Getting 'enough' vegetables

Both reviews on this page measure intake in grams per day, and both convert grams into servings using the same rule, stated in their own methods: 80 grams is one serving. That is where "a portion" comes from, and it is worth being clear about what it is and is not. It is a unit of measurement the researchers adopted so that studies asking different questions could be compared - not a biological threshold that anybody measured, and not a quantity the body treats as special. In practice - and this part comes from public-health portion guidance rather than from either review - 80 grams is about a medium apple, a large handful of green beans, two broccoli florets, a small bowl of salad, or three heaped tablespoons of most cooked vegetables. Multiply it out and the landmarks in this research become readable: the 200 grams a day that the first figure on this page is measured in is two and a half servings; the 400 grams behind the familiar "five a day" advice is five; the 800 grams at which the larger review found the lowest rate of death is ten. Measuring in grams means the headline figures count weight eaten rather than how it arrived on the plate, and neither review has anything to say about fresh versus frozen or expensive versus cheap - the word frozen does not appear in the larger review at all. Raw versus cooked was examined, though, and the answer is reassuring: analysed separately, cooked vegetables and raw vegetables were each associated with a lower rate of death, by similar amounts. The one form that stands out in the other direction is tinned fruit, which the same review associated with a higher rate of death on thin evidence - covered further down. Three practical notes follow from what was actually counted, and none of them involves buying anything. These studies counted the weight of fruit and vegetables eaten, so cooking is not the issue - cooked vegetables were associated with a lower rate of death when the larger review looked at them separately, and raw vegetables pointed the same way on much thinner evidence. Frozen was never examined either way, and tinned tomatoes and frozen peas are exactly the kind of thing this fix means. The one exception worth knowing is tinned fruit, which that review associated with a higher rate of death on thin evidence; tinned vegetables were not singled out. And the vegetables you already like beat the ones you think you should eat, because the ones you should eat go off in the drawer.

Aune D et al. 2017 · Wang DD et al. 2021

The single most useful thing in this literature is the shape of the curve, not the size of any one number. Both reviews modelled the relationship flexibly rather than assuming a straight line, and both found the same thing: it bends. The fall in mortality is steepest at the bottom of the intake range and flattens as intake rises. The larger review states it plainly - the reduction in risk was steepest up to 400 grams a day - and the second review found the curve essentially flat above about five servings. Read practically, that means the person with the most to gain from one more portion is the person currently eating almost none, and the person with the least to gain is the one already eating plenty. This is the opposite of how dietary advice is usually experienced. "Ten a day" sounds like a target that only the already-virtuous can hit, and it makes everybody below it feel that the effort is pointless until they get there. The curve says the reverse: the cheapest gains in this whole body of evidence sit at the very bottom, which is where most people are. It is also why this page sets no target. A target is a thing to fail at. The evidence supports the direction, and the direction is the useful part. The reason to anchor this to a meal rather than to a daily count is that the count is the part that fails. A number you are meant to hit every day is a number you notice missing, and the research supports no particular daily figure that both reviews on this page would agree on. What they do agree on is the shape of the curve, which is why the next portion is the honest unit rather than the fifth.

Aune D et al. 2017 · Wang DD et al. 2021

Honesty requires flagging that these two papers do not tell the same story about the ceiling. The 2017 review, pooling 95 cohort studies, found the lowest rate of death at 800 grams a day - ten servings, double the amount public-health advice recommends - and its authors said so in their title-worthy conclusion. The 2021 review, pooling 26 cohorts, found the curve flattening at about five servings, with higher intakes associated with minimal further reduction. Both teams are serious, both used similar methods, and both are looking at overlapping evidence. The most likely explanation for the gap is boring and important: hardly anybody in these studies ate ten servings a day, so the far end of the 2017 curve is built on very few people and the earlier review says as much, describing 800 grams as being at the high end of the range of intake across studies. The later team gives a different and more direct reason: it says its updated analysis included more studies than the earlier ones, which gave it greater power to trace the shape of the curve. Where a curve runs out of data it becomes an extrapolation dressed as a finding. What both reviews agree on is everything that matters for a reader: more is associated with lower mortality, the relationship is not a straight line, the returns are steepest at the low end, and there is no evidence of harm at higher intakes. Nobody who is arguing about whether the answer is five or ten is arguing about whether the answer is more than two.

