Accuracy and feasibility of initiating real-time continuous glucose monitoring in noncritically ill hospitalized patients with type 2 diabetes

Abstract

BACKGROUND

Continued use of personal continuous glucose monitoring (CGM) devices is recommended for people with diabetes receiving insulin therapy during hospitalization to achieve better glycemic control and reduce hypoglycemia. However, data on the initiation of real-time CGM (rt-CGM) in hospitalized patients remain limited.

AIM

To evaluate accuracy and feasibility of initiating Glunovo i3 rt-CGM for general-medicine inpatients with type 2 diabetes mellitus (T2DM).

METHODS

A total of 25 patients with T2DM were enrolled. The mean age was 72.8 ± 11.6 years, 64% were female, and the mean glycated hemoglobin was 10.0% ± 3.1%. The Glunovo i3 rt-CGM sensor and transmitter were placed on the patient’s abdomen. Medical personnel calibrated the device using capillary blood glucose (CBG) measurements. Paired CBG and sensor glucose values were analyzed to assess the accuracy of CGM using the mean absolute relative difference (MARD) and Clarke Error Grid Analysis (CEGA).

RESULTS

CGM was used for 7.7 ± 2.4 days, with calibration performed 2.3 ± 0.7 times/day. Among 397 paired CBG and CGM measurements, overall MARD was 12.2% ± 12.2%, with the lowest MARD observed at CBG levels > 180 mg/dL (11.0% ± 11.1%). The percentages of readings within ± 15%/15 mg/dL, ± 20%/20 mg/dL, and ± 30%/30 mg/dL were 72.8%, 82.9%, and 91.7%, respectively. Overall, 96.7% of points fell within CEGA zones A and B, with none in zone E. Time in range improved from 41.2% ± 26.2% on day 1 to 48.3% ± 22.3% during the last 72 hours (P = 0.469).

CONCLUSION

Initiation of the Glunovo i3 rt-CGM system in noncritically ill hospitalized patients with insulin-treated T2DM is feasible and demonstrates acceptable accuracy and safety for glucose monitoring during hospitalization.

Key Words: Continuous glucose monitoring; Inpatient glucose management; Accuracy; Type 2 diabetes mellitus; Insulin therapy

Core Tip: Evidence regarding the initiation of real-time continuous glucose monitoring (rt-CGM) in hospitalized adults remains scarce. This study evaluated the accuracy and feasibility of initiating the Glunovo i3 rt-CGM system for inpatient glycemic monitoring among 25 patients with type 2 diabetes mellitus (T2DM) admitted to non-intensive care unit medical wards. Among 397 paired capillary blood glucose and sensor glucose measurements, the mean absolute relative difference was 12.2%, with 96.7% of values falling within zones A and B of the Clarke Error Grid Analysis. These findings suggest that initiating rt-CGM in hospitalized patients with T2DM is feasible and demonstrates acceptable accuracy for glucose monitoring.

INTRODUCTION

Poor glycemic control, including hyperglycemia, hypoglycemia, and glycemic variability, can lead to adverse outcomes in hospitalized patients with diabetes mellitus[1]. Optimal and dedicated inpatient diabetes care is mandatory for achieving glycemic goals. Point-of-care (POC) capillary blood glucose (CBG) measurement is recommended as the standard method for glucose monitoring during hospitalization. The frequency of POC glucose testing varies based on the patient’s clinical status and severity of illness and is typically performed 3-4 times a day in noncritically ill patients[2,3]. Nevertheless, glucose testing at 4-hour to 6-hour intervals may miss asymptomatic hypoglycemia and fail to capture glycemic variability.

