Chapter 37 — Nutrition Science (Honestly): What the Evidence Actually Says About What to Eat

What Danny said in the kitchen one night

It is a Tuesday in late autumn. Danny Reyes-Park is on the line at the fermentation-focused restaurant where he works on weekends, except tonight is not a weekend, tonight is a staff meal — the cooks eat together at 4 p.m. before the doors open at 5:30 — and somebody has brought up, again, the article about saturated fat that ran in The New York Times over the weekend, the one that reversed the previous decade's reversal of the previous decade's reversal.

The chef, who is older than Danny by twenty years and has been cooking professionally since before Danny was born, listens to the conversation for about ninety seconds. Then he picks up the tongs he uses to portion the staff meal — a brothy chickpea stew with carrots and a hot olive oil drizzle — and he says, without looking up from the pot:

"We cook for taste. Nutrition is the next thing. If it has anything to do with cooking at all."

The conversation stops. The cooks eat their stew. The article does not come up again that night.

Danny writes the line down in the back of his notebook because he writes everything down, but the line bothers him for weeks. It bothers him because it is wrong in some way he cannot quite name, but it is also right in some way he cannot quite name either, and the gap between those two things turns out to be the entire content of this chapter.

What is right about the line: most of cooking, most of the time, has nothing to do with nutrition. It has to do with feeding people something that tastes good enough to bring them back to the table. The chefs who have cooked the food the world remembers — Escoffier, Jiro Ono, Hassan M'Souli, Cecilia Chiang, Edna Lewis — were not running nutrition calculations. They were chasing flavor.

What is wrong about the line: every piece of food you eat is going to be metabolized into energy or building material or waste. Nutrition is not a separate subject from cooking; it is what cooking is for, biologically, after the eating is done. To say "nutrition has nothing to do with cooking" is to say "the destination has nothing to do with the journey."

What this chapter is going to try to do is hold both of those things together at once. We are going to be honest about what nutrition science knows and what it does not know. We are going to push back, gently and firmly, against the certainty that has dominated public-health messaging on food for the last seventy years — a certainty that has been wrong in major and consequential ways, and that has hurt people. And we are going to try to get back to a place where you can sit down at a table, eat what is in front of you, and not feel like you are taking a test.

The chapter does not tell you what to eat. The chapter tells you how to think about what to eat, and which sources to trust, and which ones to disregard.

That is the most we can promise. Let's begin.

Why nutrition science is hard

Before we say anything about nutrition, we need to talk about why the field is in the state it is in.

Nutrition science is, methodologically, one of the hardest sciences to do well. To understand why a chemistry experiment can be replicated in three different labs and produce the same result, while a nutrition study can run for ten years on tens of thousands of subjects and come out the other end with a result no one can replicate, we have to understand the structural problems baked into food research.

Problem one: you cannot lock people in a metabolic ward for twenty years. The gold standard in medical research is the randomized controlled trial — the RCT — in which subjects are randomly assigned to either get an intervention or a placebo, and then followed for long enough to see what happens. This is how we learned that smoking causes lung cancer (eventually), that aspirin reduces heart attacks (clearly), that statins lower cardiovascular events in some populations (after a long argument). RCTs work because randomization eliminates almost every form of confounding bias. If you randomly assign 10,000 people to take a drug or a sugar pill, the people who took the drug are, on average, identical to the people who took the placebo in every way except the drug. So when you see a difference in outcomes, you can attribute it to the drug.

You cannot do this with food. You can do it for a few weeks (and metabolic-ward studies that run two to four weeks are how we learn about short-term metabolic effects). But you cannot keep 10,000 people on a controlled diet for a decade. Real diets vary day to day, season to season, mood to mood. People go on vacation. People get divorced and start drinking more wine. People discover air fryers. The diet adherence in long-term nutrition trials is famously poor — by year three of a five-year trial, the "low-fat group" and the "control group" are eating diets that are nearly indistinguishable.

This means that for most of the questions you actually care about — does this way of eating reduce my chance of dying of a heart attack in twenty years — you cannot run the cleanest experiment. You have to use weaker tools, and you have to be honest about the weakness.

Problem two: food-frequency questionnaires are unreliable. When researchers cannot run a controlled-diet study, they often run an observational study, in which they ask a large group of people what they eat and then follow them for years to see what happens. The instrument that does the asking is usually a food-frequency questionnaire (FFQ) — a checklist where you mark how often, on average, you eat each of about a hundred foods.

People are bad at this task. Try it yourself: how many times have you eaten broccoli in the last three months? You do not know. You are guessing. People with disordered eating systematically under-report. People who are health-conscious systematically over-report vegetables. People with a recent doctor's appointment systematically under-report ice cream. Validation studies that compare FFQ data to actual dietary intake (measured by direct observation or by analyzing biomarkers in blood and urine) typically find correlations in the 0.3 to 0.6 range — useful for very rough population-level patterns, useless for individual prediction. Some researchers, including the late John Ioannidis at Stanford, have argued that the entire FFQ-based literature is so noisy as to be unreliable.

