Continuous Glucose Monitors Without Diabetes: The Evidence

What does a continuous glucose monitor actually measure?

A continuous glucose monitor, or CGM, measures glucose in interstitial fluid, the fluid surrounding cells under the skin, and not glucose in blood. A filament sits a few millimetres below the skin surface and reports a value at a fixed interval. The Dexcom Stelo system, cleared by the US Food and Drug Administration in March 2024 as the first over-the-counter CGM, displays a measurement every 15 minutes and is worn for up to 15 days. Clearance of a device is a premarket review. Dietary supplements reach the same shelves without one.

Glucose reaches interstitial fluid after it reaches blood, so the sensor trace runs behind. Basu and colleagues measured the delay directly in 2013 using microdialysis catheters in eight healthy overnight-fasted adults and reported a mean lag of 5.3 to 6.2 minutes for tracer glucose to appear in the interstitial space. Device software smooths and partly corrects for this, adding its own processing delay.

The lag matters most when glucose is moving fast. During the steep rise after a meal, the number on the phone describes where glucose stood several minutes earlier.

How accurate is a consumer CGM, and what does MARD mean?

MARD, mean absolute relative difference, is the average percentage gap between sensor readings and a laboratory reference taken at the same moment. Current sensors report single-digit MARD. The Dexcom G7 accuracy study published by Garg and colleagues in 2022 reported 8.2% MARD for arm placement and 9.1% for abdomen placement, across 316 adults with diabetes, 619 sensors and 77,774 sensor-to-reference paired measurements.

Abbott's FreeStyle Libre 3 reported an overall MARD of 7.8% in a 2023 study of 100 participants aged 4 and older, with 93.4% of sensor values within 20% or 20 mg/dL of the Yellow Springs laboratory reference among those aged 6 and over. Abbott's over-the-counter sensors, Lingo and Libre Rio, cleared in June 2024, use the same Libre sensing platform. The FDA's clearance notice for Stelo states that the device performed similarly to other integrated CGMs.

Those validation numbers were generated in people with diabetes across a wide glucose range. A person without diabetes lives inside a narrow band, roughly 70 to 140 mg/dL (3.9 to 7.8 mmol/L), so the same percentage error consumes a much larger share of the signal. An 8% deviation at 110 mg/dL is about 9 mg/dL (0.5 mmol/L), and MARD is an average, so individual readings sit further out.

What does normal glucose look like in someone without diabetes?

Glucose in people without diabetes stays close to 100 mg/dL (5.6 mmol/L) for most of the day, with brief excursions upward after meals. Shah and colleagues published the most cited reference profile in 2019: 153 healthy non-diabetic participants aged 7 to 80, wearing a blinded Dexcom G6 for up to 10 days across 12 centres.

• Mean glucose: 99 ± 7 mg/dL (5.5 mmol/L); 104 mg/dL (5.8 mmol/L) in participants over 60
• Time between 70 and 140 mg/dL (3.9 to 7.8 mmol/L): median 96%, interquartile range 93% to 98%
• Time above 140 mg/dL: median 2.1%, roughly 30 minutes per day
• Time above 180 mg/dL (10.0 mmol/L): median 0.0%
• Time below 70 mg/dL: median 1.1%, roughly 15 minutes per day
• Coefficient of variation, the standard measure of glucose variability: 17 ± 3%
• 28% of participants recorded at least one reading in the hypoglycaemic range during the wear period

A larger community sample produced looser numbers. Spartano and colleagues analysed blinded Dexcom G6 Pro data from 1,175 Framingham Heart Study participants in 2025, of whom 560 were normoglycaemic. Mean glucose in that group was 114.5 mg/dL (6.4 mmol/L) and time between 70 and 140 mg/dL was 86.8%, with roughly three hours per day at or above 140 mg/dL. Those participants were older and heavier than the 2019 reference cohort. The 15 mg/dL gap between two well-conducted studies is a warning against treating any single normal band as fixed.

Do glucose spikes above 140 mg/dL mean something is wrong?

A reading above 140 mg/dL (7.8 mmol/L) in a person without diabetes is common and carries no established clinical meaning on its own. Normoglycaemic adults in the Framingham analysis spent about 12.3% of each day at or above that threshold, which is the ordinary consequence of eating.

Variability between healthy people is wide. Hall and colleagues gave standardised meals to 57 participants without a diabetes diagnosis in 2018 and sorted their CGM traces into three patterns of increasing variability, which the authors named glucotypes. Participants classified as normoglycaemic by laboratory testing still recorded excursions that would look alarming on a consumer app.

Specialists do not agree on how to read these reports. Spartano and colleagues asked 18 expert clinicians to interpret 20 blinded CGM reports from adults without diabetes in 2025. Interrater reliability was poor, with a Fleiss kappa of 0.36, and all 18 agreed on whether follow-up was warranted in only 3 of the 20 cases.

Can a CGM tell you which foods are right for you?

CGM data show consistently that people respond differently to identical meals. Whether acting on those differences improves health is a separate question with much weaker evidence behind it.

