Plant foods help, and the reason is plainer than any single superfood. Across fruit, vegetables, pulses and grains, people who eat more of these foods live modestly longer and suffer less heart disease. The signal is real but observational: no trial has fed whole diets for decades, so confounding caps how sure we can be. Almost all of it travels through dietary fibre and the overall eating pattern, not through any one food’s special virtue.
You cannot rank fruits or vegetables against each other on outcomes. The per-food data are too thin to separate an apple from a pear; where they resolve at all, leafy greens and citrus carry the tightest signals — not the berries and grapes that get the marketing. Refined grain, meanwhile, is not found harmful: the evidenced carbohydrate harm is free sugars, sugary drinks above all.
The two things people worry about mostly dissolve. For a generally-nourished adult eating normally-cooked food, plant “antinutrients” carry no meaningful net harm — the real exceptions are raw high-lectin beans, which need a full boil, and a few nutrient-status-specific strata. And the sugar in whole fruit is not a problem: it arrives packaged in a fibre matrix, so the free-sugar concern is juice and sugary drinks, not the fruit bowl.
For someone already eating varied whole plant foods, the choice of plant food is a refinement, not a big lever. The levers with room to move are getting fibre up toward the target and cutting free sugars — both reachable many ways.
Most of the evidence is about a pattern, not a single food
Almost every eat more plant food finding rests on the same kind of study: cohorts that follow people who already eat a lot of a whole food group and record who stays healthy. That design watches a pattern, not a food. And the people who eat the most fruit and vegetables differ in dozens of ways at once — they smoke less, weigh less, move more, drink less alcohol. A healthy population is not evidence for any one thing on its plate; the benefit could ride on the fibre, the potassium, the displaced junk food, or on no single component at all -> Is the Food Category Doing Any Work.
A sharp test shows when a food label carries no real information. Compare the spread of outcomes within a category to the spread between categories. If the foods inside the label differ from one another more than the label differs from its neighbours, the category-level number describes no actual food — it is an average over a mix nobody eats -> Is the Food Category Doing Any Work.
Pool 95 cohorts and the fruit-and-vegetable effect is real but shallow. Each extra 200 g/day (about two and a half 80 g servings) tracks roughly 8-10% lower risk of cardiovascular disease and death: all-cause mortality RR 0.90 (0.87-0.93), coronary heart disease 0.92 (0.90-0.94), with a weaker ~3% for total cancer, 0.97 (0.95-0.99) (Aune et al., 2017). But Aune attributes that effect to «a myriad of nutrients and phytochemicals, including fibre, vitamin C, carotenoids, antioxidants, potassium, flavonoids and other unidentified compounds which are likely to act synergistically» (Aune et al., 2017). So the number is genuine, and it is not clean evidence for any single fruit, vegetable, or compound.
That shallow slope is probably a floor, not a ceiling. Diet is measured by asking people, and the error is large enough to flatten real gradients toward no effect -> Measurement Error in Dietary Assessment. The direction is asymmetric: correcting for the error cannot conjure an effect out of a true null, but it can substantially strengthen a real one (Willett, 2012). The three F&V cohorts that did correct for it saw the association grow, not shrink — in China Kadoorie, cardiovascular death per daily portion of fresh fruit moved from RR 0.77 (0.72-0.83) to 0.63 (0.56-0.72) once regression dilution was accounted for (Aune et al., 2017). So a flat or null dose-response is weak evidence of no gradient, and the reported effect is, if anything, understated.
One caveat runs the other way, and it is about study design rather than biology. A fibre supplement can be dosed and blinded where a bowl of beans cannot, so a blindable isolate is a different exposure than the food and can earn a better evidence grade on design alone -> Dietary Fibre and Health, Is the Food Category Doing Any Work.
A final discipline applies before trusting the shape of any of these curves. Where a curve appears to flatten into a protective plateau, or to bend upward into harm at high intake, that arm can be a study-design artifact — reverse causation, confounding, or unequal reporting precision — rather than a real feature. Such an arm has to survive an artifact check before it earns a place in a recommendation -> The U-Shaped Association Artifact.
