Dietary Patterns

A diet fails when it leaves out essential nutrients; above that floor, the label changes little that has been measured

Two screens answer the question, and neither ranks diets. The first rules a diet out. The second asks whether a diet’s name matters once it clears the first.

A diet is worse by construction when it leaves out essential nutrients or badly mismatches energy. Any pattern can fail. Strict plant-based diets run short of B12, iodine, iron, zinc and long-chain omega-3 unless planned; typical meat-eating diets run short of fibre, folate and magnesium. Repletion helps the deficient, not the replete.

Above that floor, the diet’s name moves weight and blood pressure only a little. LDL is the exception: there, low-carb does worse than low-fat. Early gains over a usual diet mostly fade within a year, and the US dietary-guidelines review (NESR) could not grade diets defined by fat or carb share against mortality.

On weight and blood pressure over the first months, leaving a typical diet for almost any studied pattern matters more than choosing among them. Three things travel with that change: shared foods (more plants, nuts, unsaturated oils, fish; less processed meat, ultra-processed food), total energy, and sticking with the diet. No study has measured their shares, and they overlap rather than add.

One named pattern has a trial showing fewer events. In high-risk adults, a Mediterranean diet with supplied olive oil or nuts probably cut strokes; deaths did not detectably change. Low-fat and MIND trials found no significant effect on their main outcome, which rules out a large effect but not a small one.

Most other links between a pattern and death come from cohorts. In drug trials, people who stick to anything, even a placebo, already do better.

The gain is largest for people at high risk or eating poorly. For a low-risk person who already eats reasonably, the gain is small — a reason to stop optimizing. High-fat low-carb diets carry a caution for people whose LDL climbs on them.

Guideline bodies also cite an environmental case for plant-heavy diets. It is a separate question from these health findings.

How sure: medium for the differences in surrogates (weight, blood pressure, LDL); low for events and mortality.

This guide covers whole-diet patterns: Mediterranean, DASH, vegetarian and vegan, low-carb and ketogenic, low-fat, paleo, MIND, Portfolio, carnivore, and commercial programmes such as Atkins or Weight Watchers. Single foods have their own pages (Food Groups and Health Outcomes - A Dose-Response Matrix). When a food category is too mixed to carry an effect, see Is the Food Category Doing Any Work. For how diet ranks against other levers, see Layer 1 - Ranking Interventions for a Stratum.

The first screen is the one the evidence can actually enforce: does the diet supply what the body cannot make?

Some diets are worse by construction: they leave out nutrients or mismatch energy

A diet can be ruled out for leaving essential nutrients uncovered, or for supplying far too much or too little energy for years on end. The same test cannot rule a diet in: clearing the floor shows only that a diet is not deficient, not that it is good. The floor uses these two harms and no others because they are the least contested in nutrition. Saturated fat, red meat and carbohydrate share stay out of it; they return later as contested levers above the floor.

Correcting a deficiency helps the deficient, not the already replete

A supplement does two different jobs depending on where a person starts. Below the requirement, correcting the shortfall can move an outcome; above it, more buys nothing, and well past need it can harm -> Deficiency Repletion vs Enhancement. Iodine shows the lower arm most clearly. A meta-analysis of 24 studies in children aged 5 and under put the cost of iodine deficiency at about 7.4 IQ points, and the four study designs it pooled gave figures from 6.9 to 10.2 points — a spread between designs, not a confidence interval (Bougma et al., 2013). Only two of the studies were randomized, samples were small and the observational studies left confounders out, so the direction is secure and the exact size is not (Bougma et al., 2013).

A Cochrane review restricted to randomized trials sharpens that picture. In severe deficiency, two old injected-iodine trials raised child IQ by 11.21 points (95% CI 7.96 to 14.46); in mild-to-moderate deficiency, the one oral trial (53 children) found no difference (Harding et al., 2017); the injections are described in (Harding et al., 2017). Harding co-authored the Bougma review and the two share their severe-deficiency trials, so the Cochrane result refines the same evidence rather than confirming it independently (Bougma et al., 2013; inferred from Harding et al., 2017). The benefit sits where deficiency is deep.

Plant-based diets fall short on a predictable set of animal-sourced nutrients

One systematic review compared nutrient intake and blood status across 141 studies of adults on vegan, vegetarian and meat-containing diets, mostly in Europe, South and East Asia, and North America (Neufingerl & Eilander, 2021). For most of the nutrients that come mainly from animal foods, risk is highest in vegans:

Nutrient	Vegan	Vegetarian	Meat-eater	Note
B12 deficiency (mean prevalence)	~44%	~32%	0-16% (range)	serum marker, which understates deficiency
Iodine deficiency (urine <100 ug/L)	92%	37%	51%	three studies measured it — thin
Iron deficiency	15%	11%	7%	women most affected
Zinc deficiency	30%	14%	13%	phytate blocks absorption
EPA/DHA intake	27 / 4 mg/d	16 / 31 mg/d	94 / 172 mg/d	vegan/vegetarian far below 250 mg; meat-eaters about at it

Figures: (Neufingerl & Eilander, 2021), (Neufingerl & Eilander, 2021); the meat-eater B12 range from (Neufingerl & Eilander, 2021); the serum-marker caveat from (Pawlak et al., 2013).

Iron is the instructive case. Vegans eat more iron than meat-eaters (21.0 vs 13.9 mg a day), yet their ferritin is lower (31.3 vs 55.5 ug/L, and 24.3 in vegetarian women), because plant iron is absorbed at roughly 10% against 18% for a diet with meat and fish (Neufingerl & Eilander, 2021) (Neufingerl & Eilander, 2021). Intake, absorption and blood level can point in opposite directions, so an intake count that meets the requirement can still leave a shortfall. Zinc carries the opposite caveat: the authors found no reported adverse health effects in adult vegetarians with lower zinc status (Neufingerl & Eilander, 2021).

Two limits apply to the whole panel. The authors compared average intakes to requirements rather than counting people below them: «A major limitation of our review was that we could not provide reliable estimations on the adequacy of dietary nutrient intakes.» (Neufingerl & Eilander, 2021). And iodine and EPA/DHA rest on very few studies, so the 92% figure is a warning, not a firm number.

Meat-eaters fall short too, on different nutrients

The same review found meat-eaters at risk of low intakes of fibre, polyunsaturated fat, ALA (men), protein, folate, niacin, B12, calcium, vitamin E and magnesium (Neufingerl & Eilander, 2021), and every pattern fell short on one nutrient: «Average vitamin D intake was below the EAR (i.e., 10 µg/d) in all dietary patterns indicating high prevalence of inadequate intakes in the overall population.» (Neufingerl & Eilander, 2021). The authors’ summary: «There were nutrient inadequacies across all dietary patterns, including vegan, vegetarian and meat-based diets.» (Neufingerl & Eilander, 2021).

Both sets of shortfalls rest on the same kind of evidence: intake or blood level against a requirement, with no disease outcome measured on either side. Any narrow diet can fail the floor; a stricter plant-based one fails it more often on animal-sourced nutrients, which argues against excluding foods for its own sake.

Both authors work for Unilever, which markets fortified foods and vitamin and mineral supplements, among other products (Neufingerl & Eilander, 2021). The deficiency tables do not run against that interest, so they earn credit from their methods, not from their sponsor. Their framing goes further than their data: the abstract calls plant-based diets «generally better for health and the environment» (Neufingerl & Eilander, 2021), but the review measured blood markers, not disease, and holds no environmental data.

B12 is the one gap a vegan diet cannot close with food

Only microorganisms make B12, so plant foods carry none unless fortified (Pawlak et al., 2013). A review of vegetarian B12 status concluded that «Supplements, when taken in an adequate dose, are very effective in both the prevention and treatment of B12 deficiency and are very cost effective» (Pawlak et al., 2013). For vegans, that makes a B12 supplement or fortified food close to non-optional — a cheap, low-harm fix for a gap the diet itself creates. Choosing the dose needs a person’s own blood results and is a clinician’s call.

A deficient blood test is not yet a harm

Every number in the panel above is a blood level or an intake, not an illness -> Surrogate Outcomes. In one study Pawlak cites, «none of the vegetarians included in their study had clinical symptoms despite the fact that about two-thirds of the sample had B12 depletion or deficiency» (Pawlak et al., 2013). Pawlak warns against reading that as harmless: mild symptoms can go unnoticed, and iron deficiency or high folate can mask the blood signs (Pawlak et al., 2013). No held trial tests whether correcting these diet-related shortfalls changes a disease outcome. Deficiency causes harm where it is deep, as iodine shows. The mild shortfalls these surveys find are a risk to manage, not a measured harm.

Iodine in pregnancy: the trials are too thin to settle routine supplements

The Cochrane review of 14 randomized or quasi-randomized studies concluded: «There were insufficient data to reach any meaningful conclusions on the benefits and harms of routine iodine supplementation in women before, during or afer pregnancy.» (Harding et al., 2017). That is too little evidence, not evidence of no effect, and the benefit that does appear sits in severe deficiency. In one trial of 356 women, supplements also raised excessive intake: 7.0% of supplemented women against 1.6% of others (RR 4.33, 95% CI 1.24 to 15.07) (Harding et al., 2017) -> Iodine Supplementation in Pregnancy.

Energy: chronic surplus or deficit fails whatever the label

A diet that supplies far more energy than a person uses, or far less, for years fails the floor whether it is called Mediterranean, vegan or keto. The diets that make a surplus easy, through ultra-processed food and energy-dense formats, are covered below under Shared foods, total energy and staying on the diet carry the difference, and they overlap.

An all-butter or all-fast-food diet shows how the floor works. It fails because it leaves out most vitamins, minerals and fibre and makes a chronic energy surplus easy — no argument about saturated fat or red meat is needed, and none is used.

Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception

Most diets people actually choose between clear this floor. The next question is whether the name on the diet matters once they do. On weight and blood pressure it barely does: in the largest trial network held, named diets differ from one another by about a kilogram and under 2 mmHg, below the thresholds the reviewers set in advance for a difference worth having. LDL cholesterol is the one risk factor where the labels split.

Weight and blood pressure: low-carb, low-fat and the branded programmes land in the same place

Ge and colleagues placed 14 named programmes (Atkins, DASH, Zone, Ornish, Mediterranean, Weight Watchers, paleolithic and others) and three macronutrient patterns into one network meta-analysis of 121 randomized trials in 21,942 adults with overweight or obesity, and graded every comparison (Ge et al., 2020). Against usual diet at 6 months, low-carb and low-fat diets came out almost identical: 4.63 vs 4.37 kg lost, and systolic blood pressure down 5.14 vs 5.05 mmHg (Ge et al., 2020). Those gains over usual diet are real. Ge set its bar for an important difference before looking: «weight loss 2 kg, systolic blood pressure 3 mm Hg, diastolic blood pressure 2 mm Hg, LDL cholesterol 5 mg/dL» (Ge et al., 2020).

The gaps between diets fall under that bar. The largest pattern-to-pattern gap, low-carb vs moderate-macronutrient, was 1.57 kg (95% credible interval 0.86 to 2.29) and 1.66 mmHg systolic at 6 months; Atkins beat Zone by 1.38 kg (0.15 to 2.62) (Ge et al., 2020). Ge’s summary: «Differences between diets were typically small to trivial and often based on low certainty evidence» (Ge et al., 2020). The comparison ran at scale and came back near-null. That makes it a finding of no meaningful difference, at low-to-moderate certainty, not a gap in the evidence.

The Cochrane review of low-carb against balanced-carbohydrate diets (61 trials) reaches the same size. In adults without diabetes, low-carb diets probably produce about 1 kg more weight loss: 1.07 kg (0.59 to 1.55) at 3 to under 12 months and 0.93 kg (0.04 to 1.81) at 12 months or more, both at moderate certainty, a difference the review judges not clinically important (Naude et al., 2022). When both arms are prescribed the same energy, the gap shrinks further (the numbers are in Shared foods, total energy and staying on the diet carry the difference, and they overlap). DIETFITS, a 12-month trial in 609 adults, found 0.7 kg (95% CI -0.2 to 1.6) between healthy low-carb and healthy low-fat, not significant (Gardner et al., 2018).

These three estimates draw on overlapping trials and closely related contrasts, so they are one body of evidence read three ways, not three separate checks. Controlled feeding studies explain why the macronutrient label adds so little (under Total energy).

