How do I know which adolescent girls are at greatest risk for HIV?

Innovative approaches can help find vulnerable young women in need of HIV prevention services

The following is a guest post by Michelle Li, MS, and Quinn Lewis, MIA, of Data.FI project

In sub-Saharan Africa, adolescent girls and young women are disproportionately impacted by HIV and are more likely to become infected, compared to their male peers. DREAMS (Determined, Resilient, Empowered, AIDS-free, Mentored, and Safe) programming, funded through the United States President’s Emergency Plan for AIDS Relief and others, is designed to address the range of social, cultural and economic factors that make girls and young women particularly vulnerable to HIV. 

How do programs close this risk gap and prevent those infections? It’s a complex challenge that requires new techniques to locate ― at a highly granular level ― those adolescent girls and young women who are at heightened risk for HIV.

In our work on Data.FI, funded by PEPFAR, we have devised an innovative approach to estimating the size of populations at risk of acquiring HIV. Specifically, our approach is designed to:

• More accurately calculate the number of adolescent girls and young women who are at greatest risk of contracting HIV, but are not yet infected;
• Estimate the percentage of adolescent girls and young women at higher risk that DREAMS programs have reached with interventions:
• Generate highly granular maps that precisely locate the most at-risk adolescent girls and young women, and the percentage of those who are receiving interventions

How do we do this? Our solution involves combining traditional data sources, such as representative household surveys and satellite imagery, with machine-learning software to fill gaps in these data sources. First, we use traditional population-based household survey data ― e.g., from the Demographic and Health Surveys and the Population-based HIV Impact Assessments ― to calculate the proportion of adolescent girls and young women at greater risk based on known factors including early sexual debut, having multiple sexual partners, alcohol use, a history of violence, being orphaned, and not being in school — among others.

These surveys typically only identify representative populations at the national or regional levels, limiting the ability of programs to find adolescent girls and young women in need of services at say, the district level. To address this, Data.FI partner Fraym uses machine learning and artificial intelligence to combine geotagged survey data with satellite imagery data to predict survey data values for all non-enumerated areas at hyper-local levels. From this process, we are able to generate a modelled surface that depicts, at a 1-square-kilometer resolution, where  adolescent girls and young women at higher risk in each age band are likely to live. These estimates can be aggregated to generate population size estimates at program area levels, and can be visually depicted as hyper-local heat maps.

Reaching HIV epidemic control requires protecting the most at-risk adolescent girls and young women and other populations most likely to acquire HIV before they are infected. Using all the tools we have available, including artificial intelligence, allows us to better locate the most vulnerable and more equitably inform prevention services. These are girls’ lives. Staying healthy protects their future.

Michelle Li is the technical advisor for measurement and learning at Data.FI, Palladium.

Quinn Lewis is analytics team lead at Data.FI, Fraym. Data.FI helps PEPFAR and local governments improve their ability to access data for a range of HIV services, synthesize data across multiple sources, and helps drive change management processes that promote local capacities for data use. For more information, visit https://datafi.thepalladiumgroup.com/