With that caveat fixed, here is what each plant group’s evidence actually supports, outcome by outcome — and where it simply cannot single out any one food.
What each plant group does, outcome by outcome
An exposure has no single number. Each plant group moves several outcomes by different amounts, with different certainty, and the evidence individuates the food itself only sometimes. The sections below take each group in turn, outcome by outcome, with the interval and the studied range attached to every figure. All of it is observational, dose-response meta-analysis of self-reported eaters — so the ceiling is confounding, and measurement error means a flat result understates rather than overstates a real gradient.
Fruit and vegetables
Aune’s 2017 pooling of 95 cohorts (up to ~2.1 million people) reports per-serving dose-response estimates rather than high-versus-low contrasts. One serving is 80 g; the table is per 200 g/day of fruit and vegetables combined. (Aune et al., 2017)
| Outcome | RR per 200 g/day (95% CI) | I2 | cohorts |
|---|---|---|---|
| Coronary heart disease | 0.92 (0.90-0.94) | 0% | 15 |
| Cardiovascular disease | 0.92 (0.90-0.95) | 31% | 13 |
| All-cause mortality | 0.90 (0.87-0.93) | 83% | 15 |
| Total cancer | 0.97 (0.95-0.99) | 49% | 12 |
So the cardiovascular and mortality endpoints fall about 8-10% per 200 g/day; cancer is the weak arm, roughly 3% per 200 g/day. CHD carries the lowest heterogeneity of the four outcomes (I2 0%).
The curve is steepest at low intake — the first servings carry most of the effect, and the marginal value of the ninth or tenth serving is small and uncertain. Lowest observed risk sits at 800 g/day (10 servings) for the cardiovascular and mortality endpoints, and cancer flattens near 600 g/day. (Aune et al., 2017) Read 800 g/day as the sampling edge, not a demonstrated optimum: it is where the studied range thins out, so it marks a floor-for-most-benefit region rather than a point target (The Underivable Optimum).
A guidance body sets its target the same way. NNR 2023 writes the recommendation as a range — «It is recommended to consume 500–800 grams, or more, per day of vegetables, fruits and berries in total.» (Nordic Council of Ministers, 2023) — whose ceiling lands at 800 g/day, exactly where the cohort data thin, and whose “or more” declines to cap it as harm. A body reading the same observational splines arrives at a range topping out at the studied edge, not a point optimum: that sharpens the caveat above rather than settling it into a proven curve feature -> Fruit and Vegetable Intake and Health.
The subtype cells are too thin to rank one fruit or vegetable against another (n=2-6, wide CIs). Where they do resolve, the tight, steep signals sit with leafy greens and citrus — not the berries and grapes that carry most of the marketing. (Aune et al., 2017)
| Type | CHD | Stroke | Read |
|---|---|---|---|
| Green leafy vegetables | 0.72 (0.64-0.82) | 0.73 (0.57-0.94) | steepest and tightest, both endpoints |
| Citrus fruits | 0.91 (0.86-0.96) HvL | 0.78 (0.69-0.90) | robust, especially stroke |
| Grapes | 0.87 (0.56-1.37) NS | 0.57 (0.34-0.97) | CHD null; stroke on 2 studies only |
| Berries | 1.13 (0.90-1.43) NS | 1.07 (0.79-1.45) NS | point estimate above 1.0 for both |
The strawberry cell (4.66, 1.14-19.03) rests on a single study with an absurd interval and should be discarded, not read as a signal. The pooled fruit-and-vegetable effect does not distribute evenly onto a favourite fruit — where the data can individuate at all, they point at leafy greens and citrus.
State any fruit target as whole fruit. The free-sugars limit excludes intrinsic whole-fruit sugar but includes fruit juice, so a fruit recommendation and a sugar recommendation are about different things (Free Sugars Intake).
