Abstract
BACKGROUND:
In the ICU, continuous glucose monitors (CGMs) may improve glycemia and reduce the need for point-of-care blood glucose (POC BG) monitoring, but face challenges because of clinical conditions that affect accuracy.
RESEARCH QUESTION:
What is the feasibility of using POC BG calibration to improve CGM accuracy?
STUDY DESIGN AND METHODS:
This feasibility study pooled data from a retrospective study of patients with COVID-19 in the ICU and a prospective single-arm clinical trial of patients in the ICU. Our sample included 110 patients receiving IV insulin monitored using a hybrid CGM plus POC BG protocol with a factory-calibrated Dexcom G6 CGM (Dexcom, Inc.). Validation was required for initial and ongoing nonadjunctive use or for standalone use and was defined as CGM ± 20% of POC BG measurement for values of ≥ 100 mg/dL or ± 20 mg/dL for values of < 100 mg/dL. In the cohort with COVID-19, calibration was performed at the nurse’s discretion. In the prospective study, calibration was performed after persistent failure to achieve validation.
RESULTS:
A total of 55 patients (50%) underwent 167 calibrations. Those with a calibration had a mean age of 57.9 ± 13.6 years, 49% were male, 83% were White, and 60% had type 2 diabetes. After calibration, validation was achieved in 72.6%, 66.7%, and 77.8% of patients at 6, 12, and 24 hours after calibration, respectively. The mean absolute relative difference (MARD) was 25% at calibration, decreasing to 9.6%, 12.7%, and 13.2% at 6, 12, and 24 hours. Similar percentages were observed after eliminating pairs with multiple calibrations. Calibration was timely, within 5 minutes of the POC BG measurement in 70% and < 10 minutes in 83%. No statistical difference in MARD was found between timely and late calibrations or based on sensor rate of change at the time of calibration.
INTERPRETATION:
Our feasibility study demonstrated an improvement in CGM accuracy with POC BG calibrations in ICU patients. Further research is needed to understand optimal implementation strategies and impact on outcomes. CHEST Critical Care 2025; 3(4):100193
Keywords: calibration, CGM, continuous glucose monitor, COVID-19, glucose variability, hybrid CGM-POC monitor, hybrid protocol, ICU, inpatient diabetes, inpatient hyperglycemia, intensive care unit, MARD, mean absolute relative difference, point-of-care blood glucose, sensor rate of change
Continuous glucose monitors (CGMs) have revolutionized diabetes management, facilitating improvements in hemoglobin A1c (HbA1c) and reductions in hypoglycemia and diabetic ketoacidosis.1–3 During the COVID-19 pandemic, CGM use in hospitals surged, easing nursing workloads by reducing the need for point-of-care blood glucose (POC BG) monitoring, patient-provider contact, and personal protective equipment.4–11 Despite these advantages, CGM use in hospitals, particularly in ICU settings, presents challenges because of clinical conditions that affect the accuracy of interstitial glucose measurement, including rapid glucose fluctuations, hypoperfusion, acidosis, and interfering substances.12 Thus, it is of critical importance to develop strategies for optimizing safe use of CGMs in the hospital.
CGMs are devices placed on the arm or abdomen that measure interstitial glucose concentrations every 1 to 5 minutes. Glucose values are transmitted to a receiver or smartphone for up to 7 to 15 days. They provide alerts for hypoglycemia or hyperglycemia and display glucose trends, offering valuable insights for glucose management. However, interstitial glucose can differ from blood glucose values, especially during rapid fluctuations.13,14
Older CGM technologies required manual calibration, which requires the user to enter a POC BG value into the CGM, adjusting the CGM’s algorithm. However, the need for ongoing fingerstick glucose monitoring, including inconvenience and patient discomfort, can reduce patient adherence. Moreover, calibrations could introduce bias if performed late or during periods of rapid glucose rate of change.14 The subsequent factory-calibrated systems eliminated the need for user calibration, making these devices more convenient and user-friendly.13,15 The Dexcom G6 (Dexcom, Inc.), for example, uses an automated calibration function that corrects for sensor drift over its wear period based on the typical drift of an average sensor.
Studies examining current factory-calibrated systems in ambulatory settings have reported a mean absolute relative difference (MARD) of 9% to 10%, with CGM values falling within 20% of reference values 93% of the time.16,17 Thus, CGM has become widely used in ambulatory settings. In contrast, ICU studies using newer CGMs report higher MARDs of 9.7% to 20.6%.18,19 This reduced performance highlights the challenges of using CGMs in the ICU and has limited regulatory approval for inpatient use. Expert guidelines have recommended confirmatory POC BG monitoring alongside CGM use in these settings. However, the role of manual calibrations in ICUs remains unexplored.
In this feasibility study, we investigated the impact of POC BG calibrations on CGM accuracy in 2 samples of patients in the ICU and evaluated the practicality of incorporating calibration protocols into routine ICU workflows. Our primary objective was to determine whether POC BG calibration improves CGM accuracy, measured by MARD and sensor validation rates after calibration. In addition, we evaluated the effects of calibration timeliness and rate of glucose change at the time of calibration.
