Getting to Zero: Triangulating HIV Incidence Predictions from Micro-Level Network Modeling and Macro-Level Incidence Projections

Introduction: While new HIV infections in Chicago have stabilized over the past five years, eliminating them will require scale-up of biomedical interventions, including antiretroviral treatment (ART) for HIV-positive individuals, and preexposure prophylaxis (PrEP) for HIV-negative individuals. To achieve appropriate levels of scale-up, it is critical to project HIV incidence over the short term (the next ten years, for example). Young Black men who have sex with men (YBMSM) are a population that is disproportionately vulnerable to HIV infection, and eliminating new infections among YBMSM will be a significant step in HIV elimination across other vulnerable populations in Chicago. Projecting the expected number of HIV infections is a first step in evaluating interventions to assess elimination interventions. We will project the number of new HIV infections over the next decade among YBMSM in Chicago using three different approaches, and compare the ART and PrEP uptake levels needed to eliminate new HIV infections as predicted by the three methods.

Methods: We will use data on new HIV infections in Chicago, and apply two methods to project the number of new infections at the macro level: locally weighted regression (loess: locally weighted smoothing), and a projection method proposed by Bonacci and Holtgrave (B&H). Loess is a non-parametric regression technique, which allows us to fit multiple regression lines on clusters of points that appear close together. This method is flexible as it does not require the assumption that all the data can be described using a regression line with one slope. The B&H method projects future HIV incidence using the average change in incidence over the past three years. Both of these methods use macro-level incidence data at the population-level to project the number of new HIV infections. We will compare the new infections predicted by these methods, with an agent-based model for YBMSM in Chicago that is being developed as part of the BARS study. This model integrates the Repast High Performance Computing toolkit for agent-based modeling with the statnet suite in R for network modeling.

Results: We will estimate the number of HIV infections over the next 10 years predicted by each model when the levels of ART and PrEP uptake are assumed to be at current levels. We will then consider scenarios where ART and PrEP are exclusively scaled up by about 20% over a 10-year period, and estimate the number of new infections obtained. We will then consider scenarios where a combination of ART and PrEP are scaled up together. Finally, we will consider the “inverse problem” of how much ART and PrEP need to be scaled up by to eliminate new infections.

Conclusions: Modeling frameworks that operate at the micro- and macro-levels provide alternative frameworks to compare the potential efficacy of prevention interventions. These assessments can help create meaningful targets for intervention scale-up to achieve public health goals.