Identifying predictors of problematic substance use among youth living with HIV in Uganda: a machine learning approach
An article in AIDS and Behavior, published September 18, 2025
Authors: Claire Najjuuko, Rachel Brathwaite, Massy Mutumba, Saltanat Childress, Sylivia Nannono, Phionah Namatovu, Chenyang Lu, Fred M Ssewamala
This study explored the significant issue of substance use among young people living with HIV in Uganda and applied machine learning techniques to predict problematic substance use (PSU) in this group. By analyzing data from 200 young individuals aged 18 to 24, the researchers developed and tested six different predictive models to identify key risk factors for substance use and determine which model performed best. They found that the random forest model was the most effective, achieving a high level of accuracy.
The study revealed several important predictors of substance use, encompassing individual, interpersonal, and community factors. These included depression, risky sexual behaviors, low monthly income, experiences of adversity during childhood, family involvement in selling alcohol, friends facilitating access to alcohol, exposure to community campaigns against alcohol, household size, and awareness of alcohol's impact on HIV treatment. These findings underscore the potential of machine learning to identify youths at risk and inform targeted interventions. The insights from the study can help HIV researchers and service providers develop better policies and support measures to reduce substance abuse and improve HIV care among young people in Uganda.