ABSTRACT
Background The goal of the US Ending the HIV Epidemic (EHE) plan is to reduce HIV incidence by 75% by 2025 and 90% by 2030. This initiative will scale-up prevention and care activities in high-need areas like the Southeast United States. It is unknown what interventions, alone or in combination, will have the greatest impact towards reaching the EHE targets.
Methods We developed a stochastic HIV transmission model for men who have sex with men (MSM), calibrated to local surveillance estimates of HIV prevalence, PrEP utilization, and HIV care continuum levels in the Atlanta area. Model scenarios varied HIV screening rates relative to empirical levels, under two assumptions of how HIV-negative persons would be linked to PrEP initiation, and also considered relative improvements to HIV care linkage and care retention for those screening positive. We estimated both the proportion of cumulative infections averted over 10 years and the reduction in incidence rates relative to EHE targets.
Results A ten-fold relative increase in HIV screening rates (to approximately biannual screening for black and Hispanic MSM and quarterly for white MSM) would lead to 41% of infections averted under the assumption of integrated linkage to PrEP, with prevention through both increased PrEP coverage and increased HIV viral suppression. At the same relative increase in screening but under the assumption of no PrEP linkage, only 9.9% of infections would be averted, with prevention only through increased viral suppression. Improvements to HIV care retention would avert 33.5% of infections if retention rates were improved 10-fold. If both screening and retention were jointly improved 10-fold, 66.6% and 48.5% of cumulative infections would be averted under assumptions of PrEP linkage and no-linkage, respectively. Under the joint 10-fold PrEP-linked scenario, it would take 7.3 years to meet the 75% EHE target and 30.4 years to meet the 90% target for MSM in Atlanta.
Conclusions Interventions to improve HIV screening linked with PrEP for those screening negative, and HIV care retention would have a substantial impact on HIV prevention through reduction of HIV acquisition and transmission rates. However, additional interventions beyond these substantial improvements to HIV screening, PrEP coverage, and HIV care retention will be necessary to reach the EHE target of a 90% reduction in incidence for Atlanta MSM by 2030.
INTRODUCTION
Despite the availability of highly effective HIV prevention tools such as preexposure prophylaxis (PrEP) and treatment as prevention (TasP), antiretroviral therapy (ART) coverage in high-risk populations remains low in many areas in the United States.1–4 The new Ending the HIV Epidemic (EHE) plan seeks to reduce HIV incidence nationally by 75% in 5 years (by 2025) and 90% in 10 years (by 2030) by targeting federal resources towards interventions that increase access and uptake of these prevention tools in high-need regions like the Southeast.5 However, it is unknown what combinations and what amounts of interventions will be needed to meet that goal.
HIV elimination depends on reducing HIV acquisition and transmission rates through achieving adequate levels of ART coverage at the population level.6 HIV care and prevention continuum frameworks have defined the intermediate steps that lead to goals of viral suppression for those living with HIV, and high levels of PrEP adherence and persistence for those uninfected but at risk.2,7,8 Mathematical models have then been used to estimate the relationship between continuum levels and future HIV prevention targets.9 Prior models evaluating the HIV care continuum for people with HIV in the United States (U.S.) have suggested that major improvements to HIV screening, care linkage, and care retention would be needed to achieve substantial reductions in HIV transmission.10–13 Models focused on HIV-uninfected populations have projected the impact of PrEP given low coverage and poor persistence.14–16 Despite this strong modeling research base, the potential for synergy between the HIV care and prevention continua has not been sufficiently evaluated. Both continua rely on HIV screening, a gateway to PrEP initiation for persons who test negative and ART initiation for persons who test positive.8
In this study, we developed a mathematical model of HIV with an integrated HIV prevention and care continuum framework in a target population of MSM in the Atlanta area. Our primary research question was what combinations of increases in HIV screening (alone or as a gateway to PrEP initiation), HIV care linkage, and HIV care retention could meet the EHE targets of 75% and 90% reduction in HIV incidence. These projections may inform HIV implementation science and public policy about what HIV prevention strategies should be prioritized in high-incidence, high-need settings like Atlanta that are the focus of the new EHE plan.
