Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation

Abstract

Background:

Adequate glucose control is an important, yet a challenging task, in the early postoperative care after liver transplantation (LTx). Potential of continuous glucose monitoring (CGM) in this context had not been fully explored because of concerns about sensors’ precision in critical care.

Objectives:

We initiated a trial to assess feasibility, accuracy, and benefit of CGM used in addition to standard blood glucose (BG) management.

Design:

Prospective randomized trial.

Methods:

Patients undergoing LTx were included and randomized to wearing (1) unblinded (“active”) CGM which could help guiding insulin therapy, (2) blinded CGM serving as controls. Dexcom G6 was applied immediately after surgery, arterial BG values measured with blood gas analyzer served as reference and for calibration. We evaluated mean absolute relative difference (MARD), Diabetes Technology Society (DTS) Error Grid zones, bias, 15/15 agreement rate, evolvement of accuracy over time, and possible interfering factors. Fisher test, Mann–Whitney test, DTS software, and linear mixed model were used for statistical analysis.

Results:

We included 155 LTx recipients, 13 were excluded (1 died during surgery, 12 experienced sensor failure—6 from each group), data from 142 patients (80 in active and 62 in blinded CGM group) were analyzed. Overall, MARD was 9.1% in active and 10.7% in the blinded group (p < 0.0001). DTS error grid in active group had shown 90% in zone A versus 85.3% in the blinded group. These results were consistent for reference glucose ranges 3.9–10 mmol/L and above 10 mmol/L, with less precision in the low ranges (these values were however rare). Sensor accuracy peaked on days 2–3 and deteriorated over time with significant worsening from day 7 on in both groups.

Conclusion:

Our study demonstrates feasibility and acceptable accuracy of CGM comparable to reports from the outpatient care. We believe this could be attributed to sensor calibration, reflected by more favorable outcomes in the active CGM group. In addition, we demonstrate decline in accuracy over time, suggesting their use in critical care could be limited to less days to secure sufficient reliability.

Trial registration:

This study is part of an ongoing prospective trial registered on ClinicalTrials.gov (NCT05585801) and was registered on October 12, 2022.

Keywords

accuracy continuous glucose monitoring intensive care liver transplantation

Introduction

Liver transplantations (LTx) save more than 30,000 lives each year worldwide¹ and these patients depend on accurate in-hospital blood glucose (BG) monitoring as a critically important factor in their recovery.²

Even though the actual target glucose ranges for intensive and postoperative care have long been a matter of debate, substantial dysglycemia undoubtedly worsens the postoperative outcomes.^3–5 Patients undergoing LTx are not an exception.² On the contrary, their critical state and effect of immunosuppression (which in most cases includes corticosteroids) often complicate glucose control and there is evidence that hyperglycemia can be related to both short-term and long-term adverse outcomes.^6–10 Furthermore, some of them often have diabetes pretransplant, while another nonneglectable portion might develop impaired glucose control or diabetes after transplantation because of immunosuppressive therapy with steroids and calcineurin inhibitors.¹¹

Current recommendations for glucose control in these patients rely on frequent BG measurements and protocols for continuous intravenous insulin administration.^12,13 Despite the attempts for maximum safety and reliability of the analytical methods for glucose measurement and insulin dosage, navigating between glucose fluctuations enhanced by these patients’ critical state is often challenging and brings unsatisfactory results.^14–16

While in the outpatient care, introduction of continuous glucose monitoring (CGM) has completely transformed the care for patients with diabetes,¹⁷ glucose monitoring in the intensive care setting currently lags behind and relies solely on certified glucose meters, missing the potential of capturing trends, possibility of tailored insulin therapy, and parenteral nutrition, or alarm notification at critical glucose values.^12,13

A recent scoping review of CGM accuracy in intensive care¹⁸ reports substantial diversity between different sensors as it includes also older studies, various protocols and comparator methods, and explores a very heterogeneous patient population. Small studies (mostly from the peak of the COVID-19 pandemic)^19–24 had shown promising results of current CGM systems in critical state and additional supportive evidence comes from the perioperative trials^25–27 with CGM and even closed loop insulin pump therapy.^28,29 Although CGM accuracy did not meet the requirements for certified BG meters, the applicability and clinical accuracy were acceptable in most cases. Despite slight lack of precision of measurement, the potential of real-time monitoring of glycemic trends and using alarms provided a new benefit, in addition to the possibility of remote monitoring of infectious patients.

Therefore, we decided to test implementation of CGM use in a specific group of liver transplant recipients, to assess CGM applicability, accuracy, reliability, and impact on glucose control in this extremely complex and high-risk patient population. We report detailed sensor accuracy outcomes and the effect of possible interfering factors.