Wang DD et al. 2021 · Aune D et al. 2017

Solid in its direction, soft in its size, and impossible to settle with the study designs available. Start with what is strong. This is a very large body of evidence - 95 cohort studies in one review, 26 in the other, covering millions of people followed for years - and the associations turn up across Europe, the United States, Asia and Australia, in populations that eat completely differently from one another. They survive dropping any single study. The most statistically well-behaved estimate in the whole review, coronary heart disease, has a heterogeneity statistic of zero, meaning fifteen studies in different countries essentially agreed. And the mechanism is not mysterious: the larger review notes that fibre and fruit-and-vegetable intake have been shown to reduce blood pressure, the later one describes potassium as linked to lower blood pressure, and raised blood pressure is a major contributor to stroke - which is where the largest association here sits. Now the softness. The headline mortality estimate carries a heterogeneity statistic of 83% and the strongest signal in the review that small studies with striking results are over-represented in the published literature. The reviewers investigated and report that dropping the smallest and the outlying studies softened that statistical warning sign without substantially altering the summary estimates in most of the analyses - their own hedge, worth keeping, because it is not a clean bill of health for every one of them. Their published wording is that there was indication of small-study bias in several of the analyses, and that they may therefore have slightly overestimated some of the associations. Read the figures on this page with that sentence attached: the direction is well supported, the exact size may be a little generous. More fundamentally, everybody in every one of these studies chose their own diet. People who eat a lot of fruit and vegetables smoke less, exercise more, drink less and are better off - and while these studies adjust for such things, the reviewers themselves write that residual confounding cannot be excluded. Pushing the other way, the handful of studies that measured how badly food questionnaires capture what people actually eat found that correcting for it made the association stronger. So the errors do not all run in one direction, which is unusual and mildly reassuring. The honest position: something real is very likely being measured, its size is uncertain, and no observational study can close the gap.

Aune D et al. 2017 · Wang DD et al. 2021

Taking the last question first: no. Analysed separately, fruit and vegetables were each associated with a lower rate of death, and by similar amounts, so neither half is carrying the finding on its own. Do not read the separate figures as a ranking, though - 200 grams of fruit is a much larger jump within what people actually eat than 200 grams of a combined total, and the analyses rest on different sets of studies. Juice is where the two reviews genuinely conflict, and rather than pick the answer we prefer, here is both. The 2017 review found fruit juice associated with lower mortality, but that finding rests on one to two studies, which is far too thin to lean on. The 2021 review, looking at more data, found intakes of fruit juices and potatoes not associated with mortality at all, and its authors concluded explicitly that their findings support increasing fruit and vegetables "but not fruit juices and potatoes". Given a two-study finding on one side and a larger analysis on the other, treating juice as a substitute is not supported. The same review found the association held for most subgroups of fruit and vegetables with the exception of starchy vegetables such as peas and corn. One oddity worth reporting because it is genuinely strange: the 2017 review found tinned fruit associated with a higher rate of death, from three to four studies - thin evidence, possibly nothing, possibly the syrup, and mentioned here only so nobody finds it later and thinks it was hidden.

Wang DD et al. 2021 · Aune D et al. 2017

Nothing on this page can tell you that, and it matters that we say so rather than imply otherwise. Every figure here comes from prospective cohort studies: large groups of people were asked what they ate, followed for years, and their death rates compared. Nobody was assigned to eat more vegetables, so what is being described is an association between what people already ate and what happened to them afterwards, not the result of a change anybody made. The problem this creates is easy to state and impossible to solve with this kind of study. People who eat a lot of fruit and vegetables differ from people who eat few in almost every way that also affects how long they live - smoking, exercise, alcohol, income, whether they see a doctor. These studies adjust statistically for many of those things, and the reviewers still write that residual confounding cannot be excluded. A trial that randomly assigned tens of thousands of people to different diets for decades and counted deaths would settle it; no such trial exists and it is unlikely one ever will. What can be said honestly is that a very large, geographically diverse body of evidence points consistently in one direction, that the mechanism is plausible, and that eating more fruit and vegetables carries no known downside.

Yes, and if any of this applies to you the person to ask is your own doctor or pharmacist rather than a web page. If you take warfarin, the vitamin K in green leafy vegetables interacts with it - the usual advice is to keep intake steady rather than to avoid them, but a sudden change in how much spinach, kale or broccoli you eat is worth mentioning to whoever monitors your dose. If you have kidney disease you may have been given a limit on potassium, which fruit and vegetables are a major source of. If you have inflammatory bowel disease, a narrowed bowel, or you have recently had abdominal surgery, you may have been advised to keep fibre low for a period. Some people with irritable bowel syndrome find certain fruits and vegetables trigger symptoms. None of these are reasons for most people to eat less; they are reasons for some people to get individual advice first. Nothing on this page is medical advice, nothing on it diagnoses anything, and nothing on it is a treatment for any condition. The research quoted here counted deaths in large groups of people over years and was not designed to say anything about any individual.