A Multi-Agent Approach Used to Predict Long-Term Glucose Oscillation in Individuals with Type 1 Diabetes

Abstract

The glucose–insulin regulatory system and its glucose oscillations is a recurring theme in the literature because of its impact on human lives, mostly the ones affected by diabetes mellitus. Several approaches have been proposed, from mathematical to data-based models, with the aim of modeling the glucose oscillation curve. Having such a curve, it is possible to predict when to inject insulin in type 1 diabetes (T1D) individuals. However, the literature presents prediction horizons of no longer than 6 h, which could be a problem considering their sleeping time. This work presents Tesseratus, a model that adopts a multi-agent approach used to combine machine learning and mathematical modeling to predict the glucose oscillation for up to 8 h. Tesseratus can support endocrinologists and provide personalized recommendations for T1D individuals to keep their glucose concentration in the ideal range. It brings pioneering results in an experiment with seven real T1D individuals. Using the Parkes error grid as an evaluation metric, it can be depicted that 93.7% of measurements fall in zones A and B during the night period with MAE 27.77 mg/dL. It is our claim that Tesseratus will be a reference for the classification of a glucose prediction model, supporting the mitigation of long-term complications in the T1D individuals.

Keywords

• glucose oscillation
• prediction
• multi-agent
• type 1 diabetes
• personalized
• recommendation