Continuous glucose monitoring (CGM) is an emerging technology that addresses some of the limitations of POC glucose monitoring by improving the detection of hypoglycemia and glucose fluctuations[4,5]. The accuracy of CGM has been extensively studied and validated in outpatient settings. Continued use of personal CGM devices is recommended in patients with diabetes receiving insulin therapy during hospitalization, in conjunction with bedside POC glucose monitoring[3,6]. However, evidence regarding the initiation of real-time CGM (rt-CGM) in hospitalized adult patients remains scarce owing to insufficient data on its accuracy and feasibility in the hospital setting. Studies evaluating CGM accuracy in hospitalized patients, commonly assessed using the mean absolute relative difference (MARD), have demonstrated variable performance across different CGM systems and patient populations. Inpatient use of the Glunovo CGM system (Infinovo Medical Technology Co., Ltd., Suzhou, Jiangsu Province, China) has been evaluated in only one previous study; however, the characteristics and severity of the patients included were not reported[7].

This study aimed to evaluate the accuracy and feasibility of the Glunovo i3 rt-CGM system for inpatient glycemic monitoring in patients with type 2 diabetes mellitus (T2DM) admitted to non-intensive care unit (non-ICU) medical wards.

MATERIALS AND METHODS

Study design and study population

This was a single-center prospective study of hospitalized noncritically ill adult patients with T2DM. The inclusion criteria were age 18 years, a diagnosis of T2DM, admission to a non-ICU medical ward, and a requirement for insulin therapy. In routine clinical practice, insulin therapy is generally initiated for persistent hyperglycemia (blood glucose > 180 mg/dL) during hospitalization. However, the final decision to initiate insulin therapy was made by the attending physician based on the patient’s blood glucose level and overall clinical condition; no fixed glucose threshold was mandated by the study protocol.

Exclusion criteria included coagulopathy (international normalized ratio > 3), thrombocytopenia (platelet count < 20000/µL), hypotension (blood pressure < 90/60 mmHg) or poor tissue perfusion, multiorgan failure, significant edema, current treatment with dual antiplatelet therapy, and use of medications known to interfere with the accuracy of CGMs, such as hydroxyurea, intravenous ascorbic acid, or acetaminophen at doses > 4 g in 24 hours.

Study protocol

Eligible participants were enrolled from non-ICU medical wards at Siriraj Hospital, Bangkok, Thailand, between January 2022 and March 2022. All patients or their legally authorized representatives provided written informed consent before enrollment. The study protocol was approved by the Siriraj Institutional Review Board (COA No. Si 644/2021). Demographic characteristics, comorbidities, previous diabetes treatment (if any), and severity of illness, assessed using the National Early Warning Score 2 (NEWS2)[8], were collected after recruitment. The NEWS2 ranges from 0 to 20, with higher scores indicating greater clinical risk.

The Glunovo i3 rt-CGM system was used in this study. Before device insertion, medical personnel involved in patient care were trained to perform CGM calibration using the Glunovo smartphone application. A 14-day Glunovo i3 rt-CGM sensor and transmitter were then placed on the participant’s abdomen. The device was calibrated using POC CBG measurements three times on the first day and at least twice daily thereafter, in accordance with the manufacturer’s recommendations. POC CBG testing was performed using the Thai Food and Drug Administration (FDA)-approved StatStrip Xpress Glucose Hospital Meter System (Nova Biomedical, Cheshire, United Kingdom). The CGM system automatically received sensor glucose (SG) values every 3 minutes and displayed them in the application.

CBG measurements were performed according to the routine inpatient monitoring schedule: Before meals and at bedtime in patients receiving oral intake, or every 6 hours in patients who were nil per os.

During the study period, diabetes treatment regimens were determined solely by the attending physicians, and the investigators did not participate in or influence treatment decisions. Attending physicians had access to rt-CGM data. If the CGM system generated an alert for hypoglycemia (< 70 mg/dL) or hyperglycemia (> 250 mg/dL), medical staff were instructed to promptly measure CBG and make treatment decisions based on the CBG results. The sensor and transmitter were removed after 14 days of use or earlier if the patient was discharged before completing the 14-day period. All CBG and SG values, diabetes medications, and adverse events (AEs) were recorded throughout the study period.

Statistical analysis

The sample size was calculated using the formula for estimating a population mean. Based on a previous study by Sadhu et al[9], in which the overall MARD was 13.1% with a standard deviation of 10%, a sample size of 25 participants was required to achieve a precision of 4% at a 95% confidence level.

