CGM Data in the Non-Diabetic Athlete: Reading Glucose Variability Without Pathologizing Normal Physiology

A 34-year-old amateur triathlete walks into intake with two weeks of Stelo data on his phone. He has tagged every post-meal spike above 140 mg/dL in red. His morning oatmeal-and-banana pre-ride breakfast peaks him at 168. His post-ride recovery shake peaks him at 172. He has an overnight dip to 62 mg/dL on three of fourteen nights. The Stelo app's coaching prompts have told him his glucose variability is in the "needs improvement" range, and he is asking whether he should drop carbohydrate intake, whether his recovery shake is making him insulin resistant, and whether the overnight dip means he has reactive hypoglycemia.

His fasting glucose is 86. His HbA1c is 5.1. His fasting insulin is 4.2 uIU/mL. His HOMA-IR is 0.9. He is the textbook definition of a metabolically healthy athlete, and the interpretive frame the consumer CGM app handed him is a diabetic decision matrix applied to physiology the diabetic decision matrix was never built to evaluate. If the RD echoes the app's frame — "yes, your variability is concerning, cut your carbs" — the athlete ends up under-fueling a training block, his power output drops, his recovery suffers, his late-season race is compromised, and the intervention has produced exactly the clinical outcome the screening was supposed to prevent.

The consumer CGM market is now mature enough that athletes self-procuring fourteen-day or twenty-eight-day sensors are showing up routinely in the sports-nutrition intake with two weeks of data, an app's interpretation, and a set of questions the standard intake was not designed to answer. The fix is a CGM interpretation framework calibrated to athletic physiology rather than to diabetic decision matrices — one that distinguishes normal post-meal excursion from clinically meaningful dysglycemia, distinguishes exercise-induced glucose rise from insulin resistance, and routes the small minority of cases with true dysglycemia signals to the appropriate medical referral without pathologizing the majority who are just looking at normal physiology in an interface that grades it red.

This post is the CGM interpretation protocol I run when an athlete brings sensor data into the intake. The physiology that drives the curves, the calibration error of the device, the four-quadrant variability matrix, the true dysglycemia signals that warrant referral, the common counseling mistakes, and the SOAP documentation that captures the interpretation defensibly.

Why the consumer CGM app's frame is wrong for healthy athletes

Three structural reasons.

The diabetic time-in-range standard (70-180 mg/dL) does not apply to non-diabetic physiology. The 70-180 target was developed for type 1 and type 2 diabetes management — populations where glucose excursions above 180 are common, where hypoglycemia below 70 is clinically dangerous because of impaired counter-regulation, and where the management goal is keeping the patient inside the band for as much of the day as possible. The non-diabetic athlete with intact counter-regulation has glucose excursions to 140-170 routinely after carbohydrate-dense meals, has overnight nadirs in the 60-75 range with no symptoms and no impairment, and has post-exercise spikes that are gluconeogenic responses to catecholamines rather than pathologic spikes. Reading non-diabetic physiology against the 70-180 band produces a false-positive rate that approaches 80% in athlete populations and counsels intervention on what is well-described normal variability.

The CGM measures interstitial glucose, not plasma glucose, with a 5-15 minute lag and a 10-15% calibration error. A single reading of 168 mg/dL on the sensor corresponds to an actual plasma glucose somewhere in the 142-194 range depending on the calibration drift of that specific sensor, the body site of placement, the hydration state of the user, and where the reading sits in the lag window. The consumer apps display the number to single-mg/dL precision, which creates a false sense of measurement accuracy and pushes the user to react to noise. The interpretive frame for sensor data needs to read clusters, not points, and tolerate the noise band the device produces.

Exercise-induced glucose rise is not insulin resistance. A 20-40 mg/dL rise in blood glucose during high-intensity exercise is a normal counter-regulatory response — sympathetic-nervous-system catecholamines (epinephrine and norepinephrine) drive hepatic gluconeogenesis and glycogenolysis to maintain glucose availability for the working muscle. The rise occurs at the moment the muscle is taking up glucose at the fastest rate of the day and is mechanistically opposite from insulin resistance, which is impaired peripheral glucose uptake at rest. Reading the exercise-spike as evidence of insulin resistance is reading the curve backwards.

