Online prediction of subcutaneous glucose concentration for type 1 diabetes using empirical models and frequency‐band separation

Abstract

Online glucose prediction which can be used to provide important information of future glucose status is a key step to facilitate proactive management before glucose reaches undesirable concentrations. Based on frequency-band separation (FS) and empirical modeling approaches, this article considers several important aspects of on-line glucose prediction for subjects with type 1 diabetes mellitus. Three issues are of particular interest: (1) Can a global (or universal) model be developed from glucose data for a single subject and then used to make suitably accurate on-line glucose predictions for other subjects? (2) Does a new FS approach based on data filtering provide more accurate models than standard modeling methods? (3) Does a new latent variable modeling method result in more accurate models than standard modeling methods? These and related issues are investigated by developing autoregressive models and autoregressive models with exogenous inputs based on clinical data for two groups of subjects. The alternative modeling approaches are evaluated with respect to on-line short-term prediction accuracy for prediction horizons of 30 and 60 min, using independent test data.

Sensor abnormality detection based on global prediction model for type I diabetes

“…(2) A global (or universal) model [13] which is developed from glucose data for one subject and then used to make glucose predictions for other subjects. Here frequency band separation is not used.…”

“…Therefore, the identification of simple, accurate glucose prediction models has been drawing increasing attention. The combination of predictive empirical (or "data-driven") modelling techniques [8][9][10][11][12][13] with frequent and rich glucose measurements may provide an early warning of impending glucose excursions and proactive regulatory interventions for diabetes subjects. In general, the prediction model has the form of a linear dynamic model where the future glucose concentration is calculated as the linear combination of current and past glucose signals, and/or available exogenous input signals, which commonly include insulin delivery and meal carbohydrate (CHO) estimates.…”

Multivariate statistical analysis methods to investigate interindividual glucose dynamics for subjects with type 1 diabetes mellitus

“…Thus the model is not suitable for on-line applications where future glucose values are not available, which, thus, was limited regarding its practical application. Zhao et al [19] reported a global AR model for online glucose prediction based on frequency-band separation where glucose dynamics are separated into two parts, the low-frequency band and high-frequency band, where the low-frequency band includes subject-common glucose dynamics. Rahaghi and Gough [20] have recently suggested that the glucose dynamics for subjects without diabetes can be divided into four distinct frequency ranges with different periods.…”

Multiple order model migration and optimal model selection for online glucose prediction in Type 1 diabetes

“…Not only can continuous monitoring technologies alert a user when a hypoglycaemic episode or other blood glucose excursion is underway, but CGM measurements may also provide important information for the development of a prediction model to alert near‐future hyper/hypoglycaemia occurrences. Various glucose prediction methods have been reported recently. Considering the inherent complexity of the glucose‐insulin dynamic system, which is simulated by mathematical models, data‐driven (or empirical) models have been widely studied to explore the information hidden in the data to learn glucose response to various stimuli.…”

“…However, it is not useful for glucose control because the information obtained from the exogenous inputs is not employed. Zhao et al have also pointed out that AR with exogenous inputs (ARX), models with two exogenous inputs, and insulin delivery and meal carbohydrates (CHOs), were not global. An ARX model has been successfully used to develop models for the artificial pancreas (AP) .…”