FormalPara Key Summary Points

Why carry out this study?

The third generation of “Flash” technology, i.e., the FreeStyle Libre 3 system provides continuous glucose readings to the user without sensor scanning, utilizing an established third-generation glucose algorithm and a reduced size on-body sensor.

This study was performed to demonstrate that the reduced size of the on-body sensor does not adversely affect the performance of the sensor.

What was learned from the study?

Sensor performance is not adversely affected by the smaller sensor size compared to FreeStyle Libre 2 system.

The overall mean absolute relative difference (MARD) between the FSL3 values and venous plasma reference values from 95 study participants aged ≥ 6 years was 7.8%.

Introduction

The accuracy of continuous glucose monitoring (CGM) systems has significantly improved over the last two decades [1], and this improved accuracy has led to changes in how CGM data are used for the management of diabetes. Early versions of CGM systems required confirmation of the data with values obtained by traditional blood glucose monitoring (BGM) systems. Improvements in the accuracy of CGM data have ultimately resulted in the use of CGM data without a confirmatory BGM result. Recently, the US Food and Drug Administration has approved CGM systems as interoperable CGM systems, thereby enabling their integration with insulin delivery systems [2] and closed loop systems [3].

The FreeStyle Libre Flash Glucose Monitoring System (FSL; Abbott Diabetes Care, Alameda, CA, USA), which became available in 2014, was the first factory-calibrated CGM device available to people with both type 1 diabetes (T1D) and type 2 diabetes (T2D) [4,5,6,7]. This first-generation system continues to be used to facilitate diabetes management decisions, including insulin dosing. The second-generation FreeStyle Libre 2 (FSL2) system provides the additional options of hypoglycemia and hyperglycemia alarms. When configured, the FSL2 monitoring device sends data to the receiver every minute to generate an alarm if indicated. This monitoring device obtained the Conformitè Europëenne (CE) marking in 2018 [8]. The FSL2 system was updated in 2020 with a new glucose algorithm that provided improved accuracy across the measurement range, but specifically at the low end of the dynamic range [9]. In 2020, a third generation of the product, FreeStyle Libre 3 (FSL3), secured the CE marking [10]. Like FSL2, the FSL3 system provides continuous glucose readings every minute, as well as providing glucose levels, trends and alerts. The FSL3 system consists of two primary components: a sensor which transmits via Bluetooth Low Energy (BLE) and a BLE-enabled mobile application:

  • A disposable on-body assembly (sensor) that incorporates a subcutaneously implanted electrochemical glucose sensor (which incorporates wired enzyme technology) and associated electronics.

  • The user’s smartphone, with the FSL3 app enabled, which is used to start new sensors and display glucose readings. The user interface provides users with event logging service.

The FSL3 system uses the same sensing technology as the FSL2 system to monitor glucose levels in interstitial fluid. However, the FSL3 system has a new one-piece sensor applicator, and although the sensor tail is the same in both devices, the on-body sensor of the FSL3 system is about 70% smaller and the phone does not need to be scanned over the sensor to obtain a glucose result. The glucose results, in addition to glucose alarms, are automatically transmitted every minute to the FSL3 app (further referred to as the App) on a mobile phone by BLE technology and can be viewed on the respective device as required. The App provides the user with real-time glucose measurements (glucose values) accompanied by trend information (glucose arrows) on the phone display. In addition to real-time glucose results being displayed on the phone screen, the preceding 12 h of glucose data are also displayed upon opening the App. When not in communication with the FSL3 app (via loss of Bluetooth signal), the sensor can store up to 14 days of glucose data, which can be retrieved by re-establishing connection with the App.

The App uses the glucose data received via BLE to issue alarms that notify the user when the glucose level has passed a configurable high or low glucose threshold, as well as when the BLE signal is lost. The sensor and App communicate every minute to check if the glucose alarms have been triggered. To use the glucose alarms, the user must turn them on and have the phone within 10 m (33 ft) of the paired sensor, with both Bluetooth and notification settings enabled for the App.

In the present study, we evaluated the performance of the FSL3 system compared to the venous plasma reference in participants aged ≥ 6 years and to the fingerstick capillary blood glucose (BG) reference for pediatric participants aged 4 and 5 years.