The interpretive framework the consumer app applies to all three of these surfaces is calibrated for diabetic decision-making and produces clinically incorrect counseling when applied to non-diabetic athletes.

The five-dimension interpretive frame for athletic CGM data

Dimension 1: Fasting glucose stability. The 4 AM to 7 AM window in a non-stressed, non-traveling athlete should sit between 65 and 95 mg/dL with minimal variability. Fasting glucose consistently above 100 over multiple nights, or a fasting drift upward across the wear period, is the single most informative signal in the data for screening dysglycemia. Pair the sensor data with a venous fasting glucose and an HbA1c. The CGM-derived fasting estimate is correlated with venous glucose but is not a substitute for it; the venous draw is the diagnostic instrument.

Dimension 2: Post-meal excursion amplitude. The clinically informative metric is peak-minus-baseline, not the peak alone. An athlete with a baseline of 85 and a peak of 165 has a 80-point excursion. An athlete with a baseline of 110 and a peak of 165 has a 55-point excursion. The same peak number describes two different metabolic responses. Healthy non-diabetic excursions in response to a mixed meal with 60-90 g of carbohydrate typically run 40-60 mg/dL above baseline; excursions above 80 mg/dL on a moderate carbohydrate load warrant a closer look at the meal composition, the timing relative to exercise, and the underlying glucose tolerance.

Dimension 3: Time-in-range, calibrated for athletic physiology. The athletic-population reasonable target is approximately 80% of the day between 70 and 140 mg/dL, with excursions above 140 acceptable in the immediate post-meal window and excursions above 180 worth investigating but not automatically pathologic. The diabetic 70-180 standard is too permissive for screening dysglycemia in healthy adults and too rigid for screening normal physiology — it catches almost nothing in either direction.

Dimension 4: Glucose variability (coefficient of variation). The CV — standard deviation divided by mean, expressed as a percentage — captures whether the glucose system is well-regulated independent of where the mean sits. Healthy non-diabetic CV typically runs 12-20%. CV above 25% in a non-diabetic athlete suggests either unstable counter-regulation, frequent under-fueling with rebound spikes, frequent over-fueling with hyperinsulinemic crashes, or sensor noise. The CV is the highest-signal single-number summary of the wear period.

Dimension 5: Exercise-induced glucose response pattern. Distinguish the post-meal spike from the exercise spike. Mark each session in the data, note the intensity and duration, and look at the immediate glucose response. A high-intensity interval session producing a 30-50 mg/dL rise that resolves within 60-90 minutes is normal counter-regulation. A long endurance session producing a steady fall to the 60-70 range that resolves with carbohydrate intake is normal endurance physiology. A high-intensity session producing a 60+ mg/dL rise that takes 3+ hours to resolve warrants a closer look at the broader picture (caffeine dose, stress, hydration, sleep) before it becomes a glucose-regulation concern.

The four-quadrant variability matrix

The five dimensions reduce to a four-quadrant clinical decision matrix:

1. Tight variability, low excursions (CV under 18%, post-meal excursions under 60 above baseline). The metabolically efficient athlete. No intervention indicated. The athlete who showed up worried about a spike to 145 is in this quadrant 80% of the time. Re-frame the data, document the baseline, and proceed with the consult complaint.

2. Wide variability, moderate-to-high excursions (CV 20-28%, post-meal excursions 60-100 above baseline). Normal athletic physiology with high-glycemic intake patterns. The intervention, if any, is around meal composition for daytime stability — pairing carbohydrate with protein and fat, distributing carbohydrate across multiple smaller meals rather than two large ones, and timing carbohydrate density around training sessions. Not a dysglycemia signal. Re-screen with venous fasting glucose and HbA1c only if the rest of the clinical picture warrants.

3. Tight variability, high excursions (CV under 18%, post-meal excursions consistently above 80 above baseline). A pattern worth investigating. The CV is well-controlled, suggesting the counter-regulation is intact, but the excursion amplitude is consistently above the healthy range. Differential: high-glycemic-load meals at the meal sizes the athlete is consuming, low fiber-protein-fat content in the same meals, possible impaired first-phase insulin response, or possible early dysglycemia. Order venous fasting glucose, HbA1c, and fasting insulin. Calculate HOMA-IR. Refer to PCP for a 2-hour OGTT if the venous workup is borderline.