Cognition — a weaker, observational arm
Beyond the cardiovascular and cancer endpoints, a separate observational line links fruit and vegetables to lower cognitive-disorder risk in older adults. Zhou’s 2022 meta-analysis of 16 observational studies found that «a high intake of fruits and vegetables was associated with a declined in the prevalence of cognitive disorders (OR: 0.82, 95% CI: 0.75–0.90)» (Zhou et al., 2022), with modest heterogeneity (I2 35.3%). A flavonoid meta-analysis reads the same signal at the component altitude: higher dietary flavonoid intake tracks «lower risk of adverse cognitive events (pooled OR = 0.90, 95% CI: 0.83–0.98, P = 0.01)» (Peng et al., 2026).
These are not two independent lines confirming a plant-to-brain benefit. Flavonoids are a component of fruit and vegetables, so the flavonoid result is the component-side of Zhou’s whole-food signal — the same observational, self-reported, healthy-user signal at two altitudes. So the food-vs-component question stays open: is it the flavonoid, the whole matrix, or confounding by an overall healthy diet? -> Is the Food Category Doing Any Work. Both arms are measurement-error-laden and modest, and both fade at the hard diagnosis — Alzheimer’s disease is null in each (Zhou OR 0.88, 0.76–1.01; flavonoids OR 0.90, 0.69–1.17), and flavonoid dementia is null too (OR 0.97, 0.79–1.19) (Zhou et al., 2022), (Peng et al., 2026). Read this as a low-certainty candidate lever, not a demonstrated cognitive benefit -> Dementia Prevention and Modifiable Risk Factors.
Whole versus refined grain
Aune’s 2016 whole-grain meta-analysis (45 cohorts, up to ~705,000 participants) supplies the dose-response the bare guideline numbers lack. The table is per 90 g/day (3 servings). (Aune et al., 2016)
| Outcome | RR per 90 g/day (95% CI) | I2 | cohorts |
|---|---|---|---|
| Coronary heart disease | 0.81 (0.75-0.87) | 9% | 7 |
| Cardiovascular disease | 0.78 (0.73-0.85) | 40% | 10 |
| All-cause mortality | 0.83 (0.77-0.90) | 83% | 11 |
| Total cancer mortality | 0.85 (0.80-0.91) | 37% | 6 |
| Stroke | 0.88 (0.75-1.03) NS | 56% | 6 |
Benefit keeps climbing past 90 g/day. Risk reductions are observed up to 210-225 g/day (7 to 7.5 servings), with all-cause risk lowest at 225 g/day and no plateau or upper harm arm within the data. (Aune et al., 2016) So 90 g/day is a population median and a study-density marker, not an optimum — present it with this effect, interval and shape, never bare.
One outcome the Aune table omits is type 2 diabetes, and the DIfE/Boeing series fills it with the best-graded whole-grain arm in the corpus: RR 0.87 (0.82-0.93) per 30 g/day, NutriGrade high — one of only two high-graded protective cells in that 12-food-group matrix, both belonging to whole grains (Schwingshackl et al., 2017). Most of that benefit is captured by ~50 g/day, and it is not independent of the fibre story below (same overlapping cohort pool).
The NNR 2023 carve-out that allows some refined cereals at high energy requirements (athletes, heavy manual work, high-growth adolescence) is a legitimate stratum permission, not a general finding. Note that NNR blends environmental objectives into its food-group advice, so its cereal recommendations are not pure health findings and are not cited here as such.
Refined grain is not found harmful. SACN pooled cohort data to a flat null — RR 1.00 (0.98-1.01) for
both cardiovascular events and type 2 diabetes, graded No association · Moderate.
(Scientific Advisory Committee on Nutrition, 2015) Aune corroborates: per 90 g/day, refined
grain shows neither benefit nor a measured harm signal (all-cause 0.95, 0.91-0.99; CVD 0.98, 0.90-1.06
NS; CHD 1.13, 0.90-1.42 NS). (Aune et al., 2016) The whole-grain
carve-out in guidance is therefore compositional — whole grain carries more of the active fraction —
not a demonstrated harm of refined grain.