METHODS
Study Design
Our network-based mathematical model of HIV transmission dynamics for US MSM was built with the EpiModel software platform,17 software for simulating epidemics over dynamic contact networks under the statistical framework of temporal exponential random graph models (TERGMs).18 Building on our previous applications to estimate the impact of HIV PrEP for MSM,4,14 new model extensions for this study integrated novel network data streams and expanded the representation of the HIV care continuum. Full methodological details are provided in a Supplemental Appendix.
HIV Transmission and Progression
This model simulates the dynamics of main, casual, and one-time sexual partners for black, Hispanic, and white/other MSM, aged 15 to 65, in the Atlanta metropolitan area. The starting network size in the model simulations was 10,000 MSM, which could increase or decrease over time based on arrival (sexual debut) and departure (mortality or assumed sexual cessation at age 65). We used primary data from the ARTnet study to fit statistical models for summary statistics entered into the TERGMs. ARTnet was a web-based egocentric network study conducted in 2017–2019 of MSM in the US, with data from 4904 respondents reporting on 16198 partnerships with respect to partnership timing, overlap, and attributes.19 For this analysis, we included a main effect for geography of residence (city of Atlanta versus all other areas) in model predictions to represent our study target population. Parameters were weighted by census-based race/ethnicity and age distributions to account for selection biases.
For the TERGMs, predictors of partnership formation included partnership type, number of ongoing partnerships (network degree) across types, heterogeneity in degree and assortative mixing by race/ethnicity and age, and mixing by receptive versus insertive sexual position. Partnership durations were modeled for main and causal partnerships as a set of dissolution rates stratified by partnership type and homophily within age group. Other statistical models with ARTnet data were fit to predict frequency of acts within partnerships and the probability of condom use per act.
In the absence of ART, MSM progressed through HIV disease with HIV viral loads modeled as a continuous attribute that changed through the acute, chronic, and AIDS stages of disease. After infection, men could be screened for HIV and initiate ART, which would lower their HIV viral load (VL) and increase their longevity.20,21 Lower VL with sustained ART use was associated with a reduced rate of HIV transmission probability per act.22 Factors modifying the HIV acquisition probability per act included PrEP use,23 condom use,24 sexual position,25 and circumcision of the insertive partner.26
HIV Prevention and Care Continuum
We simulated an integrated HIV continuum of antiretroviral-based prevention and care, with HIV screening as the central gateway to either PrEP or ART.8 MSM engaged in HIV screening at regular intervals; we calibrated screening rates to reproduce 2017 estimates from the Georgia Department of Public Health (GDPH) surveillance report of the proportion of MSM with HIV who were diagnosed.27 These calibrated screening rates were externally validated by comparing simulations to 2017 estimates of the average time between infection and diagnosis.28 MSM screening positive could then enter the HIV care continuum (linkage and retention in ART care) while MSM who screened negative could enter the HIV prevention continuum (PrEP initiation, adherence, and persistence).
