Summary · What the studies show
Nutrition · Metabolism · Diabetes · Everyday Practice · Evidence BriefCrossover trials and four pooled reviews, 2013 to September 2026
Vegetables and protein first, starch last
Reordering a Meal Changes Its Glucose Curve — Acutely
Across trials in five countries, eating vegetables and protein before starch consistently lowers the glucose spike that follows a meal — and the largest of them, published this year, beat even an arm allowed to eat with no time limit at all. Whether that acute effect adds up to a long-term benefit has a narrower answer than it did: pooled across 17 randomised trials it moves long-term blood sugar by 0.16 percentage points, which the reviewers themselves call minimal.
- ↓73%glucose iAUC reduction over 2 hours, 11-person trial (T2DM)
- ↓42.7mg/dL lower blood glucose an hour after the meal, pooled across 17 randomised trials
- −0.16percentage points of HbA1c, the whole long-term effect in 389 patients
- n=33the largest trial yet: 3 days, continuous monitoring, 4 eating orders compared
The Acute Effect replicated in five countries, and now pooled
| Standing | What the evidence shows | Source |
|---|---|---|
| Pooled across randomised trials | Carbohydrate last lowers blood glucose by 42.7 mg/dL an hour after the meal and 13.0 mg/dL at two hours, raises GLP-1 and lengthens gastric emptying by 28 minutes. | 17 trials, 389 people (2025) |
| Largest single trial | Vegetables, then protein, then carbohydrate gave the lowest peak, the latest peak, the most time in range and the least variability — beating every other order, including eating with no time limit at all. | 33 women, 3 days of monitoring (2026) |
| Replicated independently | The same direction in New York, Singapore, the United Arab Emirates, Korea and Japan — in people with type 2 diabetes and in healthy adults alike. | Five crossover trials, 11 to 33 people each |
| Measured directly | GLP-1, the gut hormone behind the effect, rose 38% when carbohydrate came last, in the one trial that sampled it every half hour for three hours. | 16 people with type 2 diabetes (2017) |
The Long-Term Question answered, and the answer is small
| Standing | What the evidence shows | Source |
|---|---|---|
| Best evidence | Long-term blood sugar falls by 0.16 percentage points — statistically real, and described by the reviewers as a minimal effect in mild type 2 diabetes. | 17 randomised trials, 389 people (2025) |
| Superseded | An earlier pooling of 8 trials found 0.21 points with an interval that crossed zero, and concluded there was no evidence of benefit beyond ordinary dietary advice. | 8 trials, 230 people (2022) |
| Much larger, much weaker design | One Kyoto clinic reports a 1.0-point gap after five years. Its two groups chose themselves, its comparison group was told to eat vegetables first as well, and its own authors attribute the fall to the rest of the diet changing. | 138 against 104 patients, 5 years (2022) |
| Not established | No trial in this set reports a heart-attack, stroke or survival outcome. Everything measured is a blood marker or a weight. | All 20 sources on this page |
Timeline · The published record, 2013–2026
Thirteen Years, and What Each Added in order of publication
- 2013Imai, Fukui and Kajiyama review a 2.5-year, non-randomised comparison of 333 Japanese outpatients with type 2 diabetes taught to eat vegetables first, against routine care.
- 2015Shukla et al. (Weill Cornell Medicine) publish an 11-person pilot: eating vegetables and protein before starch cuts the 2-hour glucose curve by 73%.
- 2017The same group publishes a 16-person, 3-arm trial measuring GLP-1 directly: carbohydrate-last raises GLP-1 by 38% and lowers glucose iAUC by 53%.
- 2019Sun et al. publish a 16-person crossover in Singapore: vegetable-first orders again blunt the glucose response, in healthy adults.
- 2020Kubota, Yabe and colleagues publish a mechanism review: protein/fat raise GLP-1 and slow gastric emptying; fibre slows absorption without raising GLP-1.
- 2022Okami et al. publish a GRADE-rated systematic review of 8 trials, 230 patients: no clear long-term HbA1c benefit beyond standard dietary advice.
- 2022Nitta et al. follow the same Kyoto cohort to five years: HbA1c 8.5% to 7.6% in the educated group, 7.9% to 8.0% in the comparison group — and vegetables up, rice and sweets down.
- 2023Shukla et al. publish a 45-person, 16-week randomised pilot in prediabetes: significant weight loss, a non-significant HbA1c trend.
- 2024Shaheen et al. publish an 18-person UAE crossover: vegetables/protein first again lower glucose and insulin, in healthy adults.
- 2024The first study in pregnancy: 25 women with gestational diabetes eat two weeks their usual way, then two weeks vegetables first. Post-meal glucose falls about 6%.
- 2025Touhamy, Shukla and colleagues publish a real-world continuous-glucose-monitoring letter on carbohydrate-last eating.
- 2025Deguchi et al. time 41 people eating: what decides how long a meal takes, and how much it is chewed, is the meal itself — not whether the vegetables came first.
- 2025Higuchi et al. follow 18 healthy young women across whole days, splitting the carbohydrate over five meals as well as ordering each one: swings in blood sugar fall by about a quarter. Two changes at once, so the order’s own share is not separable.
- 2025Saldarriaga-Callejas et al. pool 17 randomised trials and 389 patients: glucose sharply lower after the meal, GLP-1 higher, gastric emptying slower — and HbA1c 0.16 points lower.
