A hybrid model to predict glucose oscillation for patients with type 1 diabetes and suggest customized recommendations

Abstract

The number of adults with Diabetes Mellitus has quadrupled in the last few years around the world, and its prevalence is steadily increasing. Several mathematical and agent-based models have been proposed to shape the human glucose-insulin regulatory system and its ultradian oscillations. However, current models do not have the ability to customize the prediction for each patient and to propose corrections in programmed insulin doses. In order to contribute to the improvement of these mechanisms, this work presents a model that combines both, mathematical and agent-based models to predict glucose oscillation and build customized oscillation profiles of glucose concentration in the patient bloodstream. This hybrid model uses the patient inputs, such as physical activity, insulin, as well as food intake to customize the patient profile. The reactive agents deal with the glucose concentration, the amount and type of insulin, the time and type of physical activity, as well as the amount of carbohydrates ingested. Moreover, the deliberative agent receives information from a continuous compartmentalized model of the human glucose-insulin regulatory system and interact with reactive agents to provide customized information to the patient. A proof of concept is provided based on a specific patient with type 1 diabetes. Initially, the dataset is composed of information from one patient and from a Continuous-Glucose Monitor, measuring every 5 min. The periodic data acquisition may improve the agents’ learning process and the deliberation about actual information that will re-feed the ordinary differential equations which make up the hybrid model.