In the care continuum, MSM were linked to ART based on race-specific rates drawn from the literature as initial estimates10 and then calibrated to GDPH surveillance data on race-stratified proportions of MSM linked to care within 30 days of HIV diagnosis.27 This model assumes immediate initiation of ART at care linkage based on current HIV treatment guidelines.29 After linkage, MSM could cycle off and back on ART; halting rates were initially informed by the literature10 and then calibrated to GDPH surveillance data on race-specific proportions of MSM with VL suppression.27 ART reinitiation rates were kept fixed for reasons of model identifiability (reinitiation rates were perfectly negatively correlated with halting rates). MSM who remained on ART achieved suppression after 3 months of use.21 Upon stopping ART, VL would rebound to set-point VL.30
The HIV prevention continuum consisted of initiation, adherence, and persistence in PrEP care for daily oral TDF/FTC.4 HIV-negative MSM who met indications for PrEP based on CDC guidelines were eligible to start.31 Because MSM may enter PrEP care through regular HIV screening or spontaneously,1 we modeled the link between HIV screening and PrEP in these two ways to understand the relationship between HIV screening and PrEP coverage. In the first scenario, MSM were eligible to start PrEP if they both had indications and they had HIV screened negative that week. In the second scenario, MSM were eligible to start PrEP based on indications alone, so PrEP initiation was spontaneous and not linked to regular screening; however, confirmation of HIV-negative status was conducted upon PrEP initiation. In each scenario, eligible MSM then started PrEP based on an initiation probability (i.e., initiation was not deterministic), which led to an overall coverage level of 15%, consistent with estimates of MSM in the Atlanta area.1
After PrEP initiation, adherence was modeled as three levels, with 78.4% meeting the high-adherence level32 that resulted in a 99% relative reduction in HIV acquisition risk per sexual act.23 MSM with high PrEP adherence reduced their condom use based on ARTnet estimates (see Appendix Section 4.2). PrEP discontinuation rates were based on estimates of the proportion of MSM who were retained in PrEP care at 6 months (57%),33 transformed into median time to discontinuation (224 days) assuming a geometric distribution. In addition, MSM stopped PrEP if they no longer exhibited PrEP indications (evaluated annually for active PrEP users).31
Intervention Scenarios
In our primary intervention scenarios, we varied the rates of HIV screening, ART care linkage, and ART care retention from the base race/ethnicity-stratified rates to simulate interventions for a 10-year period (2019 to 2029). For counterfactuals varying one rate, other rate parameters were held constant. For HIV screening, we modeled the two separate scenarios based on assumptions that PrEP initiation was linked to regular screening versus unlinked (i.e., spontaneous). For both scenarios, we simulated relative increases in current race-stratified screening rates as well as potential standardized target screening intervals (annual, biannual, quarterly) that would be equal across race groups. We took a similar approach in the care linkage scenarios, while we only varied retention rates relative to base rates as there are no published target intervals.
We then considered the impact of increases in screening and retention jointly. First, we varied the relative screening and retention rates together to explore the combination of the two interventions, with two separate models based on the assumptions about whether HIV screening and PrEP initiation are directly linked, as described above. Second, we simulated the most optimistic scenario considered, a 10-fold improvement in both screening and retention with PrEP linked to screening for a 50-year period to estimate how long it would take to reach the national EHE goal in the modeled population of MSM in Atlanta if these improvements were sustained.
Simulation and Analysis
For each scenario, we simulated the model 1000 times and summarized the distribution of results with medians and 95% simulation intervals (SI). Primary epidemiological outcomes were overall and race/ethnicity-stratified HIV incidence per 100 person-years at risk (PYAR) during the final intervention year. We also calculated the cumulative percent of infections averted (PIA) comparing the cumulative incidence over the intervention period for each counterfactual scenario to that in the base scenario. Process outcomes associated with each scenario were also calculated, including measures of the HIV care continuum (proportion of MSM with HIV who were screened, linked to care, and HIV virally suppressed at the end of the intervention scenario) and prevention continuum (current PrEP coverage).
RESULTS
Table 1 presents the scenarios in which PrEP initiation was linked to HIV screening. Increases to the relative screening rates were associated with lower HIV incidence and higher PIA. A 10-fold increase in screening, which corresponds to increasing to approximately quarterly screening for white MSM and biannual screening for black and Hispanic MSM, resulted in a reduction in incidence from 1.27 to 0.72 per 100 PYAR overall, with 41.2% infections averted. The impact on incidence was non-linear across the relative rates, with the greatest difference in PIA at lower relative increases in rates: the difference in the PIA comparing base rates to 2-fold (13.4%) was larger than a doubling of rates from 5-fold to 10-fold (10.8%). By race/ethnicity, black MSM had the highest base HIV incidence (2.57 per 100 PYAR), followed by Hispanic MSM (0.78 per 100 PYAR), and white MSM (0.38 per 100 PYAR). With a 10-fold relative change to screening rates, a higher percent of infections would be averted for white MSM (52.1%) compared to Hispanic MSM (45.2%) and black MSM (38.8%) because of the larger absolute increase in screening for white MSM.
