A Sensor for a Disease You Don't Have
In March 2024 the FDA cleared Dexcom's Stelo, the first continuous glucose monitor available in the United States without a prescription.1 Three months later Abbott followed with two devices at once: Libre Rio, cleared for adults with type 2 diabetes who are not on insulin, and Lingo, cleared for “consumers 18 years and older who are looking to improve their overall health and wellness.”2
Read those two Abbott clearances again, because the arrangement is genuinely strange. Lingo, the one sold to healthy people as a metabolic-optimization tool, is explicitly not cleared for anyone with diabetes. Libre Rio, the one regulated as a device for a disease, is not the one being marketed to you. The sensor pointed at the healthy population is the one the agency reviewed as a wellness product. And a 510(k) clearance is a substantial-equivalence pathway, not an efficacy pathway. Nobody had to show the FDA that wearing one of these makes a healthy person healthier, because that is not what the pathway asks.
The category took off anyway. In August 2025, Signos became the first CGM-powered system cleared for weight management, at $139 a month.3 Levels, Nutrisense, Zoe, Ultrahuman and January AI all sell some version of the same proposition: your glucose curve is a hidden report card on your metabolism, and only a sensor can read it. Commercial market research puts the over-the-counter CGM segment at roughly $528 million in 2025, with the “non-diabetic health enthusiast” slice near $197 million and growing about 17% a year.4 Those are analyst projections, not data, but the direction is not in dispute.
The statistic doing most of the selling is that only 12% of American adults are metabolically healthy. That number is real. It comes from a 2019 NHANES analysis of 8,721 adults, which found 12.2% met optimal criteria across five markers.5 What the marketing leaves out is what those five markers are: waist circumference, blood pressure, triglycerides, HDL cholesterol, and glucose. Drop waist circumference alone and the figure rises to 17.6%. The shortfall is driven mostly by abdominal adiposity and blood pressure: two things a glucose sensor cannot measure. A statistic about your waistline is being used to sell you a needle for your arm.
I want to be clear about what I am not arguing. The CGM is one of the genuinely great medical devices of the last two decades. For someone taking insulin it is the difference between flying blind and flying with instruments, and it has prevented an enormous amount of suffering. That is not the question here. The question is what happens when you take an instrument calibrated for a pathological range, aim it at a normal one, and sell the readout as self-knowledge.
What a Normal Glucose Curve Actually Looks Like
Before you can decide whether a spike is a problem, you need to know what a healthy person's day looks like. Until 2019 we barely did. The reference data came from people with diabetes, because those were the people wearing sensors.
Design: Multicenter prospective study of 153 healthy non-diabetic participants aged 7 to 80, wearing a blinded Dexcom G6 for a mean of 192 hours.6
Result: Mean glucose 99 ± 7 mg/dL. Median 96% of time spent between 70 and 140 mg/dL. Median time above 140 mg/dL: 2.1%, or roughly half an hour a day. Median time above 180 mg/dL: 0.0%. Mean coefficient of variation 17 ± 3%.
Limitation: Cross-sectional and modest in size, with no outcome follow-up. A corrigendum was published in 2021; readers should consult it alongside the original.
Those numbers set the frame for everything that follows. A healthy person lives, almost all of the time, inside a band about 70 points wide. The larger picture is the same: a 2026 pooled analysis of 3,634 people without diabetes or prediabetes from the PREDICT cohorts found a median 95.3% of time between 70 and 140 mg/dL and a median coefficient of variation of 15.3%.7 In that analysis, time in range discriminated ten-year cardiovascular risk only moderately, and multivariable models explained just 5% to 23% of the variance in CGM metrics. Most of what your sensor shows you is not explained by anything anyone has measured about you.
So a healthy person already spends about 96% of the day in range, and about thirty minutes above 140. The app is going to show you those thirty minutes, in red, with a push notification.
8.5% Error on a 70-Point Range
Here is where the physics starts to matter. A CGM does not measure blood glucose. It measures glucose in interstitial fluid, using an enzymatic reaction, and then applies a smoothing and lag-correction algorithm to estimate what your blood is doing. In a person whose glucose swings from 50 to 350 mg/dL, that estimate is plenty good enough to act on. In a person whose glucose lives between 80 and 140, the error term starts eating the signal.
