Continuous Glucose Monitoring–Guided Insulin Administration in Hospitalized Patients With Diabetes: A Randomized Clinical Trial

Diabetes is reported in 20–34% of hospitalized adult patients in general medicine and surgery units (1,2). Dysglycemia, defined as hyperglycemia, hypoglycemia, and increased glucose variability in hospitalized patients with diabetes, has been associated with adverse outcomes, such as prolonged length of stay and increased risk of infections and higher mortality rates among others (1,3). Therefore, reducing dysglycemia represents a major goal when managing hospitalized patients with diabetes.

Clinical guidelines have recommended the use of basal-bolus insulin regimen as the preferred approach to manage patients with type 2 diabetes hospitalized in general medicine and surgery units (4–6). The use of scheduled subcutaneous insulin analogs with basal (i.e., glargine, detemir, or degludec) once daily in combination with short (regular) or rapid-acting insulin (i.e., lispro, aspart, or glulisine) prior to meals is effective for the management of most patients with diabetes (7,8). However, hypoglycemia is a common adverse event of basal-bolus insulin therapy, with reported incidence rates ranging between 12 and 35% in randomized controlled studies in non–intensive care unit (ICU) settings (8,9). The development of hypoglycemia, like hyperglycemia, has been associated with higher rates of hospital complications and hospital mortality (4–6,10–12).

Bedside capillary point-of-care (POC) glucose monitoring is the standard of care to assess glycemic control in the hospital. Clinical guidelines recommend bedside POC testing before meals and at bedtime to assess glycemic control and to adjust insulin therapy in the hospital (4,5,13). Diabetes management based on POC testing has been demonstrated to be effective in randomized controlled studies (7,14), but POC glucose testing before meals and bedtime leaves extended intervals of time when glucose is not monitored, leading to missed information important for glycemic control. Continuous glucose monitoring (CGM) measures interstitial glucose every 5 min, thus providing a more complete glycemic profile during a 24 h period compared with standard POC glucose testing (15,16). Recent studies in hospitalized patients with type 2 diabetes treated with a basal-bolus regimen have reported high accuracy compared with POC testing as well as increased detection of both hypo- and hyperglycemic events with the use of CGM compared with POC glucose testing (15,17).

The availability of glucose telemetry using CGM devices to wirelessly transmit glucose data from the bedside to a centrally located monitor at the nurse’s station has been shown to improve glycemic control by reducing the frequency and time in hypoglycemia in insulin-treated patients with type 2 diabetes (18,19). These results suggest that CGM technology, by improving detection of hypoglycemia, may improve outcomes in hospitalized patients with diabetes; however, the efficacy and safety of CGM in adjusting inpatient insulin therapy have not been evaluated. Therefore, we conducted a randomized clinical trial comparing insulin adjustments by POC testing (standard of care) and by RT-CGM profile in patients treated with a basal-bolus insulin regimen.

This multicenter, noninferiority open-label randomized study was conducted in three hospitals in the U.S., including Grady Memorial Hospital and Emory University Hospital in Atlanta, GA, and University of Maryland Medical Center in Baltimore, MD. The study protocol was approved by the Emory University and University of Maryland Institutional Review Boards. Informed consent was obtained from all subjects during hospitalization prior to enrollment. This trial is registered with ClinicalTrials.gov as NCT03877068.

We screened subjects >18 years of age with type 1 or type 2 diabetes admitted to general medical and surgical services. We enrolled patients with glucose levels <400 mg/dL without laboratory evidence of diabetic ketoacidosis and with an anticipated length of hospitalization >72 h after enrollment. Key exclusion criteria included patients with acute illness who required or were expected to require ICU admission or had a planned MRI during hospitalization, clinically relevant hepatic disease (diagnosed liver cirrhosis and portal hypertension), corticosteroid therapy (equivalent to prednisone dose >5 mg/day), end-stage renal disease (dialysis), anasarca, pregnancy, or any mental condition rendering the patient unable to understand the nature, scope, and possible consequences of the study. Following the coronavirus disease 2019 (COVID-19) pandemic declaration, a modification was submitted to exclude individuals hospitalized with COVID-19 infection.

Subjects were treated with a basal-bolus insulin regimen to a target fasting and premeal glucose concentration between 140 and 180 mg/dL. All patients underwent POC testing before meals and bedtime. Subjects were randomly assigned (1:1 ratio) to one of two groups: a standard of care with participants wearing a blinded Dexcom G6 CGM with insulin dose adjusted based on capillary POC glucose monitoring (POC group) or to the real-time CGM (RT-CGM) group with insulin adjustment based on daily CGM profile. Participants were randomly allocated to one of the two groups based on a statistician-developed computer-generated randomization table that assigned patients (1:1) with block stratification according to the blood glucose value at randomization and whether it was >200 mg/dL or <200 mg/dL.

