SCH: Advancing Public Health Intervention with Data-driven Restless Multi-Armed Bandit Framework - Public health interventions are essential in mitigating the burden of both infectious and chronic diseases, including diabetes and cardiovascular conditions. Early interventions, such as timely medication, lifestyle modifications, and the dissemination of health information, can significantly improve health outcomes. However, optimizing these interventions remains a major challenge due to limited healthcare resources, variation in patient disease phenotypes and treatment response, and the complex interplay of clinical, behavioral, neighborhood, and healthcare-access factors. Identifying patients at elevated clinical risk and efficiently allocating healthcare resources requires advanced computational approaches that integrate real-world data and adaptive decision-making frameworks. This proposal focuses on developing data-driven restless multi-armed bandit (RMAB) techniques to optimize intervention strategies for diabetes, cardiovascular conditions, and maternal health in the U.S. healthcare system. Due to the complex nature of these diseases, we propose a contextual RMAB model that integrates patient context and datasets, including observational EHR data (e.g., MIMIC-III, MIMIC-IV, All of Us, and MGB Biobank/RPDR) and intervention trials and cohort studies (e.g., REAL HEALTH-Diabetes, Look AHEAD, DPP), to allocate interventions based on disease progression and patient responses. We also propose a network RMAB model to represent peer/contact relationships and measured neighborhood and healthcare-access factors, and a multi-agent RMAB model to optimize intervention strategies across decentralized healthcare providers. These models will enable more precise and efficient allocation of interventions by incorporating patient connectivity, healthcare facility constraints, and policy-driven incentives, bridging the gap between computational decision-making and real-world healthcare applications. By leveraging machine learning, network analysis, and health economics, this proposal will develop scalable and interpretable AI-driven frameworks for optimizing healthcare resource allocation. Collaborating with hospitals, universities, non-profit organizations, and government agencies, this project will support the transition of research-driven innovations into clinical practice, facilitating evidence-based decision-making to improve intervention targeting and patient outcomes at scale.
