Optimal Design for Individual Model Identification based on Ambulatory Continuous Glucose Monitoring in Patients with Type 1 Diabetes

Abstract

Obtaining an individual model parametrisation based on ambulatory continuous glucose monitoring in patients with type 1 diabetes is a key challenge for the development of an artificial pancreas. Although multiple model structures have been proposed in the literature, parameter identification from individual glycaemic profiles remains a difficult task. In this work, an optimal experimental design study is carried out, based on Fisher Information Matrix techniques, for improving identification of individual glycaemic parameters. Simple experiment conditions are proposed to find out the most appropriate and timely insulin bolus according to different meals. Based on a qualitative interpretation of the results, a clinical protocol is proposed for model identification based on ambulatory data provided by patients (continuous glucose monitoring, insulin pump and patient's logbook). Clinical studies are currently being performed in order to validate the proposed experiments. Preliminary results are shown and discussed.

Evaluation of a Novel Continuous Glucose Monitoring-Based Method for Mealtime Insulin Dosing—the iBolus—in Subjects with Type 1 Diabetes Using Continuous Subcutaneous Insulin Infusion Therapy: A Randomized Controlled Trial

“…Results seem to indicate that (1) CGM may be used to obtain an individual model of postprandial glycemic response, 12 (2) this model can be used for prandial insulin dosing, and (3) CGM-based insulin delivery results in a postprandial glycemic control similar to that achieved by standard bolus calculators. These results are encouraging and open the way to larger clinical studies aimed at validating less user-dependent strategies for prandial insulin dosing.…”

“…1). 12 In order to account for intrapatient variability, a prediction model with interval parameters 9 was calculated from the previous identified model considering 20% uncertainty in insulin sensitivity and 10% in CHO estimation. The interval model was validated during the last 3 days of CGM, where the patients had a standardized meal daily (40-g, 60-g, or 100-g CHO content, the same composition of the inpatient study).…”

“…The interval model was validated during the last 3 days of CGM, where the patients had a standardized meal daily (40-g, 60-g, or 100-g CHO content, the same composition of the inpatient study). 12 Finally, constraints on PG were posed, and the SIVIA algorithm led to a three-dimensional set of solutions that contained all the basalbolus combinations for maintaining PG between 90 and 180 mg/dL for the given meal, according to the prediction model (Fig. 2).…”

On the use of hard/soft specifications to deal with intra-patient variability in postprandial glucose control in type 1 diabetes1

“…Simpler models must be used for the practical implementation of modelbased control strategies, since no tracer data will be available for their identification during clinical practice. The model used here is a combination of Wilinska's model for subcutaneous insulin absorption (Wilinska et al, 2005), Dalla Man's model for glucose absorption from the gut (Dalla Man et al, 2006) and a modified version of Panunzi's model (Panunzi et al, 2007) for the endogenous glucoregulation (Laguna et al, 2010). This combination was chosen to get a good balance between model identifiability and model complexity.…”

Online redesign of clinical tests for the identification of type 1 diabetes models in the presence of continuous glucose monitoring systems

“…Preliminary studies involving the application in-vivo of conventional MBDoE methodologies including CGMSs (Laguna et al, 2010) show the potential of the technique, but clearly underline some major limitations of the conventional design formulations. First of all, limited preliminary information on model parameters and/or model mismatch may lead to plan scarcely informative (sub-optimal) or unfeasible tests (e.g.…”