Continuous glucose monitoring during intravenous insulin infusion treatment: Assessing accuracy to enable future clinical utility

Abstract

Aims

Continuous glucose monitoring (CGM) during intravenous insulin infusions (IVII) could reduce blood glucose (BG) testing burden in hospital, however CGM accuracy concerns exist. We aimed to assess CGM accuracy during IVII.

Methods

This multi‐centre observational study included adults with type 1 diabetes (T1D) who required IVII treatment during hospital admission whilst wearing their own CGM devices (Abbott FreeStyle Libre 2, Medtronic Guardian 3, Dexcom G6). IVII dose adjustments were performed based upon standard of care BG measures. Accuracy was assessed according to mean absolute relative difference (MARD) and Consensus error grid (CEG) analysis, using time‐matched (±5 minutes) pairs of CGM glucose and reference BG (point‐of‐care [POC], blood gas [GAS]) obtained during IVII.

Results

In total, 736 time‐matched glucose pairs were obtained from 56 hospital admissions (52% with diabetic ketoacidosis; 32% requiring intensive care). Median IVII duration was 16 hours (IQR 7.2–28). Overall MARD was 12.5% (11.9% for CGM‐POC pairs; 14.1% for CGM‐GAS pairs). In CEG analysis, 99.0% of glucose pairs were within zones A/B. Based on local hospital IVII dose titration protocols for non‐intensive care wards, if CGM measures had been used instead of POC, dose adjustments would have been the same in 77% of instances.

Conclusions

This real‐world study of adults with T1D demonstrated high concordance of CGM measures with BG during IVII. The accuracy of CGM during IVII might enable its greater clinical utility when treating inpatients receiving IVII. More inpatient studies are required to validate the use of CGM during IVII.

What's new?

What is already known?

• Use of continuous glucose monitoring (CGM) devices to guide intravenous insulin infusion (IVII) treatment in hospital could reduce reliance upon finger prick capillary blood glucose testing.

• However, regulatory approval for inpatient use of most CGM devices remains limited due to accuracy concerns.

What has this study found?

• In our multi‐centre real‐world study of adults with T1D using CGM admitted to hospital and treated with IVII, CGM glucose had high concordance with matched reference blood glucose.

What are the implications of the study?

• Future use of CGM to guide IVII treatment may improve patient experience in hospital by decreasing the burden of intensive finger prick capillary blood glucose testing.

1. INTRODUCTION

There is a high prevalence of diabetes mellitus amongst hospitalised adults. ¹ , ² , ³ Optimising blood glucose (BG) management in hospital improves clinical outcomes. ⁴ , ⁵ , ⁶ Subcutaneous insulin regimens are appropriate for the inpatient management of diabetes or hyperglycaemia in most scenarios. ⁷ , ⁸ However, intravenous insulin infusions (IVII) are often required to optimise inpatient glycaemia, especially in individuals with insulin‐requiring diabetes who develop unstable blood glucose levels or elevated blood ketone levels, or in the context of prolonged fasting in hospital. ⁹ , ¹⁰ Treatment with IVII requires intensive, often hourly, BG monitoring to safely titrate intravenous insulin doses, via a combination of capillary point‐of‐care (POC) and blood gas (GAS) BG testing. This high frequency of glucose testing to achieve target glycaemia remains burdensome for people with diabetes and hospital health professionals.

People with type 1 diabetes (T1D) are at increased risk of developing diabetic ketoacidosis (DKA) requiring hospitalisation ¹¹ or dysglycaemia during hospitalisation, and they often require IVII. In Australia, approximately 80% of people with T1D now use CGM in the community, following the introduction of a national subsidy in 2017. ¹² , ¹³ Many people with T1D who use CGM to guide insulin treatment at home continue to wear their CGM devices whilst hospitalised. ¹⁴ However, it is uncertain whether CGM accuracy is sufficient to dose insulin during IVII in hospital. Continuous glucose monitoring (CGM) technology enables minimally invasive BG estimates, via interstitial fluid glucose (ISFG) measures. Interest is rapidly developing in how the benefits of outpatient CGM use can be translated into safe and effective inpatient use. Regulatory approval for inpatient use of most CGM devices remains limited due to accuracy concerns. ¹⁴ However, people with diabetes prefer to be supported by their hospitals to continue CGM use during hospitalisation. ¹⁴ At present, clinical guidelines do not recommend the use of CGM ISFG measures to guide IVII dose adjustments. However, the potential use of a person's own CGM device during IVII treatment could reduce reliance upon intensive finger prick capillary BG testing, and thereby improve their in‐hospital experience.

