FormalPara Key Summary Points

Why carry out this study?

People with type 1 diabetes (T1D) or type 2 diabetes treated with intensive insulin therapy (T2D-IIT) increasingly use continuous glucose monitoring (CGM) systems to monitor their glucose levels and manage their diabetes.

There are two types of CGM systems—real-time CGM (rtCGM), which provides glucose values automatically to a compatible device, and intermittently scanned CGM (isCGM), which requires volitional scanning of the CGM to obtain glucose values.

This study used medical and pharmacy payor claims and linked laboratory data to analyze adherence patterns between these two CGM systems and examined the association between CGM adherence (of any type) and changes in glycated hemoglobin (A1C).

What was learned from the study?

Adherence was higher among rtCGM users and, regardless of diabetes type, the odds of adherence were over twice as high for rtCGM users compared to isCGM users.

Adherent CGM users experienced significantly greater A1C reductions compared to non-adherent CGM users.

CGM adherence could help users with T1D and T2D-IIT lower their A1C and because rtCGM users demonstrated greater adherence, this may increase their likelihood of beneficial outcomes.

Introduction

Continuous glucose monitoring (CGM) systems are increasingly used by people with type 1 diabetes (T1D) or type 2 diabetes using intensive insulin therapy (T2D-IIT). There are two types of CGM systems—real-time (rtCGM) and intermittently scanned (isCGM), which differ in their method of data retrieval and availability [1]. Accuracy of the latest CGM systems is quite high and some are nonadjunctive, meaning they can be used as the basis for diabetes management decisions without the use of blood glucose testing [1]. Select CGM systems can also integrate with automated insulin delivery systems such as insulin pumps or connected insulin pens [1, 2].

Use of CGM systems is associated with beneficial glycemic outcomes among people with T1D or T2D-IIT [3,4,5]. These glycemic outcomes are achieved through behavior change on behalf of the person with diabetes and careful adjustment and monitoring of their treatment regimen in consultation with a healthcare provider [6].

Adherence to diabetes medication is associated with improved glycemic control, lower healthcare utilization, reduced medical costs, and lower mortality rates [7, 8]. However, little is known about adherence to diabetes devices, such as CGM systems, and whether adherence differs between CGM types and, finally, whether device adherence is associated with glycemic improvements, such as decreased glycated hemoglobin (A1C). In this study, we queried a payor claims database to investigate adherence to CGM and analyzed whether it was associated with beneficial A1C outcomes.

Methods

Data Source

This was a retrospective, observational, US database study using de-identified administrative health claims and linked laboratory data from the Merative™ MarketScan® Research Database. The MarketScan® database is statistically de-identified and HIPAA (Health Insurance Portability and Accountability Act)-compliant and therefore the study was deemed exempt from an ethics committee review. The authors licensed the use of the database from MerativeTM. This database includes claims data for inpatient, outpatient, and outpatient prescription drug utilization of employees and their spouses and dependents or individuals with supplemental Medicare or Medicare Advantage insurance. Only data from commercially insured beneficiaries were included in the present study.

Cohort Definition

The population included adults ≥ 18 years of age with T1D or ≥ 30 years of age with T2D-IIT identified using International Classification of Diseases, 9th/10th Revisions (ICD-9/ICD-10), who were enrolled in commercial health plans (non-Medicare) and had no prior evidence of CGM use. The identification period spanned August 1, 2019 to March 31, 2021. CGM sensor initiation during this identification period was required and constituted the index date. Continuous health plan coverage 12 months pre-index (baseline) and 12 months post-index (follow-up period) was also required.

People with T2D were required to have evidence of bolus insulin (indicating intensive insulin therapy) in the baseline period. People with T1D were required to have basal and bolus or mixed insulin use at baseline. To mitigate cohort overlap by diabetes type, exclusion criteria specific to the T1D cohort included the use of basal insulin alone, sulfonylureas, thiazolidinediones (TZD), or combination drugs containing sulfonylureas or TZD during the study period. People with evidence of pregnancy or gestational diabetes were excluded.

Cohorts were segmented by CGM type using National Drug Codes (NDC) (Dexcom G5/G4 or G6 rtCGM system sensor components [8627005104 or 8627005303] or Abbott Freestyle Libre 14-day or Libre 2 isCGM system sensor components [57599000101 and 57599080000]). Individuals who had claims for a Dexcom device and an Abbott device during the identification period were excluded, making the cohorts mutually exclusive. For the subset of people included in the analysis of change in A1C, valid A1C values within 6 months before and after the index date were required.

Assessment of CGM Adherence

Adherence to CGM was calculated using the proportion of days covered (PDC). PDC is the Pharmacy Quality Alliance’s preferred metric for estimating adherence to therapies for chronic diseases such as diabetes [9]. PDC is calculated as the proportion of days in the measurement interval during which a person had access to the medication or device. Specifically, PDC is the sum of days covered by each prescription divided by the number of days from the first treated day to the end of the time period of interest. The maximum value for PDC is 1.0 and early fills or differences in CGM sensor lifetime do not impact PDC. In the full cohort, PDC was determined during the 12-month follow-up period through analysis of NDC codes related to sensor refills. PDC ≥ 0.8 was defined as adherent.