The gains over usual diet mostly fade within a year

At 12 months the advantage over usual diet shrinks for every pattern: weight loss falls from 4-5 kg at 6 months to about 3 kg, and «At 12 months the effects on weight reduction and improvements in cardiovascular risk factors largely disappear.» (Ge et al., 2020) Ge could not measure adherence (see Staying on the diet). Every outcome in the network is a surrogate measured for at most a year; no trial in it compared named diets on strokes, heart attacks or deaths -> Named Diet Programs Compared, Surrogate Outcomes.

LDL cholesterol is where the labels split

Against usual diet at 6 months, low-fat diets lowered LDL by 7.08 mg/dL and moderate-macronutrient diets by 5.22 mg/dL, both at moderate certainty, while low-carb diets lowered it by only 1.01 mg/dL (interval -2.96 to 4.96, crossing zero; low certainty) (Ge et al., 2020). So low-carb ties on weight and blood pressure and does worse on LDL. That is a surrogate result. What it means for people whose LDL rises sharply on high-fat low-carb eating is covered under Who gains: people at high risk, people who eat poorly, and a few groups who need specific care. The low-carb mortality cohorts are in Vegetarian, paleo and low-carb patterns rest on cohorts; carnivore rests on almost nothing.

One label kept an LDL difference past six months: «Weight reduction at the 12 month follow-up diminished, and aside from the Mediterranean diet for LDL reduction, improvements in cardiovascular risk factors largely disappeared» (Ge et al., 2020). That is one surrogate in one network. It does not show the Mediterranean diet beating DASH or a vegetarian diet on events.

Reviews of mortality, guideline bodies and the diabetes literature read it the same way

The US Dietary Guidelines Advisory Committee’s 2020 review graded protective patterns as Strong for an association with lower all-cause mortality. It graded the Mediterranean, Healthy Eating Index, DASH and plant-based scores together: «This highlights that a high-quality dietary pattern comprised of nutrient-dense foods, regardless of the label, associated with reduced all- cause mortality risk.» (Boushey et al., 2020) That is a cohort association; the next section asks how far it can be trusted. Dietary Patterns and All-Cause Mortality

Diets defined by their macronutrient split could not be graded at all (Grade not assignable), though 28 articles had studied them (Boushey et al., 2020). The committee blamed the contrasts: the macronutrient proportions being compared were often as close as 41% vs 41.7%, and measured by differing methods (Boushey et al., 2020). Where macronutrient share did track mortality, food quality travelled with it: diets inside the acceptable ranges «tended to associate with reduced all-cause mortality risk, particularly when the diets examined were of higher quality», though comparisons with and without the food context gave inconsistent findings (Boushey et al., 2020).

In PURE, the Mediterranean, HEI-2010 and DASH scores predicted mortality almost equally (HR per 20-percentile step in the score 0.96, 0.93 and 0.96; AUC 0.55, 0.55 and 0.53), all scored on PURE’s own questionnaire (Mente et al., 2023) -> Diet Quality Scores and Cardiovascular Risk. PURE’s own score did slightly better on its own data, HR 0.91 (0.89-0.93) with an AUC of 0.61 for mortality and 0.54 for major cardiovascular disease (Mente et al., 2023). Every score sorted people only weakly, and the score built on these participants had the home advantage.

The same verdict holds in type 2 diabetes. An umbrella review of hypocaloric-diet meta-analyses concluded that they «do not support any particular macronutrient profile or style over others» (Churuangsuk et al., 2021). It pools the same class of trials Ge draws on, so it extends the finding to that group rather than confirming it independently -> Diets for Weight Management in Type 2 Diabetes.

NICE assessed low-carb, very-low-carb and low-fat diets and attached no recommendation to any of them. The only diet it recommends is a low-energy diet (total or partial meal replacement, 800-1200 kcal/day, for at most 12 weeks), defined by energy. NICE adds that the energy deficit can be reached by lowering specific macronutrient content (National Institute for Health and Care Excellence, n.d.) -> Diets for Weight Loss - What NICE Recommends. Here current guidance and the evidence agree. That agreement deserves reporting as plainly as a disagreement would.

Where labels do separate, the gap follows particular foods

In the one cohort that scored both (REGARDS), Mediterranean and paleo scores separated, and the gap tracked nuts and red and processed meat (figures under How large the healthy-adherer effect can be and Paleolithic). It is one observational cohort, and it refines the label barely matters reading without overturning it.

Ge also found plant-heavy and omnivorous programmes similar on its risk factors, and set that result against US, Canadian and EAT-Lancet advice to eat plant-based diets: «To the extent that short term results might have implications for long term cardiovascular outcomes, our findings do not support this conclusion: rather, they suggest that omnivorous based diets (eg, Atkins, Zone) have a similar effect to diets that tend to be higher in plant based foods (eg, Ornish, DASH, Mediterranean).» (Ge et al., 2020). That is a statement about weight, blood pressure and lipids over a year, not about the long-term cohort outcomes of vegetarian diets.

Nearly all pattern-and-death evidence comes from cohorts, and placebo adherence in drug trials is about as large as the crude diet signal

Named diets get trials on weight and blood pressure, but almost never trials that count strokes and deaths. That limits how far the rest of the evidence can be trusted.

A Strong grade built on cohorts

The US 2020 Dietary Guidelines review (NESR) graded the link between protective dietary patterns and lower all-cause mortality Strong. Of the 153 articles behind that grade, 152 were prospective cohort studies and one was a randomized trial (Boushey et al., 2020). That one trial was PREDIMED, and it did not detectably change total deaths (its figures are under One trial cut strokes in high-risk adults; it did not detectably change the number of deaths). NESR reached Strong on its own scale after rating the cohort body on risk of bias, consistency, directness, precision and generalizability (Boushey et al., 2020).

Under GRADE, a body of observational studies starts at low certainty and climbs only on explicit grounds such as a large effect or a dose-response gradient (Schünemann et al., n.d.) -> Rating Certainty of Evidence. Both labels are accurate. Strong here describes a consistent association; it is not a statement that the pattern causes fewer deaths. So the right verb for a pattern and mortality is is associated with, never reduces.

NESR also lists limits that sit alongside its Strong grade. Most cohorts enrolled only people without chronic disease, so NESR judged it «likely that the reported results are biased towards those individuals who are healthier than the general population» (Boushey et al., 2020). It still judged the results applicable across race, ethnicity and income because of the volume and consistency of the evidence. Most studies measured diet once, at baseline, then followed people for years (Boushey et al., 2020). Several cohorts also failed to adjust for income, physical activity, smoking or BMI, and race was under-reported (Boushey et al., 2020) -> Dietary Patterns and All-Cause Mortality.

How large the healthy-adherer effect can be

People who stick to any health advice tend to do better on everything else too: they smoke less, move more, get screened. Three held results put a size on that effect.

An inert pill. Simpson pooled the placebo arms of eight drug trials (19,633 people). Those who took their placebo faithfully died less often than those who did not: «Good adherence to placebo was associated with lower mortality (0.56, 0.43 to 0.74), as was good adherence to beneficial drug therapy (0.55, 0.49 to 0.62).» (Simpson et al., 2006) A placebo has no route to fewer deaths, so that whole gap belongs to the kind of person who adheres -> Diet Quality Scores and Cardiovascular Risk.

A diet score. In PURE (147,642 adults in 21 countries), people with the highest healthy-diet score had an all-cause mortality HR of 0.54 (0.50-0.59) against the lowest, adjusted only for age and sex. Full adjustment for lifestyle and income moved it to 0.70 (0.63-0.77) (Mente et al., 2023). Simpson’s figure is an odds ratio and PURE’s a hazard ratio, so they are not the same quantity; they are near-identical in size, and both are largely unadjusted behaviour-to-death associations.

Two named patterns in one cohort. In REGARDS (a US cohort described under Paleolithic), Whalen compared top and bottom fifths of a Mediterranean and a Paleolithic diet score. Adjusted for age, sex, race and energy only, the all-cause HRs were 0.49 (0.42-0.56) and 0.59 (0.51-0.67). After further adjustment, «particularly smoking status», they moved to 0.64 (0.55-0.74) and 0.77 (0.67-0.89), and both stayed statistically significant (Whalen et al., 2017).

Read together, the three results show that a diet-to-death association no stronger than about HR 0.56 (anything from 0.56 up to 1) sits inside what the adherer effect can produce with no causal help from the diet. That does not make PURE’s adjusted 0.70 or Whalen’s adjusted figures pure confounding. Adjustment removes the part the study could measure, and Simpson’s 0.56 is an unadjusted ceiling, not the residual. It does mean that a 20-30% adjusted reduction cannot be taken as causal on its size alone. The same caution applies to PREDIMED’s adherence-adjusted estimate, reported under One trial cut strokes in high-risk adults; it did not detectably change the number of deaths.

Asking people what they ate

Every cohort above measured diet by asking people, usually with a food-frequency questionnaire. In the UK national diet survey, adults aged 16-64 reported energy intake that was, on average, 34% below the energy they actually spent, measured by doubly-labelled water; for children aged 4-10 the gap was 12% (Scientific Advisory Committee on Nutrition, 2015). Error unrelated to the outcome shrinks a real association toward no effect in a simple model; once confounders are themselves measured badly, the bias can run in either direction (Willett, 2012). So a flat or null gradient in a diet cohort is weak evidence that no gradient exists. Residual confounding and measurement error also push a pattern estimate in opposite directions, by amounts no one has measured -> Measurement Error in Dietary Assessment.

Where the randomized trials are

Few named patterns have been randomized at all, and fewer on events. The map below gives direction only; each number lives in the section named.

Pattern	Trial(s)	Outcome type	Direction	Where the numbers are
Mediterranean	PREDIMED (high-risk, no CVD at entry)	events, deaths	fewer strokes; deaths not detectably changed	One trial cut strokes in high-risk adults; it did not detectably change the number of deaths
Mediterranean	Lyon (after a heart attack, 1990s)	deaths	fewer deaths; one old trial	One trial cut strokes in high-risk adults; it did not detectably change the number of deaths
Low-fat	WHI and pooled fat-reduction trials	events	no significant effect	DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome
MIND	one 3-year trial	cognition	no significant effect	DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome
DASH	feeding trials	blood pressure	lower	DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome
Portfolio	feeding trials	LDL, apoB	lower	DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome
Named diets, low-carb vs balanced	network and Cochrane pools of weight trials	weight, BP, LDL	small differences	Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception
Vegetarian, paleo, low-carb (long term), carnivore	none	—	cohort or nothing	Vegetarian, paleo and low-carb patterns rest on cohorts; carnivore rests on almost nothing

— compiled from the per-pattern sections; no single source maps it.

When cohorts and trials disagree

For the Mediterranean diet, Dinu’s umbrella review found strong observational support for lower overall mortality, cardiovascular disease, coronary disease, heart attack and diabetes. For the five outcomes that both designs had examined, the pooled trials were «the latter showing no evidence (except for diabetes)» (Dinu et al., 2017). That trial pool is dominated by PREDIMED, and its lone diabetes trial is very likely a PREDIMED substudy, so it refines the picture rather than confirming it independently -> Mediterranean Diet and Cardiovascular Events. MIND shows the same shape: positive cohorts, a null trial (under DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome).

A split like this can mean the two designs measured different exposures: a few years of a supplied or coached diet versus decades of habitual eating. A null trial then says little about the lifelong pattern, and a positive cohort says little about cause until confounding is dealt with. Neither stream settles the question alone -> The Observational-Trial Discordance.

Attention runs opposite to evidence

Several heavily promoted patterns (keto, paleo, carnivore) sit on the thinnest hard-outcome evidence. Ketogenic and low-carb diets have no long event trial; paleo rests on one cohort; carnivore has no held trial or cohort at all. -> Surrogate Outcomes

Among named-pattern trials, only Mediterranean ones counted events and found fewer of them: PREDIMED and the older, smaller Lyon trial in heart patients. PREDIMED’s result is narrower than its reputation.

One trial cut strokes in high-risk adults; it did not detectably change the number of deaths

The Mediterranean diet is the only named pattern whose large randomized trial reduced cardiovascular events (a composite of heart attack, stroke and cardiovascular death, carried mostly by strokes); the low-fat WHI trial was null. That trial, PREDIMED, probably reduced strokes in people at high cardiovascular risk. It did not detectably change how many of them died.

