Diabetes Healthcare Professionals Use Multiple Continuous Glucose Monitoring Data Indicators to Assess Glucose Management

Abstract

Background:

Continuous glucose monitoring (CGM) offers multiple data features that can be leveraged to assess glucose management. However, how diabetes healthcare professionals (HCPs) actually assess CGM data and the extent to which they agree in assessing glycemic management are not well understood.

Methods:

We asked HCPs to assess ten de-identified CGM datasets (each spanning seven days) and rank order each day by relative glycemic management (from “best” to “worst”). We also asked HCPs to endorse features of CGM data that were important in making such assessments.

Results:

In the study, 57 HCPs (29 endocrinologists; 28 diabetes educators) participated. Hypoglycemia and glycemic variance were endorsed by nearly all HCPs to be important (91% and 88%, respectively). Time in range and daily lows and highs were endorsed more frequently by educators (all Ps < .05). On average, HCPs endorsed 3.7 of eight data features. Overall, HCPs demonstrated agreement in ranking days by relative glycemic control (Kendall’s W = .52, P < .001). Rankings were similar between endocrinologists and educators (R² = .90, Cohen’s kappa = .95, mean absolute error = .4 [all Ps < .05]; Mann-Whitney U = 41, P = .53).

Conclusions:

Consensus in the endorsement of certain data features and agreement in assessing glycemic management were observed. While some practice-specific differences in feature endorsement were found, no differences between educators and endocrinologists were observed in assessing glycemic management. Overall, HCPs tended to consider CGM data holistically, in alignment with published recommendations, and made converging assessments regardless of practice.

Introduction

In the decade since the first published randomized controlled trial demonstrated a clinically meaningful reduction in HbA1c levels using real-time continuous glucose monitoring (rtCGM),¹ the use of rtCGM has continued to gain popularity based on growing evidence from a number of studies (eg, T1D Exchange, JDRF-CGM, GLADIS, GOLD, DIAMOND, HypoDE) that show meaningful benefit for both adults and children who regularly use the technology.^2-9

Compared to traditional self-monitoring of blood glucose (SMBG), CGM provides an expanded range of derived data features that can have clinical utility beyond HbA1c measurements. While HbA1c is widely used as an indicator of recent (approximately three months) glycemic control, it is limited in its use to predict hypoglycemia risk¹⁰ or as a rapid indicator following treatment and/or behavior change.¹¹ This is particularly important in the context of calculating safe and effective doses of rapid-acting mealtime or correction insulin doses, or changing the dose, timing, or frequency of other diabetes therapies. Although SMBG addresses some of these gaps, physical and behavioral barriers (including those related to usability) can limit SMBG as a consistent and reliable tool for glycemic management.

Raw CGM data is dense, dynamic, and is nowadays collected in real time, allowing a number of data features to be derived to inform glucose management.¹² In addition to allowing calculation of unbiased average glucose levels (compared to SMBG, which often occurs with greater frequency around hypo- or hyperglycemic events) over a particular time duration, CGM data also provides detailed information about glucose variability,^13-15 change over time, and time in/out of a glucose target range (including excursions into lows and highs).^8,16 The continuous nature of CGM data also yields insights into upward or downward trends over short time periods, enabling more effective reactive and/or anticipatory behavior change.¹⁷

One of the benefits of using CGM is being able to retrospectively analyze the data and infer relative glycemic management. Such retrospective analyses can help HCPs and people with diabetes better understand how contextual factors might correspond to days of relatively better or worse glycemic management and inform treatment and behavior modifications.

Many recommendations have been made to guide diabetes healthcare professionals (HCPs) and people with diabetes on assessing retrospective CGM data and interpreting glycemic control.^12,15,18,19 However, CGM adoption is still nascent, and HCPs’ experiences using CGM may vary. Thus, an understanding of how HCPs actually interpret CGM data in clinical practice is still lacking. Moreover, given that CGM data can be evaluated among multiple, albeit nonmutually exclusive data features,²⁰ the extent to which HCPs agree on which data features are important for inferring glycemic management is also not well known.

The aim of this study is to gain a better understanding of how HCPs make use of the multitude of CGM data features when assessing glycemic management. Specifically, we wanted to survey HCPs to understand what aspects of CGM data are considered by them to be important to assessing glycemic management. We also hoped to observe empirically whether consensus exists in both the self-reported prioritization of CGM data features as well as inferring glycemic management from CGM data.