Assessment of A1C Change and Proportion Reaching Target A1C

A1C values were available for 559 out of 7669 (7.3%) of the full cohort. For inclusion in the analysis, at least one laboratory A1C value within 6 months pre- and post-index was required. Change in A1C was calculated as the mean A1C post-index minus the mean A1C pre-index during the 6-month period, with negative values indicating improvement in A1C.

In addition to assessing A1C change, the proportion reaching recommended A1C targets was also assessed. The American Diabetes Association (ADA) Standards of Care recommends an A1C goal of < 7.0% for non-pregnant adults without significant hypoglycemia [10]. The National Committee for Quality Assurance collects Healthcare Effectiveness Data and Information Set (HEDIS) survey results directly from health plans and preferred provider organizations. HEDIS collects data on multiple diabetes-related health metrics including A1C levels, with A1C control defined as < 8.0% [11, 12]. The proportion reaching these targets of < 7.0% and < 8.0% at baseline and follow-up was quantified.

Statistical Analyses

Descriptive statistics, including percentages, means, and standard deviations (SDs), were calculated for demographic characteristics and study outcomes and were presented by diabetes type (T1D or T2D-IIT) or CGM type (rtCGM or isCGM). Bivariate differences in adherence between CGM type were tested with independent sample t tests for continuous measures or Χ2 tests for categorical measures. A logit model was used to regress the dependent variable (post-index CGM adherence) on the independent variable (type of CGM) and covariates (gender, Charlson comorbidity score, insulin pump use during follow-up, and mean number of oral diabetic medications used at baseline). CGM adherence was dichotomized as PDC < 0.8 (non-adherent) or PDC ≥ 0.8 (adherent). Change in A1C between non-adherent and adherent CGM users irrespective of device type was assessed using difference-in-differences (DiD) tests. The DiD estimate indicates the magnitude and direction of change in outcome (A1C change) between the two groups. Pearson correlations assessed the association between CGM adherence and A1C change. Analyses were performed using Instant Health Data software (Panalgo, Boston, MA) and R (R Foundation for Statistical Computing, Vienna, Austria), version 3.2.1.

Results

A total of 7669 commercially insured beneficiaries with T1D or T2D treated with intensive insulin therapy (T2D-IIT) were identified. The majority were male with an average age in the low 50s (range 30–66 years) for the T2D-IIT cohort and low 40s (range 18–66 years) for the T1D cohort (Table 1).

Table 1 Demographic characteristics of the T1D and T2D-IIT cohorts
Full size table

CGM Adherence Patterns

To determine adherence rates among users of CGM systems—rtCGM or isCGM—we analyzed the PDC. After analyzing NDC codes for sensor refills over 12 months among users with T1D, the PDC was 0.72 (0.31) (mean (SD)) for rtCGM and 0.55 (0.34) for isCGM (p < 0.0001). This was comparable among users with T2D-IIT, where the PDC was 0.71 (0.30) for rtCGM sensor users and 0.56 (0.34) for isCGM sensor users (p < 0.0001).

The proportion of adherent users to rtCGM was 59.7% for users with T1D and 56.8% for users with T2D-IIT (Fig. 1a). In contrast, among isCGM users with T1D or T2D-IIT, the proportion of adherent users was 36.3% or 37.6%, respectively (Fig. 1b). Overall, regardless of diabetes type, the odds of adherence were over twice as high for rtCGM users compared to isCGM users (T1D odds ratio 2.8 (95% CI 2.4–3.3), p < 0.0001; T2D-IIT odds ratio 2.2 (95% CI 1.9–2.5), p < 0.0001). In addition, the proportion of users with very low adherence (PDC < 0.25) was approximately twice as high for isCGM users compared to rtCGM users (Fig. 1).

Fig. 1
figure 1

Adherence by CGM type. Adherence based on proportion of days covered (PDC) assessed over 12 months through National Drug Codes for A real-time continuous glucose monitoring (rtCGM) or B intermittently scanned continuous glucose monitoring (isCGM)

Full size image

Adherence and A1C

The association between CGM adherence and change in A1C was assessed by testing the DiD between non-adherent and adherent CGM users, regardless of type of CGM system (i.e., rtCGM or isCGM) (Table 2). Adherent CGM users achieved significantly greater A1C improvements compared to non-adherent CGM users.

Table 2 Relationship between CGM adherence and change in A1C
Full size table

Irrespective of diabetes type, adherent CGM users were more likely to reach target A1C. Among adherent users with T1D or T2D-IIT, 69.8% and 66.1% met the HEDIS target of < 8.0% at follow-up, respectively. Among non-adherent users with T1D or T2D-IIT, only 45.7% and 45.5%, respectively, met the same target (Table 3). Similarly, 33.6% and 28.6% of adherent CGM users with T1D or T2D-IIT met the ADA target of < 7.0% at follow-up, compared to 23.4% and 20.2% of non-adherent CGM users with T1D or T2D-IIT. In addition, in all cases, adherent users experienced a greater increase in the proportion achieving the A1C target, despite higher baseline proportions already meeting the target value (Table 3 and Fig. 2). Greater CGM adherence was associated with greater decreases in A1C at follow-up among users with T1D (r = − 0.16, CI [− 0.03, − 0.29], p = 0.02) or T2D-IIT (r = − 0.14, CI [− 0.03, − 0.24], p = 0.01).