Methods

We initiated a prospective randomized trial (ClinicalTrials.gov No. NCT05585801) including liver transplant recipients who required intensive postoperative care at the Department of Anesthesiology, Resuscitation and Intensive Care and Surgical Intensive Care unit (ICU) at the Institute for Clinical and Experimental Medicine, Prague between December 2022 and June 2024. The study was approved by the local Ethical Committee of IKEM and Thomayer Hospital (document no. 09558/22, G-22-13) and was in concordance with the Declaration of Helsinki. While the main study endpoint was the effect on glycemic control, in this first analysis, we focused on the feasibility and accuracy outcomes after 2 years of the study course.

Eligibility and randomization

We included patients undergoing LTx from a deceased donor, older than 18 years, requiring postoperative intensive care who agreed to participate in the study. The exclusion criteria included any condition that precluded the planned surgical procedures.

After providing signed informed consent before surgery, patients were randomized to two groups:

Study group wearing a nonblinded rtCGM (real-time CGM) device added to standard therapy, enabling nurses and physicians to see real-time glucose values from the CGM and adapt the therapy. (For simplicity, this study group will be referred to as “active CGM group.”)

Control group wearing blinded rtCGM paired with a receiver (which will be referred to as “blinded CGM group.”)

Randomization was performed before surgery, using the sealed envelope method according to the randomization list generated by the study statistician.

Power analysis

Power analysis was based on the pilot study including 26 patients after major abdominal surgery, 11 of them wearing a blinded rtCGM. We tested time in range (percent) in both groups, mean target time in range (6–10 mmol/L) was 66.6% (SD of 16.46) and 69.1% (SD of 14.19) in the blinded and open groups, respectively. Types I and II errors were set to 0.05 and 0.10, respectively. Target difference between time in range was supposed to be 7.5% and required numbers of patients were set for 60 in the blinded group and 82 in the nonblinded group.

Glucose monitoring and insulin therapy

A Dexcom G6 sensor and transmitter (Dexcom Inc., San Diego, CA, USA) was applied by a trained nurse immediately after surgery and paired with a Dexcom receiver allowing a blinded mode or an iPhone for unblinded readings. Based on our previous studies,^30,31 we applied the sensor in an alternative site in the infraclavicular region. We assumed that this approach would be more suitable for the needs of intensive care and less prone to dislocation or compression artifacts in sedated patients.

In the active CGM group, the rtCGM glucose values served as an additional guide for insulin administration, along with standard monitoring using a blood gas analyzer (Radiometer ABL 800; Copenhagen, Denmark) and a StatStrip BG meter (Nova Biomedical Corporation, Waltham, MA, USA). Data collected from the blinded CGM group served as controls for monitoring the outcomes of current standard therapy. Continuous intravenous insulin was administered according to local approved protocol, with a target glucose range of 6–10 mmol/L.

Reference methods and calibration

Glucose values from arterial blood samples measured with a blood gas analyzer served as reference values for our assessment and sensor calibration. Even though the Dexcom G6 sensor is factory calibrated and (according to the manufacturer) does not require recalibration, we elected to perform additional sensor calibrations, expecting that the unstable postoperative patient conditions might pose a greater challenge to maintain adequate accuracy compared to the outpatient use. All sensors were calibrated four times on the first day (every 6 h) and then once daily (every 24 h) on the following 2 days. In addition, the nurses were instructed to recalibrate the nonblinded sensors in the active CGM group, if the difference between glucose values measured with CGM and the reference method exceeded 1.5 mmol/L.

Accuracy assessment

We calculated (1) overall mean absolute relative difference (MARD), (2) percentage in different zones of the new DTS Error grid,³² (3) relative error ranges, and (4) bias in the two study groups, with respect to different calibration approach. (For more details about the CGM accuracy assessment methods, we refer to Freckmann et al.³³ scoping review).

MARD was also computed separately for three different ranges of the reference glucose values: (a) <3.9 mmol/L; (b) between 3.9 and 10 mmol/L; and (c) >10 mmol/L.

For comparison, StatStrip glucose meter accuracy was also evaluated, using the blood gas analyzer glucose values as reference.

In addition, we assessed sensor accuracy change over time. MARD data were analyzed across four predefined time intervals: day 1, days 2–3, days 4–6, and day 7 on, as we expected the most significant changes would occur in these intervals.

Evaluation of the impact of possible interfering factors

Based on current evidence on possible modifying factors, we also analyzed the effect of acetaminophen in patients treated with doses larger than 1 g per day, compared with those who did not receive this treatment. Additionally, we evaluated if computer tomography (CT) scan and additional surgical revision (with risk of signal disturbances in the operation room or during transport) would impair the sensor accuracy and analyzed MARD 48 h before and after these procedures in individual patients. We also performed a post hoc analysis focusing on possible effect of vasopressor therapy on CGM precision.

Statistical methods

Descriptive data are presented as median and interquartile range or mean and standard deviation for continuous variables and counts for categorical variables. For clinical accuracy analysis, we compared the paired glucose values from CGM and blood gas analyzer, using free available software provided by DTS website.³⁴ GraphPad Prism version 10 (GraphPad Software, San Diego, CA, USA) was used for basic statistics, applying Mann–Whitney test, t-test, and Kruskal–Wallis test for comparison of numerical data (based on distribution), Fisher test, and Chi-square test for categorical data.