Dexcom's own FDA decision summary for Stelo reports a mean absolute relative difference of 8.5% across the 70 to 180 mg/dL range, rising to 17.4% below 54 mg/dL.1 The consumer-facing claim, “93% accurate”, means that 93% of readings fall within 20 mg/dL or 20% of a laboratory reference. Roughly one reading in fourteen is off by more than a fifth. An 8.5% error on a reading of 110 mg/dL is about nine points. The routine noise floor of the device is a meaningful fraction of the entire range you are being asked to interpret.
That is the manufacturer's own number, generated under favorable conditions. Independent work in healthy volunteers is less flattering.
Design: Seven conditions, randomized, with washout, in healthy adults with BMI 18 to 30. Abbott FreeStyle Libre 2 compared against capillary reference sampling.8
Result: The sensor overestimated fasting and postprandial glucose by about 0.9 mmol/L, roughly 16 mg/dL. Peak glucose read 1.2 to 1.6 mmol/L higher. Unadjusted, the CGM overestimated time above 7.8 mmol/L about four-fold, and still roughly two-fold after baseline adjustment. A smoothie was classified as high glycemic index by the sensor and low glycemic index by capillary sampling.
Limitation: Only 15 participants, and the test product was supplied under an unrestricted grant from a beverage company. Several authors report food, beverage and pharmaceutical funding.
An older study using a previous device generation found a mean absolute relative difference of 17.6% against venous blood sampled every ten minutes for 24 hours in 34 healthy normoglycemic adults, with a median correlation between sensor and vein of 0.68.9 Sensors have improved since. They have not improved enough to make the interstitial compartment into the vascular one.
And the error is not random with respect to what you are watching. A 2026 double-blind crossover in ten healthy adults found the worst accuracy precisely during rapid glucose declines, where the mean absolute relative difference exceeded 20%.10 The sensor is least trustworthy on the down-slope, which is to say, on the back half of the spike you are staring at.
Then there is the problem that should have ended the personalization business on the day it was published.
Design: Sixteen adults without diabetes wore a Dexcom G4 Platinum on the abdomen and an Abbott Libre Pro on the arm simultaneously, as inpatients, across 760 ad libitum meals.11
Result: The mean Kendall rank correlation between the two devices for ranking meals was 0.43. Choosing which foods to cut based on one sensor rather than the other meant forfeiting 50 ± 10% of the achievable glucose reduction.
Limitation: An older device generation, and a small sample. Funded by the NIH intramural program, with no commercial interest in the result.
Device-to-device disagreement did not go away with newer hardware. A 2025 study had 23 participants wear a FreeStyle Libre 3, a Dexcom G7 and a Medtronic Simplera at the same time for 14 days. Time in range came out at 76.3%, 76.2% and 84.0% depending on which sensor you asked. Three-quarters of participants had a time-in-range figure that differed by more than five percentage points between two of the devices.12 That cohort had type 1 diabetes, so it demonstrates the principle rather than the magnitude in a healthy arm. The principle is enough: the number on your screen is partly a property of the brand you bought.
Two sensors, one body, the same 760 meals, and they agree on your worst foods about as often as a coin agrees with a coin.
Howard, Guo & Hall, AJCN 2020Your Response to an Identical Meal Doesn't Correlate With Itself
The intellectual foundation of consumer CGM is the idea of personalized glycemic response: two people eat the same banana and get different curves, therefore nutrition advice should be individualized, therefore you need a sensor. The founding papers are real and were genuinely interesting. A 2015 study from the Weizmann Institute showed that postprandial responses varied substantially between people and could be predicted better by a machine-learning model incorporating microbiome data than by carbohydrate content alone.13 The PREDICT 1 study, published in Nature Medicine in 2020 with 1,002 UK participants and a 100-person US validation cohort, reported a between-person coefficient of variation of 68% for postprandial glucose and a model correlation of 0.77.14
Both papers are real science. Both were also the commercial origin story for companies now selling subscriptions. PREDICT was funded by, and largely written by employees of, Zoe Global.