Participants in the intervention group were monitored by RT-CGM glucose telemetry system (GTS) (18,19). CGM devices and GTS components were placed by study team members. GTS has been described previously (18). In short, the system included three platforms: 1) Dexcom G6 (Dexcom, San Diego, CA) CGM device (sensor/transmitter); 2) a smartphone with internet connectivity placed in the patient’s room served as an intermittent (routing) device; and 3) a tablet (Apple iPad) located at the nursing station. The first step of data transmission was the transfer of glucose values from the CGM sensor-transmitter (both worn by the patient) using Bluetooth. Secondly, using Dexcom software applications, the information was then sent to a smartphone (Apple iPhone), which served as an intermediate-transmitting (routing) device, recording and transmitting the glucose values. The study phone, which was specifically purchased for study-related procedures, was locked in a safe box located in the patient’s room. Using commercial and password-protected internet wireless networks, glucose values were then transmitted wirelessly to the nursing station from the smartphone to a tablet (Apple iPad). We used the Dexcom G6 and Follow digital software applications in order to transmit the CGM glucose values (https://www.dexcom.com/apps).

Hypoglycemia alarms were set at 80 mg/dL at the tablet placed at the nursing station for prevention of hypoglycemia. The hypoglycemia CGM alarms were used to notify nurses of low glucose values. Following alarm activation, nursing staff were instructed to obtain a POC test to confirm CGM-detected hypoglycemia and provide 15 g of carbohydrates in response to the hypoglycemia alarm, with the goal to prevent clinically significant hypoglycemia. Nursing staff were asked to proceed with the above actions without the need of further communication with the study team members.

An additional CGM alarm was set if glucose levels were >250 mg/dL for at least 1 h. The alarm notified the research team, who determined if medication adjustments were necessary, which were then relayed to the primary team for implementation. Participants in the control group wore blinded CGM devices, which are CGM systems that have the alarms turned off and were only used to record CGM glucose values during the hospital stay. If glucose value by POC testing was found to be <80 mg/dL, 15 g of carbohydrates was given as a preventive measure to avoid clinically significant hypoglycemia.

Subjects in both groups had POC testing before meals, at bedtime, and when clinically indicated. Insulin was initiated and titrated to target a glucose range between 140 and 180 mg/dL (see study protocol in the Supplementary Appendix). Insulin dose recommendations, either for insulin initiation or titration, were based on the protocol and made by a board-certified endocrinologist. These recommendations were ultimately relayed to clinical nurses, who administrated insulin to the hospitalized patients. Correctional insulin was administrated by nursing staff following the protocol without the need of further communication with providers or study team.

Starting insulin dosage and daily insulin adjustment orders were similar in both groups and followed a basal-bolus insulin regimen previously reported (7,8). The insulin protocol is included in the Supplementary Appendix. For the control group, POC values were used for insulin titration, as per standard of care. In the intervention CGM group, the previous 24 h CGM glucose profile was used for insulin adjustment. Nurses and medical staff received general diabetes education as well as education on CGM technology provided by diabetes educators and the research team, which included how and when to remove or replace CGM sensors and transmitters and hypoglycemia prevention protocol. Participants were followed during the entire hospital stay unless deceased or transferred to the ICU.

Primary end points included differences in percentage of time in range (TIR) 70–180 mg/dL and hypoglycemia, defined as percentage of time below range (TBR) <70 mg/dL and <54 mg/dL. Other secondary outcomes included differences in percentage of time above range (TAR) >180 mg/dL and >250 mg/dL, total number of hypoglycemic episodes <70 mg/dL and <54 mg/dL (20), and number of hypoglycemic episodes per participant. We also evaluated if there was a difference in nocturnal hypoglycemia by evaluating difference in number of events <70 mg/dL and <54 mg/dL and in the percentage of TBR <70 mg/dL and <54 mg/dL occurring between 12:00 midnight and 6:00 a.m. Furthermore, we evaluated whether there was a difference in reoccurrence of hypoglycemia <70 mg/dL and <54 mg/dL, occurring during the day or nocturnal hours, and difference in mean daily glucose values.

We examined differences in glycemic variability measured by the mean amplitude of glycemic excursion (MAGE), coefficient of variation (CV), and SD. In addition, we compared differences in the length of hospital stay, total daily dose of insulin, and hospital-related complications. Hospital-related complications included any complications that occurred 48 h after randomization and included acute renal failure (increase in serum creatinine >0.5 mg/dL or need for hemodialysis), infections, cardiovascular, acute respiratory failure, venous thromboembolism, gastrointestinal bleed, surgical reintervention, and hospital mortality.