Study Design and Methods
Study Population and Design
We conducted an observational study by pooling data from 2 cohorts (1 retrospective and 1 prospective) assessing CGM use in the ICU. The first (the cohort with COVID-19) was a retrospective observational study of patients with COVID-19 admitted to a single medical ICU from May 1, 2020, through January 15, 2021. All patients who received IV insulin in this unit were monitored using a factory-calibrated Dexcom G6 CGM, where feasible. POC BG monitoring was carried out using the Nova StatStrip Glucose Hospital Meter System (Nova Biomedical StatStrip). Trained ICU nursing staff placed sensors on patients. The hybrid CGM plus POC BG monitoring protocol and outcomes were reported previously.4 In brief, the CGM could be used nonadjunctively (without confirmation by POC BG monitoring) on an intermittent basis after 2 consecutive sensor values met validation criteria of ± 20% of POC BG monitoring for values of ≥ 100 mg/dL or ± 20 mg/dL for values of < 100 mg/dL. Thereafter, validation was assessed using POC BG monitoring at least every 6 hours if predicted low-glucose or high- or low-threshold alerts came from the CGM or if a change in clinical status occurred, such as initiation of vasopressors, intubation, or nutrition. The Dexcom G6 CGM is factory calibrated, and therefore user calibration is not required. However, optional user calibration is accomplished via manual entry of a blood glucose value into the receiver.20 The protocol did not include guidance on CGM calibration; however, calibrations were performed at the nurse’s discretion when validation was not met. The retrospective study was approved by The Ohio State University Biomedical institutional review board (Identifier: 2020H0408).
The second study (the cohort without COVID-19) was a prospective single-arm clinical trial evaluating the efficacy and feasibility of CGM use in the medical ICU. Data from the initial 60 participants enrolled in the trial between August 8, 2022, and September 5, 2023, were included in this analysis. Key inclusion criteria were admission to the medical ICU and hyperglycemia requiring IV insulin, based on daily screening (Monday-Friday) of the ICU census. POC BG monitoring was carried out using the Nova StatStrip Glucose Hospital Meter System. The validation thresholds for CGM use matched those of the retrospective study. However, the protocol required calibration if validation was not achieved within 12 hours. If validation still was not achieved, calibration was repeated 6 hours later. Sensors that failed to validate within 24 hours of sensor placement were discarded and replaced. The protocols also differed in that as soon as validation was achieved, POC BG monitoring occurred every 4 hours. Confirmatory POC BG monitoring was required more frequently for a duration of 6 hours in response to changes in clinical conditions such as intubation, initiation of pressor support, new cardiovascular events, changes in nutrition support, hemoglobin of < 7 g/dL, CGM BG of < 70 mg/dL, predicted glucose of < 55 mg/dL alert, pH of < 7.3, or discrepancies between symptoms and glucose readings. Exclusion criteria were COVID-19, refractory shock (norepinephrine dose > 0.5 μg/kg/min or equivalent), active treatment for diabetic ketoacidosis or hyperosmolar hyperglycemic state, pitting edema, anasarca, cyanosis, high-dose acetaminophen or hydroxyurea use, pregnancy, use of a home insulin pump, or residence in a correctional facility. The trial was regulated by a US Food and Drug Administration Investigational Device Exemption (G220050) and was approved by The Ohio State University Biomedical Institutional Review Board (Identifier: 2021H0407). All participants, legal next of kin, or both signed informed consent forms. In both cohorts, insulin dosing was guided by the institution’s IV infusion guidelines, which have a target blood glucose range of 120 to 150 mg/dL.
Technology
The Dexcom G6 CGM consists of a sensor, transmitter, and display device. The sensor is worn on the skin and is attached to a wire that is inserted < 0.5 inches subcutaneously via a single-use applicator on the back of the upper arm, similar to previous inpatient studies.4 It measures interstitial glucose every 5 minutes for 10 days. A bluetooth-connected transmitter (Dexcom G6 CGM) sends glucose readings to a paired display device, either a manufacturer receiver or a smartphone using the G6 app.17 The system provides discrete glucose values and trend data. The app continuously uploads data to Dexcom’s remote server, where it is displayed in the Clarity diabetes management software (Dexcom Clarity), whereas receiver data must be uploaded manually after use. Hyperglycemia and hypoglycemia alerts were set per protocol (> 300 mg/dL and < 100 mg/dL, respectively), with urgent low (< 54 mg/dL) and urgent low soon (predicted glucose < 54 mg/dL within 20 minutes) preset by the manufacturer. Smart phones and receivers were stored outside patient rooms, labeled with hospital identifications.
Data Collection
We extracted data from the electronic health record including medical history, HbA1C values, inpatient medications, POC BG monitoring values, and hospitalization details. POC BG measurements were obtained using StatStrip (Nova Biomedical), which automatically time-stamps glucose values. Devices are downloaded during each shift, and glucose values and time stamps are transmitted automatically to the electronic health record. CGM data were downloaded via Clarity software into patient-specific comma separated values files. Sensor-meter pairs were defined as a CGM value and a POC BG monitoring value obtained within 5 minutes of each other. These pairs were used to assess sensor accuracy at baseline (time of calibration) and at 6, 12, and 24 hours after calibration. For analysis at each time point, sensor-meter pairs were included if they occurred within 1 hour before or after the target time, specifically between 5 to 7 hours, 11 to 13 hours, and 23 to 25 hours after calibration.
Data Analysis
The primary outcomes for this analysis included frequency of sensor validation (already described) and MARD. Continuous variables with normal distribution were reported as mean (SD), whereas those with nonnormal distribution were reported as median (interquartile range [IQR]). Categorical variables were reported as number (percentage). Differences between groups were determined using the analysis of variance for normally distributed variables and the Kruskal-Wallis test for nonnormally distributed variables. Differences between categorical variables were determined using the 2-tailed Fisher exact test. Within-group comparisons were assessed using the Wilcoxon signed-rank test for nonparametric data or a paired t test for normally distributed data. P values were not adjusted for multiple comparisons. For each sensor-meter pair, the absolute relative difference was calculated by subtracting the CGM value from the reference POC BG value and dividing by the POC BG value and expressed as a percentage. The absolute relative difference was reported for analysis of the first sensor-meter pair after calibration and for the first and second sensor-meter pair after calibration. The MARD for each patient was calculated as the mean of all available absolute relative differences within each interval of interest. The sensor rate of change was calculated by taking the difference between the sensor value recorded 30 minutes before calibration and at the time of calibration and dividing by 30. Timely calibration was defined as calibration occurring within 5 minutes of the POC BG monitoring time stamp.
