This prospective ancillary study assessed the performance of the Dexcom G7 interstitial continuous glucose monitoring (CGM) system in a cohort of critically ill adults. The objective was to compare CGM readings with arterial blood glucose under routine ICU conditions and during induced rapid glucose changes using intravenous glucose tolerance tests (IVGTT).
The investigation was an ancillary prospective study nested within the Determination of Insulin Resistance in ICU (DIRICU) study. Twenty critically ill adults were enrolled. Inclusion criteria reported in the abstract were: presence of stress hyperglycemia, no history of diabetes, and an anticipated ICU stay of at least 5 days. The study compared sensor-derived interstitial glucose values from the Dexcom G7 to arterial blood glucose measurements over the first 10 days of ICU admission.
Accuracy assessment used paired CGM and arterial blood glucose values. Performance metrics included mean bias, mean absolute relative difference (MARD), linear regression (R2), and Clarke error grid analysis. Measurements were analyzed separately for readings obtained outside of IVGTT and for those obtained during IVGTT. A post hoc temporal shift analysis (20-minute shift) was applied to evaluate whether accounting for a time lag between interstitial and blood glucose signals improved concordance during rapid glucose excursions.
Outside of IVGTT, the study collected 716 paired CGM–arterial blood glucose values. Results indicated moderate agreement between Dexcom G7 and reference glucose under these relatively stable conditions: the linear regression coefficient was R2 = 0.58. The mean bias was reported as +11 mg/dL, and MARD was 17%. Clarke error grid analysis placed 98.9% of paired values in zones A or B, indicating values that would lead to clinically acceptable or benign treatment decisions.
During IVGTT, which provokes rapid glucose excursions, the dataset comprised 1,018 paired values. Accuracy deteriorated substantially under these conditions. The R2 dropped to 0.028, indicating poor linear correlation between sensor and arterial readings. Mean bias shifted to −34 mg/dL, and MARD increased to 24%. Clarke error grid results showed only 80.3% of values in zones A or B and an elevated proportion (18.6%) in zone D, reflecting potentially dangerous discrepancies that could lead to erroneous clinical actions.
A post hoc analysis introduced a 20-minute temporal shift to CGM values to account for expected physiological and sensor delays between blood and interstitial glucose peaks during rapid changes. After temporal shifting, performance during IVGTT improved: R2 rose to 0.34, MARD decreased to 19%, and 94.7% of values fell into Clarke zones A or B. These improvements suggest that part of the observed inaccuracy during rapid excursions reflects a temporal lag of interstitial measurements relative to arterial blood glucose rather than purely measurement error.
The abstract reports no sensor-related adverse events among study participants during the monitoring period.
In this cohort of 20 critically ill adults with stress hyperglycemia and no prior diabetes, the Dexcom G7 CGM system demonstrated acceptable accuracy during relatively stable ICU conditions by common metrics (MARD 17%, 98.9% in Clarke A/B). However, performance declined markedly during rapid glucose excursions provoked by IVGTT, with higher MARD (24%), substantial negative bias, poor correlation, and an increased fraction of values in Clarke zone D. A 20-minute temporal shift partially corrected these discrepancies, supporting the role of physiologic lag between blood and interstitial compartments.
Based on these results, the authors suggest that in critically ill patients—particularly when rapid glycemic changes are likely—reliance on interstitial CGM alone may be insufficient. Hybrid monitoring strategies that combine continuous interstitial glucose monitoring with intermittent blood glucose confirmation may be preferable to ensure safe and accurate glycemic management in the ICU.
Note: The abstract provides aggregated accuracy metrics and safety findings; additional methodological details, patient-level characteristics, and extended outcomes were not reported in the abstract.