The risk of HIV transmission at each step of the HIV care continuum among people who inject drugs: a modeling study

The Centers for Disease Control and Prevention (CDC) estimates that about 1 in 11 new HIV diagnoses in the United States (U.S.) occur among those with a history of injection drug use [1]. In recent years, settings such as New York City (NYC) and Vancouver have achieved dramatic reductions in HIV incidence among people who inject drugs (PWID), through efforts including needle and syringe programs and HIV treatment scale-up [2], indicating progress for HIV prevention in high-resource settings. Despite this progress, recent outbreaks have illustrated the need for expanded efforts to prevent HIV transmission among PWID [3, 4], particularly as the use of some injection drugs, such as heroin, has increased in the U.S. [5, 6].

A crucial element of HIV prevention for PWID is the use of antiretroviral therapy (ART) in reducing the infectiousness of HIV-infected individuals [7], in what is known as treatment as prevention (TasP). Reduced individual- and population-level infectiousness have translated to profound reductions in HIV incidence among PWID in many settings [8, 9]. For TasP-based strategies to be fully effective, however, barriers to achieving viral suppression must be eliminated [10, 11]. In recent years, a heuristic has evolved, described as the “HIV care cascade” or “continuum”, which delineates the steps needed to transform an HIV-infected population from being treatment-naïve and infectious to virologically suppressed on highly active antiretroviral therapy, wherein they are functionally unlikely to transmit HIV to new partners [12].

Prior research has estimated that PWID in the earlier stages of HIV care (e.g., undiagnosed, as well as diagnosed but not on ART) might contribute significantly higher rates of transmission (as much as 10-fold), even when compared to those with ineffective or inconsistent treatment [13]. These results suggest that, although achieving viral suppression is the ultimate objective from both an HIV prevention and individual health perspective, improved diagnosis and treatment enrollment may have benefits for population-level reductions in HIV transmission, even without significant improvement in virologic suppression.

Although the estimated HIV incidence among PWID in NYC has declined substantially in recent years [14], due in part to the success of harm reduction (e.g. availability of sterile syringes) and TasP-based strategies [2], gaps in the care continuum may play an important role in ongoing new transmissions observed in this population. Estimating the stage(s) in the care continuum at which these HIV transmission events arise may provide critical information for optimizing the care continuum in NYC and other settings with persistent HIV incidence among PWID.

Using an agent-based model of the NYC population in 2012 (i.e., an individual-based microsimulation of the city’s residents), we provided the first estimate of the contribution of transmission arising from PWID at each step of the HIV care continuum. To account for all transmission events arising from PWID (including those to their sexual partners, who may or may not inject drugs), our model included a representation of the entire adult population, including significant sexual risk groups, such as men who have sex with men (MSM).

The total estimated PWID in NYC living with HIV infection in 2012 are presented in Fig. 1, along with the estimated number of PWID who had received an HIV diagnosis, were receiving treatment, and were virally suppressed in 2012. The gaps between the bars in the figure (i.e., their numerical difference), along with the final estimate for those virally suppressed, represent the populations within the HIV care continuum categories used for this analysis (Undiagnosed, Diagnosed − not on ART, Unsuppressed, Suppressed).

Estimated HIV incidence among PWID in the model was 113 per 100,000 person-years over the study period, whereas the incidence rate for all adult New York City inhabitants was 33 per 100,000 person-years. These results are similar to our empirical estimates calculated using available observational data (see Estimating HIV Prevalence and Incidence in Additional file 1), where we estimated the HIV incidence rates for PWID and the overall adult population to be 107 and 32 per 100,000 person-years, respectively. The mean number of transmission events arising from HIV-infected PWID was approximately 378 per 100,000 person-years; of these transmissions, 52% were via parenteral transmission and 48% were transmitted sexually.

