A field guide to the diet wars

PATTERN

Every named diet claims to have found the variable that matters. The evidence keeps saying something less marketable: that the differences between well-executed dietary patterns are small, and the difference between any of them and how most people actually eat is enormous. This is what the trials show, where the guidelines came from, what is genuinely still argued about, and what to do about it.

Educational, not medical advice. Where the evidence is contested, it is labelled contested and both cases are made. Every number is sourced at the end.

Head to head — across outcomes, not just weight

The scoreboard

Diets are usually compared on one outcome — weight at twelve months — which is exactly the outcome on which they differ least. Pick any two and see them across seven.

Compare (pick up to three)

Scores are a 0–10 synthesis across trials, not a single study. Evidence grade is separate from performance: a diet can score well on a measure that has only ever been tested for twelve weeks.

The trial that should have ended the argument. DIETFITS randomised 609 adults to a healthy low-fat or a healthy low-carbohydrate diet for twelve months. Weight change was −5.3 kg versus −6.0 kg — no significant difference. Neither genotype pattern nor baseline insulin secretion predicted who did better on which, which was the study's real target and its most important negative result. Within each arm, individual outcomes ranged from roughly 30 kg lost to 10 kg gained. The variation inside each diet dwarfed the difference between them.

One dial — and what it actually buys you

Where you land on the carbohydrate axis

Almost every named diet is a position on a single axis, dressed up as a philosophy. Move the dial and watch the name change — along with the metabolic, digestive and adherence consequences that the name usually leaves out.

You're describing
Fiber, realistic g/day
12-mo adherence %
Evidence depth

The quality slider matters more than the carb slider, and this is the least intuitive thing on the page. In a metabolic ward, 20 adults ate diets matched for presented calories, macronutrients, sugar, sodium and fiber — the only difference was processing. On the ultra-processed arm they ate 508 kcal a day more and gained 0.9 kg in two weeks, then lost it on the unprocessed arm. They ate 17 kcal per minute faster. Macros held constant, and the food still decided the outcome.

Two centuries — the same argument, restaged

How we got here

The low-carb versus low-fat war is older than nutrition science. It has been fought at least four separate times, and each round has been treated by its participants as brand new. Tap any entry.

1992 — the most consequential graphic in nutrition

What the pyramid got wrong

The USDA Food Guide Pyramid was not stupid. It was a reasonable summary of 1980s consensus, rendered in a form that could not express the distinctions that turned out to matter. Its failures are worth understanding precisely, because the same failure modes are alive in nutrition advice today.

Four specific failures

failure 1

It treated all grains as one thing

Six to eleven servings of "bread, cereal, rice and pasta" at the base, with no distinction between intact whole grains and white flour. Those two foods have close to opposite relationships with cardiometabolic risk. A country that read the base as permission to eat bagels was reading it correctly.

failure 2

It treated all fats as one thing

"Fats, oils and sweets — use sparingly," in a single tip category. Olive oil, nuts, oily fish and partially hydrogenated shortening were given the same instruction. The trial that would later show a 30% reduction in cardiovascular events used an extra 50 mL of olive oil a day as its intervention.

failure 3

"Low fat" became a marketing licence

The reformulation wave that followed replaced fat with refined starch and sugar in thousands of products, all of which could now carry a health claim. The low-fat era is the period in which obesity prevalence roughly doubled — not proof of causation, but not a good look either.

failure 4

Servings were the wrong unit

Nobody has ever counted servings. The unit was invented for institutional catering and imposed on a population that eats from packages and restaurants. MyPlate in 2011 finally switched to proportions on a plate, which is the first thing the guidance ever did that matched how people actually see food.

The sugar industry documents — what is established and what is not

In 2016 Cristin Kearns and colleagues published an analysis of internal Sugar Research Foundation documents in JAMA Internal Medicine. The established facts: in 1965 the SRF paid two Harvard researchers, D. Mark Hegsted and Robert McGandy, $6,500 — roughly $48,000 in 2016 dollars — for a literature review on sugar and heart disease. The SRF set the review's objective, supplied articles for inclusion, and received drafts. The resulting 1967 review in the New England Journal of Medicine singled out fat and cholesterol and downplayed sucrose. The funding was never disclosed; NEJM did not require conflict-of-interest disclosure until 1984. Hegsted went on to help draft the 1977 Senate report that became the first Dietary Guidelines and to head nutrition at the USDA.

