Accuracy of the 15.5-Day G7 iCGM in Adults with Diabetes (2025)

Study design and participants

This prospective multicenter study was conducted at six sites in the United States. Demographic information is summarized in Table 1. The protocol and consent forms were approved by local and/or centralized institutional review boards. Informed written consent was obtained from study participants before study initiation. All participants had a screening visit where baseline laboratory tests were conducted and informed signed consent was obtained. Baseline hematocrit values were in the normal range and pregnancy tests where applicable were negative (as per inclusion/exclusion criteria). This study was conducted in accordance with the provisions in applicable regulations and guidelines, including Good Clinical Practice and the Declaration of Helsinki. The study was approved by the FDA through the Investigational Device Exemption process and registered at ClinicalTrials.gov (NCT05263258).

The point estimates achieved by the G7 CGM system for the arm wear location¹³ were utilized for the sample size calculation for this study. Assuming a one-sided significance level of 0.05, study power of 80%, study attrition of 15%, and a one-sided z-test of a binomial proportion with a continuity adjustment, the required sample size was 129 participants.

Between January and August 2022, the study enrolled 130 adults (≥18 years of age) with T1D, T2D using intensive insulin therapy (IIT), or T2D on basal insulin therapy. Inclusion criteria included willingness to wear up to two 15-day G7 devices and to avoid injecting insulin or wearing an insulin pump infusion set within 3 inches of the CGM insertion site. Exclusion criteria included pregnancy, anemia, or any condition, per investigator’s discretion, that could place the participant at risk by glucose manipulations. There were no inclusion or exclusion criteria related to race, sex, body mass index (BMI), or current medications or supplements.

Study procedures

Forty participants wore a 15-day G7 sensor on the back of both upper arms, and 90 participants wore one 15-day G7 sensor on the back of one upper arm. The primary CGM (i.e., the CGM used for analyses) was the first-applied CGM. Both the study participants and staff were blinded to the CGM data, and the participants continued to use their own CGM/glucose meter device to manage diabetes during the study. All sensor insertions were asked to be performed at the clinic by participants. Participants applied the overlay patch (over patch) immediately after the sensor was successfully applied.

Participants were scheduled for four in-clinic sessions of up to 13 h each on one day in each of the following ranges: day 1–3 (hours 0 to <72), day 4–7 (hours 72 to <168), day 9–12 (hours 192 to <288), day 13–15.5 (hours 288 to <372). During the clinic session, arterialized venous blood was drawn from an intravenous catheter approximately every 10–15 min (±5 min), and glucose concentrations were obtained using a Yellow Springs Instrument (YSI) 2300 Stat Plus glucose analyzer (YSI, Inc., Yellow Springs, OH, USA).¹³ While up to 50 samples of arterialized venous blood were drawn from each participant during in-clinic visits, blood draws did not exceed 3 mL/kg of body weight. Participants with T1D or T2D-IIT underwent glucose manipulations, during which deliberate but closely supervised (by the principal investigator) insulin and glucose challenges were performed to induce hyperglycemia and hypoglycemia to achieve glucose values between 40 and 400 mg/dL as per the protocol. All 15-day G7 investigational device data were blinded from the participants, and the providers used their YSI measurements data for inducing glucose manipulations by higher food intake or increasing the amount of insulin. Participants with T2D using basal insulin therapy attended clinic sessions, but intentional glucose manipulations were not performed. CGM removals were performed on day 15.5 ± a 48-h window for scheduling purposes. There were no calibrations performed on the research devices during the study. During the removal visit, insertion sites and adhesive areas were evaluated by study staff and participants were surveyed about their CGM use history and the 15-day G7’s ease of use during the study.

Data analysis

To obtain CGM-measured glucose data for analysis, raw sensor signals were collected during the study and post-processed using a proprietary algorithm. Blood glucose samples measured using the YSI were matched with the first CGM value in the 5 min immediately after the YSI measurement. Only matched pairs with CGM values in the 40–400 mg/dL reportable range were included in the performance analyses.

