Page 270 - Glucose Monitoring Devices

Algorithm development 277

FIGURE 14.1
Case 1: 15-year-old male wearing an original Minimed CGMS. Case 2: 12-year-old female
wearing a Minimed CGMS-Gold monitor. Case 3: 16-year-old female wearing the original
Minimed CGMS. Case 4: 17-year-old female wearing a MiniMed 722 Paradigm real-time
continuous glucose monitor. Alarm “bell” along the time axis at the bottom of the graph
indicates alarming (vibratory and then audio). BG, blood glucose [34].

Algorithm development
The use of predictive algorithms to prospectively suspend insulin delivery before
hypoglycemia is a very intuitive idea similar to how patients use their own devices
in real time. Refinement and selection of appropriate algorithms for this purpose,
however, was a process that took almost a decade to progress from theory to
clinically approved devices. An early study of PLGS technology published by Buck-
ingham in 2010 demonstrates the complexity of the initial approaches to achieve
automated prevention of hypoglycemia [36]. The system tested in this study used
a voting scheme between five different algorithms to recommend predictive suspen-
sion of insulin [37]. The algorithms deployed included the following:

(1) Linear projection: a projection of future glucose based on linear regression of
the past 15 min of CGM values [37];
(2) Kalman filter: a prediction of future glucose based on Kalman filter estimation
of glucose and its rate of change; this filter attempts to minimize the effect of
sensor noise [37,38];