Nonadjunctive Use of Continuous Glucose Monitoring for Diabetes Treatment Decisions

Self-Monitoring of Blood Glucose Is Suboptimal

Self-monitoring of blood glucose (SMBG) is the de facto standard used by people with diabetes to determine insulin doses, assess their glycemic state, and make informed management decisions. The accuracy of SMBG is variable depending on the particular meter with accuracy ranging from excellent to poor under controlled conditions; less accurate results likely occur during actual use.^1,2 SMBG accuracy is dependent on good hand washing with complete drying ³ and requires the use of properly stored, unexpired test strips, and proper blood application. ⁴ Depending on the specificity of the enzyme employed in the test strip, significant impacts on SMBG accuracy can occur as a result of interfering substances, or common conditions such as anemia. ⁵ Existing SMBG technologies remain burdensome and are performed less often then recommended, ⁶ creating for many the desire for alternative technology to reduce this burden. ⁷

Introduction of CGM

Beginning in 2005, modern real-time continuous glucose monitoring (CGM) systems became commercially available. Commercial CGM employs an enzyme-coated wire that is 2 to 3 times the thickness of a human hair. The wire is inserted by the patient into subcutaneous tissue and measures interstitial glucose via generation of an electrical current when glucose reacts with the enzyme glucose oxidase. Glucose is related to this electrical current through a process of calibration whereby a finger-stick glucose value is taken by the patient and entered into the CGM system. Glucose values are displayed on a receiver, insulin pump display, or smartphone. These values are updated continuously and reported every 5 minutes, with sensors labeled for 6 or 7 days of use.^8,9 Past CGM systems, including the Dexcom G4, Medtronic Enlite, and FreeStyle Navigator, received regulatory approval in the United States specifically for adjunctive use, with labeling indicating treatment decisions are to be based on blood glucose values obtained through SMBG, not CGM readings.

Obstacles to Nonadjunctive CGM Use

Historically, the Food and Drug Administration (FDA) presumably did not allow CGM system labels to include nonadjunctive use because of concerns of inaccuracy that could lead to inappropriate treatment decisions. Factors that may impact CGM accuracy include calibration error, compression artifact, delay, drift, and interfering substances. Calibration values are critical as they are utilized to quantify the sensitivity of the sensor to glucose. If a person enters an erroneous reference capillary blood glucose (CBG) value, if their fingertip has sugar or syrup on it for example, this error will artificially increase the CBG and can interject significant error into a CGM system, ¹⁰ whereas compression of the sensor site, such as lying on sensor during sleep, can cause artificially low signals. ¹¹ Sensor drift, which can artificially increase or decrease the sensor value, is incompletely understood. Sensor drift relates to the foreign body response bringing inflammatory cells to the sensor insertion cite, resulting in drift of the sensor signal possibly from these cells consuming oxygen and glucose as well as producing compounds such as hydrogen peroxide. ¹² Sensor delay is related to the physiologic lag between interstitial glucose and blood glucose concentrations,^13,14 and is impacted by whether glucose levels are rising or falling. ¹⁵ Delay can also be imparted due to smoothing of the signal during processing. ¹⁶ Interfering substances, particularly acetaminophen and to a lesser degree vitamin C, can artificially raise sensor values when they are oxidized at the indicating electrode. ¹⁷

Improvements in CGM Technology

Advances in CGM technology have significantly mitigated the risks from these confounding factors and have thereby improved confidence in these systems. Such technology advances have included reducing needle size in an effort to reduce the foreign body response, improved manufacturing processes to reduce batch-to-batch variability, better sensor membrane chemistry to eliminate the effects of interfering compounds, and advanced signaling processing to reduce delay and filter out noise and compression artifacts. And now, despite device labeling, many people use CGM for treatment decisions without confirmatory SMBG. This is supported by clinical trials which have shown that CGM users have decreased their SMBG frequency; in data from the Type 1 Diabetes Exchange, approximately 50% of CGM users report decreasing their SMBG frequency when using CGM. ¹⁸ As CGM accuracy has improved and patients gain trust in their CGM device, nonadjunctive use (treatment decisions based on CGM alone) are expected to increase. ¹⁹

CGM-based decision making offers multiple advantages over SMBG-based decisions. CGM provides not only a discrete glucose value, but also the benefit of the glucose history including trends, rate of glucose change as well as the direction of the change. Alarms can warn the wearer if glucose is drifting too low or high when they might otherwise be unaware. This allows proactive action on the part of the wearer, preventing prolonged hypoglycemia or hyperglycemia. Furthermore, these continuous historical data are available to patients and care providers.