For sensor lifetime assessment, MARD data were analyzed separately for each group using a linear mixed model implemented in Jamovi software (version 2.6.44; GAMLj v3, Sydney, Australia) with time modeled as a four-level categorical factor (Day 1, Days 2–3, Days 4–6, Day 7 on). MARD was included as the dependent variable in the model, with a fixed effect of the time factor and a random intercept for patients to account for individual variability. Model parameters were estimated using Restricted Maximum Likelihood, and degrees of freedom were approximated using the Satterthwaite method. Statistical significance (p < 0.05) in post hoc pairwise comparisons (t tests) for the categorical time factor was Bonferroni-corrected.

Results

Patient demographics and postoperative course

We included and randomized 155 patients, 88 in the active and 66 in the blinded CGM group. Twelve patients were excluded (six in each arm) because of technical failures and hematoma formation at the insertion site within first hours of monitoring (see CONSORT diagram in Supplemental Figure 1). Data from 142 patients were analyzed, 82 in the active and 60 in the blinded group.

Baseline demographics and data describing the perioperative course and ICU stay are presented in Table 1. All patients required vasopressors and ventilatory support after the operation and all needed intravenous insulin therapy in the first days (because of their critical state, corticosteroid therapy, and intravenous parenteral nutrition).

Table 1.

Baseline characteristics, demographics, comorbidities, perioperative, and ICU data.

Characteristic	Total	CGM	Standard	p Value
Patients (n)	142	82	60
Age (years)	60 (46; 67)	60 (46; 67)	61.5 (45.5; 68)	0.55
Male gender (n/%)	91/63.7	54/64.8	37/62.8	0.61
BMI (kg/m²)	27.7 (23.7; 30.7)	27.3 (23.7; 30.4)	27.8 (24; 33.3)	0.33
HbA_1c (mmol/mol)	32 (27; 39)	31 (27; 38)	33 (28; 39.5)	0.2
APACHE II score	13 (10; 16)	12.5 (10; 15)	13 (10; 16.5)	0.99
SOFA score	10 (8; 11)	10 (8; 11)	10 (8; 10)	0.71
Child–Pugh score A/B/C (n//%)	38/59/45//27/41/32	23/33/26//28/40/32	15/26/19//25/43/32	0.7/0.7/1
MELD score	17 (13; 22)	16.5 (12; 22)	17 (13; 21)	0.96
Diabetes mellitus type 1/2 (n//%)	3/37//2.1/26	2/24//2.4/29	1/13//1.7/22	0.75/0.31
Chronic pulmonary disease (n/%)	12/1.4	8/10	4/6.8	0.51
Cancer (n/%)	44/30.8	27/32.4	17/29	0.6
Ischemic heart disease (n/%)	37/26	22/26.4	15/26	0.8
Arterial hypertension (n/%)	84/59	43/51.6	41/69.7	0.06
Heart failure (n/%)	11/7.7	4/4.8	7/12	0.13
Stroke or TIA (n/%)	1/0.7	1/1.2	0	0.39
Type of surgery (n/%)
Liver transplantation (n/%)	132/92.4	77/92.4	55/93.5
Liver and kidney transplantation (n/%)	10/7	5/6	5/8.5
ICU/hospital stay (days)	9 (7; 12)/18.5 (15; 26)	9 (7; 13)/18 (15; 25)	8 (7; 12)/19.5 (15; 26.5)	0.45/0.63
Surgery duration (min)	305 (250; 360)	305 (245; 370)	295 (253; 360)	0.99
Blood loss (mL)	2100 (1000; 4000)	2500 (1000; 4000)	2000 (1000; 4000)	0.94
Norepinephrine (n/%)	142/100	82/100	60/100	1
Duration (h)	28 (14.5; 62)	28 (16; 61)	27 (12.8; 66)	0.84
Maximal dose (µg/kg/min)	0.28 (0.2; 0.5)	0.25 (0.18; 0.5)	0.3 (0.2; 0.37)	0.98
Average dose (µg/kg/min)	0.08 (0.05; 0.12)	0.08 (0.05; 0.13)	0.08 (0.04; 0.11)	0.27
Argipressin (n/%)	12/8.7	8/9.6	4/6.8	0.52
Duration (h)	16.3 (7.7; 33)	16.3 (5.5; 33)	16 (7.6; 59)	0.58
Average dose (I.U./h)	1.65 (1.33; 2.2)	1.45 (1.3; 1.88)	2.2 (1.9; 2.8)	0.58
Mechanical ventilation (n/%)	142/100	82/100	60/100	1
Duration (h)	11 (7; 28)	12 (7; 32.5)	10 (7; 23)	0.47
Continuous renal replacement therapy (n/%)	18/12.6	8/9.6	10/17	0.23
Duration (h)	70 (55.5; 89.3)	72.5 (33; 108)	68.5 (60.8; 87)	0.22
Cumulative fluid balance (mL)	4624 (2477; 7525)	4674 (1781; 7706)	4562 (2934; 7276)	0.88
Acetaminophen N (%)//average dose (g/day)	99/69//2 (1; 3)	57/68.4//2 (1; 3)	42/71.4//2 (1; 3)	0.96
Induction immunosuppression
Methylprednisolone (n/%)//dose (mg)	142/100//500	82/100//500	60/100//500	1
Basiliximab (n/%)	17/12	8/9.6	9/9.7	0.35