Problem three: the replication crisis hits nutrition harder than most fields. The wider problem in science — that a high fraction of published findings turn out to be unreplicable when other labs try to repeat them — is bad in nutrition. Several recent meta-analyses and replication efforts have found that single-study claims about specific nutrients (chocolate prevents heart disease, eggs cause heart disease, eggs do not cause heart disease, red wine prevents heart disease, red wine causes cancer, coffee causes cancer, coffee prevents cancer) frequently fail to hold up when retested. The original studies were often statistically significant; they were also often wrong. This is not because the researchers were dishonest. It is because the field's tools are noisy, the questions are hard, and the incentives reward novelty over replication.

Problem four: industry funding distorts research. Studies funded by the sugar industry, the dairy industry, the egg industry, the meat industry, and the seed-oil industry are systematically more likely to find results that favor the funder. This has been demonstrated multiple times in meta-analyses of the funding-source effect. It does not mean every industry-funded study is wrong; it means you should weight industry-funded studies less, and you should especially weight the absence of industry-funded studies that would have shown a problem (publication bias goes both ways).

🧪 The threshold concept of nutrition science: most claims you read in headlines are weaker than the headline implies. A single nutritional epidemiology study, published in a respected journal, with a statistically significant result, is evidence — but it is weak evidence. It is one data point in a noisy field. The level of certainty that a chemistry professor brings to "the boiling point of water at sea level is 100°C" is not the level of certainty that a nutrition researcher should bring to "this food is good for you." Once you understand this, the entire experience of reading nutrition news changes.

Problem five: the same study often contradicts itself. Researchers can pre-register their analyses, but in practice many nutrition papers run dozens of statistical tests on the same dataset and report the ones that came out significant — a practice now widely criticized as inflating false-positive rates. The famous example is a 2013 paper by John Ioannidis showing that fifty random ingredients from a cookbook had each been claimed, in published research, to either cause or prevent cancer.

This sets the tone. We are not going to tell you nutrition science is useless — it is not — but we are going to be cautious about almost every individual claim, and we are going to be much more confident about claims that have been shown repeatedly, in different populations, by independent groups, using different methods. That convergent evidence is the gold of the field, and it is rarer than the volume of nutrition journalism would suggest.

The Bradford Hill framework

🔬 Advanced sidebar: Bradford Hill criteria for causation. In 1965, the British epidemiologist Sir Austin Bradford Hill — the same man who had helped establish that smoking causes lung cancer — published a list of nine considerations for inferring causation from observational data. They are not a checklist; they are a set of weights. The more of them are present, the more confident you can be that an observed correlation reflects a real causal effect.

The criteria are:

1. Strength. Is the effect large? A relative risk of 2.0 (twice the rate in the exposed group) is much harder to explain by confounding than a relative risk of 1.1.
2. Consistency. Has the result been seen in different populations, by different researchers, in different settings?
3. Specificity. Does the exposure produce a specific outcome, rather than a general one?
4. Temporality. Did the exposure precede the outcome? (Trivially required, but often confused in cross-sectional data.)
5. Biological gradient. Is there a dose-response — more exposure, more outcome?
6. Plausibility. Is there a biological mechanism that could explain the effect?
7. Coherence. Does the result fit with what is otherwise known about the disease?
8. Experiment. Has an intervention reducing the exposure also reduced the outcome?
9. Analogy. Does it look like other known causal relationships?

When you read a nutrition headline, run it through these. The smoking-and-lung-cancer link, by 1965, met all nine. Most nutrition claims meet two or three. That gap is the difference between a strong claim and a weak one.

What we know reasonably well

It is easy to leave a chapter like this feeling that nutrition is all noise. It is not. There is a small core of claims that have been seen repeatedly, across populations and methods, with plausible mechanisms and (in some cases) interventional evidence. We will name them here, briefly. Each of the next sections will go deeper.

• Trans fats from partial hydrogenation increase cardiovascular risk. This is well-established and has led to bans in many countries.
• Diets very low in fruits and vegetables are associated with worse health outcomes. The reverse — eating more fruit and vegetables — has accumulated reasonably good evidence over decades.
• Diets high in ultra-processed foods are associated with overconsumption of calories and worse metabolic markers. The mechanism is debated; the association is not.
• A pattern of eating sometimes called "Mediterranean" — varied vegetables, legumes, olive oil, some fish, some dairy, low red meat — has shown cardiovascular benefits in multiple large trials including the PREDIMED randomized controlled trial. This is some of the strongest dietary-pattern evidence we have.
• Vitamin and mineral deficiencies, where they exist, cause measurable harm. Iron deficiency anemia, B12 deficiency, vitamin D deficiency in some populations, iodine deficiency: all real and treatable.
• Antioxidant supplements, despite enormous initial promise, have largely failed to produce health benefits in randomized trials, and some have caused harm.
• The "calories in, calories out" framework is true at the level of thermodynamics but is more complicated than its critics or its defenders typically allow.

The rest of this chapter is the long version. Let's start with the macronutrients.

Macronutrients: what we eat in grams

The three macronutrients are protein, carbohydrate, and fat. Almost everything you eat is some combination of these three, plus water, plus tiny amounts of micronutrients.

Protein

Proteins are chains of amino acids — twenty different amino acids, of which the human body cannot synthesize nine and so must obtain them from food. Those nine are called the essential amino acids: histidine, isoleucine, leucine, lysine, methionine, phenylalanine, threonine, tryptophan, valine.

