Development and External Validation of a Risk Calculator to Predict 'poor Mental Health' Outcomes Among Youths Affected By HIV in Uganda

Background and Purpose

Adolescents orphaned by AIDS (AoAIDS) and Adolescents living with HIV (ALHIV) – also referred to as AIDS-affected Youth – are at higher risk of experiencing poor mental health than non-orphans and HIV-negative youth. Poor Mental health is a key risk factor for poor health functioning, morbidity and suicide. Yet, because most AIDS-affected youth tend to reside in poor and low resourced communities, they lack opportunities for clinical mental health assessment. Thus, there is a need for alternative non-clinical tools to predict which AIDS-affected youths are more likely to develop poor mental health functioning. We developed and externally validated a model to predict individualized risk of poor mental health (measured by depression and/or hopelessness) among AIDS-affected youths in Uganda, a country heavily affected by HIV/AIDS, poverty and low-resourced communities.

Methods

Longitudinal data from AoAIDS in Uganda was used to develop our predictive model. Penalized logistic regression using the Least Absolute Shrinkage and Selection Operator (LASSO) regularization technique was used to select a subset of individual, familial, health, environmental and socio-economic factors that predicted poor mental health in a sample including individuals with a history of depression and/or hopelessness, and a separate sample without a reported history of depression and/or hopelessness. The model was externally validated in a separate longitudinal sample of ALHIV.

Results

Best predictors for poor mental health in the total AoAIDS sample were gender, family cohesion, social support network, asset ownership, recent diagnosis of sexually transmitted disease (STD), physical health rating, and previous poor mental health; Area Under the Curve (AUC)=72.2; 95% CI =67.9-76.5. For AoAIDS without reported history of poor mental health, the AUC=69.0, 95% CI=63.4-74.6, and was best predicted by gender, drug use, social support network, assets ownership, recent STD diagnosis, and rating of physical health. Both models were well calibrated. Performance of the model after external validation in ALHIV sample with previous reported history of poor mental health was similar (AUC=69.7; 95% CI=64.1-75.2).

Conclusions and Implications

The model predicted individualized risk of new and future onset of depression and/or hopelessness among AIDS-affected youths reasonably well and showed good generalizability. The model may offer opportunities for the design of social work and public health interventions aimed at addressing depression and hopelessness among AoAIDS and ALHIV. Further model extension in other vulnerable AIDS-affected youth populations is recommended to improve predictions.