Categorical data are expressed as n/%, numerical data as median (interquartile range).

APACHE II, acute physiology and chronic health evaluation II score; BMI, body mass index; HbA1c, glycated hemoglobin; ICU, intensive care unit; MELD, model for end-stage liver disease score; SOFA, Sequential Organ Failure Assessment score; TIA, transitory ischemic attack.

CGM accuracy and reliability

CGM was used in 100% versus 99.7% of time in the active versus blinded CGM group. We encountered 12 cases of failure; transmitter failures in 5 patients, 7 cases of hematoma formed at the insertion site, disabling sensor function (due to severe coagulopathy). These patients were therefore not included in the analysis.

Sensor MARD reached 9.1% in the active CGM group and was significantly lower compared to MARD of 10.7% in the blinded group (p < 0.0001). Accuracy in different ranges of reference glucose values (see Table 2) was consistent in values between 3.9 and 10 mmol/L and above 10 mmol/L, but less precise in the low glucose values. In the active CGM group, 10 glucose values between 3 and 3.9 mmol/L were captured by blood gas analyzer, 3 were missed by CGM, which was showing values at the low border of normal range and one value under 3 mmol/L undetected by CGM (promptly treated with no severe consequences). In the blinded group, no reference glucose value reached level 2 hypoglycemia and out of 7 values between 3 and 3.9 mmol/L, 3 were missed by CGM.

Table 2.

CGM accuracy outcomes—overall and in different reference glucose value ranges.

Reference BG range	N/%	MARD	15/15 AR (%)	Bias (%)	MAD (mmol/L)
Active CGM
Total	1894	9.1%	82.4%	1.6%
<3.9 mmol/L	11/0.6%	20.3%	72.7%	17.6%	0.65
3.9–10 mmol/L	1481/78.2%	9%	83.3%	2.6%
>10 mmol/L	402/21.2%	8.9%	81.1%	−2.7%
Blinded CGM
Total	1573	10.7%	76.4%	0.3%
<3.9 mmol/L	7/0.4%	13.4%	85.7%	12.6%	0.5
3.9–10 mmol/L	1188/75.5%	10.9%	76%	0.8%
>10 mmol/L	378/24%	10.3%	78%	−1.2%

Reference BG values from blood gas analyzer (ABL Radiometer), Active CGM, MARD, 15/15 (proportion of values within 15% of reference for glucose >5.6 mmol/L and within 0.83 mmol/L of reference for glucose below 5.6 mmol/L), MAD (only for blood glucose values under 3.9 mmol/L).

Boldface font indicates statistical significance where p < 0.05.

Active CGM, open-labelled continuous glucose monitoring arm; AR, agreement rate; BG, blood glucose; MAD, mean absolute difference; MARD, mean absolute relative difference.

Clinical accuracy analyzed by DTS error grid is reported in Table 3 and Figure 1. In the active CGM group 90% of evaluated glucose pairs were in zone A (representing no clinical risk), compared to 85.3% pairs in zone A in the blinded group, 9.8% versus 14.2% pairs were in zone B (mild clinical risk), in the active versus blinded group, respectively. The percentage in other zones with higher risk was rather low—0.2% in zone C (moderate risk) in the active group, 0.4% in zone C, and 0.1% in high-risk zone D in the blinded group.

Table 3.

CGM clinical accuracy evaluated by DTS error grid.

Risk zone	Count		Frequency (%)
Risk zone	Active	Blinded	Active	Blinded
(A) No risk	1705	1341	90.0	85.3
(B) Mild risk	186	224	9.8	14.2
(C) Moderate risk	3	7	0.2	0.4
(D) High risk	0	1	0.0	0.1
(E) Extreme risk	0	0	0.0	0.0
Not included	0	0	0.0	0.0

CGM, continuous glucose monitoring; DTS, diabetes technology society.

Figure 1.

DTS error grid graph for active CGM group (a) and blinded CGM group (b). Risk zones: A, no risk; B, mild risk; C, moderate risk; D, High risk, E, extreme risk.

Bias was 1.6% in the active CGM group and 0.3% in the blinded group. Relative error ranges according to ISO and FDA criteria are shown in Supplemental Tables 1–4. Stat Strip glucose meter had MARD of 6.2% in the active and 5.9% in the blinded group. (For more details, see Supplemental Tables 7 and 8).

