Exogenous insulin replacement to obtain glycemic control is a hallmark for type 1 diabetes (T1D) treatment (1). The American Diabetes Association (ADA) recommends physical activity for individuals with type 1 diabetes as it improves glycemic control, decreases insulin requirements, and reduces cardiovascular complications in individuals with type 1 diabetes (2–5). However, exercise in these individuals is associated with hypoglycemia, which may be a barrier for obtaining glycemic control, and as a result many individuals with type 1 diabetes avoid engaging in regular physical activity (6, 7). Individuals with type 1 diabetes may need to consume considerable amounts of carbohydrates prior to physical activity in order to avoid exercise-induced hypoglycemia, which may reduce the potential benefits of vascular health and glycemic control that physical activity brings (8, 9). Continuous glucose monitoring (CGM) systems offer a way to frequently monitor glycemic changes throughout the day and particularly during exercise (10, 11). This provides individuals with type 1 diabetes a level of detail that cannot be achieved using capillary glucose meters, thus assisting in reduction of unnecessary glycemic fluctuations and episodes of hypoglycemia (12, 13).

Studies have indicated reduced accuracy of several CGM sensors during rapid glucose changes and low blood glucose levels, both commonly observed in individuals with type 1 diabetes and especially during exercise (14). CGM performance is clinically important since low sensor precision may lead to undetected events of hypoglycemia or unnecessary meal intake ensuing hyperglycemia. Here, we investigated the performance of two commonly used CGM systems, Dexcom G6 (DG6) and FreeStyle Libre 1 (FSL1), during plasma glucose (PG) decline and hypoglycemia, induced with or without exercise in individuals with type 1 diabetes.

All 15 participants ( Table 1 ) completed both clamp days yielding 551 and 543 DG6 sensor-PG pairs for Clamp-rest and Clamp-exercise, respectively, as well as 491 and 512 FSL1 sensor-PG pairs for Clamp-rest and Clamp-exercise, respectively. All participants placed the sensors according to specified guidelines except one who placed FSL1 on the upper thigh. For both clamp days, mean PG was kept at 6-7 mmol/L during the baseline-euglycemic phase although slightly decreasing towards the decline phase ( Figure 2 ). Target hypoglycemia was reached after 90-minutes and followed by steady-state hypoglycemia of <3.0 mmol/L. Overall, PG levels for both clamp days were comparable.

We report that exercise can affect the sensor accuracy of the investigated CGM systems by either a decrease in sensor accuracy and the ability to detect hypoglycemia as seen with DG6, or contrary by an increase in sensor accuracy as seen for FSL1 which may be of clinical relevance for physically active individuals with type 1 diabetes when choosing between CGM systems.

The overall MARD of DG6 obtained in this study under baseline euglycemia was substantially higher compared to previous studies (23–25). Shah et al. compared DG6 sensor readings with YSI glucose values in 62 participants and demonstrated a general overall MARD of 9% (23). Generally, the design in the studies were similar e.g., utilizing a YSI analyzer, obtaining arterialized venous blood, and relying on the factory calibration. Furthermore, we followed the manufacturer’s specified guidelines for inserting and initializing the sensor and doing so two days before the clamp days to avoid possible inaccuracy during the sensor warm-up time. Since the overall FSL1 MARD of our study was more comparable with previous studies (26–29), and since our study followed the same principles as previous studies, the deviating MARD values obtained from DG6 may not be explained by the design of our study. The deviation may rather be related to the applied sensors. As the higher-than-expected MARD of DG6 was observed at baseline for both test days, and that they were comparable to each other, the comparison outcomes between clamp days were considered valid.

Our results suggest that exercise has a negative effect on FSL1 performance without affecting the post-exercise sensor performance. Contrary, DG6 had an improvement in performance during exercise which persisted post-exercise. We hypothesize that the exercise-related performance increase of DG6 is caused by increased blood circulation and production of heat during exercise resulting in a subsequent increase in skin blood flow (30). This could potentially lead to a higher interstitial space fluid turnover rate and better equilibrium between plasma and interstitial space fluid, thus more accurate estimates of glucose (31). Conversely, a worsening in performance was seen for FSL1 during exercise. This may be explained by the placement of the sensor in an area of high movement and mechanical activity during exercise as opposed to the abdomen as seen for DG6 (32). Thus, the continual mechanical movement of the sensor may have outweighed the possible performance increase from increased skin blood flow.

Few studies have assessed the sensors’ performance during exercise (33–37), although they mostly relied on capillary blood glucose as a reference method. Guillot et al. showed no notable changes in DG6 sensor accuracy during exercise while wearing the sensors on the abdomen (33). Dyess et al. showed an overall decrease in DG6 sensor performance during exercise, however apart from the abdomen, the participant had the option to wear the sensor on the upper arms, buttocks, and thighs (34). When sub-analyzing for the abdomen only, Dyess et al. found an increase in sensor performance which is in accordance with our study and supports our hypothesis about the role of placement of sensors.

