FormalPara Key Summary Points

We have developed novel measures of FSL adherence and experience, and investigated how these are associated with longitudinal changes in metrics of glucose control.

Our data indicate that, as their experience grows, FSL users in the Czech Republic became more adherent in sensor reapplication.

Increased adherence is accompanied by increases in %TIR, and reductions in %TAR and glucose variability.

It is evident that low initial adherence can improve with continued use of FSL and is therefore not a strong predictor of poor glycemic outcomes.

Users with low initial adherence gradually learned more about the value of their CGM metrics over time and successfully incorporated that information into their diabetes management routine.

Introduction

Continuous glucose monitoring (CGM) systems are demonstrated to be more effective at reducing HbA1c and frequency of hypoglycemia when compared to self-monitored blood glucose (SMBG) fingerprick testing [1,2,3,4], for people with type 1 diabetes (T1D) or type 2 diabetes (T2D). Flash glucose monitoring (FLASH), using the FreeStyle Libre™ system (Abbott Diabetes Care, Witney, UK) allows users to view their current and stored glucose readings by simply scanning the FreeStyle Libre sensor using a reader or smartphone app. Importantly, the FLASH system is factory calibrated [5, 6] and does not require users to perform twice-daily SMBG tests to ensure accuracy, unlike many traditional CGM systems. Once applied, each FreeStyle Libre sensor has an active wear time of 14 days during which glucose data is collected.

The 2019 International Consensus on Time in Range [7] provides recommendations that define good daily glycemic management for subjects with diabetes, using the glucose data generated by FLASH or traditional CGM systems [8]. These include the amount of time that glucose readings are within, above, or below a defined target glucose range. These metrics can also be used in day-to-day clinical practice to understand glucose variability and risk of hypoglycemia [9]. For people with T1D or T2D, these metrics are time in range (TIR; glucose 70–180 mg/dL), time below range (TBR; glucose < 70 mg/dL or < 54 mg/dL), time above range (TAR; glucose > 180 mg/dL or > 250 mg/dL) (see Supplementary Table 1 for more details).

It has been demonstrated for users of FLASH monitoring that the daily rate of sensor scanning is correlated with improved glycemic metrics as defined by the above international consensus recommendations [10,11,12]. These daily scan rates are therefore used to measure their level of engagement with glucose monitoring and diabetes management. Thus, greater engagement is associated with improvements for TIR, TBR < 70 mg/dL, TBR < 54 mg/dL, TAR > 180 mg/dL, and TAR > 250 mg/dL. These studies have used the large, anonymized data sets compiled from FreeStyle Libre sensor use, and include the data from more than 11 million FLASH sensors, with more than 13.5 billion individual glucose readings [11]. What is not clear from these reports is the clinical relevance of a user’s experience level with FLASH, the contribution of other behaviors to glucose control, and how they might relate.

One important engagement-related diabetes management behavior is the speed or discipline with which a user replaces each sensor at the end of its 14-day wear life (i.e., their adherence to sensor wear), which we assume is driven by a desire to maintain continuity of glucose monitoring. Frequent scanning is an important part of glucose management whilst wearing an active sensor, but adherence to sensor reapplication is also critical for maintaining continuity of glucose monitoring. By measuring this adherence, we may better understand another key aspect of engagement and discipline towards glucose management that is not captured by scan frequency. Since information provided by FLASH monitoring enables education and behavior modification, a user’s experience with the system is also a likely factor in their glucose control. As users become more experienced with the system, they may learn more about how to manage their diabetes by tracking the metrics uniquely provided by CGM, and may also grow more accustomed to sensor wear. It is therefore important to understand the clinical and behavioral relevance of experience with FLASH monitoring.

In the current study, we aimed to develop novel measures of adherence and experience with FLASH monitoring for users of the FreeStyle Libre system in the Czech Republic and to understand how these are associated with longitudinal changes in metrics of glucose control.