Sensor lifetime evaluation

Analysis of sensor lifetime and accuracy evolvement (see Table 4) showed that sensors were the most precise on days 2–3 (MARD 7.7% in active and 9.7% in blinded group) and then deteriorated over time, reaching MARD of 12.2% in active group versus 12.1% in the blinded group from day 7 on. While in the active CGM group this trend was significant already between days 4 and 6, it became most apparent and gained statistical significance from day 7 on in both study groups (p = 0.001 in active, p = 0.02 in blinded group). MARD on day 1 was 8.6% in the active and 10.5% in blinded CGM group.

Table 4.

CGM lifetime and accuracy in active and blinded groups across time intervals, analyzed by linear mixed model with Bonferroni-corrected post hoc t tests (p < 0.05).

Characteristic	Active CGM	p Value	Blinded CGM	p Value
Length of CGM monitoring in the ICU (days)	9 (8; 10)		8.5 (7; 10)
Sensor usage (%)	100 (99.8; 100)		99.7 (98.6; 99.8)
MARD day 1 (%)	8.6		10.5
MARD days 2–3 (%)	7.7	0.26	9.7	0.38
MARD days 4–6 (%)	9.2	0.006	11.2	0.14
MARD day 7 on (%)	12.2	0.001	12.1	0.02

Boldface font indicates statistical significance where p < 0.05.

Active CGM, open-labelled continuous glucose monitoring arm; ICU, intensive care unit, MARD, mean absolute relative difference.

Interfering factors and calibrations

Therapy with acetaminophen (1 g/day and more—see Table 1) did not seem to impair sensor accuracy (see Supplemental Table 5) and neither did CT scan (see Supplemental Table 6). There was a slight accuracy deterioration in the blinded CGM group after surgical revision, not detected in the open-labelled arm. In addition to these known interfering factors, we also explored whether the use of vasopressors might have any substantial impact on CGM accuracy.

As all patients in our cohort were treated with norepinephrine for at least a short period of time early after the operation, we compared CGM accuracy (MARD) at three different norepinephrine infusion rates: (a) ⩽0.1 µg/kg/min, (b) 0.11–0.2 µg/kg/min, and (c) >0.2 µg/kg/min (see Supplemental Table 9). We then compared MARD across these three dose categories within each study arm, as well as between study arms at the same infusion rate.

In the active CGM arm, we observed a dose-dependent decrease in CGM accuracy with increasing vasopressor requirement, which reached statistical significance. In contrast, this trend was not observed in the blinded arm.

In the active CGM group, the sensors were additionally recalibrated (in cases of glucose value difference >1.5 mmol/L) on average once during the first 3 days, and two more times between days 4 and 10.

Adverse events

We observed seven cases of hematoma formation disabling sensor function (n = 4 in the active and n = 3 in the blinded group). Based on previous experience, we applied barrier skin film solution before sensor insertion to prevent potential skin reactions, nevertheless, six cases of mild skin irritation at the sensor insertion site occurred (n = 4 in the active and n = 2 in the blinded group). No serious adverse events were observed.

Discussion

We present detailed accuracy outcomes of a prospective trial testing the Dexcom G6 sensor in the early postoperative days after LTx. Sensors were applied to an unconventional infraclavicular site which we previously evaluated and noted to be more feasible for the ICU setting.^30,31 In this analysis, we focused solely on sensor accuracy and differences between blinded and unblinded monitoring (which allowed additional recalibrations).

Current data on CGM use in the ICU setting show a wide range of conflicting results regarding CGM feasibility and accuracy.¹⁸ Especially earlier generations of sensors failed to provide satisfactory results. On the other hand, recent small studies of perioperative CGM use,^25–27 as well as reports evaluating exceptional applications of CGM in intensive care during the COVID-19 pandemic,^19–24 demonstrate that the accuracy of contemporary CGM systems has improved and that they might be suitable for broader use beyond their originally intended outpatient setting, including in the ICU. However, these findings are limited by heterogeneous patient populations and predominantly retrospective study designs. To our knowledge, no trial to date has involved liver transplant recipients in any CGM evaluation. Given the practical challenges of early posttransplant glucose management in these patients, we therefore decided to assess CGM implementation in a prospective trial involving this homogeneous and specific patient group.

Overall, sensor MARDs of 9.1% in the active arm and 10.7% in the blinded arm are comparable both to the previous reports on Dexcom G6 in stable physiological conditions in the outpatient setting³⁵ and to the results from perioperative studies.^26,27 Accordingly, clinical accuracy evaluated by DTS error grid was satisfactory in both groups, albeit better in the open-labelled CGM arm (90% vs 85.3% in zone A for active vs blinded group).