Categorical variables were presented as frequencies and percentages. Continuous variables were presented as the mean and standard deviation when normally distributed and as the median and interquartile range (IQR) when non-normally distributed.

Paired CBG and SG measurements were analyzed to evaluate the accuracy of the CGM system. The first paired CBG-SG measurement obtained immediately after CGM insertion was excluded from the accuracy analysis because, according to the manufacturer’s specifications, the initial sensor value is calibrated using the corresponding CBG value. The primary study endpoints were the MARD, median absolute relative difference (ARD), Clarke Error Grid Analysis (CEGA), and Bland-Altman plot[10]. Secondary endpoints included time in range (TIR), time below range (TBR), time above range (TAR), AEs, and the calibration success rate.

TIR, TBR, and TAR derived from CGM data were compared between day 1 and the final 72 hours of CGM use using the paired t-test. SG values closest in time to the corresponding CBG measurements, obtained either before or after the CBG measurements, were used for analysis. Paired CBG and SG measurements associated with incorrect calibration (i.e., entry of an incorrect glucose value) or failed calibration (defined as a delay of > 10-minute delay calibration or repeated calibration using the same CBG value more than twice) were excluded from the analysis.

MARD was calculated as the average relative difference between paired SG and CBG values and expressed as a percentage[9]. We also calculated the proportion of SG values within 15 mg/dL of CBG values < 70 mg/dL or 15% of CBG values ≥ 70 mg/dL (± 15/15%). The same criteria were applied for ± 20/20%, ± 30/30%, and ± 40/40% criteria were also calculated. A two-sided P value < 0.05 was considered statistically significant. Statistical analyses were performed using IBM SPSS Statistics version 20 (IBM Corp., Armonk, NY, United States).

RESULTS

Participant characteristics

A total of 26 participants were enrolled, of whom 1 was excluded owing to a sensor error on the second day after sensor placement (Figure 1). The baseline characteristics of the remaining 25 patients are shown in Table 1.

Figure 1 Flow of participants through the study. CGM: Continuous glucose monitoring.

Table 1 Baseline characteristics of the study participants, mean ± SD or n (%) or median (interquartile range).

Baseline characteristics	n = 25
Age (years)	72.8 ± 11.6
Female sex	16 (64)
BMI (kg/m2)	23.6 ± 3.6
Comorbidities
Hypertension	20 (80)
Dyslipidemia	16 (64)
Cerebrovascular disease	5 (20)
Coronary artery disease	4 (16)
Diabetes medication
No medication	5 (20)
OHA	10 (40)
Insulin	7 (28)
Insulin with OHA	3 (12)
HbA1c (%)	10 ± 3.1
eGFR (mL/minute/1.73 m2)	37 (20, 69.5)
Steroid use	3 (12)
Total daily insulin dose before enrollment (U/kg/day)	0.5 (0.4, 0.7)
Respiratory support
No respiratory support	12 (48)
Oxygen cannula	6 (24)
HFNC	1 (4)
Mechanical ventilation	6 (24)
Organ failure
No organ failure	23 (92)
Kidney failure	2 (8)
NEWS2
0-4	17 (68)
5-6	5 (20)
≥ 7	3 (12)
Length of stay (days)	27 ± 22.3
Admission diagnosis
Infection	12 (48)
Stroke	5 (20)
Heart failure or myocardial infarction	4 (16)
Cancer	2 (8)
Diabetic ketoacidosis	2 (8)

BMI: Body mass index; OHA: Oral hypoglycemic agent; HbA1c: Glycated hemoglobin; eGFR: Estimated glomerular filtration rate; HFNC: High-flow nasal cannula; NEWS2: National Early Warning Score 2.

The mean age of the patients was 72.8 ± 11.6 years, and 64% were female. The mean body mass index was 23.6 ± 3.6 kg/m². All patients had pre-existing T2DM; 8% had experienced diabetic ketoacidosis 2 weeks before enrollment and had been transitioned to subcutaneous insulin before study entry. The most common admission diagnosis was infection. Hypertension and dyslipidemia were the most frequent comorbidities. The mean glycated hemoglobin (HbA1c) level was 10.0% ± 3.1%. The median total daily insulin dose before enrollment was 33 (IQR: 22-33) U/day or 0.5 (IQR: 0.4-0.7) U/kg/day.