Results
Participant Characteristics
The analysis included 110 participants (50 participants in the cohort with COVID-19 and 60 participants in the cohort without COVID-19), with 55% of participants identifying as male, 75% of participants identifying as White, and 22% of participants identifying as Black; with a mean (SD) age of 59.9 (13.6) years; and with a mean (SD) BMI of 32.7 (8.7) kg/m2. Overall, 70% of participants had preexisting type 2 diabetes mellitus and 16% of participants had diabetic ketoacidosis or hyperosmolar hyperglycemic state, and the mean (SD) HbA1C level was 8.32% (2.38%) (or 67 [26] mmol/mol). A total of 63% of participants received corticosteroids, 42% of participants received vasopressors, and 73% of participants received mechanical ventilation. Calibrations were performed on 55 patients (calibration cohort): 21 participants (38%) from the COVID-19 study and 34 participants (62%) from the non-COVID-19 study. Baseline characteristics were similar, except for a trend for less type 2 diabetes and more type 1 diabetes in the calibration group (Table 1).
TABLE 1 ].
Baseline Characteristics by Calibration
| Characteristic | Overall (N = 110) | Any Calibration (n = 55) | No Calibration (n = 55) |
|---|---|---|---|
| Age, y | 59.9 (13.6) | 57.9 (13.6) | 61.9 (13.4) |
| Sex | |||
| Male | 61 (55%) | 28 (49%) | 33 (60%) |
| Female | 49 (45%) | 27 (51%) | 22 (40%) |
| BMI, kg/m2 | 32.7 (8.7) | 32 (7.7) | 33.3 (9.7) |
| Race | |||
| White | 82 (75%) | 43 (78%) | 39 (71%) |
| Black | 24 (22%) | 12 (22%) | 12 (22%) |
| Asian | 1 (1%) | 0 | 1 (1%) |
| Othera | 3 (3%) | 0 | 3 (5%) |
| Admission | |||
| Floor | 8 (7%) | 2 (4%) | 6 (11%) |
| Step-down | 12 (11%) | 5 (9%) | 7 (13%) |
| ICU | 90 (82%) | 48 (87%) | 42 (76%) |
| Medical history | |||
| Diabetes | |||
| No diabetes | 18 (17%) | 12 (22%) | 6 (11%) |
| Type 1 | 14 (13%) | 11 (20%) | 3 (6%) |
| Type 2 | 77 (70%) | 32 (58%) | 45 (82%), 1 other |
| Hypertension | 89 (82%) | 41 (76%) | 48 (87%) |
| Heart failure | |||
| HFpEF | 15 (14%) | 7 (13%) | 8 (15%) |
| HFrEF | 11 (10%) | 7 (13%) | 4 (7%) |
| CAD | 25 (23%)b | 16 (29%)c | 19 (35%) |
| COPD or asthma | 35 (32%) | 13 (30%) | 22 (33%) |
| Tobacco use | 72 (66%)d | 35 (64%) | 37 (69%)e |
| Admission diagnosis | |||
| Pulmonary | 71 (65%) | 33 (60%) | 38 (69%) |
| Cardiac | 5 (5%) | 4 (7%) | 1 (2%) |
| Renal | 5 (5%) | 4 (7%) | 1 (2%) |
| Neurologic | 3 (3%) | 1 (2%) | 2 (4%) |
| Glucose related | 11 (10%) | 7 (13%) | 4 (7%) |
| Musculoskeletal | 1 (1%) | 1 (2%) | 0 |
| Gastrointestinal | 3 (3%) | 1 (2%) | 2 (4%) |
| Other | 11 (10%) | 4 (7%) | 7 (13%) |
| Hospital data | |||
| Hemoglobin A1C, % | 8.32 (2.39) (n = 98) | 8.35 (2.64) (n = 48) | 8.29 (2.15) (n = 50) |
| Dialysis | 25 (23%) | 14 (26%) | 11 (20%) |
| Intubated | 81 (74%) | 43 (78%) | 38 (69%) |
| Pressor support | 38 (43%)f | 21 (39%)g | 17 (48%)h |
| DKA or HHS | 18 (16%) | 9 (16%) | 9 (16%) |
| Steroid use | 56 (63%)f | 34 (63%)g | 22 (63%)h |
| Hospital length of stay, d | 21 (15–31)f | 21 (12–32)g | 23 (18–30)h |
| Mortality | 39 (35%) | 15 (34%) | 24 (36%) |
Data are presented as No. (%), mean (SD), or median (interquartile range). CAD = coronary artery disease; DKA = diabetic ketoacidosis; HFpEF = heart failure with preserved ejection fraction; HFrEF = heart failure with reduced ejection fraction; HHS = hyperglycemic hyperosmolar nonketotic state.
Other indicates any race other than White, Black, or Asian.
n = 108.
n = 53.
n = 109.
n = 54.
n = 89.
n = 54.
n = 35.
Participants in the COVID-19 study tended to have greater severity of illness than those in the non-COVID-19 study, as illustrated by HbA1c level (mean [SD], 8.77% [2.12%] vs 7.94% [2.51%]), renal replacement therapy (34% vs 13%), ventilator support (92% vs 58%), steroid use (74% vs 49%), and mortality (66% vs 10%) (e-Table 1).
