PROJECT SUMMARY People who present with Acquire Immunodeficiency Syndrome (AIDS) are at high risk of developing an opportunistic infection and death. Opportunistic infections (OIs) are any infection that is more frequent or more severe because of HIV-mediated immunosuppression. Different states in the United States (US) have varying rates of deaths among people with HIV (PWH) and AIDS due to variable health insurance coverage rates, demographic health disparities, and access to health care. Data science public health tools to predict disease and poor outcomes are rapidly advancing, but have not yet been sufficiently applied to improve outcomes among PWH. In this K08 Mentored Career Development Award, Dr. Catherine Bielick, a fellow physician in Infectious Diseases at the University of Virginia and rising data scientist, proposes to use artificial intelligence (AI) to 1) predict the change in OI hospitalization rates by simulating Medicaid expansion (ME) in the South and identify associated health inequities, 2) predict the change in OI-related mortality rates by simulating ME and identify associated health inequities, and 3) create a time-series machine learning model to predict poor clinical outcomes for individual PWH. The first two aims will be accomplished using the State Inpatient Database (SID), which is hospitalization-level data for over 97% of hospitals in the US. The South was chosen based on preliminary data finding an association with OI hospitalizations, mortality, and being uninsured. Demographics, diagnosis codes with presence on admission indicators, and hospital information will all be used to simulate the effect of Medicaid expansion in each state and predict the change in OI hospitalization and mortality rates for PWH. The measured impact of this simulated intervention will provide important groundwork to inform progress on state-level social determinants of health (SDOH), the need for health insurance for all PWH, and future cost-effectiveness analyses. The last aim will use multisite longitudinal electronic medical record (EMR) data called the ADVANCE network consisting entirely of underrepresented patient populations from whom this subset of PWH will be the focus. A time-series deep learning model will be used to create a risk score for individual PWH which predicts loss of viral suppression, development of an OI, or all-cause mortality in the following 6 months. Accomplishing this goal will produce a foundational tool on which future machine learning models can optimize model generalizability, safety, privacy, and responsible implementation in an EMR for real-time predictions made at an intervenable time. This proposal benefits from a strong advisory team which includes leading experts in HIV health and policy, data analytics, machine learning, biostatistics, PWH data use, and AI health care. Drawing from the mentorship, collaboration, and support from the Division of Infectious Disease, Public Health Sciences, the McManus ...