Longitudinal latent class and joint modelling of antiretroviral adherence

Mcduling, Campbell

Longitudinal latent class and joint modelling of antiretroviral adherence

Thesis / Dissertation

2025

Publisher

University of Cape Town

Department

Department of Statistical Sciences

Faculty

Faculty of Science

Abstract

This dissertation investigates the relationship between antiretroviral therapy (ART) adherence and viral outcomes in people living with HIV in South Africa using advanced statistical modeling techniques. Utilizing data from the ADD-ART study, a prospective cohort of 238 adults on ART in Cape Town, the research employs survival analysis, joint modeling, and longitudinal latent class analysis to compare di↵erent adherence monitoring tools and examine heterogeneity in adherence behaviors. Key findings include: Electronic Adherence Monitoring (EAM) and tenofovir diphosphate levels in dried blood spots were more strongly associated with viral non-suppression than self-reported adherence; joint modeling revealed a stronger association between EAM adherence and viral outcomes compared to traditional survival models, with each additional missed dose in the preceding 30 days associated with an 81% increase in the hazard of viral non-suppression; longitudinal latent class analysis identified five distinct adherence trajectory groups, with poorer or declining adherence groups experiencing significantly higher rates of viral non-suppression; baseline viral load and prior tuberculosis exposure were significant predictors of subsequent viral non-suppression, even after accounting for adherence. The results highlight the importance of using objective adherence measures, the value of advanced statistical techniques in HIV research, and the need for personalized adherence support strategies. Limitations include potential violations of model assumptions and generalizability constraints.

Keywords

statistical science

Reference:

Collections

Masters

Full item page