Predictive Glucose Trends From Continuous Glucose Monitoring: Friend or Foe in Clinical Decision Making?

Abstract

In this commentary, we briefly review the currently recommended approaches to interpretation and management of continuous glucose monitor (CGM) rate of change (ROC) trend arrows and discuss the inherent difficulty in incorporating practical recommendations for their application into routine clinical care. We have limited our review and discussion to the currently available Dexcom G5 and G6 CGM systems and Abbott’s Freestyle Libre flash glucose monitor (FGM) system, as they are the most widely used and currently approved for nonadjunctive use in the United States.

It is now well-established that personal use of continuous glucose monitoring (CGM) devices improves glycemic control, reduces hypoglycemia, and improves quality of life for people living with both type 1 (T1D) and type 2 diabetes mellitus (T2D).^1-4 In recent years, both the quality and usability of CGM devices has dramatically increased. Clinical experience, expert recommendations, and a growing body of evidence have demonstrated the safety and efficacy of incorporating CGM into personal diabetes management. However, point-of-care self-monitoring of blood glucose with glucometers (SMBG) is still the most widely used approach to glucose monitoring. CGM users have generally been encouraged to complement CGM data with SMBG data, and until recently, CGM devices have required calibration with intermittent SMBG glucose data to operate effectively.⁵ However, we all recognize a foreseeable shift toward a future where CGM data replace routine SMBG, just as SMBG previously supplanted urine glucose monitoring.

In line with this thinking, the US Food and Drug Administration (FDA) recently approved two CGM systems, Abbott’s Freestyle Libre flash glucose monitor (FGM) and the Dexcom G6 CGM, for stand-alone use (without the need for SMBG for calibration) and importantly, for nonadjunctive use in clinical decision making, barring a few specific circumstances.^6,7 With these new advancements, it is important to pause and contemplate what the future of glucose monitoring may look like for our patients with diabetes, as well as the health care teams who will be called upon to teach, review, and implement advanced glucose monitoring strategies into busy clinical practices. Despite the evidence demonstrating benefit, and growing enthusiasm for new technologies, current rates of CGM use remain relatively low in patients with diabetes, with most recent data from the T1D Exchange demonstrating that only 22% of pediatric patients with T1D utilize CGM, though this number is increasing, particularly in children <6 years old.⁸ One potential reason for lack of adoption of CGM is perceived barriers to use by clinicians. These potential barriers include high cost, variable or insufficient insurance coverage, shortages of skilled educators, limited time to review CGM data in a busy clinic and to keep up with rapid advances in the field, mistrust in diabetes technology, and limited understanding of how to best utilize and incorporate the wealth of information provided by these devices.^9,10

What has become clear in recent years is that despite barriers to the usage of such technology perceived by the medical community, people living with diabetes are interested in using CGM technology. Those who use CGM technology are also utilizing rate of change (ROC) trend arrows a great deal to inform aspects of their therapeutic decision making, even with limited guidance from their medical providers. Patients may alter both the timing and dosing (more or less) of meal-time and correction insulin and the timing and content of meal consumption depending on the ROC trend at the time of a meal. Often these decisions vary significantly between individuals.¹¹

We have finally entered the exciting era that was heralded nearly 20 years ago, when CGM devices first held the tantalizing possibility of allowing patients to anticipate future glucose trends and incorporate ROC trend arrows into real-time decision making. However, we should recognize that we remain limited by the paucity of published high quality data that informs our approach to incorporating ROC data into real-time behavioral adjustments. In particular, how we educate our patients on the adjustment of insulin dosing in response to ROC arrows is fraught with complexity. The majority of our collective recommendations are rooted in expert clinician experience with disparate view points and no current consensus guidelines based on high quality data. To add further complexity, ROC arrows have different appearances and meanings across different devices, and more recent literature has suggested that ROC trend arrows may not be a very reliable marker of future glucose change.¹²

Discussion

Several approaches to incorporating ROC arrows into clinical decision making have been published within the last few years, including the Scheiner, Pettus/Edelman, Endocrine Society, and Klonoff/Kerr perspectives.^13-16 While our commentary primarily focuses on the use of devices approved for nonadjunctive use it is worth acknowledging that many of the recommended approaches to the management of ROC trend arrows include recommendations for Medtronic devices. Medtronic’s most recent sensor, the Guardian Sensor 3, while currently lacking FDA approval for nonadjunctive use, is very likely already used in such a fashion, as was the case with earlier CGM versions by other manufacturers in the past.