Methods

Sensors and Readers

FLASH glucose monitoring using the FreeStyle Libre system measures interstitial fluid glucose levels using a factory-calibrated sensor with a wear time of up to 14 days. Data is transmitted to a dedicated reader or a smartphone app each time that they are used to scan the glucose sensor. Each scan reveals glucose readings that have been automatically saved by the glucose sensor at 15-min intervals, for the previous 8 h. Data collected by FLASH readers were included in this analysis. Only the first version of FreeStyle Libre has been available in the Czech Republic from 2018 to date. This analysis therefore does not include FreeStyle Libre 2 readers. The sensor data gathered by the reader was anonymized and uploaded to a database by the FreeStyle Libre software, which includes an agreement from each subject covering data use by the software. The FreeStyle Libre desktop software is used to generate an Ambulatory Glucose Report (AGP), which summarizes the user’s sensor data, including details about consensus metrics [9]. Amongst 16,079 FreeStyle Libre users in the Czech Republic, anonymous data were extracted for 1600 users who had at least 36 completed sensors from October 22, 2018 to December 31, 2021. Reimbursement for people with T1D was mandated in the Czech Republic beginning in October 2018. As a result of the potential effect that users’ reimbursement status has on their sensor adherence, only users whose first sensor begins after the reimbursement mandate were selected for analysis.

User Experience and Adherence with FLASH

User experience with FLASH was defined by the number of completed sensors (from 1 to 36 sensors). Adherence with FLASH was defined by the percentage of data capture and the sensor gap time, i.e., the time between the end of a given sensor and the start of the next sensor. To capture data over a period that ensures persistent behavior, user adherence was analyzed in four experience levels after initiating FLASH; Start (sensors 1–3); Early (sensors 4–6); Middle (sensors 19–21); End (sensors 34–36). Within each experience level, each user’s mean percent data capture and mean sensor gap time were calculated. Users were split into two adherence levels based on mean gap time during the start period, “low” adherence (> 24 h mean gap time, n = 723 users) and “high” adherence (≤ 8 h mean gap time, n = 877 users).

Glucose Control Measures

For a given sensor, analysis of glucose control required at least 120 h of sensor wear duration to ensure reliable interpretation of its glycemic measures. For each user and user-experience segment (start, early, middle, end), data from all such sensors were combined to calculate each user’s overall glucose control metrics at each experience level. Glucose control metrics assessed included average glucose; %TIR 70–180 mg/dL; %TBR < 70 mg/dL; %TBR < 54 mg/dL; %TAR > 180 mg/dL; %TAR > 250 mg/dL; glucose coefficient of variation (CV); average scan frequency (daily scans). These metrics were aggregated amongst the users allocated to the “low” or “high” adherence groups, and the groups’ aggregate glucose control measures were tracked across the four experience levels.

Statistical Analyses

Longitudinal change was measured by the mean of users’ differences from start to end. Because the cohort selection criteria was determined by users’ adherence during the start period, longitudinal changes in adherence metrics (% data capture, mean gap time) were evaluated from early to end (in order to avoid potential regression-to-mean effects). Cross-sectional measurements are described using the median because of the skew in their distributions. p values for mean change in mean % data capture, mean gap time, and glucose metrics were derived from paired t tests. p value for % users starting next sensor on average within 24 h were derived from two-proportion z tests.

Compliance with Ethics Guidelines

This study has used only retrospective and deidentified data, and therefore ethics committee approval was not required. Deidentified data was used only from individuals who consented by opting in on the LibreView platform to their data being used for real-world studies. Permission was granted by Abbott for use of the deidentified data and all research was conducted in accordance with the 1964 Declaration of Helsinki.

Results

Longitudinal Change in Sensor Adherence for Users with Low Initial Adherence

The low-adherence group consisted of users with a mean sensor gap time > 24 h during their start period (sensors 1–3). The median of low-adherence users’ mean gap times for this period was 118.5 h (almost 5 days). After the start period, the low-adherence group reduced their sensor gap times significantly (Fig. 1). Median gap time was 51.3 h during the early period (sensors 4–6), 13.0 h in the middle period (sensors 19–21), and 10.3 h by the end period (sensors 34–36). From the early to end periods, low-adherence users showed an average decrease in mean gap time by 117.5 h (p < 0.001).

