Machine learning-based profiles of atherosclerosis to predict disease outcomes in older HIV-infected women and men

7. PROJECT SUMMARY/ABSTRACT Recent applications of machine learning methods to epidemiologic data have led to discoveries of novel disease profiles with implications for clinical treatment and practice, including novel phenotypes of heart failure. Machine learning is attractive for large data sets because its algorithms can incorporate many more variables than standard regression methods, take into account temporality and other types of multidimensionality, and easily incorporate non-linearity and multiple interactions. Atherosclerosis is an ideal condition in which to apply machine learning methods, because its pathogenesis is known to be multifactorial, but precise causes are still not fully understood. In the context of HIV infection, its pathogenesis is further complicated by multiple comorbidities, potential adverse effects of antiretroviral therapy (ART), and HIV itself, mediated by immune activation and inflammation even in the context of long-term suppressive ART. We plan to apply the innovation of machine learning to discover new phenotypes of HIV-associated atherosclerosis that may reflect distinct pathways and affect clinical outcomes. Our Women?s Interagency HIV Study (WIHS) and Multicenter AIDS Cohort Study (MACS) cohorts are two of the most well-established epidemiologic studies of HIV infection in existence. Each cohort has conducted serial imaging of atherosclerosis since 2002 in over 3,000 participants, and across 20-30 years of follow-up, cohort members have each contributed thousands of demographic, clinical, and laboratory data points. Most cohort members are now 50-59 years of age, while substantial numbers are in their sixth or seventh decade of life, allowing us to examine disease features and risk factors across the age spectrum. We propose to use machine learning methods to develop novel phenotypes of atherosclerosis among HIV-infected WIHS women and MACS men (Aim 1), and then determine the utility of these phenotypes by assessing their associations with cardiovascular disease mortality and other age-related clinical and functional outcomes, including traditional ?geriatric? outcomes (Aims 2-3). Individuals with HIV infection have evidence of accelerated aging; prior studies have typically evaluated isolated outcomes despite the fact that aging is inherently a multifactorial and complex process. By using machine learning methods to identify novel phenotypes of atherosclerosis that may require different preventive or treatment interventions, this project has the potential to lead to new insights into the role of aging in cardiovascular disease pathogenesis among HIV-infected adults.