Page 271 - Glucose Monitoring Devices
P. 271
278 CHAPTER 14 Predictive low glucose suspend systems
(3) Hybrid infinite impulse response filter: Uses linear discrete-time signal
processing to generate output glucose predictions using previously measured
glucose concentrations without the input of insulin infusion; this generates a
prediction horizon through recursive application of the filter [37,39];
(4) Statistical prediction: Uses multiple empirical statistical models to generate a
probability of hypoglycemia; the methods employed include (a) calibration, to
convert CGM signal into an accurate blood glucose (BG), (b) prediction, using
training data and recent calibrated BG history to generate predictions and
accuracy estimates, (c) hypoglycemia alarming, to transform predictions and
accuracy estimates into a probability of hypoglycemia, which is then thresh-
olded into a binary alarm [37,40,41];
(5) Numerical logical algorithm: Feeds a three-point calculated rate of change
using backward difference approximation and current CGM value into logical
expressions to predict hypoglycemia [37].
The results from these initial trials demonstrated that the use of less voting
systems to trigger pump suspension produced better hypoglycemia protection
[36]. A subsequent in silico trial utilizing only a refined Kalman filter demonstrated
that this simplified approach could reduce hypoglycemia by almost three quarters
[42]. Outpatient safety testing of this Kalman filter-based algorithm demonstrated
a significant reduction in overnight hypoglycemia, though with a mild increase in
mean AM glucose values [43].
Commercial development of PLGS technology has resulted in two different
systems: the MiniMed predictive low glucose management (PLGM) system found
in the MiniMed 640G and as a feature in the MiniMed 670G, and the Tandem
PLGS system available as Basal-IQ [44e46]. These commercial designs (see below
for more details) are even more simplistic than those originally proposed by Bucking-
ham as part of the voting system. The PLGM design uses linear regression to estimate
future glucose values. The Basal-IQ system uses a linear regression model tuned on
the last four sensor glucose values to predict the sensor glucose 30 min into the future.
The development of PLGS technology has thus favored the selection of more
simplistic single algorithm designs that suspend basal insulin delivery based on
linear prediction models. These systems utilize slightly different rules for the
resumption of insulin delivery that may play a role in the postsuspension peak
glucose values seen with the different designs. In the next section, we will review
clinical trial data on the outpatient and real-world use of PLGS technology.
PLGS clinical studies
In addition to the continuum of developmental studies conducted earlier, there have
been numerous clinical efficacy studies conducted on the PLGM and PLGS
algorithms. Such studies have been conducted during inpatient, supervised outpa-
tient, and real-world environments. The compilation of these studies is presented
below (Table 14.1).