Performance and Safety of a 15-Day Real-Time Continuous Glucose Monitoring System in Chinese Adults With Diabetes: A Multicenter Trial

Abstract

Background:

This study evaluated the performance and safety of a 15-day real-time continuous glucose monitoring (rtCGM) system in Chinese adults with diabetes, focusing on its accuracy and clinical utility for long-term glucose management.

Methods:

A clinical evaluation was conducted using four rtCGM sensors per participant, with placement sites on both the upper arm and the abdomen. The system’s accuracy was assessed using a factory-calibrated model. Primary outcomes included the mean absolute relative difference (MARD), the 20/20% agreement rate, and the proportions of Clarke and Consensus Error Grid in zones A+B.

Results:

A total of 74 participants were screened. The MARD values were 8.44% for the upper arm and 8.91% for the abdomen under factory-calibrated model. The 20/20% agreement rates were 95.78% for the upper arm and 94.41% for the abdomen under factory-calibrated model. The Clarke and Consensus Error Grid A+B proportions were 99.53% and 99.96% for the upper arm, and 99.46% and 99.82% for the abdomen under factory-calibrated model, respectively.

Conclusion:

The CGM system showed high accuracy, robust alert performance, stable repeatability, and favorable safety over 15 days in Chinese adults with diabetes, supporting its clinical utility for glucose monitoring in these adult patients.

Keywords

continuous glucose monitoring diabetes

Introduction

Effective diabetes management depends on accurate and frequent monitoring of glucose levels. Patients with diabetes experience constant fluctuations in glucose levels due to diet, exercise, medication, and stress, necessitating vigilant self-monitoring.¹ Continuous glucose monitoring (CGM) systems are associated with improved glycemic control and reduced hypoglycemia in patients with type 1 diabetes (T1D) and type 2 diabetes (T2D).^2,3 In ambulatory adults, CGM has been associated with improved time in range (TIR), fewer hypoglycemic events, and lower glycated hemoglobin levels, supporting its use in routine diabetes care where appropriate.⁴

Although several factory-calibrated CGM systems are available, the performance data of new devices in this field are essential for their clinical acceptance. In addition to point accuracy, a comprehensive assessment of a CGM system must include its stability throughout the sensor wear period, ability to accurately detect glycemic extremes (hypoglycemia and hyperglycemia), and trend accuracy, which is a critical feature for predictive alerts and insulin dosing decisions.^5,6 Furthermore, there is insufficient evidence for extended-wear sensors compared with 10- or 14-day devices. Determining the optimal anatomical site for sensor placement remains an area of interest.⁷ The upper arm is increasingly used as an alternative to the abdominal site; however, comprehensive head-to-head comparisons of sensor performance at these sites using a factory-calibrated model are limited. Modern real-time factory-calibrated real-time continuous glucose monitoring (rtCGM) generally improves TIR and reduces hypoglycemia, but user-initiated calibration may still be useful in specific clinical scenarios or usage environments.

Despite the increasing global uptake, robust multicenter data on Chinese adults are limited. Therefore, we conducted a prospective, multicenter, clinical investigation of a 15-day rtCGM (Anytime 4Pro) versus venous plasma glucose sampled, and prespecifying analyses of wear-site effects (upper arm vs abdomen), along with primary and secondary endpoints. We hypothesized that the system would meet predefined acceptance criteria across primary endpoints and demonstrate robust secondary performance in an adult Chinese population.

Methods

Study Design and Participants

This study was a multicenter trial conducted to evaluate the performance and safety of our CGM system in Chinese patients with diabetes. Between January and May 2024, eligible participants from the First Affiliated Hospital with Nanjing Medical University, the Second Affiliated Hospital of Nanjing Medical University, the Affiliated Hospital of Xuzhou Medical University, and Huai’an Second People’s Hospital who were approved to wear the device for 15 days were enrolled in the study. Participants were aged >18 years with a clinical diagnosis of T1D or T2D. Detailed inclusion and exclusion criteria are provided in Supplemental Table S1. The study protocol was reviewed by the ethical committee (Ethics Approval Number: IRB-SOP-AF17; 2024-QX-002-LP-01; XYFY-2024-QL006-01; HEYL-H-2024-01). The study was registered with the Chinese Clinical Trial Registry (ChiCTR2400092405).