Fig. 1
figure 1

Longitudinal change in sensor adherence and sensor data capture for adult FreeStyle Libre users in the Czech Republic. Summary statistics of each group’s adherence for the “Start” period (sensors 1–3), the “Early” period (sensors 4–6), the “Middle” period (sensors 19–21), and the “End” period (sensors 34–36). Users’ mean gap time (time between the end of one sensor and activation of the next) during the Start period was used to represent their initial adherence, and was therefore used as selection criteria for the analyzed groups. For each period, each group’s adherence is described by % of users with a mean gap time ≤ 24 h (% who on average started the next sensor within 24 h of the end of the previous sensor), and by the median of users’ mean gap times. Median % data capture for each group in each period is also shown. Results are shown for all users in each of the low-adherence and high-adherence groups. For the low-adherence group, greater adherence is accompanied by greater TIR and lower TAR (Table 1) at successive experience segments, demonstrating the association between adherence and glucose control. *p < 0.001 for change from Early to End period. As a result of the use of mean gap time during the start period as a selection criterion for the groups analyzed, the evaluation of longitudinal changes in adherence metrics did not include the start period, and instead spanned from the early to the end period. p values for the difference in % users starting next sensor on average within 24 h were derived from two-proportion z tests

Full size image

By the early period, 38.5% of the low-adherence users applied a new sensor within 24 h on average, and this proportion rose to 58.0% during the middle period and to 65.0% by the end period. The 26.6% increase in this proportion from the early to end periods (p < 0.001) indicates users’ increased likelihood to be prompt or disciplined with sensor reapplication at greater levels of experience with FLASH. The high-adherence group (≤ 8 h sensor gap time) had a median sensor gap time of 2.8 h during the start period, which was maintained at the end period (3.4 h).

Longitudinal Change in Daily Scan Rates and Data Capture

Daily scan rates for both low- and high-adherence users were consistent across the sensor use periods. The median number of daily scans for low-adherence users was 12.1 scans/day in both the start and end periods. For high-adherence users, median daily scans were 13.6 scans/day in the start period and 13.1 scans/day at the end period (Table 1).

Table 1 Longitudinal change in glucose control metrics for low- and high-adherence groups with increasing experience with sensor use
Full size table

The median % data capture was consistently high (> 96%) for both groups at all experience levels (Fig. 1), indicating that, once a sensor was applied, the users in both the high- and low-adherence groups were regularly scanning at least once every 8 h.

Longitudinal Change in Glucose Metrics for Users with Low Initial Adherence

Increased adherence (reduced sensor gap time) for the low-adherence group was associated with increased %TIR (median 55.0% or 13.2 h/day at start and median 58.0% or 13.9 h/day at end, corresponding to a mean increase of 2.4% or 33.8 min/day; p < 0.001), and reduced %TAR 250 mg/dL or 13.9 mmol/L (median 11.8% or 2.8 h/day at start and median 8.5% or 2.0 h/day at end, corresponding to a mean decrease of 3.1% or 44.7 min/day; Table 1; p < 0.001). Average glucose (median 163.7 mg/dL or 9.1 mmol/L at start and median 160.1 mg/dL or 8.9 mmol/L at end) also decreased by a mean of 5.5 mg/dL or 0.3 mmol/L (p < 0.001), and CV (median 40.9% at start and median 39.2% at end) decreased by a mean of 1.7% (p < 0.001), from start to end. %TBR remained stable across the four experience levels (Table 1). A comparison of the mean and median changes in glycemic control between low-adherence users and high-adherence users supports the observation of relative improvement with increased longitudinal adherence amongst users with low initial adherence during the Start period (see Supplementary Tables 1–3).

Discussion

This real-world observational study provides a novel longitudinal analysis of data collected from October 22, 2018 to December 31, 2021 by FreeStyle Libre users in the Czech Republic and investigated change in adherence to sensor wear from the first sensor through to the 36th sensor. We show that, as users acquire more experience with FLASH monitoring, they can become more adherent to prompt application of a new sensor after the end of the previous sensor. This increase in adherence is accompanied by increased %TIR 70–180 mg/dL, reduced %TAR, and reduced glycemic variability. Although low-adherence users initially exhibited extended periods between sensor replacements, compared to high-adherence users (Fig. 1), they demonstrated high engagement with the sensors after they were applied and activated (Table 1). This observation may be due in part to a selection bias: the selection criteria required that users completed 36 sensors with sufficient sensor data in the selected periods. This increases the likelihood (for both groups) of including users with greater overall engagement with the product, who are better able to use the system. The requirement for data sufficiency in the start and early periods, as well as high initial scan frequency, also suggests that low initial adherence in the selected sample was less likely to be due to difficulty operating the device. Median data capture was high for both groups (> 96%), well above the 70% target recommended by international consensus guidelines [7, 9], with more than 11 scans/day from start to end periods of sensor use (Table 1).