Calibration Process Measures
Calibration process measures are shown in Table 2. Within the calibration cohort, a total of 167 calibrations were performed, 91 in the COVID-19 study and 76 in the non-COVID-19 study, with 2.0 calibrations (IQR, 1.0–3.5 calibrations) per patient (P = .55 between studies). The median sensor rate of change at calibration was 0.23 mg/dL/min (IQR, 0.06–0.43 mg/dL/min), with a significant difference between studies (median, 0.2 mg/dL/min [IQR, 0.07–0.37 mg/dL/min] vs 0.26 mg/dL/min [IQR, 0.075–0.58 mg/dL/min] in the cohort with COVID-19 and the cohort without COVID-19; P = .04). The median time between POC BG measurement and calibration was 3.1 minutes (IQR, 1.5–8.0 minutes) and was numerically greater in the cohort with COVID-19 (3.9 minutes [IQR, 1.6–9.8 minutes] vs 2.6 minutes [IQR, 1.4–5.0 minutes]; P = .08). In 70% of occurrences, POC BG measurements were taken within 5 minutes of calibration, and 80% were obtained within 10 minutes. After calibration, the time to the next sensor-meter pair was 0.99 hours (IQR, 0.83–1.41 hours), with a significantly shorter time in the non-COVID-19 study (P = .003).
TABLE 2 ].
Summary of Calibrations and Subsequent Sensor-Meter Agreement
| Variable | Overall (N = 110) | COVID-19 Study (n = 50) | Non-COVID-19 Study (n = 60) | P Valuea |
|---|---|---|---|---|
| Patients with calibrations | 55 (50) | 21 (38) | 34 (62) | > .99 |
| Total calibrations, No. | 167 | 91 | 76 | NA |
| Calibrations per patientb | 2 (1–3.5) | 2 (1–5) | 2 (1–3) | .55 |
| Sensor rate of change at calibration, mg/dL/min | 0.23 (0.06–0.43) (n = 163) | 0.2 (0.07–0.37) (n = 87) | 0.26 (0.075–0.58) (n = 76) | .04 |
| Time between calibration and POC BG measurement, min | 3.1 (1.5–8.0) | 3.9 (1.6–9.8) | 2.6 (1.4–5.0) | .08 |
| POC BG < 5 min of calibration | 116 (70) | 58 (64) | 58 (76) | .09 |
| POC BG 5–10 min of calibration | 17 (10) | 11 (12) | 6 (8) | .45 |
| POC BG < 10 min of calibration | 133 (80) | 69 (76) | 64 (84) | .24 |
| Next sensor-meter pair | ||||
| Time to next sensor-meter pair, h | 0.99 (0.83–1.41) | 1.04 (0.85–3) | 0.96 (0.78–1.09) | .003 |
| ARD, % | 7.09 (2.58–16.13) (n = 155) | 6.69 (3.62–15.04) (n = 82) | 7.56 (2.15–18.32) (n = 73) | .73 |
| Validation on first sensor-meter pair | 131 (85) (n = 155) | 74 (90) (n = 82) | 57 (78) (n = 73) | .05 |
| Validation on first and second sensor-meter pair | 108 (70) (n = 154) | 67 (83) (n = 81) | 41 (56) (n = 73) | .0004 |
| CGM validation | ||||
| 6 h | 82 (73) (n = 113) | 36 (73) (n = 49) | 46 (72) (n = 64) | > .99 |
| 12 h | 88 (67) (n = 132) | 40 (63) (n = 63) | 48 (70) (n = 69) | .86 |
| 24 h | 112 (78) (n = 144) | 63 (81) (n = 77) | 49 (73) (n = 67) | .23 |
| MARD | ||||
| At calibration | 25.9 (15.0–38.4) (n = 167) | 24.3 (13.0–37.1) (n = 91) | 30.1 (18.3–43.2) (n = 76) | .04 |
| 6 h | 9.35 (4.1–16.5) (n = 113) | 12.2 (4.2–21.0) (n = 49) | 8.3 (4.1–16.2) (n = 64) | .41 |
| 12 h | 12.7 (6.5–21) (n = 132) | 14.7 (6.9–24.8) (n = 63) | 11.0 (5.9–16.5) (n = 69) | .02 |
| 24 h | 13.2 (6.46–19.7) (n = 144) | 13.6 (6.3–18.0) (n = 77) | 13.2 (7.02–26.4) (n = 67) | .36 |
Data presented as No. (%), mean (SD), or median (interquartile range). The number of calibrations included vary across rows because of differences in data availability. Sensor rate of change was unavailable for 4 calibrations because of sensor initiation < 30 min before calibration (n = 163). ARD and validation at the next sensor-meter pair were based on calibrations with at least 1 sensor-meter pair 6 h after calibration (n = 155). One calibration with a second pair > 6 h after calibration was excluded from combined first and second pair validation (n = 154). CGM validation and MARD at 6, 12, and 24 h were assessed only for calibrations with a sensor-meter pair occurring within a 2-h window centered on each time point (ie, 5–7 h, 11–13 h, and 23–25 h after calibration). ARD = absolute relative difference; CGM = continuous glucose monitor; MARD = mean absolute relative difference; NA = not applicable; POC BG = point-of-care blood gas.
Not adjusted for multiple comparisons.
Among patients who required calibrations.
Effect of Calibration on Validation and Accuracy
Sensor validation and accuracy results are shown in Table 2. Validation was achieved in 85% of the first sensor-meter pairs after calibration (90% in the COVID-19 study and 78% in the non-COVID-19 study). In addition, 70% of calibrations achieved validation with the first and second sensor-meter pairs. A significantly higher proportion of sensor-meter pairs met validation criteria with the first 2 consecutive POC BG measurements after calibration in the COVID-19 study vs the non-COVID-19 study (83% vs 56%; P = .0004). Validation was achieved in 73% of calibrations at 6 hours, 67% of calibrations at 12 hours, and 78% of calibrations at 24 hours and was similar between studies.