The current recommended dietary allowance (RDA) for protein in healthy adults is about 0.8 g/kg body weight per day (about 0.36 g/lb), which works out to roughly 50–60 g for a typical adult. This was set conservatively to prevent deficiency, not to optimize health. There is now reasonable evidence that for athletes, older adults, and people in calorie deficits, somewhat higher intakes — in the range of 1.2 to 2.0 g/kg — may better support muscle maintenance and recovery. Most healthy adults in developed countries eat well above the RDA without trying.

🌍 Complete vs incomplete proteins, and why every culture solved this. Animal proteins (meat, fish, eggs, dairy) contain all nine essential amino acids in roughly the proportions humans need; they are sometimes called complete. Most individual plant proteins are lower in one or two essential amino acids — grains tend to be low in lysine, legumes tend to be low in methionine. The classical worry was that vegetarian diets would be deficient in essential amino acids.

The classical worry was wrong, and the reason it was wrong is one of the most beautiful patterns in human food history. Every culture, working independently, paired a grain with a legume.

The grain provides methionine and is short on lysine. The legume provides lysine and is short on methionine. Together they cover the full essential amino acid profile. No culture had a textbook explaining this. Every culture worked it out by living for centuries on what was available, paying attention to what kept their children growing.

(We do now know that you do not need to eat the grain and the legume in the same meal. Your body keeps amino acid pools in circulation, and you can balance protein intake across the day. The "must combine at one meal" rule was a 1970s misunderstanding that has been revised.)

This pattern is theme four of the book — food traditions are accumulated scientific knowledge — in its purest form. Long before there was a chemistry of amino acids, there was a chemistry of survival.

Carbohydrates

Carbohydrates are sugars (single units like glucose, double units like sucrose) and polysaccharides, which are long chains of sugar units. The major polysaccharides in human nutrition are starch (chains of glucose, digestible) and fiber (chains of various sugars, partly indigestible).

We cover the chemistry of starch and sugar in chapters 9 and 10. What concerns us here is how the body handles them.

When you eat a digestible carbohydrate, enzymes (mostly amylase, which we met in chapter 13) break it down into glucose, which enters your bloodstream. Your blood glucose rises; your pancreas releases insulin; insulin signals your tissues to take up glucose for energy or storage. The size and speed of the blood-glucose rise depends on what you ate. A spoonful of pure glucose syrup raises blood sugar fast and high. A bowl of steel-cut oats raises blood sugar slowly and modestly, because the oats contain fiber, the starch is in intact granules, and the matrix slows digestion.

The glycemic index (GI) tries to capture this — it is a number assigned to a food based on how high blood glucose rises after eating a standard portion, compared to a reference of pure glucose. Pure glucose is 100; white bread is around 75; rolled oats are around 55; lentils are around 30. The glycemic load (GL) adjusts the GI for portion size, which is closer to what actually matters in a meal.

GI is real, but it is oversimplified. It varies between people, between varieties of the same food, between cooking methods (a hot potato has a higher GI than a cold potato — chapter 9's retrogradation in action), and between meal contexts (fat and protein eaten alongside lower the GI of carbohydrates). For most people without diabetes, eating a varied diet that includes some whole grains, legumes, vegetables, and fruit is more useful than chasing GI numbers.

The strongest carbohydrate finding in nutrition is for fiber. Across many studies and study types, populations eating more fiber (especially from whole grains, legumes, vegetables, and fruit) have lower rates of cardiovascular disease, type 2 diabetes, and colorectal cancer. The mechanisms are multiple: fiber slows digestion, feeds gut bacteria that produce short-chain fatty acids (acetate, propionate, butyrate) the colon uses for energy, and changes the microbial composition of the gut. Most adults in industrialized countries eat about half the recommended fiber intake of 25–35 g/day. This is the closest thing to a confident dietary recommendation in this entire chapter.

The recent low-carb fashion needs to be put in historical perspective. From the 1960s through the 1990s, mainstream nutrition advised low-fat, high-carb diets — that was the era of fat-free SnackWell's cookies and pasta as a health food. From around 2000, the pendulum swung to low-carb, high-fat — that was Atkins, then keto, now carnivore. The evidence supports neither extreme as universally optimal. What it supports, repeatedly, is variety, moderation, more whole foods than processed, and individual variation. Some people respond well to lower-carb diets, especially people with insulin resistance. Some people thrive on traditional high-carb diets — Okinawans, for instance, ate diets that were 80% carbohydrate by calories and lived a long time.

Fats

This is the section where nutrition science has reversed itself most spectacularly, and where the reversal has consequences for everything you eat.

We covered the chemistry of fats in chapter 11. Briefly: fats are triglycerides — three fatty acid chains attached to a glycerol backbone. The fatty acid chains differ in length and in the number of double bonds (carbon-carbon double bonds in the chain). Saturated fats have no double bonds; monounsaturated fats have one; polyunsaturated fats have several. Trans fats are an artificial category produced when food chemists partially hydrogenated vegetable oils, twisting some of the double bonds from the natural cis configuration to a trans configuration to make the oil solid at room temperature.

