Page 225 - Glucose Monitoring Devices

228 CHAPTER 11 Retrofitting CGM traces

Outpatient-like dataset
In an outpatient setting, BG references are collected much less frequently than inpa-
tient studies. Hence, in order to emulate the few outpatient reference data here we
discarded about 80% of the references in the original dataset and retained only those
that would have been collected in our outpatient protocols [11e14]. Specifically, we
retained:
• BG references collected at calibration times;
• BG references prescribed prior to the admission and the discharge of the patient;
• BG check in response to a CGM hypo-alarm;
• BG references collected at meal time, to compute meal-bolus;
• BG references performed about 2 h after the meal.

For each admission, in the outpatient-like dataset, we retained 12ð11 13Þ
samples (reported as median and (25th 75th percentile) range), which correspond
to 21:4%ð20:0% 23:6%Þ of the samples collected during the trial.
An example of the original and the outpatient-like datasets is illustrated in
Fig. 11.2. The continuous blue trace represents the CGM readings, and red dots
denote the reference BG values that would have been collected in an outpatient
study. Gray diamonds are the YSI measurements of the original dataset that are
not retained in the outpatient-like dataset.

Accuracy outcomes metrics and statistical analysis

Original CGM and retrofitted CGM values are matched with the gold standard mea-
surement performed at same time, i.e., the test-set YSI measurements.
The main metrics used to assess the accuracy of the signals are the absolute
difference (AD) and the relative absolute difference (ARD). These metrics are
computed for each data pair. Then, the overall mean (standard deviation) is reported
for normally distributed metrics, and population median (25the75th) percentile is
reported for nonnormally distributed metrics. Normality is assessed by Lilliefors
test.
To evaluate interpatient variability of the accuracy, we computed for each patient
mean AD (MAD) and mean ARD (MARD). Then, the population mean (standard
deviation) is reported for normally distributed metrics, and population median
(25the75th) percentile is reported for nonnormally distributed metrics.
Finally, clinical accuracy is assessed with the Clark Error Grid analysis (CEG)
[22] by reporting the percentage of data points falling in zone A and with the
percentage of points falling in zone Ar of the rate grid (accurate glucose rate) of
the rate error grid of the continuous glucose error grid analysis proposed in Ref. [23].

Statistical analysis
Original and retrofitted CGM performances are compared with a paired t-test for
normally distributed metrics and with a Wilcoxon signed-rank test for nonnormally
distributed data.