Continuous glucose monitoring (CGM) accuracy can be affected by altered physiology and concurrent therapies in critically ill patients. The authors performed a secondary analysis to identify independent determinants of CGM performance using the Dexcom G6 device in adults receiving mechanical ventilation, vasopressor support and intravenous insulin. Candidate variables included sensor wear time, arterial blood glucose (ABG), noradrenaline‑equivalent (NE) dose, erythrocyte volume fraction, pH, lactate, body‑mass index (BMI) and dialysis.
This analysis used data from a prospective observational study of 40 critically ill adults under mechanical ventilation and vasopressor therapy who were receiving intravenous insulin. Paired ABG and CGM values were collected and analyzed; the dataset comprised 2,946 paired measurements. The CGM system evaluated was the Dexcom G6®.
The primary outcome was the percentage absolute relative difference (ARD) between CGM and ABG. Secondary outcomes were the proportions of CGM readings that met the ISO 15197:2013 and CLSI POCT12‑A3 accuracy criteria.
Multivariable generalized linear mixed models were fitted, including patient‑level random intercepts to account for repeated measures within patients. Continuous covariates were modeled using B‑splines to allow non‑linear relationships. The model evaluated independent associations between ARD (and the categorical accuracy metrics) and the preselected covariates: sensor wear time, ABG, NE dose, erythrocyte volume fraction, pH, lactate, BMI and dialysis.
Overall median ARD was 10.8% (IQR 5.2–18.8); mean ARD was 12.7% (95% CI 10.7–15.3) %. Regarding categorical performance, 64.5% of CGM readings met ISO 15197:2013 criteria and 56.0% met CLSI POCT12‑A3 criteria.
Sensor wear time was independently associated with ARD and with the probability of meeting ISO and CLSI accuracy criteria (P < 0.001). Predicted ARD decreased during the first approximately 50 hours after sensor insertion, while the probability of meeting ISO/CLSI criteria increased over the same period and then stabilized. These findings indicate an early postinsertion period of reduced stability in CGM performance that improves with wear time.
Arterial pH was independently associated with ARD and both categorical accuracy metrics (P ≤ 0.02). The reported relationship between pH and accuracy was approximately linear in spline analyses. In sensitivity analyses using a linear term, each 0.1‑unit increase in pH was associated with a 1.5 percentage‑point higher ARD (P = 0.009) and a 32% lower odds of meeting CLSI criteria (P = 0.001). Thus, acid–base status appears to influence CGM performance in this critically ill cohort.
NE dose and ABG were independently associated with ISO‑defined accuracy (NE dose P = 0.006; ABG P < 0.001). Spline analyses suggested reduced accuracy at higher NE doses (noted above approximately 0.6 µg/kg/min) and at lower arterial glucose values (below ≈5 mmol/L [<90 mg/dL]). The authors note that observations were limited in these extreme ranges, which constrains definitive interpretation for those subsets.
In this population of mechanically ventilated ICU patients receiving vasopressors and intravenous insulin, CGM accuracy using the Dexcom G6 improved substantially with increasing sensor wear time, with notable early postinsertion instability during the first ~50 hours. Arterial pH was independently associated with CGM performance, with higher pH linked to worse ARD and lower odds of meeting CLSI criteria in linear sensitivity analyses. NE dose and low ABG levels were also associated with reduced ISO‑defined accuracy, though data at high NE doses and very low glucose values were sparse.
These results suggest clinicians and implementers of CGM in critical care should be aware of early sensor instability and the potential influence of acid–base disturbances and vasopressor dosing on CGM accuracy. Consideration of wear time and physiologic state may inform interpretation of CGM values and decision‑support systems in the ICU.
This report is a secondary analysis from a prospective observational study; the abstract provides statistical associations but does not report causal inferences. Observations were limited in some covariate ranges (notably high NE doses and low ABG), which the authors explicitly acknowledge as constraining interpretation for those subsets. The abstract does not provide additional methodological details such as exact timing of measurements relative to interventions, calibration procedures, or patient inclusion/exclusion criteria; those details were not reported in the source abstract.
continuous glucose monitoring, CGM, Dexcom G6, sensor wear time, arterial pH, noradrenaline (vasopressor), arterial blood glucose, ARD