The question nobody asked loudly enough for five years was whether your own response is stable. If you eat the same meal twice, do you get the same curve? Because personalization requires that you do. If your response to a food is not reproducible, then a sensor cannot tell you your response to that food. It can only tell you what happened once.
Design: Thirty adults without diabetes, in two inpatient controlled-feeding studies, ate duplicate presented meals about a week apart across four dietary patterns, yielding 1,189 paired glucose responses measured on both Abbott and Dexcom sensors.15
Result: Correlation between duplicate incremental areas under the curve was 0.46 on Abbott and 0.45 on Dexcom. Intraclass correlation coefficients were 0.28 and 0.17. Bland-Altman limits of agreement spanned roughly ±30 mg/dL. Critically, the variability of a person's response to the same meal was statistically indistinguishable from the variability across different meals.
Limitation: Thirty participants, inpatient, so it may understate real-world variability rather than overstate it. Funded by the NIH intramural program with no industry money.
Sit with that last line. The spread in your glucose curve when you eat the identical meal a week later is about as wide as the spread across completely different meals. The authors' own conclusion is that personalized dietary advice based on CGM “requires more reliable methods involving aggregated repeated measurements.” Nobody is selling aggregated repeated measurements. They are selling a notification after lunch.
This is not a fringe finding. A 2025 Nature Medicine study of 55 deeply phenotyped non-diabetic participants eating seven standardized 50-gram carbohydrate meals found reproducibility ranging from an intraclass correlation of 0.26 for beans to 0.73 for pasta.16 That study did find a genuine physiological signal: participants who were insulin-resistant on gold-standard clamp testing had potato responses 179% higher and pasta responses 152% higher than insulin-sensitive participants. Personalized glycemic response is partly real. But finding it required steady-state plasma glucose testing and multi-omic phenotyping in a research unit. It did not fall out of a consumer sensor.
The same paper is quietly devastating about the hacks. The effect sizes for the famous “mitigators” in healthy people were small: a fiber preload produced a Cohen's d of 0.12, protein 0.19, fat 0.05. Those are the numbers underneath a genre of social media content.
The one randomized trial of a full personalized-nutrition program deserves a fair hearing, because it is the best evidence the industry has.
Design: Parallel randomized trial of a CGM-informed personalized nutrition program against US dietary guidelines delivered as digital resources, 177 versus 170 participants over 18 weeks.17
Result: Triglycerides fell by 0.13 mmol/L (P=0.016). Weight fell 2.46 kg more in the intervention arm. LDL cholesterol was null (P=0.521), fasting glucose was null, and HbA1c moved 0.05%.
Limitation: Funded and authored by the company selling the program. The control arm received a leaflet and a few emails against a full self-monitoring, education and support package, and only the intervention arm logged food in real time. The trial never isolated whether biological personalization outperformed generic healthy-eating advice, which is the actual claim.
Independent critics computed the weight loss at roughly 0.15 kg per week, against 0.5 to 1 kg per week for a standard energy-restricted diet, at something on the order of £280 per kilogram lost. You can lose weight by paying close attention to what you eat. That has never been the contested part.
Sources: Hengist 2025; Howard 2020; ACE and NAVIGATOR trials.11,15,18,19
Somebody Already Ran the Experiment. Twice.
Set aside accuracy and reproducibility for a moment and grant the premise. Suppose your sensor is right, and suppose your spikes are stable and real. Does flattening them make you healthier?
There is no trial that answers this in normoglycemic people. Not for cardiovascular events, not for mortality, not for diabetes incidence. Zero. But there are two very large trials that get closer than anything the wellness industry has produced, and they are rarely mentioned by the people selling sensors.
Design: 6,522 Chinese patients with coronary heart disease and impaired glucose tolerance, randomized to acarbose or placebo for a median of five years. Acarbose exists for one purpose: to blunt the postprandial glucose rise.18
Result: The primary five-point cardiovascular composite was flat, hazard ratio 0.98 (95% CI 0.86 to 1.11, P=0.73). Diabetes incidence did fall, rate ratio 0.82 (0.71 to 0.94, P=0.005).
