Introduction

In patients with diabetes mellitus (DM), daily glucose monitoring assists in evaluating the therapeutic effect, allowing patients to adapt their diet and clinicians to adjust the medications [1, 2], especially to prevent hypoglycemia, one of the known complications of hypoglycemic medications [3, 4]. Currently, venous blood glucose, fingertip instantaneous blood glucose, and glycated hemoglobin (HbA1c) are available for glucose monitoring [5]. Among them, fingertip instantaneous blood glucose is convenient and widely used, which can be self-administrated by the patients at home. However, its accuracy is easily affected by exercise, diet, drugs, or mood swings, and it can only be measured several times a day [6, 7]. The availability of an accurate, convenient, and stable glucose monitoring system for people with DM who require long term and dynamic glucose control is therefore of paramount importance.

The continuous glucose monitor (CGM) method has been proposed as a new approach to this issue over the past few years [8, 9]. CGM utilizes a disposable glucose sensor that can be inserted (7–14 days) or implanted (up to 180 days) subcutaneously for continuous glucose measurement in the interstitial fluid [10]. Currently, there are three types of CGM systems available worldwide, including real-time CGM, retrospective CGM, and intermittently scanned CGM or flash glucose monitoring [12]. In China, imported flash glucose monitoring system FreeStyle Libre (Abbott Diabetes Care, Alameda, CA) is an ideal choice for glucose monitoring outside of hospitals, since other systems require frequent finger blood calibrations [11]. However, a previous study conducted in China suggested FreeStyle Libre system underestimated venous glucose when glucose levels fluctuate quickly [13], and the accuracy of FreeStyle Libre was lower in the hypoglycemic range [14]. Besides, several patients suffered from the sensor falling off, measurement discrepancies, and cutaneous reactions [15]. Therefore, alternative CGM options should be explored.

The SiJoy GS1 CGM system (GS1 CGM, Shenzhen Sibionics Technology Co. Ltd.) measures glucose concentration in subcutaneous interstitial fluid using glucose sensors. It is implanted through an applicator into the subcutaneous tissue, and the acquired quantitative values are temporarily saved in a memory module. In addition, a glucose curve is also plotted to reflect the fluctuation trend of patients’ glucose levels, allowing continuous monitoring. Due to a lack of previous studies on GS1 CGM, this multicenter clinical trial aimed to evaluate its performance, usability, and safety in Chinese adults with DM.

Methods

Patients and study design

This multicenter clinical trial was conducted in Shandong Provincial Hospital, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, and Shenzhen University General Hospital from July 3, 2020, to November 14, 2020.

The inclusion criteria were as follows: (1) aged 18 years and above; (2) diagnosed with DM and received treatment previously, including type 1 DM, type 2 DM, or subjects with 1 or more DM risk factors. The definition of DM risk factors was according to the Chinese Guidelines for the Prevention and Treatment of Type 2 Diabetes, 2017 Edition [2].

Exclusion criteria were as follows: (1) severe hypoglycemia with loss of consciousness or epilepsy within 6 months; (2) heart failure or previous cerebrovascular disease with hemiplegic sequelae; (3) conditions (such as cognitive decline, mental illness) that make participants unable to understand informed consent or communicate with study personnel; (4) scars on the sensor wearing surface caused by severe skin scalds, burns, sunburns, trauma, ulcers, and surgeries; (5) severe skin lesions, such as extensive eczema, extensive scars, extensive tattoos, herpetic dermatitis, severe edema, and psoriasis; (6) abnormal coagulation function, anemia, or abnormal hematocrit; (8) blood donation within half a year due to the safety concern; (9) women who are pregnant, lactating, or plan to get pregnant in the next month; (10) prescription of other research drugs, devices, or instruments that will affect the objective evaluation, follow-up, and testing; (11) those who required magnetic resonance imaging (MRI) and computed tomography (CT) examinations during sensor wearing period; (12) any other condition that researcher believes needs to be excluded from the study, such as needle sickness or emotional instability.

GS1 CGM

GS1 CGM consisted of sensor package, reader package, and system software. Participants wore two GS1 CGM systems, one on each upper arm, the first one being the main sensor and the other being the auxiliary sensor. The medical personnel recorded the serial data of the main and secondary sensors.

Within 5 min after venous blood was drawn, the glucose values measured by the main and auxiliary sensors of GS1 CGM were recorded respectively [16]. An alarm function on the GS1 CGM was kept on within half an hour before and after venous blood collection. A threshold of 11.1 mmol/L was set for hyperglycemia, and 4.4 mmol/L was set for hypoglycemia. The software would inform the patient when glucose values monitored by CGM were above or below these thresholds, and medical personnel kept accurate records when alarm occurred.

