Page 288 - Glucose Monitoring Devices

Introduction 295

horizon to determine the optimal future insulin infusion rates with respect to a spec-
ified performance index. Furthermore, MPC can explicitly consider the system
constraints and multivariable interactions in the optimization problem, and the
MPC formulations are not inexorably restricted by the type of model, objective func-
tion, or constraints. As predictive controllers compute the optimal control actions on
the basis of a model and cost function, the glucose control performance is substan-
tiated by the fidelity of the glucose-insulin models. The MPC formulations
employed in many early clinical trials are synthesized using an average linear model
relating insulin to glucose concentrations. The use of a general linear model to
describe the complex nonlinear glycemic dynamics is compelled by the feasibility
and computational complexity of the controller implementation on portable systems
with limited computational and power resources. However, subjects exhibit diverse
glucose-insulin dynamics over time (intrasubject variability) and in relation to other
subjects (intersubject variability) due to diverse time-varying biological and physi-
ological characteristics of people. Adaptive and patient-individualized glucose-
insulin models can thus improve the glucose control performance [21,34e37].
An adaptive and customized MPC formulation with tailored glycemic models
developed through efficient identification techniques can improve the glucose con-
trol performance. Despite these controller enhancements, the closed-loop control
performance of glucose regulation is restricted by the exclusive reliance on glyce-
mic measurements and trends. This restrictive observation into the physiological
characteristics of people with T1DM limits the control performance, especially
when physical activity is confronted, leaving subjects to manually manage glyce-
mic variations during and after exercise. Conducting moderate-intensity aerobic
exercise lowers glucose concentrations as sensitivity to insulin and glucose uptake
to working muscles increases. Although moderating insulin administration before
exercise can alleviate exercise-induced hypoglycemia, it requires prior planning
and deliberate organization, which is demanding and disregards spontaneous activ-
ities. Neglecting the effects of physical activity in glycemic models and closed-
loop insulin control can contribute to a worsening of glycemic control [38]. These
limitations can be addressed by explicitly considering physiological measurements
from wearable devices, such as heart rate and energy expenditure, that are repre-
sentative of physical activity. Incorporating these additional variables in glycemic
models can extend the capability of the models beyond the usual univariate control
architecture for manipulating the infused insulin based solely on the glucose mea-
surements. Developing a multivariable AP (mAP) architecture where physiological
variables from wearables can characterize the glycemic effects of physical activity
offers immediate benefits for glucose control [39,40]. The improvement in the
glucose control performance is reinforced by the alleviation of manual user entries
for physical activity and meals as additional physiological variables and real-time
estimated parameters can quantify the prandial and exercise effects. The estima-
tion of physiological parameters also ensures improved glycemic control when
exercise and meals are unannounced or incorrect estimates are entered into the system.