The aim of this study was to assess CGM accuracy during real‐world IVII use in adults with T1D admitted to hospital, by comparing CGM ISFG with reference (REF) BG measures (POC and GAS).

2. METHODS

This is a subgroup analysis of the CONFIDE‐1 (CONtinuous glucose monitoring For Inpatients with DiabEtes–type 1) study, a multi‐centre retrospective observational whole‐of‐hospital study of all adults with T1D who wore their own CGM device during multi‐day acute hospital admissions between July 2020 and December 2023. ¹⁵ The CONFIDE‐1 study was conducted across three health services in Melbourne, Australia (Eastern Health [EH], Melbourne Health [MH], Peninsula Health [PH]), and included medical and surgical units, critical care and non‐critical care wards.

This subgroup analysis assessed the accuracy of CGM ISFG measures during IVII treatment. Although study participants wore their own CGM devices during their hospital admission, IVII dose adjustments were performed according to POC or GAS BG measures rather than CGM ISFG, per standard practice at each hospital.

2.1. Participants

Participants were included in this subgroup analysis if, during periods of IVII treatment, CGM ISFG measures were available from hospital‐linked web‐based CGM software accounts (LibreView, Medtronic Carelink, or Dexcom Clarity), and REF BG (POC, GAS) measures were available from hospital electronic medical records.

Participants admitted under paediatric/obstetric/palliative care/psychiatry/subacute units were excluded.

2.2. Materials

CGM devices included in this analysis were FreeStyle Libre 2 (reportable range 2.2–27.8 mmol/L; Abbott Diabetes Care Inc, VIC Australia), Medtronic Guardian 3 (reportable range 2.2–22.2 mmol/L; Medtronic, MN, USA) and Dexcom G6 (reportable range 2.2–22.2 mmol/L; DexCom Inc, CA, USA).

Capillary and arterial blood POC measures were obtained using Accu‐Chek Guide (Roche Diabetes Care, Rotkreuz, Switzerland), Nova StatStrip (Nova Biomedical, MA, USA) or FreeStyle Optium Neo H (Abbott Diabetes Care, CA, USA) glucose meters, at EH, MH and PH, respectively; GAS measures were obtained using GEM5000 (Werfen, Barcelona, Spain) or ABL90/ABL800 FLEX (Radiometer, Copenhagen, Denmark).

2.3. Statistical analyses

Time‐matched glucose pairs were obtained by pairing closest CGM ISFG and REF BG measures, within ±5 min.

To determine the clinical applicability of CGM for IVII, we first assessed the number of CGM measures outside device reportable ranges. To assess the accuracy of CGM during IVII, we then evaluated CGM‐REF glucose pairs, after excluding glucose measures outside device reportable ranges. For each glucose pair, absolute relative difference (ARD) was calculated as the absolute difference between CGM ISFG and REF BG, divided by REF BG; mean ARD (MARD) and median ARD were then calculated. Consensus (Parkes) error grid (CEG) analysis and Bland–Altman plots were performed. ¹⁶ , ¹⁷ Subgroup analyses by REF BG source, glucose strata, IVII indication and ward acuity were conducted.

Exploratory analyses of CGM accuracy in differing inpatient scenarios (admissions with DKA or requiring intensive care) were performed using the Wilcoxon Rank‐Sum test, utilising MARD results aggregated by unique admission. Comparisons of non‐parametric variables were performed using the Wilcoxon Rank‐Sum test. Baseline and outcome data were summarised as mean (SD); median (IQR); or number (%), as appropriate. Statistical analyses were performed using R version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria); “ega” package was used for CEG analysis. ¹⁸

Ethical approval was obtained from each health service's Office for Research (EH: QA23‐105‐102749; MH: QA2023122; PH: SA/103943/PH‐2023). The relevant human research ethics committees' representatives deemed individual consent was not required as only aggregate de‐identified data, which participants had previously provided consent for access by their health services, was used for quality improvement purposes.

3. RESULTS

From July 2020 to December 2023, hospital‐linked inpatient CGM data from people with T1D was available in 338 admissions. Of these, time‐matched CGM‐REF glucose data during IVII treatment was available in 56 admissions (Table 1); 32% of these admissions required intensive care unit (ICU) support. These 56 admissions were derived from 50 unique participants; three participants had two admissions each, and one had four admissions. Median (IQR) IVII duration was 16 h (7–28).

