Plasma Insulin Estimation in People with Type 1 Diabetes Mellitus

TLDR

The study aims to provide real‑time estimation of plasma insulin concentration in type 1 diabetes patients. This is achieved by augmenting Hovorka’s glucose–insulin model with continuous‑discrete extended Kalman filtering, unscented Kalman filtering, moving horizon estimation, and latent‑variable regression to jointly estimate insulin and time‑varying model parameters. Clinical data from T1DM subjects confirm that the proposed estimators accurately reconstruct plasma insulin levels.

Abstract

In this work the real-time estimation of plasma insulin concentration (PIC) to quantify the insulin in the bloodstream in patients with type 1 diabetes mellitus (T1DM) is presented. To this end, Hovorka's model, a glucose–insulin dynamics model, is incorporated with various estimation techniques, including continuous-discrete extended Kalman filtering, unscented Kalman filtering, and moving horizon estimation, to provide an estimate of PIC. Furthermore, due to the considerable variability in the temporal dynamics of patients, some uncertain model parameters that have significant effects on PIC estimates are considered as additional states in Hovorka's model to be simultaneously estimated. Latent variable regression models are developed to individualize the PIC estimators by appropriately initializing the time-varying model parameters for improved convergence. The performance of the proposed methods is evaluated using clinical data sets from subjects with T1DM, and the results demonstrate the accurate estimation of PIC.