The main results for HIV transmission within each step of the HIV care continuum are presented in Fig. 2. As shown, 33% of PWID in 2012 were undiagnosed, 30% were diagnosed but not on ART, 12% were prescribed ART but not virally suppressed, and 26% were prescribed ART and virally suppressed. For 2012, we estimated that 52.6% (95% simulation interval [95% SI]: 47.1–57.0%) of all HIV transmission events arising from PWID were from those who were undiagnosed, 36.6% (95% SI: 32.2–41.5%) from those who were diagnosed but who had not initiated ART, 8.7% (95% SI: 5.6–11.8%) from those who had been prescribed ART but failed to achieve viral suppression, and 2.1% (95% SI: 1.1–3.9%) from those enrolled on ART who have achieved viral suppression.

The undiagnosed PWID had the greatest ratio of transmission to population size, 1.61, followed by ratios of 1.23 for those Diagnosed-not on ART, 0.74 for Unsuppressed, and finally 0.08 for Suppressed.

Results of the sensitivity analyses are presented in their respective panels in Fig. 3. As shown, the largest changes in results from the main analysis were observed when assuming fewer undiagnosed cases, or assuming a large reduction in risk behavior following HIV diagnosis. These differences are largely manifested by shifting estimated transmission to and from those with undiagnosed infection. The remaining analyses differed less substantially from those of the main analysis. Supplementary output (Additional file 1: Figures. S1 and S2) are available in the Additional file 1, presenting the proportion of HIV-infected PWID that had received an HIV diagnosis over the simulation period, as well as the proportion of all PWID diagnosed with HIV that were enrolled on ART, respectively. The original model ouputs used to generate the results presented here are available in Additional file 2, Additional file 3, Additional file 4, Additional file 5 and Additional file 6.

Using an agent-based modeling approach, we estimated that approximately one-third of all HIV transmission events arising from PWID occurs among those who have been diagnosed, yet have not begun ART, with an additional one-half occurring among those who have yet to be tested. Together, these steps accounted for an estimated 89% of all HIV transmission from PWID in 2012, despite only comprising 62% of the total infected population. These results are believed to be the first to examine PWID transmission along the HIV care continuum at a city-wide level, and indicate the importance of advancing PWID further along the care continuum, increasing rates of HIV serostatus awareness, ART initiation and viral suppression.

Although we estimated two-thirds of HIV-infected PWID to have diagnosed infection throughout 2012 in NYC, the magnitude the undiagnosed PWID population may have had on transmission was far greater than its relative size. As expected, out of the four points in the care continuum examined in this study, undiagnosed PWID produced the largest ratio of transmission to population size, at about 1.6. This result suggests that early diagnosis of HIV infection among PWID may provide the greatest opportunity to reduce incident infections on a per-person basis. Although we assumed reduced risk behavior following an HIV diagnosis (which is likely one reason for this result), there are other factors that may have influenced the likelihood of transmission from diagnosed compared to undiagnosed agents. For instance, the probability of being acutely-infected is higher among those with undiagnosed HIV infection (who are more likely to be recently infected), thus increasing transmission risk in this sub-group.

It should be noted that we assumed a non-zero probability of transmission for those with viral suppression, accounting for the transmission arising from this population. Although some equate viral suppression with an inability to transmit, it is unclear if or at what level of viral suppression this becomes absolute for an efficient mode of transmission such as parenteral exposure [36, 37]. Previous modeling work has similarly assumed a small probability of transmission, even from virally suppressed individuals [13].

From a national perspective, even though previous work has estimated that as many as 93% of all HIV-infected PWID may be previously diagnosed, only about 1 in 3 HIV-infected PWID was found to be virologically suppressed [38]. Delayed engagement in care and ART initiation have been found to be highly common among PWID [39, 40]. Research has shown several barriers to engagement in care, retention on treatment, and ART adherence, such as poverty, poor access to medical care, homelessness, and incarceration [39, 41, 42]. Despite substantial barriers to enrollment and retention, PWID often maintain a level of viral suppression (for those actively engaged in ART care) comparable to other high-risk groups in the U.S. [36]. This suggests that although treatment enrollment and retention may be hindered by significant structural and systemic barriers, the benefits of treatment initiation are high—as demonstrated by the comparably minor transmission shown in our results to originate from PWID on treatment.