TABLE 1.

Baseline participant characteristics.

Variable	Results
Admissions
N	56
Admitting unit
Medical	42 (75%)
Surgical	14 (25%)
Admission type
Emergency	50 (89%)
Elective	6 (11%)
Admission details
DKA	29 (52%)
ICU	18 (32%)
Length of stay (days)	2.3 (1.8–4.9)
IVII duration (hours)	16 (7.2–28)
Participants
N	50
CGM‐reference BG pairs	736 (100%)
POC	541 (74%)
GAS	195 (26%)
Age (years)	42 (28–58)
Sex
Male	28 (56%)
Female	22 (44%)
Charlson Comorbidity Index	2 (1–4)
HbA1c
NGSP (%)	9.1 (2.4)
IFCC (mmol/mol)	76 (26)
Preadmission Insulin
Basal‐bolus	34 (68%)
CSII	11 (22%)
Other	5 (10%)
CGM device model
Abbott Libre 2	35 (70%)
Dexcom G6	10 (20%)
Medtronic Guardian 3	5 (10%)

Note: Results are expressed as mean (SD), median (IQR) or number (%), as appropriate.

Abbreviations: BG, blood glucose; CGM, continuous glucose monitoring; CSII, continuous subcutaneous insulin infusion; DKA, diabetic ketoacidosis; GAS, blood gas; HbA1c, haemoglobin A1c; ICU, intensive care unit; IVII, intravenous insulin infusion; MDI, multiple daily injections; POC, point‐of‐care; REF, reference.

Amongst the 52% of admissions with DKA, presentation biochemistry included median (IQR) blood glucose 21.7 mmol/L (17.8–27.0), capillary beta‐hydroxybutyrate 5.8 mmol/L (4.5–6.4), serum bicarbonate 14 mmol/L (11–16) and pH 7.22 (7.11–7.25).

Approximately 1% of CGM glucose measures during IVII exceeded the device's reportable range (2.2–27.8 mmol/L for Libre 2, 2.2–22.2 mmol/L for Guardian 3 and G6). After excluding these, a total 736 time‐matched CGM‐REF glucose pairs within device reportable ranges (541 CGM‐POC pairs, 195 CGM‐GAS pairs) were available for accuracy analyses, with mean (SD) REF BG 9.9 mmol/L (4.2 mmol/L).

Overall, MARD was 12.5% (median ARD 8.2%). MARD (median ARD) was 11.9% (7.7%) for the 541 CGM‐POC pairs, and 14.1% (9.4%) for the 195 CGM‐GAS pairs, respectively (Table 2). When aggregated by admission (Table S1), mean MARD was not significantly different during IVII in DKA compared to non‐DKA admissions (p = 0.20), nor in admissions requiring ICU compared to admissions not requiring ICU (p = 0.11).

TABLE 2.

Mean and median absolute relative difference of CGM measures compared with reference (synchronous) blood glucose during IVII treatment.

	MARD (SD)	Median ARD (IQR)	BG pairs
Overall	12.5 (15.2)	8.2 (2.6–16.7)	736
Reference blood glucose source
POC	11.9 (15.2)	7.7 (2.0–16.3)	541
GAS	14.1 (14.9)	9.4 (4.8–18.6)	195
CGM model
Libre 2	9.7 (13.9)	6.1 (1.6–13.0)	439
Dexcom G6	16.3 (15.4)	12.5 (5.1–23.1)	229
Guardian 3	17.3 (18.2)	10.2 (5.4–23.2)	68
IVII indication
DKA	13.8 (15.7)	8.9 (3.6–17.5)	424
Non‐DKA	10.6 (14.1)	6.8 (1.4–15.7)	312
Ward acuity
ICU	14.1 (19.2)	9.4 (2.8–19.4)	184
Non‐ICU	11.9 (13.5)	8.0 (2.5–16.0)	552
BG measurement strata (mmol/L)
<3.9	38.7 (44.2)	30.6 (10.0–50.0)	13
3.9–10	12.7 (14.4)	8.7 (3.0–17.1)	423
>10	10.9 (12.6)	6.8 (2.3–14.8)	300
≤5.6	19.8 (25.2)	11.8 (2.5–25.2)	95
>5.6	11.4 (12.7)	7.9 (2.6–15.9)	641

Abbreviations: ARD, absolute relative difference; BG, blood glucose; CGM, continuous glucose monitor; DKA, diabetic ketoacidosis; EMR, electronic medical record; GAS, blood gas; ICU, intensive care unit; IQR, interquartile range; IVII, intravenous insulin infusion; MARD, mean absolute relative difference; POC, point‐of‐care; SD, standard deviation.

