Including the Planning of Orally-Ingested Carbohydrates in a Zone Model Predictive Control for Artificial Pancreas

Artificial Pancreas (AP) is a technology that assists people with type-1 diabetes (T1D) in the challenge of administering exogenous insulin in order to maintain the blood glucose concentration in a safe range. In this work, we consider a zone Model Predictive Control (zone-MPC) with asymmetric costs, and we add to it the capability to plan and suggest the assumption of carbohydrates (CHO) to the patient. To produce sparse-in-time and quantized CHO suggestions, a Mixed Integer Programming (MIP) formulation of the MPC problem is considered. The resulting algorithm is tuned and validated on the UVa/Padova T1D Simulator, an accurate model of T1D metabolism, approved by the U.S. Food and Drug Administration (FDA) for pre-clinical testing of T1D therapies. The results show that the zone-MPC + CHO suggestions allows to increase the time spent in the blood-glucose safe range on average, with respect to the insulin-only zone-MPC. This is achieved while consistently cutting down the time in hypoglycemia: from 1.2% to 0.3% in median.