What PREDIMED tested

PREDIMED enrolled 7447 Spanish adults at high cardiovascular risk but without cardiovascular disease, so it is a high-risk primary-prevention trial, not a trial in heart patients. Most entrants had type 2 diabetes or at least three major risk factors (about half had diabetes, and about 82% hypertension). (Estruch et al., 2018)

Two groups were advised to eat a Mediterranean diet and were given free extra-virgin olive oil or mixed nuts; the control group was advised to cut fat. Over a median 4.8 years, the combined Mediterranean groups had fewer heart attacks, strokes and cardiovascular deaths taken together: HR 0.70 (0.55-0.89), a 5-year risk of 36 (olive oil) and 40 (nuts) vs 57 per 1000. The olive oil and nuts were donated by producers, which had no role in the design, analysis or reporting. (Estruch et al., 2018) See Mediterranean Diet and Cardiovascular Events.

Stroke carries the result; deaths did not detectably change

Rees’s Cochrane review split the trial’s composite into its parts and graded each one, per 1000 people over about 4.8 years (Rees et al., 2019):

Outcome	Low-fat advice -> Mediterranean, per 1000	HR (95% CI)	How sure
Stroke	24 -> 14	0.60 (0.45-0.80)	moderate
Heart attack	16 -> 12	0.79 (0.57-1.10)	low
Cardiovascular death	12 -> 10	0.81 (0.50-1.32)	low
Death from any cause	47 -> 47	1.00 (0.81-1.24)	low

The diet probably prevents about 10 strokes per 1000 high-risk people over five years. It may lower heart attacks and cardiovascular deaths, but both intervals include no effect. Total deaths were the same in both arms; the trial’s own report gives HR 0.98 (0.77-1.24). (Estruch et al., 2018) So the composite’s 30% is mostly a stroke result.

The USDA’s pattern review (NESR) summarizes PREDIMED more favourably in its prose: «Consumption of the Mediterranean diets with EVOO or mixed-nuts were significantly associated with reduced all-cause mortality risk after (median) 4.8y.» (Boushey et al., 2020) Its own evidence table gives the intention-to-treat all-cause hazard ratios as 0.90 (0.69-1.18) for olive oil, 1.12 (0.86-1.47) for nuts and 0.98 (0.77-1.24) combined, all null. (Boushey et al., 2020) This guide uses the table. -> Dietary Patterns and All-Cause Mortality

Rees also explains why the review cannot go beyond one trial here: «Only one trial reported clinical endpoints for primary prevention and this study experienced methodological issues regarding randomisation with the report subsequently being retracted and re-analysed (PREDIMED).» «In addition, both the PREDIMED trial and The Lyon Diet Heart Study supplied supplemental foods as well as dietary advice to follow a Mediterranean-style diet so the policy implications of the findings of these trials are unclear (Appel 2013).» (Rees et al., 2019) The 2018 re-analysis used propensity scores to repair the randomization problems, and the result held when the 1588 participants affected were dropped (HR 0.69, 0.53-0.92). It is an RCT with a repair, not a clean one. (Estruch et al., 2018)

What changed in the plate, and what did not

The trial supplied food. Advice to eat a Mediterranean diet, without free olive oil or nuts, has not been tested against events. PREDIMED also left energy and exercise alone: «No total calorie restriction was advised, nor was physical activity promoted.» (Estruch et al., 2018) Strokes fell without a weight-loss programme. The contrast with weight-loss trials is under Shared foods, total energy and staying on the diet carry the difference, and they overlap (Does Weight Loss Reduce Cardiovascular Events).

The dietary contrast was modest. Most participants already ate close to a Mediterranean pattern, the control group got healthy-diet advice, and the largest between-group differences were in fat type, from the supplied oil and nuts. (Estruch et al., 2018) That makes the result informative per unit of change, but it is no licence to expect the same from pouring olive oil over a poor diet. The control arm was low-fat advice, so the trial compared one fat profile with another, not fat with no fat.

The per-protocol figure is larger, and less trustworthy

Among participants who stuck to their assigned diet, the hazard ratio for the composite was 0.42 (0.24-0.63). (Estruch et al., 2018) That is a per-protocol estimate, not the effect of offering the diet. People who keep to any regimen tend to do better for reasons unrelated to it. That healthy-adherer effect is sized under Nearly all pattern-and-death evidence comes from cohorts, and placebo adherence in drug trials is about as large as the crude diet signal. So 0.42 overstates what the diet itself does, by an unknown amount.

Risk factors move a little

Against another diet, mostly low-fat advice, the Mediterranean pattern probably lowers LDL cholesterol by about 0.15 mmol/L (-0.27 to -0.02). Against minimal intervention it probably lowers blood pressure by about 2.99/2.0 mmHg, though that estimate rests on 2 trials and 269 people. Against another diet the blood-pressure gap shrinks to an imprecise -1.5 mmHg. (Rees et al., 2019) Harms were rarely recorded: «Two trials reported on adverse events where these were absent or minor (low- to moderate-quality evidence).» (Rees et al., 2019) No trial reported quality of life or costs. (Rees et al., 2019)

Lyon: a heart-patient trial with a large effect and low certainty

The Lyon Diet Heart Study asks a different question. It enrolled 605 people who already had coronary heart disease, gave Mediterranean-diet advice plus a canola-based margarine, and compared them with usual care. The trial stopped early at a March 1993 interim analysis, with follow-up extended to a mean 46 months. Total deaths fell from 79 to 35 per 1000, HR 0.44 (0.21-0.92). (Rees et al., 2019) (Rees et al., 2019)

Rees rates this low certainty: one older trial, an unclear randomization method and a modified Zelen design. (Rees et al., 2019) (Rees et al., 2019) Lyon may reduce deaths in heart patients, but its size should not be carried forward, and it should not be pooled with PREDIMED: the population, the era and the comparator all differ.

Who this transfers to

PREDIMED’s absolute gain depends on its high-risk entrants; the same relative effect buys far less at low baseline risk (Baseline Risk and the Relative-Absolute Split). The authors flag it themselves: «An additional limita- tion of our study is that participants were at high cardiovascular risk; whether the results can be generalized to persons at lower risk requires further research.» (Estruch et al., 2018) The absolute numbers for a lower-risk reader are worked through under Switching labels moves less than leaving a typical diet; at low risk both gains are small.

The Mediterranean diet is the only pattern with a positive event trial. It has not been shown to beat DASH or a vegetarian diet on events; neither has been tested that way.

Other patterns have been randomized too, but on blood pressure, cholesterol or cognition — or, when events were counted, with a null result.

DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome

Four more named patterns have been randomized. Two of them, DASH and Portfolio, move a risk factor reliably in trials lasting weeks, and neither has been tested on heart attacks, strokes or deaths. The other two were tested on a hard outcome over years: the low-fat trial counted heart disease and the MIND trial measured cognition, and both found no significant effect on the contrast they actually achieved. Each pattern below gets the same five points: best design held, risk-factor result, hard-outcome result, known harm or caveat, and what is not held.

DASH: probably lowers blood pressure by about 3-5 mmHg; its effect on events is untested

Best evidence. Two pooled analyses of randomized trials against a control diet, plus one controlled-feeding trial that sits inside both pools.

Blood pressure. The two pools give two different numbers, and neither replaces the other. Siervo pooled 20 RCTs (1917 people, 2-24 weeks) on change from baseline and found systolic -5.2 mmHg (-7.0 to -3.4) and diastolic -2.6 (-3.5 to -1.7). (Siervo et al., 2014) Filippou pooled 30 RCTs (5545 people, average baseline 134/85 mmHg, mean follow-up about 15 weeks) on attained blood pressure and found -3.2 (-4.2 to -2.3) and -2.5 (-3.5 to -1.5), with moderate certainty because many trials were unblinded or lost data. (Filippou et al., 2020) Half of Filippou’s trials are Siervo’s, and the two use different metrics, so the gap between -3.2 and -5.2 cannot be pinned on either cause. (inferred from Filippou et al., 2020; Siervo et al., 2014)

Who responds. Filippou found the same systolic fall in people with and without hypertension (-3.9 vs -3.9 mmHg, P=0.96) and no significant difference by BMI (-3.9 vs -2.6, P=0.12). (Filippou et al., 2020) That weakens Siervo’s between-trial signal of a larger fall at higher baseline blood pressure. The safer reading stratifies by overall cardiovascular risk; it does not promise a bigger mmHg drop to someone with higher blood pressure. (inferred from Filippou et al., 2020; Siervo et al., 2014)

DASH and salt do not add up. The DASH-Sodium trial fed 412 US adults with above-optimal blood pressure DASH or a typical-US control diet at three sodium levels for 30 days each, with all food supplied and weight held constant. (Sacks et al., 2001) DASH lowered systolic pressure by 5.9 mmHg (-8.0 to -3.7) at high sodium but by only 2.2 (-4.4 to -0.1) at low sodium. DASH plus low sodium, against the control diet at high sodium, gave -8.9 mmHg (-6.7 to -11.1). (Sacks et al., 2001) Adding the two single effects would have predicted about -12.6 mmHg. (inferred from Sacks et al., 2001) The authors put it directly:

«The reductions in blood pressure caused by the combination of dietary interventions were smaller than they would have been if the effects of each dietary in- tervention were strictly additive (P<0.001 for the in- teraction).»

(Sacks et al., 2001)

Both levers together still beat either alone; the combined figure comes from the trial, not from a sum. This is the one direct test held of how pattern levers combine. The trial is one of the pooled trials, so it adds structure to the pooled DASH effect rather than a second witness to it. (Filippou et al., 2020; inferred from Sacks et al., 2001; Siervo et al., 2014)

Events. No randomized trial of DASH has counted heart attacks, strokes or deaths. Cohorts in which people scored higher on DASH adherence had fewer cardiovascular events (RR 0.80, 0.76-0.85), but an umbrella review rated that evidence low (the GRADE starting level for cohort data; no serious risk of bias was identified); no absolute figure is held. (Chiavaroli et al., 2019) An adherence score also tracks many other healthy habits. The same umbrella’s blood-pressure estimate is Siervo’s figure re-imported, so it is not a third estimate. (inferred from Chiavaroli et al., 2019; Siervo et al., 2014)

Harm or caveat. None reported in the held sources. Some pooled trials combined DASH with salt restriction, weight loss or exercise (Filippou et al., 2020), so part of the -3.2 mmHg may be co-intervention.

Not held. An event trial, and any multi-year trial (mean duration 15.3 weeks; longest about one year) (Filippou et al., 2020). Other blood-pressure levers are on Sodium Intake and Blood Pressure and Blood Pressure Lowering and Cardiovascular Events; the pattern detail is on DASH Diet and Blood Pressure.

Portfolio: lowers LDL by about a sixth in people with high cholesterol, measured over about a month

Best evidence. One meta-analysis of 7 controlled-trial comparisons in 439 people with high cholesterol, adding nuts, plant protein, viscous fibre and plant sterols to a cholesterol-lowering background diet and comparing against that background diet alone, over a median of about 4 weeks. (Chiavaroli et al., 2018)

Lipids. LDL fell by 0.73 mmol/L (-0.89 to -0.56), about 17%, and apoB by 15%, both at high certainty. Body weight did not change (-0.10 kg, p=0.59). Blood pressure fell by 1.75/1.36 mmHg, at moderate certainty because the intervals included the pre-set threshold for importance. (Chiavaroli et al., 2018) In the tightly controlled feeding trials LDL fell 21%; the free-living trials pulled the pooled figure down, so the fall a person gets depends on how fully they keep all four components. (Chiavaroli et al., 2018)

Events. None counted. The 1.34-point drop in estimated (Framingham) 10-year coronary risk, about 13 per 1000 lower (5 to 22), moderate certainty, in hyperlipidaemic adults, is calculated from the lipid and blood-pressure changes, not observed. (Chiavaroli et al., 2018) (Chiavaroli et al., 2018) The authors say so:

«In this regard, there remains a need for large randomized trials of the effect of the portfolio dietary pattern on hard CV outcomes or surrogate end- points of atherosclerosis.»

(Chiavaroli et al., 2018)

Harm or caveat. No harm reported. All 7 comparisons come from one investigator group, the diet’s inventors, with food-industry funding across the author list. (Chiavaroli et al., 2018) So the pooled figure is the ceiling an efficacy-tuned, single-group evidence base produces.

Not held. An event or plaque trial (PortfolioEx is awaited), trials by independent groups, and trials in people with normal cholesterol. Detail: Portfolio Dietary Pattern and LDL Cholesterol.

