Personalized insulin therapy optimization: Guided multi-objective optimization for clinically relevant decision support

Despite significant progress in automated insulin delivery systems, therapy optimization for individuals with type 1 diabetes (T1D) remains limited by generalized strategies that overlook personal preferences and risk tolerances. Building on our prior work, this study presents an enhanced multi-objective decision support system that uses the University of Virginia Replay Simulator (UVA-RS) with improved optimization techniques to generate viable, patient-specific therapy configurations. To reflect clinical priorities, we influence the optimization toward a preferred region of interest, supporting the discovery of balanced and actionable solutions. To make the decision space more navigable, we implement a sparsification strategy that presents a streamlined yet diverse set of therapy configurations. Simulation studies using the adult cohort of the UVA/Padova T1D Simulator show that the penalized formulation increases the proportion of solutions meeting clinical targets, while the representative set captures distinct therapeutic strategies with meaningful differences in glycemic outcomes. In replay simulations, all optimized therapies substantially outperformed baseline control, increasing Time in Range from 70.5% to 87.3%–88.8% while reducing both hyperglycemia and hypoglycemia. These enhancements aim to empower patients and healthcare providers to collaboratively select insulin therapy strategies that align with individual priorities, risk profiles, and lifestyle considerations.