After 14 days of wearing, participants followed the instructions to remove the sensor and disinfect the skin in the wearing area with alcohol.

Venous blood glucose test

Participants were randomly divided into 5 groups (A-E). GS1 CGM glucose detection results and venous blood glucose detection results were collected at the early stage (group A, the first day of wearing the sensor), the early and middle stage (group B, the second-fifth day of wearing the sensor), the middle stage (group C, the sixth-ninth day of wearing the sensor), the middle and late stage (group D, the tenth-thirteenth day of wearing the sensor), and the late stage (group E, the fourteenth day of wearing the sensor). For randomization, a third-party statistical center generated a random number table through a computer randomization program, and randomly matched 70 numbers (R001–R070) to the groups A–E. Then, the third-party statistical center prepared 70 cards. Only the number (R001–R070) was displayed on the cards, and the groupings were hidden by an opaque coating. Neither the investigator nor the sponsor could know the grouping corresponding to the group number. Then, investigators at each center recruited subjects based on a competitive enrollment process, and assigned randomized cards to subjects according to the order of enrollment. Subjects were given the cards and scratched off the coating to obtain their grouping.

After randomization, every participant was assigned to take the venous blood glucose test in corresponding stage. Venous blood was drawn once every 15 min (± 3 min) for at least 7 h until 28 valid data records were collected. Intravenous blood glucose was measured with the EKF glucose detector (Biosen C-line).

Primary outcomes

Based on the comparison of glucose values measured by GS1 CGM and venous blood glucose test, the primary outcomes included the 20/20% consistency, the percentage of A+B zones of Clarke error grid and Consensus error grid, and the mean absolute relative difference (MARD%).

The 20/20% consistency was defined as the percentage of the glucose value difference between GS1 CGM and the venous blood fell within the range of ± 20% (glucose value > 4.4 mmol/L) or ± 1.1 mmol/L (glucose value ≤ 4.4 mmol/L). The target of 20/20% consistency was 65% while the applied clinical standard was 60% according to National Medical Products Administration guidelines [17]. Performance was evaluated by the consensus error grid and Clarke error grid. Zone “A” represented absolutely accurate results; zone “B” represented acceptably accurate results; and zones C, D, and E were inaccurate results with varying degrees. The percentage of A, B, C, D, and E zones was calculated as well as the percentage of A + B zones. The target percentage of A + B zones was 95% while the clinical standard was 90% [17]. MARD% was defined as the mean of the absolute deviation from the arithmetic mean for all single measured value. The target of MARD% was 18%, and the clinical standard was 20% [17].

The following subgroup analyses were performed on all primary outcomes: venous blood glucose level: < 4.4 mmol/L, 4.4–11.1 mmol/L, and > 11.1 mmol/L; DM type: type 1, type 2, and others of DM; center: Shandong Provincial Hospital as center 1, Sun Yat-sen Memorial Hospital, Sun Yat-sen University as center 2, and Shenzhen University General Hospital as center 3.

Secondary outcomes

Secondary outcomes included success rate of abnormal glucose alarm, success rate of abnormal glucose measurement, sensor stability, sensor repeatability, and product usability.

The abnormal glucose alarm was triggered when the venous blood glucose value was lower than 4.4 mmol/L or higher than 11.1 mmol/L. In such case, the GS1 CGM was supposed to alarm the participants. The abnormal glucose alarm success rate was defined as the percentage of the main sensor of GS1 CGM successfully triggering abnormal glucose alarm within 15 min before or 30 min after venous blood extraction. The abnormal glucose measurement success rate was the percentage of the main sensor of GS1 CGM successfully measured abnormal glucose within 30 min before and after venous blood extraction. Sensor stability was determined by 20/20% consistency and MARD% at different stage. Sensor repeatability was evaluated by the paired absolute relative difference (PARD%), which was calculated as the difference between main and auxiliary sensor at the same time point divided by mean value of main and auxiliary sensor at the same time. Product usability was evaluated by questionnaire completed by participants. The questionnaire consisted of 18 items, including ability to understand the operation process of the system, independent and convenient operation, and overall experience. The score for each item was as follows: strongly agree (5 points), agree (4 points), uncertain (3 points), disagree (2 points), and strongly disagree (1 point), with a total score of 90 points.

Adverse events

Adverse events (AE) and serious AE (SAE), including the type, severity, number, and incidence, were recorded. After the equipment was removed, the evaluation of the skin condition of the wearing part was conducted by medical personnel for signs of redness, swelling, pain, itching, and bleeding.