What is that active fraction? Probably fibre. Reynolds tracks whole grain and total fibre in the same studies and finds whole grain acts largely through its fibre, not as a separate lever (Dietary Fibre and Health). So steer by fibre content, not by the word “whole grain.”
That fibre lens exposes a genuine oddity in the grades. SACN rates fibre isolates and gum supplements
Effect · Adequate — its top strength grade — while whole grain is graded lower, Moderate for
cardiovascular disease and Limited for stroke and hypertension.
(Scientific Advisory Committee on Nutrition, 2015), (Scientific Advisory Committee on Nutrition, 2015) This is not a verdict that a supplement beats
food. Isolates can be randomised at a chosen dose and blinded; a fibre-bearing food mostly cannot, so it
is observed rather than trialled. The better grade tracks the better design, not the better food
(Is the Food Category Doing Any Work) — and the isolate’s proven effect is on the LDL surrogate, not
the cohort mortality endpoint.
One caution on the refined carbs are bad intuition: the evidenced carbohydrate harm is free sugars, above all sugary drinks — a different exposure from the starchy refined-grain category. That harm, and the glycaemic-load nuance for the already insulin-resistant, belong to the fruit-sugar section below and are not derived here.
Pulses and legumes
Where SACN measured legumes it found nulls — legume fibre to type 2 diabetes RR 1.01 (0.98-1.04) per 1 g/day, legume fibre to colorectal cancer RR 0.98 (0.94-1.02), non-soy legume to CVD RR 0.96 (0.90-1.03). (Scientific Advisory Committee on Nutrition, 2015) These are weak evidence of absence, not evidence of no effect. The increment is 1 g/day of legume fibre against 7 g/day for total fibre, so the slice is one-seventh as wide and close to uninformative about what happens at real intakes. About fourteen legume outcome cells sit unstudied, so the expectancy test applies: silence here is un-studied, not un-associated.
ESC gives the first food-level number, but it is a surrogate: a daily portion of pulses lowers LDL-C by about 0.2 mmol/L. (European Society of Cardiology, 2021) The paired claim that pulses are associated with lower CHD is stated without a magnitude or an interval, and only the surrogate limb carries a number (Surrogate Outcomes).
The DIfE/Boeing 12-food-group dose-response series now supplies the outcome-level figures the surrogate lacked — a weak answer, but an answer. Per 50 g/day of legumes: all-cause mortality RR 0.96 (0.90-1.01), NutriGrade moderate (Schwingshackl, Schwedhelm, et al., 2017); CHD 0.96 (0.92-1.01), moderate (Bechthold et al., 2017); type 2 diabetes 1.00 (0.92-1.09), low (Schwingshackl, Hoffmann, et al., 2017); incident hypertension 0.98 (0.95-1.01), very low (Schwingshackl, Schwedhelm, Hoffmann, Knüppel, et al., 2017). Every interval crosses or touches 1.00 — a small inverse-to-null linear signal, not a demonstrated benefit, on low-to-moderate self-reported cohort data with the usual measurement-error attenuation .
A purpose-built legume review now answers the question the food-group series could only glance at. Thorisdottir’s 2023 SR+MA for NNR2023 pooled 47 studies — 31 cohorts of about 2.1 million adults plus 16 trials — against both hard endpoints and blood lipids. The cohort arm is near-null: high-versus-low coronary heart disease RR 1.00 (0.95-1.05), stroke 0.98 (0.91-1.05), type 2 diabetes 0.90 (0.77-1.06), and «No clear dose-response association was found for any of the outcomes» (Thorisdottir et al., 2023) — graded «limited – no conclusion» under WCRF criteria.
The trial arm, though, moves a surrogate: pooled RCTs feeding at least 120-150 g/day of legumes lowered LDL-cholesterol by 0.19 mmol/L (-0.27 to -0.11) (Thorisdottir et al., 2023). So legumes shift the LDL marker at real intervention doses while the hard-endpoint cohort cells stay flat — a surrogate lever, not a demonstrated mortality or event benefit (Surrogate Outcomes).