Our results for NYC differed notably from earlier work by Skarbinsky, et al., who used a compartmental model to estimate HIV transmission along the care continuum for the entire U.S. [13]. We found a substantially higher contribution from PWID with undiagnosed infection (52.6% vs. 16.3%), but this difference was largely anticipated due to the stark differences between the estimated proportion of undiagnosed HIV infection among PWID in NYC (using the NHBS) and assumptions regarding the country as a whole made by Skarbinsky et al.

There are several limitations to our approach. The most significant of these is the challenge in correctly estimating the number of undiagnosed infections in NYC, among PWID, and other subpopulations. Some of our estimates for HIV diagnosis are based on a small sample size (particularly among female PWID). Additionally, specifying the behavior of diverse communities of PWID is inherently difficult (and at times based on older data), particularly considering the large range in potential sexual and injection risk behavior and network mixing [32]. In addition to the difficulty inherent in correctly specifying the HIV risk of diverse populations, it is possible that complex and confounding relationships exist between certain risk behaviors and different stages of the HIV care continuum (e.g., those less likely to engage in needle-sharing might also be more likely to be on HIV treatment and achieve viral suppression). Complex relationships such as this were not incorporated into our analyses. Future analyses may benefit from increased specification of the relationship between likelihood of engaging in risk behavior and stages along the HIV care continuum.

Methods for determining the likely route of HIV transmission among newly diagnosed cases are well documented [18‐20]; however, their accuracy is not well understood, which makes it difficult to compare our results regarding the proportion of transmission attributable to sexual or parenteral exposures. This issue is inherently important among PWID, who have such disparate risk behaviors. Although our model calibration process largely dictated the proportion of transmission via a given route, empirical estimates among PWID are difficult to ascertain [43], and this would not have altered the total estimated incidence for the analysis, which was based on NYC DOHMH surveillance among all of those with reported IDU. The NYC DOHMH classifies any new infections among individuals reporting a history of IDU under either the IDU or MSM/IDU transmission categories. Both reporting bias and inclusion of individuals who were not actively injecting during the time of their HIV acquisition may have affected our results. Since our model only classifies those currently engaging in IDU as PWID, differences in classification by the NYC DOHMH and our model may have led to inaccurate estimates for incidence among PWID.

Although newer surveillance estimates suggest that the number of HIV-infected PWID in NYC may be overestimated [35], it is unclear what impact these results may have on the proportion of undiagnosed cases among PWID and others, even if levels of treatment engagement and viral suppression may be higher than previously estimated. Our analysis suggests that higher levels of virological suppression (i.e., 70% vs. 38%) substantially shifts estimates of transmission towards the undiagnosed population, but this assumes the size of the undiagnosed population remains as it was in the main analysis.

Certain model assumptions may have also influenced the results in potentially meaningful ways. For example, HIV testing was presumed to be 100% accurate and with no window period, leaving the possibility that some infections in the model were diagnosed earlier than they may have otherwise been with older generation testing. Finally, ART is presumed to have an immediate effect on suppression of viral load and disease progression, meaning that individuals on ART experience the benefits of full or partial adherence immediately, potentially overestimating the preventive effect of ART, which may take several weeks or months to become established [44]. Other modeling or observational approaches may be feasible to confirm our findings; however, observational approaches would require a substantially enhanced surveillance system, with robust estimates for risk behavior, viral sequencing, and treatment outcomes among PWID. Although more parsimonious models may be able to avoid some of the limitations of complex agent-based modeling, they may have difficulty reproducing the many advantages of this approach, including the reproduction of the sexual and injection risk network among PWID and with non-PWID partners.

Among PWID in NYC in 2012, an estimated 89% of new HIV infections originate from persons who are not on ART. Our results demonstrate the importance of early diagnosis, as well as interventions to engage previously diagnosed PWID in care and treatment to further suppress population-level HIV transmission. With an estimated 53% of infections arising from undiagnosed PWID, further efforts should also be made to expand frequent HIV testing, which has been shown in prior research to be effective at forestalling transmission [45], but may also decrease transmission via reduced risk behavior post-diagnosis [46‐48]. Future HIV prevention research should focus on the elimination of identified and potential barriers to the testing, diagnosis, and retention of PWID on HIV treatment.