Comprehensive Summary
This study applied six machine learning models to identify predictors of problematic substance use (PSU) among 200 Ugandan youth (ages 18–24) living with HIV. The random forest model performed best (AUROC = 0.78, AUPRC = 0.75). Key predictors included depression, risky sexual behavior, low socioeconomic status, adverse childhood experiences, household size, exposure to alcohol education modules, and knowledge of alcohol’s effects on HIV treatment. The findings underscore how social, psychological, and environmental factors intersect to elevate PSU risk. The authors propose that such models could help identify high-risk youth in HIV-affected regions and inform tailored prevention strategies.
Outcomes and Implications
Machine learning offers a way to identify youth at highest risk of problematic substance use (PSU), particularly among those living with HIV in Uganda. Early detection enables targeted interventions that account for psychological, social, and environmental risk factors such as depression, adverse childhood experiences, and alcohol use. Addressing PSU is vital for maintaining HIV treatment adherence and improving health outcomes. Public health officials could also use these models to design culturally sensitive prevention programs, strengthening care delivery for vulnerable populations.