Closed-loop MPC-based artificial pancreas: Handling circadian variability of insulin sensitivity

Closed-loop glycemic control algorithms have demonstrated the ability to improve glucose regulation in patients with type 1 diabetes mellitus (T1DM), both in silico and clinical research trials. However, many of the proposed control strategies have been developed under time-invariant linear models, without considering the well-know intraday parametric variability. In this work, a recently published pulsatile Zone Model Predictive Control (pZMPC) is analyzed under a real observed pattern of intraday insulin sensitivity (SI ), according to the latest updates of the FDA-approved UVA/Padova simulator. To this end, nominal information of the postprandial insulin sensitivity is explicitly included in the control-oriented model to be taken into account in the predictions. The proposed control strategy is evaluated in silico under the FDA-approved UVA/Padova simulator, the performance analysis is done using statistical metrics and it is compared with the invariant version of the aforementioned controller. Results show a significant improvement in the analyzed metrics when the time-variant control-oriented model is considered in the predictions, which justifies the proposed increment in the controller complexity.