To determine whether the 15-day G7 system met the special controls required for iCGM systems, the following seven primary endpoints were analyzed: (1) proportion of dichotomized status of CGM reading in comparison to YSI within 20% to be >87%; (2) for CGM readings <70 mg/dL, proportion of dichotomized status of CGM reading in comparison to YSI within 15 mg/dL to be >85%; (3) for CGM readings <70 mg/dL, proportion of dichotomized status of CGM reading in comparison to YSI within 40 mg/dL to be >98%; (4) for CGM readings 70–180 mg/dL, proportion of dichotomized status of CGM reading in comparison to YSI within 15% to be >70%; (5) for CGM readings 70–180 mg/dL, proportion of dichotomized status of CGM reading in comparison to YSI within 40% to be >99%; (6) for CGM readings >180 mg/dL, proportion of dichotomized status of CGM reading in comparison to YSI within 15% to be >80%; and (7) for CGM readings >180 mg/dL, proportion of dichotomized status of CGM readings in comparison to YSI within 40% to be >99%. Each primary endpoint was tested at a one-sided significance level of 0.05. Point estimates and lower bound from the one-sided 95% confidence interval (CI) for each of the seven primary endpoints were derived from the generalized estimating equations (GEE) method.

For analysis of accuracy across CGM glucose ranges, the absolute difference from the YSI glucose measurement was calculated for CGM readings <70 mg/dL. For CGM readings ≥70 mg/dL, the absolute relative difference (ARD) of the CGM reading compared with the YSI glucose reading was calculated. Sensor heterogeneity was assessed by analyzing each sensor’s mean ARD (MARD) and by determining the proportion of sensors having >80% of their values within ±20 mg/dL of YSI values <70 mg/dL or within 20% of YSI values ≥70 mg/dL (hereafter referred to as %20/20). The related accuracy metrics %15/15, %30/30, and %40/40 were calculated similarly. To align with the International Organization of Standardization guidelines, %15/15, %20/20, %30/30, and %40/40 were also calculated with 100 mg/dL as the dividing glucose concentration (i.e., for %15/15: the proportion of CGM values within ±15 mg/dL of YSI values <100 mg/dL or within 15% of YSI values ≥100 mg/dL). These alternative metrics are presented in Tables 2–8. Performance was assessed across glucose concentration ranges and days of sensor wear. For analysis of accuracy at various CGM rates of glucose concentration change, rate of change (RoC) was calculated as the difference between consecutive glucose values per unit of time (units of milligrams per deciliter per minute). MARD and %20/20 were calculated for the following RoC ranges: <−2, −2 to <−1, −1 to <0, 0 to 1, >1 to 2, and >2. Clinical accuracy was characterized by consensus error grid (CEG) analysis.¹⁵

Detection of hypoglycemia and hyperglycemia, recognition of true alert or detection rates, and performance of the Urgent Low Soon alerts were calculated as previously described.¹³ Briefly, the true alert rate quantifies the proportion of hypoglycemia or hyperglycemia alerts corroborated by a YSI value at or below (hypoglycemia alert) or at or above (hyperglycemia alert) the threshold setting within ±15 min of the alert. The true detection rate quantifies the proportion of hypoglycemia or hyperglycemia events (defined by the YSI measurement) that were associated with a CGM alert within ±15 min of the event. Urgent Low Soon alert will occur if a user’s glucose level is predicted to fall ≤55 mg/dL within 20 min.

Heterogeneity of accuracy across participant subgroups was analyzed using a GEE model to adjust for intrasubject correlation with a significance level of 0.15. Device safety was assessed by evaluation of all device-related adverse events. Ease of use and satisfaction with CGM use were evaluated via survey questions assessed at insertion and removal. All analyses were performed using SAS software, version 9.4 (SAS Institute, Inc., Cary, NC, USA).

Study population

A total of 130 individuals were enrolled in the study (Table 1). Most participants were female (53.1%), had T1D (86.9%), and were overweight or obese (67.7%). Seven of the 130 participants did not complete all study requirements; 3 missed ≥1 clinic visit, 2 withdrew from the study, and 2 were not able to complete the study because both sensors (one primary and one backup) fell off the body before completion of clinic sessions. The analysis population included all participants with ≥1 CGM-YSI matched pair, which included all 130 participants. From these 130 participants, there were 20,441 valid CGM-YSI matched pairs. After excluding 131 matched pairs that were outside of the CGM’s reportable range of 40–400 mg/dL, there were 20,310 matched pairs used for accuracy analyses.

