"Despite substantial advances in antiretroviral therapy (ART), many people living with HIV (PLWH) continue to experience intermittent viremia, viral rebound, incomplete immune recovery, and increased chronic disease burden despite treatment availability. Current HIV care approaches frequently rely on isolated clinical measurements and reactive management strategies that may inadequately capture evolving patterns of HIV treatment experiences over time. Earlier identification of individuals at elevated risk for adverse outcomes could facilitate targeted interventions and improve long-term health outcomes.
The study will leverage the rich longitudinal structure of the MACS/WIHS Combined Cohort Study (MWCCS), one of the largest and most comprehensive HIV cohort studies in the United States, to develop an explainable artificial intelligence (XAI) framework for longitudinal prediction of HIV treatment instability and immune non-recovery. The central hypothesis is that longitudinal temporal patterns in virologic, immunologic, behavioral, psychosocial, and treatment related characteristics can be integrated using explainable machine learning approaches to generate clinically interpretable predictions of future HIV treatment instability.
Specific aims are to: (1) develop longitudinal representations of HIV treatment experiences using repeated clinical, behavioral, psychosocial, and treatment-related measures; (2) develop and evaluate explainable longitudinal machine learning models for predicting future treatment instability and immune non-recovery; and (3) evaluate clinical utility through risk stratification and assessment of translational potential.
Unlike traditional regression approaches that primarily estimate average population-level associations, the proposed framework will integrate longitudinal information to capture nonlinear relationships, temporal dependencies, and individualized risk patterns while maintaining clinical interpretability through explainable AI methods. Expected outcomes include development of longitudinal HIV treatment representations, validated predictive models, clinically interpretable patient-level risk profiles, and preliminary evidence supporting future precision HIV care applications. Findings from this pilot study will generate preliminary data and computational pipelines supporting future NIH R21 and R01 applications focused on explainable AI and predictive HIV care."
Project Summary provided by investigator.