The main objectives of this study were to compare differences in the percentage of TIR 70–180 mg/dL and hypoglycemia, defined as percentage of TBR <70 mg/dL and <54 mg/dL between patients managed by RT-CGM (intervention group) and capillary POC testing (standard of care group). The data were summarized as mean ± SD for continuous variables and count (percentage) for discrete variables unless otherwise specified. We compared baseline, clinical characteristics, insulin therapy, and hospital clinical outcomes between treatment groups. We made the comparisons using the nonparametric Wilcoxon test for continuous variables and χ² tests (or Fisher exact tests) for discrete variables. Because some patients did not have any hypoglycemia during the study, to compare the time in hypoglycemia between the two study groups, we applied the zero-inflated β regression method and used the zero-inflated negative binomial regression method to compare differences in hypoglycemia. A P value of <0.05 was considered significant. We performed the statistical analyses using SAS 9.4.

We consented 185 eligible general medicine and surgery patients (Fig. 1). Of them, nine subjects in the control group and three subjects in the intervention group were excluded due to lack of CGM data (one subject), sensor failure (one subject), or administrative withdrawal (one subject refused to wear the CGM device after abdominal surgery). Overall, 173 subjects completed the study and had CGM data: 85 subjects randomly assigned to the standard of care (POC group) and 88 subjects to the intervention (CGM) group (Fig. 1 and Supplementary Appendix). Among them, six subjects in the standard of care and five subjects in the intervention group were further excluded because of a hospital stay <24 h after CGM placement. In the final analysis, we included 162 subjects. Demographic data for both groups are detailed in Table 1. There were no significant differences in age, sex, race, BMI, type and duration of diabetes, admission HbA_1c, blood glucose at the time of the randomization, diabetes outpatient regimen, or primary admitting diagnosis between groups (Table 1).

There were no significant differences in TIR 70–180 mg/dL among subjects managed in the CGM group compared with those managed by the POC standard of care (54.51% ± 27.72 vs. 48.64% ± 24.25; P = 0.14) (Table 2). There were no differences in mean daily glucose (183.2 ± 40 vs. 186.8 ± 39 mg/dL; P = 0.36) or total daily insulin dose (40.7 ± 29.5 vs. 36.1 ± 28.1 units/day; P = 0.33). There were nonstatistically significant differences in TAR >180 mg/dL (44.80% ± 27.89 vs. 49.21% ± 25.50; P = 0.26) and in TAR >250 mg/dL (16.24% ± 19.63 vs. 17.08% ± 17.59; P = 0.45) for participants in the CGM intervention and POC group, respectively.

We observed a nonsignificant reduction in TBR <70 mg/dL (0.69% ± 2.15 vs. 2.15% ± 5.91; P = 0.43) and TBR <54 mg/dL (0.32% ± 1.33 vs. 1.00% ± 3.74; P = 0.35) in the CGM group compared with POC group (Table 2). A similar trend was observed in the percent of patients with CGM values <70 mg/dL (36% vs. 39%; P = 0.68) or <54 mg/dL (14% vs. 24%; P = 0.12), as well as in the number of hypoglycemic events per patient <70 mg/dL (0.65 ± 1.26 vs. 1.15 ± 2.24 events/patient; P = 0.36) and <54 mg/dL (0.22 ± 0.59 vs. 0.56 ± 1.46 events/patient; P = 0.11) between the CGM and POC groups.

In terms of nocturnal hypoglycemia, we observed nonstatistically significant reductions in percentage of TBR <70 mg/dL (0.22% ± 0.84 vs. 0.76% ± 2.67; P = 0.90), percentage of TBR <54 mg/dL (0.13% ± 0.75 vs. 0.35% ± 1.57; P = 0.35), or the number of hypoglycemic events per patient <70 mg/dL (0.20 ± 0.49 vs. 0.34 ± 0.83 events/patient; P = 0.71) or <54 mg/dL (0.08 ± 0.36 vs. 0.23 ± 0.72 events/patient; P = 0.14).

