increase in scenarios with lower persistence, because the number of individuals with partial protection increased. We found this model assumption unrealistic, and therefore its population impact implausible. In our model, the two components of PrEP prevalence (initiation and persistence) were disentangled in order to evaluate the individual contribution of each factor. We were therefore able to vary LAI-PrEP persistence without artificially inflating the person-time of LAI-PrEP use, thereby isolating the effects of initiation and persistence. Our model shows that, in order to achieve sufficient coverage of PrEP use in the population, persistence must be optimized for DO-PrEP and LAI-PrEP.
Our model represents an optimistic scenario of increased persistence among those using LAI-PrEP compared to DO-PrEP, but persistence will likely remain a challenge for all persons accessing PrEP. The reasons for discontinuation of DO-PrEP in the real-world include decreased HIV risk perception, non-adherence to the clinical protocol for PrEP, insurance barriers and experiencing side-effects [9, 52]. We assumed that persistence would improve with LAI-PrEP, in part because the 8-week intervals between dosing would encompass periods of temporary decreased risk in which some individuals would otherwise discontinue DO-PrEP. We also assumed that the lack of daily adherence would decrease fatigue, making it more likely that individuals would be motivated to maintain the protocol. This assumption may be unrealistic since more frequent clinical care visits (i.e., six annual visits for LAI-PrEP compared to four for DO-PrEP) may be more burdensome for some. However, the differing clinical procedure of LAI-PrEP may attract individuals whom are not willing to use DO-PrEP. Implementation strategies for LAI-PrEP should anticipate challenges to persistence and include support programs to help patients evaluate risk and ensure longitudinal persistence when appropriate.
Limitations
Our model has a number of important limitations. The first is uncertainty of the LAI-PrEP parameters, since Phase III clinical trials are ongoing. Our model parameters may be misspecified which would cause our primary results to be biased. However, we tested a range of plausible parameters for LAI-PrEP pharmacokinetics and efficacy, which had only a small effect on the model results. Our model is also a simplification of the protocol for LAI-PrEP used in the ongoing Phase III clinical trial. Specifically, we did not include the 4-week oral dosing phase of cabotegravir that is used to ensure safety. We also assumed that the peak plasma concentration for cabotegravir would be achieved after a single intramuscular dose, instead of the two initial doses at a 4-week interval. Similarly, we did not model an oral ramp-down phase for individuals discontinuing LAI-PrEP under the supervision of their clinician (opposed to spontaneous discontinuation). This later phase of LAI-PrEP is used to prevent infection during the waning tail of LAI-PrEP protection. Increased complexity and clinical burden of the protocol will likely result in lower persistence, decreasing the person-time of LAI-PrEP coverage in the population. Finally, we did not model resistance to cabotegravir among individuals that acquire HIV infection following discontinuation of LAI-PrEP, while the drug remains present but provides only partial protection. This phenomenon presents a new challenge compared to DO-PrEP, which rapidly metabolizes after discontinuation. However, resistance was estimated to be rare in the earlier model of cabotegravir-based LAI-PrEP [32].
Conclusions
LAI-PrEP has the potential to further transform the HIV prevention paradigm, building on the progress since DO-PrEP became available. The superiority of LAI-PrEP in our model is tied to improvements in adherence and persistence rather than pharmacological efficacy. The maximal benefit of LAI-PrEP will be realized only if uptake is robust and individuals are persistent following initiation. These are not trivial challenges, but lessons learned from the implementation of DO-PrEP can inform strong public health programs to promote and support the use of LAI-PrEP among indicated persons. As LAI-PrEP becomes available, uptake and persistence should be closely monitored along with DO-PrEP to identify opportunities to optimize implementation of each.
ACKNOWLEDGEMENTS
This work was supported by National Institutes of Health grants R21 MH112449, R01 AI138783, and P30 AI050409 (Emory University Center for AIDS Research).
REFERENCES
Subject Area
Reviews and Context
0
Comment
0
TRIP Peer Reviews
0
Community Reviews
0
Automated Services
0
Blogs/Media
Author Videos