The short version: the central premise of consumer glucose monitoring is that your personal glucose response to a food is a stable, measurable property of you. The best test of that premise put two sensor brands on 30 people without diabetes and fed them the same meals twice. The agreement between the two readings of the identical meal was ICC 0.14 to 0.31, and the variability in response to an identical meal was no smaller than the variability across completely different meals (p=0.38 and p=0.60).

If your response to the same breakfast on two days differs as much as your response to two different breakfasts, then the number on your phone is not telling you about the food.

Separately: there is no published prospective study we could find that measured glucose excursions in people without diabetes and then followed them to see whether those excursions predicted anything. And when 18 expert clinicians were shown the same 20 glucose reports from people without diabetes, they could not agree on who needed follow-up — Fleiss kappa 0.36.

Three corrections before we start

These matter because all three claims circulate widely, including in earlier drafts of our own notes.

1. The "Selvin 2023" accuracy study was not in healthy people. The Clinical Chemistry paper with n=172 and two sensors worn simultaneously enrolled adults with type 2 diabetes not on insulin. It is good evidence that sensors disagree with each other; it is not evidence about the normoglycaemic range.

2. The "societies permit 30% error in healthy people" claim is a misreading. The Australian and New Zealand diabetes societies' proposed minimum standard does contain a row allowing up to 30% of readings in the 3.9–10.0 mmol/L band to miss by more than ±15% — but that document is explicitly about people with diabetes and insulin dosing. It proposes nothing about wellness use. The correct framing is below.

3. The key replication study is a preprint. The duplicate-meal study is on medRxiv and we could find no peer-reviewed version. It is also small. We lead with it anyway because it is the only study that directly tests the premise — but you should know its status.

Does your glucose response to a food mean anything?

The foundational claim

**Zeevi D, Korem T, Zmora N, et al. (senior authors Segal E, Elinav E). "Personalized Nutrition by Prediction of Glycemic Responses." Cell 2015. An 800-person main cohort, a 100-person validation cohort, a 26-person dietary intervention, and 46,898 meals**. A machine-learning model using 137 features including gut microbiome composition.

• Between-person variability in response to identical meals was high — the headline finding
• Algorithm correlation with measured response R ≈ 0.68–0.70, versus 0.38 for carbohydrate counting
• The 26-person intervention found that algorithmically "good" versus "bad" personalised diets produced different postprandial responses (p<0.05)

This is a genuinely important paper and it launched the field. The Weizmann Institute exclusively licensed the technology to DayTwo, with Lihi Segal as co-founder and CEO. We could not retrieve the paper's own funding statement or Eran Segal's commercial role, and are not characterising them.

What the paper established: people differ from each other in their glucose responses. What it did not establish: that any individual's response to a given food is reproducible enough to act on.

The duplicate-meal test

Hengist A, Ong JA, McNeel K, Guo J, Hall KD. "Imprecision nutrition? Duplicate meals result in unreliable individual glycemic responses measured by continuous glucose monitors across four dietary patterns in adults without diabetes." medRxiv 2023, doi:10.1101/2023.06.14.23291406.

n=30 adults without diabetes. Abbott Libre Pro worn by 14, Dexcom G4 Platinum by all 30. Participants ate duplicate meals across four dietary patterns. Funded by the NIH Intramural Research Program (NIDDK) — no commercial funder. Status: preprint; we found no peer-reviewed publication.

Test–retest of the identical meal:

Identical meals versus different meals — standard deviation of response:

The authors' conclusion: "Individual postprandial CGM responses to duplicate meals were unreliable in adults without diabetes."

An ICC of 0.14 means essentially no reproducibility. The Bland-Altman limits mean the same meal could read 30 mg/dL (1.7 mmol/L) higher or lower on a second occasion — which is larger than most of the differences people act on.

Limits, stated plainly: n=30, older sensor models (G4 and Libre Pro rather than G7 and Libre 3), a US metabolic-ward setting, and preprint status. A larger study with current sensors could land differently. But this is the only direct test of the premise we could find, it was publicly funded, and it came out negative.

We could not locate further independent replication work. That is itself notable for a field this commercially active.

Accuracy in the range healthy people live in

What MARD does and doesn't tell you

MARD — mean absolute relative difference — is the average percentage gap between the sensor and a reference blood measurement. The headline figures:

• Dexcom G7: 8.7% in adults against a laboratory reference
• Abbott Libre 3: 7.9% (manufacturer-reported; the methodology differs from competitors', and that difference has been criticised)

Both figures come to us through secondary sources rather than the pivotal papers, which we could not fetch.