Methods

Study Design

This was a pivotal, non-randomized, single arm, multi-center, non-significant risk, prospective evaluation of the performance of the FSL3 system. The protocol and informed consent forms underwent ethical review and were approved by a duly constituted Advarra Institutional Review Board (Pro00049169) [11]. All subjects provided written informed consent prior to any study activities. The clinical trial was conducted in accordance with the Helsinki Declaration of 1964 and its later amendments. A total of 108 participants aged ≥ 4 years were enrolled in the study at four US clinical research centers. Sample size was calculated based on detecting a clinically relevant difference for the purpose of statistically characterizing clinical point accuracy. Six participants failed the screening criteria, one participant withdrew consent, and one participant did not have sensor data that could be paired with reference glucose measurement. Thus, of the 108 participants enrolled, 100 were assessed as evaluable for inclusion in the effectiveness analysis, and 101 participants provided responses to the usability questions.

Participants wore two sensors (one on the back of each upper arm) for up to 14 consecutive days following sensor application. The point accuracy of the FSL3 system was evaluated by comparing the interstitial glucose readings obtained from the first applied sensor with evaluable data to the plasma venous YSI (YSI 2300 STAT PLUS Glucose and Lactate Analyzer; YSI Incorporated, Yellow Springs, OH, USA; referred to further as the YSI reference) reference value for participants aged ≥ 6 years (n = 95). The YSI reference measurements were performed at each study site during up to three in-clinic sessions, each lasting for up to 8 h to obtain venous blood for YSI measurements every 15 min when glucose is between 70 mg/dL (3.9 mmol/L) and 250 mg/dL (13.9 mmol/L) and every 5 min when glucose is < 70 mg/dL or > 250 mg/dL. YSI quality control (QC) tests were performed throughout the duration of the plasma venous testing to ensure that the analyzers at each study site had the same measurement quality. These in-clinic days were allocated to participants such that sensor wear days 1, 2, 3, 7, 8, 9, 12, 13, and 14 were all included in the accuracy analysis. Participants aged 4 and 5 years (n = 5) used a FreeStyle Libre Reader for capillary BG reference in one in-clinic setting for up to 4 h. Sensor data were collected using a smartphone. At the time of sensor insertion and removal, study staff recorded any adverse events at the sensor insertion site as well as participant-reported adverse events. Study participants (or their caregivers) were asked to respond to a usability questionnaire at both the time of insertion and time of removal of the sensor.

Data Analysis

Only matched data pairs for which the sensor results were within the reportable range (40–500 mg/dL, or 2.2–27.8 mmol/L) were used for the performance evaluation. The agreement levels were calculated relative to a glucose concentration difference (in mg/dL) when the reference glucose value was < 70 mg/dL and relative to a normalized concentration difference (in %) ≥ 70 mg/dL, at different ranges: ± 15%/± 15 mg/dL (0.8 mmol/L), ± 20%/± 20 mg/dL (1.1 mmol/L), and ± 40%/± 40 mg/dL (2.2 mmol/L). The mean absolute relative difference (MARD) was calculated as the absolute value of the average percentage difference between the paired sensor and reference glucose values. Clinical accuracy was examined by consensus error grid (CEG) analysis [12].

Paired sensor-YSI reference data were grouped into four wear periods: (1) beginning (days 1, 2, and 3); (2) early middle (days 7 and 8); (3) late middle (days 9 and 12); and (4) end (days 13 and 14). The sensor-to-sensor precision was determined by pairing results obtained within 5 min from two sensors worn simultaneously with matched wear time for subjects for whom these data were available. The average coefficient of variation (CV) and paired absolute relative difference (PARD) were calculated by taking the arithmetic mean of all glucose reading pairs.

The lag between the sensor glucose reading and the YSI reference was evaluated by performing least square linear regression of the difference between the sensor and YSI readings versus the sensor rate of change. The sensor rate of change was calculated as the instantaneous rate of change (i.e., using the sensor reading matched to YSI, along with the previous and following values, covering a total of 30 min). The slope of the regression line is the mean lag time.

All analyses were performed using SAS software, version 9.4 (SAS Institute, Inc., Cary, NC, USA).

Results

Participant demographic data are provided in Table 1. Among the 100 evaluable study participants, 56% were female, 56% were adults, 44% were pediatric patients, and 83% had T1D. An insulin pump was used by 53% of the study participants. Characteristics of the study participants are presented in Table 2. The age of participants ranged from 4 to 78 years, and the body mass index ranged from 15.2 to 48.7 kg/m2. Glycated hemoglobin levels ranged from 5.2% to 13.4%.