Consider, for example, a person with diabetes preparing to go to bed for the night. She performs a nightly SMBG with a blood glucose value measured at 100 mg/dL; the decision of what to do next is unclear. This person would not know whether a snack should be eaten to avoid a nighttime low or if there should be no treatment since her blood glucose is within her target range. Now consider the same situation with CGM; if the blood glucose is trending downward, then a bedtime snack is advisable. If the CGM indicates the glucose trend is rising, a snack is likely not required and may cause morning hyperglycemia.

Evidence for Improved Outcomes With CGM

There is strong evidence supporting the reduction in hyper- and hypoglycemia with consistent CGM use.^20-22 In the STAR-3 trial, comparing the use of multiple daily insulin injections (MDI) to sensor-augmented pump use, it was the increased frequency of sensor use, and not the use of pumps per se, that was associated with a greater reduction in A1C. ²³ This association of reduced A1C with CGM use was confirmed by the SWITCH crossover study, which demonstrated a marked improvement in A1C during the intervention period with real-time CGM use. The reduction in A1C largely dissipated when the CGM was blinded, and returned when the CGM data were again unblinded. ²⁴ They found that when CGM systems were used in real-time, subjects demonstrated more intensive manipulation of insulin delivery, with an increase in the number of insulin boluses, more frequent use of bolus calculators, and increased use of temporary basal rates and basal suspensions. In a large survey of experienced CGM users across the United States, subjects were probed to describe how they use their glucose data to reduce their A1C and decrease hypoglycemia. ²⁵ Survey participants reported several common themes including making large insulin dose modifications, adjusting insulin timing, prophylactically treating potential hypoglycemia with carbohydrates, and responding to low and high glucose alerts. Many respondents also lowered their individualized glucose targets after CGM was initiated.

Level of CGM Accuracy Required for Nonadjunctive Use

On review of the available data, CGM-based decision making can clearly improve glucose control and reduce glucose excursions. However, clinical safety for nonadjunctive CGM use has not yet been proven in a trial setting. To determine the threshold for safe CGM use as a nonadjunctive device, Kovatchev and colleagues conducted extensive in-silico testing. ²⁶ Compiling data from 56 pump users, they were able to simulate the impact of 7 CGM accuracy levels, with a mean absolute relative difference (MARD) of 3-22%. An abrupt increase in the risk of hypo- and hyperglycemia was noted in their simulations at an MARD of > 10%. Wilinska and Hovorka presented a similar finding when modeling the use of CGM values to drive insulin delivery in an ICU setting using 3 established glucose control protocols. ²⁷

The published accuracy of the recently released Dexcom 505 algorithm meets this proposed MARD safety threshold of less than 10% in clinical testing across a range of glucose levels from 40 to 400 mg/dL (2.2-22 mmol/L). ²⁸ The Randomized Trial Comparing Continuous Glucose Monitoring With and Without Routine Blood Glucose Monitoring in Adults with Type 1 Diabetes (REPLACE-BG) is currently underway, evaluating the nonadjunctive use of the Dexcom system with the 505 algorithm in adults with type 1 diabetes (T1D) using insulin pumps. A total of 225 subjects will be randomized in this study, half randomized to CGM with confirmatory SMBG treatment, and the other half to CGM-based decision-making. The CGM + SMBG group will wear a CGM device daily and verify CGM readings with a SMBG measurement prior to making any diabetes management decisions. The CGM only group will use the CGM values without a SMBG confirmation in most situations. CGM values will not be used for treatment decisions at times when the risk for inaccuracy are higher, including the first 12 hours following initial sensor placement, after taking acetaminophen, or when CGM values are discordant with the user’s symptoms. The primary outcome for this trial will be the time spent in the target range of 70 to 180 mg/dL, measured with CGM over the 6 months of the study. The results of this study will better define the safety of nonadjunctive use of the Dexcom CGM data for diabetes management decisions.

Current Usage of CGM in Decision Making and Insulin Dosing

CGM systems are already being used for diabetes treatment decisions in Europe. The Abbott Navigator II is approved for use in determining insulin doses when glucose is not changing rapidly and the Abbot Libre Flash Glucose Monitor, while not a true CGM because it does not provide continuous data and glucose trends, is approved for treatment decisions if the person is not hypoglycemic, if glucose is not changing rapidly, and if symptoms are concordant with the system readings. Most recently, the Dexcom G5 Mobile CGM system is now approved in Europe to be used nonadjunctively. The MiniMed 530G insulin pump system and the Medtronic Veo insulin pump can suspend insulin if the sensor falls below a threshold and the user doesn’t respond to an alarm. ²⁹ Likewise the Medtronic 640G insulin pump system provides a predictive low glucose suspend if the patient is not responding to alarms and the system believes a hypoglycemic event may be occurring in the near future. However, with these systems, patients are still expected to primarily base their dosing decisions on SMBG measurements.