Two things to understand about these numbers.

First, they were established in diabetes cohorts, with reference samples spanning hypoglycaemic, normal and high ranges. We could not fetch the pivotal papers to confirm the range breakdown, so treat "established in diabetes populations" as our strong expectation rather than a verified claim.

**Second, MARD is a relative error, and that flatters performance at low glucose.** A fixed absolute error of 0.5 mmol/L is a 3% error at 15 mmol/L and a 10% error at 5 mmol/L. People without diabetes spend nearly all their time in the low part of the range — around 97.8% of time between 70 and 180 mg/dL, per the normative data below. A pooled MARD dominated by high readings is not the error rate you are experiencing. That reasoning is ours, not a sourced finding, but the arithmetic is not controversial.

We could find no MARD at all for Abbott Lingo or Dexcom Stelo — the two products marketed specifically to people without diabetes.

What the diabetes standard actually says

The Australian Diabetes Society, ADEA, ADIPS, ANZSPED and NZSSD published "Strengthening safety through regulatory standardisation for continuous glucose monitoring systems (anzCGM) in Australia and New Zealand" (versions dated December 2025 and April 2026).

Its proposed minimum performance table:

Read the second row carefully: a device can meet this standard while up to 30% of readings between 3.9 and 10.0 mmol/L are off by more than ±15%. At a true 5.5 mmol/L, ±15% is ±0.8 mmol/L. Nearly a third of readings may sit outside that.

But the scope matters, and this is where the claim gets mangled. The document covers people with diabetes — type 1, type 2, gestational and other — and addresses insulin dosing and glycaemic management. It says nothing about people without diabetes or about wellness use. We found no ADS, ADEA or NZSSD standard for non-diabetic CGM.

So the honest statement is: the minimum standard proposed for diabetes use in Australia and New Zealand allows up to 30% of readings in the 3.9–10 mmol/L band to err by more than 15%. That is the band healthy people live in, and that is why the figure is relevant. It is not a standard the societies have proposed for wellness users, and it should not be reported as one.

What normal actually looks like, and why experts can't agree

**Spartano NL, et al. J Clin Endocrinol Metab 2025;110(4):1128–1134. n=1,175 (560 normoglycaemic, 463 prediabetic, 152 diabetic), blinded Dexcom G6 Pro, 7+ days. Funded by NHLBI and NIDDK; Dexcom supplied discounted sensors.**

In people without diabetes:

• Mean glucose 114.5 mg/dL (6.4 mmol/L), SD 11.5
• About 97.8% of time between 70 and 180 mg/dL
• About 12.1% of time above 140 mg/dL — roughly three hours a day
• About 1.2% above 180 mg/dL — more than fifteen minutes a day

Read that again: a completely normal person spends about three hours a day above 140 mg/dL and a quarter of an hour above 180. Spikes are not pathology. They are what a functioning glucose system looks like.

And the experts cannot agree on what to do with them. A companion study by the same group showed 20 challenging Dexcom G6 Pro reports from people without diabetes to 18 expert clinicians, alongside HbA1c and fasting glucose:

• Agreement on who needed follow-up: Fleiss kappa 0.36 — poor
• More than half recommended follow-up for anyone with >2% time above 180 mg/dL, despite normal HbA1c and fasting glucose

The authors' conclusion was that normative data are urgently needed. If 18 specialists reading the same report disagree this much, a consumer reading their own is not going to do better.

(We should flag a citation oddity: one source lists this paper as J Diabetes Sci Technol 2025;20(3):727–735, a volume number that looks wrong for 2025. Verify before citing.)

Do spikes in healthy people predict anything?

We could find no prospective cohort that measured CGM excursions in people without diabetes and followed them for incident diabetes or cardiovascular disease. Our searches found reviews and cross-sectional work, nothing longitudinal. Absence from our search is not proof of absence — but the burden of proof here belongs to the companies selling the monitors.

What does exist is cross-sectional. **Bakhshi B, et al. Diabetes Care, online 17 September 2026**: n=1,356 adults without diabetes or cardiovascular disease, blinded Dexcom G6 Pro, up to 10 days.