Limitation: Funded by Bayer, which makes acarbose. The population had established coronary disease, so it is higher-risk than a typical consumer.
NAVIGATOR tested the same idea with a different drug. 9,306 people with impaired glucose tolerance plus cardiovascular disease or risk factors, randomized to nateglinide, a postprandial-targeted insulin secretagogue, for a median of five years. No reduction in diabetes incidence, no reduction in the core cardiovascular composite, and more hypoglycemia.19
Pool them. 15,828 people, at substantially higher risk than anyone buying a Lingo subscription, given drugs whose entire mechanism is the thing your app is coaching you toward, for five years. The cardiovascular payoff was nothing. The diabetes-prevention payoff in ACE was real but modest. That asymmetry is the honest story: blunting postprandial glucose may delay a diagnosis in someone already heading toward one. It has never been shown to save a heart, and it has never been tested at all in someone whose HbA1c is normal.
Association data keeps arriving. A Framingham analysis published in Diabetes Care in September 2026 links higher mean glucose and more time above 140 mg/dL to hypertension and cholesterol risk in people without diabetes.20 It is cross-sectional, the devices were supplied by a manufacturer at a discount, and it cannot distinguish glucose as a cause from glucose as a marker of the adiposity and inactivity that travel with it.
Which leaves the interpretation problem. In 2025, eighteen expert clinicians, the names on the CGM consensus guidelines, were independently asked to review twenty CGM reports from people without diabetes, with HbA1c and fasting glucose provided. Overall agreement was a Fleiss kappa of 0.36, which is poor. For people with more than 2% time above 180 mg/dL despite a normal HbA1c and normal fasting glucose, recommendations for follow-up ranged from 56% to 100% of clinicians, with no consensus.21
The world's leading CGM experts cannot agree on what a healthy person's report means. Your app tells you anyway, in a push notification, within ninety seconds.
On Spartano et al., J Diabetes Sci Technol 2025One of the Three Viral Tricks Actually Works
The CGM boom produced a parallel content economy of spike-flattening hacks: vinegar before meals, vegetables before carbohydrates, a walk afterward. I have no interest in dunking on all three, because one of them has better evidence than most things I review.
Walking after meals is the real one. A 2023 meta-analysis of eight three-armed randomized crossover trials in 116 people found that exercise after eating beat inactivity with a standardized mean difference of 0.55, and beat the same exercise done before the meal (SMD 0.47). Exercise before eating was no better than sitting still.22 Timing mattered: sooner after the meal was better. A separate meta-analysis found that light-intensity walking breaks lowered postprandial glucose substantially more than merely standing, and a 2026 review of 53 studies found interruptions every 15 to 20 minutes were optimal.23 This is consistent, replicated, mechanistically obvious muscle physiology. It is also entirely free, and you do not need a sensor to know to do it.
Vinegar is real in dysglycemia and borderline in health. A 2022 systematic review and meta-analysis of 46 studies found vinegar reduced the acute glucose response in people with impaired glucose tolerance or type 2 diabetes, with a standardized mean difference of −0.53. In healthy volunteers the effect was −0.27 with a confidence interval touching zero, and the authors rated the evidence very uncertain given heterogeneity and publication bias.24 Isolated short-chain fatty acids did nothing at all. Longer-term apple cider vinegar analyses report bigger effects, but in diabetic populations with extreme heterogeneity, and one subgroup analysis found fasting glucose went slightly up in apparently healthy participants.25
Food order flattens an hour and pays it back later. An 18-trial meta-analysis of eating vegetables before carbohydrates found glucose lower by 1.21 mmol/L at 30 minutes and 0.52 mmol/L at 60 minutes, and higher at 120 minutes, by 0.24 mmol/L, with insulin also higher.26 The certainty of evidence was rated very low and eight of eighteen trials were at high risk of bias. In type 2 diabetes, where the practice has been studied hardest, seventeen trials found carbohydrate-last eating moved HbA1c by 0.16%.27 You are reshaping a graph. Whether the twenty-four-hour metabolic consequence is favorable is unmeasured.