Accuracy metrics and iCGM performance

The overall MARD for the 15-day G7 CGM system was 8.0%. The overall %15/15, %20/20, %30/30, and %40/40 agreement rates were 87.7%, 94.2%, 98.9%, and 99.8%, respectively. The results for the consensus error grid of CGM versus YSI are presented in Figure 1. Overall, 99.9% of sensor readings are in zones A and B of the CEG.

Primary endpoints and performance goals for special control metrics (i.e., iCGM metrics) are summarized in Table 2 alongside the endpoints achieved in this study. The 15-day G7 CGM system exceeded all performance goals and met iCGM criteria.

Individual sensor performance is summarized in Figure 2. Among the 130 individual sensors, 122 (93.8%) had >80% of CGM-YSI matched pairs that met %20/20 accuracy.

Accuracy across wear periods, glucose ranges, and participant characteristics

Accuracy across four wear periods throughout the 15.5-day sensor lifetime is summarized in Table 3 and Table 4. Sensor accuracy across the 70–180 mg/dL glycemic range and deliberately induced hypoglycemic and hyperglycemic ranges is summarized in Table 5 and Table 6. Accuracy remained high throughout the sensor life and across glucose ranges. In addition, after the 27-min warm-up, the %20/20 accuracy over the first 12 h and 24 h of wear was 83.4% and 89.4%, respectively.

There were no statistically significant differences in accuracy between participant subgroups, including sex, age, diabetes type, baseline A1c, BMI category, or study site.

Accuracy at various rates of glucose concentration change

Sensor accuracy was assessed over six different categories of rates of glucose concentration change (Table 7 and Table 8). Accuracy was highest (i.e., lowest MARDs and highest %20/20 agreement rates) when glucose concentrations were changing by <1 mg/(dL·min). When the glucose RoC was <−2 mg/(dL·min) (n = 269 matched pairs), the %20/20 agreement rate was 85.9%, and when glucose RoC was >2 mg/(dL·min) (n = 767), the %20/20 agreement rate was 92.4%.

Threshold and Urgent Low Soon alert performance

The true alert rate estimates the probability of the user experiencing an event (defined by the YSI measurement based on the alert setting) within ±15 min of the CGM alert. Of the 5019 alerts arising when the hypoglycemia alert threshold was set to 70 mg/dL, 82.0% were true alerts, indicating the CGM value ≤70 mg/dL was corroborated by a YSI value ≤70 mg/dL within ±15 min. Of the 12,496 alerts arising when the hyperglycemia alert threshold was set to 240 mg/dL, 95.4% were true alerts, indicating the CGM value ≥240 mg/dL was corroborated by a YSI value ≥240 mg/dL within ±15 min.

The true detection rate estimates the probability that the CGM provides an alert within ±15 min of the user experiencing an event (defined by the YSI measurement). Of the 2125 events occurring when the hypoglycemia alert threshold was set to 70 mg/dL, 92.2% of the hypoglycemia events were true detections, defined as a YSI value ≤70 mg/dL corroborated by a CGM value ≤70 mg/dL. Of the 5390 hyperglycemia events occurring when the hyperglycemia alert threshold was set to 240 mg/dL, 94.3% were true detections, defined as a YSI value ≥240 mg/dL corroborated by a CGM value ≥240 mg/dL.

An additional analysis was performed to determine how often level 1 hypoglycemia (YSI glucose ≤70 mg/dL) was detected with the hypoglycemia alert threshold set to 55 mg/dL. Of the 1218 alerts arising when the hypoglycemia alert threshold was set to 55 mg/dL, 89.2% were associated with a YSI measurement ≤70 mg/dL within ±15 min.

Of the seven unique Urgent Low Soon alerts occurring during the in-clinic sessions, six (85.7%) were followed by a YSI value ≤70 mg/dL within 30 min.

Adverse events and user experience

There were no serious adverse events related to the study device among enrolled participants. Three participants (2.3%) reported four mild to moderate device-related adverse events. These included two reports of local infection (one moderate and one mild), one report of mild skin irritation (edema, erythema, and medical adhesive-related skin irritation), and one report of moderate pain or discomfort during sensor wear.

Users surveyed at insertion (Table 9) found the insertion process to be easy or very easy (126/128; 98.4%). Most participants (109/128; 85.2%) had used personal CGM systems in the past, with Dexcom G6 as the most frequently used CGM (76/109; 69.7%). Among those with a history of personal CGM use, most (100/109; 91.7%) answered at the time of removal that the 15-day G7 system was about the same or easier to use than their personal CGM system.