The trans-fat story is settled. Consumption of industrial trans fats raises LDL cholesterol, lowers HDL, and increases cardiovascular risk in essentially every type of study that has looked. The FDA classified partially hydrogenated oils as no longer "generally recognized as safe" in 2015 and required removal from the U.S. food supply by 2018. Many other countries have similar bans. Pat — the chemistry teacher you met in chapter 11 — has been retiring her butter-vs-margarine demonstration each year and updating her notes. The trans-fat story is what convinced many people that nutrition science could in fact arrive at a confident answer; it could; this was one of those times.

The saturated-fat story is not settled.

For about fifty years — from the 1950s, when Ancel Keys's Seven Countries Study helped popularize the diet-heart hypothesis, through the early 2000s — mainstream nutrition advised reducing saturated fat as a primary intervention to prevent cardiovascular disease. The original argument was: saturated fat raises LDL cholesterol, LDL cholesterol is associated with atherosclerosis, therefore reducing saturated fat reduces heart disease.

The first half of the chain holds — saturated fat, on average, raises LDL. The middle holds — LDL is, on average, associated with atherosclerosis. The last step is where the evidence got messier. Over the 2010s, several large meta-analyses (including ones by Ronald Krauss, Russell de Souza, and others) found that the direct link between saturated fat intake and cardiovascular events was weaker than the LDL chain implied. Some studies found a small effect; some found no effect; some found that the effect depended on what people replaced the saturated fat with.

The replacement question turned out to matter enormously. If you remove saturated fat and replace it with unsaturated fat (olive oil, canola, nuts, seeds), cardiovascular risk goes down. If you remove saturated fat and replace it with refined carbohydrates (which is what large-scale food reformulation in the 1980s and 1990s actually did, swapping butter for trans-fat margarine and adding sugar to make low-fat products taste palatable), cardiovascular risk does not go down and may go up. The famous example is the Women's Health Initiative Dietary Modification Trial, a randomized trial of nearly 49,000 postmenopausal women that ran from 1993 to 2005 and tested a low-fat dietary intervention. The trial found no significant reduction in cardiovascular disease, breast cancer, or colorectal cancer in the intervention group — a finding that, when published in 2006, surprised the field and contributed to the broader rethinking of the low-fat hypothesis.

The current reasonable position is something like this: trans fats are bad and have been correctly removed; saturated fat is probably worse than unsaturated fat for cardiovascular risk on average, but the effect size is smaller than the 1980s assumed; the most consequential variable is what you replace fat with, not the absolute fat content of the diet. Olive oil, the centerpiece of the Mediterranean pattern, has held up best.

🔬 Advanced sidebar: the saturated-fat / heart-disease question, a 70-year history. Ancel Keys's 1950s correlation between saturated-fat intake and heart-disease mortality across countries was real but cherry-picked (he reported seven countries; data existed for many more, and the broader pattern was weaker). The diet-heart hypothesis became dietary policy in the United States by the late 1970s, contributing to the 1980 Dietary Guidelines for Americans. The 1980s and 1990s saw mass food reformulation — low-fat, high-sugar, trans-fat margarine. Cardiovascular mortality did decline over this period, but the credit is shared among many factors (statins, smoking cessation, hypertension treatment, surgical advances), and the dietary contribution is hard to isolate. The 2010s saw meta-analyses softening the saturated-fat verdict, and the 2020s have seen what is sometimes called the "saturated-fat litigation" play out in nutrition journals. The current consensus is that replacing saturated fat with mono- and polyunsaturated fat is beneficial, but demonizing saturated fat in absolute terms was an overreach. Coconut oil and butter are not poisons; they are also not health foods. Use both in moderation as part of a varied diet.

The omega-3 fatty acids (EPA, DHA, ALA) found in fatty fish, walnuts, flaxseed, and some other foods have a smaller but consistent body of evidence for cardiovascular benefit, especially from food rather than supplements. The supplement story is more mixed: large trials of fish oil capsules have found inconsistent effects, and the very high-dose prescription preparations may help specific patients, but the broad consumer-supplement effect on healthy populations is small to nil.

Micronutrients: vitamins and minerals

Vitamins and minerals are the small molecules and elements you need in milligrams or micrograms per day. Their effects are real and, in deficiency, dramatic.

The vitamins divide into water-soluble (the B-complex vitamins B1, B2, B3, B5, B6, B7, B9, B12, and vitamin C) and fat-soluble (A, D, E, K). Water-soluble vitamins are typically excreted when in excess; fat-soluble vitamins are stored in liver and adipose tissue, and excessive intake of A, D, and (to a lesser extent) E can cause toxicity.

Minerals you need most often: iron, calcium, magnesium, potassium, zinc, iodine, selenium, copper, and a handful of others. Sodium and chloride are also minerals, but you almost never need to think about them as a deficiency in a typical Western diet — see chapter 3 for the long discussion of sodium, which we will not repeat.

The honest summary on vitamins and minerals:

• Real deficiency causes real disease. Scurvy from vitamin C deficiency, beriberi from B1, pellagra from B3, rickets from D, anemia from iron or B12, goiter from iodine. These are still problems in some populations.
• For people eating a varied diet of whole foods, most vitamin and mineral intakes are adequate.
• Some populations have specific risks: vegans need a B12 supplement (cobalamin is essentially absent from plants); pregnant people benefit from folate supplementation; people in northern latitudes with limited sun exposure may need vitamin D, especially in winter; older adults often have lower B12 absorption; people with certain medical conditions or taking certain medications may have specific needs.
• Mega-dose supplementation has been the disappointing story of nutrition for the last forty years.