4. Wide variability, low mean glucose (CV above 25%, mean below 90). The under-fueled-athlete pattern. Frequent rebound spikes from low baseline, glycogen-depleted physiology driving high-amplitude counter-regulatory responses, and possible co-occurring [low energy availability](/blog/screening-athletes-for-low-energy-availability). The intervention is fueling assessment and the [LEA screening protocol](/blog/screening-athletes-for-low-energy-availability), not glucose-management counseling. Pair with the [RED-S workup](/blog/red-s-in-male-athletes-clinical-differential) in male athletes or the LEAF-Q in female athletes.

When to refer to medical

Four signals warrant a medical referral rather than dietetic management.

Fasting glucose consistently above 100 across multiple sensor wear periods, especially paired with HbA1c above 5.7. This is the impaired-fasting-glucose / pre-diabetes range and belongs in PCP or endocrinology workup with venous confirmation and an OGTT.

Post-meal excursions consistently above 200 mg/dL on moderate carbohydrate loads (50-80 g of carbohydrate in a mixed meal). The 200 threshold is the diabetic post-prandial cutoff; consistent excursions above it in a non-diabetic athlete on moderate carbohydrate intake warrant a clinical workup.

Symptomatic hypoglycemia. Tremor, sweating, palpitations, confusion, or syncope occurring at the same time as a CGM-recorded glucose under 55, with reproducibility across multiple events, requires medical workup — the differential includes reactive hypoglycemia, late dumping syndrome, insulinoma in rare cases, and medication-induced hypoglycemia. The asymptomatic dip into the 60s during sleep is not in this category.

Glucose variability above CV 35% in an athlete with no obvious explanation (no under-fueling pattern, no extreme intake variability, no acute illness). The wide-variability signal in the absence of behavioral explanation warrants a venous workup and a clinician evaluation.

Common counseling mistakes

Applying the diabetic 70-180 time-in-range standard to non-diabetic athletes. Produces false positives at scale. The athletic-calibrated 70-140 target with allowance for post-meal excursions is the appropriate frame.

Counseling a low-carbohydrate diet on the basis of CGM excursion data alone. A healthy athlete with normal HbA1c and normal fasting insulin who is excursing to 165 after oatmeal is not insulin resistant and does not benefit from carbohydrate restriction. The endurance-sport population requires 6-10 g/kg/day of carbohydrate for performance; reducing intake on the basis of a misinterpreted CGM curve compromises training adaptation.

Reading an overnight dip to the 60s as reactive hypoglycemia. Asymptomatic interstitial-glucose readings in the 60-75 range during sleep are within normal non-diabetic physiology. The reactive-hypoglycemia diagnosis requires symptomatic hypoglycemia with reproducibility, plasma glucose confirmation, and a structured workup; a single overnight dip on a consumer CGM is not it.

Treating the exercise-induced spike as evidence of insulin resistance. Mechanistically backwards. The exercise spike is counter-regulatory and resolves spontaneously; it does not require intervention.

Counseling on a single 14-day wear without a baseline. The first sensor wear catches a wide range of physiology depending on travel, illness, sleep, training load, and dietary pattern. A second 14-day wear at a different point in the training year is what distinguishes the athlete's baseline from a noisy window.

Ignoring the 10-15% sensor error band and reacting to individual readings. The interpretive frame reads clusters and patterns, not single points. An isolated spike to 175 that does not repeat across the wear period is sensor noise; a repeated pattern of 165-180 peaks after the same meal type is signal.

Failing to integrate the CGM with venous labs. The sensor data is a hypothesis-generating instrument. The diagnostic workup for dysglycemia is venous fasting glucose, HbA1c, fasting insulin, and where indicated a 2-hour OGTT. A clinical decision made on CGM data alone, without venous confirmation, has neither the accuracy nor the documentation depth the case requires.

