Model Identification And Blood Glucose Control For Patients With Type 1 Diabetes:A Clinical Research

The past decade has witnessed a sharp increase in the number of individuals with type 1 diabetes mellitus (T1DM) worldwide. Patients with T1DM have no endogenous insulin secretion because of the autoimmune destruction of pancreatic islet β cells and need intensive insulin therapy. Glucose control that maintains blood glucose concentrations within the normal range can suppress the development and progression of microvascular and cardiovascular complications and increase the life expectancy of patients with T1DM.There is no cure for T1DM, the only way to decrease blood glucose level is insulin infusion. Artificial pancreas, which can deliver insulin continuously, utilizes the real-time blood glucose concentrations, so it is closed-loop control. Hence, it can significantly decrease the frequency of hypoglycemia compared with conventional daily multiple subcutaneous insulin infusion therapy. Many studies have proposed various closed-loop control algorithm, some of which have been evaluated in clinical trials.A closed-loop glycemic control trial has been designed in this paper, ten patients were enrolled and completed the whole trial. Customized model and control parameters were designed for each patient based on the glucose data. Then a learning-type model predictive control (L-MPC) algorithm was used to control the blood glucose concentration. Based on the control performance on the previous day, one can adjust the control setpoint to improve the control performance day by day. The statistical results validated the feasibility and effectiveness of the learning-type artificial pancreas.Considering the influence of daily exercise and alcohol intake, we also imposed exercise and alcohol disturbance to patients. The clinical results showed the learning-type AP had good robustness with respect to disturbances.Previous studies on glucose-insulin model considered the time delay of insulin absorption to be constant. Considering the fact that the time lag for insulin absorption of one patient could change over a timescale, it is assumed that the time-varying delay of insulin input follow a Markov chain. By using the expectation-maximum (EM) algorithm, the coefficients of input and output and the time-varying delay were obtained. The accuracy of the model identified by using EM algorithm outmatched that by using least square algorithm.