Fully Automated Artificial Pancreas for Adults with Type 1 Diabetes using Multiple Hormones: Exploratory Experiments

ABSTRACT Objective A fully-automated insulin-pramlintide-glucagon artificial pancreas that alleviates the burden of carbohydrate counting without degrading glycemic control was iteratively enhanced until convergence through pilot experiments on adults with type 1 diabetes. Methods Nine participants (age 37±13 years, HbA1c 7.7±0.7%) completed two 27-hour interventions: a fully automated multi-hormone artificial pancreas and a comparator insulin-alone artificial pancreas with carbohydrate counting. The baseline algorithm was a model-predictive- controller that administered insulin and pramlintide in a fixed ratio, with boluses triggered by a glucose threshold, and administered glucagon in response to low glucose levels. Results The baseline multi-hormone dosing algorithm resulted in non-inferior time in target range (3.9-10.0 mmol/L) (71%) compared to the insulin-alone arm (70%) in two participants, with minimal glucagon delivery. The algorithm was modified to deliver insulin and pramlintide more aggressively to increase time in range and maximize the benefits of glucagon. The modified algorithm displayed a similar time in range for the multi-hormone arm (79%) compared to the insulin-alone arm (83%) in two participants, but with undesired glycemic fluctuations. Subsequently, we reduced the glucose threshold that triggers glucagon boluses. This resulted in inferior glycemic control for the multi-hormone arm (81% vs 91%) in two participants. Thereafter, a model-based meal detection algorithm to deliver insulin and pramlintide boluses closer to mealtimes was added and glucagon was removed. The final dual-hormone system had comparable time in range (81% vs 83%) in the last three participants. Conclusion The final version of the fully-sautomated system that delivered insulin and pramlintide warrants a randomized control trial.