Expert Clinical Interpretation of Continuous Glucose Monitor Reports From Individuals Without Diabetes

Abstract

Background:

Clinical interpretation of continuous glucose monitoring (CGM) data for people without diabetes has not been well established. This study aimed to investigate concordance among CGM experts in recommending clinical follow-up for individuals without diabetes, based upon their independent review of CGM data.

Methods:

We sent a survey out to expert clinicians (n = 18) and asked them to evaluate 20 potentially challenging Dexcom G6 Pro CGM reports (and hemoglobin A1c [HbA1c] and fasting venous blood glucose levels) from individuals without diabetes. Clinicians reported whether they would recommend follow-up and the reasoning for their decision. We performed Fleiss Kappa interrater reliability to determine agreement among clinicians.

Results:

More than half of expert clinicians (56-100%, but no clear consensus) recommended follow-up to individuals who spent >2% time above range (>180 mg/dL), even if HbA1c <5.7% and fasting glucose <100 mg/dL. There were no observed trends for recommending follow-up based on mean glucose or glucose management indicator. Overall, we observed poor agreement in recommendations for who should receive follow-up based on their CGM report (Fleiss Kappa = 0.36).

Conclusions:

High discordance among expert clinicians when interpreting potentially challenging CGM reports for people without diabetes highlights the need for more research in developing normative data for people without diabetes. Future work is required to develop CGM criteria for identifying potentially high-risk individuals who may progress to prediabetes or type 2 diabetes.

Introduction

There is high consumer interest in using continuous glucose monitoring (CGM) for health and wellness even among people without diabetes. ¹ The US Food and Drug Administration has recently approved CGM sensors for over the counter purchase (without a prescription), making them more accessible for individuals without diabetes.^2,3 It is expected that an increasing number of people without diabetes will bring CGM reports to their providers for concerns related to glucose fluctuations. Health care providers need to be prepared for interpretation of CGM data for people without diabetes. Currently, there are no guidelines for how to interpret CGM data or understanding of the utility of CGM in this population. ⁴

One can extrapolate, from guidelines for interpreting CGM reports from individuals with diabetes, that important initial CGM metrics to focus on may be time in range and time spent above and below range.^5-7 While time in tight range (70-140 mg/dL) has been proposed as a normal range for healthy individuals, ⁸ our recent large cohort study (n > 1000) indicates that individuals without diabetes or prediabetes spend more than 10% of their CGM readings, on average, above 140 mg/dL. ⁹ Moreover, at the time of the present interpretation study, the default settings for ambulatory glucose profile (AGP) reports defined time in range as 70 to 180 mg/dL. It is critical to consider that clinicians will rely heavily on the available framework (ie, default AGP report metrics) when interpreting CGM data for this population. Therefore, much thoughtfulness is required when considering what metrics should be presented in CGM reports.

Although CGM sensors have not been approved to diagnose or predict prediabetes or diabetes, ¹⁰ this possible application for CGM holds promise for the future. ¹¹ In the absence of specific guidelines for CGM interpretation in those without diabetes, diabetologists and endocrinologists with CGM expertise can help advise on CGM data interpretation in this new population. Therefore, we designed a study to evaluate how expert clinicians interpret CGM data from individuals without known diabetes based on AGP reports available at the time of the survey.

Methods

We recruited 18 trained clinicians (mostly diabetologists or endocrinologists) with expertise using CGM. They were included if they had extensive clinical experience interpreting CGM from people with diabetes and authored a research article on the topic of CGM within the last five years. In April 2024, we asked clinicians to evaluate 20 factory-calibrated Dexcom G6 Pro CGM reports (that includes the one-page AGP report) from individuals without known diabetes who wore the sensor for eight or more days (including the first partial day) without any manual calibration during the wear period. This criterion provided seven full days of CGM sensor wear (70% of the possible ten days), which is one of the criteria suggested in consensus recommendations.^5,12

In a survey, clinicians were first asked to report whether they would “recommend follow-up with a health care provider” based on the one-page Professional CGM AGP report alone (Supplemental Figure 1). Clinicians could select one of three options: “yes,” “no,” or “not enough data for recommendation due to potentially faulty CGM device” in a REDCap survey.^13,14 We then asked them to evaluate the full Professional CGM report, hemoglobin A1c (HbA1c), and fasting venous blood glucose levels corresponding with each report (Supplemental Figure 2). We asked them whether they would change their decision based on the new information and to describe what factors they considered when evaluating their recommendation.

