Abstract
Mental health conditions and substance use are prevalent among people with HIV (PWH), are correlated with one another, and associate with viral non-suppression independently; their joint association with viral non-suppression may be under-studied because of data sparsity. We conducted a machine learning-based clustering analysis to characterize groups of patient-reported mental health symptoms and substance use based on their relationship with HIV viral suppression. Participants in the Johns Hopkins HIV Clinical Cohort reported symptoms of depression, anxiety, and post-traumatic stress, and recent use of alcohol, cocaine, amphetamine, non-prescribed opioids, and cannabis (2013-2023). We fit a random forest model with the viral suppression status as the outcome against self-reported items as predictors and used a forest-derived similarity measure to group participants into three clusters. The cluster with the lowest viral suppression rate (74.5%) had the highest depression symptom score (median score 4, interquartile interval [IQI] 1-8) and anxiety symptom score (median score 2, IQI 0-7) along with the greatest prevalence of recent cocaine (99.9%) and opioid (28.0%) use. The cluster with the highest HIV viral suppression rate (81.1%) had the lowest depression symptom score (median 1, IQI 0-4) and anxiety symptom score (median 0, IQI 0-2) and lowest proportion of recent cocaine (0%) and opioid (2.5%) use. Clinically meaningful groups of PWH with heterogenous mental health and substance use characteristics were formed using a machine learning-based clustering approach. PWH with mental health symptoms and substance use represent an important subpopulation for interventions to improve antiretroviral treatment outcomes.