What the authors explicitly did not claim: there is no direct evidence that the industry wrote or altered the manuscript, and the evidence of influence on its conclusions is circumstantial. The honest reading is that this is a documented and serious conflict of interest whose downstream effect on policy is plausible but not demonstrated — which is a different and less satisfying claim than the one usually made about it online.

The Ancel Keys "cherry-picking" story, handled carefully

The popular version: Keys had data on 22 countries, selected the 7 that fit his hypothesis, and launched the low-fat era on a fraud. The actual history is messier. The 22-country dataset came from Yerushalmy and Hilleboe's 1957 critique of Keys, not from Keys' own work. The Seven Countries Study was a prospective cohort designed from 1958 onward, and countries were chosen for feasibility, contrasting diets and local research partners — a defensible design choice for a study that required decades of field cooperation.

That does not make Keys right about everything. The diet–heart hypothesis as originally stated was too simple, the ecological correlations were weaker than presented, and the leap from population-level associations to universal individual advice was not warranted by the evidence he had. Both things are true: the fraud story is a myth, and the science was oversold. Nutrition history is full of cases where the debunking is as sloppy as the original claim.

A replacement — sorted by confidence, not by food group

A better food guide

Every food guide ever published sorts by category — grains, proteins, dairy — which is a supply-chain taxonomy, not a health one. This one sorts by the two things that should actually govern your attention: how sure are we, and how much does it matter. Tap any food.

Read the top-right corner and stop there if you want. Everything in it is high-confidence and high-impact, and there are only about six items. The lower-left quadrant is where nearly all nutrition content lives — supplements, timing, superfoods, seed oil discourse — and where almost none of the available benefit does. The attention economy and the evidence base are inversely correlated, reliably enough that you can use it as a filter.

If you want it as a list

Do thisHow muchConfidence
Eat legumes most daysHalf a cup cooked. Cheapest fiber, protein and micronutrients available anywhere.high
Make olive oil your default fatTwo to four tablespoons daily, extra virgin, used liberally rather than sparingly.high
A handful of nuts30 g. One of the two supplemented arms in the only large dietary-pattern trial with hard endpoints.high
Get fiber to 30–40 gFrom food. The largest single nutrient gap in the Western diet.high
Cut ultra-processed foodThe one intervention that changed intake with macros held constant.high
Oily fish twice a weekSardines, mackerel, salmon, anchovies.moderate
Vegetables at every meal, fruit wholeWhole fruit, not juice. The matrix is most of the point.high
Intact whole grains over flourOats, barley, farro, whole rye. "Whole grain flour" is still flour.moderate
Minimise processed meatBacon, salami, hot dogs. Distinct from unprocessed red meat.high
Minimise sugar-sweetened drinksThe single most replaceable item in most people's diet.high
Alcohol: less is betterThe old J-curve does not survive Mendelian randomisation. No amount is protective.moderate

The answer, scoped precisely

If you want one winner

Here is the strongest claim anyone can honestly make, and it is narrower than the headline version: the Mediterranean pattern is the only dietary pattern with a large randomised trial showing a reduction in hard cardiovascular endpoints. Not surrogate markers. Not weight. Heart attacks, strokes and cardiovascular deaths.

PREDIMED, in numbers

Design
7,447 Spanish adults aged 55–80 at high cardiovascular risk but free of disease at entry. Randomised to Mediterranean diet plus extra-virgin olive oil, Mediterranean diet plus mixed nuts, or a control arm advised to reduce fat. Median follow-up 4.8 years.
Result
Adjusted hazard ratio 0.69 (95% CI 0.53–0.92) for the combined Mediterranean groups versus control. Roughly a 30% relative reduction in the composite of myocardial infarction, stroke and cardiovascular death.
In absolute terms
96 events in the olive oil arm (3.8%), 83 in the nut arm (3.4%), 109 in control (4.4%). That is about one percentage point over five years — which is a real and worthwhile effect, and a much smaller number than "30%" makes it sound.
The asterisk
The 2013 paper was retracted in 2018 after baseline distributions flagged irregularities in randomisation — some sites had randomised by household or clinic rather than individually. It was reanalysed without relying on the randomisation assumption and republished. The effect estimates barely moved. This is what good self-correction looks like, and it is also a genuine limitation.
What it does not show
That this generalises to a 30-year-old, or to a population that does not already eat this way, or that any single component is responsible. The control arm received advice to reduce fat, not an active competing diet — so this is Mediterranean versus usual care, not Mediterranean versus keto.