In CEG analysis (Table 3, Figure 1), 99.0% of glucose pairs were within the clinically acceptable zones A/B (Zone A: no effect on clinical action; Zone B: little/no effect on clinical outcome; Zone C: likely to affect clinical outcome; Zone D: could have significant medical risk; Zone E: could have dangerous consequences ¹⁶ ). In admissions with DKA and admissions in ICU, 98.8% and 97.8% of glucose pairs, respectively, were within CEG zones A/B (Table 3).

TABLE 3.

CGM accuracy during IVII according to CEG analysis.

	Zone A	Zone B	Zone C	Zone D	Zone E
Overall	83.6% (615)	15.5% (114)	0.7% (5)	0.3% (2)	0% (0)
Reference blood glucose source
POC	84.5% (457)	14.6% (79)	0.7% (4)	0.2% (1)	0% (0)
GAS	81.0% (158)	17.9% (35)	0.5% (1)	0.5% (1)	0% (0)
Inpatient clinical scenario
DKA	80.4% (341)	18.4% (78)	0.9% (4)	0.2% (1)	0% (0)
ICU	85.3% (157)	12.5% (23)	1.1% (2)	1.1% (2)	0% (0)

Note: Results are expressed as % (number of glucose pairs).

Abbreviations: CEG: consensus error grid; CGM: continuous glucose monitoring; DKA: diabetic ketoacidosis; GAS: blood gas; ICU: intensive care unit; IVII: intravenous insulin infusion; POC: point‐of‐care.

FIGURE 1.

CEG comparison of CGM interstital fluid glucose versus reference blood glucose in people with T1D treated with IVII (a: POC; b: GAS), according to insulin infusion indication (red: DKA; blue: Non‐DKA). CEG, Consensus error grid; CGM, Continuous glucose monitoring; DKA, Diabetic ketoacidosis; GAS, Blood gas; IVII, Intravenous insulin infusion; POC, Point‐of‐care; T1D, Type 1 diabetes.

In Bland–Altman plot analyses (Figure 2), CGM demonstrated a mean difference (and 95% limits of agreement) of +0.03 mmol/L (−3.30, +3.36) compared to POC and −0.12 mmol/L (−4.41, +4.27) compared to GAS.

FIGURE 2.

Bland–Altman plot of difference between CGM interstital fluid glucose and reference blood glucose measures during IVII (a: POC; b: GAS). 95% limits of agreement (red dotted lines) calculated as mean ± 1.96 SD. CGM, Continuous glucose monitoring; GAS, Blood gas; IVII, Intravenous insulin infusion; POC, Point‐of‐care; SD, Standard deviation.

Amongst CGM‐POC glucose pairs obtained on non‐ICU wards, utilising the CGM reading instead of corresponding POC reading to titrate IVII dosing according to each hospital's respective variable rate ward insulin infusion policies (Figure S1A,B) would have resulted in the same dose adjustment 77% (307/398) of the time.

The 12 CGM ISFG measures outside the device's reportable range were all elevated glucose levels that exceeded the device's upper reportable limit, but were detectable by time‐matched REF BG measures, and mostly occurred during the first 24 h of admissions with DKA.

4. DISCUSSION

In adults with T1D, CGM use during IVII demonstrated an overall MARD of 12.5% (including MARD 13.8% in DKA). Importantly, overall 99% of glucose pairs were in CEG zones A/B, suggesting clinical outcomes would have been similar had IVII dose adjustments been based upon CGM glucose instead of reference POC/GAS glucose. Future prospective studies of CGM‐guided IVII dosing could be feasible in the near future and further larger prospective studies are anticipated.

This study is one of the larger studies to date investigating CGM accuracy during IVII, and one of the few comparing CGM with both POC and GAS measures. Some of the first CGM studies in hospital were performed during the COVID‐19 pandemic and suggested potential utility of CGM during IVII. Those studies were predominantly conducted in participants with type 2 diabetes. ¹⁹ , ²⁰ , ²¹ , ²² These studies, comprising 5–24 IVII‐treated participants, reported MARD 10.4%–14.8% for Dexcom G6 compared to POC glucose. ¹⁹ , ²⁰ , ²¹ , ²² Importantly, our study assessed CGM accuracy in people with type 1 diabetes, who may be prone to greater glycaemic variability. One DKA‐specific study compared CGM (Freestyle Libre Professional) to pooled POC/serum/whole blood glucose measures and reported MARD 19.6%. ²³

In our study, few CGM glucose measures (1%) were outside the CGM device's reportable range during IVII treatment, providing reassurance regarding clinical applicability, including for DKA. However, it is worth noting that the ISFG upper reportable range for Libre 2 is 22.2 mmol/L in North America, but is 27.8 mmol/L for the rest of the world (including Australia and the United Kingdom). Thus, the clinical applicability during periods of severe hyperglycaemia may be slightly lower in North America.