Methods

We presented de-identified real-life example datasets from ten CGM users to HCPs (ie, endocrinologists and diabetes educators) familiar with interpreting CGM as part of their regular clinical practice via an online survey (SurveyMonkey Inc; San Mateo, CA; www.surveymonkey.com). Clinician participants were recruited from a convenience sample, namely by word of mouth through the co-authors’ professional networks, by invitation to clinicians who visited Glooko’s product booth at a national meeting of endocrinologists, and by email solicitation to an organization of diabetes educators.

Each CGM dataset spanned seven continuous days and was randomly queried from the Glooko data warehouse (Glooko, Inc., Mountain View, CA; www.glooko.com). The datasets were presented in a graphical format (ie, the daily glucose profile component of the ambulatory glucose profile [AGP]) and were agnostic of each user’s diabetes type (though most likely to be from users with type 1 diabetes) and current therapies. The study was approved by the El Camino Hospital Institutional Review Board, and all HCPs provided informed consent and were compensated for their time.

The survey asked the HCPs to view each set of graphical CGM data traces and to rank order each day’s qualitative glycemic management from “best” to “worst”: “In the survey below, you’ll see ten questions that each present a graphical overview of one week’s worth of CGM data for one person with diabetes. We would like you to evaluate sample CGM data and rank order one week’s worth of data from best (1) to worst (7) in terms of glycemic control. Please use your best clinical judgment when making this inference.” In this task, daily CGM traces were overlaid on top of each other, and each participant ranked each day’s CGM trace based on relative glycemic control. For each CGM dataset (ie, each seven-days of CGM data), we quantified the extent of agreement among the HCPs. Agreement of rank judgments among HCPs was quantified using Kendall’s coefficient of concordance (W), a nonparametric statistic that assesses agreement among raters. We aggregated the average W for the group as a whole, as well as separately by practice (ie, endocrinologists and educators). We also assessed the extent to which endocrinologists and educators performed the rank judgments similarly (ie, between-group rank similarity) using R², Cohen’s kappa, and mean absolute error (MAE).

Following the ranking of days by relative glycemic management, the survey also asked the HCPs, “which were the most important factors determining your rankings above (select all that apply)?” and provided the following list: variance, average glucose, hypoglycemia, hyperglycemia, time in range, time out of range, high point of day, and low point of day. These data features were described conceptually; we did not define absolute thresholds for what was to be considered to be hypo- or hyperglycemia. The “in target range” was implicitly shown to be 70-180 mg/dL on the CGM graphs. These results were summarized for all HCPs as well as separately by practice.

Results

In the study, 57 HCPs (29 endocrinologists; 28 diabetes educators) participated. The HCPs had 16 years (median; interquartile range [IQR]: 6-26 years) of self-reported clinical practice experience and eight years (median; IQR: 3-10 years) of self-reported experience with CGM. The HCPs reported seeing 25 CGM-using patients per typical month in their clinical practice (median; IQR: 10-40). No differences were observed between endocrinologists and educators in any of these characteristics (Table 1). The geographic distribution of the HCPs is also shown in Table 1.

Table 1.

Study Sample Characteristics.

	All HCPs (n = 57)		Endocrinologists (n = 29)	Educators (n = 28)
Clinical experience (y)	Median	16	15	19
	IQR	6-26	4-26	13.5-26.25
CGM experience (y)	Median	8	5	8.5
	IQR	3-10	2-10	5-10.5
# of CGM patients (per mo)	Median	25	20	25
	IQR	10-40	10-40	10-41.25
Study completion time (min)	Median	30	22	36
	IQR	17-49	15-37	28-55
# of CGM data features endorsed	Mean	3.7	3.1	4.2
	IQR	3-5	2-4	3-5
CGM assessment concordance (Kendall’s W)	Median	.53	.54	.51
	IQR	.49-.56	.53-.56	.44-.59
Geographic distribution	West	40.4%	24.1%	57.1%
	Southwest	22.8%	17.2%	28.6%
	Midwest	15.8%	27.6%	3.6%
	Northeast	10.5%	13.8%	7.1%
	Southeast	8.8%	13.8%	3.6%

Study participants (HCPs) have a broad range of clinical practice and CGM experience.

On average, HCPs spent 30 minutes (median) to complete the study survey (ie, to review seven days of CGM graphs from ten people with diabetes; IQR: 17-49 minutes; Table 1). A difference was observed between endocrinologists and educators on time spent on the survey (median 22 vs 36 minutes, respectively; Wilcoxon W = 262, P = .022).