In a more detailed analyses of accuracy throughout different ranges of the reference BG values, the results were consistent with the overall MARD for ranges between 3.9 and 10 mmol/L and above 10 mmol/L. As for the low BG range (below 3.9 mmol/L), the accuracy was notably lower, but we assessed only 11 pairs of values in the active group (including one undetected level 2 hypoglycemia) and 7 pairs in the blinded group (less than 1% of the total count throughout all ranges), as hypoglycemic episodes were rare because of careful frequent monitoring by nursing staff.

Our evaluation of sensor lifetime brought new insight into the accuracy evolution over time. We observed best precision between days 2 and 3 with a gradual decline. This deterioration became clearly significant from day 7 on in both groups. On the other hand, CGM readings were reliable on the first day. These findings are in contrast with previous evidence³⁶ suggesting worse accuracy after sensor initiation, which improves and stabilizes over time. This level of accuracy on day 1 could be attributed to frequent calibration according to the protocol (every 6 h) in both study groups. These lifetime trends raise the question whether the optimal length of Dexcom G6 in the context of critical care could be shorter than recommended for outpatient care to achieve better precision and reliability.

We must admit that this analysis of accuracy over sensor lifetime was not a predefined study endpoint; it was not included in the initial statistical considerations and power analysis.

This question arose while we were gathering more clinical experience with CGM use, and we therefore decided to explore and assess it in a retrospective, exploratory manner. As a result, the evidence from our findings is certainly not strong enough to support any official recommendations for clinical practice. However, we believe it highlights the need for further prospective studies focusing specifically on changes in accuracy over time.

We decided to use arterial BG values from a blood gas analyzer as the reference measurement. We are aware that the precision of blood gas analyzers varies greatly, and this protocol might not be applicable worldwide. Accuracy of different glucose measuring methods in critical care is still debated and can be impaired by many different interfering factors.³⁷ Based on previous evidence and internal validation outcomes, we considered this method more accurate than a point-of-care BG meter, while still practical for frequent routine sampling by the nursing staff.

Despite Dexcom G6 being already factory-calibrated, we performed routine recalibrations during the first 3 days. In addition, the open-labelled sensors were recalibrated if the difference of glucose values between CGM and reference exceeded 1.5 mmol/L. Throughout the study, these sensors were additionally recalibrated (in cases of glucose value difference >1.5 mmol/L) on average 3 times, on average once during the first 3 days, and two more times between days 4 and 10. We believe that this different approach resulted in better accuracy outcomes in the active CGM group and might have affected the lifetime trends reported above. However, the actual calibration protocol needs to be further explored and validated and might differ between various CGM systems. In addition, we are aware that our methods—such as the specific calibration protocol and the use of an arterial blood gas analyzer accurate enough to serve as a reliable reference—might not be applicable to all ICU settings, clinical conditions, or scenarios.

Some of our patients underwent CT scans and surgical revisions. While the CT scans did not seem to affect the sensor accuracy, mild deterioration was observed after surgical revision in the blinded CGM group. This difference might also be due to the different calibration approach. Acetaminophen (previously reported as potential interfering with the glucose-oxidase based sensors³⁸) did not affect the accuracy in either group.

As intensive care poses different challenges to CGM precision than the outpatient setting and introduces the possibility that new interfering factors may emerge, we also assessed the potential impact of vasopressors and peripheral tissue hypoperfusion. In the active arm, sensor accuracy decreased with increasing norepinephrine dose; however, this trend was not apparent in the blinded arm. This difference might be explained by the different calibration approach and by the possibility that reduced calibration frequency in the blinded arm masked the additional effect of norepinephrine therapy. Nevertheless, the question of the vasopressor effect on CGM accuracy clearly warrants further research.

Our study lacks CGM measurements during transplant surgery itself. We deliberately chose this approach because of previous experience of signal loss and other sensor disturbances in the operating room during pilot testing. In addition, the preparation for LTx is often a time-limited acute situation, not providing enough time for sensor warm-up and adaptation before substantial glucose excursions occur during the anhepatic and neohepatic phase of LTx. Therefore, we believed that the first hour after operation would be a more suitable time for sensor application and its further practical use.

Another limitation of our study is the loss of data from the 12 malfunctioning sensors. These sensors had to be excluded from the analysis, which might have affected the results.

Our patient cohort was specific. Liver transplant recipients are rarely included in glucose monitoring clinical studies because of their unstable clinical state and complex needs.² On the other hand, this group of patients allowed us to evaluate CGM feasibility and accuracy in highly unphysiological conditions, extreme even for the intensive care setting. Furthermore, liver transplant recipients experience substantial BG fluctuations, and this problem had not been sufficiently addressed so far.² Even though the accuracy of Dexcom G6 did not meet the criteria recommended for point-of-care BG meters,³⁹ clinical accuracy turned out to be satisfactory and the CGM provided insight into the dynamics and trends of BG changes. We hypothesize that these results would prove to be similar in other patient populations requiring critical care.