CGM Reports Selected for Survey

We selected 20 CGM reports from a sample of individuals living in a community setting, without known diabetes and not taking glucose-lowering medication. CGM reports were further excluded if from individuals with HbA1c ≥6.5% or fasting blood glucose ≥126 mg/dL. We chose ten CGM reports from individuals with prediabetes, defined as HbA1c 5.7% to 6.4% and/or fasting glucose 100 to 125 mg/dL, and ten reports from individuals without prediabetes (HbA1c and fasting glucose below those thresholds). Hb A1c and fasting glucose were measured from blood collected on the same day, a few hours before CGM sensors were connected to these individuals. CGM reports were not selected at random. Instead, we chose CGM tracings with a diverse range of features (eg, substantial time spent in high or low glucose levels, mismatch between glucose management indicator [GMI] and HbA1c, brief hyperglycemic events vs prolonged hyperglycemic events, “sawtooth” patterns, elevated or variable overnight glucose levels). Any identifying information was redacted from CGM reports and we provided no additional information about medical history or other data. Individuals who provided these reports gave written informed consent and the institutional review board at Boston University Medical Center approved the study protocols.

Professional CGM reports were downloaded from Dexcom Clarity software. ¹⁵ At the time this study was conducted (April, 2024), CGM AGP reports are a single page that includes information about the time spent in glucose ranges (<54, 54-69, 70-180, 181-250, >250 mg/dL), the average (mean) glucose, GMI (which is directly calculated from mean glucose, but meant to be a comparison measure of HbA1c), ¹⁶ coefficient of variation, percent of time the CGM was active, an AGP figure showing a summary of glucose values across the course of an average day, and smaller daily glucose profiles. ⁵ The full CGM report includes larger daily glucose profiles shown with different formatting, with daily time spent in glucose ranges and other daily and hourly statistics, an overlay of daily glucose profiles, standard deviation, the number of days with CGM data, and average number of calibrations per day. The full CGM reports also include any alerts, activity/exercise, meals, insulin doses, and so on. But these features were not available for individuals wearing the CGM sensors in our current study because sensors were provided in a blinded mode. Screenshots showing an AGP report and part of a CGM report are shown in Supplemental Figures 1 and 2.

Statistical Analysis

We reported the order in which CGM reports were presented in the survey, the HbA1c and fasting blood glucose corresponding with each CGM report, and the percent of clinicians who recommended follow-up or reported a potentially faulty device. We also reported the average Fleiss Kappa interrater reliability^17,18 for agreement among clinicians, based on their recommendations for follow-up (vs no follow-up or not enough data to recommend). For each of the 20 CGM reports, we also presented the CGM metrics that were provided on the AGP report.

Results

Individuals providing the CGM reports had a mean age of 60.6 years (range 47-74 years), mean body mass index (BMI) of 28.0 kg/m² (n = 6 with obesity, BMI ≥30 kg/m²), and 9/20 were women. The overall agreement to recommend follow-up for an individual without diabetes when presented with their CGM report, HbA1c, and fasting glucose value among clinical experts was low (Fleiss Kappa = 0.36). We observed universal agreement among clinician experts for only 3/20 CGM reports (Tables 1 and 2), including only a single report that all experts agreed not to recommend follow-up (CGM report number 8, Table 1). This individual did not have prediabetes, had mean CGM glucose <100 mg/dL, and spent 0% time <54 mg/dL and 0% time >180 mg/dL. All clinicians agreed to recommend follow-up for two individual CGM reports (CGM report numbers 6 and 17, Table 2). Both reports were from individuals with prediabetes (HbA1c >5.7%), mean CGM glucose >150 mg/dL (GMI >7%), and 18% to 32% time >180 mg/dL. It was also notable that high mean glucose or GMI did not always coincide with high HbA1c or high time >180 mg/dL. In one case without prediabetes (CGM report number 10), where GMI and mean glucose were high (6.7% and 140 mg/dL), but time >180 mg/dL was only 1%, only 33% of experts recommended follow-up.

Table 1.

Expert Clinicians’ Final Follow-up Recommendation Rates for CGM Reports From Individuals Without Prediabetes, Displayed in Descending Order by Mean Glucose (and GMI).