Participants in the CGM group with one or more hypoglycemic events <70 mg/dL had less recurrent hypoglycemic events <70 mg/dL (1.80 ± 1.54 vs. 2.94 ± 2.76 events/patient; P = 0.04) and lower percentage of TBR <70 mg/dL (1.89% ± 3.27 vs. 5.47% ± 8.49; P = 0.02) compared with the control POC group (Fig. 1A). The incidence rate ratio for inpatient hypoglycemia <70 mg/dL was estimated as 0.53 (95% CI 0.31–0.92). The group difference in percentage of TBR <70 mg/dL was also confirmed by the zero-inflated β regression (P < 0.001), which accounts for diabetes status for the zero-inflation component. Similarly, subjects in the CGM group who experienced nocturnal hypoglycemia <70 mg/dL had less nocturnal reoccurrence of hypoglycemic events <70 mg/dL (1.21 ± 0.43 vs. 1.93 ± 0.92 events/patient; P = 0.02) and lower percentage of TBR <70 mg/dL (1.30% ± 1.71 vs. 4.27% ± 5.15; P = 0.004) compared with the control POC group (Fig. 1B). Among those that experienced hypoglycemia <54 mg/dL, RT-CGM intervention led to less frequent hypoglycemic events, with an estimated incidence ratio for hypoglycemia <54 mg/dL of 0.37 (95% CI 0.17–0.83).

There were no differences in glycemic variability between the POC group and CGM group, as measured by the CV (27% ± 8 vs. 26% ± 9; P = 0.33), SD (50.4 ± 16.2 vs. 46.8 ± 18 mg/dL; P = 0.28), and mean amplitude of glycemic excursion (65.02 ± 39.10 vs. 61.24 ± 32.41; P = 0.73) (Table 2).

There was no statistically significant difference in the entire hospital length of stay between the CGM and POC groups (median 8.0 days [2.0, 67.0] vs. 8.0 days [3.0, 45.0]; P = 0.79). In addition, there were no differences in hospital-related complications, mortality, or any adverse events related to sensor insertion between groups, with three participants in the CGM group having minor bleeding and two participants in the standard of care having minor sensor bleeding and three sensor applicator malfunctions (P = 0.72). Among subjects with minor bleeding in the CGM group, one subject was treated with aspirin and enoxaparin, one subject received aspirin and heparin, and one was not on aspirin or anticoagulants. In the POC group, one subject was treated with aspirin and enoxaparin and one received enoxaparin.

In this randomized controlled trial, we explored the safety and efficacy of RT-CGM in guiding daily insulin adjustment in hospitalized patients with type 1 and type 2 diabetes treated with a basal-bolus insulin regimen. Our results indicate that RT-CGM resulted in a similar improvement in glycemic control and TIR compared with capillary POC standard of care. We observed a trend toward a reduction of hypoglycemic events and TBR, as well as in hyperglycemic events in the CGM group compared with POC testing. In addition, among patients with one or more hypoglycemic events, the use of CGM resulted in a significant reduction in the number of recurrent hypoglycemic events and time in hypoglycemia compared with the POC group. Importantly, the reduction in recurrent hypoglycemic events was seen overnight, a time that POC is rarely checked. These results indicate that the use of RT-CGM is safe in guiding insulin adjustment among hospitalized insulin-treated patients with diabetes.

Recent observational and prospective studies have explored the accuracy of intermittently scanned and RT-CGM in hospitalized patients with diabetes (21,22). These studies have shown acceptable accuracy and a greater ability to detect hypoglycemia with CGM, particularly asymptomatic and nocturnal hypoglycemia, when compared with capillary POC glucose testing (19,21,23). A recent prospective study with the use of intermittently scanned CGM reported a mean absolute relative difference of 14.8% and an error grid analysis of 98.8% of glucose pairs within zones A and B in general medicine and surgery patients (23). A more recent and larger prospective study in 205 insulin-treated patients with 4,067 matched CGM and POC glucose pairs, which used the same CGM system that we also used in our study, demonstrated acceptable accuracy, with a reported mean absolute relative difference of 12.8% and error grid analysis of 98.7% of values in zones A and B with the use of Dexcom G6 RT-CGM (17). To overcome the accuracy limitation of POC capillary glucose testing, an ongoing study is testing the effectiveness of the Dexcom RT-CGM system compared to arterialized venous blood sample measurements using the Yellow Springs Instrument (NCT02880267).

The use of inpatient CGM to guide insulin treatment has historically raised concerns about the limited knowledge on CGM interpretation by hospital clinicians as well as the potential for overinterpretation of CGM data, which may lead to insulin stacking and resultant hypoglycemia (24). However, in the current study, following a standard protocol to adjust insulin, we demonstrate that RT-CGM can safely be used without an increased risk of iatrogenic hypoglycemia. In agreement with our results, an interim analysis of a randomized trial reported that RT-CGM with GTS can reduce the number and time in hypoglycemia in high-risk hospitalized patients with diabetes (19) as well as reduce significant hyperglycemia (25). In addition, our study supports recent diabetes guidelines and consensus by a panel of experts on hospital diabetes care, which reported that CGM may allow better and safer management of inpatients with hyperglycemia by effectively identify trends toward hypoglycemia and hyperglycemia (24).