- 2026Kim, Jang and Lee publish a systematic review of 6 studies, 107 healthy adults, finding the same acute pattern.
- 2026A review in Belfast reads six studies in people with diabetes and reaches the same direction, rating the certainty of the evidence low.
- 2026Choi et al. run the largest trial yet: 33 young women, four eating orders, three consecutive days, continuous monitoring. Vegetables first wins on every measure, including against unrestricted eating.
The argument · Why order changes the curve
Why Order Changes the Curve the mechanism, and its limits
The explanation repeated across these trials is hormonal and mechanical, not a matter of which food is eaten. Protein eaten first stimulates the gut hormone GLP-1, which slows the stomach’s emptying and prompts insulin when the carbohydrate finally arrives; fibre eaten first slows absorption in the intestine without touching GLP-1 at all. The two are separate mechanisms that add up, which is why vegetables and protein before starch does more than either alone. Until recently both halves rested on single trials of a dozen or so people. Pooling 17 randomised trials now puts numbers on them: GLP-1 higher by 8.21 pmol/L, and the stomach taking 28 minutes longer to half-empty. One thing the mechanism is not is eating more slowly. When 41 people were timed and their chews counted, what decided how long a meal took was the meal — a pizza against a rice-and-broccoli lunchbox — and not whether the vegetables came first or last.
An Acute Effect Is Not a Long-Term One
A 73% lower curve over two hours and a 0.16-point move in a marker averaged over months are answers to two different questions, measured in two different ways. Both are on this page; neither is evidence for the other.
Why the Five-Year Figure Is So Much Bigger
This article’s own reading, not a finding: a study that lets the whole diet change measures eating order plus everything it leads to, while a trial that holds the meal fixed measures the order alone. The Kyoto cohort records its patients eating more vegetables and fewer sweets over those five years, and its authors say that is where the improvement came from.
What others add · The trials, side by side
The Acute-Effect Trials, Side by Side five trials, four countries
| Trial | Participants | Finding |
|---|---|---|
| New York, 2015 | 11 with type 2 diabetes | Glucose curve 73% lower over 2 hours; insulin 48.5% lower. |
| New York, 2017 | 16 with type 2 diabetes | Glucose curve 53% lower over 3 hours; GLP-1 38% higher. |
| Singapore, 2019 | 16 healthy adults | Every vegetable-first order blunted the response; vegetable, meat, rice released the most GLP-1. |
| United Arab Emirates, 2024 | 18 healthy adults | Glucose curve 40.9% lower; insulin 31.7% lower; rated fuller at 1 and 2 hours. |
| Korea, 2026 | 33 young women at raised risk | Three days of continuous monitoring: lowest peak, most time in range, least variability of four orders. |
The Long-Term Question, Argued Both Ways one clinic, 17 trials, and the distance between them
Five-year cohort
1.0
Pooled randomised trials
0.16
What separates them
6×
Pregnancy: The Newest Evidence, and the Weakest three studies since 2024
The fastest-growing part of this literature is gestational diabetes, and it is where the designs are weakest. The first study, in 2024, put 25 women through two weeks of their usual diet and then two weeks of vegetables first, and reported post-meal glucose about 6% lower — with no comparison group at all. The second, in 2025, compared 54 women in two arms and reported clearly lower post-meal glucose in the food-order arm; its abstract calls the allocation randomised and its methods section says it was not, and every outcome about the birth itself — mode of delivery, shoulder dystocia, intensive-care admission, birth weight — came back statistically non-significant, although the abstract lists them as results. The third, in 2026, is a record of 517 Japanese women whose dietitians taught meal order together with splitting the day’s carbohydrate across five meals, which is two changes at once. Post-meal glucose in pregnancy is the one claim this strand supports. Nothing here is evidence about babies.
What a Clinician Says It Is Worth
“It’s definitely a hack, but it’s worth trying, and it could be a helpful tool to add to existing strategies for managing blood sugar,” Nancy Oliveira, a registered dietitian who manages the nutrition service at Brigham and Women’s Hospital, told Harvard Health this April. She adds that the research is still preliminary and limited to short-term studies, and that a large portion of refined carbohydrate may cancel the benefit.
Conclusion · What this means, and does not
What the Studies Had People Do a description, not a prescription
- In the trials that tested a structured order, vegetables were eaten first, typically for about five minutes.
- Protein followed — meat, fish, egg or tofu, depending on the trial.
- Starchy staples — rice, bread, noodles — came last, usually within a 15–20-minute total eating window.
- The Kyoto clinic that ran the longest follow-up asked for more than 120 g of vegetables per meal, raw or cooked, with potato, pumpkin and corn counted as starch and eaten last.
- None of the trials studied people with reflux, sarcopenia or stomach ulcers, and no published source recommends a different order for those conditions.
In Summary
Eating vegetables and protein before starch lowers the glucose spike after that meal. That much is well established: five countries, 17 pooled randomised trials, and a mechanism measured directly. Over months and years it moves long-term blood sugar by about 0.16 percentage points in people with mild type 2 diabetes — real, and small. It costs nothing, changes no ingredient, and is one of the few dietary suggestions that asks a person to give up nothing at all.
What This Page Is Not
A summary of published research, not medical advice. No trial cited here enrolled anyone with reflux, sarcopenia or a stomach ulcer, and none reports a heart-attack, stroke or survival outcome.