A Statistical Model for Inference of Recent and Incident HIV Infection Using Surveillance Data on Individuals Newly Diagnosed With HIV Infection in Scotland

Abstract

Background: To inform global ambitions to end AIDS, evaluation of progress toward HIV incidence reduction requires robust methods to measure incidence. Although HIV diagnosis date in routine HIV/AIDS surveillance systems are often used as a surrogate marker for incidence, it can be misleading if acquisition of transmission occurred years before testing. Other information present in data such as antibody testing dates, avidity testing result, and CD4 counts can assist, but the degree of missing data is often prohibitive. Methods: We constructed a Bayesian statistical model to estimate the annual proportion of first ever HIV diagnoses in Scotland (period 2015–2019) that represent recent HIV infection (ie, occurring within the previous 3–4 months) and incident HIV infection (ie, infection within the previous 12 months), by synthesizing avidity testing results and surveillance data on the interval since last negative HIV test. Results: Over the 5-year analysis period, the model-estimated proportion of incident infection was 43.9% (95% CI: 40.9 to 47.0), and the proportion of recent HIV infection was 21.6% (95% CI: 19.1 to 24.1). Among the mode of HIV acquisition categories, the highest proportion of recent infection was estimated for people who inject drugs: 27.4% (95% CI: 20.4 to 34.4). Conclusions: The Bayesian approach is appropriate for the high prevalence of missing data that can occur in routine surveillance data sets. The proposed model will aid countries in improving their understanding of the number of people who have recently acquired their infection, which is needed to progress toward the goal of HIV transmission elimination.