The continuous glucose monitor has spent the past few years moving out of the diabetes clinic and into the wellness aisle, where a small disc on the back of an arm now produces a curve that people photograph and post. The pitch usually arrives with rules attached: start with vegetables, never eat fruit alone, keep the line flat. Those rules are mostly about a single meal.
Two papers published three weeks apart look at that question from different distances. The first is a small observational study of what actually happened across consecutive meals in everyday life. The second is a mapping of every peer-reviewed study of glucose monitors in people with prediabetes. Together they sketch the device’s real state of evidence: it produces a signal, the signal is not yet a verdict, and the most confident advice circulating about it runs ahead of both.
What the monitor saw across 342 meals
The free-living study, published in Nutrients on 7 November, is a secondary analysis of one arm of a weight-loss trial run at the National University of Singapore — an analysis the authors describe as observational and not pre-specified. Fifty-five adults with a body mass index at or above 23 wore a Dexcom G6 sensor for ten days and photographed their food in an app that matched each entry to a national nutrient database. Median age was 27, just over half were women, and the median BMI was 26.4.
From 1,135 logged eating occasions the authors built 342 usable “pairs” — one eating occasion followed by the next on the same day, with no third occasion intruding on the two-hour window after the second. Each participant contributed a median of five, and the most frequent sequence was lunch into dinner.
The analysis then asked a narrow question: holding the second meal’s own carbohydrate, energy and starting glucose constant, did anything about the first meal predict the size of the glucose rise after the second? Twelve features of the preceding meal were tested one at a time, each scaled to a one-standard-deviation difference within the same person.
Three associations, and the one that faded
Three came out. A preceding meal with a higher share of energy from total sugar was followed by a next-meal glucose response 29.6% larger (95% confidence interval 16.9% to 45.2%). A longer gap between the two meals was followed by a larger response as well: an extra 1.9 hours — one standard deviation in this sample — went with 29.1% more (15.0% to 48.8%). A preceding meal with more of its energy from protein went with a smaller response, 13.9% less (−23.1% to −4.2%), but that estimate did not survive an extra adjustment for the second meal’s own protein, fat and fiber, so the authors treat it as the least secure of the three conclusions.
Saturated fat pointed the same way (+14.8%) with an interval crossing zero, and did not clear the false-discovery threshold set for twelve simultaneous tests. Everything else was flat: meal duration, meal energy, glycemic load, the preceding meal’s own glucose response, weighted glycemic index, fiber density, carbohydrate share and total fat share.
That glycemic index and glycemic load did nothing is the result least likely to appear in a social-media explainer, and the authors say so: the sugar association “was not accompanied by similar associations for glycemic index, glycemic load, carbohydrate percentage, or preceding-meal iAUC.” Total sugar may be standing in for something they could not isolate. Added sugar was not separated from sugar naturally present in food, and fructose intake, hepatic glucose production and insulin sensitivity were not measured — the mechanisms that would make the finding biological rather than statistical.
One outside check was run against a separate public dataset of 615 meal pairs from 45 participants. The meal-gap association held there, at a similar magnitude. The protein association did not.
Why a longer gap tracked a bigger rise
The gap finding has the deepest experimental literature behind it. A randomised crossover in 26 healthy adults found delaying lunch by two hours raised the post-lunch peak. Breakfast-omission studies in type 2 diabetes found much smaller post-lunch responses when breakfast had been eaten, with free fatty acid concentrations before lunch tracking the size of the response. A high-protein breakfast has been shown to lower the glucose response to bread eaten four hours later.
The mechanism proposed is that the previous meal’s insulin and incretin release, and the suppression of circulating fatty acids, blunt the next meal’s excursion — which would mean a longer fast removes that brake. None of those hormones were measured here, and the authors are careful about what a long gap means in a free-living dataset: it may also mark a missed meal, an unusual bout of activity, or food that was eaten and never logged.