HIV Incidence Associated with Changes to Relative Screening Rates and Average HIV Screening Interval, Overall and Stratified by Race/Ethnicity, with PrEP Initiation Linked to HIV Screening
The mechanisms for HIV incidence reduction in these scenarios are shown in Supplemental Tables 14–17. HIV prevention occurred through a combined reduction in risks of HIV acquisition and transmission. Reduced HIV acquisition risk occurred through screening-driven increases in PrEP coverage, from 14.9% in the base scenario to 67.0% with a 10-fold increase in screening. As screening occurred more often, MSM had a greater opportunity to initiate PrEP during periods of indications. Standardizing screening rates and the offer of PrEP across race groups removed disparities in PrEP coverage. Reduced HIV transmission rates occurred through increased VL suppression that lowered the probability of transmission. Screening specifically increased the prevalence of suppression among all MSM (from 48.9% in the base scenario to 55.8% in the 10-fold scenario) given fixed linkage and retention rates. Annual screening increased the proportion diagnosed from 83.4% to 96.9%, and the median delay between infection and diagnosis decreased from 3.5 years to 2.2 years. However, screening alone did not remove baseline disparities in rates of viral suppression.
Table 2 shows the same set of screening rate counterfactuals under the assumption that PrEP initiation is not linked to HIV screening. In these scenarios a 10-fold increase in screening rates led to 9.9% of all infections being averted. This was approximately one-quarter the impact of the PrEP-linked scenario with the same relative increase in screening (9.9% versus 41.2% infections averted). By race/ethnicity, the reductions in incidence did not vary substantially for the relative rate counterfactuals. For the target screening intervals, annual screening averted 8.2% infections and quarterly screening averted 10.8% infections.
HIV Incidence Associated with Changes to Relative Screening Rates and Average HIV Screening Interval, Overall and Stratified by Race/Ethnicity, with Random PrEP Initiation (Not Linked to HIV Screening)
Supplemental Tables 18–21 demonstrate that, unlike in the PrEP-linked scenarios, PrEP coverage did not vary with increases to HIV screening: it remained constant at ∼15% overall. Here the only prevention mechanisms are through reduction in transmission risk through changes to VL suppression levels, from 50.8% in the base scenario to 56.6% in the 10-fold scenario. Because these changes in VL suppression were similar to those in the PrEP-linked scenarios, the difference in infections averted between the two scenario sets was attributable to increases in screening-driven PrEP coverage.
Table 3 presents the model scenarios for HIV care linkage and retention. Compared to base rates of HIV care linkage within 30 days of diagnosis, a 2-fold increase in rates had little impact on HIV incidence. Even immediate linkage averted fewer than 1% of infections compared to current estimated linkage rates. The impact of care linkage interventions did not vary by race/ethnicity. Mechanisms are shown in Supplemental Tables 22–25. Although doubling the rate of care linkage would increase the proportion linked to care within one month of diagnosis from 64.8% to 90.1%, this by itself (i.e., with HIV care retention fixed) had no impact on the overall levels of VL suppression.
HIV Incidence Associated with Changes to HIV Care Linkage Rates and HIV Care Retention Rates, Overall and Stratified by Race/Ethnicity
Alternatively, increases in care retention, resulted in substantial reductions in HIV incidence. Overall, a 2-fold increase in retention rates would avert 16.3% of infections and a 10-fold increase would avert 33.5% of infections. The time to first ART stoppage in the reference scenario was 3.1 years, 3.5 years, and 6.2 years for black, Hispanic, and white MSM (see Supplemental Table 12), so a 10-fold increase would represent remaining on ART for 31, 35, and 62 years on average, respectively. By race/ethnicity, there were only minor differences in the PIAs across the rate scenarios, with 33.1% to 35.5% averted in the 10-fold improvement scenario. The mechanisms are shown in Supplemental Tables 26–29. Increases in retention result in large increases in the fraction of HIV-diagnosed MSM and all MSM with HIV with VL suppression: a 10-fold rate increase would result in 89.6% of diagnosed MSM achieving VL suppression. As the race/ethnicity-stratified tables show, although there were large differences in VL suppression in the base scenarios, a 10-fold increase in the retention rates led to nearly 90% of those with diagnosed infection in each race/ethnicity group achieving viral suppression.