For FSL1, Moser et al. showed higher MARD during exercise in 19 individuals with type 1 diabetes compared to capillary blood glucose (35). Likewise, Fokkert et al. showed an increase in MARD for FSL1 during exercise in 23 individuals with type 1 diabetes compared to capillary blood glucose (36). However, Giani et al. showed no performance difference between rest and exercise in 17 young individuals with type 1 diabetes (37). As FSL1 is placed on the upper arm, an area of high activity during exercise, the decrease in sensor performance may be of this reason. We can only speculate if the performance of FSL1 would remain unchanged or increase if placed on the abdomen during exercise compared to rest. However, Charleer et al. showed a worsening in FSL1 performance when placed on the abdomen (38). Although sensor placement may have a notable role, factors such as varying populations included, different exercise forms and intensities employed, different extent of glycemic excursions and varying PG reference methods utilized may contribute to differing results.

According to Clarke error grid analysis, most of the data points in our study were in the clinically safe zones A+B for both sensors regardless of the clamp day, although the percentage of data points in the upper zone D was surprisingly high for DG6 during Clamp-rest. During Clamp-exercise, DG6 had a shift in data points from the clinically unacceptable zones (solely located in the upper D zone) to the clinically acceptable zones with an increase in zones A+B by almost 10 percentage points compared to Clamp-rest. This indicates that exercise potentially improves sensor accuracy and/or sensor lag time during rapid PG decline and hypoglycemia. In contrast, FSL1 showed a slight decrease of data points in the clinically acceptable zones during Clamp-exercise compared to Clamp-rest. FSL1 had almost a doubling of data points in zone D for Clamp-exercise indicating exercise having a clinically relevant negative effect. Thus, Clarke error grid analysis is consistent with the MARD results and could be explained in the same manner.

The previously mentioned studies also evaluated the clinical safety of DG6 and FSL1 using Clarke Error grid analysis. Guillot et al. showed good clinical reliability for DG6 during exercise with 99% of all values located in the clinically safe zones A+B (33). Dyess et al. found that individuals who wore DG6 on their buttocks during exercise had an increase of values in zone D while an increase of values in zones A+B was seen when wearing the sensor on the abdomen (34). For FSL1, Giani found 97% of sensor readings fell in zones A+B during rest and 98% during exercise indicating a marginal difference between the two settings (37). Moser et al. found 91% of sensor values were in zones A+B during rest while 78% for FSL1 during exercise indicating a decrease in the clinical accuracy of the sensor during exercise (35). Throughout all the studies where exercise negatively impacted the sensor performance, sensor values specifically increased in the upper zone D similarly to our study indicating an increase in failure to detect hypoglycemia.

The strengths of the present study include the cross-over design, the direct comparison of exercise versus rest and the conduction in a controlled clinical research facility. Sensors were initialized two days prior to the clamp day to prevent possible sensor inaccuracies. Unlike other studies that used capillary blood glucose, our study utilized PG measured by YSI 2900 as a glucose reference method which is often cited as the gold standard (39). Furthermore, the sensors were not user-calibrated but relied on the factory calibration mimicking a real-life setting with individuals doing the same.

A limitation to our study is the small number of participants and that the study only included male adults thus limiting the generalizability to females and other age groups. One participant had their FSL1 placed on the upper thigh which could potentially influence on the results, although Charleer et al. showed a minimal difference between the placement of FSL1 on the upper thigh and the upper arm (38). Another limitation is the rather high MARD observed for Dexcom G6 which could influence the results and that MARD does not take sensor errors into account (e.g., consistent higher glucose estimates). To overcome this, an adjunctive analysis called precision absolute relative deviation, which requires the insertion of an identical parallel sensor, could potentially have added value to our study (40). A limitation of Clarke Error grid analysis is the rather stringent limits between the zones which newer error grids seek to mitigate (41). Finally, the physiological effect of physical activity on glucose levels is rather complex and different at differing exercise types, intensities, and durations. Thus, the observed performance of the sensors cannot be generalized to other exercise intensities or durations.

In conclusion, the two commonly used sensors DG6 and FSL1 showed different responses to exercise in relation to PG decline and hypoglycemia in individuals with type 1 diabetes. Exercise negatively impacted FSL1 sensor performance during both declining PG and hypoglycemia, whereas DG6 had more accurate sensor readings during exercise and post-exercise. Individuals with type 1 diabetes and healthcare practitioners should be aware of the potentially negative or positive impacts of exercise on CGM sensor accuracy in detecting clinically relevant episodes of hypoglycemia.

Parts of the following study were presented as an oral presentation at the European Association of the Study of Diabetes (EASD) 58th annual meeting in Stockholm, Sweden, September 2022

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the Scientific Ethical Committee of the Capital Region of Denmark. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

KM: Data curation, Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Project administration, Investigation, Formal analysis. PH: Writing – review & editing, Supervision, Resources, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. SE: Writing – review & editing, Validation, Resources, Methodology. JF: Formal analysis, Software, Supervision, Validation, Writing – review & editing. UP-b: Writing – review & editing, Methodology, Validation, Supervision, Conceptualization. FK: Conceptualization, Methodology, Resources, Supervision, Validation, Writing – review & editing. TV: Writing – review & editing, Validation, Supervision, Project administration, Funding acquisition, Conceptualization. AA: Conceptualization, Formal analysis, Methodology, Supervision, Validation, Writing – review & editing.