The first approach to incorporating ROC trend data into insulin dosing decision making was the DirecNet Applied Treatment Algorithm (DATA) method, which was published more than a decade ago. After 10 years, this small study using the Abbott Freestyle Navigator CGM system remains the only approach studied in a controlled clinical trial and suggested incremental adjustments in bolus insulin dosing of between 10-20% of the intended bolus dose based on ROC arrow directionality at the time of bolus.¹⁷ In the 30 children and adolescents enrolled in the DirecNet study, no significant increase in severe hypoglycemia was demonstrated utilizing the DATA approach. While this method is relatively conservative, with a favorable safety outcome as indicated by no increase in severe hypoglycemia rates, it may prove difficult to implement and translate into concrete insulin dose recommendations for patients of lower health literacy or numeracy where calculations based on percentages of a dose are required or for those individuals on multiple daily injections (MDI) of insulin given dose constraints of 0.5 unit-1 unit increments available to MDI users.

In the first published recommendation for management of ROC arrows aimed at both patients and health care providers, Scheiner published an alternate suggested method, based on his clinical experience using Medtronic and Dexcom CGM devices.¹³ He recommended utilization of a patient’s insulin sensitivity factor to either add or subtract a fixed amount of bolus insulin to offset anticipated changes in glucose, as estimated by ROC trend arrows. While his recommendations for adjustment for hyperglycemia are again relatively conservative, it is our opinion that his recommended modifications for downward-trending sensor glucose may not be significant enough to avoid hypoglycemia for those utilizing the Dexcom system, given his definition of a modestly changing glucose as one changing 2-3 mg/dL/min, indicated by one 90° up or down arrow.

Subsequently, Pettus and Edelman suggested an approach to insulin dosing adjustment utilizing 30-minute “anticipated” glucose levels based on ROC arrows for the Dexcom G5 system.¹⁴ They also make recommendations regarding timing of meal consumption and insulin administration in scenarios where glucose levels may be changing rapidly, such as “waiting for the bend” after the administration of meal time insulin in the case of rapidly increasing glucose, or delaying administration of meal-time insulin in the case of rapidly decreasing glucose to mitigate potential hypoglycemia. They do not make recommendations for insulin bolus adjustments in the event of sensor glucose changes >3 mg/dL/min and pragmatically advise of certain situations in which patients should confirm CGM data with SMBG data before making a decision.

In contrast, Klonoff and Kerr proposed a method where fixed doses of insulin are added to the planned insulin bolus.¹⁶ Their “simplified approach” recommends fixed 1-unit, 1.5-unit, and 2-unit changes to bolus insulin dosing based on rates of glucose change, ranging between 1-2 mg/dL/min, 2-3 mg/dL/min, and >3 mg/dL/min ranges, respectively. They suggest their recommendations could be incorporated by a wider CGM user base including those using the Dexcom G4-6 systems and several Medtronic systems. Their recommendation utilizes 45-minute, “mid-point,” anticipated glucose values for the ROC arrows and then application of both the “rules of 1500 and 1960” (sensitivity factor estimation calculations) in an effort to determine a minimum total daily dose of insulin where this recommendation would be expected to avoid hypoglycemia. While the application of the “rules of 1500 and 1960” attempts to define a range of total daily doses of insulin intended to mitigate hypoglycemia and account for calculated insulin sensitivity, this approach has not been formally assessed and may not be appropriate at the extremes of obese, insulin-resistant individuals or young, lean individuals with exquisite insulin sensitivity. With that in mind, it is more straightforward for those with limited numeracy or health literacy, though using 0.5-unit increments may limit its utilization among MDI users.