Conclusion

CGM with the Dexcom G6 sensor was feasible and clinically accurate even in a complicated group of patients treated at the ICU after LTx. Our findings also support the benefit of additional sensor calibration and applicability of the alternative infraclavicular sensor placement. The protocol for calibration and insulin dose adjustment based on CGM data as well as potential shortening of the number of days of sensor use to achieve better precision still require more evidence for CGM to be safe and reliable for routine practice. Nevertheless, if properly validated using a suitable reference method, we believe that implementing CGM along with standard BG monitoring can facilitate reaching desired glycemic targets and improving the quality of care.

Supplemental Material

sj-docx-1-tae-10.1177_20420188251405372 – Supplemental material for Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation

Supplemental material, sj-docx-1-tae-10.1177_20420188251405372 for Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation by Marek Protus, Barbora Voglová Hagerf, Antonin Jabor, Janka Franekova, Lenka Nemetova, Eva Uchytilova, Veronika Indrova, Jana Beckova, Alex Macek, Martina Doleckova, Veronika Svirlochova, Milos Mraz, Martin Haluzik, Peter Girman, Martina Viravova, Michael A. Kohn, David C. Klonoff and Eva Kieslichova in Therapeutic Advances in Endocrinology and Metabolism

Supplemental Material

sj-docx-2-tae-10.1177_20420188251405372 – Supplemental material for Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation

Supplemental material, sj-docx-2-tae-10.1177_20420188251405372 for Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation by Marek Protus, Barbora Voglová Hagerf, Antonin Jabor, Janka Franekova, Lenka Nemetova, Eva Uchytilova, Veronika Indrova, Jana Beckova, Alex Macek, Martina Doleckova, Veronika Svirlochova, Milos Mraz, Martin Haluzik, Peter Girman, Martina Viravova, Michael A. Kohn, David C. Klonoff and Eva Kieslichova in Therapeutic Advances in Endocrinology and Metabolism

Footnotes

Acknowledgements

The authors give special thanks for all support from the Transplantation Surgery Department of Institute for Clinical and Experimental Medicine, and the nursing teams of the Anesthesiology, Resuscitation and Intensive Care Department and Intensive Care Ward of the Transplantation Surgery Department.

Declarations

ORCID iDs

Marek Protus

Barbora Voglová Hagerf

Antonin Jabor

Janka Franekova

Lenka Nemetova

Eva Uchytilova

Veronika Indrova

Jana Beckova

Alex Macek

Veronika Svirlochova

Milos Mraz

Peter Girman

Michael A. Kohn

David C. Klonoff

Eva Kieslichova

Supplemental material

Supplemental material for this article is available online.

References

Terrault

Francoz

Berenguer

, et al. Liver transplantation 2023: status report, current and future challenges. Clin Gastroenterol Hepatol 2023; 21(8): 2150–2166.

Alvarez-Sotomayor

Satorres

Rodríguez-Medina

, et al. Controlling diabetes after liver transplantation: room for improvement. Transplantation 2016; 100(10): e66–e73.

Van den Berghe

Wilmer

Milants

, et al. Intensive insulin therapy in mixed medical/surgical intensive care units. Diabetes 2006; 55(11): 3151–3159.

Van den Berghe

Wouters

Weekers

, et al. Intensive insulin therapy in critically ill patients. N Engl J Med 2001; 345(19): 1359–1367.

de Vries

FEE

Gans

Solomkin

, et al. Meta-analysis of lower perioperative blood glucose target levels for reduction of surgical-site infection. Br J Surg 2017; 104(2): e95–e105.

Wallia

Parikh

Molitch

, et al. Posttransplant hyperglycemia is associated with increased risk of liver allograft rejection. Transplantation 2010; 89(2): 222–226.

Wallia

Parikh

O’Shea-Mahler

, et al. Glycemic control by a glucose management service and infection rates after liver transplantation. Endocr Pract 2011; 17(4): 546–551.

Wallia

Schmidt

Oakes

, et al. Glycemic control reduces infections in post-liver transplant patients: results of a prospective, randomized study. J Clin Endocrinol Metab 2017; 102(2): 451–459.

Ammori

Sigakis

Englesbe

, et al. Effect of intraoperative hyperglycemia during liver transplantation. J Surg Res 2007; 140(2): 227–233.

10.

Wang

Wei

Zhou

, et al. Hyperglycemia aggravates acute liver injury by promoting liver-resident macrophage NLRP3 inflammasome activation via the inhibition of AMPK/mTOR-mediated autophagy induction. Immunol Cell Biol 2020; 98(1): 54–66.

11.

Brodosi

Petta

Petroni

, et al. Management of diabetes in candidates for liver transplantation and in transplant recipients. Transplantation 2022; 106(3): 462–478.

12.

American Diabetes Association Professional Practice Committee. 16. Diabetes Care in the Hospital: Standards of Care in Diabetes-2025. Diabetes Care 2025; 48(Suppl. 1): S321–S334.

13.