Low-fat: cutting total fat did not reduce heart disease in the largest trial or in pooled trials

Best evidence. The Women’s Health Initiative trial randomized 48,835 postmenopausal women aged 50-79 to a low-fat eating pattern or no diet change and followed them for a mean 8.1 years. The target was 20% of energy from fat, made up with vegetables, fruit and grains, with no weight-loss goal, so fat was mainly replaced by carbohydrate. (Howard et al., 2006)

Risk factors. LDL fell by 3.55 mg/dL and diastolic pressure by 0.31 mmHg; HDL, triglycerides, glucose and insulin did not change. (Howard et al., 2006)

Events. Coronary heart disease HR 0.97 (0.90-1.06), stroke 1.02 (0.90-1.15), all cardiovascular disease 0.98 (0.92-1.05). (Howard et al., 2006) The women never reached the target: the gap between arms was about 10.7 points of energy from fat at year 1 and about 8.2 at year 6.

«However, the achieved difference in percentage en- ergy from fat was only about 70% of de- sign assumptions.»

(Howard et al., 2006)

The authors projected only about 40% power for that smaller contrast, and predicted that an LDL change this small would cut cardiovascular risk by only 2-4%. (Howard et al., 2006) So the trial shows no meaningful effect of this particular contrast; it does not show that diet does nothing for the heart. (inferred from Howard et al., 2006)

The pooled trials. Hooper’s Cochrane review found the same for the whole class of fat-reduction trials against usual diet: RR 0.97 (0.87-1.08) across 50,655 participants. (Hooper et al., 2012) That pool is dominated by WHI, which alone contributed over 48,000 women without cardiovascular disease (Hooper et al., 2012), so it largely restates the WHI result rather than generalizing it. The same review found that changing the type of fat did reduce events, which is a different exposure (overall RR 0.86, 0.77-0.96, moderate certainty, driven by the fat-modification trials):

«The findings are suggestive of a small but potentially important reduction in cardiovascular risk on modification of dietary fat, but not reduction of total fat, in longer trials.»

(Hooper et al., 2012)

PREDIMED’s comparison group received low-fat advice, which fits the same split. What the split means for components is taken up below under Shared foods, total energy and staying on the diet carry the difference, and they overlap.

Harm or caveat. No harm on events (stroke HR 1.02). The tested swap was fat for carbohydrate, which is not the pattern current guidance recommends, as the WHI authors note. (Howard et al., 2006)

Not held. No absolute event figure is held, and the direct trials of reducing fat against changing fat type (three small trials, 912 participants, 28 events) are too small to read. Detail: Low-Fat Dietary Pattern and Cardiovascular Disease. (Hooper et al., 2012)

MIND: a 3-year trial found no significant benefit for cognition; it rules out a large effect, not a small one

Best evidence. One RCT: 604 adults aged 65 or older with a family history of dementia, overweight and eating a suboptimal diet, randomized to the MIND diet or a control diet for 3 years. Both arms were asked to cut calories slightly. (Barnes et al., 2023)

Cognition and brain scans. Global cognition improved in both groups; the difference was 0.035 standard units (-0.022 to 0.092, P=0.23). Brain MRI measures did not differ either. The trial had over 90% power to detect a difference of 0.02 SD per year in the rate of decline. (Barnes et al., 2023)

«CONCLUSIONS Among cognitively unimpaired participants with a family history of dementia, changes in cognition and brain MRI outcomes from baseline to year 3 did not differ significantly between those who followed the MIND diet and those who followed the control diet with mild caloric restriction.»

(Barnes et al., 2023)

Why the null is narrow. The control group also changed its diet and lost about as much weight (-4.8 kg vs -5.0 kg), so the trial compared MIND with a diet that was itself improving, not with a poor diet. (Barnes et al., 2023) Any trial result has to be read against its comparator (The Comparator Problem).

In cohorts of 26,103 people, each 1-SD higher MIND score was associated with 0.042 standard units better cognition (0.020 to 0.065). (Huang et al., 2023) The trial did not supply causal support for the cohort association; the two figures measure different quantities (cohort level per SD vs 3-year trial change). The trial rules out a large 3-year effect in this group; it does not show the direction is wrong. (inferred from Barnes et al., 2023; Huang et al., 2023)

Harm or caveat. No harm reported. The volunteers were selected and motivated, which limits how far the result travels in either direction. (inferred from Barnes et al., 2023)

Not held. A longer trial, or one in a less selected population. Detail: MIND Diet - Observational Benefit vs Randomized Null and MIND Diet and Cognitive Decline.

Named commercial programmes

Branded programmes (Atkins, Zone, Weight Watchers and others) have been randomized only on weight and risk factors over 12 months or less, and no hard outcome has been measured. The numbers are under Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception above.

The remaining patterns have no randomized evidence on hard outcomes at all, only cohorts, and the cohorts are uneven.

Vegetarian, paleo and low-carb patterns rest on cohorts; carnivore rests on almost nothing

No trial has randomized people to a vegetarian, paleo, low-carb or carnivore pattern and counted deaths or heart attacks. What exists is cohorts, and they are uneven: many for vegetarian diets, one for paleo, a pooled set for carbohydrate intake, none for carnivore. Each pattern below gets the same five checks in the same order: best design, risk-factor result, hard-outcome result, known harm, and what is missing.

Vegetarian and vegan

Best design: prospective cohorts, pooled in one umbrella review of 21 systematic reviews; no general-population trial on events. The 21 reviews share primary cohorts, so the count measures coverage, not 21 separate tests. (Landry et al., 2024)

Risk factors: in observational comparisons, vegans ran lower than omnivores on systolic blood pressure (-2.56 mmHg, -4.66 to -0.45, very low certainty) and LDL (-0.49 mmol/L, -0.62 to -0.36, low certainty). The randomized LDL comparison in healthy adults was null: -0.13 mmol/L (-0.37 to 0.12). (Landry et al., 2024)

Hard outcomes. In the Adventist Health Study 2 (AHS-2: 73,308 Seventh-day Adventists, 2,570 deaths, mean 5.8 years), vegetarians combined had a 12% lower death rate than non-vegetarians, HR 0.88 (0.80-0.97). Pesco-vegetarians showed the largest reduction, 0.81 (0.69-0.94); vegans 0.85 (0.73-1.01), an interval that crosses 1; semi-vegetarians 0.92 (0.75-1.13), null. In absolute terms the gap is about one death per 1000 person-years: 6.61 in non-vegetarians against 5.33-5.61 in vegan, lacto-ovo and pesco-vegetarians (semi-vegetarians 6.16). (Orlich et al., 2013) The benefit was concentrated in men; the sex difference is covered under Who gains: people at high risk, people who eat poorly, and a few groups who need specific care.

Pooled across cohorts, a vegetarian pattern is associated with lower CVD incidence, RR 0.85 (0.79-0.92), and lower CHD incidence, 0.79 (0.71-0.88), both at moderate certainty, the highest grade any arm reached. CVD mortality 0.92 (0.85-0.99) and CHD mortality 0.76 (0.68-0.85) are low certainty. Total stroke was not reduced, 0.90 (0.77-1.05), and vegans’ stroke point estimate sat above 1, 1.17 (0.69-1.99). (Landry et al., 2024) The CVD-mortality figure is fragile: a second pooling of the same cohorts gave the same 0.92 with an interval crossing 1, and the umbrella reported the significant one, judging that review higher quality. (Landry et al., 2024)

Strictness bought nothing extra in AHS-2: pesco and lacto-ovo did as well as vegan, while semi-vegetarian was null. That points the same way as the floor section (Some diets are worse by construction: they leave out nutrients or mismatch energy), where nutrient risk climbs as more food groups are cut.

Why the certainty stays low. AHS-2 vegans were leaner than non-vegetarians (BMI 24.1 vs 28.3), 98.8% drank no alcohol and 85% had never smoked, against 83.4% and 75.7% of non-vegetarians. The authors adjusted for a long list of these factors and still wrote: «Potential for uncontrolled confounding remains.» (Orlich et al., 2013) Diet was measured once, at baseline. (Orlich et al., 2013) The US 2020 dietary-patterns review also named plant-based patterns as protective (Boushey et al., 2020), but it re-pools the same Adventist cohorts and shares an author with AHS-2, so it repeats this evidence rather than adding to it.

Vegetarian names different diets. The British EPIC-Oxford cohort found no mortality advantage for vegetarians. Its vegans ate about 27 g of fibre a day; AHS-2 vegans ate 45-47 g. (Orlich et al., 2013) A mortality figure attaches to a particular vegetarian diet in a particular population, not to the word. -> Is the Food Category Doing Any Work

Known harm: strict plant-based diets carry the floor’s adequacy risks (B12 above all); see Some diets are worse by construction: they leave out nutrients or mismatch energy and Vitamin B12 Status in Vegetarian and Vegan Diets.

Funding: the umbrella review was funded partly by the Academy of Nutrition and Dietetics’ vegetarian practice group, a stake in the exposure it studied. (Landry et al., 2024) That earns the same scrutiny as an industry-funded paper: its data stand, its framing is read with care. Most arms were still graded low or very low, and the null stroke results were reported.

Not held: a general-population trial on events. Vegetarian Dietary Patterns and Mortality

Paleolithic

Best design: one cohort. REGARDS followed 21,423 US Black and White adults aged 45 and over, with the stroke belt oversampled, for a median 6.25 years (2,513 deaths). Diet was a paleo score built from one baseline food questionnaire, not a prescribed diet. The authors describe it as the first study of a paleo pattern and mortality. (Whalen et al., 2017)

Risk factors: no paleo risk-factor data held.

Hard outcomes: the highest paleo-score fifth had lower all-cause mortality than the lowest, HR 0.77 (0.67-0.89), fully adjusted. The CVD arm crossed 1: 0.78 (0.61-1.00). Much of the signal sat in two foods. Removing nuts from the score weakened the all-cause association by 15.6% (to HR 0.89), and removing red and processed meat weakened it by 14.3% (to 0.88); both intervals then touched 1. (Whalen et al., 2017) The same nuts and processed meat that the other patterns share carry most of what the paleo label shows here. -> Dietary Patterns and All-Cause Mortality

Known harm: among current smokers the paleo score tracked higher risk (interaction P = 0.04), but the stratum intervals included 1 and the authors attribute it to chance. (Whalen et al., 2017) The score penalizes dairy and grains; no adequacy data for that are held.

Not held: a second cohort, any pooled review, any trial on events. Confidence is low.

Low-carbohydrate: hard outcomes

Best design: cohorts and pooled cohort analyses, all with diet from food questionnaires. No trial has counted events. The weight and LDL trials are under Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception.

Hard outcomes. In cohorts, carbohydrate intake and death trace a U. In the ARIC cohort the lowest mortality sat at 50-55% of energy from carbohydrate. Pooled with seven other cohorts, both ends were higher than moderate intake: HR 1.20 (1.09-1.32) for low intake (<40%) and 1.23 (1.11-1.36) for high (>70%). At the low end, mortality tracked what replaced the carbohydrate. Swapping carbohydrate for animal fat or protein tracked higher mortality, HR 1.18 (1.08-1.29); swapping it for plant fat or protein tracked lower, 0.82 (0.78-0.87). (Seidelmann et al., 2018) The 18-country PURE cohort found no association between low carbohydrate intake (below 50% of energy) and health outcomes, and its authors read that as no support for very-low-carb diets either. (Dehghan et al., 2017)

A 38-study pool scored diets on a composite low-carb index (low carbohydrate, high protein, high fat, from any source). The highest-scoring group had more coronary heart disease, RR 1.43 (1.18-1.72), but on only three cohorts. «No significant association for total CVD, all-cause or cardiovascular mortality was observed.» All-cause mortality was 1.03 (0.96-1.10). (Qin et al., 2023) These pools share cohorts with one another, so they refine one body of evidence rather than confirm it twice.

What this licenses. The cohort data do not show that low-carb raises mortality, and they do not show it is safe for everyone. They point to the replacement food: animal-based low-carb tracks worse, plant-based better. -> Low-Carbohydrate vs Balanced-Carbohydrate Diets

Known harm: high-fat low-carb carries a caution for people whose LDL rises sharply on it; see Who gains: people at high risk, people who eat poorly, and a few groups who need specific care.

Not held: an event trial.

Carnivore and very-low-carb extremes

No long hard-outcome cohort and no adequacy data on an animal-foods-only diet are held, so the evidence is insufficient. The only directional signal is a risk factor borrowed from the low-carb trials, where low-carb diets did not significantly lower LDL while low-fat diets did. Among the patterns people ask about, this end of the range is the least studied. Saying so is the finding. -> Named Diet Programs Compared

Stroke: two results that do not compete

PREDIMED, the large primary-prevention event trial of a Mediterranean pattern, cut strokes; vegetarian cohorts show no stroke benefit. These are different designs testing different diets in different people. Neither result ranks one pattern above the other.