Time spent on survey was not correlated with years in practice, years of CGM experience, or number of monthly CGM patients. Years in practice was, however, correlated with years of CGM experience (Spearman’s ρ = .62, P < .001) but not with the number of monthly CGM users they reviewed (Spearman’s ρ = .08, P = .57). Years of CGM experience was correlated with number of monthly CGM users (Spearman’s ρ = .29, P = .027). Years in practice, CGM experience, and number of CGM patients seen monthly were not correlated with the endorsement of any of the CGM data factors evaluated.

Hypoglycemia and glucose variance were the most frequently endorsed priority features (91.2% and 87.7% of all participants, respectively). Hyperglycemia (43.8%), time in range (57.9%), time out of range (40.3%), and mean daily glucose levels (24.6%) were also frequently reported as key indicators for CGM assessment (Figure 1).

Figure 1.

Frequency of clinicians endorsing each CGM feature. Hypoglycemia (91.2%) and variance (87.7%) were nearly unanimously endorsed by all clinicians. Educators endorsed “time in range” (78.6% vs 37.9%), “low point of day” (25.0% vs 0%), and “high point of day” (21.4% vs 0%) more frequently than endocrinologists did.

There were differences between endocrinologists and diabetes educators in the endorsement of time in range (educators: 78.6%; endocrinologists: 37.9%; P = .0045), high glucose point (educators: 21.4%; endocrinologists: 0%; P = .027), and low glucose point of the day (educators: 25%; endocrinologists: 0%; P = .014).

As a group, clinicians prioritized a mean of 3.7 of the eight listed CGM data features for assessing glycemic management from CGM data. Educators endorsed more data features (mean = 4.2) than endocrinologists (mean = 3.1; P = .011). Overall, 21% of HCPs endorsed one to two different data features in ranking daily glucose management indicators, 51% endorsed three to four data features, and 28% endorsed five to eight data features (Figure 2).

Figure 2.

Number of CGM features endorsed by clinicians. Collectively, most HCPs endorsed three to four data features as important. As seen in the cumulative data, educators tended to endorse more data features than endocrinologists: 57.1% of educators (vs 34.5% of endocrinologists) endorsed four or more data features, and 92.9% of educators (vs 65.5% of endocrinologists) endorsed three or more data features.

As a group, the HCPs demonstrated agreement in how they ranked daily glycemic management from CGM data (mean Kendall’s W = .52, P < .001). Diabetes educators and endocrinologists had comparable concordances in rank judgments (mean Kendall’s W = .51, .54, respectively; Wilcoxon W = 59, P = .53). Educators and endocrinologists also demonstrated between-group similarity in rank judgments (R² = .90, Cohen’s kappa = .95, MAE = .4 [all Ps < .05]; Mann-Whitney U = 41, P = .53).

Discussion

HCPs comprising endocrinologists and diabetes educators prioritized similar aspects of the CGM data overall when assessing glycemic management despite having multiple, albeit not mutually independent, data features to guide interpretation. The HCPs also showed moderate agreement in how they interpreted real-world CGM graphs to assess glycemic management.

Although we observed a degree of interrater variability in the use of multitude of CGM-derived data features,²¹ hypoglycemic excursions and glucose variance were nearly unanimously endorsed as important indicators by the HCPs. This is an encouraging sign of clinical consensus in what can be a potentially subjective exercise.

Apart from hypoglycemic excursions and glucose variance, time in range, hyperglycemia excursions, and time out of range were also frequently endorsed as priority data features by both endocrinologists and educators. While these and other data features can be informative for assessing glycemic management, they are likely considered on a context-dependent manner based on patient status, treatment goals, and therapies. Given the multitude of CGM data features available, such as those investigated in this study as well as those that are part of the standard AGP,^22,23 more work needs to be done to validate these different data features. It is noteworthy that significant progress has been made recently with regard to time in range with its recent validation as an endpoint for diabetes clinical trials^24,25 and guidelines for treatment targets.²⁶ Still, as more CGM data features become validated, it remains to be seen how these data features will be used in clinical practice and research, and standardized into electronic medical records.

Nearly 80% of surveyed HCPs endorsed using three or more CGM data features when assessing glucose management. This suggests that HCPs were inclined to leverage the richness of CGM data features to arrive at a more comprehensive and holistic understanding of glycemic management. Having access to the range of CGM data features can also be beneficial for adjusting insulin or medications and informing self-management.