The disappointing story deserves its own section.

The antioxidant collapse

In the 1980s and 1990s, observational studies found that people who ate diets rich in fruits and vegetables — and therefore in antioxidant compounds like vitamin C, vitamin E, beta-carotene, and selenium — had lower rates of cardiovascular disease and cancer. This made biological sense: oxidative damage to DNA and lipids is implicated in both diseases, and antioxidants quench reactive oxygen species in test tubes.

The hypothesis was straightforward: if antioxidants in food are protective, then antioxidant supplements should also be protective. Pharmaceutical and nutraceutical companies took this seriously. So did public-health researchers. Several large randomized trials were launched.

The results were not what anyone expected.

The Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC), published in 1994, randomized about 29,000 Finnish male smokers to receive vitamin E, beta-carotene, both, or neither. The beta-carotene group had higher lung cancer rates than placebo. The vitamin E group had no benefit on cancer overall.

The Beta-Carotene and Retinol Efficacy Trial (CARET), published in 1996, was halted early when its 18,000 subjects (smokers and asbestos workers) given beta-carotene plus retinol showed increased lung cancer mortality compared to placebo.

The Selenium and Vitamin E Cancer Prevention Trial (SELECT), published in 2009 and updated through 2014, randomized about 35,000 men to selenium, vitamin E, both, or placebo to test prostate cancer prevention. Vitamin E supplementation was associated with a 17% increase in prostate cancer over time. The trial was halted early.

A 2007 meta-analysis by Bjelakovic et al., published in JAMA, examined 47 randomized trials of antioxidant supplements (beta-carotene, vitamin A, vitamin E, vitamin C, selenium) covering more than 180,000 subjects. The meta-analysis found that beta-carotene, vitamin A, and vitamin E supplementation significantly increased all-cause mortality. The increase was small but real.

This is one of the most consequential findings in modern nutrition, and it has not penetrated public awareness in proportion to its importance. The hypothesis that taking antioxidant supplements at doses higher than you would get from food is generally a bad idea. Whole foods, with their balance of antioxidants and pro-oxidants, vitamins and minerals, fiber and phytochemicals, behave differently from isolated mega-doses. The story of "superfood" antioxidant claims that flooded the 2000s — açaí, goji, mangosteen, pomegranate juice — was largely overhyped marketing built on the rubble of the supplement hypothesis.

⚠️ The exception, important and narrow: targeted vitamin and mineral supplementation for specific deficiencies, under specific medical supervision, is reasonable and effective. Folate before pregnancy, B12 in vegan diets, vitamin D in deficient individuals, iron for diagnosed iron-deficiency anemia. The collapse of the antioxidant hypothesis is about prophylactic mega-dose supplementation in people without deficiency. Read the difference.

Sodium, hydration, caffeine

We promised in chapter 3 that we would not relitigate sodium here. Briefly: sodium is essential, the average American eats more than is necessary, the population-level recommendation to reduce intake remains the position of major health bodies, and the evidence is somewhat weaker than that recommendation suggests. The strongest case for sodium reduction applies to people with hypertension. For normotensive people, the evidence is more equivocal. We refer you back to chapter 3 for the full discussion. 🔗

Hydration is a gentler topic. The famous "eight glasses of water a day" rule has no clear scientific origin and no good evidence behind it. The body's thirst mechanism is well-tuned: drink when thirsty. Most people in temperate climates with reasonable diets meet their water needs from the combination of beverages and the water content of food. The exceptions are predictable: hot environments, intense exercise, certain illnesses, older adults whose thirst signaling is sometimes blunted, and athletes who can lose multiple liters of sweat in a session. There is little evidence that "extra" hydration in healthy people produces any benefit, and very high water intake without electrolytes can cause hyponatremia (dangerously low blood sodium). Drink when thirsty. Stop counting glasses.

Caffeine, the subject of much of chapter 21, is one of the most-studied compounds in nutrition. The current evidence for healthy adults supports intakes up to about 400 mg of caffeine per day (roughly four 8-oz / 240-mL cups of brewed coffee) without identifiable harm; many studies suggest small benefits at this range, including a modest reduction in cardiovascular and some neurological disease risks. Higher intakes can cause anxiety, sleep disruption, and palpitations in sensitive people. Pregnancy guidelines recommend lower intakes (usually 200 mg/day or less). Caffeine sensitivity varies considerably between individuals because of genetic variation in caffeine-metabolizing enzymes. Self-knowledge wins.

Dietary patterns: what works, in the studies that exist

We have spent the chapter so far at the level of nutrients. The more recent move in nutrition science has been to study dietary patterns — the combinations of foods people actually eat — rather than individual nutrients in isolation. This has been a productive turn.

Mediterranean. The dietary pattern with the strongest evidence base is what gets called Mediterranean. The category is loose, but the core elements include vegetables and fruits eaten at most meals, legumes and whole grains as staples, olive oil as the primary fat, fish and seafood several times a week, modest amounts of dairy (often as cheese and yogurt), modest red wine with meals (in cultures where wine is part of the tradition), and red meat as an occasional rather than daily food.