If the label does this little, the differences that do show up probably travel with something else.

Shared foods, total energy and staying on the diet carry the difference, and they overlap

When a pattern beats a typical diet, three things travel with the gain: the foods most studied patterns share, how much energy people end up eating, and whether they keep eating that way. None of the three has been measured as a share of the effect, and they are not separate levers: dropping ultra-processed food changes both the foods and the calories, and adherence scales everything. So the three can be named and taken apart, but not ranked.

Shared foods

The patterns that cohort studies link to lower all-cause mortality share a common core: vegetables, fruits, legumes, nuts, whole grains, unsaturated vegetable oils and fish, with relatively little red and processed meat, high-fat dairy, and refined carbohydrates or sweets. The 2020 US dietary-guidelines review (NESR) graded this class Strong on its own scale, from cohorts (Boushey et al., 2020). The grade attaches to the whole pattern, not to any food inside it: the review itself found the meat, dairy and refined-carbohydrate elements the least consistent across studies (Boushey et al., 2020) -> Dietary Patterns and All-Cause Mortality.

PURE’s six-food healthy-diet score shows the same thing from the other side. Adding red meat or whole grains to the score changed nothing, and the authors conclude that «a healthy diet can be achieved in a number of ways which does not necessarily require either including or excluding any specific food category.» (Mente et al., 2023) Within the pattern, the foods behave as near-substitutes. That robustness belongs to the score; it does not prove that each food acts on its own. In a cohort score, it fits equally well with the whole gradient being confounding -> Is the Food Category Doing Any Work.

A named diet can be split into two parts: what it removes and what it distinctively adds. Most popular diets remove ultra-processed food, sugary drinks and refined snacks whatever their ideology, and the evidence on intake and weight supports that removal. The signature addition (the beef of carnivore, the fat of keto) is what separates one junk-free diet from another. The evidence on that addition is often thin or runs the other way. Low-carb diets lower LDL less than low-fat, or not at all, and raise it in some people; fibre-bearing plant foods drop out with nothing in their place. Against a junk-heavy comparator the removal dominates, and the addition is credited along with it. No source measures how a real diet’s effect splits between the two, so this is a direction, not a quantity -> What a Diet Removes vs What It Adds.

The Mediterranean trial is the case where an addition held up. PREDIMED supplied olive oil or nuts and cut strokes against a low-fat-advice control that was itself not a junk-heavy diet. So at least part of the credit sits with what was added (stroke and death figures are under One trial cut strokes in high-risk adults; it did not detectably change the number of deaths above). The trial cannot separate the oil from the nuts from the rest of the diet.

Fat follows the same logic: changing the type of fat moved events where cutting total fat did not. The Hooper and WHI results are under DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome above. Saturated fat stays a contested component above the floor, never a floor criterion. WHO’s guidance on it splits by what replaces it, and its strength differs by replacement: strong for polyunsaturated fat, conditional for plant monounsaturated fat and for fibre-rich carbohydrate (World Health Organization, 2023) -> Saturated Fat Intake and Replacement. Single-food dose-response figures live in Food Groups and Health Outcomes - A Dose-Response Matrix.

Total energy

Ultra-processed food changes how much people eat even when the menu is matched. In an inpatient crossover, 20 weight-stable adults ate ultra-processed or unprocessed meals matched for presented calories, energy density, macronutrients, fibre, sugars and sodium, freely, for two weeks each. On the ultra-processed diet they ate 508 ± 106 kcal/day more and gained 0.9 kg; on the unprocessed diet they lost 0.9 kg (Hall et al., 2019). The trial is small and short, measured intake and weight rather than disease, and did not match non-beverage energy density or protein, which likely contributed to the effect (protein leverage could explain at most half) (Hall et al., 2019). It shows that processing probably drives over-eating; it does not show that processing causes heart disease -> Ultra-Processed Food and Health Outcomes.

The macronutrient split, by contrast, barely changes body fat or energy expenditure once calories are fixed. A pooled analysis of 32 controlled-feeding studies (563 people) that swapped carbohydrate for fat at equal energy found energy expenditure 26 kcal/day higher and fat loss 16 g/day greater on the lower-fat diets — the opposite of the low-carb prediction, and tiny. Hall’s reading: «In other words, for all practical purposes “a calorie is a calorie” when it comes to body fat and energy expenditure differences between controlled isocaloric diets varying in the ratio of carbohydrate to fat.» Hall leaves open other health effects of the split: «Nevertheless, it is possible that isocaloric diets differing in carbohydrate and fat may have overall health effects unrelated to total body fat or energy expenditure.» (Hall & Guo, 2017) -> What Drives Fat Gain - Energy Balance vs the Carbohydrate-Insulin Model.

Free-living trials show the same pattern. In the Cochrane low-carb review, trials that gave both arms a similar energy prescription found low-carb ahead by 0.48 kg (0.85 to 0.11; 27 trials, no heterogeneity). Trials pitting free-eating low-carb against an energy-restricted control found 1.84 kg (3.07 to 0.62; 7 trials, high heterogeneity). Both sets ran 3 to <12 months, in adults without type 2 diabetes (Naude et al., 2022). The energy subgroups overlap with the depth-of-restriction subgroups, and the review judged the spread not clinically important. The larger figure is not an artefact of the comparator either: a restricted control should bias against low-carb, so it fits an appetite effect as well -> Low-Carbohydrate vs Balanced-Carbohydrate Diets.

WHO’s total-fat guidance looks like a contradiction and is not. Reduced-fat arms in free-living non-dieting adults lost 1.42 kg (1.73 to 1.10) against usual fat intake, high certainty, and WHO attributes the effect partly to lower total energy intake (World Health Organization, 2023b). Cutting the most energy-dense macronutrient lowers weight when energy is free to fall; swapping macronutrients at fixed energy does almost nothing. One mechanism probably does most of the work (WHO leaves room for others), under two conditions.

In type 2 diabetes the format that controls intake beats any composition. Very-low-energy diets (400-500 kcal/day for 8-12 weeks) lost 6.6 kg more than a standard low-energy diet at 3 months (9.5 to 3.7), from two small trials with completer data only, so the figure is uncertain. Formula meal replacement lost 2.4 kg more (3.3 to 1.4) over 12-52 weeks, moderate certainty (Churuangsuk et al., 2021).

The review reads energy as the pathway, but that is its interpretation, not a measured result: «Planned analysis of associations between changes in ener- gy intake and weight changes from baseline, to differentiate effects of energy restriction and dietary regimen, proved impossible from the published information.» (Churuangsuk et al., 2021) The meal-replacement trials also prescribed equal energy to both arms, which points to programme structure and adherence as much as to fewer calories -> Diets for Weight Management in Type 2 Diabetes.

For type 2 diabetes remission, the clearest trial result came from a total diet replacement formula that was high in carbohydrate by share. DiRECT enrolled adults diagnosed within the past 6 years and gave the intervention arm a low-energy formula diet (825-853 kcal/day; 59% carbohydrate, 13% fat, 26% protein) for 3 months, extendable to 5 (Lean et al., 2018): «Diabetes remission was achieved in 68 (46%) participants in the intervention group and six (4%) participants in the control group (odds ratio 19·7, 95% CI 7·8–49·8; p<0·0001).» (Lean et al., 2018) At 24 months, 53 of 149 people who started the programme were in remission, against 5 of 149 controls (M. E. J. Lean et al., 2019).

No randomized trial of the low-carb route to remission is held. The diabetes umbrella review states: «No RCT has evaluated LCDs/ketogenic diets for type 2 diabetes remission.» (Churuangsuk et al., 2021) Remission followed an energy deficit, and in DiRECT it «varied with weight loss» (M. E. Lean et al., 2018). The formula was 59% carbohydrate, but at 825-853 kcal that is only about 120-125 g a day. With no randomized comparison of a low-carb route, the evidence points at energy and weight without testing the macronutrient label -> Total Diet Replacement and Type 2 Diabetes Remission.

Body weight versus what is on the plate

PREDIMED moved events with no weight-loss component: it used no calorie restriction and no exercise promotion (see One trial cut strokes in high-risk adults; it did not detectably change the number of deaths above). The largest weight-loss trial went the other way. Look AHEAD randomized 5,145 overweight or obese adults with type 2 diabetes to intensive lifestyle intervention or diabetes education. Weight loss was 8.6% vs 0.7% at one year and 6.0% vs 3.5% at the end, yet cardiovascular events ran at about 18.3 vs 19.2 per 1000 person-years, HR 0.95 (0.83 to 1.09); the trial stopped for futility at a median 9.6 years (Look AHEAD Research Group, 2013) -> Does Weight Loss Reduce Cardiovascular Events.

The weight gap between arms shrank to about 2.5 percentage points, LDL was the one risk factor the intervention did not improve, and the trial was powered (>80%) to detect an 18% difference (Look AHEAD Research Group, 2013). With both arms also receiving good diabetes care, that null is narrower than it sounds.

Look AHEAD is also one of 54 trials in a meta-analysis of weight-loss interventions in 30,206 adults with obesity, almost all of them low-fat, weight-reducing diets. The pool repeats the cardiovascular null (new cardiovascular events RR 0.93, 0.83 to 1.04, high quality) but finds fewer deaths: «For the primary outcome, high quality evidence showed that weight loss interventions decrease all cause mortality (34 trials, 685 events; risk ratio 0.82, 95% confidence interval 0.71 to 0.95), with six fewer deaths per 1000 participants (95% confidence interval two to 10).» (Ma et al., 2017)

Look AHEAD carries 54.6% of the weight in that mortality pool, and without it the result stands, RR 0.78 (0.63 to 0.96) (Ma et al., 2017). Cardiovascular events (RR 0.93, 0.83 to 1.04) and cardiovascular deaths (RR 0.93, 0.67 to 1.31, eight trials) were not detectably reduced (Ma et al., 2017). So the pool cannot show where the death benefit comes from, and the cardiovascular intervals are too wide to rule out a cardiac share. Because the trials were mostly low-fat diets, they also cannot separate weight loss from the change in diet -> Does Weight Loss Reduce Cardiovascular Events.

Set side by side, PREDIMED and Look AHEAD suggest that in PREDIMED, events moved with what people ate rather than with how much they weighed. The populations, countries and comparators differ, so this is a contrast between two trials, not a tested comparison.

Staying on the diet

Named-diet gains fade by 12 months, probably partly because adherence fades. Ge’s network meta-analysis could not measure it: most trials did not report adherence, which may have been low by 12 months. Ge reads its estimates as what average adherence buys: «Full adherence would probably yield larger effects in improvement of weight loss and cardiovascular risk factors.» (Ge et al., 2020) Ge adds that whether people outside trials would stick to these diets better or worse is speculation. So the decay probably reflects adherence at least in part. A person choosing between two diets is choosing between two sets of advice at the adherence they can sustain -> Named Diet Programs Compared.

Adherence also shows up in the trials that counted events. In WHI the low-fat arm drifted back toward usual intake (under DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome above). In PREDIMED the per-protocol estimate outruns the ITT one (under One trial cut strokes in high-risk adults; it did not detectably change the number of deaths above). Adherence is unmeasured in most of this evidence, so it is the thinnest of the three levers.

The levers do not add

DASH-Sodium is the only direct test of stacking within one trial. DASH and sodium reduction together lowered blood pressure less than the two effects added up (numbers under DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome above). So no figure in this guide sums component, energy and adherence effects into a total, and none should be built from them. The levers share parts: a diet that drops ultra-processed food changes its foods and its calories at once, and only adherence turns either into an effect.

Switching labels moves less than leaving a typical diet; at low risk both gains are small

Put the two contrasts side by side — one named diet against another, and any of them against a typical diet — and the sizes are not close. Moving off a usual diet onto almost any studied pattern shifts weight and blood pressure at 6 months by more than the pre-set importance thresholds in the largest comparison. Choosing one named pattern over another shifts them by less. Neither contrast has been measured as a difference in deaths for a low-risk person.

The two contrasts on one table

Each figure below comes from the section named beside it; the table repeats headline numbers so the two columns can be read side by side. “Not measured” means no held study ran that comparison.

