Model-based personalization scheme of an artificial pancreas for Type 1 diabetes applications

Automated controllers designed to regulate blood glucose concentrations in people with Type 1 diabetes mellitus (T1DM) must avoid hypoglycemia (blood glucose <;70 mg/dl) while minimizing hyperglycemia (>180 mg/dl), a challenging task. In this paper, a model-based control design approach with a personalized scheme based on readily available clinical factors is applied to a linearized control-relevant model of subject insulin-glucose response profiles. An insulin feedback strategy is included with specific personalization settings and variations in a tuning parameter, τ_c. The control strategy is challenged by an unannounced meal disturbance with 50 g carbohydrate content. A set of metrics are introduced as a method of evaluating the performance of different controllers. In-silico simulations of ten subjects in the Food and Drug Administration accepted Universities of Virginia and Padova metabolic simulator indicate that the personalization strategy with a τ_c setting of 270 minutes gives very good controller performance. Post-prandial glucose concentration peaks of 183 mg/dl were achieved with 97% of the total simulation time spent within a safe glycemic zone (70-180 mg/dl), without hypoglycemic incidents and without requiring a time-consuming model identification process.