They also separate this acute effect from time-restricted eating, a different question on a different timescale: repeated daily eating windows have generally improved insulin sensitivity in trials, while extending an overnight fast from 12 to 16 hours raised the breakfast glucose response in one small study, in women but not men. An observational study of meal spacing cannot adjudicate between the two.
What has actually been tested in prediabetes
The second paper is a scoping review, published in Frontiers in Endocrinology on 28 October, that searched PubMed and mapped 48 peer-reviewed studies of continuous glucose monitoring in adults with prediabetes. It is a map rather than a verdict — no meta-analysis, no quality scoring — and what it shows is a field of small, short studies with no answer to its central question.
Twenty-nine of the 48 studies tested some kind of CGM-guided intervention, with a median of 33 prediabetes participants (range 6 to 253). Thirty-one percent ran for a week or less, 41% for one to thirteen weeks, and two for longer than half a year. Across them a pattern recurred: monitor measures such as time in range improved, while glycated hemoglobin and fasting glucose often did not. The authors offer three explanations — most trials were too short for an HbA1c change to register, some enrolled people already at near-ceiling control (one began at 99.5% time in range), and real-time feedback designs were sometimes compared against blinded ones.
Only one included trial followed participants for two years, and it was the only one reporting a lasting between-group difference in HbA1c. Standardisation is poor: most used the consensus 70–180 mg/dL band for time in range, but several substituted tighter bands (70–140, 54–140) without saying why. One study found professional CGM disagreeing with venous glucose from an oral glucose tolerance test by a clinically meaningful margin. Another, with 8,315 participants, found single fasting glucose measurements classifying people differently once repeated CGM data were considered.
And the question that would matter most to anyone buying a sensor for prevention has not been studied at all: none of the 48 used CGM to predict who progresses to type 2 diabetes. The review calls that consistent with current professional guidance, which holds the evidence insufficient to support CGM for the purpose, and puts the first over-the-counter sensor cleared for prediabetes at roughly 89 to 99 US dollars a month.
The number on the screen is not a diagnosis
Part of the confusion is definitional. The thresholds that define normal, prediabetes and diabetes come from fasting blood draws, a two-hour oral glucose tolerance test and HbA1c — not from the peak of a post-meal curve. The National Institute of Diabetes and Digestive and Kidney Diseases lists normal as A1C below 5.7%, fasting plasma glucose of 99 mg/dL or below and 139 mg/dL or below two hours into a tolerance test; prediabetes as 5.7% to 6.4%, 100 to 125 and 140 to 199 respectively. On self-testing it is blunt: “Blood testing equipment that you can buy over the counter, such as a blood glucose meter, cannot diagnose diabetes.”
The same institute describes monitors as something a doctor may recommend “if you need insulin to manage type 1 diabetes, type 2 diabetes, or another form of diabetes” — narrower than the consumer market now implies.
For scale, the free-living study’s own participants — adults with excess weight, not diabetes — averaged 113.5 mg/dL, spent 90.3% of readings between 70 and 140 mg/dL, and peaked at 8.9 mmol/L (about 160 mg/dL) after meals. MedlinePlus notes about one in three adults has prediabetes and that lifestyle changes can delay or prevent progression, which is why the tests that diagnose it still start in a clinic.
What we did not find
No randomised trial has manipulated the interval and composition of a previous meal while standardising the next one, which is what the Singapore authors say is needed before any of their associations can be read as advice. No study has shown that glucose monitoring in people without diabetes changes a health outcome. There is no validated cut-off for a “spike” in a person without diabetes. The fiber null here sits awkwardly beside the population-level fiber gap, a separate argument about diet quality rather than about any one meal’s curve.
What the two papers support is narrower than either the ads or the dismissals: what you ate before appears to shape what happens next, a longer gap between meals appears to sharpen the next rise, and the monitor records both without yet telling anyone what to do about it. For questions about meal timing and late eating, the practical guidance has not moved much — our look at eating late covers the ground that the evidence does support. And as with anything touching blood glucose, a conversation with a healthcare professional beats a screenshot, as our medical disclaimer sets out.
