Modeling and Prediction in Diabetes Physiology

Diabetes is a group of metabolic diseases characterized by the inability of the organism to autonomously regulate the blood glucose levels. It requires continuing medical care to prevent acute complications and to reduce the risk of long-term complications. Inadequate glucose control is associated with damage, dysfunction and failure of various organs. The management of the disease is non trivial and demanding. With today’s standards of current diabetes care, good glucose regulation needs
constant attention and decision-making by the individuals with diabetes. Empowering the patients with a decision support system would, therefore, improve their quality of life without additional burdens nor replacing human expertise. This thesis investigates the use of data-driven techniques to the purpose of glucose metabolism modeling and short-term blood-glucose predictions in Type I Diabetes Mellitus (T1DM). The goal was to use models and predictors in an advisory tool able to produce personalized short-term blood glucose predictions and on-the-spot decision making concerning the most adequate choice of insulin delivery, meal intake and exercise, to help diabetic subjects maintaining glycemia as close to normal as possible. The approaches taken to describe the glucose metabolism were discrete-time and continuous-time models on input-output form and statespace form, while the blood glucose short-term predictors, i.e., up to 120 minutes ahead, used ARX-, ARMAX- and subspace-based prediction.