What Was Measured, and How
The review pooled randomised controlled trials lasting at least 28 days, in adults, all measuring the same thing: flow-mediated dilation. That is the ultrasound test described earlier in this hub — squeeze an artery shut with a cuff, release it, and measure how much the vessel widens in response to the surge of blood. A healthier lining widens more.
Everything here is therefore expressed as percentage points of FMD compared with a control group. The searches ran to August 2024, 133 studies met the criteria, 85 had usable numbers for pooling, and 7,721 people took part across trials lasting from 28 days to two years.
Two deliberate exclusions are worth knowing about. Supplements, protein powders and nutrient extracts were all excluded — this is a review of food. And anything combined with an exercise programme was excluded too, so nothing here is exercise benefit wearing a vegetable costume.
The Seven That Worked
Fourteen food groups and eleven dietary patterns were assessed. Seven showed a consistent improvement. Here they are, largest effect first — and then a column that changes how you should read the first one.
| Food or pattern | FMD change (95% confidence interval) | Trials |
|---|---|---|
| Tea | +3.22 points (2.01 to 4.42) | 4 |
| Mediterranean diet | +2.32 points (1.86 to 2.78) | 7 |
| Cocoa | +1.68 points (0.08 to 3.29) | 15 |
| Soy | +1.58 points (0.32 to 2.83) | 6 |
| Nuts and seeds | +1.39 points (0.17 to 2.60) | 15 |
| Vegetables | +1.37 points (0.34 to 2.40) | 7 |
| Fruit | +0.86 points (0.16 to 1.55) | 24 |
Read the Right-Hand Column First
The temptation is to read the table top-down and start drinking tea. Look across instead.
Tea produces the largest effect in the review and rests on four trials. Fruit produces the smallest and rests on twenty-four.
That is not a reason to dismiss the tea result. It is a reason to hold it more loosely than its position suggests. Small numbers of trials produce unstable averages, and a single unusual study carries far more weight in a pool of four than in a pool of twenty-four. The review's own certainty ratings reflect this: across all seven findings the GRADE assessment came out somewhere between moderate and low, never high.
The range in brackets is where the true effect probably sits. Narrow is confident; wide is not.
Compare the Mediterranean diet (1.86 to 2.78 — tight, and comfortably clear of zero) with cocoa (0.08 to 3.29 — enormous, and its lower edge almost touching nothing at all). Both are reported as positive findings. They are not equally solid findings, and the numbers say so if you look.
By that test the most dependable result in the whole review is not the one at the top. It is the Mediterranean pattern: a decent-sized effect with the narrowest interval of the seven.
The Pattern Beats the Ingredient
There is something worth noticing in that. The Mediterranean diet outperformed almost every individual food on the list, and it is not a food at all — it is a way of eating that contains most of them.
Vegetables, fruit, nuts, seeds, olive oil, pulses, and comparatively little processed meat. Each component shows a modest effect when tested alone. Assembled into a pattern and eaten for months, the effect is larger than most of its parts and far more consistent between trials.
That is an encouraging finding for anyone trying to eat well, because it takes the pressure off any single item. Missing the blueberries in a given week matters much less than the overall shape of what you eat.
What Did Not Make the List
Seven categories showed consistent benefit. Seven did not: oats, coffee, herbs and spices, oils, margarine and butter, dairy, eggs, and meat.
Read that carefully, because it is easy to over-interpret. Absence of a consistent effect on this one measure is not evidence of harm, and it is not the same as proof of no effect. Several of those groups were assessed on very few trials indeed — coffee on two, meat on two, herbs and spices on two. Two trials cannot establish much in either direction.
Oats are the instructive case. Four trials, no consistent FMD signal — yet the effect of oat beta-glucan on LDL cholesterol is among the better-established findings in nutrition. Oats are doing something useful; it is simply not showing up in this particular measurement. A food can be worth eating for reasons a single test does not capture.
FMD is a surrogate marker — a stand-in for what actually matters, which is whether people go on to have heart attacks and strokes. An improvement in it is encouraging rather than conclusive.
This hub has been careful throughout not to convert FMD points into a risk reduction, and that caution applies here too. What the table shows is that these foods change something measurable and relevant in the right direction, in trials lasting a month or more. That is genuinely worth knowing. It is not the same as a promise.