Figure 1 visualizes the PIA when screening rates and retention rates were jointly varied up to 10-fold from current estimates. The two panels contrast the effects under assumptions of PrEP-linked screening. If both screening and retention were increased 10-fold, 66.6% and 48.5% of infections were averted under assumptions of PrEP linkage and no-linkage, respectively. In the PrEP-linked scenario, the vertical contours at lower levels of screening demonstrate the greater relative benefit of increased screening (with PrEP initiation). In the PrEP-unlinked scenario, the horizontal contours show the relatively limited benefit of increased screening (without PrEP increases) compared to HIV care retention. Numerical results are provided in Supplemental Table 30. In the model with model screening and retention rates increased 10-fold, the proportion of MSM with HIV who were diagnosed exceeded 99% and the proportion of the diagnosed who were virally suppressed was 89%.
Percent of cumulative infections averted (PIA) over 10 years as a function of relative improvements to HIV screening rates and HIV care retention rates, under a scenario where the offer of PrEP is linked to screening negative (left panel) or that the offer of PrEP occurs as a separate activity with no linkage to HIV screening (right panel). The maximum PIA achieved (upper right corners of each panel) were 66.6% and 47.9%, respectively.
Because only a 67% cumulative incidence reduction was achieved after 10 years in the most optimistic scenario in Figure 1 (the upper right corner of the left panel), Figure 2 displays simulating this scenario out for 50 years to evaluate how long it would take to reach the EHE 2025 75% reduction target and the 2030 90% reduction target in point incidence. The reference model held the HIV prevention and care continuum levels constant whereas the 10×10 model increased HIV screening (with linked PrEP) and retention rates 10-fold. We projected that 75% incidence reduction target would be achieved in year 2027 (7.3 years post-intervention initiation) and the 90% incidence reduction target would be achieved in 2050 (30.4 years post-intervention initiation).
HIV incidence rate per 100 person-years at risk (PYAR) over 50 years comparing a scenario in which the HIV prevention and care continuum was fixed (Reference Model) to a scenario in which there were 10-fold increases in rates of HIV screening (with linked PrEP) and care retention (10×10 Model). The EHE target of a 75% reduction in the incidence rate is achieved in the 10×10 Model in year 2027 (7.3 years post-intervention initiation) and the 90% incidence reduction target is achieved in 2050 (30.4 years post-intervention initiation).
DISCUSSION
In this study, we found that major improvements to an integrated HIV prevention and care continuum could avert over two-thirds of HIV infections expected among MSM in Atlanta over the next decade. HIV screening interventions combined with HIV PrEP linkage would provide a major prevention benefit compared to HIV screening not tied to PrEP Although it may be possible to reach the 2025 75% incidence reduction target within the next decade with this combination of large improvements in both the treatment and prevention continua, reaching the 90% was not projected to be possible in this timeframe. Even achieving the EHE goal in high-need areas like the Southeast within 30 years would require a substantial acceleration of HIV prevention and care service delivery.
Several mathematical modeling studies have evaluated how meeting HIV care continuum targets in the United States impacts HIV incidence,34,35 while others have evaluated the effects of specific HIV prevention tools,14,15 but few modeling studies have evaluated packages of interventions and their synergistic effects on an integrated HIV prevention and care continuum. Previous care continuum models for the U.S. have generally found that improving HIV care retention would be higher impact than focusing on HIV screening or linkage,10,36,37 although that has not been universal.11,38
We found that the relative prevention benefit of screening versus retention depend upon assumptions of separate versus integrated HIV prevention and care continua. PrEP-linked screening, by lowering both acquisition and transmission risks, resulted in an impact of screening comparable to that of improving rates of retention for MSM with HIV. Previous models of HIV screening and linkage interventions for MSM with HIV have only included the reduction of HIV transmission risk through accelerated suppression of HIV VL.10 The transmission risk reduction benefits of HIV screening and linkage interventions compared to HIV care retention interventions are transient. The average time between infection and diagnosis, and between diagnosis and care linkage are approximately 2.5 years and 3 months.28 HIV retention interventions may now span decades under appropriate ART,39 and our 10-fold retention increase scenario represented nearly lifetime maintenance on ART.