The device (Jiangsu Yuwell POCTech Biotechnology Co., Ltd, Jiangsu, China) consisted of a disposable sensor, a transmitter, software, and accessories. The CGM system demonstrates the following performance characteristics: the sensor requires a one-hour initialization period and offers an effective service life of 15 days. The in vivo measurement range for glucose has an overall system reportable range of 1.7 to 27.8 mmol/L. The disposable sensor was sterilized by electron-beam irradiation. The transmitter is powered by a built-in battery and operates within an environment of 5°C to 40°C and 15% to 90% relative humidity, while the sensor is designed for the normal human body temperature range. In addition, the CGM system is equipped with a dedicated adhesive overlay patch designed to enhance sensor stability. The CGM system components are shown in Figure 1.

Figure 1.

Illustration of the anytime 4Pro CGM device. Sensor with charger (left), applicator (right), and representative depiction of the Yuwell Anytime app (middle).

Study Procedures and Data Collection

The participants with diabetes received training and signed an informed consent form. After screening and enrollment, the subjects wore the device by themselves; four sensors were worn on both sides of the same implantation site on the upper arm and abdomen and were monitored and observed for 15 days.

One day was randomly selected for intensive venous blood collection from “days 1-2 after initialization”, “days 7-9 during wearing,” and “the last 24 hours of wearing.” All participants fasted for at least eight hours for intensive venous blood collection. To enrich hypoglycemic data, patients with T1D underwent intentional glucose modulation via slight increases in their routine insulin doses. This was performed under strict medical supervision to target glucose levels <3.9 mmol/L for no more than 1 hour. Investigators monitored glucose reduction in real-time, providing immediate carbohydrate intervention if glucose fell too rapidly or patients reported discomfort. After the product is initialized, blood was collected every 15 ± 3 minutes according to the randomized intensive blood collection time points. Venous blood collection lasted for at least seven hours, and venous plasma was analyzed within 15 minutes using an EKF glucose/lactic acid analyzer. Continuous glucose monitoring values were recorded within three minutes of venous blood collection. Because the EKF value is only measured every 15 minutes, low glucose value cannot always be obtained, which can lead to comparison failures. When the EKF glucose value is less than 4.4 mmol/L, intravenous blood sampling was added within 10 ± 5 minutes after the blood sampling time point. In the clinical calibration mode, the fasting fingertip glucose level was measured in the morning to calibrate the sensor once daily. The subjects were required to complete the calibration of the four products within five minutes of collecting fingertip blood in the clinical calibration mode setting.

Study Outcomes

The primary effectiveness outcomes included the percentage of CGM readings within 15/15%, 20/20%, and 40/40% of the reference values, percentage of measuring points in zones A + B of the Clarke error grid analysis, percentage of measuring points in zones A + B of the consensus error grid analysis, mean absolute difference (MAD) and mean absolute relative difference (MARD). The consistency rate of CGM readings and venous blood glucose values (20/20% standard) refers to the percentage of sensor values that fell within ±20 mg/dL for glucose concentrations ≤4.4 mmol/L and 20% for glucose concentrations >4.4 mmol/L.⁸

For error grid analyses, greater than 95% of the A + B area was considered the standard. The distribution of the measurement points in zones A to E was analyzed. Zone A showed no effect on the clinical action. Zone B had little or no effect on clinical outcomes. Zone C is likely to affect clinical outcomes. The remaining zones D and E may have significant medical or dangerous consequences. The number of data points in each zone and their percentages, as well as the sum of zones A and B, were calculated.⁹

The MAD is provided for low glucose values (≤3.9 mmol/L), and MARD is provided for glucose values >3.9 mmol/L. For the MARD, based on the available guidelines, a value of less than 18% was considered the standard for this study.¹⁰ The equation for the MARD is as follows:

A R D_{K} = 100 % \times \frac{| y_{C G M} (t_{K}) - y_{E K F} (t_{K}) |}{y_{E K F} (t_{K})}

M A R D = \frac{1}{N} \sum_{k = 1}^{N} A R D_{K}

The secondary outcomes assessed in our study were alert accuracy, sensor stability, sensor repeatability, and trend accuracy. As for the hypoglycemia alert accuracy, hypoglycemia is defined as EKF venous glucose concentration < hypoglycemia alert threshold, and the alert threshold is set to 3.9 mmol/L (70 mg/dL). Hyperglycemia is defined as EKF venous glucose concentration > hyperglycemia alert threshold, and the alert threshold is set to 10 mmol/L (180 mg/dL). The sensor stability was assessed by analyzing the MARD values during different phases of the 15-day wear period: beginning (days 1-2), middle (days 7-9), and end (day 15). Trend accuracy was evaluated by comparing the rate of glucose change measured by the CGM system with that determined by the EKF reference. The correct detection rate of hypoglycemia/hyperglycemia refers to the percentage of occasions when the EKF blood glucose level was above/below the alert setting, and the CGM initiated an alert within 15 minutes before or after the EKF blood glucose level was above or below the alert setting. Device safety was assessed by evaluating all device-related adverse events.