During the COVID-19 pandemic, given the potential advantages of CGM devices, the U.S. Food and Drug Administration issued a nonobjection statement for inpatient CGM use. As a result, CGM devices have been increasingly used for remote glucose management to reduce risk of exposure for hospital staff and patients and to reduce personal protective equipment utilization. By enabling remote glucose monitoring, CGM devices have been found to be safe and represent a valid alternative to POC glucose testing (26,27). However, CGM use in the hospital setting is still considered investigational and is not Food and Drug Administration–approved. This is in part due to the concern of the accuracy of CGM devices in critically ill patients or those with hypotension, hypoxemia, acidosis, anasarca, or vasoconstriction (24) and the lack of accurate studies comparing CGM values and laboratory measurements. To overcome the accuracy limitation of POC testing, an ongoing study is testing the effectiveness of the Dexcom RT-CGM system compared to arterialized venous blood sample measurements using the Yellow Springs Instrument (NCT02880267).

Our study has some limitations, including a relatively small sample size and a low number of hypoglycemic episodes observed. Future studies should include a larger sample size with power to test whether using RT-CGM devices with a low glucose alarm can safely achieve tighter glucose control, reducing hypoglycemia and better clinical outcome (NCT05135676). The current study notably included a mixed population of both general medical and surgical patients with type 1 and type 2 diabetes, as well as a significant proportion of minority patients, who are frequently an underrepresented population in clinical trials. With the increased use of CGM among ambulatory patients, one can expect an increasing number of patients choosing to use CGM while hospitalized. Thus, this study provides significant evidence on the methodology and implementation for further interventional studies using RT-CGM systems among hospitalized patients with diabetes.

In summary, our results indicate that the inpatient use of RT-CGM is safe and effective in guiding insulin dosing, resulting in trends toward improvement in hypoglycemia and hyperglycemia, as well as a significant reduction of recurrent hypoglycemic events compared with POC glucose testing.

Funding. This investigator-initiated study was funded by Dexcom. E.K.S. is supported in part by a Veterans Affairs Merit award from the U.S. Department of Veterans Affairs Clinical Science Research and Development Service (1I01CX001825). R.J.G. is partially supported by research grants from National Institutes of Health (NIH)/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grants P30DK111024 and 1K23DK123384-03. P.V. is supported in part by NIDDK grant 1K23DK113241. G.D. is supported by the NIDDK under award number 1K23DK122199-01A1. F.J.P. is supported in part by NIDDK grants K23GM128221 and P30DK111024-06. G.E.U. is partly supported by research grants from the NIH/National Center for Advancing Translational Sciences (Clinical and Translational Science Award program grant UL 3UL1TR002378-05S2) and the NIH/National Center for Research Resources (NIH/NIDDK 2P30DK111024-06).

The contents do not represent the views of the U.S. Department of Veterans Affairs or the U.S. government. The funder had no role in study design, data collection, data interpretation, or in manuscript writing.

Duality of Interest. E.K.S. has received unrestricted research support from Dexcom (to Baltimore VA Medical Center and University of Maryland) to conduct clinical trials. R.J.G. received research support to Emory University for investigator-initiated studies from Novo Nordisk, Dexcom, and Eli Lilly and Company and consulting fees from Sanofi, Eli Lilly and Company, and Weight Watchers, outside of this work. P.V. has received consulting fees from Merck and Boehringer-Ingelheim. G.D. has received research support from Insulet. F.J.P. has received research support from Merck, Dexcom, and Insulet and consulting fees from Boehringer Ingelheim, Eli Lilly and Company, Dexcom, and Medscape, Inc. G.E.U. has received research support (to Emory University) from Astra Zeneca, Bayer, and Dexcom. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.K.S. and G.E.U. researched the data and wrote the initial drafts of the manuscript. A.U., W.Z.C., B.A., L.G.S., I.M., S.L., E.M., C.G., K.M., C.C., R.M., W.H.S., M.C.P.-G., and S.C. served either as study coordinators screening and randomly assigning research candidates and collecting data or as clinical providers managing subjects daily, including weekends, ensuring successful completion of the study protocol. R.J.G., P.V., A.L.M., G.D., M.F., T.I., F.J.P., L.G.S., I.M., and S.C. reviewed and edited the study proposal and manuscript. L.P. analyzed the data. G.E.U. wrote the research proposal and researched the data. G.E.U. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3–7 June 2022.