Decision accuracy of simultaneously used real‐time CGM versus intermittently scanned CGM around exercise in type 1 diabetes: A secondary analysis of the ULTRAFLEXI‐1 study

Abstract

Aims

Continuous glucose monitoring (CGM) systems have become important technologies to improve glycaemia in people with type 1 diabetes (T1D). However, it has been shown that during rapid glucose change, sensor performance can deteriorate. Comparative data on sensor performance during high rates of glucose change, such as during exercise, between a real‐time continuous glucose monitor (rtCGM) and an intermittently scanned continuous monitor (isCGM) remain limited.

Methods

Twenty‐two people with T1D (8 women, age 42 ± 11 years, HbA_1c 59 ± 8 mmol/mol (7.6 ± 0.8%)) simultaneously used an rtCGM (Dexcom G6) and an isCGM (Freestyle Libre 1). Sixty‐minute exercise sessions were performed on a cycle ergometer at moderate intensity, and glucose values from both CGM systems were compared against capillary reference blood glucose measurements (EKF S‐Line; EKF Diagnostics, Germany). Data were assessed using the Median Absolute Relative Difference (MedARD) with interquartile range, as well as the Diabetes Technology Society Error Grid (DTS EG).

Results

During exercise, the MedARD was 14.6% [7.0;23.8] for rtCGM (2304 comparison points) vs. 11.6% [5.6;19.6] for isCGM (2266 comparison points) (p < 0.0001). When stratified by glycaemic range, the MedARD was 39.2% [31.8;46.8] vs. 27.0% [17.0;34.6] for time below range (<70 mg/dL) (p = 0.0001), 16.1% [8.1;24.8] vs. 12.8% [6.4;20.4] for time in range (70–180 mg/dL) (p < 0.0001) and 9.5% [4.7;16.0] vs. 8.0% [3.8;13.7] for time above range (>180 mg/dL) (p = 0.0064) for rtCGM vs. isCGM.

Conclusion

In this head‐to‐head comparison of rtCGM and isCGM, isCGM demonstrated superior performance during exercise in adults with T1D.

What's new?

What is already known?

• CGM accuracy decreases during exercise due to rapid glucose changes and due to higher lag time; standardised, rate‐of‐change–aware assessment is recommended, but head‐to‐head rtCGM vs. isCGM data during exercise that follow the latest guidelines are scarce.

What this study has found?

• In 22 adults, isCGM (Libre 1) showed lower MedARD than rtCGM (Dexcom G6) during exercise.

What are the implications of the study?

• This analysis supports the relevance of systematic head‐to‐head comparison of CGM systems under exercise conditions.

1. INTRODUCTION

Continuous glucose monitoring (CGM) systems represent an integral therapy management component for individuals with type 1 diabetes (T1D). ¹ Regular CGM use has been shown to improve glycaemic control by reducing HbA_1c, time below range (TBR; <70 mg/dL), time above range (TAR >180 mg/dL), glycaemic variability and increasing time in range (TIR; 70–180 mg/dL) and has also been linked to improved quality of life. ² Considering that people with T1D base their therapy decisions on sensor glucose readings, sensor performance should be as accurate as possible to avoid sensor inaccuracy‐induced dysglycaemia. ³ However, it is well documented that during rapid glucose change, like during physical activity and exercise, sensor performance is challenged, which might increase the risk of hypoglycaemia. ⁴ As CGM systems assess glucose levels in the interstitial fluid, there is a physiological delay before the glucose diffuses into the blood vessels or from the blood vessels into the interstitial space. ⁵ , ⁶ , ⁷ Therefore, blood glucose concentrations and interstitial fluid concentrations measured, for example during moderate‐intensity exercise, face a lag time of around 12 min. ⁸ The accelerated rate of glucose change (RoC) due to exercise in combination with the physiological lag time between the blood and the interstitial space poses a challenge for glycaemic management during exercise in people with T1D. ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ Which is why especially decision accuracy (displayed rtCGM values) is of particular importance when compared to analytical accuracy (accuracy of downloaded data that is smoothed) ¹¹ when it comes to exercise, since therapy decisions are based on these values.