Table 1 Demographics of the study participants
Full size table
Table 2 Characteristics of the study participants
Full size table

There were 6845 matched data pairs compared to YSI reference values available for analysis from study participants aged ≥ 6 years. Deming regression of the sensor results versus the YSI reference had a slope of 1.03 and intercept of − 8.1 mg/dL (− 0.45 mmol/L) with a correlation coefficient of 0.96. For the participants aged 4 and 5 years, there were 72 matched data pairs compared to the self-monitoring blood glucose (SMBG) reference, and the Deming regression had a slope of 0.96, an intercept of 6.3 mg/dL (0.35 mmol/L) with a correlation coefficient of 0.93. The ratio of the measurement errors of reference method (x) and test method (y) was set at 1, and no weighting applied in the Deming regression.

The overall performance of the FSL3 system against the reference by age of the study participant is provided in Table 3. Overall the MARD of the FSL3 system against the YSI reference was 7.8%; for the adult and pediatric participants, the MARD was 7.5% and 8.6%, respectively. Overall, the 20 mg/dL/20% agreement was 93.4% against the YSI reference; for the adult and pediatric participants, the agreement was 94.9% and 89.7%, respectively. Among the 95 primary sensors with YSI reference, 73 (76.8%) sensors had > 90% of the sensor-reference matched pairs within 20 mg/dL or 20% of the reference and 72 (75.8%) sensors had MARD of < 10%. The MARD for the participants aged 4 and 5 years was 10.0% against the SMBG reference. The results for CEG analysis are presented in Fig. 1, which shows that 92.1% of the sensor readings are in zone A and 99.9% are in zones A and B of the CEG.

Table 3 Summary of performance by age group
Full size table
Fig. 1
figure 1

Consensus error grid analysis plot for CGM vs. YSI glucose values. CGM Continuous glucose monitoring, YSI plasma venous blood glucose reference using the YSI 2300 STAT PLUS Glucose and Lactate Analyzer (YSI reference)

Full size image

Table 4 presents the accuracy and bias measures of the data at different glucose ranges compared to the YSI reference. At glucose levels < 70 mg/dL, the accuracy was assessed as difference in mg/dL and ≥ 70 mg/dL as percentage difference. Since this study did not include glucose manipulations, the number of CGM-YSI pairs at glucose levels < 54 mg/dL (3.0 mmol/L) are limited. At glucose levels ≥ 54 mg/dL (3.0 mmol/L), > 90% of the results were within 20 mg/dL or 20% of YSI across the measurement range. Analysis of sensor accuracy at different glucose rates of change shows MARD of 10.9% at < − 2 mg/dL/min, 10.0% at − 2 to − 1 mg/dL/min, 7.5% at − 1 to 1 mg/dL/min, 8.4% at 1 to 2 mg/dL/min, and 9.7% at > 2 mg/dL/min.

Table 4 Accuracy and bias measures against the YSI reference
Full size table

The percentage within 20% or 20 mg/dL of the YSI reference and MARD performance throughout the wear duration is presented in Table 5. Reference data were collected over 9 of the 14 days of sensor wear and were grouped into four time windows consistent with the previously reported time windows for FSL2 system [8]. Both system agreement with YSI reference and MARD remained consistent across the wear period with > 90% of the data within 20 mg/dL or 20% and MARD of < 9% across the wear period.

Table 5 Accuracy and bias measures against the YSI reference as a function of sensor wear period
Full size table

The accuracy of the FSL3 system was not affected by diabetes type, type of insulin administration, wear period or study site. The analysis of variance (ANOVA) for the system agreement results for each of the above subgroups had a p value of > 0.05. The average sensor-to-sensor precision (CV) of 4.7% was determined by pairing results from the two sensors worn by the same participant simultaneously with matched wear time. For glucose levels < 100 mg/dL (5.6 mmol/L), the mean standard deviation (SD) was 5.3 mg/dL (0.3 mmol/L). The mean CV for glucose levels ≥ 100 mg/dL was 4.4% The PARD was 6.7%. Precision was consistent at different glucose rates of change. Pediatric participants (aged 6–17 years) showed a mean precision of 5.7% (PARD 8.0%) compared to 4.2% (PARD 6.0%) for the adult participants.