The strongest single piece of evidence is the PREDIMED trial, a Spanish randomized study of about 7,500 people at high cardiovascular risk that compared a Mediterranean diet supplemented with extra-virgin olive oil or with mixed nuts to a low-fat control. The trial was stopped early in 2013 because the Mediterranean groups showed about a 30% reduction in major cardiovascular events. (PREDIMED was later reanalyzed after methodological concerns about randomization were raised, and the reanalysis in 2018 confirmed the cardiovascular benefit, with the effect somewhat attenuated.) Multiple other studies — including the Lyon Diet Heart Study and several large cohort studies — have supported the same general pattern.

The Mediterranean pattern is not a single thing. Italian, Greek, Spanish, Lebanese, Moroccan, and southern French versions all differ. What they share is not a specific food list but a structure: a lot of plants, modest portions of animal foods, mostly unprocessed, eaten slowly, often shared. Some of the benefit may not be in the macronutrient profile at all but in the pattern of eating.

DASH. The Dietary Approaches to Stop Hypertension pattern was designed as an intervention for blood pressure and tested in randomized trials in the late 1990s. It emphasizes fruits, vegetables, low-fat dairy, whole grains, lean protein, and low sodium. It works for blood pressure — that finding is well-replicated — and is broadly similar to the Mediterranean pattern with some differences in dairy and fat.

Blue Zones. The Blue Zones are five regions identified by journalist Dan Buettner and demographer Michel Poulain as having unusually high concentrations of long-lived people: Sardinia (Italy), Okinawa (Japan), Nicoya (Costa Rica), Ikaria (Greece), and Loma Linda (California, among Seventh-day Adventists). Their diets vary, but share patterns of mostly plant foods, legumes as a staple, modest portions of meat (often less than 5% of calories), and meal patterns embedded in close social and family ties. The popular framing has sometimes overpromised — there are real methodological concerns about birth-record reliability in some Blue Zones, and the diet-vs-genetics-vs-social-network contributions are hard to disentangle — but the pattern of mostly-plant, modestly-portioned, socially-shared eating is consistent with other evidence.

The Pollan summary. Michael Pollan's now-famous formulation, "Eat food. Not too much. Mostly plants," is a journalist's compression that maps surprisingly well to the evidence. Eat food is a gesture toward the ultra-processed-food finding (more on which below). Not too much is total energy balance. Mostly plants is the consistent finding from Mediterranean, DASH, Blue Zones, and large cohort studies. As one-line summaries go, it is a defensible one.

The fad cycle. Keto, Atkins, paleo, and carnivore diets are the current incarnations of low-carbohydrate eating. The honest summary: short-term weight loss is real on these diets, often greater than on low-fat diets at the same calorie intake. Some metabolic markers (HbA1c, triglycerides, HDL) often improve, especially in people with insulin resistance. Long-term sustainability is the main problem — adherence drops sharply at 6 months and beyond, and the diets that "work" are usually the diets people can stay on. Cardiovascular long-term outcomes for very-low-carb diets are not well-established, with concerns about the saturated-fat / animal-protein loading in some versions. The ketogenic diet has a separate, well-established medical use for treatment-resistant epilepsy in children.

Plant-based and vegan. Vegetarian and vegan dietary patterns have observational evidence for somewhat lower cardiovascular disease and certain cancer rates compared to omnivorous diets, though the comparison populations often differ in many other ways (vegetarians in cohort studies tend to smoke less, drink less, exercise more — the healthy user effect). Specific nutrient considerations matter: vegans need a reliable B12 source (supplement or fortified food), should pay attention to iron and zinc (lower bioavailability from plants), and may benefit from algae-based DHA if not eating fish. With those addressed, vegan diets are nutritionally complete.

Intermittent fasting. The various intermittent-fasting protocols (16:8 daily window, 5:2 pattern, alternate-day) have shown modest metabolic improvements in shorter trials. The current evidence suggests that for weight loss and metabolic markers, intermittent fasting works about as well as ordinary calorie restriction at equivalent calorie intake — which is to say, it works for some people because the time-restricted window helps them eat fewer calories. Sustainability is the main variable.

Calories, satiety, and the ultra-processed-food story

Energy balance — calories in, calories out — is true at the level of thermodynamics. The first law of thermodynamics applies to humans the same way it applies to combustion engines: energy is conserved. If you store fat, the energy in that fat came from your food.

But the framework is incomplete in two important ways.

First, "calories out" is not a constant. Total daily energy expenditure varies with body composition, activity, ambient temperature, hormones, and adaptive thermogenesis. When people lose weight, their resting metabolic rate often drops more than predicted by their lower body mass — the body defends its previous weight. This is one reason long-term weight loss is so hard.

Second, "calories in" is regulated. Hunger and satiety are regulated by hormones (leptin, ghrelin, GLP-1, peptide YY, cholecystokinin), gut signals, blood glucose, and central nervous system processing. Different foods have different effects on satiety per calorie. A bowl of beans and a candy bar at the same calorie count produce very different satiety responses; the beans typically suppress hunger more for longer.

This brings us to the story that has emerged most clearly in nutrition research over the last decade: ultra-processed foods.