An important conclusion from our study is that initial low adherence with sensor use is not a strong predictor of glycemic outcomes nor of eventual adherence. As users with low initial adherence gained experienced using the FreeStyle Libre system, they demonstrated improved adherence, which was associated with improved glycemic metrics for %TIR, %TAR, and %CV, when measured from the first to 36th sensor. Their mean longitudinal increase in TIR matches their mean decrease in TAR > 180 mg/dL (2.4%), which suggests that management of hyperglycemia was a factor in the relationship between adherence, experience, and glucose control. Their longitudinal increase in adherence is evidence of behavior change, which in turn suggests the possibility that behavior modification was a driver in the reduction of hyperglycemia.

It can be speculated that, as their experience grows, users learn more about the insights that time-in-range metrics provide, increasing the value they place on CGM data and motivating them to replace each sensor. As discussed, difficulty with system operation was unlikely to be a factor in initial low adherence for the analyzed cohort (they were able to capture sufficient data with the sensor and use the desktop software to view their AGP). Thus, the main result of increased adherence in the analyzed sample is likely greater data continuity, and consequently more-accurate and precise information about time-in-ranges. As such, the results appear consistent with the idea that users with low initial adherence gradually learned more about CGM metrics, found value in the data, and successfully incorporated that information into their diabetes management routine.

It is important to note that the glycemic measures observed during the start period do not reflect the glycemic status of the user prior to starting with FLASH monitoring. The IMPACT randomized controlled trial indicated that glycemic metrics, such as %TIR and %TBR, changed from baseline within 2 weeks after starting FLASH monitoring [13]. Our study does not have baseline measures and we cannot draw conclusions about change from before and after the start period of FLASH monitoring. However, we have evidence that, irrespective of change from baseline, there is a strong association between increased experience, increased adherence to sensor reapplication, and improved glucose control.

Limitations

Strengths of our study include the size of the data set and the large cohort of FLASH users included in the analysis, as well as the real-life setting, the consistent technology being used, and the broad inclusion criteria. A limitation is a potential selection bias due to the requirement that users completed 36 sensors with sufficient data in each period, which is more likely to select (in both groups) for users with greater overall engagement with the technology. Thus, the effects of improved skill in device operation, which may also come with more experience, may be underrepresented in the analyzed sample. We also accept that by modelling the adherence by the length of sensor gap time there is a risk that extended gap times may be a consequence of adhesive skin reactions or accidental sensor displacement, which are not accounted for in our model. However, by including data from users with 36 completed sensors and a high % data capture profile, we believe that we have minimized these risks. Other limitations include the anonymized data structure, which lacks demographic information other than including FLASH users in the Czech Republic. Nor does the data include information on gender, age, duration of disease, clinical parameters (including HbA1c test data or socioeconomic indicators) or type of diabetes (although it can be assumed that the large majority of people included in our analysis have T1D because partial or full reimbursement of FLASH in the Czech Republic has been mandated since October 2018 only for this group of patients).

Conclusions

This real-world study is the first longitudinal investigation of the impact of the FreeStyle Libre system for people with diabetes in the Czech Republic, with a focus on evaluating adherence and experience with FLASH technology. This has been possible by analyzing selected groups of users with either low or high adherence with FLASH monitoring during the start period. We have shown that, with experience over time, users of the FreeStyle Libre system can modify their behavior and can become more adherent in sensor reapplication, and this is accompanied by longitudinal increases in %TIR, and reductions both in %TAR and glucose variability. This shows that low initial adherence to use of FLASH monitoring is not a strong predictor of poor glycemic outcomes.