Mediterranean, vegetarian, vegan or keto: which diet protects health?

Mediterranean, vegetarian or vegan? Few health questions are argued so passionately. Look closely, though, and you find less a winner than a shared foundation: every eating pattern with good data relies on vegetables, legumes, whole grains, nuts and plant oils, and every one cuts back on processed meat. This article separates what trials have shown from what rests on observation, and where plant-based diets leave gaps. It also looks at the ketogenic diet and at the much-cited Blue Zones.

Why nutrition research is so hard

Most of what we know comes from observational cohorts: tens of thousands of people fill in food-frequency questionnaires, which depend on memory and self-report, and are followed for years. Then there is confounding: vegetarians tend to smoke less, exercise more and drink less alcohol, so a lower heart risk may reflect their lifestyle rather than their plate. Long randomised trials with hard endpoints such as heart attack or death are expensive and rare. See reading health news and absolute vs relative risk for why "30% lower risk" means little without the starting number.

The Mediterranean diet: hard endpoints from randomised trials

PREDIMED: retraction and republication

The Spanish PREDIMED trial assigned 7,447 people aged 55 to 80 at high cardiovascular risk, but without known heart disease, to a Mediterranean diet with extra-virgin olive oil, one with nuts, or advice to eat less fat. The original 2013 report in the New England Journal of Medicine was retracted in 2018 because of protocol deviations. Household members had been enrolled into the same group without randomisation, and at one of eleven sites participants were assigned without randomisation. The authors reanalysed the data, also excluding 1,588 participants with known or suspected deviations, and republished.

After a median of 4.8 years, major cardiovascular events (heart attack, stroke or cardiovascular death) occurred at 8.1 (olive oil) and 8.0 (nuts) per 1,000 person-years, compared with 11.2 in the control group. In relative terms that is about 30% fewer events (hazard ratios 0.69 and 0.72); in absolute terms, roughly three fewer events per 1,000 people per year. Stroke accounted for most of the difference; overall mortality showed no clear difference.

After a heart attack: Lyon and CORDIOPREV

Two trials studied people with existing heart disease (secondary prevention). In the French Lyon Diet Heart Study, after a first heart attack, the Mediterranean group had 14 cardiac deaths or non-fatal heart attacks over a mean of 46 months, against 44 in the group on a "prudent" Western diet. The Spanish CORDIOPREV trial (Lancet 2022) compared a Mediterranean diet with a low-fat diet in 1,002 patients with coronary heart disease over seven years: 87 versus 111 major cardiovascular events, or 28.1 versus 37.7 per 1,000 person-years. The benefit appeared in men; among the 175 women there was no difference. It was a single-centre trial, but points the same way.

Vegetarian and vegan: consistent signals, mostly observational

No large trial of vegetarian or vegan diets has used heart attack or death as its endpoint; the data come from cohorts:

• EPIC-Oxford (48,188 people, 18 years): vegetarians had less coronary heart disease (10 fewer cases per 1,000 people over 10 years) but slightly more stroke (3 more per 1,000 over 10 years), mostly haemorrhagic stroke (bleeding in the brain).
• Bones: in the same cohort, vegans had more than twice the risk of hip fracture of meat eaters (hazard ratio 2.31), or 14.9 more cases per 1,000 people over 10 years.
• Adventist Health Study-2 (73,308 Seventh-day Adventists, US): vegetarians as a group had 12% lower mortality (hazard ratio 0.88), more clearly in men than in women. In an already health-conscious population, that is a signal, not proof.
• Type 2 diabetes: a meta-analysis of nine cohorts with 307,099 people found a 23% lower risk with strongly plant-based eating, and 30% lower with "healthy" plant-based patterns (fruit, vegetables, whole grains, legumes, nuts).

In a meta-analysis of 30 trials, vegetarian or vegan diets lowered LDL cholesterol by about 10% and apolipoprotein B by 14%. In a trial of 22 pairs of identical twins (JAMA Network Open 2023), LDL was 13.9 mg/dL lower after eight weeks on a vegan diet than in the twin eating an omnivorous diet. That is a real but limited effect; our article on statins in primary prevention works with a reduction of about 30% for statin therapy.

Nutrients to watch

The German Nutrition Society (Deutsche Gesellschaft für Ernährung, DGE) updated its position on vegan diets in 2024. For the healthy adult general population, it says a vegan diet can be a health-promoting diet, provided that a vitamin B12 supplement is taken and the other potentially critical nutrients are adequately supplied, if necessary also through supplements – one of the few clear cases for supplements (see supplements: what the evidence says). Those named include the omega-3 fatty acids EPA and DHA, calcium, zinc, iodine and, in women, iron. Vitamin D often needs supplementing whatever one eats (see vitamin D deficiency). Protein should come from varied sources such as legumes, whole grains, nuts and soy.

Keto and low-carb: slightly faster at first, not better in the long run

A ketogenic ("keto") diet cuts carbohydrates, typically to 20 to 50 g a day, so far that the liver turns fat into ketone bodies, which the body burns instead of glucose: a state called ketosis. Moderate low-carb diets are less strict and need not induce ketosis.

Weight: a meta-analysis of 13 long-term randomised trials found 0.91 kg more weight loss on a very-low-carbohydrate ketogenic diet than on a low-fat diet, with slightly higher LDL cholesterol. In the US DIETFITS trial (609 overweight adults), a healthy low-carb and a healthy low-fat diet did similarly well after twelve months (−6.0 versus −5.3 kg, not statistically significant); by then, the low-carb group still got about 30% of its energy from carbohydrates.

