Introduction Baltimore City has one of the highest HIV incidences and prevalences in the United States. An HIV testing program, implemented in several emergency departments (EDs), has accounted for 11% of newly diagnosed HIV cases from 2008–2013. We derive an agent-based model (ABM) for HIV transmission in Baltimore City, and use this to determine the significance of ED-based HIV testing on HIV transmission.

Methods An agent-based computational simulation was performed via the Python programming language, using 523,113 agents to represent the 13+ population of Baltimore City. The simulation was calibrated using HIV prevalence and incidence data culled from 2007 to 2013 City surveillance data. During each timestep, agents interacted with other agents. Agents were assigned one of three categories: seronegative, seropositive aware, or seropositive unaware, and individual risks were assigned from these categories, with seropositive unaware agents being 3.5 times more likely to transmit the disease. ED testing changed unaware agents to aware, and rates of testing were varied in order to study the effects on overall incidence. A subsequent sensitivity analysis was performed using different ranges of parameters, and a range of incidence projections was calculated.

Results Baltimore City HIV incidence decreased from 1,052 new cases (0.207%) in 2007 to 356 (0.068%) in 2013. Our model was able to approximate HIV incidence over time as observed from 2007 to 2013. Overall HIV incidence is forecast to decrease from 0.068% in 2013 to 0.042% in 2020 (95% CI: 0.015–0.079). It is further demonstrated that doubling capacity of the ED-based testing programs would likely avert 35 additional HIV transmissions from 2014 to 2020.

Conclusion We conclude that ABM provides an effective means of describing an epidemic with a highly heterogeneous population, and additionally that the ED-based testing program has had a significant impact on curtailing HIV transmission in Baltimore City.

Disclosure of interest statement This work was supported by a National Institutes of Health grant [K01AI100681 to Y-HH]. No conflicts of interest are reported.