Abstract Body

Understanding social dynamics behind HIV testing may provide novel ways to achieve high testing coverage more efficiently. We tested whether social network data predicted testing uptake in a community wide mobile HIV testing campaign.

We built a social network for a rural Ugandan community with 4810 adults where, as part of the SEARCH Trial (NCT01864603), a hybrid approach of community health campaigns (CHC) followed by home based testing (HBT) for non-attendees tested 94% of adult stable residents (82% by CHC, 12% by HBT). In a baseline census, adult residents (>=15 years old) provided contact information for up to 6 friends in each of five social domains (health, money, emotional, food, and free time). Named contacts were matched to a census enumeration to build the network. Associations between testing uptake and network characteristics, including local network density (proportion of realized edges among all possible), clustering (proportion of trios that are fully connected), and number of named contacts from each social domain, were evaluated using logistic regression with cluster-robust standard errors, adjusting for age, gender, occupation, education, marital status, and wealth. Structural features of a subnetwork of HIV+ adults were compared to 1000 simulated populations of the same prevalence and network topology.

The community-wide network contained 25148 links across all domains, with 96% of individuals connected in a single cluster (Figure 1). After adjusting for demographics, subjects with fewer named contacts in any domain and subjects with a denser local network were both significantly less likely to attend CHC and more likely to fail to test at either CHC or HBT. The greatest risk for not testing was conferred by higher network density (aOR 2.9, 95%CI 1.47, 5.74) and fewer named free time contacts (aOR 1.24, 95%CI 1.11, 1.38). Increased risk for CHC nonattendance was also associated with high local network clustering (aOR 1.78, 95%CI 1.06, 2.97). A subnetwork of 100 HIV+ adults was denser and had more clusters of size >=4 than expected by chance, and showed multiple social links to non-testers.

Network-based testing strategies that target dense and clustered local networks and increase social outreach may achieve high coverage more efficiently. Social network data identified clusters of HIV+ individuals with social links to multiple non-testers, potentially informing identification of hidden high-risk individuals.