Use of Continuous Glucose Monitor in Critically Ill COVID-19 Patients Requiring Insulin Infusion

An Observational Study

Eileen R. Faulds; Andrew Boutsicaris; Lyndsey Sumner; Laureen Jones; Molly McNett; Keaton S. Smetana; Casey C. May; Elizabeth Buschur; Matthew C Exline; Matthew D. Ringel; Kathleen Dungan


J Clin Endocrinol Metab. 2021;106(10):e4007-e4016. 

In This Article


In this study, we report that a hybrid POC plus CGM model can be safely implemented in critically ill COVID-19 patients. There was a substantial reduction in the number of POC BG measurements per day, sensor validation was achieved in nearly all patients within 24 hours, and nurses demonstrated active use of the CGM device. The hybrid protocol resulted in acceptable glycemic control with minimal hypoglycemia, though accuracy was suboptimal for standalone use.

Our cohort of patients was very ill, indicated by admission laboratory values and need for mechanical ventilation, vasopressors, and renal replacement therapy. Patients who received the hybrid protocol generally had marked hyperglycemia and very high insulin requirements. Despite these factors, sensor validation was achieved in all patients, leading to an estimated 71% reduction in POC testing as measured against standard of care (24 POC measures/day). Davis and colleagues reported a similar reduction in frequency of POC glucose (63%) in 9 critically ill COVID-19 patients using a hybrid protocol using Dexcom G6, a computerized IV insulin algorithm, and decision support for sensor validation embedded in the HER.[31] Agarwal and colleagues reported a 60% reduction in POC testing using a hybrid model using Dexcom G6 that recommended POC testing during hypoglycemia and hyperglycemia or otherwise once daily.[32] In another report of 11 patients (excluding hyperglycemic emergencies, not on vasopressors), POC testing was reduced 33% using a protocol that allowed standalone CGM (Dexcom G6 or Medtronic Guardian Connect) use only when the CGM value was in target range and a minimum POC BG frequency of 4 hours.[33] Chow and colleagues assessed 30 COVID-19 positive ICU patients, in which the first 24 hours of data were masked.[34] Thereafter, Dexcom CGM values could be used if it was <20% of the POC. In the study, the number of daily POC glucose values did not decline, but detailed insulin infusion data were not available and frequency of monitoring was only 5.5 times per day during the masked period. Our findings provide additional reassurance for broader use of the hybrid POC + CGM protocol and extended benefit through reduced nursing workload even for patients without IV infusion pump externalization by reducing the frequency of POC testing. In addition, potential benefits to the patient include less pain, blood loss, and sleep interruptions as well as reduced costs.[15]

There was some discrepancy in protocol interpretation and CGM use during the initial validation period. Although the majority of sensors (n = 14) were validated within 2 hours, progression to nonadjunctive use was not unanimous. In 2 instances, CGM glucose was used prematurely, whereas in other instances (n = 5), nurses continued POC BG monitoring after initial CGM validation had been achieved. It is interesting to note that progression to nonadjunctive use within the protocol's guideline improved over time and the 2 instances of premature use occurred after several months of CGM use in the ICU. Another recent study also reported additional POC testing outside the protocol indications.[33] In our work, nurses were not only documenting CGM values in the EHR (around 20 times per day) but were using the values to guide treatment decisions as evidenced by a median of 11 or more titration values per day documented in the EHR. In this cohort, devices were all placed by nursing leaders or the consult service. Thus, additional training and improvements in workflow are needed to expand this task to the bedside nurse. Additionally, integration of CGM data into the EHR, computerized decision support providing calculation of sensor-meter agreement, and incorporation of trend arrow data could increase benefit.[31] Previously, we reported overall favorable nursing acceptance using a focus group format;[23] in other research, two-thirds of nurses surveyed believed that the addition of CGM improved patient care.[34] Within the study institution, nursing feedback and experience was instrumental in the decision by nursing leadership to assume the budget for purchase of all future CGM devices.