The NOVA classification, developed by Brazilian researchers in the early 2010s, sorts foods into four groups by degree of processing. Group 4 — "ultra-processed foods" — is defined as industrial formulations made primarily from substances extracted from foods (sugars, oils, starches), modified ingredients (hydrogenated fats, modified starches), and additives (flavors, colors, emulsifiers, sweeteners), with little to no whole food remaining. Examples: many breakfast cereals, packaged snacks, sodas, instant noodles, mass-produced breads, reconstituted meat products, most "diet" and "low-fat" packaged foods.

Several lines of evidence have converged on the finding that ultra-processed foods drive overconsumption.

The most striking experiment was a small but rigorous metabolic-ward study by Kevin Hall's NIH lab, published in 2019. Twenty subjects were assigned to two-week periods of either an ultra-processed diet or an unprocessed diet, matched for calories, sugar, fat, fiber, and macronutrients on offer. Subjects were instructed to eat as much or as little as they wanted. On the ultra-processed diet, they spontaneously ate about 500 calories per day more than on the unprocessed diet, and gained weight; on the unprocessed diet, they ate less and lost weight. The diets were nutritionally matched. The processing was the variable.

The hypothesis is that ultra-processed foods produce overconsumption through some combination of: - Hyperpalatability: engineered combinations of fat, sugar, and salt that exceed what nature offers and exceed the brain's normal satiety calibration. - Speed of eating: soft textures, low chewing requirements, and rapid eating that outpaces satiety signaling. - Energy density and low fiber: more calories per gram, less stretch on stomach receptors. - Microbiome effects: emulsifiers and non-nutritive sweeteners may alter gut bacteria in ways that affect appetite and metabolism (this is suggestive but not yet settled).

The take-home is becoming hard to avoid: a diet built around ultra-processed foods tends to produce overconsumption almost regardless of macronutrient breakdown. This is, in many ways, a more important finding than any of the macronutrient debates. The actionable advice that emerges is also Pollan-shaped: cook more from ingredients you recognize. Eat less out of packages.

This has implications for the "low-carb" and "low-fat" wars. Both schools, when they are at their best, tend to push toward less ultra-processed food (because most ultra-processed food is high in both refined carbs and seed-oil-derived fats). When they are at their worst, both schools approve of ultra-processed versions of their preferred macronutrient profile (low-carb keto bars, low-fat SnackWell's), and this is where the ultra-processed problem reasserts itself.

What this chapter cannot tell you

We need to name several things this chapter does not pretend to address.

Eating disorders. Some readers come to nutrition information with histories of disordered eating — anorexia, bulimia, binge eating disorder, orthorexia. For those readers, anything that frames foods in terms of "good" and "bad," anything that promises optimization, anything that turns eating into a daily test, can be harmful. If you recognize yourself in this description, the most useful single thing this chapter can offer is permission to disengage from nutrition discourse for a while and work with a professional. The chapter is not a substitute for that work, and we are not going to pretend it is.

Weight stigma. The Health at Every Size perspective, developed by clinicians and researchers including Lindo Bacon and Lindo Bacon's collaborators (and now elaborated by many others), points out that weight-focused health interventions have a poor track record and that weight stigma — the social, medical, and economic discrimination experienced by people in larger bodies — has documented harms on its own. The science here is still being argued, but the data on weight stigma producing worse health outcomes is reasonably good. The advice that follows: focus on behaviors that are within your control (eating patterns, movement, sleep, stress, social connection) rather than on a number on a scale, and recognize that body diversity is real and not a moral failing.

Food access. Most American adults — and many billions of people globally — do not have unrestricted access to fresh produce, whole grains, fish, and the rest of the Mediterranean shopping list. Food insecurity, defined as inconsistent access to enough food for an active healthy life, affects roughly 10–15% of U.S. households in a given year. Food deserts and food swamps (neighborhoods where ultra-processed food dominates the available choices) shape what people eat as much as personal preference does. Telling someone in a food-insecure household to "eat more fish" is not nutrition advice; it is a tone-deaf prescription. The systemic question of food access is a real public-health problem and is not solved by individual-level dietary recommendations.

🌍 A note on Eurocentrism in nutrition advice. Mainstream nutrition science emerged from American and Northern European institutions, and its examples and food categories often reflect that origin. The Mediterranean pattern is well-studied because Mediterranean populations were studied. The same kind of detailed evidence does not exist for many traditional cuisines that are, by every reasonable mechanism, equally healthful — Japanese, Korean, Vietnamese, traditional Indian, Mexican, West African. When you read that "the Mediterranean diet" is the best-studied healthy pattern, the right addition is the best-studied healthy pattern that has been studied. There is no reason to think that a traditional Oaxacan diet of corn, beans, squash, tomatoes, chiles, and the occasional fish is nutritionally inferior. It has been less measured.

This matters when we talk about Maya. Maya's mother makes egusi soup with spinach and pumpkin seeds and palm oil and dried fish. Maya's aunt, when Maya visited Lagos as a teenager, served her a millet gruel with peanuts and dates that her aunt called fura da nono. Both meals nourished healthy people for generations before any nutrition researcher had looked at them. The framework Maya needs to evaluate her own family's food is not "is this a Mediterranean diet"; it is "is this a varied, mostly-whole-food diet that has sustained healthy people." By that standard, almost every traditional cuisine the world has produced qualifies. The framework that says otherwise is the one to question.

