Page 287 - Glucose Monitoring Devices

294 CHAPTER 15 Automated closed-loop insulin delivery

tions of the control algorithms and insulin dosage computations involved in the in-
sulin delivery systems, highlights the disturbances affecting glycemic control, and
introduces the various paradigms developed to address the challenges to tight glyce-
mic control.
The emergence of continuous glucose monitoring (CGM) sensors, providing
real-time measurements of subcutaneous glucose levels, signified a major step to-
ward improved diabetes monitoring and treatment [7]. CGM enables frequent feed-
back to make corrections and appropriate changes in insulin delivery. CGMs enable
users to take preventative measures and make adjustments in insulin therapy based
on real-time interstitial glucose readings and alerts for impending hypoglycemia or
hyperglycemia excursions [8e11]. Sensor-augmented pump therapy that combines
CGMs with continuous subcutaneous insulin infusion (CSII) pumps is shown to
improve glycemic control compared with multiple daily injection therapy [11,12]
and provides increased functionality that includes personalized bolus calculators,
preprogrammed temporary insulin infusion suspension based on preset hypoglyce-
mic thresholds, and automated insulin delivery.
Closed-loop control of glucose concentrations through the pairing of CGM sen-
sors and CSII pumps using control algorithms builds upon the concept of sensor-
responsive insulin delivery and is a topic of significant interest [13]. Closing the
loop between glucose concentration sensing and insulin infusion through control
algorithms, termed automated insulin delivery or artificial pancreas (AP) systems,
allows the algorithms to automatically adjust the insulin infusion in real time based
on feedback from the CGM sensors [1,14e17]. The core of the AP system is the con-
trol algorithm that computes the appropriate amount of insulin to administer to
subjects [18,19]. Various control algorithms are developed for AP systems to auton-
omously manipulate the subcutaneous delivery of insulin on the basis of real-time
sensor glucose levels, including proportional-integral-derivative (PID) control,
fuzzy logic control, neural networks, and model predictive control (MPC)
[9,15,16,20e30]. The classical PID controller manipulates insulin delivery by
assessing the deviation of current glucose measurements from the target glucose
level (the proportional component), the area under the curve between measured
and target glucose levels (the integral component), and the rate of change in the
measured glucose level (the derivative component). Fuzzy logic control adjusts
the insulin infusion rate based on approximate encoded rules that mathematically
express the empirical clinician knowledge acquired by diabetes practitioners.
Artificial neural networks approximate nonlinear uncertain systems that are then
readily exploited in the synthesis of nonlinear controllers. Among the AP control
algorithms, MPC, a control strategy based on optimal control concepts, has become
increasingly prevalent because they have theoretically proven closed-loop stability
properties, are readily able to handle complex multivariable systems, and can sys-
tematically deal with state and input constraints [14,31e33].
MPC algorithms utilize dynamic models of the system in the optimization prob-
lem to predict the future evolution of the glucose measurements over a finite-time