None of these three lines is independent corroboration of the Aune-family plant-food evidence. The DIfE/Boeing series shares team, food-group definitions, and an overlapping cohort pool with it; Thorisdottir shares much of the same cohort base and the NNR review team, and reports high-versus-low contrasts rather than the series’ per-50 g slopes — so the agreement across all three is partly mechanical and not the same quantity -> Food Groups and Health Outcomes - A Dose-Response Matrix. The honest reading shifts from the corpus cannot answer to the corpus now holds a dedicated but near-null legume answer on hard endpoints, with a real LDL effect at RCT doses.
As a protein source, plant DIAAS is low — peas 64, wheat 40, against whole milk powder 122. (Food and Agriculture Organization of the United Nations, 2013) Hitting a protein target from pulses therefore takes more food mass, because each gram delivers fewer digestible indispensable amino acids. That is a mass cost, not a harm.
Nuts
Nuts are a plant food, and the same Aune lineage reports an inverse all-cause association (RR 0.78, 0.72-0.84, per 28 g/day). (Aune et al., 2016b) The increment differs from the other groups, so nuts are not separately rankable against them on these numbers.
That is the benefit side. The rest of the plant-food question is two worries — antinutrients and fruit sugar — and both shrink under the same isolate lens.
The antinutrient scare is mostly mis-scaled — with named exceptions
The alarming antinutrient studies share a design. They dose an isolated compound — a purified lectin, phytate powder — into an animal or a cell line, often at levels no cooked meal delivers. Petroski and Minich’s review of six compound classes makes the point directly: the effect of an isolated compound differs from the same compound inside a whole-food matrix, and most antinutrient research uses isolated compounds in animal models that do not represent a balanced diet (Petroski & Minich, 2020), (Petroski & Minich, 2020). Whole-food human trials mostly do not reproduce the harm.
This is a transportability failure. An effect measured in one setting — isolated, raw, animal — does not carry to another — matrixed, cooked, human. It is the same error the wiki flags whenever a food’s effect is read off a context it never occurs in -> Is the Food Category Doing Any Work, Measurement Error in Dietary Assessment.
The expectancy test seals the staple case. Legumes and whole grains feed billions of people every day. A large chronic harm from properly prepared plant staples would be visible by now. It is not.
Across the six classes, Petroski’s finding is uniform: ordinary preparation — soaking, sprouting, fermenting, boiling, cooking — cuts the compounds down (Petroski & Minich, 2020).
| Class | Main concern | Defused by | Stratum that still needs care |
|---|---|---|---|
| Lectins | gut damage; acute poisoning from raw/undercooked legumes | boiling/autoclaving (a full boil for the high-lectin legumes) | anyone eating raw or undercooked kidney beans |
| Phytate | binds zinc/iron/calcium, lowering mineral uptake | soaking, germinating, fermenting, cooking | monotonous high-phytate diets with marginal mineral status |
| Tannins | inhibit non-heme iron absorption in isolation | matrix effect; whole-diet studies show no iron-status link | low iron stores, especially menstruating women |
| Oxalates | kidney-stone formation | cooking plus adequate calcium | recurrent stone-formers and high urinary-oxalate excretors |
| Goitrogens | thyroid interference | cooking reduces progoitrin | suboptimal-iodine status |
| Phytoestrogens | endocrine disruption | source- and processing-dependent | infants (small size, immature gut) |
For most classes, in a varied diet, that makes the concern academic. For lectins in the highest-lectin legumes it is not optional. Kidney beans and soybeans need a full boil or autoclave; a lower cooking temperature does not destroy the lectin, and undercooked beans have caused documented mass poisonings (Petroski & Minich, 2020). That is an acute toxicity, not a chronic-diet claim.
So the honest message is not that antinutrients are harmless, but that properly prepared plant foods in a balanced diet carry no meaningful net antinutrient harm — with properly prepared doing real work for raw high-lectin beans.
