A Bilevel Optimization Approach to Integrating Personalized Treatment with Healthcare Resource Planning

This award will contribute to the advancement of the national health and welfare by developing innovative methods to personalize the treatment of diabetic retinopathy (DR), the leading cause of blindness in American adults. Despite the increasing availability of emerging therapies that enable various effective DR treatment options, current clinical guidelines continue to rely on a one-size-fits-all approach. This award supports the development of data-driven methods that can lead to better utilization of emerging treatment techniques while addressing the unique clinical, systemic, and social needs of individual patients. With diabetic patients accounting for more than 10 percent of the US population, the successful establishment of personalized DR treatment guidelines will directly benefit numerous individuals. This project is expected to have a particular impact on the health and welfare of underrepresented patients by promoting equitable access to treatment. This award will fund a doctoral student in interdisciplinary research through multilateral collaboration and create educational materials for both operations research and healthcare professionals. This research aims to develop personalized DR treatment guidelines through a bilevel programming approach, which integrates two key perspectives on treatment design: (i) optimizing patient-specific treatment schedule over time through constrained, multi-agent Markov decision processes at the lower level, and (ii) determining eye clinic-level operational interventions to enhance resource allocation and provide personalized access to care at the upper level. The complex interaction between eye clinics and patients will be approximated by a large-scale mixed-integer program, which will be solved using decomposition methods and patient-specific heuristics calibrated with data from the electronic medical record system and a patient simulation model. Computational challenges in the lower-level problems will be addressed via dimensionality reduction techniques exploiting a unique solution structure and inverse optimization techniques identifying model parameters that facilitate easy-to-compute decision rules. This research will lead to advancements in the modeling and solution approaches for bilevel, multi-agent sequential decision-making. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.