Table 3 Proportion meeting ADA and HEDIS A1C targets
Full size table
Fig. 2
figure 2

Proportion meeting A1C targets. Cumulative distribution plots of A1C at baseline and follow-up among A non-adherent users with type 1 diabetes (T1D), B adherent users with T1D, C non-adherent users with type 2 diabetes treated with intensive insulin therapy (T2D-IIT), and D adherent users with T2D-IIT

Full size image

Discussion

This study indicates that adherence based on PDC was significantly higher among rtCGM users compared to isCGM users in both T1D and T2D-IIT populations. In addition, greater adherence to CGM was associated with significant and clinically meaningful improvements in A1C [13]. This supports the ADA’s Standards of Care in Diabetes which recommends CGM for adults with diabetes using intensive insulin therapy to manage their diabetes [14]. Users’ higher adherence to rtCGM is relevant to the recommendation that CGM devices should be used as close to daily as possible for maximal benefit [14].

Measurement of adherence is critical to understanding treatment efficacy in the real world, but adherence metrics for durable medical devices, such as insulin pumps, can be poor indicators of their effectiveness or efficiency. CGM systems, however, do lend themselves to pharmacy refill analyses similar to pharmaceutical medications owing to their generally monthly refill schedule. Adherence to diabetes therapies is important to assess because A1C could remain high because of low treatment efficacy, low treatment adherence, or a combination of both [15]. Our analysis found an association between higher adherence to CGM systems and improved glycemic outcomes. This suggests that CGM systems can contribute to favorable and clinically meaningful outcomes, but adherence may be important to optimize glycemic control.

Our analysis found that the odds of adherence were over two times greater among users of rtCGM systems than for users of isCGM systems. This higher adherence to rtCGM could be due to feature optionality on rtCGM systems, such as a predictive hypoglycemia alert and real-time data availability without the need to physically scan the sensor. Individual motivations for using CGM could also play a role in adherence with potentially higher adherence among rtCGM users interested in preventing severe hypoglycemic or hyperglycemic events. In addition, differences in insurance coverage or out-of-pocket costs could also contribute to adherence patterns. About twice as many isCGM users had very low adherence (PDC < 0.25), compared to rtCGM, indicating higher rates of sporadic or short-term use.

This lower adherence among isCGM users compared to rtCGM users is notable because CGM use is associated with reductions in A1C and a lower risk of acute and long-term diabetes complications [3, 4, 16,17,18,19]. It is unknown whether people using insulin who stopped using CGM returned to self-monitoring of blood glucose (SMBG) regimens or administered insulin with little to no information about their glucose levels.

Our results found a greater decrease in A1C among people with diabetes who were more adherent to CGM. Beyond the individual benefits of lowering A1C [16, 19,20,21,22], improved population-level outcomes are important to healthcare payors. A recent analysis of the cost-effectiveness of rtCGM vs. isCGM among people with T2D using multiple daily injections of insulin (a subset of T2D-IIT) from a US payor perspective found rtCGM to be cost-saving over isCGM [23]. The larger A1C decreases, on average, among adherent CGM users may further contribute to increased cost-effectiveness in this population. However, it is also important to note that while about two-thirds of those adherent to CGM met the HEDIS target of A1C < 8.0%, only about one-third met the ADA target of A1C < 7.0%. Helping more people with diabetes who use insulin reach the ADA recommendation may require increasing availability of diabetes education programs and use of newer technologies such as hybrid closed-loop systems.

Limitations of this study include the all-US cohort, the lack of racial and ethnic data, and the relatively low proportion of the cohort with A1C data. The findings may not be generalizable beyond people with commercial insurance coverage in the USA. An additional limitation is that this is a retrospective data analysis. Medication changes over the 6-month observation window could have impacted the observed A1C results. Adherence was assessed through calculation of PDC, which is an indirect estimate of adherence obtained from pharmacy refill data. Actual utilization could differ from observed utilization and the degree to which people used the CGM devices to monitor their glucose levels and change their behavior cannot be assessed. Our analysis removed people with a pharmacy claim for more than one type of CGM system and thus we did not assess patterns of switching between systems. The analysis did not include extensive assessment of the use of insulin pumps and hybrid closed-loop systems beyond statistical adjustment for pump use in our multivariate model. Additional research is needed to assess CGM adherence in relation to other diabetes technologies. As new CGM devices, with new features and functionality, enter the market, their adherence will have to be assessed and the current findings may not generalize to these devices.

Conclusions

Our findings indicate that for people with T1D or T2D-IIT, higher adherence to CGM could help them lower their A1C. RtCGM users were also significantly more likely to be adherent to CGM, potentially increasing the likelihood of beneficial outcomes. Because of CGM’s association with A1C improvements and fewer acute and long-term complications, it is expected that high CGM adherence could further improve outcomes and lower individual and payor healthcare costs.