A few named strata still warrant care. These are contraindications in a named group, not a general reason to avoid the food:
- low iron stores, especially menstruating women — take tannin-rich tea between meals, not with them;
- recurrent kidney-stone formers — watch total oxalate load;
- marginal-iodine status — cook goitrogenic brassicas rather than eating them raw in bulk;
- infants — limit concentrated phytoestrogen exposure (small body size, immature gut).
The sign flips by nutrient-status stratum: the same tannin that is inert for a replete adult blocks iron in someone already short of it -> Deficiency Repletion vs Enhancement.
One caution on all of this. The evidence is a single narrative review, graded low, so its reassurance earns the same scrutiny as the alarm it corrects — the verdict leans on the classes where whole-food human evidence exists, not on mechanistic optimism. And no patient-important outcome moves either way. The value is licensing someone to stop avoiding prepared beans and grains, not a promise that eating them treats anything.
One tie-in is worth naming. The digestibility-acting compounds here — phytate, tannins, trypsin inhibitors — are the same set FAO names as the plant-protein digestibility discount, and the same soaking-and-cooking step that lowers the toxicity worry also raises protein quality -> Protein Quality and the DIAAS Score.
The other worry is sugar — specifically, whether the sugar in fruit is a problem.
Whole fruit is not a sugar problem
The worry about sugar in fruit dissolves at the definition. WHO’s free-sugars limit excludes the intrinsic sugar inside intact fruit and includes the sugar in fruit juice (World Health Organization, 2015). Whole fruit sits outside the counted exposure; juice sits inside it. A blanket “cut sugar” instruction loses the one distinction that matters here. See Free Sugars Intake.
The reason is a dose-and-form story, not a sugar-content one. A large free-fructose bolus — from sugar-sweetened drinks, fruit juice, or high-fructose corn syrup — drives fat synthesis in the liver (hepatic de novo lipogenesis, the MASLD-relevant harm). Whole fruit delivers a modest fructose dose wrapped in a fibre matrix that slows absorption and blunts that hepatic flux. So a fruit’s sugar-to-fibre ratio does not decide its outcome — the matrix does. The lever is cutting free fructose in beverages, not avoiding whole fruit.
-> Fatty Liver MASLD and Weight Loss
Glycaemic index tells the same story from the other side. GI and GL are surrogates, not outcomes, and they ride together with fibre and whole-grain content in whole-food diets (Jenkins et al., 2024). Jenkins found that low-GI diets and high-fibre, high-whole-grain diets gave near-identical disease associations in the same cohorts — which is what proxies for one underlying pattern produce. So glycaemic response is not an independent lever, and a fruit’s GI does not carry its outcome (inferred from Jenkins et al., 2024). GI plausibly bites in a stratum, not the population: the already insulin-resistant, where postprandial excursions matter — a management finding, not a general-diet lever. See Surrogate Outcomes, Glycaemic Index and Glycaemic Load and Chronic Disease.
Fruit juice is the honest complication, and it splits by outcome. On the metabolic, hepatic, and dental channel, juice sits inside the harmful free-sugar exposure — the same free-fructose bolus, plus caries.
Yet on cardiovascular endpoints, Aune’s cohort evidence found juice inversely associated with stroke (high-vs-low RR 0.67 [0.60-0.76]; per-100 g 0.72 [0.63-0.83]) and CHD (high-vs-low 0.79 [0.63-0.98]) (Aune et al., 2017). Both readings hold once you match the outcome axis, so this is a distinction, not a contradiction. Do not read juice as uniformly fine — its free-sugar load still dominates the metabolic channel. And the robust processing signal in these data is whole-versus-processed, not fruit-versus-juice: tinned fruit carried a positive (harm) association with cardiovascular disease (Aune et al., 2017).
The evidence cannot say which specific fruit to pick. Whether a grape’s glycaemic load makes it worse than a raspberry is a named gap — the per-fruit subtype cells are too thin (two to six studies, wide intervals) to rank one fruit against another on outcomes (Aune et al., 2017). The evidence separates whole fruit from processed forms, not one fruit from the next.