Pat in the classroom

Pat Hammond teaches a four-week unit on nutrition at the end of her general chemistry class, the year I observed it. The unit is called "Be a Skeptic, Not a Cynic." She is explicit, in her classroom, that the goal is not to leave the students distrusting all food information; the goal is to leave them able to read.

She teaches them to look at a nutrition headline and ask:

1. Is it a single study or a meta-analysis?
2. How big was the study?
3. Was it observational or randomized?
4. How long did it run?
5. Who funded it?
6. What did the researchers actually measure, and what claim is the headline making?
7. Does the claim match what other research on the same question has found?

Then she gives them three weeks of news clippings she has accumulated and asks them to score each one. By the end of the unit, the students are amazingly good at this, and they are also funnier — they have started a class joke about the Daily Mail headline rotation, which one of them has charted as alternating between "X causes cancer" and "X prevents cancer" for several common foods.

Pat says, on a Friday afternoon when the unit ends, that this is the most important thing she teaches. Not because nutrition is more important than chemistry (she does not believe that), but because nutrition is where the students will, every single week, encounter scientific claims for the rest of their lives. The skill she is building is not a fact set. It is a posture. A skeptical, generous, careful posture toward claims about food.

🍳 Kitchen Lab tease — How to read a nutrition headline. The full version is in exercises.md. The gist: pick three nutrition headlines from this week's news. For each one, find the original study (the link is usually buried somewhere in the article; if not, try Google Scholar). Check sample size, duration, study type, funding, and the actual effect size. Write a one-sentence honest version of the headline. Compare it to the version that ran. Notice how often the honest version is much weaker — and how often it would not have made it past the editor.

Pickles, probiotics, and what the evidence on fermented foods actually shows

We mentioned in chapter 30 — and will mention again in chapter 33 — that fermented foods are having a moment in popular health discourse. The chapter you are reading would be incomplete without a sober update on the probiotics question.

The honest summary: there is real and growing evidence that the gut microbiome matters for many aspects of health, and there is some evidence that habitually consuming fermented foods is associated with somewhat better metabolic and inflammatory markers. The strongest single piece of evidence is a 2021 randomized trial from Stanford by Sonnenburg and colleagues, which found that adults assigned to a high-fermented-food diet for 10 weeks showed increased microbiome diversity and reduced markers of inflammation compared to a high-fiber control.

The weak parts of the popular probiotic story:

• The specific bacteria in commercial yogurt or sauerkraut may or may not survive stomach acid in your gut. Some do; many do not. The strain matters.
• "Probiotic supplements" with billions of CFU on the label are largely unregulated for actual content and viability.
• Claims that any specific probiotic prevents or treats any specific disease — outside narrow indications like C. difficile recurrence — are mostly not well-supported.
• "Boosting your immune system" is not a meaningful clinical claim. The immune system is not a battery.

The reasonable position: a varied diet that includes some fermented foods (yogurt, kefir, sauerkraut, kimchi, miso, kombucha, tempeh) is almost certainly fine and probably modestly helpful. The mechanism is plausible. The effect size in a healthy person is likely small. The fermented vegetables track in this book is interesting and pleasurable on its own merits, and you do not need to overclaim its health benefits to enjoy it.

🥖 The bread track gets a similar update: whole-grain bread has reasonably good evidence for cardiovascular and metabolic benefits relative to refined-grain bread; sourdough has some evidence for lower glycemic response than commercial yeasted bread of the same flour, probably because of the lactic acid lowering pH and slowing starch digestion. Bread is not a medicine. It is also not a poison. Most people can eat bread that is made from real ingredients without anxiety.

Closing — Aroon's grandmother and the longer view

I asked Aroon Sornprasit, the Thai chef in Toronto, about the discussion in this chapter. He listened to me read the Pat-classroom section. He thought for a while, the way he does. Then he said, in his usual small dry voice:

"My grandmother fed three generations on what she could grow and what she could trade for. She did not read studies. She watched her children. She fed them what kept them strong. She changed what she cooked when she saw a child get tired or pale. She knew what was good for them. She knew this from looking."

Aroon's grandmother is gone now. She was, by every objective measure, a successful nutritionist for the people she fed.

The lesson is not that we should distrust modern nutrition science. The lesson is that nutrition science, at its best, is trying to articulate, in formal language, what attentive people figured out by paying attention. The pattern of "mostly plants, varied, in modest portions, with people you love" was not invented by epidemiologists. It was uncovered by every culture that survived long enough to have grandmothers.

The advice this chapter leaves you with, if it leaves you with any: cook food made of food. Eat it with people. Pay attention to how you feel. Ignore the headlines that contradict each other every week. Trust the long, slow signals — your sleep, your energy, your gut, your strength, your enjoyment — more than you trust any single study.

The next chapter is the one about where food is going. Cultured meat. Precision fermentation. The kitchen of 2050. A lot of what comes next will change. The principle that closes this chapter — eat food, mostly plants, with people you love — has been right for ten thousand years. It will probably still be right when the cultured-meat reactors are humming.

Turn the page.