Time in range percentages were displayed as dark red (<54 mg/dL), bright red (54-69 mg/dL), green (70-180 mg/dL), yellow (181-250 mg/dL), orange (>250 mg/dL). Follow-up rate is shaded if >50% of clinicians recommended follow-up.

Abbreviations: CV, coefficient of variation; CGM, continuous glucose monitoring; GMI, glucose management indicator; HbA1c, hemoglobin A1c.

Table 2.

Expert Clinicians’ Final Follow-up Recommendation Rates for CGM Reports From Individuals With Prediabetes, Displayed in Descending Order by Mean Glucose (and GMI).

Time in range percentages were displayed as dark red (<54 mg/dL), bright red (54-69 mg/dL), green (70-180 mg/dL), yellow (181-250 mg/dL), orange (>250 mg/dL). CGM report number and biomarkers from individuals with elevated HbA1c (5.7-6.5%) and/or fasting glucose (100-126 mg/dL) are bolded for emphasis. Follow-up rate is shaded if >50% of clinicians recommended follow-up.

Abbreviations: CV, coefficient of variation; CGM, continuous glucose monitoring; GMI, glucose management indicator; HbA1c, hemoglobin A1c.

The only observed trend that a majority of experts (56-100%) appeared to be following was to recommend follow-up for individuals who spent >2% time >180 mg/dL (Tables 1 and 2). Daily tracings from each CGM report were categorized by whether they spent >2% time >180 mg/dL and their prediabetes status (Supplemental Figures 3-6). A potential anomaly from this trend was CGM report number 2 (with the highest time >180 mg/dL), where we observed that only 83% of clinical experts recommended follow-up. It is clear from Supplemental Figures 1, 2, and 5 that most of the time above range in CGM report number 2 occurs on the last few days. The three clinical experts who did not recommend follow-up suggested that the CGM sensor may have been faulty, especially due to the apparent discordance from HbA1c.

We did not observe other trends for clinicians recommending follow-up based on mean glucose, GMI, or coefficient of variation. For example, features such as brief hyperglycemic events in an individual without prediabetes that resolved rapidly were concerning to some clinicians, but not others (Figure 1a). Furthermore, clinicians did not agree as to which CGM reports contained potentially erroneous or spurious data (ie, faulty device). The strongest agreement for suggesting a sensor was faulty was 39% of experts for CGM report number 3 (Figure 1b). Notably, another 39% recommended follow-up for this report, because apparent low glucose levels could either be spurious or have real physiological causes such as insulinoma or reactive hypoglycemia.

Figure 1.

Informative examples of CGM reports and expert clinicians’ reasons for their recommendations. (a) CGM report order number 9: highest percent time above range (>180 mg/dL) without prediabetes. (b) CGM report order number 3: the CGM report most frequently suggested to represent a faulty device.

Sensitivity analysis of clinicians given a first pass to evaluate all CGM reports without seeing the HbA1c and fasting blood glucose showed that >50% of clinicians recommended follow-up for all CGM reports that had at least 1% time >180 mg/dL (data not shown). After clinicians were provided with HbA1c and fasting glucose levels, fewer clinicians recommended follow-up for almost all CGM reports from individuals without prediabetes.

We summarized the reasons clinicians gave for their decisions in Table 3. Interestingly, many of the reported reasons some clinicians gave for recommending follow-up were the same reasons other clinicians gave for not suggesting follow-up or suggesting that there was a potentially faulty CGM sensor (eg, discordance between CGM metrics and HbA1c/fasting blood glucose). CGM report number 2 also exemplifies a feature mentioned by a few CGM experts where the last few days appear different than the rest of the report (Supplemental Figure 2).

Table 3.

A Summary of Expert Clinician Recommendations and Reasons Given for Recommending Follow-up or Suggesting a CGM Sensor Was Potentially Faulty.

Types of follow-up recommended by expert clinicians:

• Repeat blood biomarkers and CGM; consider OGTT or finger-stick measurements  • Determine if lifestyle advice is needed; consult dietitian  • Assess BMI, BP, CVD risk factors/global risk for insulin resistance and CVD

Reasons for recommending follow-up:  • Discordance between CGM mean glucose (or GMI) and HbA1c/fasting blood glucose  • Possible insulin deficiency: postprandial excursions, hyperglycemia, time >180 mg/dL, high variability  • Possible insulin resistance: high CGM later in the day or other features listed above  • Possible insulinoma: substantial hypoglycemia not expected for individuals without diabetes  • Possible GCK-MODY: elevated overnight glucose  • Contributing factors to recommend follow-up: younger age, family history of diabetes, other risk factors

