Impact of Inaccurate Meal Announcements on Glycaemia Control with or without an Artificial Pancreas.

Type 1 Diabetes (T1D) therapy's goal is to maintain glycaemia in the safe euglycaemic range. Recent technological advancements help the patient in this hard task: Sensor Augmented Pump (SAP) systems combine an insulin pump for continuous subcutaneous insulin infusion with a glucose sensor. Nevertheless, the patient remains in charge of insulin dosing. Artificial Pancreas (AP) is a more advanced technology, where a closed-loop control algorithm automatically adjust the pump insulin infusion based on the sensor feedback. With SAP and with most of the available AP systems the patient is requested to estimate the carbohydrates (CHO) content of a meal and to announce it to the system. However, human-provided information about the upcoming meal is prone to various errors, including: CHO estimation error, announcement delay and missed meal announcement. In this work, we compare the robustness of SAP and an AP therapy in handling meal-related errors, investigating (in-silico) and quantifying how much each of the three previous factors influences glycaemia control. In particular, we adopt an AP based on a predictive control algorithm (Model Predictive Control, MPC). While both SAP and the MPC-based AP experience a progressive performance degradation as human errors increase, the AP exhibits a more gradual performance decline, confirming the expected superior robustness with respect to SAP and providing a quantitative analysis of its magnitude.Clinical relevance- Evidences on the impact of human errors on glycemic control achieved by SAP and AP in T1D management might be translated into better guidelines for clinical adoption, could allow clinicians to perform effective cost-benefit evaluation and could facilitate the definition of target populations that benefit the most from each treatment.