• Mean glucose and time above 140 mg/dL associated with 21–26% higher odds of hypertension or dyslipidaemia
• Mean glucose associated with 53% higher odds of elevated 10-year cardiovascular risk score
• Glycaemic variability was not associated with cardiovascular risk

This is association at a single point in time, in which the outcome is itself a risk score rather than an event. It is very recent and we have not verified the author list or details. Independent commentators quoted alongside it made the point that matters: "we don't know how to act on differing glucose patterns."

Note which metric failed: variability — the thing a CGM measures that a blood test can't — showed no association. Mean glucose, which a cheap HbA1c already captures, did.

Fasting glucose and HbA1c are the validated risk markers. CGM-derived excursion metrics are not, and have no agreed action thresholds — which is exactly what the expert-disagreement study above demonstrates.

Does CGM-guided eating work?

The pooled picture

**Richardson KM, et al. Int J Behav Nutr Phys Act 2024;21:145. 25 RCTs, n=2,996. Only 3 (12%) were in people without diabetes**, and those were in overweight or obese participants.

Pooled across all populations:

• HbA1c −0.28% (95% CI −0.42 to −0.15), p<0.001
• Weight −0.7 kg (−1.4 to 0.0), p=0.066 — not significant
• BMI −0.4 kg/m² (−0.9 to 0.0), p=0.080 — not significant

11 of the 25 trials (44%) reported conflicts of interest with CGM companies — Abbott, Dexcom and Medtronic with five each.

And the pooled result is dominated by type 2 diabetes studies, so it does not transfer.

The non-diabetic picture specifically

**Zhong T, et al. Eur J Med Res 2026;31:397. 23 studies including 7 RCTs, 1,074 non-diabetic participants**.

• Weight and BMI in the RCTs: SMD −0.25 (−0.63 to 0.12), p=0.19 — not significant
• HbA1c: improvements in prediabetes at 1 and 2 years (p=0.007, p=0.033)
• "No appreciable glycemic benefit" in healthy normoglycaemic people
• Mean glucose: SMD −0.54 (−1.02 to −0.07), p=0.03

The authors' framing is worth adopting: CGM functions as "a precision biofeedback tool integrated within structured lifestyle programs," not as a standalone weight-loss intervention.

**Basiri R, Cheskin LJ. Nutrients 2024;16(23):4005 is the kind of trial that gets cited as proof. n=30 (15 vs 15), overweight or obese adults with prediabetes, 30 days**, funded by a university public health college. Mean glucose fell 129.1→121.6 mg/dL versus 131.1→129.5; GMI 6.4→6.2% versus 6.4→6.4% (p=0.02); time in range 95.1→97.9%.

Time in range was already 95% at baseline. Both arms received nutrition therapy. There was no weight outcome and no clinical outcome, and it ran for a month. It is a pilot, and it should be described as one.

How many RCTs exist in healthy normoglycaemic people? We could not establish a precise count. Both reviews suggest very few, all small.

The Australian market and what regulates it

Vively

Vively's co-founder has described the company as "exempt from the TGA, similar to a Fitbit or Oura ring" as a wellness app, with approval to be sought later for specific healthcare applications. That quote reaches us via a summary of an SBS article; get the original wording before relying on it.

Vively's own website states: "Vively is a general wellness platform and does not provide medical advice, diagnosis, or treatment."

To its credit, Vively's own blog says something more candid than most of the marketing in this category: "CGMs can have error margins of up to 20%, and there's no agreed-upon standard for what constitutes abnormal glucose levels in non-diabetic individuals." That is an accurate statement of the problem, published by a company selling the solution.

Its medical director has told the ABC that the research is "still early."

Pricing, 1 October 2026 — and the pages conflict. The homepage advertised a CGM program at $99 per 14-day sensor on a promotion listed as running "through September 7," which may have expired, plus a $99 baseline health check. An undated help page lists sensors at $199 for one, $358 for two, $499 for four on an auto-renewing annual plan. Check live before committing. The underlying device brand is not stated on the pages we read.

Abbott Lingo

Sold in Australia through Withings at A$130 for 2 biosensors and 2 applicators (about four weeks). The Australian page showed sold out when we checked.

For adults 18 and over not on insulin, no prescription. The page states it is not recommended for people with problematic hypoglycaemia or a history of eating disorders — a caution worth taking seriously given the harms section below.

We could not confirm Lingo's TGA or ARTG status. If it matters to you, the register is searchable at tga.gov.au.

Everything else

Levels is a US service with a mandatory US$199 annual fee plus device costs. Nutrisense offers 3-, 6- and 12-month plans; Australian availability unconfirmed. Dexcom Stelo — we could not confirm Australian availability or pricing.

