SmartCGMS as a Testbed for a Blood-Glucose Level Prediction and/or Control Challenge with (an FDA-Accepted) Diabetic Patient Simulation

Abstract

Diabetic patient desires to avoid hypo- and hyperglycemic episodes, which result from insufficient insulin production. As the diabetes disease progresses, it requires an advance control of external insulin administration with an insulin pump. Given the importance of blood-glucose level prediction for the insulin therapy, there is a Blood-Glucose Level Prediction Challenge. This prediction is based on a post-mortem dataset, which include a number of signals related to the daily life of a diabetic patient. We propose replacing these post-mortem signals with an in-silico diabetic patient. For this purpose, we can use the SmartCGMS continuous glucose monitoring and controlling framework together with an FDA-accepted diabetic patient simulation. As a result, a competing researcher have the same conditions as a developer of a real-life insulin pump, connected to a real diabetic patient. When using SmartCGMS, simulated, prototyped and real devices can work together. This approach reduces the difference between laboratory and practical results, thus increasing the level of realism for the entire challenge. As a report on the current SmartCGMS state, we describe the previously unpublished features, which enable an improved glucose level prediction and/or control challenge.