"Today, a diabetic patient must follow a very constraining therapy with many blood glucose measurements, dose calculations and insulin injections. The ideal would be to have a single system that can conduct all of these operations without requiring any intervention", says Prof. Peter Diem, Chairman and Head of the Department of Endocrinology, Diabetes and Clinical Nutrition at Inselspital. Such a system needs to continuously measure and calculate the level of glucose present in the blood and to constantly deliver insulin via an infusion pump. The amount of infused insulin is determined by an algorithm that estimates the patient needs based on the measured glucose levels, the time of day or the expected activities and that will adapt the pump infusion rate accordingly.

Insulin need depends on daily activity and food profile

"Approaches taken so far do not resolve fundamental difficulties: the patients' variability, uncertainties related to system disturbances, e.g. food intake and physical activity, and errors related to the used devices,” says Dr. Stavroula Mougiakakou, Head of the Diabetes Technology Research Group at the University of Bern’s ARTORG Center. “The proposed algorithm is easy to use, introduces the concept of real-time personalisation based on reinforcement learning, is able to tackle inter- and intra-patient variability, and can compensate for the effects of uncertain events."