Using Machine Learning to Predict Depression among Adolescents Living with HIV in Uganda.

Area of Science:

• Digital Health
• Mental Health Technology
• Adolescent Health

Background:

• Adolescents living with HIV (ALWHIV) experience high rates of depression.
• Limited mental health services in Uganda necessitate innovative detection solutions.
• Machine learning (ML) offers a promising avenue for early depression risk identification in ALWHIV.

Purpose of the Study:

• To utilize machine learning models for predicting depression risk among ALWHIV in Uganda.
• To identify key predictors of depression in this vulnerable population.

Main Methods:

• Cross-sectional data from 833 ALWHIV (aged 10-17) in Southern Uganda were analyzed.
• Seven ML models were evaluated using 10-fold cross-validation, with performance assessed by AUROC, AUPRC, accuracy, sensitivity, specificity, and F1-score.
• SHapley Additive exPlanations (SHAP) analysis identified feature importance.

Main Results:

• Depression prevalence was 30.97% among participants.
• The Random Forest model demonstrated the highest performance (AUROC = 0.79).
• Key depression predictors included hopelessness, HIV stigma, shame, self-esteem, and support from teachers and caregivers, indicating significant psychosocial influences.

Conclusions:

• ML models show strong potential for predicting depression in ALWHIV.
• Psychosocial challenges are critical factors in adolescent depression.
• External validation is recommended to enhance generalizability of ML models for ALWHIV depression.