Reasons for NOT recommending follow-up:  • HbA1c, GMI, mean glucose, TIR/TBR/TAR all in target  • Even with postprandial CGM spikes, HbA1c is in good control (ie, discordance)  • Postprandial hyperglycemia resolves quickly  • CGM metrics may not be accurate without diabetes  • Sporadic nocturnal hyperglycemia may be normal, possibly representative of late-night snacking  • Some clinicians who suggested no follow-up did mention that lifestyle recommendations may be helpful

Reasons for suggesting potentially faulty device:  • Discordance between CGM metrics and HbA1c/fasting blood glucose (sensor overestimation?)  • High CGM values only on last few days of wear  • Substantial hypoglycemia not expected for people without diabetes—compression lows/sensor error  • High overnight glucose not expected  • Contributing factors to an unclear report: substantial missing data (<70% capture rate), less than ten days wear

Abbreviations: BMI, body mass index; BP, blood pressure; CVD, cardiovascular disease; CGM, continuous glucose monitoring; GCK-MODY, glucokinase-maturity-onset diabetes of the young; GMI, glucose management indicator; HbA1c, hemoglobin A1c; OGTT, oral glucose tolerance test; TAR, time above range; TIR, time in range; TBR, time below range.

Discussion

Results from our study suggest that even clinicians with extensive CGM experience do not agree with one another on how they interpret or provide follow-up recommendations for individuals without diabetes based on CGM data. The one potential take-away for individuals without elevated HbA1c or fasting glucose is that a majority of clinicians (>50%) recommended follow-up with a health care provider if a CGM report shows that they spent >2% time above range (>180 mg/dL). With CGM sensors being increasingly marketed to individuals without diabetes, our findings may have major implications for an increased burden on health care providers to interpret CGM reports. We recently reported in >500 middle- and older aged adults with normoglycemia (HbA1c <5.7% and fasting glucose <100 mg/dL) from the Framingham Heart Study that roughly 20% of this population spend >2% time above 180 mg/dL (and the prevalence doubles for those with prediabetes). ⁹ Based on the expert clinician recommendations we reported in the current study, as CGM sensors become more widely available for the pool of individuals without diabetes, there may be a large number who are recommended to receive follow-up based on results from their CGM report. However, unfortunately, there is no clear consensus on the criteria for recommending follow-up.

Currently, CGMs are only recommended to manage established diabetes, ¹⁹ with interpretation suggested to focus mainly on time spent in different glucose ranges.^5-7 There are some important considerations for health care providers who are tasked with interpreting CGM reports from individuals without diabetes. The first is that we should not expect GMI (a metric that corresponds closely with mean CGM glucose) to reflect HbA1c, as these measures have been reported to be quite discordant in this population.^20,21 Therefore, at present, GMI cannot act as a surrogate for HbA1c in the diagnosis/screening for diabetes, but GMI and mean glucose may also provide information that is complementary to HbA1c. ¹⁶ Providers should also be aware of differences in sensitivity and accuracy among different brands and models of sensors^22,23 and that some sensors require manual calibration. There are numerous other individual-specific factors that may have an impact on sensor glucose metrics, including various medications, intercurrent illness, sensor wear sites, diet/physical activity, pregnancy, chronic kidney disease, and the well-known concerns with sensor data within the first 12 to 24 hours of wear. ¹⁹ Furthermore, it is important that anyone interpreting CGM data is aware that CGM sensors can display what users call “compression lows” (a sharp drop in glucose levels due to pressure, such as that exerted when someone rolls onto the sensor while sleeping).^24,25 Lack of familiarity with CGM sensors and their quirks may lead providers who are less familiar with CGM data to recommend referral for inappropriate follow-up with endocrinologists, who are already overburdened without enough providers to meet current demands. ²⁶

A major strength of our study comes from the unique aspects of our sample of CGM reports that included both extreme examples and typical examples from individuals without diabetes. But our true strength comes from the expertise of those 18 clinicians who completed the CGM interpretations. The expert clinicians we surveyed were mostly specialist diabetologists/endocrinologists who care for patients with more complex diabetes, and most only rarely treat individuals without diabetes. This is important context because some of the expert clinicians may have required a higher bar for recommending follow-up, while others may have erred on the side of caution due to the lack of data on long-term outcomes for individuals without diabetes. These differences in interpretation and recommendations were expected and discordance has even been observed when clinical CGM experts interpret CGM results from individuals with diabetes. ¹⁵ We observed remarkable differences of opinion among clinicians in terms of how to interpret elevated postprandial glucose levels (both brief and prolonged) or substantial time spent below range (<70 mg/dL), including what glucose levels may be considered “elevated” or “low.”