Now the costs, which are not only financial.
The price of a number you already have
Lingo runs about $1,079 a year, Signos about $1,668. A self-pay HbA1c costs roughly $28 to $59 and a fasting glucose $23 to $34. A year of sensors buys twenty to thirty-eight HbA1c tests. None of it is covered by insurance for wellness use.
Isobornyl acrylate dermatitis
The adhesive in these sensors sensitized people to isobornyl acrylate, identified in 2017. In a review of 1,036 Belgian patients, 5.5% had cutaneous adverse events and 3.8% had confirmed allergic contact dermatitis to the allergen. Barrier films failed in 11 of 16 patients who tried them.
Fear as a feature
In a 2025 mixed-methods study of 56 CGM users, 68% reported fear of developing type 2 diabetes when they saw an elevated reading, and higher eating-disorder symptom scores tracked with device-related distress. Nearly all of that sample had diabetes or prediabetes, so it is suggestive rather than definitive.
No graded recommendation exists
The ADA's 2026 Standards of Care lists Stelo and Lingo and notes that anyone may buy them. It issues no graded recommendation for their use in people without diabetes, because there is no evidence to grade. NICE recommends CGM only for people using insulin.
On that dermatology point, note the structure of the trade-off rather than the raw number. A 3.8% sensitization rate among people who need the device to dose insulin safely is an acceptable cost. At consumer scale, in a market projected toward billions of dollars, the same rate sensitizes a large number of people to an acrylate they will then react to in nail products, adhesives and medical tape, for no medical benefit. That is a real dermatologic externality of a wellness trend, and it is the part of this story that lands in my old specialty.
The professional position is best described as conspicuous silence. The American Diabetes Association's 2026 Standards of Care acknowledges that over-the-counter CGMs exist and that people without diabetes can buy them.28 It stops there. A 2024 narrative review in Diabetic Medicine went further, examining the commercial model (detect abnormal glucose, change behavior, improve metabolic health) and concluding there was a lack of consistent, high-quality evidence for any of the three steps. Those authors urged that commercial claims about CGM use in people without diabetes “be labelled as misleading.”29 That is unusually blunt language for a journal.
And what does the pooled trial evidence show? A 2024 meta-analysis of 25 randomized trials in 2,996 people, with and without diabetes, found HbA1c fell 0.28% and time in range rose 7.4 percentage points, with null effects on BMI and weight; only four of twenty-five trials measured diet at all.30 A 2026 meta-analysis restricted to 1,074 non-diabetic individuals found BMI null, and, the line that matters, HbA1c significantly improved in prediabetic subgroups but showed no significant improvement in core glycemic parameters among healthy normoglycemic individuals.31
That is the shape of the whole field. Something real is happening in prediabetes. In people with normal glucose, the effect disappears into the noise: the same noise the device itself generates.
One more thing, because it is the clearest illustration of how this industry treats evidence. Signos registered a 380-person randomized trial of its platform for weight management. The trial is complete or nearly so. In June 2026 the company issued a press release claiming up to 7% weight loss over six months, “rivaling real-world GLP-1 weight loss.” The underlying paper is not that randomized trial. It is a retrospective cohort in which only 38% of enrolled users had six-month weight data, comparing periods when the same users were engaged against periods when they were not.32 They ran a randomized trial and marketed the engagement analysis. When a company has the better study and publicizes the weaker one, that tells you something about what the better study found, or at least about what they expect you to check.
An Excellent Instrument Pointed at an Unproven Question
I am rating consumer CGM use in people without diabetes Insufficient Data, and I want to explain why it is not the harsher rating, because two sub-claims arguably deserve one.
It is not Marketing Hype, because three things are genuinely true. The device is superb in insulin-treated diabetes. The prediabetes signal is consistent across meta-analytic subgroups. And the pooled behavior-change data shows a small but real HbA1c effect. Something is happening when people wear these. We simply have no evidence it matters in a person whose HbA1c is normal.