Honarmand

Sirimaturos

Hirshberg

, et al. Society of critical care medicine guidelines on glycemic control for critically ill children and adults 2024. Crit Care Med 2024; 52(4): e161–e181.

14.

Roth

Sommerfeld

Birkenfeld

, et al. Blood sugar targets in surgical intensive care. Dtsch Ärztebl Int 2021; 118(38): 629–636.

15.

Vogt

Bally

Perioperative glucose management: current status and future directions. Best Pract Res Clin Anaesthesiol 2020; 34(2): 213–224.

16.

Preiser

Chase

Hovorka

, et al. Glucose control in the ICU: a continuing story. J Diabetes Sci Technol 2016; 10(6): 1372–1381.

17.

Galindo

Aleppo

Continuous glucose monitoring: the achievement of 100 years of innovation in diabetes technology. Diabetes Res Clin Pract 2020; 170: 108502.

18.

Nielsen

Grigonyte-Daraskeviciene

Olsen

, et al. Accuracy of continuous glucose monitoring systems in intensive care unit patients: a scoping review. Intensive Care Med 2024; 50(12): 2005–2018.

19.

Sadhu

Serrano

, 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.

20.

Agarwal

Mathew

Davis

, et al. Continuous glucose monitoring in the intensive care unit during the COVID-19 pandemic. Diabetes Care 2021; 44(3): 847–849.

21.

Chow

Kelly

Rieff

, 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.

22.

Davis

Faulds

Walker

, 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.

23.

Faulds

Boutsicaris

Sumner

, et al. Use of continuous glucose monitor in critically ill COVID-19 patients requiring insulin infusion: an observational study. J Clin Endocrinol Metab 2021; 106(10): e4007–e4016.

24.

Longo

Elias

Khan

, et al. 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.

25.

Nair

Dellinger

Flum

, et al. A pilot study of the feasibility and accuracy of inpatient continuous glucose monitoring. Diabetes Care 2020; 43(11): e168–e169.

26.

Tripyla

Herzig

Joachim

, et al. Performance of a factory-calibrated, real-time continuous glucose monitoring system during elective abdominal surgery. Diabetes Obes Metab 2020; 22(9): 1678–1682.

27.

Perez-Guzman

Duggan

Gibanica

, et al. Continuous glucose monitoring in the operating room and cardiac intensive care unit. Diabetes Care 2021; 44(3): e50–e52.

28.

Krutkyte

Roos

Schuerch

, et al. Fully closed-loop insulin delivery in patients undergoing pancreatic surgery. Diabetes Technol Ther 2023; 25(3): 206–211.

29.

Herzig

Suhner

Roos

, et al. Perioperative fully closed-loop insulin delivery in patients undergoing elective surgery: an open-label, randomized controlled trial. Diabetes Care 2022; 45(9): 2076–2083.

30.

Voglová Hagerf

Protuš

Németová

, et al. Alternative site of RT-CGM sensor application for abdominal surgery in the infraclavicular region. J Diabetes Sci Technol 2023; 17(6): 1728–1730.

31.

Voglová Hagerf

Protus

Nemetova

, et al. Accuracy and feasibility of real-time continuous glucose monitoring in critically ill patients after abdominal surgery and solid organ transplantation. Diabetes Care 2024; 47(6): 956–963.

32.

Klonoff

Freckmann

Pleus

, et al. The diabetes technology society error grid and trend accuracy matrix for glucose monitors. J Diabetes Sci Technol 2024; 18(6): 1346–1361.

33.

Freckmann

Eichenlaub

Waldenmaier

, et al. Clinical performance evaluation of continuous glucose monitoring systems: a scoping review and recommendations for reporting. J Diabetes Sci Technol 2023; 17(6): 1506–1526.

34.

Diabetes Technology Society. DTS Error Grid Analysis Tool, www.diabetestechnology.org/dtseg/ (accessed 22 December 2025).

35.

Welsh

Psavko

Zhang

, et al. Comparisons of Fifth-, Sixth-, and Seventh-Generation continuous glucose monitoring systems. J Diabetes Sci Technol 2024; 18(1): 143–147.

36.

Wright

Williams

Friedman

, et al. Accuracy of continuous glucose monitors for inpatient diabetes management. J Diabetes Sci Technol 2023; 17(5): 1252–1255.

37.

Arias-Rivera

Raurell-Torredà

Fernández-Castillo

R-J

, et al. Blood glucose monitoring in critically ill adult patients: type of sample and method of analysis. Systematic review and meta-analysis. Enferm Intensiva 2024; 35(1): 45–72.

38.

Heinemann

Interferences with CGM systems: practical relevance?

J Diabetes Sci Technol 2022; 16(2): 271–274.

39.

International Organization for Standardization. ISO15197:2013 in vitro diagnostic test systems: requirements for in vitro blood glucose monitoring systems for self-testing in managing diabetes mellitus, https://www.iso.org/standard/54976.html (2013, accessed 22 December 2025).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB

0.05 MB