Predictive modelling of HIV infection risk among people who inject drugs in north-west Nigeria: A multidimensional analysis of sociodemographic, psychosocial, and behavioral determinants

It is vital to accurately help understand who is at risk of HIV infection for proper prevention planning and resource allocation. In the current study, 79% accuracy was achieved in the predictive modelling of the risk of HIV infection in People Who Inject Drugs in North-West Nigeria using machine learning. With this degree of predictive capacity, prevention can be improved to the point that policymakers can limit new infections by tailoring interventions to target the likely locations of new outbreaks and spread. More generally, this information can help optimize the use of limited resources across the region. The analysis used 50,000 anonymized records collected over three years across three states. Sociodemographic, psychosocial and behavioral risk factors were incorporated in the models. Logistic Regression won against Random Forest and XGBoost with an AUC score of 0.8578. Needle sharing, depressive symptoms, trauma exposure, and low social support were key predictors of HIV infection. Findings underscore the significance of including mental health screening and trauma-informed harm reduction within HIV prevention for PWID in this region.