Illness severity was assessed using the NEWS2[8]. Most patients (68%) had low scores (≤ 4), indicating mild disease, while 32% had higher scores, indicating a greater risk of clinical deterioration.

CGM analysis

The mean duration of CGM use was 7.7 ± 2.4 days (range: Up to 11 days), with no CGM data available beyond 11 days. The mean frequency of calibration was 2.3 ± 0.7 times per day. None of the participants experienced AEs related to CGM sensor insertion or wear.

MARD

The overall MARD across all paired SG and CBG measurements was 13.5% ± 15.7%. After excluding 117 pairs because of incorrectly entered CBG calibration values, calibration performed > 10 minutes after CBG measurement, or repeated calibration using the same CBG value more than twice, 397 pairs remained for analysis. In this set, the MARD was 12.2% ± 12.2%, and the median ARD was 9.1% (IQR: 3.7%-15.5%). MARD varied across glucose ranges: It was lowest at CBG levels > 180 mg/dL (11.0% ± 11.1%) and the highest in the hypoglycemic range, as shown in Table 2. MARD values were also assessed according to the day of sensor wear, as shown in Table 3, with the lowest value observed on day 9 (6.8% ± 7.1%). Because few measurements were available on later days, days 9-11 were combined; the MARD for these days was 11.3% ± 18.3%.

Table 2 The absolute relative difference according to capillary blood glucose range.

Index	Capillary blood glucose range (mg/dL)			Total
	< 70	70-180	> 180
Paired values (n)	12	179	206	397
MARD (%), mean ± SD	28.4 ± 25.6	12.6 ± 11.4	11.0 ± 11.1	12.2 ± 12.2
ARD (%), median (Q1, Q3)	27.1 (10.2, 39.7)	9.3 (4.0, 17.2)	8.2 (3.4, 14.5)	9.1 (3.7, 15.5)
95%CI	12.2-44.7	10.9-14.2	9.5-12.5	11.0-13.4

ARD: Absolute relative difference; 95%CI: 95% confidence interval; MARD: Mean absolute relative difference; Q1: First quartile; Q3: Third quartile.

Table 3 Mean absolute relative difference according to day of sensor wear.

Day	Paired values (n)	MARD (%), mean ± SD	ARD (%), median (Q1, Q3)	95%CI
1	67	15.7 ± 12.7	13.7 (5.8, 22.7)	12.6-18.8
2	62	12.1 ± 8.4	11.0 (5.7, 16.5)	10.0-14.3
3	58	9.6 ± 9.1	7.1 (3.2, 13.1)	7.2-12.0
4	59	11.0 ± 11.4	8.3 (4.4, 14.2)	8.1-14.0
5	44	12.6 ± 15.2	6.7 (3.1, 17.2)	8.0-17.3
6	37	11.7 ± 12.4	7.6 (2.6, 14.3)	7.5-15.8
7	28	13.5 ± 12.7	11.7 (3.9, 17.5)	8.6-18.4
8	15	11.1 ± 11.1	8.1 (3.4, 11.4)	5.0-17.3
9-11	27	11.3 ± 18.3	5.9 (1.7, 14.7)	4.0-18.5
Total	397	12.2 ± 12.2	9.1 (3.7, 15.5)	11.0-13.4

ARD: Absolute relative difference; 95%CI: 95% confidence interval; MARD: Mean absolute relative difference; Q1: First quartile; Q3: Third quartile.

Agreement analysis

Table 4 shows the agreement between the paired SG and CBG measurements. The overall agreement rates according to the 10/10%, 15/15%, 20/20%, 30/30%, and 40/40% criteria were 54.2%, 72.8%, 82.9%, 91.7%, and 96.0%, respectively. When analyzed according to glucose range, 72.1% of SG measurements were within the 15/15% agreement criterion for CBG levels of 70-180 mg/dL, compared with 76.2% for CBG levels > 180 mg/dL.

