Monitoring HIV incidence is vital for characterizing the epidemiology and trajectory of HIV epidemics and impacts of prevention efforts. Standard methods for measuring incidence such as cohort studies take considerable time and cost and are often not feasible in settings, leading to a reliance on new HIV diagnoses, inherently a biased measurement. We compare HIV incidence estimation using 3 different methods/data sources which might leverage available program data in a variety of settings.
We used data derived from a cluster-randomized trial among people who inject drugs from 12 Indian cities to estimate HIV incidence. First, we used a validated multi-assay algorithm (MAA) to define recent HIV infection within the trial’s follow-up cross-sectional samples (Aug 2016-Apr 2017) accrued using respondent-driven sampling (RDS). Second, we estimated incidence from PWID that participated in two (confirmed via biometrics) cross-sectional RDS samples – baseline (Jan-Dec 2013) and follow-up (Aug 2016-Apr 2017) – and were serologically HIV negative at baseline. Third, we estimated incidence from initially HIV-negative PWID clients who received one or more repeat HIV tests at integrated care centers (ICCs) (Jun 2014-Feb 2017) in 6 of the 12 cities. The goal was to test clients every 6 months. ICCs also provided opioid agonist therapy and other PWID services (e.g., needle exchange).
Across all cities, MAA-estimated incidence was generally highest, followed by the serial cross-sectional, with ICC estimates being substantially lower. MAA annual incidence ranged from 18.5% (New Delhi) to zero (Imphal), serial cross-sectional incidence from 16.1% (Kanpur) to 0.3% (Imphal), and ICC incidence from 7.3% (Aizawl) to 0.1% (Imphal). On average, the serial cross-sectional estimate was 19% lower than the MAA (range: -60% to +190%) and 20% higher than the ICC (range: -32% to +953%). While estimates were variable, rank order generally stayed the same across the estimates (Figure). Spearman rank correlation was 0.94 for the MAA-serial cross-sectional estimates, 0.83 for MAA-ICC, and 0.66 for serial cross-sectional-ICC estimates.
While HIV incidence estimates within a given city were variable by method, the rank order by incidence was consistent. While use of facility-based data will generally underestimate population incidence, using cross-sectional population-based data to estimate HIV incidence can prioritize where resources may optimally be directed.