Addressing Gaps in HIV Preexposure Prophylaxis Care to Reduce Racial Disparities in HIV Incidence in the United States

Samuel M. Jenness; Kevin M. Maloney; Dawn K. Smith; Karen W. Hoover; Steven M. Goodreau; Eli S. Rosenberg; Kevin M. Weiss; Albert Y. Liu; Darcy W. Rao; Patrick S. Sullivan


Am J Epidemiol. 2019;188(4):743-752. 

In This Article


In this modeling study, we found that implementation of PrEP could reduce absolute disparities in HIV incidence between BMSM and WMSM even despite current racial gaps in HIV PrEP care. Further disparity reduction with PrEP could be achieved with interventions targeting each of the modeled PrEP continuum steps for BMSM. Major gains in overall HIV incidence reduction and disparity elimination with PrEP would require targeting initiation factors (awareness, access, and prescription) over engagement factors (adherence and retention) for BMSM, given the currently observed PrEP continuum.

Many HIV-prevention interventions successful at reducing HIV incidence are challenged by simultaneously addressing persistent HIV racial disparities. Systemic racial gaps in clinical care for testing and treatment of HIV[44] have led to an HIV-prevention landscape in which white and higher-income MSM disproportionately benefit.[45] We rooted our model structure and parameters in robust data to estimate how empirical representations of the PrEP continuum could affect HIV incidence over the next decade in a high-burden, low-resource population of younger BMSM.[26] Our model suggests that it is possible to reduce, although not entirely eliminate, disparities in HIV incidence by race while at the same time reducing HIV incidence overall with PrEP.

To guide public health policy, we used a 5-step PrEP care-continuum framework to conceptualize gaps in PrEP care.[3] First, we found that awareness of PrEP was the step most strongly associated with incidence reduction for BMSM, partially due to the marginally declining conditional probabilities for the subsequent steps. Several studies have found reduced interest of BMSM in PrEP,[12,13] related to lack of knowledge about PrEP and perceived stigma in using it.[14,46,47] New technologies, such as mobile phone applications, are currently being developed to address this step. Second, PrEP access given awareness could increase infections averted by 4.1% if raised to levels observed among WMSM in our model. Access-related interventions include patient assistance programs to cover medication costs;[48] however, PrEP requires ongoing monitoring services covered through health insurance, which may be a barrier for some BMSM.[15–17] Third, we found a relatively minimal effect for prescription rates conditional on access because the observed gap was only 10% (73% versus 63%). While all indicated BMSM seeking a PrEP prescription should receive one, this will depend on indications for PrEP being accurately queried by clinicians who are willing to prescribe PrEP. BMSM are less likely to be "out" to their doctors,[49] and some clinicians may be less willing to prescribe to BMSM than to WMSM.[50] Clinical training on PrEP patient assessment is greatly needed. Fourth, adherence is critical to both the impact and efficiency of PrEP, with a substantial effect on the NNT. Race/ethnicity has been strongly associated with suboptimal PrEP dosing.[11,51] Long-acting formulations like injectable cabotegravir might be of benefit to BMSM with adherence barriers.[52] Finally, greater retention in PrEP care was strongly associated with both infections averted and lower NNT in our model. PrEP discontinuation for reasons other than lapsed indications has been an increasing challenge in clinical practice as PrEP users mature;[53] lessons learned from managing patients with suboptimal levels of retention in HIV medical care may guide considerations of how to limit PrEP discontinuation.[3]

We quantified disparities on an absolute index that subtracts the standardized incidence rate for WMSM from the BMSM incidence rate and a relative index that takes their ratio. Many policy documents use the latter: The National HIV/AIDS Strategy, for example, sets a goal to reduce racial disparities in new HIV diagnoses by a relative measure.[6] In a dynamic intervention context, however, we would suggest that ratios are less suitable than differences for 3 reasons. First, the population-level burden of disease is quantified by the incidence rate of disease per unit of person-time. Using the absolute disparity allows one to express disparities with this same denominator. Second, there are parallels in using the absolute index with the choice of risk differences versus relative risks to quantify public health impact of a risk factor in epidemiologic studies.[54] Third, the ratio scale is unstable when the denominator is small relative to the numerator, as it is here. Ratio scales may be misleading for some interventional scenarios in these cases when reducing the difference in the number of incident infections between races has the counterintuitive effect of increasing the disparity ratio. Therefore, we recommend that disparities be quantified as absolute differences.


Our model conceptualizes racial disparities by simulating a 2-race population of MSM (of younger non-Hispanic black and white MSM) in the Atlanta area. The conclusions drawn from this study are therefore most applicable to this target population. Deviations from random sampling of MSM in this target population from the 2 network/behavioral studies[11,21] could have resulted in biases in the estimates of model parameters in these domains. Specifically, because most parameters represent marginal probability and rate estimates, the resulting statistics in our models depend on the specific distribution of covariates in our particular study sample. An overrepresentation of young MSM in these studies, for example, could have resulted in upwardly biased behavioral risk parameters if positively correlated with age. Clinical and biological parameters were drawn from the secondary literature; aggregating multiple data streams into a single model requires strong assumptions about exchangeability, the implications of which have recently been examined in the methodological literature.[55] However, a related strength of our study with respect to parameterization is its rigorous Bayesian model calibration and validation methods to evaluate and adjust for sources of parameter uncertainty through fitting the model projects to external HIV and sexually transmitted infection prevalence and incidence data. Additionally, our model may be limited by the assumption that routine HIV screening is the primary point for entry into PrEP, based on the requirement that HIV testing be performed before PrEP initiation.[37] Initiation of PrEP before specific sexual risk events has also been observed,[10] and our future work will explore variations in reasons for starting PrEP. Finally, the continuum parameters were also based on 2 studies with race-stratified estimates, and these BMSM study populations may not represent other populations of HIV-uninfected BMSM in the United States. Further parameter data are needed for other geographic settings to transport these findings to other MSM populations.[56]