Page 230 - Glucose Monitoring Devices
P. 230
Accuracy of retrofitted CGM versus number of references available 233
Adults Pediatrics
15 15
mean MAD mean MAD
Mean Absolute Difference [mg/dl] 10 5 Mean Absolute Difference [mg/dl] 10 5
0 0
CGM Retrofitted CGM CGM Retrofitted CGM
Adults Pediatrics
15 mean MARD 15 mean MARD
Mean Absolute Relative Difference [%] 10 5 Mean Absolute Relative Difference [%] 10 5
0 0
CGM Retrofitted CGM CGM Retrofitted CGM
FIGURE 11.6
Boxplot of MAD (top panel) and MARD (bottom panel) population distribution of the
original CGM and of the retrofitted CGM obtained with five SMBG per 12-h session (each
gray dot represents a patient admission). The improvement granted by the method is
clearly visible.
references was chosen since it is likely that at least five references in 12 diurnal
hours are collected when the CGM sensor is used adjunctively to SMBG, e.g.,
two SMBG measurements for calibration plus three SMBG checks around meal
time per day. Indeed, five [4,7] SMBGs were collected from 07:00 to 19:00 by
the patients in our artificial pancreas studies conducted in free-living conditions
for 2 months [27] and 1 month [28]. More precisely, in 59.1% of the 12-h periods
going from 07:00 to 19:00 the patients collected five or more SMBGs, and only
in 23.8% of these 12-h periods less than four SMBG measurements were performed.
Accuracy of retrofitted CGM versus number of references
available
In this section, we investigate how the accuracy improvement achieved by retrofit-
ting Dexcom G5 in real-life conditions is affected by the number of SMBG measure-
ments available. In fact, our method leverages on these BG measurements to