Outcome	Between named diets	Pattern vs a usual or control diet
Weight	~1 kg; largest gap 1.57 kg at 6 mo, below the 2 kg bar (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception)	4-5 kg at 6 mo, ~3 kg at 12 mo (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception)
Systolic BP	largest gap 1.66 mmHg, below the 3 mmHg bar (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception)	~5 mmHg at 6 mo (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception); DASH about 3-5 mmHg (DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome)
LDL	low-carb -1.01 (interval crossing zero) vs low-fat -7.08 mg/dL vs usual diet (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception); Mediterranean -0.15 mmol/L vs another diet (One trial cut strokes in high-risk adults; it did not detectably change the number of deaths)	low-fat -7.08, moderate-macronutrient -5.22 mg/dL (Above the floor, named diets differ by about a kilogram and under 2 mmHg; LDL is the exception); Portfolio -0.73 mmol/L, hyperlipidaemic (DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome)
Events, deaths	not measured — no head-to-head event trial	PREDIMED vs low-fat advice, high risk: strokes 24 -> 14, deaths 47 -> 47 per 1000 over ~4.8 y (One trial cut strokes in high-risk adults; it did not detectably change the number of deaths); WHI low-fat vs usual diet: CHD HR 0.97, null (DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome); cohorts only otherwise (Vegetarian, paleo and low-carb patterns rest on cohorts; carnivore rests on almost nothing)

Weight, BP and LDL figures: (Ge et al., 2020). PREDIMED per-1000 rows: (Rees et al., 2019). WHI: (Howard et al., 2006). DASH range: two partly-overlapping pooled estimates, neither picked (DASH and Portfolio lower blood pressure and LDL; low-fat and MIND trials found no significant effect on their main outcome). for the side-by-side reading.

LDL is the one row where macronutrient labels split: low-carb lowered LDL least; among named diets only Mediterranean differed from usual diet. The events row holds no between-label comparison at all; Mediterranean appears there because PREDIMED tested it on events, not because it beat another pattern on them.

Absolute numbers, where the source gives them

Three event figures exist in a per-1000 form or could have; they use different horizons, so read each on its own.

• PREDIMED: in the table above.
• Adventist Health Study-2: a death-rate gap of about 1 per 1000 person-years (6.61 in non-vegetarians vs 5.33-5.61 in vegan, lacto-ovo and pesco-vegetarians; semi-vegetarians 6.16), a yearly rate from a cohort, so not the same quantity as PREDIMED’s five-year count (Orlich et al., 2013).
• PURE: for the top-vs-bottom diet-score contrast (mortality HR 0.70, 0.63-0.77), unadjusted death rates by score fifth are given, 12.14 vs 4.10 per 1000 person-years bottom vs top (Mente et al., 2023); no adjusted absolute difference is held, and none is computed here.

The same relative effect buys less at lower risk

An absolute benefit is a relative reduction applied to a person’s own baseline risk, so it shrinks as that risk falls even when the relative effect holds. GRADE puts it plainly: «Recommendations, however, may differ across subgroups of patients at different baseline risk of an outcome, despite there being a single relative risk that applies to all of them.» (Schünemann et al., n.d.) This needs no subgroup claim; it is arithmetic on prognostic data alone (route a) -> Baseline Risk and the Relative-Absolute Split.

If the relative effect carried over unchanged, a person with a quarter of PREDIMED’s stroke risk would avoid about a quarter as many strokes; whether it does carry over is untested. WHI, the large low-fat pattern trial in a mostly primary-prevention population, was null on CHD, though its thin achieved fat contrast means it cannot cleanly test the baseline-risk point (Howard et al., 2006). No held dietary-pattern trial shows a hard-outcome benefit in a genuinely low-risk population -> Cardiometabolic Interventions and Hard CV Outcomes in Low-Risk People.

The weight-loss trials held here show no proven cut in cardiovascular events either: Look AHEAD (above) was null even in adults with type 2 diabetes. Their all-cause signal is under Body weight versus what is on the plate -> Cardiometabolic Interventions and Hard CV Outcomes in Low-Risk People.

Where diet pattern sits among the levers

For a low-risk person already eating reasonably, the choice of pattern is a lever but not a large one. Smoking, by contrast, carries an all-cause hazard of about 3 for those who continue, and quitting by around 40 recovers about a decade of life (Jha et al., 2013). Inactivity and adiposity are the other levers this wiki treats as dominant -> Layer 1 - Ranking Interventions for a Stratum. These figures are in different units and do not form a ranking, but the gap in size is not subtle.

The diet-score curve points the same way. In PURE, most of the gain comes from moving people up from a poor diet, and it flattens above a middling score. The next section asks who sits where on that curve. -> Diet Quality Scores and Cardiovascular Risk

So the Mediterranean trial evidence is a bound, not a promise. It shows what a pattern change can do in high-risk adults: strokes carry the gain, and deaths did not detectably change. For someone at low risk who already eats reasonably, the remaining gain from switching or perfecting a pattern is small. That is a reason to stop optimizing, not a failure to find a better diet.

Who gains: people at high risk, people who eat poorly, and a few groups who need specific care

Where a person sits on those numbers depends on a few of their features. Two of them, baseline risk and how poorly someone eats now, change how much a better diet buys. A few others change what is safe or what needs checking. One popular idea, matching the diet to a person’s genes or insulin response, was tested in a trial and did not hold up.

High baseline risk: the absolute gain needs it

PREDIMED, the large primary-prevention event trial of a named pattern, enrolled a high-risk group. Its adults had no cardiovascular disease at entry but carried diabetes or several risk factors. Its stroke reduction is stated per 1000 at that baseline under One trial cut strokes in high-risk adults; it did not detectably change the number of deaths. The same relative effect in a lean, normotensive non-smoker would prevent far fewer strokes, and no held trial measured it there (see Switching labels moves less than leaving a typical diet; at low risk both gains are small above).

A poor current diet: most of the gain sits at the bottom of the scale

PURE scored 147,642 people in 21 countries on six protective foods (fruit, vegetables, nuts, legumes, fish, mainly whole-fat dairy), one point per food eaten above the cohort median, range 0-6. Each one-quintile step up the score was associated with 9% lower mortality, HR 0.91 (0.89-0.93), and 6% lower major cardiovascular disease, HR 0.94 (0.92-0.97) (Mente et al., 2023) (Mente et al., 2023). The paper’s headline «8% lower» mortality is a different number; it matches the pooled estimate across its four prospective cohorts, HR 0.92 (0.90-0.93). The headline’s 244,597-person total also counts two case-control studies, which give no mortality estimate (Mente et al., 2023) (Mente et al., 2023).

For the composite outcome, the curve was steeper below the global median. The results place the flattening near a score of 3; the discussion puts it at 4 of 6, which the authors describe as the level beyond which there is modest additional gain (Mente et al., 2023) (Mente et al., 2023). The knee sits mid-range, not at the edge of the studied 0-6 range.

So the person who gains most is a low consumer of these foods moving to a moderate score. Someone already at 4 or above buys little by pushing higher (Mente et al., 2023). The regions where the association with events was strongest — South Asia, Africa and China — also had the lowest mean scores (2.1, 2.6, 3.1). Their steeper gradient is mostly position on the curve, not a regional difference in how diet works.

Three limits travel with this knee. It sits on a coarse seven-point score built from median cut-offs, not on grams of any food, and that scoring itself compresses the top of the range. Food-frequency questionnaires blur gradients, so the flattening above 4 may partly be what the instrument cannot resolve. And the data are observational: the association shrank substantially on adjustment (under Nearly all pattern-and-death evidence comes from cohorts, and placebo adherence in drug trials is about as large as the crude diet signal). The knee is a located feature of one cohort’s curve, not an optimum, and it does not settle whether diet curves generally flatten (Diet Quality Scores and Cardiovascular Risk, The Underivable Optimum).

Atherogenic-lipoprotein responders on high-fat low-carb: check the lipids

Low-carb diets that are high in fat can raise LDL

The Cochrane review of low-carb versus balanced-carb diets ends with a stated caution: «In people with lipid disorders and variability with atherogenic lipoprotein response, caution in recommending low- carbohydrate and consequent high-fat diets is warranted.» (Naude et al., 2022) In DIETFITS, the low-carb arm raised LDL and the between-group change favoured low-fat by about 5%, while HDL rose and triglycerides fell more on low-carb (Gardner et al., 2018). For anyone with a lipid disorder, or whose LDL or apoB climbs on this diet, measure the response rather than assume it.

The caution matters because the harm accumulates. LDL and other apoB-carrying particles cause atherosclerotic disease. The European Atherosclerosis Society consensus states that lowering LDL particle concentration by any route reduces events in proportion to both the size of the LDL drop and how long it lasts. The condition is that the LDL change reflects a real change in particle number and carries no off-target harm (Ference et al., 2017). Read in reverse, a diet that keeps LDL raised for years adds exposure year by year (LDL ApoB and Cumulative Exposure).

In the metabolic syndrome, diabetes and high triglycerides, LDL-C and LDL particle number can come apart, because small, dense LDL carries less cholesterol per particle. In those conditions the consensus says apoB or particle number «may more accurately reflect the causal effect of LDL on ASCVD» (Ference et al., 2017). For an insulin-resistant person trying high-fat low-carb, LDL-C alone may under-read the risk, so apoB or particle number is worth checking alongside it. -> LDL ApoB and Cumulative Exposure

This is a contraindication for one group, not a verdict on low-carb for everyone. On weight, low-carb and balanced-carb diets come out about a kilogram apart, and the cohort link to death depends on what replaces the carbohydrate; both are reported in the earlier sections. Low-carb is neither shown to raise mortality nor shown safe for every lipid profile.

Vegans, pregnancy and menstruating women: the floor bites hardest

The numbers are in the floor section (Some diets are worse by construction: they leave out nutrients or mismatch energy). Vegans run the highest rates of low B12, iodine, iron, zinc and long-chain omega-3 status. Women show the iron gap most, because plant iron is absorbed less well even when intake is higher (Nutrient Adequacy of Plant-Based Diets, Vitamin B12 Status in Vegetarian and Vegan Diets).

Pregnancy is where the two iodine findings meet. One set of studies finds most vegans below the iodine adequacy cut-off, on thin data. A Cochrane review finds routine iodine supplementation in pregnancy insufficiently evidenced, with the benefit concentrated in severe deficiency (Iodine Supplementation in Pregnancy). The two answer different questions, prevalence and trial benefit, and no held source measures the vegan-and-pregnant group directly. The overlap is named here, not quantified; dosing belongs with a prescriber.

Matching the diet to genes or insulin secretion: tested, not supported

DIETFITS randomized 609 non-diabetic adults with overweight or obesity to healthy low-fat or healthy low-carb for 12 months. It pre-specified two personalization hypotheses as primary aims: a three-SNP genotype pattern, and insulin secretion measured 30 minutes after a glucose challenge. Power, estimated post hoc, was about 90%. Neither interaction was significant (diet x genotype P=.20; diet x insulin secretion P=.47). «In the context of these 2 common weight loss diet approaches, neither of the 2 hypothesized predisposing factors was helpful in identifying which diet was better for whom.» (Gardner et al., 2018)

Personalizing a diet beyond the average result needs positive evidence that the effect differs by subgroup; a plausible mechanism is not enough, and here the powered test came back empty. The null holds when both arms emphasize diet quality, in non-diabetic adults, and for this insulin measure; smaller trials using fasting insulin are not covered by it (Low-Carbohydrate vs Balanced-Carbohydrate Diets).

Sex: the vegetarian cardiovascular association was concentrated in men, in one cohort

In the Adventist Health Study 2, «Associations in men were larger and more often significant than were those in women.» (Orlich et al., 2013) A formal test found a diet-by-sex interaction for cardiovascular mortality (P=.01), but not for all-cause mortality. Men’s cardiovascular-death HR was 0.71 (0.57-0.90); women’s was 0.99 (0.83-1.18) (Orlich et al., 2013). This is a difference in the association by sex, not a baseline-risk artifact. But it is a route-(b) subgroup signal from one observational cohort, unreplicated and unexplained by intake: the authors found no striking difference between male and female vegetarians’ nutrient intakes. So it is a hypothesis, not a reason to advise men and women differently (Vegetarian Dietary Patterns and Mortality).

Age: below about 35, the evidence is too thin to grade

The US dietary guidelines review graded the pattern-and-mortality evidence for adults and older adults only; for younger populations, roughly under 35, it found the evidence insufficient (Boushey et al., 2020). That is a gap, not a null (Dietary Patterns and All-Cause Mortality).