In general, two types of CGM systems are known: real‐time CGM (rtCGM) and intermittently scanned CGM (isCGM). While rtCGM provides glucose values continuously, isCGM requires a scan to display the actual glucose value, the accompanied trend arrow and the retrospective glucose profile. ⁵

In order to harmonise CGM performance assessment across studies, a recent review by Eichenlaub et al. ¹² proposed that in addition to the distribution, comparator datasets should include RoC values in order to achieve a more accurate reflection of physiologically dynamic states. Despite these developments in methodological guidance, studies that have assessed the performance of different CGM systems in parallel, i. e., in the same individuals under identical conditions, remain scarce. ¹³ This methodological limitation is particularly present in the field of exercise, where robust in‐parallel comparison of rtCGM and isCGM is still insufficiently represented. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷

Therefore, in this secondary analysis of the ULTRAFLEXI‐1 trial, we compared the performance of the Dexcom G6 rtCGM system (Dexcom Inc., San Diego, CA) and the Freestyle Libre 1 isCGM system (Abbott, USA) during 60‐min moderate‐intensity ergometer cycling sessions, conducted three times per week, over a period of 8 weeks, in adults with T1D using previously established and newly suggested CGM performance measures.

2. METHODS

This is a pre‐registered analysis of the ULTRAFLEXI‐1 study (German Clinical Trial Register, DRKS; DRKS00018065), ¹⁸ , ¹⁹ which was a randomised, single‐centre, four‐period, crossover clinical trial that investigated the effects of insulin glargine U‐300 and insulin degludec U‐100 on glycaemia in adults with T1D. In total, each participant completed 24 exercise sessions over the course of 8 weeks. Each trial arm consisted of six exercise sessions conducted over a 2‐week period. In total, 25 people were randomised in this trial; in this exploratory analysis, we included 22 participants who used both rtCGM and isCGM in parallel regardless of the randomisation sequence. The local ethics committee of the Medical University of Graz (31–551 ex 18/19) approved the study protocol. The ULTRAFLEXI‐1 study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. Prior to participation, potential participants were informed about the study protocol and provided written informed consent.

2.1. Eligibility criteria

In the ULTRAFLEXI‐1 study, men and women living with T1D, clinically for at least 12 months, aged 18–65 years with a body mass index of 18.0–29.9 kg/m, ² HbA1c ≤10%, and treated with multiple daily insulin injections for at least 12 months were eligible for inclusion. Additional inclusion criteria were a C‐peptide level ≤0.3 nmol/L and a mass‐specific peak oxygen consumption (VO_2peak) >20 mL/kg/min. Key exclusion criteria included recurrent severe hypoglycaemia (defined as more than one severe hypoglycaemic event during the past 12 months), hypoglycaemia unawareness was assessed by the investigator and hospitalisation for diabetic ketoacidosis within the previous 12 months.

2.2. Study design

At the screening visit, after eligibility had been confirmed and written informed consent obtained, a maximum incremental cardiopulmonary exercise (CPX) test was performed on a cycle ergometer to determine VO_2peak, as well as the first (LTP₁) and second lactate turn points (LTP₂), which were used to define the target intensity for the subsequent exercise sessions.

All participants used both an rtCGM (Dexcom G6) and an isCGM (Freestyle Libre 1; Freestyle Libre 2 was not available when the study was conducted), with sensors inserted on the back of the upper arm. Participants could choose which CGM system to wear on which arm but were not allowed to wear both on the same arm. Throughout the study period, participants completed 60‐min moderate‐intensity cycle ergometer exercise sessions in the afternoon, with the exercise intensity set at the midpoint between LTP₁ and LTP₂ (60%–65% of VO_2peak). Capillary blood samples were collected from the earlobe to assess blood glucose (reference measurement) and lactate concentrations (20 μL capillary blood; EKF S‐Line; EKF Diagnostics, Germany). The samples were taken from the earlobe to get consistent and practical sampling during exercise. ²⁰ Samples and CGM value documentation were obtained at the following time points: 3 min before exercising (resting), after a 3‐min warm‐up, every 6 min during exercise and 3 min post‐exercise (resting). With every measurement point, the isCGM has been scanned and the displayed values for both systems were written down. In order to provide more robustness due to temporal interdependence between paired points, sampling has been reduced to 12‐min intervals for this analysis to align the data to current recommendations ¹² , ²¹ , ²² and to lessen autocorrelation while retaining statistical power. With respect to the accuracy of the reference method, the EKF system was chosen as it demonstrated comparable performance to the YSI reference standard, supporting its suitability for CGM performance evaluations. ²³ Interstitial glucose concentrations were measured by means of Dexcom G6 (Dexcom G6, Dexcom, USA) and the Freestyle Libre 1 (Abbott, USA). Throughout all exercise sessions, therapy recommendations were followed according to the joint position statement of the European Association for the Study of Diabetes and the International Society for Paediatric and Adolescent Diabetes. ⁶ , ⁷

2.3. Pre‐exercise standardisation

The ULTRAFLEXI trial was a randomised, single‐centre, four‐period, crossover trial that compared basal insulin Glargine 300 U/mL and insulin Degludec 100 U/mL for time below range <70 mg/dL in two different doses (100% and 75% of the regular dose) when used around spontaneous exercise sessions in adults with T1D. Each participant was allocated to four trial arms of 6 days of evening‐exercise sessions over a period of 2 weeks (3 sessions randomly assigned within 7 days). Participants received a phone call from the research team at 08:00 that contained information to inject a regular basal insulin dose (100%) or 75% of the regular basal insulin dose at 10:00 and that an exercise session will be performed at 18:00. For the following exercise days within the same trial arm (14 days), the same basal insulin was used with the same basal dose (100% or 75% of regular basal insulin dose). On non‐exercise days (within the 2‐week exercise period), participants stayed on the same type of basal insulin, administering the regular basal insulin dose (100%). Additionally, participants followed a structured therapy regime that is replicated for each exercise day within each trial arm: basal dose injection at 10:00, lunch with 1 g.kg‐1 carbohydrates (CHO), regular bolus insulin dose between 12:00–14:00, small brunch (~15–30 grams CHO), regular bolus insulin dose between 15:00–16:00, dinner with 1 g.kg‐1 CHO, regular or up to 25% reduced bolus insulin dose (based on the participants' preference) between 20:00–21:00. Participants were asked to eat the same type and number of meals at the same time point on the days of exercise.

As suggested in the 2020 EASD/ISPAD position statement on CGM use during physical activity and exercise, ⁶ , ⁷ exercise testing was only started when blood glucose concentration was above 126 mg/dL 15 min prior to the start. When the blood glucose concentration was below 126 mg/dL, 15–30 g of CHO (gel or juice) were given. Blood glucose was measured again after 15 min, and in case blood glucose was still below 126 mg/dL, this procedure was repeated as often as needed. If the blood glucose concentration was above 270 mg/dL, a blood ketone measurement was performed. If blood ketone levels were below 1.5 mmol/L, the exercise testing was allowed to start. If ketones >1.5 mmol/L were detected, then this exercise session was cancelled and not repeated. When blood glucose was below 70 mg/dL, a hypoglycaemic episode was reported, and oral carbohydrates were provided and recorded (type, amount and time). A new blood glucose value was measured approximately 10 min after the carbohydrate administration. If the next blood glucose value was still below 70 mg/dL, the carbohydrate administration and blood glucose measurement procedure were repeated.

2.4. Statistical analyses

Data were analysed via Prism Software version 8.0 (GraphPad, La Jolla, CA, USA). After testing for normal distribution via Shapiro–Wilk normality test, continuous glucose monitoring values were compared against reference blood glucose measurements (EKF S‐Line; EKF Diagnostics, GER) for median absolute relative difference (MedARD) analysis [interquartile range] and the Bland–Altman method (bias and 95% limits of agreement (LoA)) were performed to compare glucose in exercise conditions as well as for glycaemic ranges (TIR, TBR and TAR) of rtCGM and isCGM. An unpaired t‐test or Mann–Whitney test were performed for absolute relative difference (|ARD|) comparison of the two CGM systems. Additionally, comparator data distribution analysis by means of the dynamic glucose region plot (DGRP) was performed as proposed by Eichenlaub et al. ¹² using Python (Version 3.12). The Diabetes Technology Society Software was used for the Diabetes Technology Society Error Grid (DTS EG) as proposed by Klonoff et al. ²⁴

3. RESULTS

A total of 22 participants were included in this analysis and characteristics are presented in Table 1.

TABLE 1.

Participant characteristics.

Characteristic	Value
Men, N (%)	14 (63.6)
Women, N (%)	8 (36.4)
Age (years) a	42 ± 11
BMI (kg/m2) a	23.6 ± 3.2
HbA1c (mmol/mol) a	59 ± 8
HbA1c (%) a	7.6 ± 0.8
Diabetes duration (years) a	17 ± 11
Total daily insulin dose (units)	36 [28.8;49.3]
Total daily basal dose (units)	18 [15.0;27.0]
Total daily bolus dose (units)	18 [14.5;24.3]
VO2max (mL/min/kg)	35 [28.8;42.8]

^a Results are given as mean ± SD.

3.1. CGM‐derived metrics

A descriptive representation of the overall CGM‐derived metrics and frequencies during the exercise sessions for the Freestyle Libre 1 and Dexcom G6 can be found in Tables 2 and 3.

TABLE 2.

CGM‐derived metrics for Freestyle Libre 1 and Dexcom G6.

	Mean glucose (mg/dL)	SD	Median [IQR]	CV %	% <54	% <70	% 70–180	% >180	% >250
Dexcom G6	155.5	48.6	147.7 [120.7;183.8]	31.3	0.0	0.4	72.6	27.0	5.3
Freestyle Libre 1	150.9	52.6	143.0 [112.0;181.0]	34.9	0.2	1.6	73.3	25.1	5.3

Abbreviations: CV, coefficient of variation; SD, standard deviation.

TABLE 3.

Median absolute relative difference (MedARD) and interquartile range (IQR) between interstitial glucose (rtCGM and isCGM) and reference BG for overall, exercise and resting conditions and for glycaemic ranges and p‐values for differences in absolute relative difference (|ARD|) comparison.

Area	rtCGM system	isCGM system	p‐value
	MedARD, %	MedARD, %
	[IQR]; (n)	[IQR]; (n)
Overall	14.6% [7.2;23.8] (2304)	11.6% [5.6;19.6] (2266)	<0.0001
In range (70–180 mg/dL)	16.1% [8.1;24.8] (1799)	12.8% [6.4;20.4] (1763)	<0.0001
Below range (<70 mg/dL)	39.2% [31.8;46.8] (35)	27.0% [17.0;34.6] (33)	0.0001
Above range (>180 mg/dL)	9.5% [4.7;16.0] (470)	8.0% [3.8;13.7] (470)	0.0064

3.2. Median absolute relative difference (MedARD)

During exercise, 2304 comparison points were available for rtCGM and 2266 for the isCGM, based on simultaneous reference blood glucose measurements and deleting failures. Significant differences in absolute relative difference (|ARD|) were observed under overall in‐exercise conditions as well as when glucose levels were within range (70–180 mg/dL) and below range (<70 mg/dL) and above range (>180 mg/dL).

3.3. Bland–Altman analyses

In exercise, the mean difference (bias) was 16.34 ± 28.85 (95% CI: −40.20 to 72.88) for the Dexcom G6 and 12.01 ± 24.6 (95% CI: −36.21 to 60.23) for Freestyle Libre 1, showing higher glucose values in comparison to the reference method. Bland–Altman analysis for Dexcom G6 and Freestyle Libre 1, including bias ± SD of bias and 95% levels of agreement for glucose values (rtCGM to blood glucose (BG) and isCGM to BG) is presented in Figure 1.

FIGURE 1.

Bland–Altman analysis for Freestyle Libre 1; (a) overall, (b) in range (70–180 mg/dL), (c) below range (<70 mg/dL) and (d) above range (>180 mg/dL) and the Dexcom G6; (e) overall, (f) in range (70–180 mg/dL), (g) below range (<70 mg/dL) and (h) above range (>180 mg/dL). BG, blood glucose; isCGM, intermittently scanned continuous glucose monitoring.

3.4. Comparator data distribution

As proposed by Eichenlaub et al., ¹² BG and RoC measurement pair distribution during exercise are shown by means of the dynamic glucose region plot (DGRP). In total, 1844 blood glucose and RoC pairs were found where all three measurements, rtCGM, isCGM and reference blood glucose, were available. In this analysis, the first data point had to be discarded (no previous value to calculate the rate of change), which is why the size per shift is reduced, resulting in fewer values available per session and overall. Recommended percentages for the regions (≥7.5% for each: BG values <70 mg/dL, >300 mg/dL and BG–RoC combinations, indicating rapid changes toward hypo‐ or hyperglycaemia) and percentages achieved are presented in Figure 2.

FIGURE 2.

Dynamic glucose region pPlot for BG and RoC pairs for all time points available at the same time for Dexcom G6, Freestyle Libre 1 and reference BG measurements. BG, blood glucose; MARoC, mean absolute RoC; RoC, rate of change.

3.5. Diabetes technology society error grid (DTS error grid)

With regard to risk zones (Zone A: no clinical risk, Zone B: slight risk, Zone C: moderate risk, Zone D: high risk, Zone E: extreme risk) based on the Diabetes Technology Society Error Grid (DTS EG), ²⁴ the following frequencies (counts) were observed for the Dexcom G6 during exercise: Zone A: 59.6% (n = 1373), Zone B: 38.5% (n = 887), Zone C: 1.9% (43), Zone D: 0.0% (1) and Zone E: 0.0% (0). The following frequencies (counts) were found for Freestyle Libre 1 during exercise for the different zones: Zone A: 70.6% (n = 1600), Zone B: 28.8% (n = 652), Zone C: 0.6% (13), Zone D: 0% (1) and Zone E: 0% (0) (Figure 3).

FIGURE 3.

Diabetes Technology Society Error Grid (DTS EG) ²⁴ for (a) Dexcom G6 and (b) Freestyle Libre 1 in exercise measurements comparing CGM values and capillary blood.

4. DISCUSSION

This analysis of the ULTRAFLEXI‐1 trial demonstrated that the performance of the Dexcom G6 was inferior to that of the Freestyle Libre 1 during 60‐min moderate‐intensity exercise sessions on a cycle ergometer. These findings are of particular importance as therapy decisions are based on those readings people with T1D see in that moment (decision accuracy ¹¹ ) versus the data that gets analysed by the diabetes teams/health care professionals (retrospective ‘smoothed’ data; analytical accuracy ¹¹ ) for CGM metrics and therapy assessment.

Considering the overestimation observed in both rtCGM and isCGM compared to reference blood glucose values as shown in the Bland–Altman analysis, the width of the LoA during exercise, and the deterioration of decision accuracy during exercise underline the importance of considering trend arrows during exercise and confirmatory blood glucose measurements during exercise. This also reinforces the guidance and position statements by Moser et al. (2020, 2024) ⁶ , ⁷ , ²⁵ , ²⁶ on CGM and automated insulin delivery (AID) use during exercise. It gives recommendations for therapy adjustments (carbohydrates and insulin) but also suggests raising glucose alarm thresholds to reduce hypoglycaemia risk.

When looking at the Diabetes Technology Society Error Grid (DTS EG), ²⁴ both systems showed low rates of clinically meaningful error (Zone C ≤ 2%) indicating that the probability of incorrect action was minimal. This means that, in this case, analytical superiority for MedARD does not automatically imply superior safety or decision quality.

In a previous study, ²⁷ we analysed the performance of the Freestyle Libre 1 and found similar results for the 19 participants included: accuracy deteriorated during 45 min moderate‐intensity exercise sessions (470 measurement points; MARD: 29.8%), particularly when glucose levels were low (<70 mg/dL; MARD: 45.1%), while accuracy improved during hyperglycaemia (>180 mg/dL; MARD: 16.3%). A study by Lundemose et al., ¹⁷ that compared the Dexcom G6, Guardian 4 and the Freestyle Libre 2 during 60 min moderate‐intensity ergometer cycle sessions and during home use found a MARD of 12.6% for the Dexcom G6 (79 measurement points), 10.7% for the Guardian 4 (73 measurement points) and 17.2% for the Freestyle Libre 2 (31 measurement points) (p = 0.31) during exercise. This analysis evaluated the MARD by pairing the BG with its closest CGM value (analytical accuracy ¹¹ ). In contrast to our data, this comparison showed a higher accuracy with the rtCGM compared to the isCGM, but the comparison showed no significant differences. When assessed in the home phase, the Freestyle Libre 2 showed significantly higher MARD when compared to the other systems (Dexcom G6: 10.2%, Guardian 4: 11.9%, Freestyle Libre 2: 16.3%, p = 0.02). However, the use of the Freestyle Libre 2 in this study—the successor model to the Freestyle Libre 1 applied in our study—limits comparability. Aberer et al., ¹⁴ analysed the performance of the Freestyle Libre 1, the Dexcom G4 Platinum and Medtronic Enlite CGM in 12 people with T1D under simulated daily‐life conditions (including food intake, exercise and under hypo‐ and hyperglycaemic conditions). Across all conditions MARDs were 13.2% for Freestyle Libre 1 (462 measurement points), 16.8% for Dexcom G4 (540 measurement points) and 21.4% for the Enlite CGM (502 measurement points). During the two 15 min moderate‐intensity exercise sessions separated by 5 min of resting, the MARDs were 8.7% for Freestyle Libre 1 (13 measurement points), 15.7% for Dexcom G4 (24 measurement points) and 19.4% for the Enlite CGM (22 measurement points), based on paired sensor‐reference glucose values, indicating highest accuracy for the Freestyle Libre 1. Importantly, plasma glucose was used as the reference standard in this study. Cuerda Del Pino ¹⁵ investigated the accuracy of the Dexcom G6 and Freestyle Libre 2 during 30 min aerobic exercise sessions and 11 min high intensity interval training (HIIT) and observed a MARD of 31.98% for the Dexcom G6 and 23.21% for the Freestyle Libre 2 during aerobic exercise and a MARD of 14.03% for the Dexcom G6 and 21.09% for the Freestyle Libre 2 during HIIT. In this study, the Freestyle Libre 2 demonstrated significantly higher accuracy than the Dexcom G6 during aerobic exercise (p = 0.0129), whereas the Dexcom G6 demonstrated better accuracy than the Freestyle Libre 2 during HIIT (p = 0.0019). This result aligns with our findings during aerobic exercise but not during HIIT, although again, a successor model was used and plasma glucose served as the references, whereas our study relied on capillary blood. However, with regard to comparability, it should be noted that some studies discussed do not specifically mention whether decision accuracy or analytical accuracy ¹¹ has been evaluated, which limits comparability. Overall, comparing CGM decision accuracy across studies remains challenging, due to the lack of standardised protocols, as previously emphasised by Eichenlaub et al. ¹²

In contrast to the aforementioned studies and to the best of our knowledge, this is the largest analysis to date evaluating the decision accuracy of rtCGM and isCGM use during exercise in parallel setup, allowing for direct comparison under standardised conditions. With over 2000 measurement data pairs for each system collected during the 60 min moderate‐intensity exercise sessions on a cycle ergometer, this is the biggest CGM decision accuracy analysis conducted when comparing rtCGM and isCGM under exercise in‐lab conditions.

When looking at the dynamic glucose region plot (DGRP) data distribution recommendations by Eichenlaub et al., ¹² only the ‘Alert low’ recommendation was met but not for the categories ‘Neutral’, ‘Alert high’, ‘BG high’ and ‘BG low’. This is due to the specific focus on decision accuracy during exercise conditions and its physiological nature, hence distributions as the full range of glycaemic suggestions were not able to be met. It should also be noted that this is a secondary analysis of the ULTRAFLEXI‐1 study, which was designed to investigate two different basal insulins and was not primarily intended to examine the measurement decision accuracy of CGM systems during exercise. This is also why the EKF system was used for reference measurements, although it is well known that YSI is the frequently used choice and represents the standard reference system. ²⁸

Another limitation of this study is that the isCGM used (Freestyle Libre 1) is no longer commercially available in several countries, which is why the clinical relevance is limited here for isCGM. And even though the Freestyle Libre 1 showed favourable performance, these findings are limited by the fact that intermittent scanning requires active user action, which cannot be accomplished during every kind of exercise and physical activity. Additionally, since there is no direct translation to Freestyle Libre 2 or Freestyle Libre 3, results should be interpreted with caution. Methodically it also has to be noted that there is a difference in frequency and synchronisation of the two systems as the Dexcom G6 provides values every 5 min whereas the Freestyle Libre 1 required scanning for every measurement point, which is why the time points are not perfectly aligned and introduces potential bias favouring Freestyle Libre 1. Also, it has to be noted that these findings only apply for this specific kind of exercise and under in‐lab conditions. It shows that there is still a need for a standardised approach, so procedures and numbers are more comparable and why guidelines like Eichenlaub's approach ¹² are of particular relevance. This also underlines the need to include accuracy analyses under different conditions like different kind of exercise sessions in lab studies but also in real‐world studies.

5. CONCLUSION

The Dexcom G6 rtCGM system demonstrated lower decision accuracy compared to the Freestyle Libre 1 isCGM system during three moderate‐intensity exercise sessions per week over an 8‐week period in the 22 tested adults with T1D.

AUTHOR CONTRIBUTIONS

SS wrote the original draft, HS and OM conceived the trial, AM, CSourij, ULS, PNP, NT, FAberer and HK performed the study/investigation and collected the data. All authors revised and edited the manuscript, contributed to the discussion and approved the final version.

FUNDING INFORMATION

This study was funded by Sanofi as an IIT. Dexcom Inc. provided the rtCGM.

CONFLICT OF INTEREST STATEMENT

SS has received a doctoral scholarship from the German National Academic Foundation (Studienstiftung des deutschen Volkes), as well as research and conference travel support from Ypsomed. FA received speaker honoraria from AstraZeneca, Bayer, Sanofi, Novo Nordisk, Eli Lilly, AMGEN, MSD and Boehringer Ingelheim and represents an advisory board member of Sanofi, Bayer, Eli Lilly and Novo Nordisk. HS has received consulting fees and speaker's honoraria from Amgen, Amarin, Boehringer Ingelheim, NovoNordisk, Novartis, Daiichi Sankyo, Eli Lilly and Cancom. He has received research funding from Boehringer Ingelheim, Novo Nordisk, Eli Lilly (paid to the Medical University of Graz, Austria) and has received lecture fees from Medtronic, Eli Lilly, Novo Nordisk, Sanofi, TAD Pharma, Theras, Diatec, Ypsomed, Dexcom, AstraZeneca, Insulet and Perfood; he is on the advisory board for Sanofi, TAD Pharma, Glaice, Perfood, Medtronic and Dexcom. OM reports the following conflict of interest: Clinical trial support: Sêr Cymru II COFUND Fellowship/European Union, Novo Nordisk A/S, Novo Nordisk AT, Abbott Diabetes Care, Sanofi, Dexcom, Team Novo Nordisk, SAIL, Maisels Brauerei, Medtronic AT, EFSD/EASD, Falke, BISp, perfood, Ypsomed, Sinocare. Presenters' honoraria: Medtronic AT, Medtronic Int., Eli Lilly, Novo Nordisk, Sanofi, TAD Pharma, ADA, Diatec, Berufsverband deutscher Internist*innen, Dexcom, Astra Zeneca, Ypsomed, Insulet, Diabetologen Hessen, Abbott. Conference travel support: Novo Nordisk A/S, Novo Nordisk AT, Novo Nordisk UK, Medtronic AT, Sanofi, EASD, OEDG, DDG. Advisory board: Sanofi, TAD Pharma, Dexcom, Perfood, Medtronic. The remaining authors declare no conflicts of interest. All other authors declare no conflict of interest related to this study.