"Learning" Can Improve the Blood Glucose Control Performance for Type 1 Diabetes Mellitus.

Abstract Background: A learning-type artificial pancreas has been proposed to exploit the repetitive nature in the blood glucose dynamics. We clinically evaluated the efficacy of the learning-type artificial pancreas. Methods: We conducted a pilot clinical study in 10 participants of mean age 36.1 years (standard deviation [SD] 12.7; range 16–58) with type 1 diabetes. Each trial was conducted for eight consecutive mornings. The first two mornings were open-loop to obtain the individualized parameters. Then, the following six mornings were closed-loop, during which a learning-type model predictive control algorithm was employed to calculate the insulin infusion rate. To evaluate the algorithm's robustness, each participant took exercise or consumed alcohol on the fourth or sixth closed-loop day and the order was determined randomly. The primary outcome was the percentage of time spent in the target glucose range of 3.9–8.0 mmol/L between 0900 and 1200 h. Results: The percentage of time with glucose spent i...