Despite refractory hyperglycemia at baseline and high insulin requirements, glucose control was achieved overall. Patients achieved 64% TIR on day 1 and 72% TIR thereafter, which is consistent with previous studies in which 46%[34] and 71%[31] TIR was achieved. Although sensor hypoglycemia was detected in multiple patients, the clinical significance of this is unclear because it was generally of very brief duration and was only confirmed by POC value in 1 patient, who was otherwise at very high risk for hypoglycemia. Nevertheless, in critically ill patients, the preference would be for maximizing sensitivity for detecting hypoglycemia over specificity.

Overall, accuracy was acceptable but was generally lower than reported in studies of noncritically ill patients.[35] In addition, accuracy is less robust at lower glucose values, thus underscoring the rationale for a hybrid model. However, it is important to emphasize that this was not an accuracy study. The hybrid model requires more POC BG monitoring during periods of sensor-meter disagreement, making it appear that the overall accuracy is lower than if sampling of sensor-meter pairs occurred systematically at regular intervals. We attempted to control for this by nesting values by day (reporting the mean of each patient day rather than treating sensor meter pairs as independent observations). Furthermore, MARD and MAD were similar to values on day 1, suggesting that once initial validation was achieved, progression to intermittent POC validation (every 6 hours) with interim nonadjunctive CGM use was safe and acceptable, even on day 1 of sensor wear. Other reports among COVID-19 ICU patients reported similar or slightly better accuracy.[32,33]

We did not identify an effect of other clinical variables on sensor accuracy, including oxygen saturation, mean arterial pressure, and vasopressor use. This is in agreement with other data.[36,37] However, there was a limited number of sensor-meter pairs during hypotension or hypoxemia and an underlying effect cannot be excluded. Moreover, it is possible that only profound hypoperfusion or cardiac arrest is of concern.[31] The presence of insulin as a home medication was associated with lower MARD. Although patients overall had very high insulin requirements during their hospitalization, fewer than half were on insulin before admission. It is possible that those not previously on insulin were experiencing more profound stress hyperglycemia, which may be associated with higher MARD. Additional issues reported in other studies include interference attributed to electrocautery, cooling blankets, and mechanical compression.[31,33]

This study was limited by the observational design and thus adherence to the hybrid protocol and frequency of POC measures were not rigorously surveyed. Because patients spent the majority of time close to target glucose range (140–180 mg/dL), we did not have robust assessment of sensor accuracy in the hypoglycemia or hyperglycemia range, and cannot definitively say the presence of hypoxia, hypotension, or other factors would not prove to significantly affect CGM accuracy in a larger sample. Although patients achieved a seemingly acceptable percentage of TIR, there is no consensus on standard TIR among this population, making comparisons difficult. The reference method was POC glucose using a device that is FDA-approved for ICU patients, rather than a central laboratory reference method. Although the use of a laboratory reference method would have been impractical in this real-world setting, the use of POC glucose does limit the comparison of MARD to industry standard. In addition, the goal was to establish the feasibility of a hybrid protocol, not to justify standalone use. Within the protocol itself, thresholds for CGM validation were chosen based upon prior knowledge and a balance between risks and benefits; however, the optimal criteria for inpatient CGM use are unknown. The FDA's standard-of-care recommendation for hospital glucose meters is 98% of values ±15% for glucoses ≥75 mg/dL.[38] This guidance is not specific to CGM and it is important to note that meter and CGM accuracy cannot really be compared given the additional data (ie, trends, alarms, frequency of measurement) provided by CGM systems. Recently, the FDA outlined a new 510K (premarket approval) route for CGM systems designated as "integrated CGM" with additional special controls governing accuracy. According to this guidance, overall accuracy is defined as 87% of values within ±20%.[38] Again, this guidance is not specific to inpatient use, and future research is needed to establish optimal inpatient metrics for accuracy. We did not include a control group from concern that a suitable control group of COVID-19 patients receiving IV insulin was not available or otherwise would not be comparable because of the tendency to rely on subcutaneous insulin, even in the setting of critical illness. Placement on the upper arm rather than the abdomen was necessary given the frequent pronation in this population, but may somewhat limit the generalizability. In addition, we were unable to formally evaluate the presence of any sensor site issues because of the retrospective nature of the study and lack of standardized documentation.