One limitation of our study is that our selection of a unique set of CGM reports introduces selection bias. This methodology was important in order to observe how experts would interpret different CGM features in this population, while keeping burden on our experts low. If we had provided a random sample of CGM reports from individuals without diabetes, there may have been more consensus on which CGM reports require follow-up, but only because there would inevitably be fewer reports with extreme features. Also of note, for the measurement of HbA1c and fasting glucose, blood was collected on the morning before CGM sensors were applied; therefore, the CGM data may not be truly representative of the glycemic burden that was demonstrated in HbA1c, which is more reflective of the glycemic burden during the months preceding measurement. Additionally, we only used CGM reports collected from a single type of CGM sensor worn by middle-aged and older adults without diabetes, so the results from our study can only be generalized to a specific sensor and population.

It is clear from even the few recent large studies of individuals without diabetes, each with sample sizes >400 and using different CGM sensors, that one may expect different CGM results by sensor and specific population characteristics. For example, ≥15% of individuals without diabetes from the Sugar Challenge Study (mean age ~35 years) and more than half of middle-aged to older adults in the Framingham Heart Study (~60 years old) exhibited maximum glucose values >180 mg/dL.^9,27 Furthermore, three other studies of different populations of individuals without diabetes or prediabetes (young and middle-aged athletes, young pregnant women, and older adults with mean age >80 years) exhibited ~2% to 5% time >140 mg/dL, compared to >10% time on average for participants without diabetes or prediabetes in the Framingham Heart Study.^9,28,29 The differences in CGM metrics reported among these studies of individuals without diabetes should be expected in different participant populations (by age, sex, body composition, diet and exercise, or by pregnancy status and other conditions) and with different sensors used.

Although there is a growing interest in time spent 70 to 140 mg/dL (time in tight range) as a target for individuals with diabetes, ¹² we focused our survey on recommendations experts may give based on the default layout of the AGP report at the time of the present study (April 2024), which displayed time in range 70 to 180 mg/dL. In the intervening time since our survey was conducted, new CGM sensors have become available without a prescription. Both of these new devices were launched using 70 to 140 mg/dL as the default goal in their smartphone applications for individuals without diabetes, while the downloadable AGP report currently still defaults to the 70 to 180 mg/dL range for all users. Although it may have been interesting to directly assess whether CGM experts would use time in tight range (70-140 mg/dL) as a metric to help guide their recommendations for follow-up, this was beyond the scope of our study.

More individuals without diabetes are becoming interested in tracking their glucose levels using CGM. Motivations for CGM wear in these individuals are diverse and include a desire to promote healthy behaviors and avoid prediabetes/diabetes. ¹ One study conducted in Spain reported that individuals without diabetes at baseline, but with higher CGM time spent >140 mg/dL, were more likely to develop type 2 diabetes over a five-year follow-up period. ¹¹ Studies with individuals at risk for developing type 1 diabetes (at stages 1 or 2) have also shown that certain CGM features predict progression to type 1 diabetes (stage 3).^30,31 Machine learning approaches may be useful in identifying CGM patterns or traits that predict diabetes development and progression once large studies with sufficient follow-up data are available. Although promising, current data do not provide enough information to support using CGM sensors as a screening tool for diagnosing diabetes over other clinical data; but it is likely that features observed with CGM could give health care providers reason to recommend their patients receive additional diabetes screening, especially when taken in a clinical context that may include other factors associated with diabetes risk. In the current study, we wished to observe clinical recommendations based purely on CGM (and prediabetes status) so we did not provide any clinical context, but we acknowledge that including medical history, lifestyle behaviors, and comorbidities will be a crucial part of future studies designed to set clinical standards for recommendation based on CGM.

Conclusions

Our results are important given the increased interest in using CGM among individuals without diabetes. Unfortunately, we observed poor overall agreement among expert clinicians in how to interpret CGM reports in this population. We must now await follow-up from CGM studies in individuals without diabetes to identify the CGM features that are representative of a higher risk of developing prediabetes or diabetes.