But the claim that a consumer CGM reveals your personal food responses is falsified by its own measurement instrument. An intraclass correlation of 0.17 to 0.28 for duplicate meals. A rank agreement of 0.43 between two sensors on one body. A measurement that cannot reproduce itself in the same person one week later cannot be the substrate of personalization. That specific claim, sold at $1,000 to $1,700 a year, is hype.
And the claim that flattening spikes makes a healthy person healthier is not merely unproven. It is the thing 15,828 higher-risk patients were randomized to test for five years, with no cardiovascular benefit. Absence of evidence is not evidence of absence: Peter Attia makes that argument honestly, and discloses that he advises Dexcom. But the burden of proof sits with the people charging the subscription, and after two years of over-the-counter availability they have not picked it up.
If you want to know your metabolic status, an HbA1c and a fasting glucose cost about fifty dollars and are interpretable by an actual clinician. If you want to flatten your postprandial glucose, walk for fifteen minutes after dinner; that one has the meta-analyses. If you are curious and can afford a sensor, wear one for two weeks as an education in your own physiology, then take it off. What you should not do is treat the curve as a verdict on a food, because the device cannot deliver that verdict reliably, and nobody has shown the verdict would matter.
A continuous glucose monitor is a triumph of engineering aimed at a question nobody has shown is worth asking in a healthy person, and in the narrow band where healthy people live, the device's error bars are a meaningful fraction of the thing it is measuring.
- 1. US Food and Drug Administration. 510(k) Decision Summary, Dexcom Stelo Glucose Biosensor System (K234070). Cleared 5 March 2024. Reported MARD 8.5% across 70–180 mg/dL; 17.4% below 54 mg/dL; n=130, 15-day wear. accessdata.fda.gov
- 2. US Food and Drug Administration / Abbott. Clearance of Lingo and Libre Rio over-the-counter continuous glucose monitoring systems. 10 June 2024. Lingo indicated for consumers 18+ seeking to improve overall health and wellness; not indicated for diabetes. abbott.mediaroom.com
- 3. Signos Inc. FDA clearance of the Signos CGM-powered weight management system. 20 August 2025.
- 4. Global Market Insights. Over-the-Counter CGM Market Size Report. 2025. Commercial market research; directional estimates, not peer-reviewed data. gminsights.com
- 5. Araujo J, Cai J, Stevens J. Prevalence of optimal metabolic health in American adults: NHANES 2009–2016. Metab Syndr Relat Disord. 2019;17(1):46–52. n=8,721; 12.2% (95% CI 10.9–13.6). PubMed 30484738
- 6. Shah VN, DuBose SN, Li Z, et al. Continuous glucose monitoring profiles in healthy nondiabetic participants: a multicenter prospective study. J Clin Endocrinol Metab. 2019;104(10):4356–4364. n=153, ages 7–80. See also the 2021 corrigendum, DOI 10.1210/clinem/dgab837. doi:10.1210/clinem/dgab837
- 7. Bermingham KM, Smith HA, Duncan EL, et al. Glycemic profiles in adults without diabetes: pooled analysis of the PREDICT cohorts. Nat Commun. 2026;17:4496. n=3,634. Zoe-affiliated cohort; cross-sectional.
- 8. Hutchins KM, Betts JA, Thompson D, et al. Continuous glucose monitor accuracy and glycemic index classification in healthy adults. Am J Clin Nutr. 2025;121(5):1025–1034. Randomized crossover, n=15. Test product supplied under unrestricted industry grant.
- 9. Akintola AA, Noordam R, Jansen SW, et al. Accuracy of continuous glucose monitoring measurements in normoglycemic individuals. PLoS One. 2015;10(10):e0139973. n=34; MARD 17.6% vs. 10-minute venous sampling. doi:10.1371/journal.pone.0139973
- 10. Hamilton K, Bain A, Bracken R, et al. Continuous glucose monitor accuracy during rapid glycemic change in healthy adults. Nutrients. 2026;18. Double-blind randomized crossover, n=10. MARD exceeded 20% during declines faster than 2 mg/dL/min.