The MARD was 26% (IQR, 15.0%−38.4%) at the time of calibration (and significantly higher in the non-COVID-19 study vs the COVID-19 study; P = .04). After calibration, MARD was 9.4% at 6 hours, 12.7% at 12 hours, and 13.2% at 24 hours (Table 2). The COVID-19 study showed a significantly higher MARD at 12 hours compared with the non-COVID-19 study (P = .02), but not at 6 or 24 hours. All MARDs after calibration were significantly lower than at the time of calibration (P < .0001) (Fig 1). Separate analysis after eliminating any sensor-meter pairs with repeat calibrations within each respective observation period yielded 77 calibrations and similar trends (e-Table 2).
Figure 1 –
Mean absolute relative difference at prior to (time 0) and after calibration.
Effect of Timely Calibration
When comparing calibrations performed within 5 minutes of POC BG measurement, between 5 and 10 minutes of the POC BG measurement, and > 10 minutes after the POC BG measurement, no statistically significant differences were found in validation frequency or MARD at 6, 12, or 24 hours (Table 3). However, 67% of calibrations were performed within 5 minutes, 87% of calibrations were performed within 10 minutes, and only 13% of calibrations were performed after 10 minutes.
TABLE 3 ].
Effect of Timely Calibration (Set at < 5 Min and < 10 Min) on Sensor-Meter Agreementa
| Variable | Calibration ≤ or > 5 min | Calibration ≤ or > 10 min | ||||
|---|---|---|---|---|---|---|
| Timely (≤ 5 min; n = 52) | Late (> 5 min; n = 25) | P Valueb | Timely (≤ 10 min; n = 67) | Late (> 10 min; n = 10) | P Valueb | |
| Validation on first sensor-meter pair | 41 (79) (n = 52) | 20 (91) (n = 22) | .32 | 54 (82) (n = 66) | 7 (87) (n = 8) | 1.0 |
| Validation on first and second sensor-meter pair | 37 (71) (n = 52) | 17 (81) (n = 21) | .56 | 49 (74) (n = 66) | 5 (71) (n = 7) | 1.0 |
| Validation | ||||||
| 6 h | 30 (68) (n = 44) | 14 (82) (n = 17) | .35 | 38 (70) (n = 54) | 6 (86) (n = 7) | .66 |
| 12 h | 31 (67) (n = 46) | 15 (71) (n = 21) | .79 | 41 (67) (n = 61) | 5 (83) (n = 6) | .66 |
| 24 h | 40 (87) (n = 46) | 16 (76) (n = 21) | .30 | 53 (87) (n = 61) | 3 (50) (n = 6) | .06 |
| MARD | ||||||
| At calibration | 27.3 (19.2–40.2) (n = 52) | 24.34 (19.3–33.5) (n = 22) | .58 | 27.1 (19.9–39.6) (n = 66) | 24.11 (11.1–36.0) (n = 8) | .32 |
| 6 h | 8.7 (4.5–16.9) (n = 44) | 10.2 (4.8–13.2) (n = 17) | .97 | 8.7 (4.5–16.8) (n = 54) | 12.6 (5.4–13.6) (n = 7) | .56 |
| 12 h | 8.5 (5.3–18.6) (n = 46) | 14.2 (6.0–19.2) (n = 21) | .34 | 8.9 (5.2–18.7) (n = 61) | 16.1 (10.7–23.4) (n = 6) | .24 |
| 24 h | 10.8 (5.4–15.9) (n = 46) | 13.2 (7.9–23.2) (n = 21) | .18 | 12.5 (6.0–17) (n = 61) | 10.1 (5.1–25.2) (n = 6) | .97 |
Data are presented as No. (%) or median (interquartile range) unless otherwise indicated. Timely calibration is defined as the entry of point-of-care blood glucose value within 5 min or 10 min of obtaining the point-of-care blood glucose, as determined by time-stamped data from device download. MARD = mean absolute relative difference.
Only calibrations without additional calibrations in the defined period (6 h, 12 h, or 24 h) are included.
Not adjusted for multiple comparisons.
Effect of Baseline Sensor Glucose and Rate of Change
The frequency of validation and MARD were similar regardless of sensor rate of change < ± 1 mg/dL/min vs > ±1 mg/dL/min at calibration (Table 4). Calibrations performed when initial POC BG was > 180 mg/dL showed higher 24-hour validation rates compared with those with lower initial glucose levels (88% vs 72%; P = .02), but no significant differences were found at other time points or with a cutpoint of 100 mg/dL (Table 5). Moreover, no significant differences were found in MARD at 6, 12, or 24 hours across baseline glucose thresholds.
TABLE 4 ].
Effect of Sensor Rate of Change at the Time of Calibration on Sensor-Meter Agreementa
| Variable | ROC, mg/dL/min | P Valueb | |
|---|---|---|---|
| < ± 1 | ≥ ± 1 | ||
| Validation | |||
| 6 h | 42 (75) (n = 56) | 2 (50)(n = 4) | .29 |
| 12 h | 44 (72) (n = 61) | 2 (40) (n = 5) | .16 |
| 24 h | 53 (84) (n = 63) | 2 (67) (n = 3) | .43 |
| MARD | |||
| At calibration | 25.8 (18.8–33.3) (n = 70) | 52.2 (43.5–64.2) (n = 5) | .0007 |
| 6 h | 9.8 (4.8–16.5) (n = 56) | 3.2 (0.8–8.9) (n = 4) | .07 |
| 12 h | 10.5 (5.2–18.7) (n = 61) | 7.7 (6.0–14.8) (n = 5) | .67 |
| 24 h | 11.5 (6.2–17.4) (n = 63) | 19.4 (0.6–38.8) (n = 3) | .54 |
Data are presented as No. (%) or median (interquartile range) unless otherwise indicated. MARD = mean absolute relative difference; ROC = rate of change.
Only calibrations without additional calibrations in the defined time period (6 h, 12 h, or 24 h) are included.
Not adjusted for multiple comparisons.