Table 4 Agreement between paired sensor glucose and capillary blood glucose measurements according to capillary blood glucose range.

Agreement level	± 10/10% (mg/dL)	± 15/15% (mg/dL)	± 20/20% (mg/dL)	± 30/30% (mg/dL)	± 40/40% (mg/dL)
Total	54.2%	72.8%	82.9%	91.7%	96.0%
95%CI	49.1-59.1	68.1-77.1	78.8-86.4	88.5-94.2	93.5-97.7
< 70 mg/dL	25.0%	25.0%	41.7%	58.3%	75%
95%CI	5.5-57.2	5.5-57.2	15.2-72.3	27.7-84.8	42.8-94.5
70-180 mg/dL	52.5%	72.1%	81.6%	90.5%	96.1%
95%CI	44.9-60.0	64.9-78.5	75.1-87.0	85.2-94.4	92.1-98.4
> 180 mg/dL	57.3%	76.2%	86.4%	94.7%	97.1%
95%CI	50.2-64.1	69.8-81.9	81.0-90.8	90.6-97.3	93.8-98.9

95%CI: 95% confidence interval.

CEGA

The paired SG and CBG measurements were stratified into five zones, as shown in Figure 2 and Table 5. Zone A represents SG values that deviate from CBG values by not more than 20% or values in which both SG and CBG are in the same hypoglycemic range (< 70 mg/dL). Zone B represents SG values that deviate from CBG values by > 20% but would lead to benign or no treatment-related consequences[10].

Figure 2 Clarke Error Grid Analysis of paired sensor glucose and capillary blood glucose measurements. The paired sensor glucose and capillary blood glucose measurements were stratified into five zones. Overall, 96.7% of data points fell within zones A and B, which are considered clinically acceptable. Overcorrection of high glucose levels occurred in 0.8% of measurements (zone C). Failure to detect hypoglycemia and hyperglycemia occurred in 1.5% and 1.0% of measurements, respectively (zone D). No erroneous measurements were observed (zone E).

Table 5 The paired sensor glucose and capillary blood glucose measurements were stratified into five zones.

Zone	Percentage (%), n = 397
Zone A (clinical accuracy)	83.1
Zone B (no or benign treatment)	13.6
Zone C (overcorrection)	0.8
Zone D (dangerous failure to detect and treat)	2.5
Failure to detect hyperglycemia	1.0
Failure to detect hypoglycemia	1.5
Zone E (erroneous measurement)	0

A total of 96.7% of data points fell within zones A and B, which were considered clinically acceptable. Overcorrection of high glucose levels occurred in 0.8% of measurements (zone C). Failure to detect hypoglycemia and hyperglycemia occurred in 1.5% and 1.0% of measurements, respectively (zone D). No erroneous measurements (zone E) were observed in this study.

Bland-Altman plot

The Bland-Altman plot is shown in Figure 3. The mean difference between paired SG and CBG measurements was -0.5 mg/dL, with 95% limits of agreement from -66.3 mg/dL to 65.3 mg/dL. The differences appeared more widely dispersed at higher glucose levels.

Figure 3 Bland-Altman plot of paired sensor glucose and capillary blood glucose measurements. Each point represents one paired measurement. The horizontal axis shows the mean of sensor glucose (SG) and capillary blood glucose (CBG), and the vertical axis shows their difference (SG-CBG), both in mg/dL. The mean difference was -0.5 mg/dL, and the 95% limits of agreement were -66.3 mg/dL and 65.3 mg/dL. CBG: Capillary blood glucose; SG: Sensor glucose.

CGM-derived parameters of glycemic control

The overall mean TIR, TAR, and TBR were 45.1% ± 26.1%, 53.8% ± 26.6%, and 1.1% ± 2.3%, respectively. TIR, TBR, and TAR were analyzed for each day of sensor wear. The TIR was 41.2% ± 26.2% on day 1 and increased to 48.3% ± 22.3% during the last 72 hours of sensor wear (P = 0.469; 95% confidence interval: -11.46% to 23.71%). The proportions of level 1 and level 2 hypoglycemia were 3% and 1%, respectively. No skin reactions or serious device-related cutaneous AEs were observed during the study.