- 11. Howard R, Guo J, Hall KD. Imprecision nutrition? Different simultaneous continuous glucose monitors provide discordant meal rankings for incremental postprandial glucose in subjects without diabetes. Am J Clin Nutr. 2020;112(4):1114–1119. n=16, 760 meals. NIH NIDDK intramural funding. doi:10.1093/ajcn/nqaa198
- 12. Freckmann G, Wehrstedt S, Eichenlaub M, Pleus S, et al. Concurrent use of three continuous glucose monitoring systems. Diabetes Care. 2025;48(7):1213–1217. n=23; cohort had type 1 diabetes. doi:10.2337/dc25-0129
- 13. Zeevi D, Korem T, Zmora N, et al. Personalized nutrition by prediction of glycemic responses. Cell. 2015;163(5):1079–1094. doi:10.1016/j.cell.2015.11.001
- 14. Berry SE, Valdes AM, Drew DA, et al. Human postprandial responses to food and potential for precision nutrition (PREDICT 1). Nat Med. 2020;26(6):964–973. n=1,002 UK + 100 US. Funded and staffed by Zoe Global Ltd. doi:10.1038/s41591-020-0934-0
- 15. Hengist A, Ong JA, McNeel K, Guo J, Hall KD. Imprecision nutrition? Intraindividual variability of glucose responses to duplicate presented meals in adults without diabetes. Am J Clin Nutr. 2025;121(1):74–82. n=30, 1,189 responses. NIH intramural funding; NCT03407053, NCT03878108. doi:10.1016/j.ajcnut.2024.10.007
- 16. Wu Y, Ehlert B, Metwally AA, et al. Individual variations in glycemic responses to carbohydrates and underlying metabolic physiology. Nat Med. 2025;31(7):2232–2243. n=55 deeply phenotyped non-diabetic participants. doi:10.1038/s41591-025-03719-2
- 17. Bermingham KM, Linenberg I, et al. Effects of a personalized nutrition program on cardiometabolic health: a randomized controlled trial. Nat Med. 2024;30:1888–1897. n=347, 18 weeks. Zoe-funded and Zoe-authored. doi:10.1038/s41591-024-02951-6
- 18. Holman RR, Coleman RL, Chan JCN, et al. Effects of acarbose on cardiovascular and diabetes outcomes in patients with coronary heart disease and impaired glucose tolerance (ACE): a randomised, double-blind, placebo-controlled trial. Lancet Diabetes Endocrinol. 2017;5(11):877–886. n=6,522, median 5.0 years. Funded by Bayer AG. doi:10.1016/S2213-8587(17)30309-1
- 19. Holman RR, Haffner SM, McMurray JJ, et al. Effect of nateglinide on the incidence of diabetes and cardiovascular events (NAVIGATOR). N Engl J Med. 2010;362(16):1463–1476. n=9,306, median 5.0 years. doi:10.1056/NEJMoa1001122
- 20. Bakhshi B, Lin H, Sultana N, et al. Beyond traditional glycemic measures: CGM glycemic profiles reveal associations with cardiometabolic risk in individuals without diabetes. Diabetes Care. Published online 17 September 2026. DOI 10.2337/dc26-1197. Framingham Heart Study; cross-sectional; devices discounted by manufacturer. doi:10.2337/dc26-1197
- 21. Spartano NL, Prescott B, Walker ME, et al. Expert clinical interpretation of continuous glucose monitor reports from individuals without diabetes. J Diabetes Sci Technol. 2025. 18 clinicians, 20 reports; Fleiss kappa 0.36. doi:10.1177/19322968251315171
- 22. Engeroff T, Groneberg DA, Wilke J. After dinner rest a while, after supper walk a mile? A systematic review with meta-analysis on the acute postprandial glycemic response to exercise before and after meal ingestion. Sports Med. 2023;53(4):849–869. 8 randomized crossover trials, n=116. doi:10.1007/s40279-022-01808-7