Where this lands in the SOAP

Subjective section format:

```

CGM Data Review (sensor type, wear dates, days of usable data):

• Sensor: [Stelo / Lingo / Libre 3 / Dexcom G7]
• Wear period: [YYYY-MM-DD to YYYY-MM-DD]
• Days of usable data: [X / 14]
• Athlete-reported reason for self-procurement: [text]

Interpretation by dimension:

• Fasting glucose stability (4-7 AM): mean [X], range [X-Y]
• Post-meal excursion amplitude (peak minus baseline): mean [X], range [X-Y]
• Time-in-range (70-140, athletic-calibrated): [X%]
• Coefficient of variation: [X%]
• Exercise-induced spikes: [pattern description]
• Quadrant: [1-4 from variability matrix]

Venous lab pairing:

• Fasting glucose: [X mg/dL, date]
• HbA1c: [X%, date]
• Fasting insulin: [X uIU/mL, date]
• HOMA-IR: [calculated value]

Medical red flags: [list or "none identified"]

Clinical impression: [normal athletic variability / high-excursion pattern

warranting venous workup / dysglycemia signal warranting referral /

under-fueling pattern warranting LEA workup]

Action: [reframe data / venous workup / PCP referral / LEA screening]

Follow-up: [date, plan]

```

Assessment integrates the CGM interpretation with the venous data, the training-cycle context, and the athlete's stated concern. Plan documents the counseling delivered, the workup ordered, and the re-evaluation cadence. See [SOAP notes for sports dietitians](/blog/soap-notes-for-sports-dietitians) for the broader documentation framework.

Where platform tooling helps

The bottleneck in CGM interpretation at scale is the data marshalling — fourteen to twenty-eight days of sensor data exported as a CSV, annotated against the athlete's food log, training log, sleep log, and travel log, parsed into the five-dimension interpretive frame, scored against the four-quadrant matrix, and integrated with the venous labs the workup actually relies on. The intake that has to do this by hand drops the interpretation on busy weeks, and the case ends up with the athlete being told something the data does not actually support.

The leverage is a CGM interpretation module that ingests the sensor CSV, auto-computes the five dimensions, applies the four-quadrant matrix calibrated for non-diabetic athlete physiology, flags the true dysglycemia signals against medical-referral thresholds, integrates with the food log and training log to annotate the excursions with the meal and session that produced them, and pre-populates the SOAP documentation with the interpretation and the workup recommendation. The RD's job becomes the clinical interpretation and the conversation, not the spreadsheet.

The chart trail is also defensible — every interpretation logged, every referral documented, every counseling decision tied to the data the decision was made on.

The bottom line

Consumer CGMs put a fourteen-day sensor and a misleading interpretive frame on every metabolically healthy athlete who reads a wellness newsletter, and the sports-RD intake is now routinely seeing athletes walk in with two weeks of data and a set of questions the standard intake was not built to answer. The fix is a CGM interpretation framework calibrated to athletic physiology — fasting stability, excursion amplitude, athletic-calibrated time-in-range, coefficient of variation, and exercise-induced response pattern — read against a four-quadrant decision matrix that distinguishes metabolically efficient athletes from those with under-fueling patterns, normal physiology from early dysglycemia, and consumer-app noise from clinical signal.

The athlete who walks in worried about a spike to 168 after oatmeal, with a normal HbA1c and a normal fasting insulin, is in quadrant 1 of the matrix 80% of the time and needs the data re-framed, not his diet restricted. The athlete in quadrant 4 with the wide-CV, low-mean pattern needs the [LEA workup](/blog/screening-athletes-for-low-energy-availability), not glucose-management counseling. The small minority of athletes in quadrant 3 with consistently high excursions and tight variability warrant the venous workup the consumer CGM cannot do on its own.

[Calsanova's Dietitian plan](/signup?role=dietitian) ships a CGM interpretation module with sensor-CSV ingestion, five-dimension auto-scoring, athletic-calibrated time-in-range, four-quadrant matrix routing, integrated food and training log annotation, venous-lab pairing, and pre-populated SOAP documentation. Start your 30-day free trial and turn consumer CGM data from a source of athlete anxiety into a clinical instrument that catches the cases the consumer app interprets wrong.