Why it probably works

Notice that it is not a restriction. It is an addition pattern: more olive oil, more nuts, more legumes, more fish, more vegetables. The things it displaces get displaced by being crowded out rather than forbidden, which is very likely why adherence in trials is the best of any pattern tested. Every mechanism you would want is present at once — high fiber, high monounsaturated fat, polyphenols, low glycemic load, low processing, high satiety per calorie — and no trial has ever succeeded in isolating which one matters.

Where other patterns win

  • DASH beats it for blood pressure specifically, and has the better trial evidence for that one outcome.
  • Ketogenic beats it for short-term glycemic control and type 2 diabetes remission at 6–12 months. No pattern comes close on HbA1c in the first year.
  • Whole-food plant-based matches or beats it on LDL, and has the strongest environmental case.
  • Whatever you will actually eat beats all of them at 24 months, which is the finding nobody wants to publish.
And the honest caveat about the caveat. "Mediterranean diet" is a construct — it describes a mid-century Cretan pattern that Cretans no longer eat, and it is operationalised differently in almost every study. What survives across all those operationalisations is a short list: olive oil, legumes, nuts, vegetables, fish, whole grains, little processed food, little sugar. When people say the Mediterranean diet wins, that list is what is winning. The name is packaging.

Seven days — with the fiber counted, because nobody ever counts it

What that actually looks like

Roughly 2,000 kcal, built to hit 35–45 g of fiber without trying, and designed so that cooking once feeds you twice. Tap a day.

The five habits underneath the plan

  • A pot of legumes on Sunday. One batch of chickpeas or lentils covers four of the seven days and is the single biggest lever on the fiber number.
  • Olive oil used as a food, not a lubricant. Two to four tablespoons a day, poured over things after cooking as well as during.
  • A jar of nuts within reach of where you actually get hungry. 30 g is a small handful.
  • Vegetables bought pre-prepped if that is what makes them get eaten. Frozen and tinned count. The perfect version that rots in the drawer counts for nothing.
  • Fish twice a week, tinned is fine. Sardines and anchovies are cheaper, more sustainable and higher in omega-3 per gram than most fillets.

A tangent worth taking — the crank who was accidentally right

The Great Masticator

Around 1900, an American businessman named Horace Fletcher instructed the world to chew every mouthful until it liquefied — roughly a hundred chews for a shallot — and to spit out whatever refused to dissolve. He called it a cure for most of human illness. He was wrong about nearly everything, and he had hold of something that took another 119 years to measure properly.

What he got wrong

  • There is no absorption efficiency unlocked by chewing. You do not halve your nutritional requirements.
  • The famous odourless-stool claim was an artefact of eating almost no fermentable residue, not a sign of purity.
  • The whole apparatus — a hundred chews, spitting out fibrous remainder, protein as poison — was a rule system built on a fictional mechanism.

Adherents included Rockefeller, Henry James and Kafka. Upton Sinclair described dinner parties conducted in total silence.

What he stumbled into

Strip out the mechanism and look at the behaviour: he was deliberately lowering kilocalories per minute. And that turns out to be one of the better-evidenced variables in the whole field — better than most of what the low-carb and low-fat camps spent a century arguing about. In the NIH metabolic ward trial, ultra-processed food was eaten about 17 kcal per minute faster than unprocessed, with macronutrients, sugar, sodium and fiber all held constant. Across hundreds of foods the gradient holds: roughly 35 kcal/min unprocessed, 54 processed, 69 ultra-processed.

Why speed matters: the brake has a delay

Satiety signalling is not instantaneous and never has been. Cholecystokinin, GLP-1 and PYY are released as nutrients reach the small intestine, and they take on the order of fifteen to twenty minutes to reach concentrations that meaningfully suppress appetite; even gastric distension reaches the brainstem with a lag. So there is a window at the start of every meal during which no feedback exists yet. The only question that matters is how much energy you can deliver into it.