DISCUSSION

We conducted this study to evaluate the accuracy of the Glunovo i3 rt-CGM system in hospitalized patients with T2DM who are noncritically ill. Clinical conditions, admission diagnosis, illness severity, hemodynamic status, and intensive care unit status can affect CGM reliability[6]. The overall MARD in our study was 12.2% ± 12.2%, which is comparable to values reported in similar inpatient settings. Gómez et al[11] reported a MARD of 12.9% using the Medtronic iPro2 system (Medtronic Diabetes, Minneapolis, MN, United States), while Davis et al[12] reported a MARD of 12.8% using the Dexcom G6. A study from China using the Glunovo i3 rt-CGMthe same system used in our studyreported a slightly lower MARD of 8.89% in hospitalized patients with T2DM; however, admission diagnoses and illness severity were not reported[7]. Patients in that study were younger and had lower baseline HbA1c levels than those in our cohort. In contrast, our participants were predominantly older adults with poorer baseline glycemic control and more complex clinical conditions, which may have partly contributed to the higher MARD observed in our study.

Our MARD result had better accuracy than those reported by Wright et al[13], who evaluated flash glucose monitoring using the FreeStyle Libre 1 and Libre 2 systems (Abbott Laboratories, Abbott Park, IL, United States) in the inpatient setting. Their reported MARD values were 21.4% ± 11.8% for the FreeStyle Libre 1 model and 17.7% ± 10.5% for the Libre 2 model. However, their study included patients with type 1 diabetes, T2DM, and diabetes of unknown type, whereas all participants in our study had T2DM. Furthermore, their study had a relatively shorter duration of CGM use, with a median duration of 2 days compared with 7.7 days in our study. MARD also varied by day of sensor wear. Although the lowest value in the original day-by-day analysis was observed on day 9, only 10 paired measurements from five patients were available that day, with even fewer patients contributing data on days 10 and 11. The late-day estimates should be interpreted cautiously because they are based on small, changing groups of patients and may not reliably indicate a change in sensor accuracy over time.

The MARD in the hypoglycemic range was relatively high (28.4% ± 25.6%) in our study. This finding may be attributable to the limited number of hypoglycemic observations, resulting in imprecise estimates, and/or reduced sensor accuracy at low glucose concentrations. Similar findings have been reported with other CGM systems, with reported MARD values in the hypoglycemic range ranging from 17.7% to 28.0%[12-14]. Therefore, when CGM readings indicate hypoglycemia, SG values should be confirmed using capillary or laboratory plasma glucose measurements before making clinical decisions. Furthermore, the limited number of paired SG and CBG measurements on days 9-11 may have resulted in MARD estimates that do not fully reflect the device's actual performance under these conditions.

The agreement between paired SG and CBG values in our study showed that the agreement rate (15/15%) was 72.1% for CBG levels of 70-180 mg/dL. This result fell within the acceptable range defined by the United States' FDA accuracy criteria for integrated CGM systems[15,16]. However, these accuracy criteria were developed for personal ambulatory CGM use, and there is currently no international consensus regarding accuracy standards for inpatient CGM use.

CEGA was also used to assess the clinical accuracy of CGM. In our study, 96.7% of data points fell within zones A and B of the CEGA, which are considered clinically acceptable zones. This finding was consistent with previous studies, in which the combined proportion of data points in zones A and B ranged from 91.9% to 100%[11-14,17].

Although TIR increased numerically from day 1 to the final 72 hours of sensor wear, this change was not statistically significant. As the sample size was calculated based on the primary endpoint (MARD), the study may have been underpowered to detect changes in secondary glycemic outcomes. In addition, the substantial interindividual variability in TIR, reflected by the large standard deviations and the wide confidence interval, may have further limited the ability to detect a statistically significant difference. Moreover, as attending physicians had access to the rt-CGM data, treatment decisions may have been influenced by glucose trends or alerts generated by the device. However, the study was not designed to determine whether the observed change in TIR resulted from CGM-informed treatment adjustments, routine insulin titration, improvement in the underlying illness, changes in nutritional intake, or other aspects of inpatient care. Therefore, the observed increase in TIR should not be interpreted as evidence of a direct therapeutic effect of rt-CGM.