The Weight Loss Result
Now back to the thread the previous article left hanging.
This review found that weight loss of less than 10% of body weight produced no change in FMD at all. Meanwhile several ordinary foods, eaten without any weight loss attached, moved it by one to three points.
That does not overturn the earlier finding that dietary weight loss improves endothelial markers — the two reviews measured different things in different ways, and the earlier one carried very high heterogeneity in any case. But it does sit neatly alongside the odd result described earlier: that the improvement there had no relationship to how much weight anybody actually lost.
Two separate lines of evidence now point the same way. The food appears to be doing the work. The weight change is what tends to happen at the same time.
A great many people improve their diet, feel better, and then judge the whole effort a failure because the scales moved less than they hoped.
Their vessels were very probably responding anyway. The lining is reacting to what arrives in the bloodstream, not to a number on a bathroom floor.
Where Cocoa Sits Now
Cocoa came second in this hub's earlier ranking of interventions, on the strength of a large flavanol-specific analysis. Here it comes third of seven, on fifteen trials, with that very wide interval.
Both things are true, and the difference is what was being tested. The earlier analysis pooled trials of cocoa flavanols at controlled doses; this one pooled cocoa and chocolate products as people actually consume them, which is a much messier intervention. The tighter the thing you test, the cleaner the answer — and the further it sits from a real shopping basket.
The practical reading has not changed: high-cocoa, low-sugar, and the flavanols are what you are there for.
So What Goes on the Plate
Taking the evidence at the weight it deserves rather than the order it appears in:
- Build the pattern first. The Mediterranean shape — vegetables and pulses at the centre, fruit, nuts and seeds, olive oil, plenty of plants, little processed meat — has the most reliable evidence of anything in the review. Get that right and the individual items matter less.
- Vegetables daily, and include the nitrate-rich ones. Rocket, spinach, beetroot, lettuce and celery feed the pathway described at the start of this hub, on top of whatever else they are doing.
- A handful of nuts or seeds. Fifteen trials, a solid effect, and about the least demanding change on this list.
- Soy is legitimate. Six trials and a respectable effect — tofu, tempeh, edamame, soy milk. It tends to be argued about far more than the evidence warrants in either direction.
- Fruit, reliably. The smallest effect and by some distance the best-evidenced, which is a fair trade.
- Tea, cheerfully but without ceremony. The biggest number in the review and the thinnest support for it. It is a pleasant, nearly free thing to do that may well help. Treat it as a bonus rather than a strategy.
- Cocoa as a treat that earns its place. High percentage, low sugar, small amounts.
And the thing that outranks the entire table remains what it was two articles ago: movement. Exercise produced larger and more consistent FMD improvements than any food here. Food and exercise are not competing — but if only one change is going to happen this month, the evidence is clear about which one to pick.
The Honest Caveats
Three, briefly, because this hub does not hide them.
Certainty is moderate to low. Not high, for any of the seven. The GRADE ratings were dragged down by inconsistency between trials, imprecision, and study quality.
Bias was a real problem. The reviewers report that many of the included studies carried either "some concerns" or a high risk of bias. Food trials are hard to blind — a participant generally knows whether they are eating nuts — and that limitation runs through the whole field.
Funding deserves a mention. One of the authors declares a food-industry advisory role and industry funding, disclosed openly in the paper and stated to be unconnected to this work. That is ordinary in nutrition research and not a reason to discard the findings. It is a reason to prefer the results with tight intervals and many trials over the eye-catching one with four.
A 2025 review of 133 trials found seven dietary approaches that consistently improve blood vessel function: tea, the Mediterranean pattern, cocoa, soy, nuts and seeds, vegetables and fruit. Tea produced the largest effect but on only four trials, while fruit produced the smallest on twenty-four — so the ranking by size and the ranking by confidence are not the same list, and the Mediterranean pattern is the most dependable finding in it. Weight loss under 10% of body weight changed nothing, which supports what the previous article suspected: the food is doing the work rather than the weight. Certainty across all of it is moderate to low, FMD is a stand-in for the outcomes that actually matter, and exercise still beats every food on the list.