One argument for plant-heavy patterns does not concern health at all.

Guideline bodies cite an environmental case for plant-heavy diets; it is a separate question from health

Guideline bodies now say openly that plant-heavy eating is partly an environmental recommendation. EASD’s diabetes panel writes that «plant-based and minimally processed foods are considered to have the lowest environmental impact» (Diabetes and Nutrition Study Group of the EASD, 2023). ESC advises eating less meat «From both a health and an environmental point of view», then notes that restricting red meat «may have little or no effect on major cardiometabolic outcomes», while substituting plant foods «does improve LDL-C concentrations» (European Society of Cardiology, 2021). Both bodies name the environmental reason; neither says how much it weighed against the health evidence -> Which Objective Moved This Recommendation. This guide records the direction (plant-heavy patterns carry the lower environmental load, on these bodies’ account) and stops there: it holds no environmental data, so it neither prices that load nor counts it for or against health.

PURE tested the health side directly. Scored on PURE’s own food-frequency questionnaire, the Planetary score (PURE’s scoring of the EAT-Lancet reference diet) did not track mortality: HR 1.00 (0.98-1.03) per quintile, with an AUC of 0.52, barely above chance. The other diet scores tested showed small protective associations (Mente et al., 2023).

The authors point to the Planetary score’s design: «The planetary diet score is the most restrictive and substantially limits animal foods, whereas the PURE diet score permits a moderate amount of animal foods» (Mente et al., 2023) — for example a cup of milk or yogurt and some meat daily. Dairy, and to a degree fish, carried PURE’s protective signal. It is unresolved whether that is a real weakness of the EAT-Lancet pattern or an artifact of scoring it in a population where those foods were protective. The result rests on one cohort and one questionnaire -> Diet Quality Scores and Cardiovascular Risk.

The Planetary null does not contradict the vegetarian cohorts’ CVD benefit (see Vegetarian, paleo and low-carb patterns rest on cohorts; carnivore rests on almost nothing). The two use different scores — a per-quintile contrast against a vegetarian-versus-non-vegetarian one — and a planetary diet is not a vegetarian diet. Both stand; neither ranks the other.

The separation runs both ways. The environmental case cannot be borrowed as health support; the floor section (Some diets are worse by construction: they leave out nutrients or mismatch energy) flags that move in a plant-diet adequacy review’s framing sentence. And a null health score does not weaken the environmental case, which is a separate question.

Put together, the decision comes down to a few steps.

What to do with this

The steps run in order: clear the floor, then pick a diet you will keep, then spend effort only where the evidence moved events.

1. Check the floor first. If you eat mostly or wholly plant-based, cover B12 (vegans: treat it as close to non-optional), and look at iodine, iron and long-chain omega-3. If you eat mostly meat, look at fibre, folate and magnesium. Either way, look at vitamin D, and eat enough energy and not chronically too much. -> Nutrient Adequacy of Plant-Based Diets, Vitamin B12 Status in Vegetarian and Vegan Diets

2. Above the floor, choose the pattern you will stick with. The label is a weak lever: named diets differ by about a kilogram and under 2 mmHg (LDL is the exception), and early gains fade within a year. -> Named Diet Programs Compared

3. Make the shared moves. More vegetables, fruit, pulses, nuts, whole grains, unsaturated oils and fish; less processed meat, refined starch and ultra-processed food. These are what the better-evidenced patterns have in common. No study has measured their share, and they overlap, so do not add their effects up.

4. At high cardiovascular risk, the Mediterranean pattern with olive oil or nuts is the one with a positive event trial. It probably cut strokes; it did not detectably change the number of deaths, and the trial supplied the food. -> Mediterranean Diet and Cardiovascular Events

5. On a high-fat low-carb diet, check your LDL or apoB response. Of the three macronutrient patterns, only low-carb failed to lower LDL significantly against a usual diet (-1.01 mg/dL, interval crossing zero; low-fat -7.08). In DIETFITS, LDL rose on low-carb. The caution applies most to people with an atherogenic lipoprotein response. -> LDL ApoB and Cumulative Exposure

6. At low risk and already eating reasonably, the remaining gains are small — a reason to stop optimizing. -> Baseline Risk and the Relative-Absolute Split

Several questions the evidence cannot yet answer are worth naming.

What the evidence cannot answer yet

Between patterns. No trial has compared two named patterns on deaths or events, and none is likely: a decades-long head-to-head diet trial cannot realistically be run. Two workarounds remain: emulate that trial from cohort data with its assumptions stated, or compare components rather than labels -> Named Diet Programs Compared. No review has pooled a pattern-to-mortality effect size either — NESR graded its cohort body without pooling it — so no absolute death figure for a healthy pattern exists. And the main pattern-to-mortality reviews held here (NESR, Dinu, Landry) re-pool overlapping primary studies (PREDIMED, AHS-2, the Nurses’ and Health Professionals’ cohorts), so they are not independent of one another -> Dietary Patterns and All-Cause Mortality.

Per pattern. The gaps are uneven, and this guide does not guess their direction.

• DASH and Portfolio: no event trial for either. The Portfolio trial under way measures carotid plaque volume, not events -> Portfolio Dietary Pattern and LDL Cholesterol.
• Mediterranean: advice without supplied food has never been tested on events -> Mediterranean Diet and Cardiovascular Events.
• Paleo: one cohort.
• Carnivore and very-low-carb extremes: no long hard-outcome cohorts and no nutrient-adequacy data.
• General population: no pattern trial on mortality in reasonably healthy people.

Across all patterns. Adults under about 35 are not covered. No study measures how a pattern shapes the course of decline — whether it shortens the years spent ill or only lengthens them — because the outcomes counted are deaths and events. And no source splits a named diet’s effect into what it removes (often ultra-processed food and sugary drinks) and what it adds -> What a Diet Removes vs What It Adds.

This guide has not been checked against what happened to people who followed it. It grades whether its claims are coherent and faithful to their sources, not whether following it made anyone better off. The loop between the advice and what happened is open.

Evidence box

Question	’Do the major named dietary patterns differ in their effect on patient-important outcomes, and if so which axis carries the difference — the pattern label itself, the shared components common to most evidence-based patterns, total energy/adiposity, or adherence — how large is the difference, how certain, and where is the evidence RCT-grade rather than confounded cohort? Or does the evidence not distinguish the patterns at all?‘
Evidence included	45 sources — 24 gold, 18 high, 1 moderate, 2 method
Overall certainty	Medium (see Rating Certainty of Evidence)
Source-selection note	1 source(s) below the gold/high evidence bar feed this page: Dehghan (cohort, moderate). Their roles and limitations are described in the text.
Last updated	2026-10-07 · Independently reviewed: No · Full edit history

References

Barnes, L. L., Dhana, K., Liu, X., Carey, V. J., Ventrelle, J., Johnson, K., Hollings, C. S., Bishop, L., Laranjo, N., Stubbs, B. J., Reilly, X., Agarwal, P., Zhang, S., Grodstein, F., Tangney, C. C., Holland, T. M., Aggarwal, N. T., Arfanakis, K., Morris, M. C., & Sacks, F. M. (2023). Trial of the MIND Diet for Prevention of Cognitive Decline in Older Persons. New England Journal of Medicine, 389(7), 602–611. https://doi.org/10.1056/nejmoa2302368

Bougma, K., Aboud, F., Harding, K., & Marquis, G. (2013). Iodine and Mental Development of Children 5 Years Old and Under: A Systematic Review and Meta-Analysis. Nutrients, 5(4), 1384–1416. https://doi.org/10.3390/nu5041384

Boushey, C., Ard, J., Bazzano, L., Heymsfield, S., Mayer-Davis, E., Sabaté, J., Snetselaar, L., Van Horn, L., Schneeman, B., English, L., Bates, M., Callahan, E., Venkatramanan, S., Butera, G., Terry, N., & Obbagy, J. (2020). Dietary Patterns and All-Cause Mortality: A Systematic Review. https://doi.org/10.52570/nesr.dgac2020.sr0108

Chiavaroli, L., Nishi, S. K., Khan, T. A., Braunstein, C. R., Glenn, A. J., Mejia, S. B., Rahelić, D., Kahleová, H., Salas-Salvadó, J., Jenkins, D. J. A., Kendall, C. W. C., & Sievenpiper, J. L. (2018). Portfolio Dietary Pattern and Cardiovascular Disease: A Systematic Review and Meta-analysis of Controlled Trials. Progress in Cardiovascular Diseases, 61(1), 43–53. https://doi.org/10.1016/j.pcad.2018.05.004

Chiavaroli, L., Viguiliouk, E., Nishi, S., Blanco Mejia, S., Rahelić, D., Kahleová, H., Salas-Salvadó, J., Kendall, C., & Sievenpiper, J. (2019). DASH Dietary Pattern and Cardiometabolic Outcomes: An Umbrella Review of Systematic Reviews and Meta-Analyses. Nutrients, 11(2), 338. https://doi.org/10.3390/nu11020338

Churuangsuk, C., Hall, J., Reynolds, A., Griffin, S. J., Combet, E., & Lean, M. E. J. (2021). Diets for weight management in adults with type 2 diabetes: an umbrella review of published meta-analyses and systematic review of trials of diets for diabetes remission. Diabetologia, 65(1), 14–36. https://doi.org/10.1007/s00125-021-05577-2

Dehghan, M., Mente, A., Zhang, X., Swaminathan, S., Li, W., Mohan, V., Iqbal, R., Kumar, R., Wentzel-Viljoen, E., Rosengren, A., Amma, L. I., Avezum, A., Chifamba, J., Diaz, R., Khatib, R., Lear, S., Lopez-Jaramillo, P., Liu, X., Gupta, R., … Mapanga, R. (2017). Associations of fats and carbohydrate intake with cardiovascular disease and mortality in 18 countries from five continents (PURE): a prospective cohort study. The Lancet, 390(10107), 2050–2062. https://doi.org/10.1016/s0140-6736(17)32252-3

Diabetes and Nutrition Study Group of the EASD. (2023). Evidence-based European recommendations for the dietary management of diabetes. Diabetologia, 66(6), 965–985. https://doi.org/10.1007/s00125-023-05894-8

Dinu, M., Pagliai, G., Casini, A., & Sofi, F. (2017). Mediterranean diet and multiple health outcomes: an umbrella review of meta-analyses of observational studies and randomised trials. European Journal of Clinical Nutrition, 72(1), 30–43. https://doi.org/10.1038/ejcn.2017.58

Estruch, R., Ros, E., Salas-Salvadó, J., Covas, M.-I., Corella, D., Arós, F., Gómez-Gracia, E., Ruiz-Gutiérrez, V., Fiol, M., Lapetra, J., Lamuela-Raventos, R. M., Serra-Majem, L., Pintó, X., Basora, J., Muñoz, M. A., Sorlí, J. V., Martínez, J. A., Fitó, M., Gea, A., … Martínez-González, M. A. (2018). Primary Prevention of Cardiovascular Disease with a Mediterranean Diet Supplemented with Extra-Virgin Olive Oil or Nuts. New England Journal of Medicine, 378(25). https://doi.org/10.1056/nejmoa1800389

European Society of Cardiology. (2021). 2021 ESC Guidelines on cardiovascular disease prevention in clinical practice. European Heart Journal, 42(34), 3227–3337. https://doi.org/10.1093/eurheartj/ehab484

Ference, B. A., Ginsberg, H. N., Graham, I., Ray, K. K., Packard, C. J., Bruckert, E., Hegele, R. A., Krauss, R. M., Raal, F. J., Schunkert, H., Watts, G. F., Borén, J., Fazio, S., Horton, J. D., Masana, L., Nicholls, S. J., Nordestgaard, B. G., van de Sluis, B., Taskinen, M.-R., … Catapano, A. L. (2017). Low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies. A consensus statement from the European Atherosclerosis Society Consensus Panel. European Heart Journal, 38(32), 2459–2472. https://doi.org/10.1093/eurheartj/ehx144

Filippou, C. D., Tsioufis, C. P., Thomopoulos, C. G., Mihas, C. C., Dimitriadis, K. S., Sotiropoulou, L. I., Chrysochoou, C. A., Nihoyannopoulos, P. I., & Tousoulis, D. M. (2020). Dietary Approaches to Stop Hypertension (DASH) Diet and Blood Pressure Reduction in Adults with and without Hypertension: A Systematic Review and Meta-Analysis of Randomized Controlled Trials. Advances in Nutrition, 11(5), 1150–1160. https://doi.org/10.1093/advances/nmaa041

