Continuous glucose monitoring outside diabetes, and what the evidence will support

Consumer continuous glucose monitoring sits on a real scientific foundation and is routinely marketed several steps ahead of it. This brief sets out where the line currently falls, so that product, clinical and marketing teams can work from the same read of the evidence rather than negotiating it claim by claim at the end of a campaign.

It is written for a non-specialist internal audience. It is not clinical guidance, and any health claim drawn from it should be reviewed by a qualified clinician before publication.

Summary for decision makers

Three findings drive everything below.

1. Individual variation in postprandial response is real, large and well documented. This is the strongest plank under personalised nutrition and it is genuinely impressive science.
2. The evidence that wearing a sensor changes behaviour or improves outcomes in people without diabetes is thin. Not negative, thin. The trials have mostly been done in other populations.
3. Interpretation standards for non-diabetic glucose traces are not settled. There is no agreed definition of what a normal, healthy trace looks like, which constrains what an app can responsibly tell a user their data means.

What the evidence supports

Postprandial responses differ substantially between people eating identical meals

The PREDICT 1 study, published in Nature Medicine in 2020, measured responses in 1,002 twins and unrelated healthy adults in the United Kingdom. Following standardised, identical meals, it found large between-person variation, reported as 103 percent for blood triglyceride, 68 percent for glucose and 59 percent for insulin.¹

This is a solid, well-powered result from a well-designed study, and it is the finding a personalised nutrition product is entitled to build on. Two people eating the same thing really do respond differently, and the difference is large enough to matter.

Genetics explains less of that variation than people assume

The same study reported that genetic variants had a modest impact on predictions, 9.5 percent for glucose.¹ This is a useful point for a company whose product measures rather than sequences, because it implies that a genotype-only approach would miss most of what drives the response. It should be stated as what it is, a finding about the predictive contribution of genetic variants in that model, rather than inflated into a claim that genetics does not matter.

CGM-guided feedback improves glycaemic measures where it has been tested

A 2024 systematic review and meta-analysis in the International Journal of Behavioral Nutrition and Physical Activity pooled 25 randomised controlled trials with 2,996 participants. CGM-based biological feedback reduced HbA1c by 0.28 percent, with a 95 percent confidence interval from 0.15 to 0.42, and increased time in range by 7.4 percent.²

That is a genuine effect. The critical qualifier is the population it came from.

What the evidence does not yet support

That these results transfer to people without diabetes

In the same meta-analysis, most participants had type 2 diabetes. Only 3 of the 25 included studies, 12 percent, were conducted in populations without diabetes.² A pooled HbA1c effect derived overwhelmingly from people with diabetes cannot be presented as evidence of benefit in a metabolically healthy consumer, and HbA1c is in any case a weak outcome in a population whose HbA1c is normal to begin with.

The authors say so themselves, noting that further research in populations without diabetes is needed and that other outcome measures, including glycaemic variability and behaviour change, may be more appropriate in that group.²

That wearing a sensor reliably changes what people eat

This is the mechanism the entire consumer category depends on, and it is the least well evidenced part of the chain. The review found that only 4 of 25 studies evaluated the effect of CGM on dietary change and only 5 of 25 evaluated physical activity, which was too inconsistent to support a meta-analysis of the behavioural pathway at all.²

The honest position is that behaviour change is a plausible and largely untested mechanism. A company can say that it is what the product is designed to produce. It cannot yet say that the literature demonstrates it.

Behaviour change is the mechanism the entire consumer category depends on, and it is the least well evidenced link in the chain.

That there is an agreed standard for interpreting a healthy person's trace

Clinical targets such as time in range were developed for diabetes management. Applying them to someone whose glucose regulation is normal produces a number with no established clinical meaning, and the absence of agreed interpretation guidance for non-diabetic CGM reports is an active topic in the field rather than a settled one.³ Any in-app scoring of a user's trace is therefore a product design decision presented as a measurement, and should be labelled accordingly.

Claim by claim

The following formulations are supportable as written.

• People vary substantially in how their blood sugar responds to the same meal, with variation of around 68 percent reported between individuals in a study of over a thousand adults.
• Genetic variants explained a modest share of that variation, which is part of why measurement adds information that a genetic test alone does not.
• In people with diabetes, feedback from continuous glucose monitoring has been shown in randomised trials to improve glycaemic control.

The following are not supportable on the current evidence and should be rewritten or removed.

• Any claim that CGM improves health outcomes in people without diabetes.
• Any claim that the product has been clinically proven to change eating behaviour.
• Any claim that a glucose spike in a healthy person indicates harm, disease risk or metabolic dysfunction, in the absence of an agreed interpretive standard.
• Any use of time in range or similar diabetes-derived targets as a health score for a non-diabetic user without stating that these thresholds come from diabetes care.

The gaps worth funding

If the goal is to be able to make stronger claims in two years than today, three studies would do most of the work.

1. A randomised trial in people without diabetes with a behavioural primary outcome. Measured dietary change, not HbA1c, over a period long enough to see whether early engagement persists.
2. A retention and durability study. The commercially relevant question is whether any change survives sensor removal, and almost nothing addresses it.
3. Interpretation validation. Establishing what a given pattern in a healthy adult's trace actually predicts would convert the product's central feature from a design choice into a measurement.

Recommendations

Four, in order of how much risk they remove.

1. Separate the two evidence bases in all external copy. The personalised response science is strong. The wearable-changes-behaviour science is not. Marketing that fuses them inherits the weaker one's exposure.
2. State population and outcome whenever a trial is cited. "In people with type 2 diabetes, HbA1c fell by 0.28 percent" is defensible. The same figure without its population is not.
3. Label interpretive features as guidance rather than diagnosis. Scores, grades and spike warnings are product logic and should read as such until interpretation standards exist.
4. Put a clinician's name on anything a user might act on. Named clinical review is both the regulatory safeguard and, increasingly, a competitive signal in a category where readers have learned to be sceptical.

None of this makes the category weaker. A product that states exactly what its evidence supports and exactly where the research is still open is more credible to an informed buyer than one asserting a benefit the literature has not yet tested.