- 23. Buffey AJ, Herring MP, Langley CK, Donnelly AE, Carson BP. The acute effects of interrupting prolonged sitting time with walking on postprandial metabolism. Sports Med. 2022;52(8):1765–1787. See also Gale JT, et al. Obes Rev. 2026 (53 studies). doi:10.1007/s40279-022-01649-4
- 24. Cherta-Murillo A, Pugh JE, Alaraj-Alshehhi S, et al. The effects of short-chain fatty acids on glycemic control in humans: a systematic review and meta-analysis. Am J Clin Nutr. 2022;116(2):335–361. 46 studies, n=913. Evidence rated very uncertain. doi:10.1093/ajcn/nqac085
- 25. Hadi A, Pourmasoumi M, Najafgholizadeh A, et al. The effect of apple cider vinegar on lipid profiles and glycemic parameters: a systematic review and meta-analysis. BMC Complement Med Ther. 2021;21:179. 9 studies; I²=91%. doi:10.1186/s12906-021-03351-w
- 26. Kampmann-Gongoll L, Schlesinger S, Noreik M, et al. Effect of eating vegetables before carbohydrates on glucose, insulin and glycemic control: a systematic review and meta-analysis of randomized controlled trials. Clin Nutr. 2026;66:106775. 18 RCTs; very low certainty of evidence. doi:10.1016/j.clnu.2026.106775
- 27. Saldarriaga-Callejas S, Ratan Y, Pasqualotto E, et al. Food order and glycemic outcomes in type 2 diabetes: a systematic review and meta-analysis. Acta Diabetol. 2026;63(3):399–411. 17 RCTs, n=389; HbA1c −0.16%. doi:10.1007/s00592-025-02586-0
- 28. American Diabetes Association. 7. Diabetes technology: Standards of Care in Diabetes—2026. Diabetes Care. 2026;49(Suppl 1):S150. Lists OTC CGM devices; issues no graded recommendation for use in people without diabetes. doi:10.2337/dc26-S007
- 29. Oganesova Z, Pemberton J, Brown A. Innovative solution or cause for concern? The use of continuous glucose monitors in people not living with diabetes: a narrative review. Diabet Med. 2024;41(11):e15369. doi:10.1111/dme.15369
- 30. Richardson KM, Jospe MR, Bohlen LC, Crawshaw J, Saleh AA, Schembre SM. Use of continuous glucose monitors to support behaviour change: a systematic review and meta-analysis. Int J Behav Nutr Phys Act. 2024;21:145. 25 RCTs, n=2,996. doi:10.1186/s12966-024-01692-6
- 31. Liao X, Li Y, Tang S, Xiao Y, Yu X, Huang R, Zhong T. Continuous glucose monitoring in non-diabetic individuals: a systematic review and meta-analysis. Eur J Med Res. 2026. 23 studies (7 RCTs), n=1,074. BMI null; benefit confined to prediabetic subgroups. doi:10.1186/s40001-026-03920-0
- 32. Dixon W, Kim S, Shryack GE, Levonian D, Gusz D, Fouladgar-Mercer S, Skyler JS. Weight outcomes among users of a CGM-based digital weight management platform. J Diabetes Sci Technol. 2026. Retrospective cohort; 1,147 of 3,007 enrolled users had 180-day weight data. Company-authored.
- 33. Pyl J, Dendooven E, Van Eekelen I, et al. Prevalence and prevention of contact dermatitis caused by FreeStyle Libre: a monocentric experience. Diabetes Care. 2020;43(4):918–920. n=1,036; 3.8% confirmed allergic contact dermatitis to isobornyl acrylate. See also Herman A, Aerts O, Baeck M, et al. Contact Dermatitis. 2017;77(6):367–373. doi:10.2337/dc19-1354
- 34. Richardson KM, Jospe MR, Somerville J, Felrice J, Schembre SM. Understanding the benefits and psychological burdens of using continuous glucose monitoring for lifestyle change. Obes Res Clin Pract. 2025;19(5):417–426. n=56; 96.4% of the sample had diabetes or prediabetes. doi:10.1016/j.orcp.2025.10.003