Gardner, C. D., Trepanowski, J. F., Del Gobbo, L. C., Hauser, M. E., Rigdon, J., Ioannidis, J. P. A., Desai, M., & King, A. C. (2018). Effect of Low-Fat vs Low-Carbohydrate Diet on 12-Month Weight Loss in Overweight Adults and the Association With Genotype Pattern or Insulin Secretion: The DIETFITS Randomized Clinical Trial. JAMA, 319(7), 667. https://doi.org/10.1001/jama.2018.0245

Ge, L., Sadeghirad, B., Ball, G. D. C., da Costa, B. R., Hitchcock, C. L., Svendrovski, A., Kiflen, R., Quadri, K., Kwon, H. Y., Karamouzian, M., Adams-Webber, T., Ahmed, W., Damanhoury, S., Zeraatkar, D., Nikolakopoulou, A., Tsuyuki, R. T., Tian, J., Yang, K., Guyatt, G. H., & Johnston, B. C. (2020). Comparison of dietary macronutrient patterns of 14 popular named dietary programmes for weight and cardiovascular risk factor reduction in adults: systematic review and network meta-analysis of randomised trials. BMJ, m696. https://doi.org/10.1136/bmj.m696

Hall, K. D., Ayuketah, A., Brychta, R., Cai, H., Cassimatis, T., Chen, K. Y., Chung, S. T., Costa, E., Courville, A., Darcey, V., Fletcher, L. A., Forde, C. G., Gharib, A. M., Guo, J., Howard, R., Joseph, P. V., McGehee, S., Ouwerkerk, R., Raisinger, K., … Zhou, M. (2019). Ultra-Processed Diets Cause Excess Calorie Intake and Weight Gain: An Inpatient Randomized Controlled Trial of Ad Libitum Food Intake. Cell Metabolism, 30(1), 67-77.e3. https://doi.org/10.1016/j.cmet.2019.05.008

Hall, K. D., & Guo, J. (2017). Obesity Energetics: Body Weight Regulation and the Effects of Diet Composition. Gastroenterology, 152(7), 1718-1727.e3. https://doi.org/10.1053/j.gastro.2017.01.052

Harding, K. B., Peña-Rosas, J. P., Webster, A. C., Yap, C. M., Payne, B. A., Ota, E., & De-Regil, L. M. (2017). Iodine supplementation for women during the preconception, pregnancy and postpartum period. Cochrane Database of Systematic Reviews, 2017(3). https://doi.org/10.1002/14651858.cd011761.pub2

Hooper, L., Summerbell, C. D., Thompson, R., Sills, D., Roberts, F. G., Moore, H. J., & Davey Smith, G. (2012). Reduced or modified dietary fat for preventing cardiovascular disease. Cochrane Database of Systematic Reviews, 2017(11). https://doi.org/10.1002/14651858.cd002137.pub3

Howard, B. V., Van Horn, L., Hsia, J., Manson, J. E., Stefanick, M. L., Wassertheil-Smoller, S., Kuller, L. H., LaCroix, A. Z., Langer, R. D., Lasser, N. L., Lewis, C. E., Limacher, M. C., Margolis, K. L., Mysiw, W. J., Ockene, J. K., Parker, L. M., Perri, M. G., Phillips, L., Prentice, R. L., … Kotchen, J. M. (2006). Low-Fat Dietary Pattern and Risk of Cardiovascular Disease: The Women’s Health Initiative Randomized Controlled Dietary Modification Trial. JAMA, 295(6), 655. https://doi.org/10.1001/jama.295.6.655

Huang, L., Tao, Y., Chen, H., Chen, X., Shen, J., Zhao, C., Xu, X., He, M., Zhu, D., Zhang, R., Yang, M., Zheng, Y., & Yuan, C. (2023). Mediterranean-Dietary Approaches to Stop Hypertension Intervention for Neurodegenerative Delay (MIND) Diet and Cognitive Function and its Decline: A Prospective Study and Meta-analysis of Cohort Studies. The American Journal of Clinical Nutrition, 118(1), 174–182. https://doi.org/10.1016/j.ajcnut.2023.04.025

Jha, P., Ramasundarahettige, C., Landsman, V., Rostron, B., Thun, M., Anderson, R. N., McAfee, T., & Peto, R. (2013). 21st-Century Hazards of Smoking and Benefits of Cessation in the United States. New England Journal of Medicine, 368(4), 341–350. https://doi.org/10.1056/nejmsa1211128

Landry, M. J., Senkus, K. E., Mangels, A. R., Guest, N. S., Pawlak, R., Raj, S., Handu, D., & Rozga, M. (2024). Vegetarian dietary patterns and cardiovascular risk factors and disease prevention: An umbrella review of systematic reviews. American Journal of Preventive Cardiology, 20, 100868. https://doi.org/10.1016/j.ajpc.2024.100868

Lean, M. E. J., Leslie, W. S., Barnes, A. C., Brosnahan, N., Thom, G., McCombie, L., Peters, C., Zhyzhneuskaya, S., Al-Mrabeh, A., Hollingsworth, K. G., Rodrigues, A. M., Rehackova, L., Adamson, A. J., Sniehotta, F. F., Mathers, J. C., Ross, H. M., McIlvenna, Y., Welsh, P., Kean, S., … Taylor, R. (2019). Durability of a primary care-led weight-management intervention for remission of type 2 diabetes: 2-year results of the DiRECT open-label, cluster-randomised trial. The Lancet Diabetes &amp; Endocrinology, 7(5), 344–355. https://doi.org/10.1016/s2213-8587(19)30068-3

Lean, M. E., Leslie, W. S., Barnes, A. C., Brosnahan, N., Thom, G., McCombie, L., Peters, C., Zhyzhneuskaya, S., Al-Mrabeh, A., Hollingsworth, K. G., Rodrigues, A. M., Rehackova, L., Adamson, A. J., Sniehotta, F. F., Mathers, J. C., Ross, H. M., McIlvenna, Y., Stefanetti, R., Trenell, M., … Taylor, R. (2018). Primary care-led weight management for remission of type 2 diabetes (DiRECT): an open-label, cluster-randomised trial. The Lancet, 391(10120), 541–551. https://doi.org/10.1016/s0140-6736(17)33102-1

Look AHEAD Research Group. (2013). Cardiovascular Effects of Intensive Lifestyle Intervention in Type 2 Diabetes (N Engl J Med 2013;369(2):145-154). New England Journal of Medicine, 369(2), 145–154. https://doi.org/10.1056/nejmoa1212914

Ma, C., Avenell, A., Bolland, M., Hudson, J., Stewart, F., Robertson, C., Sharma, P., Fraser, C., & MacLennan, G. (2017). Effects of weight loss interventions for adults who are obese on mortality, cardiovascular disease, and cancer: systematic review and meta-analysis. BMJ, j4849. https://doi.org/10.1136/bmj.j4849

Mente, A., Dehghan, M., Rangarajan, S., O’Donnell, M., Hu, W., Dagenais, G., Wielgosz, A., A. Lear, S., Wei, L., Diaz, R., Avezum, A., Lopez-Jaramillo, P., Lanas, F., Swaminathan, S., Kaur, M., Vijayakumar, K., Mohan, V., Gupta, R., Szuba, A., … Yusuf, S. (2023). Diet, cardiovascular disease, and mortality in 80 countries. European Heart Journal, 44(28), 2560–2579. https://doi.org/10.1093/eurheartj/ehad269

National Institute for Health and Care Excellence. (n.d.). Weight management: Evidence review F – effectiveness of different diets in achieving and maintaining weight loss. https://www.nice.org.uk/guidance/ng246/evidence

Naude, C. E., Brand, A., Schoonees, A., Nguyen, K. A., Chaplin, M., & Volmink, J. (2022). Low-carbohydrate versus balanced-carbohydrate diets for reducing weight and cardiovascular risk. Cochrane Database of Systematic Reviews, 2022(1). https://doi.org/10.1002/14651858.cd013334.pub2

Neufingerl, N., & Eilander, A. (2021). Nutrient Intake and Status in Adults Consuming Plant-Based Diets Compared to Meat-Eaters: A Systematic Review. Nutrients, 14(1), 29. https://doi.org/10.3390/nu14010029

Orlich, M. J., Singh, P. N., Sabaté, J., Jaceldo-Siegl, K., Fan, J., Knutsen, S., Beeson, W. L., & Fraser, G. E. (2013). Vegetarian Dietary Patterns and Mortality in Adventist Health Study 2. JAMA Internal Medicine, 173(13), 1230. https://doi.org/10.1001/jamainternmed.2013.6473

Pawlak, R., Parrott, S. J., Raj, S., Cullum-Dugan, D., & Lucus, D. (2013). How prevalent is vitamin B₁₂deficiency among vegetarians? Nutrition Reviews, 71(2), 110–117. https://doi.org/10.1111/nure.12001

Qin, P., Suo, X., Chen, S., Huang, C., Wen, W., Lin, X., Hu, D., & Bo, Y. (2023). Low-carbohydrate diet and risk of cardiovascular disease, cardiovascular and all-cause mortality: a systematic review and meta-analysis of cohort studies. Food &amp; Function, 14(19), 8678–8691. https://doi.org/10.1039/d3fo01374j

Rees, K., Takeda, A., Martin, N., Ellis, L., Wijesekara, D., Vepa, A., Das, A., Hartley, L., & Stranges, S. (2019). Mediterranean-style diet for the primary and secondary prevention of cardiovascular disease. Cochrane Database of Systematic Reviews, 2019(3). https://doi.org/10.1002/14651858.cd009825.pub3

Sacks, F. M., Svetkey, L. P., Vollmer, W. M., Appel, L. J., Bray, G. A., Harsha, D., Obarzanek, E., Conlin, P. R., Miller, E. R., Simons-Morton, D. G., Karanja, N., Lin, P.-H., Aickin, M., Most-Windhauser, M. M., Moore, T. J., Proschan, M. A., & Cutler, J. A. (2001). Effects on Blood Pressure of Reduced Dietary Sodium and the Dietary Approaches to Stop Hypertension (DASH) Diet. New England Journal of Medicine, 344(1), 3–10. https://doi.org/10.1056/nejm200101043440101

Schünemann, H., Brożek, J., Guyatt, G., & Oxman, A. (n.d.). GRADE Handbook: for grading quality of evidence and strength of recommendations. https://gradepro.org/handbook/

Scientific Advisory Committee on Nutrition. (2015). Carbohydrates and Health. https://www.gov.uk/government/publications/sacn-carbohydrates-and-health-report

Seidelmann, S. B., Claggett, B., Cheng, S., Henglin, M., Shah, A., Steffen, L. M., Folsom, A. R., Rimm, E. B., Willett, W. C., & Solomon, S. D. (2018). Dietary carbohydrate intake and mortality: a prospective cohort study and meta-analysis. The Lancet Public Health, 3(9), e419–e428. https://doi.org/10.1016/s2468-2667(18)30135-x

Siervo, M., Lara, J., Chowdhury, S., Ashor, A., Oggioni, C., & Mathers, J. C. (2014). Effects of the Dietary Approach to Stop Hypertension (DASH) diet on cardiovascular risk factors: a systematic review and meta-analysis. British Journal of Nutrition, 113(1), 1–15. https://doi.org/10.1017/s0007114514003341

Simpson, S. H., Eurich, D. T., Majumdar, S. R., Padwal, R. S., Tsuyuki, R. T., Varney, J., & Johnson, J. A. (2006). A meta-analysis of the association between adherence to drug therapy and mortality. BMJ, 333(7557), 15. https://doi.org/10.1136/bmj.38875.675486.55

Whalen, K. A., Judd, S., McCullough, M. L., Flanders, W. D., Hartman, T. J., & Bostick, R. M. (2017). Paleolithic and Mediterranean Diet Pattern Scores Are Inversely Associated with All-Cause and Cause-Specific Mortality in Adults. The Journal of Nutrition, 147(4), 612–620. https://doi.org/10.3945/jn.116.241919

Willett, W. (2012). Nutritional Epidemiology. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199754038.001.0001

World Health Organization. (2023a). Saturated fatty acid and trans-fatty acid intake for adults and children: WHO guideline. https://www.who.int/publications/i/item/9789240073630

World Health Organization. (2023b). Total fat intake for the prevention of unhealthy weight gain in adults and children: WHO guideline.