The mean lag time between the sensor and the venous reference was 1.8 ± 4.8 min for the 95 sensors with YSI reference. There were no serious adverse events reported during the study. Eight participants reported mild to moderate device-related adverse events, including bleeding (3), edema (1), erythema (5), induration (2), and itching (1). None of the erythema or edema events exceeded 2 on a Draize Scale of 0–4 [13].

The responses of the study participants or their caregivers to usability questions following sensor insertion and at the end of sensor wear are summarized in Table 6. The majority of the study participants or their caregivers (98 of the 99 respondents) agreed that applying the sensor was easy. Of the 99 respondents, 82 reported that applying the sensor was less painful than a routine fingerstick. Of the 80 respondents, 68 said that it was the easiest sensor to apply. In total, 99 of the 101 respondents indicated that wearing the sensor was painless.

Table 6 Participant responses to usability questions
Full size table

Discussion

The FreeStyle Libre CGM system has facilitated access to glucose sensing technology with more than 4 million individuals using the device to manage their diabetes [14]. The FSL3 system has a high degree of accuracy, specifically at the hypoglycemic ranges, optional hypoglycemia and hyperglycemia alarms, and automatic transmission of BG data to the smart phone app. The FSL3 system has the same sensor technology as that of the previous generation FSL2 system with a significantly smaller on-body sensor.

Studies of the FSL2 system with the same glucose algorithm have demonstrated a performance of 93.2% and 92.1% of the results within ± 20% or ± 20 mg/dL of the YSI reference for the adult and pediatric populations, respectively [9]. We found that the size of the on-body sensor of FSL3 compared to the FSL2 system has not significantly affected the overall performance of the sensor.

Accuracy of the sensor was found to be stable over the sensor wear period without any difference in overall performance relative to age, type of diabetes and insulin administration. The overall MARD of the system is 7.8%, compared to the FSL2 system (9.2% and 9.7% for the adult and pediatric participants, respectively) indicating that the change in the on-body component has not adversely affected the sensor performance [8]. The MARD value is also similar to that reported for other commercially available CGM systems [15,16,17].

The mean lag time between the sensor and the venous reference (1.8 ± 4.8 min) is comparable to that presented for the FSL2 system (adult and pediatric population lag times were 2.4 ± 4.6 and 2.1 ± 5.0 min, respectively). Since both systems use the same glucose algorithm, lag times are expected to be similar.

The FSL3 system has several features that distinguish it from the FSL2 system. The FSL3 system has a one-piece sensor applicator, and the on-body sensor is about 70% smaller than that of the FSL2 system. Also, the glucose results are continuously and automatically transmitted to the smartphone every minute; no scanning of the smartphone over the sensor is required. The data display on the phone screen presents the preceding 12 h of glucose data. The FSL3 sensor can store data up to 14 days. The FSL3 system provides an easy to use and comfortable sensor wear experience for up to 14 days, without the need for fingerstick measurements. These features are anticipated to simplify the usability of the device and improve user experience.

Strengths and Limitations

The study has a number of strengths and limitations. The study assessed the performance of the FSL3 system under conditions of natural glycemic variations. The study included 100 evaluable patients with diabetes aged ≥ 4 years to assess the accuracy of the system across the dynamic range over the sensor wear period.

A limitation of the study was it did not include insulin or meal challenges, thereby limiting the glucose rate of change available in the overall accuracy assessment. The study enrolled only patients who used multiple daily injections (MDI) or a pump for insulin delivery and, therefore, the data do not include patients who are not on intensive insulin therapy. The performance presented is against a reference measurement performed by trained operators; therefore, the influence of variability in reference measurements performed by patients when they are in home environment on the overall perception of accuracy is not known.

Conclusions

Our evaluation of the performance of the FSL3 system showed that the change to the on-body component has not adversely affected sensor performance compared to the FSL2 system, with an overall MARD of 7.8%, with 93.4% of the data within 20 mg/dL or 20% of the YSI reference, and stable performance over 14 days of use. With simplified user experience and demonstrated accuracy of the system, it is anticipated that the FSL3 system has the potential to further increase the adoption of sensor-based technology use for glucose management for people with diabetes.