Oral Abstracts From Adherence 2023 Conference.

Abstract

Background: To achieve ending the HIV Epidemic goals, key populations, including sexual minority men, need to adhere to evidence-based biomedical interventions including antiretroviral therapy as treatment (ART) or preexposure prophylaxis (PrEP). The present study integrates traditional and machine learning methods to evaluate whether a common set of factors can predict adherence to ART for both treatment and prevention. Method: Participants included 365 sexual minority men taking antiretroviral therapy as treatment or PrEP in South Florida. Survey respondents provided information on adherence to treat-ment or PrEP and demographic, psychosocial, and behavioral factors potentially associated with adherence. Data were analyzed using machine learning algorithms that are simple to interpret such as Classi fi cation and Regression Tree and LASSO regression variable selection, techniques that require specialized extra steps to look inside “ black box models ” like Multivariate Adaptive Regression Spline (MARS) and Random Forest models, and traditional stepwise logistic regression to identify factors associated with adherence. Results: Taking ART for HIV treatment or PrEP was not an important predictor for adherence in any of the models. Rather, the models suggested that a common set of predictors can be used to predict adherence to ART for both treatment and PrEP. Race/ethnicity was identi fi ed by all models as an important predictor of adherence. Additionally, depressive symptoms, anxiety symptoms, and substance use were identi-fi ed as an important adherence predictor by at least three (of fi ve) models. Determinants less commonly identi fi ed as important for adherence were alcohol use (CART and LASSO only), sexual orientation (CART only), self-esteem, and condomless sex (Random Forest and LASSO only). Conclusion: