CGM for Athletes: What the Data Can and Cannot Tell You

Oak
cgm glucose athletes wearables data-analysis
Long-legged glowing herd animals running through a misty valley, each with an amber shoulder organ leaving lagging light trails.

If you are an athlete without diabetes and you want a continuous glucose monitor (CGM), wear a Dexcom G7 or an Abbott FreeStyle Libre 3, place it on the back of the upper arm, and treat the first 24 hours of each sensor as discardable. Both are factory-calibrated, both report at short intervals (five minutes for G7, one minute for Libre 3), and both export clean CSV. The more important answer is what to do with the trace: CGM accuracy degrades during hard exercise, and day-to-day glucose patterns in athletes are less reproducible than most people assume, so any conclusion you draw from a single session or a single day is probably an artifact. The useful unit of analysis is a repeated, controlled comparison across several weeks, not a screenshot of one curve.

What the sensor measures, and why that matters during exercise

A CGM does not measure blood glucose. A filament sitting in subcutaneous interstitial fluid generates a current proportional to glucose concentration via glucose oxidase, and firmware converts that current to a displayed value using a factory calibration curve plus smoothing and lag-compensation. Two consequences follow. First, interstitial glucose trails plasma glucose, with the gap widening when glucose is changing fast, which is exactly the situation during a glycemic ramp after a gel or during a hypoglycemic drop. Second, the smoothing that makes a resting trace look tidy is a low-pass filter, and it will blunt and delay real excursions.

Exercise stresses every one of these assumptions. Comparisons against laboratory analyzers show that CGM error in athletes is larger during exercise than at rest, with systematic bias in addition to random noise 1. Accuracy also varies with intensity: agreement with venous reference degrades as effort and blood lactate rise, so the sessions you most want to instrument are the ones where the sensor is least trustworthy 2. Earlier work in athletic populations found that CGM performance was materially worse than the accuracy claims derived from resting clinical studies 3. Practically, this means you should not read a 15 mg/dL difference between two interval sessions as a real metabolic signal.

The reproducibility problem is separate and, for training decisions, larger. When endurance athletes wore CGMs across repeated days, day-to-day glycemic metrics varied substantially even without any deliberate intervention 4. If your within-person, between-day variance is that wide, a single A/B comparison of two breakfasts has essentially no power. You need replication.

Choosing a device

Dexcom G7 and FreeStyle Libre 3 are the two sensible choices for a non-diabetic athlete, and the differences that matter are about data plumbing rather than sensor chemistry.

The G7 has a 10-day wear period plus a 12-hour grace period, transmits over Bluetooth Low Energy continuously, and exports through Dexcom Clarity as a CSV with five-minute values. It also supports optional fingerstick calibration, which is useful if you have a reliable meter and want to correct a sensor with visible offset. The Libre 3 has a 14-day wear period and one-minute resolution, which gives you five times the sample density during a workout, but it cannot be calibrated, so a biased sensor stays biased for two weeks. For alignment with power files, one-minute data is genuinely better. For long-run stability, calibratability is worth more.

Eversense 365 is an implanted sensor with a one-year wear period, inserted and removed by a clinician in an office procedure. It solves adhesion and knocked-off-sensor problems completely, which is attractive if you swim, wrestle, or play contact sports. It costs more than disposable sensors and pricing depends heavily on insurance and provider, so get a written quote that includes both the sensor and the insertion and removal procedures before committing. Supersapiens, which layered athlete-facing analytics on Abbott’s Libre Sense, shut down its consumer service, so plan on building your own analysis rather than depending on a sports-specific app.

Site and mechanics matter more than people expect. The back of the upper arm is the validated site for both sensors. Use an overpatch if you sweat heavily or wear a wetsuit. Compression artifacts are real: lying on the sensor, or resting a barbell or a foam roller against it, transiently restricts local perfusion and produces a false drop that recovers within minutes once pressure is released. Overnight compression lows are the single most common source of spurious “nocturnal hypoglycemia” in otherwise healthy people.

Getting the raw data out

Dexcom Clarity exports a CSV in which the columns you care about are Timestamp (YYYY-MM-DDThh:mm:ss), Event Type, and Glucose Value (mg/dL). Two traps: values below 40 and above 400 appear as the strings Low and High, and the file interleaves calibration and event rows with sensor readings. Filter on Event Type == "EGV", coerce the glucose column with pd.to_numeric(errors="coerce") after mapping Low to 39 and High to 401, and localize timestamps to the timezone the receiver was in, because Clarity writes local wall time with no offset.

LibreView exports a similar file with a Record Type column where 0 marks automatically recorded historic readings and 1 marks scan values. Keep type 0 for time-series work. Units follow your account region, so check whether you are reading mg/dL or mmol/L before you compute anything.

For analysis, resample everything onto a common 5-minute grid, linearly interpolate gaps shorter than 15 minutes, and leave longer gaps as NaN rather than filling them. Then merge with training data. Garmin, Wahoo, and most head units write .fit files that fitparse or fitdecode will unpack into per-second records with power, heart rate, and cadence. Resample those to the same 5-minute grid with means, and join with pandas.merge_asof(tolerance=pd.Timedelta("2min"), direction="nearest"). Log meals and their carbohydrate content with timestamps in a separate table, because a glucose curve without an annotated input is uninterpretable.

The metrics worth computing are mean, standard deviation, coefficient of variation (SD divided by mean, as a percentage), percent of time in a defined band, and incremental area under the curve over a fixed post-meal window using the trapezoid rule with the pre-meal value as baseline. Use 2-hour or 3-hour iAUC windows consistently. MAGE and similar excursion indices are sensitive to the smoothing you apply first, so if you report one, report the smoothing parameters too.