Statistical Analysis

All statistical calculations were performed using SAS 9.4 software. The sample size was determined based on predefined accuracy endpoints, including a 20/20% agreement rate (AR) of 90% (against an 87% threshold), a MARD ≤ 15%, and ≥ 90% of sensor values within zones A and B of the Clarke/consensus error grid. A sample size of 60 participants was calculated to provide >80% power at a two-sided significance level of 0.05. To satisfy regulatory requirements for glycemic distribution, the study further ensured the collection of at least 450 hyperglycemic pairs (≥10.0 mmol/L) and 60 hypoglycemic pairs (≤3.9 mmol/L). Accounting for an anticipated 20% dropout rate, a total of 74 participants were planned for enrollment.

Rigorous bias control measures were integrated into the study design. Selection bias was minimized by enrolling representative participants (diabetes patients, adults ≥18 years needing glucose monitoring). Procedural consistency was maintained through cross-center standardized operating procedures and daily calibration of EKF instruments. Furthermore, to control for statistical bias, a 1:1:1 randomization protocol was adopted to determine the timing of intensive sampling days, ensuring the integrity of the comparative analysis.

To describe the quantitative indicators, mean, standard deviation, median, minimum, maximum, and quartile values were calculated. The enumeration data were described statistically using frequency (composition ratio). According to the data distribution, the t-test (for data with homogeneity of variance and normal distribution) or Wilcoxon rank-sum test was used for intergroup comparison of quantitative data. The chi-square test or precise probability method was used for categorical data (if the chi-square test was not applicable), and the Cochran-Mantel-Haenszel test was used for ranked data. Statistical tests were two-sided (unless otherwise stated), and P ≤ .05 was considered statistically significant.

Results

Participant Characteristics

A total of 74 participants were screened across the four centers, and two participants were excluded after screening. Seventy-two participants (45 males and 27 females) completed all study visits. Among these participants, 43 had T2D and 29 had T1D, with an age range of 18 to 72 years old. The baseline characteristics of all included participants are presented in Supplemental Table S2. Among 288 sensors, 287 were worn for the full 15 days (99.7%), while one sensor fell off after five days of wear (Supplemental Fig S1).

Primary Outcomes Under Factory-Calibrated Model

Agreement Analysis

The AR was assessed by comparing the real-time CGM glucose values with the venous glucose values measured using the EKF. The “paired points” represent all valid matching pairs between CGM readings and EKF values.

As summarized in Table 1, under the factory-calibrated mode, the CGM system showed high overall accuracy. For upper arm placement, the overall 20/20% AR was 95.78% (4268/4456), with a MARD of 8.44%. In the hypoglycemic range (<3.9 mmol/L), the system showed robust performance with a 20/20% AR of 99.12% and a MAD of 0.28 mmol/L. The ARs for the euglycemic (3.9-10.0 mmol/L) and hyperglycemic (>10.0 mmol/L) ranges were 95.00% and 96.48%, respectively. For abdominal placement, the overall accuracy remained high, though slightly lower than the upper arm, with an overall 20/20% AR of 94.41% (4207/4456) and a MARD of 8.91%. The 20/20% AR in the hypoglycemic range was 95.91%, with a MAD of 0.34 mmol/L. Accuracy in the euglycemic and hyperglycemic ranges was 93.16% and 96.48%, respectively. Across both placement sites, the 40/40% AR exceeded 99% for nearly all categories, and reached 100% in the hyperglycemic range.

Table 1.

Accuracy of Anytime 4Pro CGM System Versus EKF Reference Under Factory-Calibrated Model.