In 2017, the Endocrine Society published another approach which recommends additional bolus doses based on anticipated glucose change and typical insulin sensitivity for both adults as well as children and adolescents.^15,22 The authors make note of the variable insulin sensitivities at various stages of life and suggest several pragmatic considerations such as utilizing a higher bedtime target of 130 mg/dL in children and adolescents and reducing insulin doses by at least 50% in older and frail adults in certain circumstances. In addition, they recommend avoiding adjustment of insulin dosing based on ROC trend arrows for the 4 hours following a meal bolus, as well as modifying the timing of insulin administration around meals in the setting of rapidly changing glucose.

Recently, intriguing data questioning the accuracy of ROC indicator arrows on both the Dexcom G5 and Freestyle Libre systems was published by Freckmann et al. They reported that ROC trend arrows match measured tissue glucose change in only approximately 60% of cases.¹² Of most concern is that in over 10% of cases, particularly around the time of carbohydrate ingestion and insulin administration, ROC trend projection fidelity with measured glucose was remarkably poor. While the relative sensor inaccuracy in the initial 12-24 hours of CGM wear that we are familiar with was again borne out in this study, of significant concern is that even with prolonged multiday use, sensor-derived glucose trends varied from calculated glucose trends based on fingerstick glucose measurements by 2 or more trend indicator categories 10% of the time. Notably, the majority of ROC indicators were shown to overestimate actual change, which may prove to be particularly problematic when it comes to recommendations regarding increases in insulin dose adjustment, potentially predisposing individuals to hypoglycemia, a reason for which many patients begin to utilize CGM in the first place.

While both the Dexcom G6 CGM and Freestyle Libre FGM systems recommend SMBG in the event that glucose alerts and/or readings do not match patient symptoms or if sensor glucose readings are suspected of being inaccurate, with the exception of the method published by Pettus and Edelman, there is little reference to confirming CGM data with SMBG data in the published approaches.^18,19 Much of the benefit of incorporating CGM ROC trends into decision making is as a result of the continuous nature of the sensor glucose data, such that the user knows where they are coming from, and to some extent, can estimate where their glucose is headed. This is a significant advantage over intermittent SMBG. However, given the concerns raised by recent data regarding fidelity of ROC trend projections and the understanding that many of the currently available CGM systems tend to falsely overreport hypoglycemia, albeit to variable degrees, are we ready to confidently recommend CGM as a stand-alone replacement for SMBG?^20,21

In addition, does the current SMBG value matter if we are most interested in the glucose trend? Which value should the patient believe in the first 12-24 hours of sensor wear, and will this change as the “warm-up” period shortens with newer technologies? Furthermore, how do we account for the documented variability between glucometers and test strips and that we are fundamentally measuring glucose in different body compartments (blood vs interstitial fluid) when asking patients to confirm CGM glucose values with SMBG values? Insulin sensitivity may not be consistent throughout the day and an increased insulin adjustment may be required in the early morning or after a high-fat meal, in comparison to recognizing potential risk in the late evening and overnight period, when there is an increased vulnerability to hypoglycemia. Should we apply caution to increasing the insulin dose based on increasing ROC in the overnight period? Similarly, should we recommend a reduction in insulin dosing based on the ROC trend when the sensor is reading “falsely” high relative to SMBG data? We have all faced these questions in daily practice, and it is challenging to give an accurate answer for many of these important questions. Given all of the disparate, complex recommendations that we have briefly reviewed, as well as the limited high-quality data that underpin these recommendations, we believe that it is prudent to consider the limitations of our current approach toward ROC arrow therapeutic decisions, especially when giving recommendations to patients using these arrows.

Conclusion

CGM is extremely valuable and remains underutilized in personalized diabetes management. The available systems continue to improve and more widespread adoption is expected to continue with increased provider and patient interest and comfort using CGM devices. However, we need to be cognizant of the limitations of incorporating ROC trend arrows into clinical recommendations and more rigorous studies are needed to better inform our future recommendations.