A Novel Realistic Simulation Framework for Rigorous Validation of Insulin Therapies: integrating stochastic and data-driven variability

Preclinical evaluations of glycaemic control algorithms often rely on simulators that underestimate real-world metabolic variability, leading to overly optimistic performance assessments. A novel type 1 diabetes (T1D) simulation framework, DT1-UAN v2, was developed to generate physiologically realistic and adaptive scenarios for an improved assessment of insulin therapies. DT1-UAN v2 integrates two disturbances of intrapatient vari-ability: eight stochastically parameterised insulin sensitivity (SI) patterns and sixty data-driven profiles for the rate of blood glucose appearance (Ra). Incorporating these features exposed vulnerabilities in insulin therapies that previously relied on static physiological parameters. In silico trials on virtual T1D subjects, under both open-and closed-loop therapies, revealed contrasts between outcomes obtained with DT1-UAN v2 and the earlier DT1-UAN v1. One particular test showed that a well-known closed-loop control therapy produced inferior perfor-mances (− 17.57% in TIR or + 121.78% in TAR) under a set of realistic and altered conditions defined in v2 compared to v1. Control strategies that appeared effective in non-realistic scenarios often produced different results in the enhanced framework, uncovering underestimated risks of post-prandial hypoglycaemia or sus-tained hyperglycaemia. These findings demonstrate that incorporating realistic SI and Ra variability is essential for rigorous preclinical testing, as simplified models can conceal clinically relevant risks. The proposed frame-work provides a valuable tool for identifying hidden design flaws and enabling the development of more adaptive, and safer control strategies for people with T1D. This enhanced realism provides a critical foundation for testing and optimising personalised insulin treatments.