Placement	Glucose range,mmol/L	Paired pairs, N	15/15%^a, %	20/20%^a, %	40/40%^a, %	MAD^b, mmol/L	MARD,%	Bias^c,%
Upper arm	All	4456	88.17	95.78	99.89	NA	8.44	1.56
	<3.9	342	98.25	99.12	99.71	0.28	NA	–0.59
	3.9-10.0	2720	86.43	95.00	99.85	NA	8.85	2.24
	>10.0	1394	89.10	96.48	100.00	NA	7.51	–0.30
Abdomen	All	4456	85.35	94.41	99.82	NA	8.91	1.18
	<3.9	342	92.11	95.91	98.54	0.34	NA	3.10
	3.9-10.0	2720	82.10	93.16	99.89	NA	9.50	0.81
	>10.0	1394	90.03	96.48	100.00	NA	7.66	1.36

Abbreviations: MAD, mean absolute difference; MARD, mean absolute relative difference; NA, not applicable.

Glucose values ≤ 4.4 mmol/L, absolute differences between CGM and EKF reference values were calculated; glucose values > 4.4 mmol/L, relative differences were calculated.

MAD is provided for glucose values ≤ 3.9 mmol/L, and MARD is provided for glucose values > 3.9 mmol/L.

Bias values given as mean relative difference.

Error Grid Analysis

Clarke Error Grid Analysis

Under the Clarke error grid analysis for upper arm placement, of the 4456 paired values, an overwhelming majority were located in clinically acceptable zones A and B, accounting for 4435 (99.53%) of the data points. Specifically, 4272 (95.87%) readings fell into zone A representing measurements within ±20 mg/dL of the reference. An additional 163 (3.66%) readings were in zone B, indicating that although there was a greater relative difference at lower glucose levels, the error was still not clinically meaningful. Only 21 (0.47%) readings were found in clinically questionable or dangerous zones C, D, and E. For abdominal placement, Clarke error grid analysis showed similarly high agreement, with 4432 (99.46%) paired values falling within zones A and B; 4226 (94.84%) readings were in zone A, and 206 (4.62%) were in zone B. A small number of 24 (0.54%) readings were located in zones C, D, or E (Figure 2a and b, Supplemental Table S3).

Figure 2.

Error grid analysis under factory-calibrated model. (a, b) Clarke error grid analysis for upper arm (a) and abdomen (b); (c, d) consensus error grid analysis for upper arm (c) and abdomen (d).

In the consensus error grid analysis, which often provides a more refined evaluation, the performance was even higher. For the upper arm, 4454 (99.96%) of the paired values were situated within zones A and B. Of the readings, 4327 (97.11%) were in zone A, and 127 (2.85%) were in zone B. Crucially, only two (0.04%) readings fell into problematic zones C, D, or E. Abdominal placement also demonstrated exceptional performance in the consensus error grid analysis, with 4448 (99.82%) of the paired values within zones A and B. Of the total readings, 4239 (95.13%) readings were in zone A and 209 (4.69%) were in zone B. Only eight (0.18%) readings were found in zones C, D, or E (Figure 2c and d, Supplemental Table S3).

Further details on the error grid analysis and supplementary data (Supplemental Table S3) confirm the high agreement between the CGM system and the EKF reference. Figure 2 and Supplemental Table S3 present a detailed breakdown of the error grid analyses, including both Clarke and consensus error grids, for the upper arm and abdomen placements.

Secondary Outcomes Under Factory-Calibrated Model

Alert Accuracy

The accuracy of the CGM system in detecting hypoglycemic (<3.9 mmol/L) and hyperglycemic (>10 mmol/L) events is summarized in Table 2. For hypoglycemic alerts on the upper arm, 56 alerts were issued, with a success rate of 92.86%. Correspondingly, 342 hypoglycemic events were detected, with a success rate of 94.44%. In the abdomen, 47 hypoglycemia alerts were issued, with a high success rate of 97.87%, and 342 events were detected, with a success rate of 79.53%.

Table 2.

Accuracy of Alert and Detection Under Factory-Calibrated Model.

Placement	Glucose threshold	No. of alerts	Success rate of alert (%)	No. of detections	Success rate of detection (%)
Upper arm	Hypoglycemia (<3.9 mmol/L)	56	92.86	342	94.44
Upper arm	Hyperglycemia (>10 mmol/L)	167	97.01	1394	97.42
Abdomen	Hypoglycemia (<3.9 mmol/L)	47	97.87	342	79.53
Abdomen	Hyperglycemia (>10 mmol/L)	172	95.53	1394	97.70

For hyperglycemia alerts, the upper arm demonstrated a success rate of 97.01% for 167 alerts and 97.42% accuracy in detecting 1394 events. The abdomen showed a success rate of 95.53% for 172 alerts and 97.70% accuracy in detecting 1394 events.