Blind window kcal before feedback
Brake engages min in
Eaten by then kcal
vs unprocessed kcal

What this model is and isn't

Illustrative, not predictive. It assumes a 12-minute latency before ingested energy begins registering as satiety, a saturating signal, and a meal that ends when that signal crosses a notional threshold. Real meals end for a dozen reasons that have nothing to do with physiology — the plate is empty, the meeting started, everyone else stood up. The point of the model is the shape, and one identity inside it that is worth noticing: the extra energy consumed at a faster rate is simply the latency multiplied by the difference in rate. Twelve minutes times an extra 34 kcal/min is roughly 400 kcal, arriving before anything can object. That is the same order as the effect Hall measured, from a mechanism nobody was arguing about.

The second problem, which gets missed. Ultra-processed food also produces less gastric distension per calorie — low fiber, low water, low volume, high density. So the signal is not only arriving late, it is arriving fainter per calorie. Someone told to "listen to their fullness cues" while eating this food is being asked to detect a weak signal inside a short window. When that fails it gets read as an awareness failure or a willpower failure. It is neither. The task was rigged before they sat down.

The useful half of Fletcher

Structural beats effortful

Fletcher's protocol failed for the reason nearly all eating advice fails: it charged a fee at every single meal, forever. Chewing-count interventions work beautifully in a laboratory and evaporate on contact with a Tuesday. The durable version of the same insight costs you something once — at the shop, in the kitchen, in what is within arm's reach — and then keeps paying without further attention.

This is the general principle and it is worth more than any specific finding on this page. An intervention paid once, at setup, compounds. An intervention paid at every instance depletes. When someone relapses on a plan, the usual explanation is motivation; the more common actual explanation is that they chose a design that required a decision three times a day and the decisions ran out.

Same goal, two designs

Notice what the structural column has in common. Almost every entry works by changing what is available or default rather than what is permitted. Nothing is forbidden in any of them. That matters practically — restriction-based designs generate rebound and preoccupation — and it matters clinically, because a plan built on prohibition is a plan that can be failed. A plan built on availability can only be more or less set up.

The other lever — and where it must be handled carefully

Attention, not rules

Eating rate is a property of food. There is a second lever that is a property of the eater — interoceptive awareness — and the two are multiplicative rather than redundant. But this is also the point at which well-meant advice most often turns into something harmful, so the distinction below is not decorative.

Read this before the rest of the section

Any instruction to eat slowly, count chews, or monitor fullness can be recruited into restriction. In someone with a history of disordered eating, a chewing rule is a compensatory ritual with a scientific costume on, and a fullness scale is a permission system. Interoceptive accuracy is also genuinely degraded by both restriction and binge cycles — so "just listen to your body" asks for a signal that may not currently be legible. None of what follows is a portion-control technique, and it should not be used as one. If eating is distressing rather than merely suboptimal, this is territory for a clinician, not a webpage.

Fletcher's version

Rule-based and externally counted. A hundred chews. Spit out what resists. Success and failure are defined by compliance with a number, which means the practice can be performed perfectly while you learn nothing at all about your own experience — and can be escalated indefinitely, which is exactly the property you do not want in an eating rule.

The clinical version

Awareness-based and internally referenced. Mindfulness-based eating approaches ask you to notice several distinct signals rather than obey a target: physical hunger, the pleasure of the taste itself, stomach fullness, and general body satiety. These are separable, they change at different rates during a meal, and most people have never distinguished them. There is no number to hit and nothing to comply with.

Four signals, checked with curiosity rather than scored

What the trials actually show, and why the "disappointment" is the finding. Mindfulness-based eating interventions produce large and durable reductions in binge frequency and emotional eating — in the original work most participants no longer met binge eating disorder criteria — with inconsistent effects on weight. That result is usually reported as a partial failure. Read it the other way: it is the mechanism telling you what it is. This is an eating-regulation intervention with a plausible energy-intake side effect, not an energy-intake intervention. A tool that reliably improves someone's relationship with eating is worth having on its own terms, and mis-selling it as a weight technique is how it gets discarded when the scale doesn't move.

Putting the two levers together

Practise awareness with ultra-processed food and the check-ins have almost nothing to detect — weak distension, fast delivery, a hedonic curve that outruns the sampling. Practise it with legumes, intact grains, whole fruit and nuts and the substrate does half the work: the signal is stronger, it arrives earlier, and the attention has something real to land on. Change the food so the body can be heard; train the attention so the body gets listened to. Neither half works nearly as well alone, which is probably why a century of arguing about macronutrients produced so much less than either.