The use of IVII facilitates improved glucose control in hospital and can reduce hospital complications and mortality. ²⁴ , ²⁵ Fixed‐rate IVII is recommended for the management of DKA and hyperosmolar hyperglycaemic state. ¹⁰ , ²⁶ Variable‐rate IVII is used for a variety of indications in people with diabetes or hyperglycaemia, including perioperative diabetes management, and in people who are not eating or drinking normally or who have severe illness. ⁹ , ²⁷ However, the frequent BG monitoring required to safely dose IVII remains problematic.

On the other hand, utilisation of CGM during IVII could afford a number of benefits, including improvements in glycaemia, nursing workflow, and patient experience. Improved glycaemia during IVII as a result of CGM use ¹⁹ , ²⁸ may result from increased glucose measurement frequency and glucose trend data availability. Use of CGM to guide IVII dosing decisions has been reported to reduce nursing workload, as well as healthcare costs related to staff time and reduced usage of glucose testing equipment. ²⁹ Additionally, nursing staff report satisfaction with and preference for CGM use over frequent finger prick testing. ¹⁴

CGM use during IVII treatment may also improve the in‐hospital experience for people with diabetes. The need for intensive BG monitoring during IVII, often via hourly finger prick POC testing, can be associated with discomfort, as well as disrupted sleep. In hybrid models of glucose monitoring, the use of CGM measures to dose IVII treatment, accompanied by occasional POC validation checks, can reduce POC testing frequency by up to 70%. ²⁰ , ²¹ , ²² In our study, with a median IVII duration of 16 h, using CGM during IVII could result in approximately 11 fewer finger prick capillary glucose tests per admission, thereby fostering better patient‐centred care.

The limitations of this study include its retrospective design and lack of gold standard reference plasma glucose measures. Additionally, reference blood glucose measures were used for insulin infusion dosing decisions rather than CGM measures. Finally, the majority of DKA presentations were of moderate severity; thus, results might not be applicable to severe DKA typically managed in ICU.

5. CONCLUSIONS

In a real‐world study of adults with T1D using CGM who were admitted to hospital and received IVII treatment, CGM ISFG had high concordance with BG measures, with most (99%) measures being in Zones A/B in CEG analysis. The accuracy of CGM with IVII suggests a majority of insulin dose adjustment decisions during IVII treatment could be safely based upon CGM measures. The accuracy of CGM in hospital might enable its greater clinical utility when treating inpatients receiving IVII. More inpatient studies are required to validate the use of CGM during IVII.

AUTHOR CONTRIBUTIONS

R.W., M.K., A.C. and S.F. were involved in the conception and design of the study. M.K., A.C., C.C., D.R. and S.F. supported methodological decisions. R.W. and B.K. performed the data acquisition. R.W. performed the statistical analysis and wrote the first draft of the manuscript. M.K. and S.F. provided critical revisions to the manuscript. All authors made a significant contribution to finalising the manuscript and approved the final version of the manuscript. S.F. provided overall study supervision and is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

FUNDING INFORMATION

R.W. is the recipient of an Australian Commonwealth Government RTP Scholarship, ACADI PhD Grant, Fred Knight Research Scholarship, Rowden White Scholarship and Gordon P Castles Scholarship.

CONFLICT OF INTEREST STATEMENT

S.F. contributes to the advisory panel for Viatris Inc. and Pfizer Inc.; has received honoraria for speaker fees from AstraZeneca, Boehringer‐Ingelheim, Lilly, Novo Nordisk; and received honoraria for advisory fees from Medtronic, Mylan, Pfizer and Sanofi. M.K. has received honoraria for speaker fees from AstraZeneca.

Supporting information

Figure S1.

Table S1.

Wang R, Kyi M, Krishnamoorthi B, et al. Continuous glucose monitoring during intravenous insulin infusion treatment: Assessing accuracy to enable future clinical utility. Diabet Med. 2025;42:e70076. doi: 10.1111/dme.70076