Improved Accuracy of Continuous Glucose Monitoring Systems in Pediatric Patients with Diabetes Mellitus: Results from Two Studies

Abstract

Objective: This study was designed to evaluate accuracy, performance, and safety of the Dexcom (San Diego, CA) G4(®) Platinum continuous glucose monitoring (CGM) system (G4P) compared with the Dexcom G4 Platinum with Software 505 algorithm (SW505) when used as adjunctive management to blood glucose (BG) monitoring over a 7-day period in youth, 2-17 years of age, with diabetes.

Research design and methods: Youth wore either one or two sensors placed on the abdomen or upper buttocks for 7 days, calibrating the device twice daily with a uniform BG meter. Participants had one in-clinic session on Day 1, 4, or 7, during which fingerstick BG measurements (self-monitoring of blood glucose [SMBG]) were obtained every 30 ± 5 min for comparison with CGM, and in youth 6-17 years of age, reference YSI glucose measurements were obtained from arterialized venous blood collected every 15 ± 5 min for comparison with CGM. The sensor was removed by the participant/family after 7 days.

Results: In comparison of 2,922 temporally paired points of CGM with the reference YSI measurement for G4P and 2,262 paired points for SW505, the mean absolute relative difference (MARD) was 17% for G4P versus 10% for SW505 (P < 0.0001). In comparison of 16,318 temporally paired points of CGM with SMBG for G4P and 4,264 paired points for SW505, MARD was 15% for G4P versus 13% for SW505 (P < 0.0001). Similarly, error grid analyses indicated superior performance with SW505 compared with G4P in comparison of CGM with YSI and CGM with SMBG results, with greater percentages of SW505 results falling within error grid Zone A or the combined Zones A plus B. There were no serious adverse events or device-related serious adverse events for either the G4P or the SW505, and there was no sensor breakoff.

Conclusions: The updated algorithm offers substantial improvements in accuracy and performance in pediatric patients with diabetes. Use of CGM with improved performance has potential to increase glucose time in range and improve glycemic outcomes for youth.

FIG. 1.

Mean absolute relative difference (MARD) (%) comparison of continuous glucose monitoring against (a) reference YSI across in-clinic study Days 1, 4, and 7 and (b) self-monitoring of blood glucose across home use Days 1–7. For continuous glucose monitoring accuracy against the YSI, MARD improved from 21% on Day 1 to 16% on Day 4 and then to 15% on Day 7 for the G4 Platinum (G4P); MARD improved from 13% on Day 1 to 8% on Day 4 and then to 10% on Day 7 for the Software 505 algorithm (SW505). For continuous glucose monitoring accuracy against self-monitoring of blood glucose, MARD improved from 19% on Day 1 to 12% on Day 7 for the G4P; MARD improved from 15% on Day 1 to 11% on Day 7 for the SW505.

FIG. 2.

Clarke Error Grids for continuous glucose monitoring (CGM) versus reference YSI (a and b) and self-monitoring of blood glucose (SMBG) (c and d) for the G4 Platinum (a and c) and G4P Platinum with Software 505 algorithm (b and d). The Clarke error grid results indicate superior performance with the Software 505 algorithm compared with the G4 Platinum in the comparison of CGM versus reference YSI glucose values and CGM versus SMBG. For the G4 Platinum, 68% of values fell in Zone A, and 98% fell in Zones A and B; for the Software 505 algorithm, 90% of values fell in Zone A, and 99% fell in Zones A and B. For the comparison of CGM with SMBG values, 75% of values fell in Zone A, and 98% fell in Zones A and B for the G4 Platinum; 86% fell in Zone A, and 98% fell in Zones A and B for the Software 505 algorithm.

FIG. 3.

Parkes Error Grids for continuous glucose monitoring (CGM) versus reference YSI (a and b) and self-monitoring of blood glucose (SMBG) (c and d) for the G4 Platinum (a and c) and G4P Platinum with Software 505 algorithm (b and d). Assessment of CGM accuracy using the Parkes Error Grid yielded improved performance with the Software 505 algorithm compared with the G4 Platinum, with greater percentages of CGM values falling within the clinically accurate Zone A and the combined Zone A plus the benign error Zone B for CGM versus reference YSI and CGM versus SMBG. Notably, 100% of the CGM values fell with Zones A and B for CGM versus YSI and for CGM versus SMBG with the Software 505 algorithm.

FIG. 4.

Bland–Altman density plots of continuous glucose monitoring (CGM) versus reference YSI glucose (a and b) and self-monitoring of blood glucose (SMBG) (c and d) measurements for the G4 Platinum (a and c) and the G4 Platinum with Software 505 algorithm (b and d). In Bland–Altman density plots, depicting the bias of CGM to YSI and of CGM to SMBG (in mg/dL) for the G4 Platinum and the Software 505 algorithm, the bias was centered around 0, with higher density for the Software 505 algorithm compared with the G4 Platinum for both CGM versus YSI and CGM for SMBG. The majority of the bias for the Software 505 algorithm fell within the modified International Organization for Standardization area of %20/20.