The ABC put unsubsidised CGM costs at roughly A$200–300 a month in August 2025, which matches the Vively annual-plan arithmetic.

For contrast: the National Diabetes Services Scheme subsidises CGM for specific groups — adults with type 1 diabetes with recurrent severe hypoglycaemia or impaired awareness, and women with type 1 who are pregnant or planning pregnancy are among those named in the professional guidance. We did not verify current NDSS eligibility directly and you should check ndss.com.au rather than relying on this summary. The point stands regardless: the people for whom CGM has established clinical value get help paying for it, and the people buying it as a wellness product pay full price.

Harms

**Brown A, Pemberton J, et al. Diabetic Medicine 2024 — a narrative review of 25 studies spanning 1980 to 2023. It concluded that CGM use in people without diabetes "could cause anxiety about what is normal in terms of diet and blood sugar levels," with "a risk of developing eating disorders, such as orthorexia." Pemberton added that accuracy discrepancies can cause "unintended stress and potential psychological and behavioural implications."** We could not retrieve the full journal citation and are relying on the institutional press release.

The expert commentary is unusually blunt for this field:

• Bernard Hansel (Hôpital Bichat): "I am totally opposed to the normalization of the use of monitors... it is not necessarily beneficial."
• Partha Kar (NHS national diabetes adviser): there is "no strong evidence that CGMs help people without the condition."
• Amy-Lee Bowler (nutrition lecturer, via SBS): users may become "obsessed with the data" and restrict food too far. The term "glucose rexia" is reportedly in some clinical use.
• Jessica Weiss (Diabetes WA, via ABC): "our body does an excellent job of regulating glucose," and minor spikes and dips are "normal and very safe."

Vively disputes seeing this pattern among its own users.

On the specific worry that people will cut out fruit and wholegrains because of a glucose spike: this is the mechanism the experts above are gesturing at, and it is plausible — a bowl of oats or a banana will produce a visible rise in a healthy person, and the app will flag it. But we found no study demonstrating this behaviour change, and no published case reports of eating disorders triggered by CGM in people without diabetes. We are presenting it as a well-reasoned concern raised by clinicians, not a documented harm. If anyone has run that study, we could not find it.

Two harms we did not research: adhesive and skin reactions, and the consequences of acting on a false low or high reading. We have no data on either.

And the harm that is documented: the expert-disagreement study. If 18 specialists cannot agree whether a given report warrants follow-up, a consumer interpreting their own data is making a judgement that the field has not equipped anyone to make.

Who might actually benefit

We want to be honest about where the evidence is absent rather than treating absence as a verdict.

Prediabetes — the best signal there is. The Zhong review found HbA1c improvement in prediabetes subgroups at one and two years, from a small number of small trials. The Basiri pilot points the same way. If you have a diagnosis of prediabetes and you are using a CGM inside a structured programme with a dietitian, that is the use case with the most support behind it — which is still not much.

Healthy normoglycaemic people. "No appreciable glycaemic benefit," per the same review. The remaining argument is educational value, which is real but should not be sold as a health outcome. Two weeks of wearing one to learn what your body does is a defensible purchase. A subscription is harder to justify.

Endurance athletes. A sports dietitian quoted by SBS called the evidence "inconclusive." There is a review on CGM in sport from the Australian Catholic University that we did not read. Verdict: unknown, not disproven.

Reactive hypoglycaemia. This is the use case where a CGM is doing something a blood test cannot — catching a transient low that correlates with symptoms. We found no evidence either way, and we are not going to pretend that absence of evidence is evidence of uselessness. If you get reliable symptoms a couple of hours after eating, a short period of monitoring to see whether your glucose is actually dropping is a reasonable thing to discuss with a GP.

Family history of diabetes. No evidence found. An HbA1c and fasting glucose are cheap, validated, Medicare-rebatable and have decades of outcome data behind them. Start there.

What the professional bodies say

The short answer: almost nothing, which is itself informative.

• Diabetes Australia's position statement on glucose self-monitoring does not address CGM in people without diabetes.
• The ADS/ADEA/NZSSD anzCGM standards are silent on non-diabetic use. We found no ADS position statement on it.
• The RACGP — we found a 2023 response on NDSS products and no position on non-diabetic CGM.
• The ADA — we did not find a statement specific to people without diabetes.

The Spartano group, whose co-authors include many of the field's leading clinicians, called for normative data and risk-progression research. That is a field saying it does not yet know what its own measurements mean in healthy people.

