Subspace-based Model Identification of Diabetic Blood Glucose Dynamics

Abstract

Diabetes Mellitus is a chronic desease characterized by the inability of the organism to autonomously regulate the blood glucose level due to insulin deficiency or resistance, thus leading to serious health damages. In order to keep tight glucose control treatment happens to be based essentially on insulin delivery. In common practise, a main limiting factor in adequate choice of insulin dosing is the lack of reliable predictions of blood glucose evolution. In this work, state-space models of various orders were investigated and evaluated for their capability of description of one diabetic subject blood glucose evolution over 24 hours. It turned out that models of low complexity are capable of detecting even abrupt changes in calibration data but they are not sufficient for validation data. Blood glucose levels could be predicted fairly well up to 30 minutes ahead on validation data.

Glycemic trend prediction using empirical model identification

“…Very recently the problem of identifying such a model has been tackled from a data-driven perspective mainly using simulated data from model in the literature [22], [23]. Indeed, fitting actual T1DM subject data to the models has been treated to a much less extent (e.g., [11], [24], [25], [26]) given the difficulties in gathering appropriate patient records. This paper dealt with a unique dataset which is being collected [12] and looks promising for future exploitation.…”

“…As far as meal modeling is concerned, the proposed strategy in [11] seems to be a theoretically sound attempt to account for differences in meal compositions. However, it was not possible to apply it in this context due to unavailable meal composition information.…”

“…In this paper, given the impossibility of identifying and ThA11.1 consequently adapting the above mentioned parameters to the subject due to lack of data, mean population parameters values were considered, as introduced in [11]. As far as the insulin is concerned, despite the fact that there are quite well-known models for insulin dynamics in the literature [16], [17], no modeling for subcutaneous insulin absorption was embraced.…”

“…Purpose of this paper is to expand on [11] and investigate individual-specific predictive models from T1DM patient records. Two strategies were approached: Firstly, Kalman estimators based on identified state-space models were designed; Secondly, direct least-squares identification of various order ARX-and ARMAX-based predictors was done.…”

Adaptive subspace-based prediction of T1DM glycemia

“…However, none of these works considered the dynamic interplay between previously injected insulin, meal intake and eventually exercise. Attempts at including in the prediction model insulin and carbohydrates can be found in [7], [8], [9], [10], [11] and [12]. A different approach is that of [13] and [14] focusing on insulin sensitivity prediction and consequently on the resulting blood glucose prediction by means of stochastic models for critically ill patients.…”