Zeevi and colleagues monitored 800 people for a week in 2015 and recorded responses to 46,898 meals, finding high variability in post-meal glucose between individuals eating the same food. The PREDICT 1 study reported by Berry and colleagues in 2020 replicated the pattern in 1,002 UK adults with independent validation in 100 US adults: post-meal glucose responses varied 68% between people, meal macronutrient composition explained 15.4% of the variance in glycaemic response, gut microbiome composition 6.0%, and genetic variants 9.5%.

One randomised trial has tested the approach. Ben-Yacov and colleagues assigned 225 adults with prediabetes to a personalised postprandial-targeting diet or a Mediterranean diet for six months, with six months of follow-up, and reported greater improvement in daily time above 140 mg/dL and in HbA1c in the personalised arm. Three limits apply. The participants had prediabetes rather than normal glucose tolerance, the intervention was a supervised algorithm rather than a retail app, and the endpoints were glycaemic surrogates rather than clinical outcomes. Surrogate endpoints limit other lifestyle literatures in the same way, including randomised trials of sauna bathing.

Can a sensor tell you whether you have prediabetes?

No. The American Diabetes Association's Standards of Care in Diabetes 2025 states that there is presently insufficient evidence to support the use of continuous glucose monitoring for screening or diagnosis of prediabetes or diabetes. Diagnosis rests on venous laboratory tests, and the thresholds are fixed.

• Diabetes: fasting plasma glucose at or above 126 mg/dL (7.0 mmol/L)
• Diabetes: 2-hour plasma glucose at or above 200 mg/dL (11.1 mmol/L) during a 75 g oral glucose tolerance test
• Diabetes: HbA1c at or above 6.5% (48 mmol/mol)
• Prediabetes: fasting plasma glucose 100 to 125 mg/dL (5.6 to 6.9 mmol/L), 2-hour plasma glucose 140 to 199 mg/dL (7.8 to 11.0 mmol/L), or HbA1c 5.7% to 6.4%
• Except in cases of unequivocal hyperglycaemia, diagnosis requires two abnormal results, either from two different tests or from the same test on separate occasions

A consumer sensor value appears nowhere on that list. Anyone whose readings concern them should ask a clinician for laboratory testing rather than conclude anything from an app.

How much of the trace is measurement bias?

Controlled comparison suggests that consumer sensors read high in people without diabetes. Hutchins and colleagues ran a randomised crossover trial in 15 healthy adults across seven laboratory visits in 2025, comparing a FreeStyle Libre 2 against capillary finger-prick sampling after 50 g carbohydrate loads from fruit-based products.

Sensor estimates were 0.9 ± 0.6 mmol/L (about 16 mg/dL) higher than capillary estimates at fasting and 0.9 ± 0.5 mmol/L higher after meals. The sensor overestimated time above 7.8 mmol/L (140 mg/dL) by roughly fourfold, falling to roughly twofold after correcting for the fasting offset. The glycaemic index assigned to a smoothie was 69 by sensor and 53 by capillary blood. The trial was small, used one sensor model, and was funded by a drinks manufacturer that the authors state had no role in design, analysis or publication.

Time above range, the metric most consumer apps foreground, is therefore the metric most likely to be inflated in a person with normal glucose regulation.

What is the risk of over-interpreting the readings?

The main risk of wearing a CGM without diabetes is behavioural: cutting foods, often fruit, dairy or whole grains, in response to excursions that carry no demonstrated clinical meaning. Four sources of error compound in the same direction. Sensors read roughly 16 mg/dL high in healthy adults, MARD sits near 8%, readings lag blood by 5 to 6 minutes, and healthy people genuinely spend two to three hours a day above 140 mg/dL. Wearable sleep scores raise the same over-interpretation problem.

Direct evidence on eating behaviour in healthy CGM users is thin. The concern appears in clinical commentary rather than in trials that have measured it, and no published trial has quantified rates of food avoidance among consumer sensor users. A 2025 systematic review of seven studies covering about 1,127 non-diabetic participants, three of them randomised, concluded that CGM shows promise for personalising lifestyle changes while noting limited direct evidence of effect on established cardiovascular risk factors.

Some uses do not depend on interpreting single numbers. Comparing two versions of the same breakfast over several weeks, or the effect of a post-meal walk on the same meal eaten twice, uses the sensor as a relative instrument, which is closer to what its error profile supports.

Who should not wear one?

Over-the-counter sensors are cleared for adults who do not use insulin, which excludes several groups outright.

• Insulin users need a prescription CGM. Abbott's Libre Rio is cleared for adults with type 2 diabetes not using insulin, and Lingo is a general consumer product
• People with problematic hypoglycaemia: the FDA's clearance notice states Stelo is not for them, because it is not designed to alert users to dangerously low glucose
• People with a history of disordered eating, or anyone who anticipates responding to daily numbers with dietary restriction, should discuss the decision with a clinician first
• Anyone with symptoms of high or low blood sugar, including unexplained thirst, weight loss, blurred vision or repeated faintness, needs clinical assessment rather than a sensor

Adhesive reactions and insertion-site irritation are the common physical complaints.