Stability and Repeatability of Sensor

Stability across the 15-day wear period was assessed at pre-specified windows: early (days 1-2), mid (days 7-9), and end (day 15) periods. As shown in Table 3, for the upper arm, the MARD remained stable, being the lowest during the middle of the sensor life (7.72% on days 7-9) and slightly higher at the beginning (8.56%) and end (9.05%). The abdominal sensor showed consistent performance throughout the experiment, with MARD values of 9.23% (early), 8.85% (mid), and 8.66% (end).

Table 3.

Stability of Anytime 4Pro CGM System Versus EKF Reference During Clinic Session Period Under Factory-Calibrated Model.

Placement	Clinic session period	Sample size (N)	Paired pairs (n)	MARD (%)
Upper arm	Beginning (days 1-2)	24	1474	8.56
	Middle (days 7-9)	24	1526	7.72
	End (day 15)	24	1456	9.05
Abdomen	Beginning (days 1-2)	24	1474	9.23
	Middle (days 7-9)	24	1526	8.85
	End (day 15)	24	1456	8.66

Abbreviations: MARD, mean absolute relative difference.

The repeatability was quantified using the paired absolute relative difference (PARD) between two simultaneously worn sensors at the same location. Upper arm placement showed a PARD of 5.90% compared with 6.80% for the abdomen, indicating superior measurement consistency (Table 4).

Table 4.

Repeatability of CGM System Under Factory-Calibrated Model.

Placement	Paired pairs (n)	PARD (%)
Upper arm	4456	5.90
Abdomen	4456	6.80

Abbreviations: PARD, paired absolute relative difference.

Trend Accuracy

As shown in Table 5, the accuracy of the glucose trend arrows in the system was assessed by comparing the rate of change in the CGM system with that of the reference system. The results were presented in a grid format, with the percentage in each cell representing the proportion of time for which the trend of the CGM system agreed with the reference trend.

Table 5.

Trend Accuracy of Anytime 4Pro CGM System Under Factory-Calibrated Model.

Placement	CGM rate range (mmol/L/min)	EKF rate range (mmol/L/min)						Total (n)
Placement	CGM rate range (mmol/L/min)	<–0.11	–0.11 to –0.06	–0.06 to 0	0-0.06	0.06-0.11	>0.11	Total (n)
Upper arm	<–0.11	26 (50.98%)	14 (27.45%)	11 (21.57%)	0 (0.00%)	0 (0.00%)	0 (0.00%)	51
	–0.11 to –0.06	13 (4.66%)	159 (56.98%)	87 (31.18%)	18 (6.45%)	2 (0.71%)	0 (0.00%)	279
	–0.06 to 0	17 (1.06%)	126 (7.82%)	1126 (69.90%)	303 (18.81%)	35 (2.17%)	4 (0.25%)	1611
	0-0.06	6 (0.43%)	25 (1.79%)	339 (24.25%)	855 (61.16%)	133 (9.51%)	40 (2.86%)	1398
	0.06-0.11	0 (0.00%)	0 (0.00%)	20 (6.58%)	101 (33.22%)	149 (49.01%)	34 (11.19%)	304
	>0.11	0 (0.00%)	0 (0.00%)	3 (1.77%)	25 (14.79%)	33 (19.52%)	108 (63.91%)	169
Abdomen	<–0.11	22 (36.67%)	22 (36.67%)	15 (25.00%)	1 (1.67%)	0 (0.00%)	0 (0.00%)	60
	–0.11 to –0.06	17 (6.05%)	134 (47.69%)	108 (38.43%)	18 (6.41%)	4 (1.42%)	0 (0.00%)	281
	–0.06 to 0	15 (0.97%)	134 (8.60%)	1038 (66.58%)	327 (20.98%)	41 (2.63%)	4 (0.26%)	1559
	0-0.06	7 (0.50%)	25 (1.79%)	384 (27.47%)	798 (57.08%)	138 (9.87%)	46 (3.29%)	1398
	0.06-0.11	1 (0.28%)	9 (2.55%)	38 (10.76%)	127 (35.98%)	127 (35.98%)	51 (14.45%)	353
	>0.11	0 (0.00%)	0 (0.00%)	3 (1.86%)	31 (19.25%)	42 (26.09%)	85 (52.80%)	161

