Artificial Intelligence and Counterfactually Actionable Responses to End HIV (AI-CARE-HIV)

ABSTRACT This work responds to the current federal and the NIH Office of AIDS Research priority areas, diagnose–treatprevent–respond, to Ending the Human Immunodeficiency Virus (HIV) Epidemic, focusing on Florida, which has the highest incidence of HIV infections in the US. About 40% of people with HIV in Florida do not reach undetectable viral load, and certain population clusters—such as residents of rural areas, all facing barriers to timely care—experience higher risk of unfavorable outcomes. Besides well-known risk factors contributing to drops in care retention and ART success, part of such remains unexplained and cannot be actioned upon. Advances in artificial intelligence (AI) and increasing availability of large real-world data (RWD), e.g., electronic health records (EHRs) and claims data, are ideal for developing models for precision health. However, the full capabilities of AI are still hampered by the fact that EHRs are not well integrated with other relevant data sources, containing information on behavioral and contextual factors, key drivers of HIV care continuum progression. Notably, circumstances such as alcohol or substance use and domestic violence are infrequently represented in structured EHRs, but can be systematically identified from clinical narratives using natural language processing (NLP). Another critical problem with AI built on RWD is that, due to inherent systematic errors due to confounding or selection in observational data like EHRs, the models might identify wrong effects for interventions. Thus, alternative predictions (i.e., counterfactuals) of naïve AI systems can be mistaken, potentially leading to harm. Causal inference methods can be coupled with AI to address RWD data issues. The overarching goal of this project is to develop “AI-CARE-HIV,” an actionable counterfactual AI framework to improve HIV outcomes in Florida, through multimodal model augmentation that includes patient circumstances and ecological measurements. This framework can be used to develop a causally-sound (under certain assumptions), actionable model usable for planning and implementing clinical, public health interventions. We will develop the project through the OneFlorida+ Clinical Research Consortium, which collates RWD data from >16.8M Floridians, and specifically the OneFlorida+ HIV cohort (N=71,363). Our project aims to: (1) Enhance the cohort by incorporating large-scale socioeconomic, ecological factors (9,000+) and prospectively validate new ones using NLP, including stigma and behavioral circumstances; (2) Create multimodal risk scores from clinical, behavioral, socioeconomic and ecological factors, identify population-level causal effects for candidate interventions apt to improve care retention, develop individualized counterfactual AI models for HIV outcomes; (3) Plan–with healthcare providers, State officials, citizen scientists–actual clinical, public health interventions anchored on counterfactual AI, using implementation science and standardized protocols. Our team includes multidisciplinary expertise supported by OneFlorida+, the Fl Dept of Health, and community entities. We expect impact at multiple levels, from infrastructure enhancement to public health benefit.