Skip to content

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

Oak
A lab machine flexing a strip of engineered muscle on a belt beside a glowing amber fluid column that pulses slightly behind it.

A continuous glucose monitor is useful to a runner for one narrow purpose: comparing your own fueling strategies against each other, on repeated runs, under conditions you control. It is a poor instrument for reading glucose moment-to-moment during hard efforts, because interstitial fluid lags blood and accuracy degrades exactly when glucose is moving fast. Validation work in nondiabetic recreational runners found that agreement with capillary reference values is worse during running than at rest.1 Treat the sensor as a tool for repeated-measures experiments across days. It is not a live fuel gauge.

What the sensor is measuring

A CGM does not sit in a vein. The sensor sits in subcutaneous interstitial fluid. Glucose diffuses from plasma into that compartment, which introduces a physiological delay of roughly five to ten minutes, plus whatever smoothing the manufacturer’s algorithm applies. During steady state the offset is small and predictable. During a 400m interval or the first twenty minutes after a gel, plasma is moving faster than interstitial fluid can follow. The reported value is a lagged and filtered estimate.

This matters for interpretation more than most consumer material admits. A review of CGM in endurance athletes makes the point directly: the measurement reflects the balance of hepatic glucose output, muscle uptake and gut absorption, and none of those are recoverable from the trace alone.2 A flat line during a long run means production and uptake happen to be matched at that moment. It does not mean you are fueled.

There is a second constraint. Blood glucose is tightly regulated, and in a person without diabetes it stays inside a narrow band under a wide range of loads. Reviews of CGM use in people without diabetes have noted that the clinical meaning of small excursions in this population is not established.3 You should expect your interesting signal to be small relative to sensor noise, which is precisely why single runs tell you nothing and repeated runs tell you something.

Sensor choice

We would use a Dexcom G7 for running work, with Abbott’s Libre 3 Plus as a reasonable alternative. Both report manufacturer MARD (mean absolute relative difference against a laboratory reference) under ten percent in adults, which is comparable. The differences that matter for this use case are elsewhere.

The G7 samples every five minutes and has a documented local Bluetooth connection, which means third-party apps such as xDrip+ can read it directly without going through the vendor cloud. Wear is ten days plus a twelve-hour grace period. The Libre 3 Plus samples every minute and runs fifteen days, which gives you finer temporal resolution during intervals. Its data path is more closed, though Juggluco and xDrip+ can read the sensor over BLE in practice.

Placement is a real variable for runners. Both sensors are labeled for the back of the upper arm. If you sleep on that side you will generate compression artifacts overnight, which look like sudden drops to 55 mg/dL with an equally sudden recovery. Alternate arms between sensors and record which arm you used, because that column will explain outliers later.

Getting the raw data out

Start with the vendor export, then decide whether you need streaming.

Dexcom Clarity exports CSV from the web interface. The relevant columns are Timestamp (YYYY-MM-DDThh:mm:ss), Event Type, Event Subtype and Glucose Value (mg/dL). Rows with Event Type of EGV are the sensor readings. Values at the rails appear as the strings Low and High rather than numbers, so cast with pd.to_numeric(..., errors="coerce") and handle those explicitly instead of dropping them.

LibreView exports a CSV with a Record Type column, where 0 is the automatic historic reading and 1 is a scan. Filter to type 0 for the continuous series. The column is Historic Glucose mg/dL or the mmol/L equivalent depending on account region, so check units before you compare anything across exports.

If you want one-minute resolution and an API, run Nightscout. It gives you a Mongo-backed store and a REST endpoint at /api/v1/entries.json?count=100000&find[dateString][$gte]=2026-05-01, which returns records with an sgv field in mg/dL and date as epoch milliseconds. xDrip+ uploads to it from either sensor. This is more setup, and worth it if you plan to run this for a season.

Aligning glucose with your run files

Your running data lives in .fit files from Garmin, Coros or Wahoo. Parse them in Python with fitdecode or fitparse, pulling record messages for timestamp, heart_rate, power, cadence and enhanced_altitude, plus the session message for totals.

The alignment bug that ruins these analyses is time zones. FIT timestamps are UTC seconds since 1989-12-31. Dexcom Clarity exports in the device’s local time with no offset attached. If you travel, or race in another zone, you will silently shift your glucose trace by hours. Localize the CGM timestamps to the zone the phone was in, convert everything to UTC, and assert on the result.

cgm = cgm.set_index("ts").tz_localize("America/New_York").tz_convert("UTC")
cgm_1min = cgm["glucose"].resample("1min").interpolate("time", limit=10)
run = run.tz_convert("UTC")
merged = pd.merge_asof(run, cgm_1min.reset_index(),
                       on="ts", direction="nearest",
                       tolerance=pd.Timedelta("3min"))

Use merge_asof with a tolerance rather than an unbounded join, so that gaps in sensor coverage appear as nulls instead of as stale values stretched across twenty minutes.