For the upper arm, the AR and error rate (ER) were 63.56 and 36.44%, respectively. When EKF glucose rate rapidly decreases (<–0.11 mmol/L/min), there were no instances of CGM glucose rate rapid increases (>0.06 mmol/L/min). The CGM glucose rate (<–0.06 mmol/L/min) was also 0% when EKF glucose rate increased rapidly (>0.11 mmol/L/min). For the abdomen, AR and ER were 57.82% and 42.18%, respectively. When EKF glucose rate rapidly decreases (<–0.11 mmol/L/min), only 0.28% CGM rate rapidly increases (>0.06 mmol/L/min). When the EKF glucose rate rapidly increases (>0.11 mmol/L/min), there were no instances of CGM glucose rapid decreases (<–0.06 mmol/L/min). The trend accuracy results from the upper arm and abdomen suggest that CGM can provide relatively accurate data for making diabetes treatment decisions.

Impact of Different Calibration Models

The performance of the CGM system under various calibration frequencies is summarized in Table 6 and Supplemental Table S4. Compared to the factory-calibrated model (overall MARD of 8.44% for upper arm and 8.91% for abdomen), calibrating the sensor once a week or once a day resulted in a marginal improvement in the MARD for the upper arm (7.90% and 7.89%, respectively). A similar slight improvement was observed for abdominal placement with daily calibration (MARD, 8.65%). Despite these slight reductions in MARD, the 20/20% ARs remained robust and relatively stable across all models, the overall 20/20% agreement of upper arm was 95.56% with weekly calibration and 95.60% with daily calibration; the overall 20/20% agreement of abdomen was 93.31% with weekly calibration and 93.36% with daily calibration. Notably, in the hypoglycemic range (<3.9 mmol/L), the system maintained high clinical safety across both manual calibration models, with 20/20% ARs exceeding 99.4% for the upper arm and 96.3% for the abdomen. Furthermore, the 40/40% ARs reached or neared 100% across all stratified glucose ranges under both weekly and daily calibration protocols.

Table 6.

Accuracy of Anytime 4Pro CGM System Versus EKF Reference Under Different Calibration Models.

Calibration model	Placement	Glucose range (mmol/L)	Paired pairs, N	15/15%^a, %	20/20%^a, %	40/40%^a, %	MAD^b, mmol/L	MARD,%	Bias^c,%
Calibration once a week	Upper arm	All	4456	88.55	95.56	99.91	NA	7.90	0.90
		<3.9	342	96.49	99.42	100.00	0.29	NA	–0.66
		3.9-10.0	2720	87.21	94.78	99.85	NA	8.29	1.32
		>10.0	1394	89.24	96.13	100.00	NA	7.11	–0.32
	Abdomen	All	4456	84.96	93.31	99.75	NA	8.72	–0.03
		<3.9	342	93.57	96.20	99.71	0.38	NA	0.92
		3.9-10.0	2720	82.90	92.28	99.67	NA	9.26	–0.55
		>10.0	1394	86.87	94.62	99.93	NA	8.16	0.72
Calibration once a day	Upper arm	All	4456	88.76	95.60	99.93	NA	7.89	0.60
		<3.9	342	96.49	99.42	100.00	0.290	NA	–0.45
		3.9-10.0	2720	87.76	95.11	99.89	NA	8.27	1.00
		>10.0	1394	88.81	95.62	100.00	NA	7.14	–0.34
	Abdomen	All	4456	85.17	93.36	99.78	NA	8.65	–0.18
		<3.9	342	93.57	96.20	99.71	0.32	NA	0.73
		3.9-10.0	2720	83.16	92.32	99.71	NA	9.18	–0.73
		>10.0	1394	87.02	94.69	99.93	NA	8.08	0.78

Abbreviations: MAD, mean absolute difference; MARD, mean absolute relative difference; NA, not applicable.

Glucose values ≤ 4.4 mmol/L, absolute differences between CGM and EKF reference values were calculated; glucose values > 4.4 mmol/L, relative differences were calculated.

MAD is provided for glucose values ≤ 3.9 mmol/L, and MARD is provided for glucose values > 3.9 mmol/L.

Bias values given as mean relative difference.