Designing comparisons that survive the noise

Given the measurement error during exercise and the day-to-day variance, the only designs that produce interpretable answers are repeated and randomized. Pick one question, for example whether 60 g/h versus 90 g/h of carbohydrate changes the glucose trajectory across a two-hour steady-state ride at a fixed power. Run each condition at least three times, alternating in a randomized order, at the same time of day, on the same sensor where possible, with the previous evening’s meal held constant. Compare the distribution of the per-session summary statistic, not the individual curves.

This is also where the limits of the field sit. Reviews of CGM in sport agree that the tool is promising for understanding individual fueling responses and carbohydrate availability, and that it has no validated performance target and no evidence yet that trace-guided fueling improves outcomes in athletes without diabetes 56. Physiological analyses reach the same conclusion: glucose is one downstream signal among many, and low readings during prolonged exercise often reflect normal hepatic and muscular flux rather than a fueling failure 78. Use the data to characterize your own responses, not to chase a number.

If you see repeated fasting values you did not expect, symptomatic lows, or a pattern that concerns you, that is a conversation with a physician, who can order a venous glucose and HbA1c. A consumer CGM is not a diagnostic device and nothing in your export should be used to start, stop, or change any medication.

Questions people also ask

Can you lift weights wearing a CGM?

Yes. Place the sensor on the back of the non-dominant upper arm and use an overpatch. Expect false dips whenever a bar, dumbbell, or bench presses against the site, and expect a genuine rise in glucose during and after heavy resistance work driven by catecholamine-mediated hepatic glucose output.

Does your blood sugar go up when you lift weights?

Commonly, yes. Short, intense efforts raise glucose because hepatic output exceeds muscle uptake during the session. This is a normal acute response in people without diabetes and it does not indicate a problem.

What is the optimal blood glucose level for an athlete?

There is no validated optimal value or target range for athletes without diabetes, and reviews of the literature say so explicitly 59. Treat your own repeated baseline as the reference and look for changes in it over time rather than comparing yourself to a published band.

Is 140 mg/dL in the morning high?

A single CGM morning reading is weak evidence of anything, because sensor error, compression artifact, and normal dawn cortisol and growth hormone dynamics all contribute. A persistent pattern across many mornings on multiple sensors is worth discussing with a clinician, who can confirm with a venous fasting glucose and HbA1c.

What CGM is better than Dexcom?

For athletes, the FreeStyle Libre 3 is the main alternative, and its advantage is one-minute resolution and 14-day wear. Its disadvantage is that it cannot be user-calibrated. Eversense 365 is the choice if sensor retention during contact or water sports is your binding constraint.

Do athletes have higher HbA1c?

HbA1c reflects red blood cell lifespan as well as average glucose, and endurance training can alter erythrocyte turnover, so HbA1c and CGM-derived mean glucose sometimes disagree in trained people 7. If the two conflict, that is a reason to ask a physician for interpretation rather than to pick whichever number you prefer.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Helen Bauhaus, Pinar Erdogan, Hans Braun, et al. Continuous Glucose Monitoring (CGM) in Sports—A Comparison between a CGM Device and Lab-Based Glucose Analyser under Resting and Exercising Conditions in Athletes. International Journal of Environmental Research and Public Health, 2023. https://doi.org/10.3390/ijerph20156440 ↩

  2. Kristina Skroce, Lauren V. Turner, Federico Y. Fontana, et al. Assessing the Accuracy of a Continuous Glucose Monitoring System Across Varying Exercise Intensities and Blood Lactate Concentrations in Healthy Male Athletes. Journal of Diabetes Science and Technology, 2024. https://doi.org/10.1177/19322968241292363 ↩

  3. Felicity Thomas, Christopher G. Pretty, Matthew Signal, et al. Accuracy and performance of continuous glucose monitors in athletes. Biomedical Signal Processing and Control, 2017. https://doi.org/10.1016/j.bspc.2016.08.007 ↩

  4. Amy-Lee M. Bowler, Louise M. Burke, Vernon G. Coffey, et al. Day-to-Day Glycemic Variability Using Continuous Glucose Monitors in Endurance Athletes. Journal of Diabetes Science and Technology, 2024. https://doi.org/10.1177/19322968241250355 ↩

  5. Amy-Lee M. Bowler, Jamie Whitfield, Lachlan Marshall, et al. The Use of Continuous Glucose Monitors in Sport: Possible Applications and Considerations. International Journal of Sport Nutrition and Exercise Metabolism, 2023. https://doi.org/10.1123/ijsnem.2022-0139 ↩ ↩2

  6. Joanna Wasik, Marta Armuła, Claire Bongage, et al. Continuous Glucose Monitoring in Healthy Athletes: Current Evidence, Practical Applications and Limitations: A Narrative Review. Quality in Sport, 2026. https://doi.org/10.12775/qs.2026.69.74446 ↩

  7. Mikael Flockhart, Filip J. Larsen. Continuous Glucose Monitoring in Endurance Athletes: Interpretation and Relevance of Measurements for Improving Performance and Health. Sports Medicine, 2023. https://doi.org/10.1007/s40279-023-01910-4 ↩ ↩2

  8. Simon Helleputte, Tim Podlogar, Javier Gonzalez. Application potential of continuous glucose monitoring (CGM) in elite endurance athletes without diabetes: What do physiology and current evidence tell us?. Performance Nutrition, 2025. https://doi.org/10.1186/s44410-025-00013-7 ↩

  9. Emilia Browarska, Katarzyna Superson, Katarzyna Miemczyk, et al. Continuous Glucose Monitoring (CGM) as a Tool for Training Quality Optimization: A Systematic Review of Metabolic Management in Non-Diabetic Athletes. Quality in Sport, 2026. https://doi.org/10.12775/qs.2026.67.74262 ↩