The hard case — low carb without wrecking the colon

Keto, done right

Keto has a structural problem that its advocates rarely name: nearly all fiber is carbohydrate, so restricting carbohydrate restricts fiber by default, and typical keto lands at 10–15 g a day. The way out is that fiber is carbohydrate you don't absorb. It costs nothing glycemically. The real constraint is digestible carbohydrate — which means the design problem is maximising fiber per net carb while keeping the fermentable share low enough that you don't buy the gas.

Fiber per net carb — the keto toolkit

Ratio of fiber grams to net carbohydrate grams. Anything above about 3:1 is essentially free fiber on a low-carb budget. Note where the vegetables sit — that is why keto diets quietly shed them.

Four scenarios, with the digestive consequences modelled

The bile acid point, which explains most of what people call keto diarrhea. High fat means high bile acid secretion. Bile acids that reach the colon are a secretory stimulus — they make the epithelium pump chloride and water into the lumen. Take psyllium with the fattiest meals rather than on an empty stomach: the gel binds bile acids, and one intervention then fixes both the early diarrhea and the later constipation.

The butyrate problem, and a partial rescue

Colonocytes take 60–70% of their energy from butyrate, made by fermenting carbohydrate you didn't absorb. Strict keto starves that supply. Two partial answers: raw potato starch stirred cold into dressing or a shake (heating destroys it) restores resistant starch at almost no net carb cost; and colonocytes can oxidise circulating beta-hydroxybutyrate, which is structurally close to butyrate, so deep ketosis may substitute in part. That second mechanism is plausible and not firmly established — treat it as a reason for optimism, not a plan.

Three practical notes

  • Magnesium citrate does double duty. Keto drives natriuresis and magnesium loss, and magnesium is an osmotic laxative. One supplement, two problems.
  • Sugar alcohols are not interchangeable. Erythritol is ~90% absorbed and renally excreted, so it is the gentlest — though a 2023 platelet-reactivity signal is worth knowing about. Allulose is well tolerated below roughly 0.5 g/kg. Maltitol will ruin an afternoon.
  • Don't get your fiber from keto breads and bars. That is inulin: rapidly fermented, no stool bulk, maximum gas per gram. You pay the whole cost and receive none of the laxation.

Rabbit holes — sorted by how much harm they did

The graveyard

Worth touring not for the laugh but for the pattern. Almost every one of these follows the same structure: a single mechanism, stated with total confidence, by someone with something to sell, explaining a problem that is genuinely miserable and genuinely hard to fix.

The tell that generalises. Every entry above proposes a single mechanism — one enzyme, one pH, one blood group, one macronutrient, one chewing count — and derives a total dietary system from it. Real physiology almost never works that way, and the confidence with which the mechanism is asserted is usually inversely proportional to the evidence for it. If a diet can be explained completely in one sentence and admits no exceptions, that is the signal.

Live disagreements — where confident people are overstating

What is actually still argued about

These are not settled, and anyone telling you otherwise is telling you about themselves. For each, the best case on both sides, and what would actually resolve it.