Discussion

In this multicenter investigation of Chinese adults with T1D and T2D, a 15-day rtCGM system demonstrated high analytical and clinical performance relative to venous plasma glucose, and no serious device-related adverse events were observed. The system met all pre-specified primary criteria, achieving a 20/20% agreement of 95.78% for the upper arm and 94.41% for the abdomen, an overall MARD of 8.44% for the upper arm and 8.91% for the abdomen, and low clinical risk according to the error grids (Clarke A + B of 99.53% and 99.46% for the upper arm and abdomen, respectively; Consensus A + B of 99.96% and 99.82% for the upper arm and abdomen, respectively). The performance was consistent across the factory and user calibration schedules (weekly and daily). Secondary endpoints including alert performance, trend accuracy, stability across time-in-wear, and paired-sensor repeatability also met the acceptance thresholds. Furthermore, the rtCGM accuracy results showed strong performance, even at hypoglycemic and hyperglycemic extremes. No safety concerns were raised during the study period.

The inclusion of the upper arm and abdomen as valid sensor insertion sites in the study design represents a significant design choice with practical implications. Under the factory-calibrated model, the results of the CGM system showed that the MARD of the upper arm was 8.44% with a 20/20% agreement rate of 95.78%, and that of the abdomen was 8.91% and a 20/20% agreement rate of 94.41%, indicating numerically higher accuracy at the arm. Meanwhile, error grid analyses corroborated clinical acceptability at both sites (consensus A + B: 99.96% arm, 99.82% abdomen; Clarke A + B: ≥99.4%), implying a very low frequency of clinically consequential misclassifications. Therefore, unlike many competing systems that are primarily validated or limited to a single site, the results of this head-to-head study suggest that the multisite capability of the CGM system offers enhanced flexibility and choice for both clinicians and patients. In scenarios where one site may be compromised (eg, owing to skin conditions, surgical interventions, or patient preference), the availability of an alternative site ensures the continuity of glucose monitoring. This versatility is particularly valuable for expanding the accessibility and usability of CGM across broader patient demographics.

Accordingly, the CGM system offers optional calibration. This dual approach is a strategic advantage: it caters to the majority of users who benefit from the convenience of free calibration while simultaneously offering an avenue for further precision enhancement for those who require or desire it. This flexibility allows clinicians to fine-tune the accuracy of the system for individual patients, thereby meeting diverse therapeutic needs and ensuring optimal glucose management.

The system demonstrated excellent stability over a 15-day wear period, with consistently low MARD values and no significant degradation trends. Sustained accuracy throughout the lifespan of a sensor is critical for long-term glucose trend analysis and reliable data collection. The low PARD values (5.90% for the upper arm and 6.80% for the abdomen) indicated high repeatability, suggesting that successive measurements were highly consistent, further bolstering confidence in the reliability of the data. Trend accuracy analysis revealed that the CGM system accurately captured glucose rate changes across a wide spectrum, from rapid declines to significant increases. The strong diagonal distribution in the trend analysis tables demonstrated the capability of the system to predict the direction and magnitude of glucose fluctuations, which is crucial for proactive diabetes management and timely intervention. This ability to reliably track glucose trends complements accurate glucose readings and provides a comprehensive picture of glycemic control.

The CGM system exhibits significant strengths, including prospective multicenter design; high-frequency venous reference (every 15 ± 3 minutes); excellent accuracy (MARD ~8.4-8.9%), high clinical acceptability as evidenced by error grid analyses (>99.8% in zones A + B of the consensus error grid), and robust performance in a factory-calibrated mode. Its 15-day wear period offers cost-effectiveness and enhanced data collection, whereas the flexibility of optional calibration and dual insertion sites (upper arm/abdomen) caters to diverse patient needs. However, the main limitation of our study is the relatively small size of sample, and the limited hypoglycemic data may affect the evaluation of device performance at low blood glucose levels. Another limitation of this study is the use of averaged data from dual sensors for accuracy calculations, a methodology adopted to strictly comply with Chinese National Medical Products Administration 2018/2023 guidelines. We recognize that international standards typically favor the selection of a single sensor (eg, first inserted or specific arm) to represent real-world performance. While our results provide a robust overview of system reliability under current regulatory frameworks, they should be interpreted with this methodological distinction in mind. Our future studies will be optimized to use single-sensor analysis to ensure better alignment with international standards. Other potential limitations include the need for further validation across broader patient populations and in fully free-living conditions, a direct comparative study with other leading systems, and a detailed assessment of the impact of optional calibration.