TABLE 5 ].
Differences in Validation and MARD Based on Initial POC BG Value
| Variable | Initial POC BG > or < 180 mg/dL | Initial POC BG < or > 100 mg/dL | ||||
|---|---|---|---|---|---|---|
| > 180 | < 180 | P Valuea | < 100 | >100 | P Valuea | |
| Validation on first sensor-meter pair | 51 (91) (n = 56) | 86 (81) (n = 106) | .11 | 2 (18) (n = 11) | 128 (85) (n = 151) | .6 |
| Validation on first and second sensor-meter pair | 42 (77) (n = 56) | 70 (67)(n = 104) | .28 | 4 (36) (n = 11) | 106 (71) (n = 149) | .7 |
| Validation | ||||||
| 6 h | 28 (80) (n = 35) | 54 (69) (n = 78) | .26 | 3 (37) (n = 8) | 77 (73) (n = 105) | .7 |
| 12 h | 30 (65) (n = 46) | 58 (67) (n = 86) | .84 | 2 (25) (n = 8) | 82 (66) (n = 124) | .7 |
| 24 h | 46 (88) (n = 52) | 66 (72) (n = 92) | .02 | 3 (37) (n = 8) | 107 (79) (n = 136) | .4 |
| MARD | ||||||
| 6 h | 11.5 (4.4–15.9) (n = 35) | 12.6 (3.7–20.3) (n = 78) | .9 | 9.64 (2.8–19.9) (n = 8) | 12.3 (4.1–16.5) (n = 10) | .9 |
| 12 h | 15.8 (5.5–21.7) (n = 46) | 15.6 (6.9–19.9) (n = 86) | .9 | 21.1 (11.3–34.5) (n = 8) | 15.3 (6–21) (n = 124) | .2 |
| 24 h | 13.3 (6.6–16.8) (n = 52) | 17.2 (6.5–24.8) (n = 92) | .17 | 14 (2.7–26.4) (n = 8) | 15.8 (6.8–19.1) (n = 136) | .6 |
Data are presented as No. (%) or median (interquartile range) unless otherwise indicated. MARD = mean absolute relative difference; POC BG = point-of-care blood glucose.
Not adjusted for multiple comparisons.
Discussion
This feasibility study was a comprehensive evaluation of the role for calibrations when adapting a factory-calibrated device for use in a hybrid CGM plus POC BG monitoring protocol in the ICU. Notably, the study demonstrated that calibrations are feasible and can be performed routinely by nursing in a timely manner (within 5–10 minutes of the POC BG measurement).
Moreover, calibration results in improved MARD and provided a greater chance of reaching the threshold for sensor validation required by the protocols for nonadjunctive use. The findings lay the groundwork for optimizing implementation of hybrid glucose monitoring approaches for managing hyperglycemia in ICU settings.
The most significant and novel finding of our study was the substantial improvement in MARD after calibration. In a patient population with characteristics that typically hinder CGM accuracy, this finding suggests that a protocol that incorporates calibration when validation criteria are not met may allow for greater periods of nonadjunctive use of CGMs. Of note, the overall MARD was still higher than that observed in ambulatory settings, thus underscoring the usefulness of hybrid monitoring protocols that call for periodic POC BG monitoring.
Currently, no US Food and Drug Administration-approved CGM exists for inpatient use, and data on their safety in the ICU are limited.21 Nevertheless, as CGM use grows in the ambulatory setting, patients may expect to continue use in the hospital. A 2020 consensus guideline by the Diabetes Technology Society recommends that hospitalized patients may continue using their personal CGM devices, but encourage hospital systems to implement protocols for supporting their use safely.22 In 2022, the Endocrine Society suggested adjunctive CGM use (with confirmatory POC BG monitoring) to guide adjustments in insulin dosing in hospitalized noncritically ill adults with diabetes at high risk of hypoglycemia, provided that resources and training are available.23 However, the American Diabetes Association notes that current research on clinical outcomes, safety, and cost-effectiveness is insufficient to endorse the widespread use of CGMs in hospitalized patients. A recent consensus statement suggests that careful manual calibration during stable glucose periods may improve CGM accuracy.24
An important finding in our comparison of patients requiring CGM calibration vs those who did not is that baseline characteristics were similar, except for the type of diabetes, which favored a higher prevalence of type 1 diabetes among those who received calibrations. This could be explained by greater frequency of glucose excursions and glucose variability affecting sensor accuracy in type 1 diabetes.25,26
It should be emphasized that the studies differed in that calibrations were performed according to protocol in the non-COVID-19 study, but at the discretion of the nurse in the COVID-19 study. In a previously published study, ICU nurses caring for this cohort of patients with COVID-19 reported little to no familiarity with CGMs before inpatient use.27 Therefore, knowledge and discretion on when and how to perform calibration likely was gleaned from personal and collective nursing experience over the course of inpatient use. Despite the difference in calibration training and approach, most calibrations in both studies were completed within 5 minutes. Moreover, calibration resulted in improvements in MARD in both study groups.
Interestingly, calibration timing relative to the POC BG measurement did not influence MARD significantly. This observation should be approached with caution, because 67% of calibrations were performed within 5 minutes and 86% of calibrations were performed within 10 minutes of the POC BG measurements. However, it indicates that integration of calibration into routine nursing workflows is feasible even in resource-constrained settings. Moreover, calibration improved MARD and validation frequency even when performed > 5 minutes after the POC measurement. This finding requires confirmation in larger samples and other environments.
CGM accuracy is influenced by the rate of change in glucose levels, because rapid fluctuations can lead to inaccurate CGM readings as a result of the physiological lag between blood and interstitial glucose levels.28 Hence, manufacturer guidelines recommend against calibration in the presence of trend arrows indicating rapid sensor rate of change.29 Although we did not find a significant difference in frequency of validation or MARD after calibration by sensor rate of change at baseline, a larger sample size with a larger range in sensor rate of change is needed to confirm these findings.
