The increasing uptake of continuous glucose monitoring (CGM) use offers a business opportunity, with an estimated CGM usage in 2021 of 4M users worldwide, mainly in USA and Europe. These estimates consider not only type 1 diabetes (T1D), either on insulin pens (MDI) or pump therapy, but also type 2 diabetes (T2D) patients requiring insulin therapy. Expected growth will continue with double digits beyond 2030, with the expansion of coverage for T2D and Asian markets like China and Japan, currently with relatively low usage of CGM but with very significant market potential.
CGM not only paved the way to current advanced pumps such as the artificial pancreas, but also to supervision and decision support systems (DSS) directed also to MDI users, the vast majority of patients (including T2D under insulin therapy).
Tools for patient supervision and decision support are still embrionary are there is still need for improved systems minimizing the risk of hypoglycemia. Besides, they lag behind the needs of people with T1D since they are mainly directed to pump users and/or require intensive manual intervention by patients for data collection which algorithms are based on, in detriment of usability and inducing low compliance. Usability concerns have been raised by developers of state-of-the-art DSSs due to data collection burden.
This project aims at advancing Technological Readiness Level of results from the research team in the scope of learning-based prediction systems to support insulin therapies in type 1 diabetes, addressing both machine learning tools for prediction-based decision support (Subproject 2 by University of Girona, UdG) and glucose prediction without manual input data collection (Subproject 1 by Universitat Politècnica de València, UPV). As outcome, a product prototype and demonstrator per subproject will be built.
This Proof of Concept project proposal stems from the project Solutions for the improvement of efficiency and safety of the artificial pancreas by fault-tolerant multivariable control architectures mSAFE-AP (DPI2016-78831-C2-1-R), composed of two subprojects by UPV and UdG. The mSAFE-AP project aimed at the design of an efficient and safe artificial pancreas in normal free-living use, by means of new multivariable reconfigurable fault-tolerant control architectures. As part of the objectives, the development of new tools for patients supervision, including the classification and detection of free-living scenarios was addressed, in order to provide with a set of tools to warn an artificial pancreas system about risks and automatically mitigate their effects through the modulation of the control algorithm. In this sense, the extension of the concept of fault beyond instrumentation to include patients anomalous metabolic states and human factors, as well as the development of advanced methods for an improved prediction of glucose and risks associated to insulin dosing was addressed.
This objective was developed in three tasks addressing the development of classifiers of the metabolic state and behavior of the patient, glucose predictors with increased prediction accuracy for longer prediction horizons and minimal user input, and risk predictors at different time scales based on machine learning techniques, giving rise to the results associated to this proposal.
LEAP project will bring these technologies closer to the market, ultimately advancing towards new personalized therapies for diabetes.