Which analytic methods tell us which adolescents are most vulnerable to HIV?

We know that adolescent girls and young women in sub-Saharan Africa are at more risk of becoming infected with HIV than are young men in their age group (10-24 years old) and that among those ages 15-19, girls accounted for five in six new HIV infections. Such new infections are a driver of the continued HIV epidemic.

We also know not all adolescent girls and young women exhibit HIV risk characteristics and not all of these individuals are at equal risk. To target prevention program resources, such as the DREAMS Initiative supported by the U.S. President’s Emergency Plan for AIDS Relief, how can we more accurately identify where to find the young women most at risk, so that our programs and resources are mobilized where they are most needed?

The Data for Implementation (Data.FI) project, funded by USAID, tested a methodological approach to size estimation in three high-HIV burden countries: Eswatini, Haiti and Mozambique. The results are now published in PLOS ONE. We sought to discover how we can better visualize diverse patterns of risk — spatial heterogeneity — in order to predict the number and proportion of at-risk adolescent girls and young women at hyperlocal levels. If successful, we would know where to target prevention programs to protect these young women; we would know more about differential effects of multiple risk factors; and we would be able to assess trade-offs between allocating resources to areas of highest risk versus areas now known to have the largest number of at-risk adolescent girls and young women.  

We explored risk factors by country and, separately, by age band using three different vulnerability classification methods: any-risk approach, regression and latent class analysis. We then applied machine learning and artificial intelligence to combine geotagged survey data with satellite imagery data to predict risk at a 1-square-kilometer resolution. From this process, we were able to generate a modelled surface that provides a picture of where high-risk adolescent girls and young women in each age band are likely to live.

We found we could target and prioritize prevention efforts among adolescent girls and young women by correlating two known quantities: risk/vulnerability characteristics and geographic/spatial mapping. To our knowledge, this is the first study to compare risk profile methods and simultaneously to add satellite data that can pinpoint HIV risk patterns at a 1-square- kilometer level to map hyperlocal risk rather than population-level data on HIV incidence or prevalence.

We know it is important to understand the needs and phases (e.g., start up, scale-up, maintenance) of any HIV prevention program to determine the most appropriate risk mitigation approaches. But we also know — very importantly — that identifying and engaging high-risk adolescent girls and young women before they are infected is a leap forward for prevention programs and for ending the HIV epidemic. 

Evidence of both risk and spatial heterogeneity means that program planners can and should simultaneously consider epidemic typologies of risk and geography to better target prevention efforts with high potential to curb the disease.