Strengths include an evaluation of CGM performance across diverse cohorts of critically ill patients and comparison of both protocol-based and non-protocol-based calibration practices highlighting the practicality of integrating calibrations into routine clinical workflows. Another strength of our study is the availability of precise time stamps for POC BG measurements, interstitial glucose measurements, and manual calibrations.
The study is limited by its small sample size of 110 patients, with calibrations performed in only 55 patients, restricting the generalizability of findings. Although arterial blood gas often is considered the gold standard for ICU glucose measurement,30 we used the Nova StatStrip, which is approved by the US Food and Drug Administration for capillary blood glucose testing in critically ill patients, but may offer lower accuracy by comparison.31 Potential confounders, such as baseline characteristics, may have influenced outcomes. Differences in calibration protocols between cohorts present another limitation, but also a means of comparison. In addition, the findings may not be generalizable to other CGM devices, including those from different manufacturers or newer models with distinct calibration algorithms.
Interpretation
This study demonstrated that CGM calibration is feasible in the ICU and leads to improvement in MARD and sensor validation. These findings suggest that incorporating calibration into hybrid monitoring protocols may enhance CGM accuracy and may support further investigation of intermittent nonadjunctive CGM use in these settings. Additional research with larger, randomized cohorts is essential to confirm these findings and to refine protocols for broader clinical application.
Supplementary Material
Take-Home Points.
Study Question:
This feasibility study assessed whether sensor calibration improved continuous glucose monitor (CGM) accuracy in the ICU.
Results:
Calibration was shown to be feasible and associated with improvement in sensor validation for intermittent nonadjunctive use and mean absolute relative difference.
Interpretation:
The study suggested that point-of-care blood glucose calibrations may be useful to improve CGM accuracy in patients in the ICU.
Acknowledgments
Funding/Support
This study was supported by Dexcom LLC, The Ohio State University Clinical and Translational Science Institute, the National Center for Advancing Translational Sciences of the National Institutes of Health [Grant UM1TR004548], and the National Institute of Nursing Research of the National Institutes of Health [K23 Mentored Patient-Oriented Research Career Development Award K23NR020051]. M. M. is supported by the National Institutes of Health [Grant R01AG073408-01A1], the United States Department of Defense, and the Patient-Centered Outcomes Research Institute.
Financial/Nonfinancial Disclosures
The authors have reported to CHEST Critical Care the following: E. R. F. receives research funding from Dexcom, Inc., and Insulet; is a consult for Dexcom, Inc.; and receives honoraria from Dexcom, Inc., and Medscape. M. M. receives research funding from Dexcom, Inc., including salary support for research related to the current work, and royalties from Elsevier. K. M. D. receives research funding from Dexcom, Abbott, Insulet, Endogenex, Novo Nordisk, and Sequel Med Tech; consulting fees from Eli Lilly, Dexcom, Inc., Insulet, Oppenheimer, and Elsevier; honoraria from the Academy for Continued Healthcare Learning, Med Learning Group, Medscape, and Impact Education; and royalties from UptoDate. None declared (M. N. R., B. L., M. E., C. J. L., L. J., A. R.).
Role of sponsors:
The sponsor reviewed part of the study design and provided feedback but did not have a role in the final study protocol, data collection, analysis or preparation of the manuscript.
ABBREVIATIONS:
- CGM
continuous glucose monitor
- HbA1c
hemoglobin A1c
- IQR
interquartile range
- MARD
mean absolute relative difference
- POC BG
point-of-care blood glucose
Footnotes
This article was presented as a poster at the Ohio River Region Society of Endocrinology meeting, July 27, 2024, Columbus, Ohio.
Additional information: The e-Tables are available online under “Supplementary Data.”
References
- 1.Maiorino MI, Signoriello S, Maio A, et al. Effects of continuous glucose monitoring on metrics of glycemic control in diabetes: a systematic review with meta-analysis of randomized controlled trials. Diabetes Care. 2020;43(5):1146–1156. [DOI] [PubMed] [Google Scholar]
- 2.Grunberger G, Sherr J, Allende M, et al. American Association of Clinical Endocrinology clinical practice guideline: the use of advanced technology in the management of persons with diabetes mellitus. Endocr Pract. 2021;27(6):505–537. [DOI] [PubMed] [Google Scholar]
- 3.Karges B, Tittel SR, Bey A, et al. Continuous glucose monitoring versus blood glucose monitoring for risk of severe hypoglycaemia and diabetic ketoacidosis in children, adolescents, and young adults with type 1 diabetes: a population-based study. Lancet Diabetes Endocrinol. 2023;11(5):314–323. [DOI] [PubMed] [Google Scholar]
- 4.Faulds ER, Jones L, McNett M, et al. Facilitators and barriers to nursing implementation of continuous glucose monitoring (CGM) in critically ill patients with COVID-19. Endocr Pract. 2021;27(4):354–361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Chow KW, Kelly DJ, Gupta R, Miller JD. Use of continuous glucose monitoring to assess parenteral nutrition-induced hyperglycemia in an adult patient with severe COVID-19. JPEN J Parenter Enteral Nutr. 2021;45(1):208–211. [DOI] [PubMed] [Google Scholar]