Sample size and data collection

Assuming the target 20/20% consistency was 65%, two-sided α = 0.05, and considering the drop-out rate of 20%, about 1200 measurement values were required to obtain 1 power of 80%. The estimation that each participant can provide at least 20 measurement values lead to at least 60 participants.

Statistical analysis

SAS version 9.4 (SAS Institute Inc., Cary, NC) was used for statistical analysis. Continuous variables with normal distribution were expressed as mean ± standard deviation (SD), and those with non-normal distribution were expressed as median (range). Categorical variables were expressed as frequency (percentage). Paired t test or Wilcoxon rank sum test was adopted for comparison of continuous variables, chi-square test or Fisher’s exact test was utilized for categorical variables, and Cochran-Mantel-Haenszel test was performed for grading variables. Two-sided p-values <0.05 were considered statistically significant.

Results

Patients’ characteristics

A total of 78 subjects were screened for the study. Four cases of abnormal coagulation function, 2 cases of anemia, 1 case of positive urine pregnancy test, 1 case of scar with a diameter of 1cm on the lateral side of the left upper limb, were excluded, and finally 70 cases were enrolled. A total of 69 cases completed the study, with a drop-out rate of 1.43% (1/70) due to the auxiliary sensor dropping.

The mean age of participants was 41.5 ± 13.2 years, among which 29 (42.0%) were male. 56.5% of them had type 1 DM, 34.8% had type 2 DM, and 8.7% were others for DM (Table 1)

Table 1 Patients’ characteristics
Full size table

Primary outcomes

The 20/20% consistency was 91.82%, and the lower limit of 95% confidence interval was 90.50%. The MARD% was 8.83% ± 4.03% (Table 2).

Table 2 Primary outcomes
Full size table

The percentage of A+B zones of Clarke error grid was 99.22%, and the lower limit of 95% CI was 98.72% (Table 2, Fig. 1). The percentage of A, B, and D zones was 89.80%, 9.43%, and 0.78%, respectively, and no matched glucose values fell in C and E zones. Besides, the percentage of A+B zones of consensus error grid was 99.90%, and the lower limit of 95% CI was 99.63% (Table 2, Fig. 2). The percentage of A, B, and C zone was 97.62%, 2.28%, and 0.10%, respectively, and no matched glucose values fell in D and E zones.

Fig. 1
figure 1

Clarke error grid

Full size image
Fig. 2
figure 2

Consensus raster grid

Full size image

Subgroup analysis

The 20/20% consistency for venous blood glucose level < 4.4 mol/L, 4.4–11.1 mmol/L, and > 11.1 mmol/L were 82.91%, 92.71%, and 91.49% (95% CI lower limit: 74.84%, 91.19%, and 88.59%), respectively. The percentage of A+B zones of Clarke error grid were 88.03%, 100.0%, and 99.79% (95% CI lower limit: 80.74%, 99.73%, and 98.82%), respectively. The percentage of A+B zones of consensus error grid were 98.29%, 100.00%, and 100.00% (95% CI lower limit: 93.96%, 99.73%, and 99.22%), respectively. The MARD% of the three subgroups were 15.03 ± 18.06%, 8.38 ± 7.51%, and 8.64 ± 7.39% (95% CI upper limit: 13.04%, 7.19%, and 7.70%), respectively.

Secondary outcomes

For venous blood glucose < 4.4 mmol/L, the abnormal glucose alarm success rate was 84.62% (99/117) and the abnormal glucose measurement success rate was 90.53% (86/95). For venous blood glucose > 11.1 mmol/L, the abnormal glucose alarm success rate was 95.16% (452/475) and the abnormal glucose measurement success rate was 88.52% (401/453) (Table 3).

Table 3 Secondary outcomes
Full size table

The 20/20% consistency in the early stage (day 1), the early and middle stage (days 2–5), the middle stage (days 6–9), the middle and late stage (days 10–13), and the late stage (day 14) were 95.00%, 96.25%, 91.20%, 88.95%, and 85.71%, respectively. The MARD% in the early stage (day 1), the early and middle stage (days 2–5), the middle stage (days 6–9), the middle and late stage (days 10–13), and the late stage (day 14) were 8.73 ± 2.65%, 7.84 ± 2.97%, 8.50 ± 5.08%, 9.38 ± 4.07%, and 11.86 ± 3.14%, respectively (Table 3). A total of 1684 matched glucose value from main and auxiliary sensors were collected, and the PARD% was 7 ± 7%. The score of product usability was 86.59 ± 5.17 out of 90 (Table 3).

Adverse events

Adverse events occurred in one participant (Table 4), with an incidence rate of 1.43% (1/70). The main manifestation was mild fever due to cold, which might be irrelevant to sensor implantation. The participant was appropriately treated and completed the trial eventually. No serious adverse events occurred.