Our study demonstrates that initiating rt-CGM in non-ICU hospitalized patients is feasible and can support glycemic control during admission when combined with POC glucose confirmation. However, several challenges were identified during implementation in routine clinical practice. Although healthcare professionals received training on CBG calibration, a high frequency of calibration errors and delayed calibrations was observed. This likely reflects limited familiarity with CGM use and the high workload in hospital wards. These challenges may be mitigated through additional staff training, increased familiarity with routine CGM use, patient-performed calibration when appropriate, and the use of factory-calibrated CGM systems where feasible.

In addition, the smartphone paired with the CGM sensor remained with the patient, and the CGM system was not integrated with the nursing station. Consequently, integrating CGM alerts for hypoglycemia and hyperglycemia into a centralized monitoring or telemetry system linked to the nursing station may facilitate timely recognition and clinical intervention.

This study has several strengths. First, although several studies have evaluated the accuracy of CGM in hospitalized patients and demonstrated moderate reliability, with MARD values typically ranging from 12% to 21%, most have focused primarily on accuracy assessment rather than clinical implementation. In addition, many studies were conducted in controlled settings or used blinded CGM, limiting their applicability to real-world inpatient care. Evidence supporting the initiation of CGM during hospitalization to guide glycemic management, particularly in noncritically ill patients, remains limited. Our study addresses this gap by demonstrating that rt-CGM can be feasibly initiated in hospitalized patients and used to support glycemic monitoring during hospitalization when combined with POC glucose confirmation.

Moreover, our study included a relatively longer duration of CGM use and a substantial number of paired SG and CBG measurements compared with previous studies. Gómez et al[11], Reutrakul et al[17], and Wright et al[13] reported mean or median CGM wear durations of 4.3 days, 4.3 days, and 2 days, respectively. In comparison, the mean CGM duration in our study was 7.7 days.

This study has several limitations. First, the sample size was relatively small and was initially calculated based on data from patients with coronavirus disease 2019 (COVID-19) infection[9] as evidence on CGM use in hospitalized patients without COVID-19 infection was very limited when the study was designed. Nevertheless, recalculation using data from a comparable clinical setting yielded the same required sample size[13]. Second, the single-arm design without a comparator group limited our ability to directly compare the performance of the study device with that of other CGM systems. Third, we excluded paired measurements associated with incorrect calibration from the MARD analysis. These exclusions were based solely on deviations from the prescribed calibration procedure rather than on the magnitude or direction of the differences between SG and CBG values and were therefore not intended to selectively exclude less accurate sensor measurements. Nevertheless, excluding these pairs may have introduced some selection bias toward better apparent accuracy. The use of newer-generation factory-calibrated CGM systems may minimize calibration-related errors and improve the feasibility of CGM use in hospitalized patients.

Fourth, the study was registered retrospectively because it was initiated urgently during the COVID-19 pandemic to facilitate timely data collection. This may limit transparency regarding prespecified study procedures and outcomes. However, the study protocol and predefined study procedures were established before data collection, and the statistical analysis was performed according to the predefined study objectives without post hoc modification. Finally, we did not assess healthcare provider or patient satisfaction, user acceptance, or the impact of CGM implementation on clinical workflow. Future studies should evaluate these implementation outcomes, as they may influence the successful adoption of CGM in hospitalized patients.

CONCLUSION

In our preliminary data, initiation of the Glunovo i3 rt-CGM system in noncritically ill hospitalized patients with insulin-treated T2DM is feasible and demonstrates acceptable overall accuracy; however, its performance in the hypoglycemic range warrants caution and confirmation with POC glucose measurements before clinical decision-making.

ACKNOWLEDGMENTS

The authors gratefully acknowledge Miss Khemjira Karaketklang, Research and Academic Service Unit, Department of Medicine, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand, for her valuable assistance with statistical analysis.