Page 231 - Glucose Monitoring Devices
P. 231
234 CHAPTER 11 Retrofitting CGM traces
improve CGM, and as a consequence, an accuracy degradation should be expected
as less of them become available.
For this analysis, we employ the same original datasets presented in the previous
section, but, at difference with the previous section, we let the number of SMBGs
provided to the method vary from N SMBG ¼ 10 to N SMBG ¼ 2 per 12-h session,
spanning both the cases when more or less than NSMBG ¼ 5 are available. For
the case N SMBG ¼ 2 per 12-h session, the two calibration SMBG measurements
were used. More than five SMBGs are expected to be collected by a compliant
patient, verifying the CGM reading with an SMBG before taking any therapeutic
decision, i.e., at each meal, 2 h after each meal and in case of hypo or hypoglycemia
requiring interventions; less than five SMBGs are likely collected by a noncompliant
patient or in a 12-h portion including the night. Furthermore, only two SMBGs per
day (calibrations) are available in case of nonadjunctive use of CGM, a possible
future scenario given by the recent US Food and Drug Administration panel meeting
where the panel expressed a positive opinion about the change of the label of Dex-
com G5 sensor from adjunctive to nonadjunctive [3].
Fig. 11.7 shows the how the retrofitted CGM accuracy changes when the number
of SMBG varies. Upper panels report the mean absolute deviation (mean AD) for the
two populations, while the lower panels reports the mean ARD. When many SMBGs
are available (8e10 in a 12-h session), the retrofitted trace is nearly as accurate as the
SMBG: for N SMBG ¼ 10 references per 12-h session, MAD is reduced by about 38%
in adults and by 45% in adolescents and similarly MARD is reduced by 30% and
45%, respectively. The benefit of retrofitting decreases gradually as the number of
SMBG decreases: for N SMBG ¼ 5 references per 12-h session, the improvements
in both metrics are around 30%, as previously discussed; when only N SMBG ¼ 2
references per 12-h session are available the reduction is lower but still statistically
significant: MAD decreases from 16.2 mg/dL to 13.6 mg/dL (P < .001) in adults
(about 16%) and from 18.1 mg/dL to 15.7 mg/dL (P <.001) in adolescents (about
13%); similarly, MARD reduction is from 9.0% to 8.2% (P <.001) in adults (about
9%) and from 10.7% to 9.5% (P < .001) in adolescents (about 11%).
Conclusions
The retrofitting algorithm, combining a few BG reference samples with frequent,
possibly noisy, and biased CGM data, provides an accurate and quasi-continuous
BG profile. In this chapter, we provided details on how to implement the algorithm,
and we showed several practical applications, e.g., to retrospectively improve CGM
data collected in outpatient clinical trial, but also CGM data collected in free-living
conditions.
Appendix: data preprocessing
Both reference measurements and CGM time series are possibly affected by outliers
or unreliable/wrong data. With regard to BG reference outliers, they are mostly due