Diabetes: in the Keto-Med crossover trial, 40 adults with prediabetes or type 2 diabetes ate a well-formulated ketogenic diet and a Mediterranean diet for twelve weeks each. HbA1c fell similarly on both (−9% versus −7%). On keto, LDL cholesterol rose by 10% (versus −5%), and intake of fibre, folate, vitamin C and magnesium was lower; afterwards, participants drifted towards the Mediterranean pattern. The American Diabetes Association's Standards of Care 2026 (ADA, US) say carbohydrate reduction can be considered for some adults with diabetes, but that the effect on HbA1c is mainly short-term. The German Diabetes Society (DDG) lists low-carb among suitable ways to lose weight in its 2021 nutrition recommendations.

LDL: some people, mostly lean and active, see very large LDL rises on carbohydrate restriction; researchers call them "lean mass hyper-responders". Whether that is harmless is unresolved: the only plaque study in this group has no control group, and its journal issued an expression of concern about it in 2026.

Medical use: in children with drug-resistant epilepsy, the ketogenic diet is used as a specialist therapy. A 2020 Cochrane review (13 trials, 932 participants) found more seizure freedom and far more children with at least halved seizures (risk ratio 5.80) than with usual care, though the trials were small and unblinded and the certainty of evidence low.

Blue Zones: a good story on shaky data

The term "blue zone" was coined in 2004 by the demographer Michel Poulain and the physician Giovanni Pes for a mountainous area of Sardinia with a striking number of very old people. The journalist Dan Buettner popularised it through National Geographic and added Okinawa (Japan), Loma Linda (California, with its Seventh-day Adventist community), Nicoya (Costa Rica) and Ikaria (Greece), where, he says, people reach 100 at about ten times the US rate. His "Power 9" shared habits include everyday movement, a sense of purpose, eating until about 80% full, a plant-heavy diet built on beans with meat about five times a month, moderate wine, and close family, community and faith ties.

The critique: the demographer Saul Newman (University College London) found that records of extreme age in the US, Italy, England, France and Japan cluster where record-keeping is poor, poverty high and life expectancy short. In the US, cases of people over 110 fell sharply once states introduced universal birth certificates; the Sardinian blue zone province of Ogliastra ranked second-lowest in Italy for survival to age 55. In 2010, a review in Japan turned up more than 230,000 registered centenarians who could not be found, and many Okinawan records were reportedly destroyed in the Second World War. Newman's work, a preprint without peer review, won the satirical Ig Nobel Prize for demography in 2024.

Blue Zones researchers counter that ages were checked against several independent sources (civil and church registers, genealogies, military and electoral lists) and that the zones rest on whole populations surviving to 90 and beyond, not on a few record-holders; they concede that Okinawa's advantage has faded as lifestyles modernised (Austad and Pes, The Gerontologist 2025). "Blue Zones" is also a brand, with books and health programmes in more than 50 US communities; a US hospital group acquired the company in 2020.

The takeaway: the habits attributed to the Blue Zones (mostly plants and legumes, daily movement, close social ties) are supported by independent evidence (see loneliness as a health risk). But the regions are no proof of extraordinary longevity, and their diets cannot be copied as a formula; see separating longevity science from marketing and reading health news.

The common ground, and a trap

The DGE's 2024 food-based guidelines are already strongly plant-based: just over three-quarters plant foods, five portions of vegetables and fruit daily, legumes at least weekly, a handful of nuts daily, whole grains, and at most 300 g of meat and sausage a week. The EAT-Lancet Commission's "planetary health diet" (2019, updated October 2025) is similar, with 0 to 200 g of red meat a week.

In 2015 the International Agency for Research on Cancer (IARC) classified processed meat as carcinogenic (Group 1) and red meat as probably carcinogenic (Group 2A); each 50 g of processed meat eaten daily raised colorectal cancer risk by 18%. The grouping reflects certainty of evidence, not size of risk.

The trap: "vegan" does not automatically mean healthy. In an inpatient randomised trial (Cell Metabolism 2019), 20 adults spent two weeks each on ultra-processed and unprocessed diets matched for calories, sugar, salt and fibre. On the ultra-processed diet they ate 508 kcal more per day and gained 0.9 kg; on the unprocessed one they lost 0.9 kg. Vegan sausages, crisps and soft drinks match none of the patterns that do well in studies. Wine, incidentally, is not a required part of the Mediterranean diet; on "healthy" drinking see the moderate drinking myth.

What this means for you

• Moving toward a Mediterranean pattern: olive oil as your main fat, vegetables and a handful of nuts every day, legumes several times a week, whole grains instead of white flour, fish instead of sausage. Diet is also central to the metabolic syndrome.
• If you are vegan: take a vitamin B12 supplement; this is non-negotiable. Discuss with your doctor whether checks of B12 status, iron, iodine or vitamin D make sense, and include calcium-rich foods for your bones.
• Pregnancy, breastfeeding, children, old age: a vegan diet only with qualified nutrition counselling.
• If you are considering keto: if you take diabetes medication, especially an SGLT2 inhibitor, only after talking to your doctor; have your LDL cholesterol checked. Keto largely excludes legumes, fruit and whole grains, the very foods behind the eating patterns with the best data.
• When shopping: look at how processed a food is, not just at a "plant-based" label. Fibre also feeds the gut microbiome.

This article is general information and not individual medical advice. Decisions about tests and treatment belong in a conversation with your own physician.