However, even ambitious levels of screening alone may be insufficient to meet the EHE incidence reduction goals in this population. The current EHE plan has intermediate targets of 95% of persons living with HIV to be diagnosed and 95% of the diagnosed to be virally suppressed.5 We found at least a 5-fold increase in current screening rates would be needed to achieve the former target, while a 10-fold retention increase only yielded 89% for the viral suppression target in this population. Although we kept HIV care reengagement rates fixed at current estimated levels for model identifiability, improved care reengagement may also necessary to achieve the VL suppression target.
Our joint 10×10 scenario (Figure 2) suggests that combined of improvements to HIV screening linked with PrEP initiation and HIV care retention could yield major reductions in HIV incidence within the first decade (even coming close to meeting the 2025 75% incidence reduction target), but with slower improvements after that. This is consistent with our past modeling that suggested most prevention benefits accumulate immediately after initiation.14 Challenges in meeting the 2030 90% EHE incidence reduction target may also reflect that our model was parameterized to the HIV epidemic for MSM in Atlanta, with high HIV incidence and low rates of PrEP and VL suppression at baseline. This may also suggest that more targeted approaches to HIV screening linked with PrEP and ART are also needed, including those aimed at detecting emerging HIV clusters.40
Limitations
There are at least three limitations with this analysis. First, current HIV screening practice is more heterogenous than the form represented in the model. Whereas we simulated race/ethnicity-stratified rates of interval-based screening, real-world HIV screening events occur through regular clinical check-ups and risk events (such as acquiring a new partner).41 Additionally, HIV surveillance data suggests a bimodal distribution of screening, where some MSM get routine screening and some MSM remain undiagnosed until AIDS.27,42 We did not model this heterogeneity because of the complexity in representing the interaction of testing typologies with HIV progression rates; it would also complicate the design of the model scenarios. Second, we did not investigate the cost-effectiveness of the modeled interventions. Despite the epidemiological advantage of HIV screening linked with PrEP, the cost of medication would be considerable if PrEP coverage increased as we projected.43 However, economic analyses for PrEP may be in a state of flux because of the unknown pricing of generic TDF/FTC as it becomes available in 2020, as well as new programs that incentivize PrEP access as part of the EHE initiative.44 Third, our target population (MSM in the Atlanta area) was selected based on the availability of data for model parameterization, the burden of HIV, and the focus of the EHE there. Some of the broader conclusions about the relationship between HIV screening and PrEP, and between combinations of interventions on HIV incidence, may differ in other populations and may not be generalizable to the feasibility of achieving the EHE targets nationally.
Conclusions
The plan to End the HIV Epidemic was released in 2019 to combat persistent challenges in reducing HIV incidence in the U.S. The strategy of first targeting areas with the highest number of new HIV diagnoses means that EHE will also need to be implemented in places with many challenges and existing missed opportunities for detection, prevention, and control of HIV.5 EHE’s ambitious 2030 HIV prevention goals require substantial increases in access to and use of HIV diagnostic, prevention, and treatment tools. Understanding how these tools work together, across an integrated HIV prevention and care continuum, is critical to optimizing their implementation in a way that addresses the EHE goals.
Data Availability
Model data and analysis scripts are available on our Github repository linked below.
Footnotes
Funding This work was supported by Centers for Disease Control and Prevention cooperative agreement number U38 PS004646 and National Institutes of Health grants R21 MH112449 and R01 AI138783.
Disclaimer The findings and conclusions in this manuscript are those of the authors and do not necessarily represent the official positions of the Centers for Disease Control and Prevention
Conflicts of Interest None
REFERENCES
