CROI 2016 Abstract eBook

Abstract Listing

Poster Abstracts

Conclusions: Older partners, engagement in transactional sex and <80% school attendance/month significantly increased the risk of HIV acquisition among this cohort of young South African women. These findings confirm the importance of keeping girls in secondary school but also focusing on reducing transactional sex and older partners to prevent new HIV infections.

937 Correlates of HIV in Large Population Surveys: A Comprehensive HIV Association Study Eran Bendavid 1 ; Chirag Patel 2 ; Jay Bhattacharya 1 1 Stanford Univ, Stanford, CA, USA; 2 Harvard Univ, Boston, MA, USA

Background: Identification of at-risk individuals is a major challenge for global HIV public health goals such as “90-90-90” (90% testing, 90% on treatment, and 90% virally suppressed). We develop a newmethod for identifying HIV risk factors based on big data approaches, and show an application using nationally representative HIV survey in Zambia. Methods: We develop generalizable methods for simultaneously exploring the association of very many correlates of HIV infection status. We use Zambia’s 2013-2014 Demographic and Health Survey, a nationally representative survey that included an HIV test, as the main source of data. Each individual’s HIV status was tested against 699 indicators derived from the surveys. All indicators with at least 90% completion rate were used. We split the data into a training and testing datasets, and accepted variables with Bonferroni-adjusted association either dataset (p<7x10 -9 in our data) and p<0.05 in the other dataset. We then tested the explanatory power of the novel models to estimate the discrimination between true positive and false positive cases. Results: We analyzed data on 14,719 women between 15 and 49 years old with HIV prevalence of 13.1%. After survey weighting and Bonferroni adjustments, we identified 26 indicators that were strongly associated with HIV infection. In addition to familiar risk factors such as residence in rural areas and age, self-identifying as widowed (OR 9.6, p<1x10 -47 ) or divorced (OR 3.9, p<1x10 -22 ), working in a service job (OR 3.4, p<1x10 -10 ), and drinking alcohol (OR 2.0, p<1x10 -15 ) were strong positive risk factors. Protective associations included currently breastfeeding (OR=0.4, p<1x10 -27 ), and living in households with larger number of residents (OR 0.7 for each additional person, p<1x10 -15 ). In a predictive model, the identified associations were more likely to yield true positive rather than false positive predictions. Conclusions: We develop a new approach for high-dimensional association studies of HIV risk in household surveys. We identify under-recognized risk factors for HIV in Zambia, including being a widow and drinking alcohol. This approach could be expanded to improve risk identification throughout sub-Saharan Africa.

Poster Abstracts

Figure 1: Associated HIV risk factors above the Bonferroni threshold (out of 699)

widow

year of birth

divorcee

etoh

396

CROI 2016