Metrics worth computing

Three quantities carry most of the information.

The first is incremental area under the curve over the ninety minutes after a fueling event, computed with the trapezoid rule against the pre-fuel baseline. Use numpy.trapezoid on the one-minute resampled series. This compares gels, drink mixes and solid food on a single scale.

The second is rate of change in mg/dL per minute, estimated over a rolling fifteen-minute window with a least-squares slope rather than a two-point difference. Slopes are where the lag bites hardest, so interpret them as directional rather than exact.

The third is the low-end exposure: minutes below 70 mg/dL, separated into during-run and overnight buckets. A case study tracking a national-level female marathon runner through a real race with CGM shows how much structure appears in a race trace that a single fingerstick would miss.4 Work in sub-elite endurance athletes has also used CGM to characterize post-session and overnight glucose patterns as a window on recovery load.5

Design the experiment before you collect. Pick one route and one target pace. Run it six times: three with fueling strategy A, three with strategy B, alternating. Compare mean iAUC and mean pace-normalized heart rate across arms. A controlled trial in amateur marathon runners comparing carbohydrate supplementation protocols is a useful template for what the comparison should look like.6 With n of three per arm you are detecting large effects only, and that is the right expectation.

Failure modes that produce fake insights

Sensor warm-up is the first. The initial twelve to twenty-four hours after insertion drift, so discard day one entirely rather than trying to correct it.

Dehydration and skin temperature shift interstitial dynamics, which means a hot long run and a cool easy run are not measured under the same conditions. A narrative review of CGM in healthy athletes catalogs these limitations and cautions against reading small differences as meaningful.7

Finally, there is the interpretive failure. Runners see a post-run rise and conclude something is wrong. Post-exercise hyperglycemia after hard efforts is a normal consequence of catecholamine-driven hepatic glucose output. The trace is describing your physiology. It is not grading it.

If you have diabetes, everything above is secondary to clinical management. A systematic review of diabetes and long-distance running is the place to start reading with your clinician.8 If you see repeated readings below 60 mg/dL or symptoms of hypoglycemia at rest, stop analyzing and see a physician. A CGM is not a diagnostic device in this context and nothing here is medical advice.

Questions people also ask

What is the best glucose monitor for athletes?

The Dexcom G7 if you want open local Bluetooth access and a five-minute cadence, and the Libre 3 Plus if you want one-minute resolution and fifteen-day wear. Accuracy is close enough between them that data access should decide it.

Can I use a CGM to decide when to take a gel mid-race?

We would not. The lag and the reduced accuracy during exercise mean the number you see is minutes behind reality.1 Use the sensor in training to build a fueling schedule, then execute the schedule on the clock.

Does a flat glucose line mean I am well fueled?

No. Glucose is the balance of liver output, muscle uptake and gut absorption, and a stable trace can mean those are cancelling out while glycogen falls.2

Is CGM data meaningful for a runner without diabetes?

It is meaningful comparatively. Absolute thresholds developed for diabetes care do not transfer, and the interpretation of small excursions in people without diabetes remains unsettled.3

Footnotes

  1. Lesley J. Mason, Timothy Hartwig, David Greene. Validating the Use of Continuous Glucose Monitors With Nondiabetic Recreational Runners. International Journal of Sports Physiology and Performance, 2024. https://doi.org/10.1123/ijspp.2024-0102 ↩ ↩2

  2. 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

  3. David C. Klonoff, Kevin T. Nguyen, Nicole Y. Xu, et al. Use of Continuous Glucose Monitors by People Without Diabetes: An Idea Whose Time Has Come?. Journal of Diabetes Science and Technology, 2022. https://doi.org/10.1177/19322968221110830 ↩ ↩2

  4. Fuminori Takayama. Assessment of Blood Glucose Responses in a Female National-Level Marathon Runner Using Continuous Glucose Monitoring during a Real-World Marathon Race. BioMed, 2023. https://doi.org/10.3390/biomed3030031 ↩

  5. Diana G. Kulawiec, Tony Zhou, Jennifer L. Knopp, et al. Continuous glucose monitoring to measure metabolic impact and recovery in sub-elite endurance athletes. Biomedical Signal Processing and Control, 2021. https://doi.org/10.1016/j.bspc.2021.103059 ↩

  6. Peng Zhang, Zhongke Gu, Jiale Lv, et al. Effects of Different Carbohydrate Supplementation on Marathon for Amateur Runners: A Controlled Trial. 2026. https://doi.org/10.21203/rs.3.rs-4696071/v2 ↩

  7. 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 ↩

  8. Lorin Braschler, Mabliny Thuany, Thomas Züger, et al. Diabetes Mellitus and Long-Distance Running: A Systematic Review. Sports Medicine - Open, 2026. https://doi.org/10.1186/s40798-026-01084-z ↩

Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.