HIV Risk Increases with Positive Ties in Highly Connected Sociospatial PWID Network
Abstract
Background: People who inject drugs (PWID) bear high HIV and hepatitis C virus (HCV) burden and account for some of the most explosive epidemics globally. While individual risk factors for infection are well understood, less is known about network and spatial factors. Moreover, network studies have been limited by focusing on immediate ties (egocentric network) rather than the broader sociometric/spatial networks. Methods: 2,512 PWID were recruited via a chain referral method in 2017-19 in New Delhi, India. An index initiated sampling and was asked to recall who they injected with in the past month and was provided referrals for those partners (index’s egocentric network). Similarly, each recruit named and recruited their recent injection network (recruit’s egocentric network and index’s sociometric network). Participant biometrics identified duplicates and cross-network linkages. All completed a survey, provided blood and information on injection locations; these data were used to generate spatial networks. Sociometric injection networks were created and analyzed using bespoke Python code. Individual and network-level factors were analyzed for associations with prevalent HIV infection; machine learning was used to nominate predictors. Results: Median age was 26; 99.1% were male. HIV prevalence at baseline was 36.9% and 7.4% were virally suppressed; HCV antibody prevalence was 65.1%. The networks of 8 of 11 indexes merged into one network (Figure). Average degree (number of injection partners) was 2.1 (range: 0–47), network diameter was 39 and average path length was 14. Of 928 HIV-positive participants at baseline, 64.6% were directly connected with at least one other HIV-positive PWID. Of 1,634 HCV-positive participants at baseline, 86.8% were directly connected with at least one other HCV-positive PWID. The odds of HIV increased with each additional HIV-infected ego in a network (OR=1.21) and injecting at a specific hotspot (OR=1.86), factors that were independent of individual needle sharing (OR=1.89) and injection frequency (OR=1.36; all p<0.001). Conclusions: These are among the first data to comprehensively characterize the complete sociometric injection network of PWID in an urban setting. We observed a highly connected network structure where HIV and HCV prevalence were associated with network connections and spatial overlap after adjusting for other predictors. These data have implications for the success of network-based prevention/treatment strategies.