Some practice-specific differences were observed in this study. Diabetes educators prioritized more CGM data features than did endocrinologists. Educators also spent more time reviewing the presented CGM data than did the endocrinologists. In particular, we found that a greater proportion of educators endorsed time in range and the high and low points of the day than did endocrinologists. Taken together, these observations might reflect qualitative differences between the goals and approaches of the two groups. For example, endocrinologists might be more inclined to focus on identifying relevant indicators for titrating insulin and other medications while educators might be more interested in using the data to facilitate behavior modification recommendations. There may also be issues related to time available in a clinical encounter and the demands of information beyond glucose and for other medical conditions in routine clinical endocrine practice. One limitation is that the practice-specific differences we observed could also be due to biases within our study sample. While we attempted to recruit a broad range of participants, the resulting samples, when comparing between endocrinologists and diabetes educators, may not have been well balanced by geography, years of clinical experience, practice/training background, etc. Further research in larger samples is needed to determine whether differences observed in this study, which may be practice- and context-dependent, correspond to substantive between-group differences in the self-management advice given and their effectiveness.

Time spent on survey completion was not correlated with any clinical experience-related indicators. This could suggest that a performance floor/ceiling exists in how long HCPs take to evaluate CGM data. However, one limitation of the study was that the task did not impose a constrained time limit, and we observed a wide range in survey completion times. Some HCPs might have completed the survey intermittently and not during one continuous sitting. Thus, survey completion time might not be an appropriate indicator of performance or expertise in the current study, and additional research is warranted to reach more definitive conclusions.

Positive correlations between self-reported clinical practice experience and CGM experience, and between CGM experience and the number of CGM users seen in a month support the notion that HCPs with more overall experience in practice also had more familiarity with CGM. Subsequently, those with greater CGM experience might also be more likely to prescribe CGM to patients.

Although our findings provide an indication of how clinicians interpret CGM data to assess glycemic control, some limitations and challenges should be considered. First, we acknowledge that the survey task was not necessarily representative of how CGM data is used in clinical practice. The forced-choice nature of the survey was designed to collect consistent and quantifiable responses about how HCPs evaluate retrospective CGM data to assess glycemic control, whereas in practice CGM data can be evaluated in a number of qualitatively different ways to inform medication adjustments and lifestyle recommendations, depending on treatment goals and patient profiles. For example, CGM data review is often used not only for creating an integrated metric to describe the quality of glycemic control, but also to inform insulin dose and timing adjustments, modify alert settings, and demonstrate the effects of insulin and other lifestyle factors on glycemic control. Nonetheless, our current findings provide an initial empirical survey of how HCPs currently prioritize different features of CGM data, and may be informative for HCPs who are new to or contemplating prescribing CGM.

While CGM yields a multitude of potentially informative data features that have been associated with demonstrated glycemic improvements, several barriers (eg, prescriptions limited to people with type 1 diabetes and/or on insulin, lack of reimbursement coverage by health insurance companies) currently challenge the widespread adoption of CGM (and its data) as the standard basis for assessing and supporting glycemic management.^16,27,28 However, trends in pricing and reimbursement policies and the scope of CGM prescription continue to expand the use of CGM.^9,29-32

As CGM becomes more widely adopted, one potential challenge to overcome might be that HCPs need to be able to develop familiarity and expertise in its use and data interpretation.³³ The relatively greater complexity associated with using CGM (eg, insertion/implantation of sensors, maintaining the site, calibration) and interpreting its data may also be factors limiting CGM use in the clinic.¹⁶ Continued clinical training and education coordinated by professional organizations as well as industry partners will play an important role in providing clinicians with the knowledge and training necessary to fully utilize emerging tools such as CGM (and, eg, the rich glucometric data that are now a part of the standard AGP reports). In addition, the role of digital health companies will also be important to help bridge technological knowledge and workflow efficiency gaps by providing digital and computational tools that facilitate the ingestion, integration (ie, across multiple devices), display, interpretation, and analyses of data. Such digital solutions have the potential to help clinicians maximize CGM’s potential in decision support by minimizing the burden and complexity of CGM data interpretation when patient interaction time during clinic visits can be limited.

Embracing and leveraging the multidimensionality of CGM data can lead to a more comprehensive understanding of glycemic management, as well as to the use of more informative clinical outcomes metrics to support diabetes care. Understanding how to best leverage the richness of CGM data can also lead to more precise insulin and medication adjustment and targeted behavioral change for self-management.^1,2 While several factors currently hamper the prescription, adoption, and utilization of CGM, many of these barriers can be addressed by improving, simplifying, and automating the analyses and interpretation of dense and high-dimensional CGM data. This is an area where digital health and computational tools can be particularly beneficial to support decisions in the clinic now and in the future.³⁴