Table 4 Adverse events
Full size table

Discussion

In this multicenter clinical trial, GS1 CGM demonstrated satisfactory performance, usability, and safety. Our results showed the 20/20% consistency of 91.82%, the MARD% of 8.83 ± 4.03%, and the percentage of A + B zones of Clarke and consensus grid of 99.22% and 99.90%, which met the requirements in the Guiding Principles for Technical Review of Registration of Continuous Glucose Monitoring System [17], regardless of venous glucose level, DM type, and trial center. The abnormal glucose measurement and alarm success rate, stability, repeatability, and usability were also promising. Adverse events occurred in one participant (mild fever), with an incidence rate of 1.43%, and no SAEs occurred, indicating the good safety of GS1 CGM. Based on the results, GS1 CGM was approved for marketing in Mainland China now.

According to two recent international consensuses, the performances of GS1 CGM meet the needs of clinical applications [18, 19]. Compared with other popular CGMs, FreeStyle Libre demonstrated an overall MARD between 13.2 and 18% [20, 21], long-term implant system Eversense demonstrated an overall MARD between 8.8 and 14.8% [22], and Dexcom G5 displayed 16.3% [21]. Our study demonstrated a MARD% of 8.83 ± 4.03%, which was comparable with other CGMs. It is reported that CGM will underestimate the venous glucose, and the accuracy of CGM decreased over time [23]. Indeed, our results showed that the 20/20% consistency decreased from 95.00% on day 1 to 85.71% on day 14, and the MARD% increased from 8.73 ± 2.65% on day 1 to 11.86 ± 3.14% on day 14, which was the same as other CGMs and should be further improved. Previous studies suggested CGM might not be a suitable tool for assessing glucose in prediabetes [24]. However, our study did not focus on the high-risk patients. The further study will be conducted in the future.

It was reported recently that accuracy of FreeStyle Libre [14] as well as other CGM systems [25,26,27] was lower in the hypoglycemic range. Our subgroup analysis suggested that 20/20% consistency rates were 82.91% for blood glucose < 4.4 mmol/L, with lower limit of the 95% CI of 74.84%, which is acceptable in clinical practice and comparable to the results of FreeStyle Libre. However, future improvements are needed. It is important to note, despite the measurement vulnerability of the CGM system, the success rate of clinically significant hypoglycemia detection was much higher (90.53%). Thus, the acceptable alarm success rate (84.62% and 95.16%) and measurement success rate (90.53% and 88.52%) have also been achieved.

In order to evaluate the repeatability, two sensors were implanted for each participant on the lateral edge of biceps brachii muscle of the left and right arms in our study. A total of 1684 matched glucose value of main and auxiliary sensors were collected, and the PARD% (7%) was in the range of clinical standard.

Regarding safety, previously reported AEs related to sensor implantation were bruising, erythema, allergic reaction, or pain/discomfort [28]. Eversense study reported one SAE being secondary to inability to remove the sensor. In a recent randomized controlled trial on type 1 DM [29], excessive correction of high or low glucose levels and alert fatigue leading to silencing of alerts were added to the list of adverse events. In our study, no SAEs or withdrawals related to AEs occurred. Aside from one case of losing sensor (described as falling off by itself), one AE (mild fever) occurred, and the AE was determined not relevant to the sensor by investigator. However, the design of our study did not include evaluating the risk of excessive glucose correction, and more personalized approach to glucose control may be more appropriate, as a single glucose target or range may not be optimal for all patients at all times [30].

This study has some limitations. Firstly, it was one-arm clinical trial without compare to other conventional blood glucose monitoring methods; however, we used the venous blood for comparison, which also could show the accuracy of GS1 CGM. Future real-world studies with large sample size and comparison with other CGMs are warranted. Second, the sample was small, but each patient with more than 28 points, which was enough to calculate. In addition, our study did not involve children, pregnant women, and other special groups, as well as the accuracy of glucose detection in different subpopulations. The accuracy of GS1 CGM must be further evaluated in different groups of patients, and especially in critically ill patients as it was previously reported to be relatively low [31]. Moreover, our study did not evaluate the influence of decreased accuracy during the later period on patients. Thus, the long-term blood glucose monitoring effect and prognosis of the equipment for diabetic patients need to be further confirmed. Finally, usability and cost-effectiveness of the GS1 CGM as well as other similar systems [32] can be improved.

Conclusions

GS1 CGM demonstrated satisfactory consistency, stability, repeatability, usability, and safety, and the product performance was able to satisfy the needs of clinical application. Mild adverse event was reported in one case and no SAE or withdrawal due to AE occurred.