Page 228 - Glucose Monitoring Devices
P. 228

Retrofitting real-life adjunctive data  231




                     As for the previous section, we start from datasets offering frequent SMBG and
                  YSI data, called “original dataset.” All YSI references and most SMBGs are then
                  discarded to emulate real-life data availability, obtaining what we will denote as
                  “real-life-like dataset.” Discarded YSI are used for testing.


                  Original datasets
                  The retrofitting algorithm was tested on the data collected in 51 adult subjects,
                  reported in Ref. [25], and in 46 adolescents, 13e17 years old, presented in
                  Ref. [26]. Both group of subjects wore the Dexcom G5 for 7 days and had 12 h
                  in-clinic session on either day 1, 4, or 7. During the admission, accurate BG refer-
                  ences were collected with YSI instrument (Yellow Springs, OH) about every 15 min
                  on arterialized venous samples and capillary SMBG about every 30 min using the
                  Bayer USB Contour Next meter.
                     Remark: We should point out that the data in both Refs. [25,26] were actually
                  collected using the Dexcom G4 equipped with the software 505 (also known as
                  G4AP) and not with the Dexcom G5. Nevertheless, the two models can be consid-
                  ered, for the purpose of this chapter, completely equivalent. Sensing and signal
                  processing technologies are instead identical, and the only difference in the two
                  products is in the data transmission hardware: G4 with software 505 requires an
                  ad-hoc receiver, while G5 allows direct data transmission and processing on the
                  patient’s smartphone.

                  Real-life-like datasets
                  To mimic a real-life outpatient setting, only N SMBG ¼ 5 references per 12-h sessions
                  are retained. In fact, five SMBG are likely to be collected during the daytime when
                  the CGM sensor is used adjunctively to SMBG: two SMBG measurements for
                  calibration (one at the beginning and one at the end of the session), plus three
                  SMBG checks related to the meals. The five retained SMBG are selected by uniform
                  SMBG sampling, i.e., retaining one SMGB measurement every X available ones,
                  with X suitably chosen to get the desired total number, N SMBG . YSI references
                  are used solely as gold standard to assess the accuracy of CGM and retrofitted CGM.

                  Accuracy outcomes metrics and statistical analysis
                  Original CGM and retrofitted CGM values are matched with the YSI measurements
                  performed at same time. Outcome metrics and statistical analysis are analogous to
                  those of the previous section.

                  Results
                  Fig. 11.5 reports the boxplots of AD and ARD distribution. By retrofitting, the mean
                  of AD is significantly reduced in both populations: from 16.2 mg/dL to 10.7 mg/dL
                  (P <.001) in adults (about 34%) and from 18.1 mg/dL to 11.9 mg/dL (P < .001) in
   223   224   225   226   227   228   229   230   231   232   233