Predicting Neurobehavioral Outcomes in People with HIV

We aimed to identify complex, multidimensional, longitudinal biopsychosocial phenotypes (MLBPSPs) in people with HIV (PWH) and evaluate their associations with baseline clinical characteristics. We included 506 PWH in the multi-site CHARTER study who underwent assessments at four visits, six months apart. Using machine learning, we identified four MLBPSP clusters based on means and non-linear trajectories of biopsychosocial characteristics. These characteristics included neurocognition, depressed mood, self-reported cognitive symptoms, and activities of daily living at each visit. The largest MLBPSP cluster (C1, N = 231) had the best average scores across all domains and remained stable over 18 months of follow-up. Other clusters showed varying degrees of cognitive impairment, depressed mood, and functional disability. In multivariable analyses, several baseline clinical characteristics, including chronic pulmonary disease, distal neuropathic pain, polypharmacy, and creatinine levels, significantly predicted one or more adverse MLBPSP trajectories. These findings have implications for HIV care by identifying PWH at risk for future adverse trajectories. The results may lead to insights informing future personalized interventions targeted to vulnerable subpopulations of PWH.