Another growing body of literature that supports nonadjunctive use of CGM for dosing decisions comes from the field of artificial endocrine pancreas (AP) research. The AP is a system for automating the delivery of insulin and in some cases glucagon in response to CGM data. An AP system is typically composed of at least 1 CGM system, an insulin pump and optionally a glucagon pump for bihormonal systems, and a controller that runs on either the pump hardware or a separate device such as a smartphone. The recent advances in CGM accuracy have essentially eliminated the need for redundant sensors and groups such as at Oregon Health & Science University have transitioned from using 2 Dexcom sensors ³⁰ to a single Dexcom G4 sensor. ³¹

For AP systems, SMBG values are only used to calibrate the CGM system and not to alter dosing, except in the case of severe hypo- or hyperglycemia. Thus, AP trials offer a perspective on the risk versus reward impact that dosing based on nonadjunctive CGM could potentially have for people with T1D. Recent studies have shown that use of a bi-hormonal AP^32,33 and insulin only AP^34-36 both offer substantial benefit to people with T1D. Russell et al showed that use of a bi-hormonal AP for 5 days can significantly reduce mean glucose levels in people with T1D (133 mg/dL) compared with the use of sensor-augmented pump therapy (159 mg/dL). Thabit et al showed that for 58 adults with T1D who used a single-hormone AP over 12 weeks, glucose control and lower hypoglycemia were observed as well as a reduced hemoglobin A1C (difference of -0.3%) compared with sensor-augmented pump therapy. Importantly, in the Thabit et al study, 3 severe hypoglycemic events occurred, but only when the AP was not in use. ³⁶ While the AP is not yet commercially available, numerous groups are moving in the direction of commercialization as summarized by Kowalski. ³⁷ Results from these studies indicate that the use of CGM for nonadjunctive dosing is possible and when used properly can lead to significant benefit to patients with regards to improved time in euglycemia and reduced hypoglycemia.

Conclusions and Future Considerations

Use of CGM in people with T1D has historically been low but is significantly increasing. ¹⁸ Using CGM without the need for frequent CBGs and the ability to make more informed diabetes treatment decision with CGM data is a major shift in the management of T1D. How to best use this wealth of data is not yet known and clear guidelines are lacking. The American Diabetes Association guidelines indicate CGM is a useful tool to lower A1C or for those with hypoglycemia unawareness and/or frequent hypoglycemic episodes. ³⁸ These guidelines, however, make no recommendations on how to instruct patients to use CGM data; the guidelines just note the need for robust diabetes education. In the landmark JDRF CGM trial assessing the impact of CGM use on glucose control in T1D, subjects wearing unblinded sensors were instructed to increase/decrease boluses by 10-20% if glucose was significantly rising/falling. Surveys have indicated patients in nonresearch settings often make much more dramatic adjustments than these based on trend arrows. ²⁵

The optimal way to adjust insulin based on CGM data is complex. It depends on the timing of the last bolus, meal, and activity level, among many other factors. In the future, insulin dosing recommendations will best be handled with decision support systems that can automatically recommend dosing adjustments based on a host of factors. Decision support systems will run on smartphones or be embedded in insulin pumps or CGM systems and potentially be connected to cloud-based servers. Such decision support systems will be capable of storing a vast array of historical data from the patient that resides on these cloud servers. These data will potentially be available to care providers during patient visits and to patients in real time to help them make better decisions to manage their glucose levels. Access to such large amounts of data from patients presents both an opportunity and a challenge to the community with regard to how such data will be managed and appropriately collected, transmitted, and ultimately presented to patients and care providers.

In summary, for systems meeting accuracy thresholds (MARD <10%), the enhanced information available with CGM use support the use of CGM-based treatment decisions. Extensive modeling, patient surveys and common clinical experience suggest that many people using these systems are already using CGM for treatment decisions. In light of this, improved understanding of CGM use to advance clinical care recommendations for patients is needed, including reinforcement that if CGM readings do not match clinical symptoms a SMBG value should be obtained before making a treatment decision.