- 6.Davis GM, Faulds E, Walker T, et al. Remote continuous glucose monitoring with a computerized insulin infusion protocol for critically ill patients in a COVID-19 medical ICU: proof of concept. Diabetes Care. 2021;44(4):1055–1058. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Agarwal S, Mathew J, Davis GM, et al. Continuous glucose monitoring in the intensive care unit during the COVID-19 pandemic. Diabetes Care. 2021;44(3):847–849. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Chow KW, Kelly DJ, Rieff MC, et al. Outcomes and healthcare provider perceptions of real-time continuous glucose monitoring (rtCGM) in patients with diabetes and COVID-19 admitted to the ICU. J Diabetes Sci Technol. 2021;15(3):607–614. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gómez AM, Henao DC, Muñoz OM, et al. Glycemic control metrics using flash glucose monitoring and hospital complications in patients with COVID-19. Diabetes Metab Syndr. 2021;15(2):499–503. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Longo RR, Elias H, Khan M, Seley JJ. Use and accuracy of inpatient CGM during the COVID-19 pandemic: an observational study of general medicine and ICU patients. J Diabetes Sci Technol. 2022;16(5):1136–1143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Sadhu AR, Serrano IA, Xu J, et al. Continuous glucose monitoring in critically ill patients with COVID-19: results of an emergent pilot study. J Diabetes Sci Technol. 2020;14(6):1065–1073. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Buschur EO, Faulds E, Dungan K. CGM in the hospital: is it ready for prime time? Curr Diab Rep. 2022;22(9):451–460. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Hoss U, Budiman ES. Factory-calibrated continuous glucose sensors: the science behind the technology. Diabetes Technol Ther. 2017;19(S2):S44–S50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Acciaroli G, Vettoretti M, Facchinetti A, Sparacino G. Calibration of minimally invasive continuous glucose monitoring sensors: state-of-the-art and current perspectives. Biosensors (Basel). 2018;8(1):24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Forlenza GP, Kushner T, Messer LH, Wadwa RP, Sankaranarayanan S. Factory-calibrated continuous glucose monitoring: how and why it works, and the dangers of reuse beyond approved duration of wear. Diabetes Technol Ther. 2019;21(4):222–229. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Wadwa RP, Laffel LM, Shah VN, Garg SK. Accuracy of a factory-calibrated, real-time continuous glucose monitoring system during 10 days of use in youth and adults with diabetes. Diabetes Technol Ther. 2018;20(6):395–402. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Shah VN, Laffel LM, Wadwa RP, Garg SK. Performance of a factory-calibrated real-time continuous glucose monitoring system utilizing an automated sensor applicator. Diabetes Technol Ther. 2018;20(6):428–433. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Nielsen CG, Grigonyte-Daraskeviciene M, Olsen MT, et al. Accuracy of continuous glucose monitoring systems in intensive care unit patients: a scoping review. Intensive Care Med. 2024;50(12):2005–2018. [DOI] [PubMed] [Google Scholar]
- 19.Faulds ER, Dungan KM, McNett M. Implementation of continuous glucose monitoring in critical care: a scoping review. Curr Diab Rep. 2023;23(6):69–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Dexcom. Can I calibrate Dexcom G7? Dexcom, Inc., website. Accessed December 12, 2024. https://www.dexcom.com/faqs/can-i-calibrate-dexcom-g7? [Google Scholar]
- 21.Visser MM, Vangoitsenhoven R, Gillard P, Mathieu C. Diabetes technology in the hospital: an update. Curr Diab Rep. 2024;24(8):173–182. [DOI] [PubMed] [Google Scholar]
- 22.Spanakis EK, Cook CB, Kulasa K, et al. A consensus statement for continuous glucose monitoring metrics for inpatient clinical trials. J Diabetes Sci Technol. 2023;17(6):1527–1552. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Korytkowski MT, Muniyappa R, Antinori-Lent K, et al. Management of hyperglycemia in hospitalized adult patients in non-critical care settings: an Endocrine Society clinical practice guideline. J Clin Endocrinol Metab. 2022;107(8):2101–2128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Shaw JLV, Bannuru RR, Beach L, et al. Consensus considerations and good practice points for use of continuous glucose monitoring systems in hospital settings. Diabetes Care. 2024;47(12):2062–2075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Ward WK, Engle JM, Branigan D, El Youssef J, Massoud RG, Castle JR. The effect of rising vs. falling glucose level on amperometric glucose sensor lag and accuracy in type 1 diabetes. Diabet Med. 2012;29(8):1067–1073. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Moser O, Riddell MC, Eckstein ML, et al. Glucose management for exercise using continuous glucose monitoring (CGM) and intermittently scanned CGM (isCGM) systems in type 1 diabetes: position statement of the European Association for the Study of Diabetes (EASD) and of the International Society for Pediatric and Adolescent Diabetes (ISPAD) endorsed by JDRF and supported by the American Diabetes Association (ADA). Pediatr Diabetes. 2020;21(8):1375–1393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Faulds ER, Dungan KM, McNett M, et al. Nursing perspectives on the use of continuous glucose monitoring in the intensive care unit. J Diabetes Sci Technol. 2023;17(3):649–655. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Freckmann G. Basics and use of continuous glucose monitoring (CGM) in diabetes therapy. Journal of Laboratory Medicine. 2020;44(2):71–79. [Google Scholar]
- 29.Dexcom. Understanding accuracy with your Dexcom G6 and Dexcom ONE continuous glucose monitoring system. Dexcom, Inc., website. Accessed December 12, 2024. https://s3.us-west-2.amazonaws.com/dexcompdf/OUS+Specific+PDFs/MAT-4868_understanding_accuracy_with_your_dexcom_g6_and_dexcom_one_cgm_system.pdf [Google Scholar]
- 30.Evans L, Rhodes A, Alhazzani W, et al. Surviving Sepsis Campaign: international guidelines for Management of Sepsis and Septic Shock 2021. Crit Care Med. 2021;49(11):e1063–e1143. [DOI] [PubMed] [Google Scholar]
- 31.Klonoff DC, Umpierrez GE, Rice MJ. A milestone in point of care capillary blood glucose monitoring of critically ill hospitalized patients. J Diabetes Sci Technol. 2018;12(6):1095–1100. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.