Where the numbers came from

Sources

ClaimSource
PREDIMED: 7,447 participants, median 4.8 y, adjusted HR 0.69 (0.53–0.92) combined; 96/83/109 events; retraction and republicationEstruch R, Ros E, Salas-Salvadó J, et al. N Engl J Med 2018;378:e34 (corrected and republished from N Engl J Med 2013;368:1279–90).
DIETFITS: 609 adults, 12 months, −5.3 kg low-fat vs −6.0 kg low-carb, ns; no genotype or insulin-secretion interactionGardner CD, Trepanowski JF, Del Gobbo LC, et al. JAMA 2018;319:667–79.
Ultra-processed diet: +508 ± 106 kcal/day, +0.9 kg in 2 weeks, macros and fiber matched; eating rate 17 kcal/min fasterHall KD, Ayuketah A, Brychta R, et al. Cell Metab 2019;30:67–77. Replicated with a larger effect (+813 kcal/day) by Hamano et al., Diabetes Obes Metab 2024.
Critique of the UPF trial: arms differed in added sugar, saturated fat, sodium, protein and soluble fiber despite matching intentCorrespondence in Cell Metab 2019 (Ludwig et al.).
Two-year comparison: low-carb −4.7 kg, Mediterranean −4.4 kg, low-fat −2.9 kg; best glycemic control on MediterraneanShai I, Schwarzfuchs D, Henkin Y, et al. N Engl J Med 2008;359:229–41 (DIRECT).
Carbohydrate and mortality: U-shaped, nadir around 50–55% of energy; animal-based low-carb worse, plant-based betterSeidelmann SB, Claggett B, Cheng S, et al. Lancet Public Health 2018;3:e419–28.
Sugar Research Foundation paid Hegsted and McGandy $6,500 (≈$48,000 in 2016 dollars) for the 1967 NEJM review; SRF set the objective, supplied articles and received drafts; no direct evidence it wrote or altered the manuscriptKearns CE, Schmidt LA, Glantz SA. JAMA Intern Med 2016;176:1680–5.
Eating rate gradient: ~17 kcal/min faster on ultra-processed food; ~35 / 54 / 69 kcal/min across unprocessed, processed and ultra-processed foodsHall KD et al., Cell Metab 2019;30:67–77; Forde CG, Mars M, de Graaf K. Ultra-processing and eating rate. Curr Dev Nutr 2020;4:nzaa019.
Satiety peptide latency: CCK, GLP-1 and PYY require roughly 15–20 min to reach appetite-suppressing concentrationsStandard gut–brain axis physiology; reviewed in Cummings DE, Overduin J. J Clin Invest 2007;117:13–23.
Sensory-specific satiety: hedonic value of a given food declines steeply across a portionRolls BJ, Rolls ET, Rowe EA, Sweeney K. Physiol Behav 1981;27:137–42.
Mindfulness-based eating awareness: large reductions in binge frequency, most participants no longer meeting BED criteria; inconsistent weight effectsKristeller JL, Wolever RQ. Eat Disord 2011;19:49–61; Kristeller J, Wolever RQ, Sheets V. Mindfulness 2014;5:282–97; meta-analyses by Katterman SN et al. Eat Behav 2014 and Carrière K et al. Obes Rev 2018.
Fletcherism: chewing protocol, adherents, and the absence of any absorption mechanismFletcher H, Glutton or Epicure, 1899; Levenstein H, Revolution at the Table, 1988.
Fiber dose–response: 25–29 g/day band, 15–30% lower all-cause and CV mortality highest vs lowestReynolds A, Mann J, Cummings J, et al. Lancet 2019;393:434–45.
Alcohol's apparent J-curve does not survive genetic instrumental-variable analysisMendelian randomisation studies of alcohol and cardiovascular outcomes, incl. Holmes MV et al. BMJ 2014;349:g4164; and subsequent replications.
Seven Countries Study design; the 22-country dataset originated in a critique of Keys, not in his own selectionKeys A, Seven Countries Study, 1958 onward; Yerushalmy J, Hilleboe HE. N Y State J Med 1957;57:2343–54.
Ketogenic diet for epilepsy originated at Mayo Clinic in 1921; revived in 1994 via the Charlie FoundationWilder RM. Mayo Clin Bull 1921;2:307. Freeman JM et al., Johns Hopkins ketogenic diet program publications.
Blood type diet: adherence to any of the "blood-type" patterns produced metabolic benefit independent of actual blood typeWang J, García-Bailo B, Nielsen DE, El-Sohemy A. PLoS One 2014;9:e84749.
Blue Zone longevity data quality concerns (pension fraud, missing birth records) in regions with the oldest reported populationsNewman SJ, preprint series on supercentenarian data reliability, 2019–2024.
USDA Food Guide Pyramid 1992; MyPyramid 2005; MyPlate 2011; Harvard Healthy Eating Pyramid 2001 and Healthy Eating Plate 2011USDA Center for Nutrition Policy and Promotion; Willett WC, Eat, Drink, and Be Healthy, 2001.

PATTERN · an educational reference. Interactive scores are syntheses across trials and are labelled as estimates where they are estimates. Nutrition evidence is largely observational, effect sizes are modest, and confident single-cause claims — in any direction — should be read as marketing until shown otherwise. Nothing here is a substitute for advice from a clinician or dietitian who knows your history. Companion piece to TRANSIT, on the applied science of dietary fiber.