Conclusion

In summary, a 15-day rtCGM system demonstrated high accuracy and low clinical risk in Chinese adults, with robust stability and high repeatability across wear sites. By adding wear site, repeatability, and calibration strategy evidence, this study complements prior extended-wear data and provides population-specific best practices.

Supplemental Material

sj-docx-1-dst-10.1177_19322968261445103 – Supplemental material for Performance and Safety of a 15-Day Real-Time Continuous Glucose Monitoring System in Chinese Adults With Diabetes: A Multicenter Trial

Supplemental material, sj-docx-1-dst-10.1177_19322968261445103 for Performance and Safety of a 15-Day Real-Time Continuous Glucose Monitoring System in Chinese Adults With Diabetes: A Multicenter Trial by Mei Zhang, Yibing Lu, Hongwei Ling, Wen Hu, Xuqin Zheng, Mei Huang, Meng Zhao, Xiaoqing Wang, Guanqun Zhang, Yanan Zhang and Tao Yang in Journal of Diabetes Science and Technology

Footnotes

Acknowledgements

The authors thank the individuals who participated in the study and the research staff at the investigational sites.

Abbreviations

AR, agreement rate; CGM, continuous glucose monitoring; ER, error rate; MAD, mean absolute difference; MARD, mean absolute relative difference; PARD, paired absolute relative differencepaired absolute relative difference; rtCGM, rea-time continuous glucose monitoring; T1D, type 1 diabetes mellitus; T2D, type 2 diabetes mellitus; TIR, time in range

ORCID iD

Mei Zhang

Author Contributions

Y.T. served as the principal investigator (PI), conducted the study, and led the writing, reviewing, and editing of the article. Z.M. conducted the study, performed formal analysis, methodology, validation, visualization, and contributed to writing, reviewing, and editing. L.Y.B., L.H.W. and H.W. acted as local site PIs and assisted in writing, reviewing, and editing. Z.X.Q., H.M., M.Z., W.X.Q., G.Q.Z., and Y.N.Z. provided conceptualization, methodology, and supervision.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supplemental Material

Supplemental material for this article is available online.

References

Ewunetu

Belay

Abere

, et al. Home self monitoring of blood glucose using a glucometer and its determinants among diabetes mellitus. Sci Rep. 2025;15(1):17217. doi:10.1038/s41598-025-02196-4

Langendam

Luijf

Hooft

, et al. Continuous glucose monitoring systems for type 1 diabetes mellitus. Cochrane Database Syst Rev. 2012;1(1):CD008101. doi:10.1002/14651858.CD008101.pub2

Seidu

Kunutsor

Ajjan

, et al. Efficacy and safety of continuous glucose monitoring and intermittently scanned continuous glucose monitoring in patients with type 2 diabetes: a systematic review and meta-analysis of interventional evidence. Diabetes Care. 2024;47(1):169-179. doi:10.2337/dc23-1520

Battelino

Danne

Bergenstal

, et al. Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care. 2019;42(8):1593-1603. doi:10.2337/dci19-0028

Alva

Brazg

Castorino

Kipnes

Liljenquist

Liu

Accuracy of the third generation of a 14-day continuous glucose monitoring system. Diabetes Ther. 2023;14(4):767-776. doi:10.1007/s13300-023-01385-6

Alva

Bailey

Brazg

, et al. Accuracy of a 14-day factory-calibrated continuous glucose monitoring system with advanced algorithm in pediatric and adult population with diabetes. J Diabetes Sci Technol. 2022;16(1):70-77. doi:10.1177/1932296820958754

Garg

Kipnes

Castorino

, et al. Accuracy and safety of Dexcom G7 continuous glucose monitoring in adults with diabetes. Diabetes Technol Ther. 2022;24(6):373-380. doi:10.1089/dia.2022.0011

Zhou

Zhang

, et al. Performance of a new real-time continuous glucose monitoring system: a multicenter pilot study. J Diabetes Investig. 2017;9(2):286-293. doi:10.1111/jdi.12699

Bedini

Wallace

Pardo

, et al. Performance evaluation of three blood glucose monitoring systems using ISO 15197: 2013 accuracy criteria, consensus and surveillance error grid analyses, and insulin dosing error modeling in a hospital setting. J Diabetes Sci Technol. 2015;10(1):85-92. doi:10.1177/1932296815609368

10.

Clarke

Cox

Gonder-Frederick

, et al. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care. 1987;10(5):622-628.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.06 MB

0.00 MB