What we'd tell someone about to spend $200 a month

1. Get an HbA1c and a fasting glucose first. They are validated, cheap, Medicare-rebatable with a GP's request, and backed by decades of outcome data. CGM-derived metrics have none of that.

2. Know that spikes are normal. A healthy person spends about three hours a day above 140 mg/dL. If you buy a monitor expecting a flat line, you will find "problems" that are not problems.

3. Don't trust a single reading, or a single meal. In the duplicate-meal study, the same meal read up to 1.7 mmol/L differently on a second occasion, and the reproducibility was ICC 0.14–0.31. The "this food spikes me" conclusion is mostly noise.

4. If you buy one, buy it once. Two weeks of wearing a sensor to see what your body does is a defensible curiosity purchase. An ongoing subscription buys you a measurement with no validated action threshold.

5. Don't cut out whole foods because of a curve. No clinician we found recommends this and several specifically warn against it. Oats, fruit and legumes producing a visible rise in a healthy person is physiology working.

6. If you have any history of disordered eating, don't. Abbott's own Lingo page says the same thing, and so does the published review.

7. If you have prediabetes, use it inside a programme, not instead of one. That is where the only positive evidence sits, and the evidence is for the programme with CGM attached, not CGM alone.

Frequently asked

Are CGMs accurate in people without diabetes? Less than the headline MARD figures suggest. Those figures — about 8–9% for current sensors — were established in diabetes cohorts across the full glucose range, and MARD is a relative error that flatters performance at the low end where healthy people live. The minimum standard proposed for diabetes use in Australia and New Zealand allows up to 30% of readings between 3.9 and 10 mmol/L to be off by more than 15%. There is no published accuracy data at all for Abbott Lingo or Dexcom Stelo, the two products marketed to people without diabetes.

Can a CGM tell me which foods are bad for me? The evidence says no. A publicly funded study fed 30 adults without diabetes duplicate meals and found that the glucose response to the identical meal was no more consistent than the response to different meals, with intraclass correlations of 0.14 and 0.31. The same meal read up to 1.7 mmol/L differently on a second occasion. It is a preprint and it is small, but it is the only direct test of the premise we could find.

Is it bad to have glucose spikes? In a healthy person, spikes are physiology. Blinded monitoring of 560 normoglycaemic adults found they spent about 12% of the day — roughly three hours — above 140 mg/dL, and about 1.2% above 180. No prospective study we could find has shown that excursions in people without diabetes predict any outcome, and the one large cross-sectional study found glycaemic variability unassociated with cardiovascular risk.

Will a CGM help me lose weight? Not on its own. A meta-analysis of seven RCTs in 1,074 non-diabetic participants found no significant effect on weight or BMI (SMD −0.25, p=0.19). The broader pooled estimate across all populations found weight change of −0.7 kg, which did not reach significance. The authors of the non-diabetic review described CGM as useful inside a structured lifestyle programme, not as a standalone intervention.

Are these TGA-approved? Not necessarily, and that is the point. At least one Australian provider has described itself as exempt from TGA regulation as a wellness app, comparable to a fitness tracker, and its website states it does not provide medical advice, diagnosis or treatment. We could not confirm the regulatory status of individual products and recommend searching the ARTG at tga.gov.au if it matters to you. Products subsidised for type 1 diabetes through the NDSS are in a different category entirely.

Can wearing a CGM cause problems? A narrative review of 25 studies concluded it could cause anxiety about what is normal and carried a risk of disordered eating, including orthorexia. Several senior clinicians have warned publicly about glucose surveillance in healthy people. Abbott's own consumer product page advises against use by people with a history of eating disorders. There are no published case reports we could find, so the concern is clinical judgement rather than documented harm — but it is coming from the people who treat both conditions.

What should I do instead? An HbA1c and fasting glucose through your GP. They cost a fraction of a CGM subscription, they are the markers every risk calculator and treatment guideline is actually built on, and if they come back abnormal you will be dealing with a number that has an agreed action threshold behind it.

Last reviewed 1 October 2026. Commercial conflicts are flagged on every study where the source discloses them. Prices are Australian retail at the date of review and at least one provider's pricing pages contradicted each other on the day we looked. Where we could not verify a figure or find a study, we have said so rather than filling the gap — and three widely circulated claims about this topic, including two in our own earlier notes, turned out to be misreadings and are corrected above. This page is general information, not medical advice.