Predicting Glycemia in Type 1 Diabetes Mellitus with Subspace-Based Linear Multistep Predictors

Abstract

A major challenge for a person with diabetes is to adapt insulin dosage regimens and food intake to keep blood glucose within tolerable limits during daily life activities. The accurate prediction of blood glucose levels in response to inputs would support the patients with invaluable information for appropriate on-the-spot decision making concerning the management of the disease. Against this background, in this paper we propose multistep data-driven predictors to the purpose of predicting blood glucose multiple steps ahead in the future. We formulate the predictors based on the state-space construction step in subspace identification methods for linear systems. The clinical data of 14 type 1 diabetic patients collected during a 3-days long hospital visit were used. We exploited physiological models from the literature to filter the raw information on carbohydrate and insulin intakes in order to retrieve the inputs signals to the predictors. Predictions were based on the collected CGMS measurements, recalibrated against finger stick samples and smoothed through a regularization step. Performances were assessed with respect to YSI blood glucose samples and compared to those achieved with a Kalman filter identified from data. Results proved the competitiveness of the proposed approach.