So how much does any of this matter, set against the big levers?
Where plant-food choice sits on your list of levers
If you already eat a varied plate of whole plant foods, the big plant-food levers are already pulled. Whether you reach for an apple or a pear, one more serving or two, changes little and is uncertain either way. That is not a failure to find something more — it is the finding. The ceiling here licenses you to stop optimising, and stopping is itself a decision -> Layer 1 - Ranking Interventions for a Stratum.
Two levers still have room to move.
Fibre is the real one. Most adults eat well short of the target: UK mean intakes sit 10-11 g/day below the 30 g/day AOAC reference value for men, and 13 g/day below for women (Scientific Advisory Committee on Nutrition, 2015) — roughly a 40% shortfall. Closing it is a genuine behaviour change, not a tweak. And you can reach it many ways: any high-fibre food, or a fibre supplement, gets you there. Total fibre carries the benefit, not how many different foods you eat -> Dietary Fibre and Health. Eating a range of foods loosely helps cover micronutrient needs, but that is a whole-diet adequacy heuristic, not a requirement — and a single high-fibre food already reaches the fibre target on its own.
Cutting free sugars — sugary drinks above all — is the other. It is the one carbohydrate exposure with an evidenced harm, and it is a different lever from anything on the plant-food plate.
Do not read a burden ranking as a large personal effect. GBD ranks low whole-grain intake the #1 dietary risk factor for DALYs worldwide — 82 million DALYs, about 3 million deaths (Afshin et al., 2019). That rank is about prevalence, not per-person gain. A near-universal shortfall (the global mean intake is 23% of optimal) multiplied by a modest observational risk ratio puts whole grains at the top. Almost everyone falls short, so a small individual effect sums to a large population burden. The rank tells you where a population is losing years, not how many you personally stand to gain -> Whole Grains Refined Grains and Pulses.
Two other axes ride along, and this appraisal stops at naming them. Cost and environmental load both shift when you move toward whole plant foods. But the wiki holds no price or carbon data and never nets them against the health finding. The trade-off exists; weighing it is yours.
What to do
Pull two levers. Get fibre up toward ~30 g/day — by any route, a high-fibre food or a supplement — and cut sugary drinks. Where a food comes in whole and processed forms, prefer the whole one: tinned fruit tracks higher cardiovascular risk, and fruit juice reads inside the free-sugar exposure on the metabolic side -> Free Sugars Intake.
Drop two worries. Antinutrients in properly cooked food carry no meaningful net harm for a well-nourished person. Neither does the sugar in whole fruit — it arrives matrix-packaged, not as a free-sugar bolus.
One question stays open; one has closed part-way. The evidence still cannot rank individual fruits or vegetables against each other on outcomes — the per-food data are too thin to tell an apple from a pear. What pulses do beyond the LDL surrogate now has a dedicated answer: a purpose-built legume SR+MA (Thorisdottir 2023) finds near-null cohort hard-endpoints and a real LDL drop (-0.19 mmol/L) at 120-150 g/day, echoing the DIfE/Boeing series’ small inverse-to-null cells — a surrogate lever, not a demonstrated mortality or event benefit.
Evidence box
Question ’Among fruits, vegetables, pulses and grains, what does the evidence show about each food-group”s effect on each patient-important outcome — direction, magnitude, for whom, how certain — is any sub-group or specific food better- or worse-evidenced, or does the outcome evidence not resolve at the individual-food level? And: what do the plant “antinutrients” do to a patient-important outcome at realistic intakes (and how does preparation change it), and does a fruit”s sugar-to-fibre profile change its outcome?‘ Evidence included 19 sources — 15 gold, 2 high, 1 moderate Overall certainty Low (see Rating Certainty of Evidence) Source-selection note 1 source(s) below the gold evidence bar feed this page: Petroski (narrative review, moderate). Each labelled by tier; none load-bearing for the core claims. Last updated 2026-09-04 · Independently reviewed: No · Full edit history