The Best Diet for Longevity, and the Meal Plan Problem

The dietary intervention with the best randomised evidence for helping people live longer is not the Mediterranean diet, not intermittent fasting, and not any macronutrient ratio. It is swapping table salt for a potassium-enriched substitute. That trial enrolled nearly 21,000 people and found a 12% lower rate of death from any cause.

Almost nobody talks about it. Meanwhile the topics that dominate nutrition discussion — carbs versus fat, protein targets, eating windows, ideal meal timing — have either weak evidence, evidence pointing at something other than what people think, or no mortality evidence at all.

This article works through what has actually been tested, what each macronutrient does under scrutiny, and then attempts the daily and weekly plan you probably came here for. That last part needs a warning attached, so it comes with one.

What has actually been randomised

Nutrition is dominated by observational research: measure what people say they eat, wait, count deaths. That design cannot separate the diet from the person eating it. People who eat more vegetables also smoke less, earn more, and exercise more.

So it is worth being strict about which dietary interventions have been randomised and what they measured.

Trial	People	Intervention	What it measured	Result
SSaSS 2021	20,995	Potassium-enriched salt substitute	Stroke, death	Death 0.88 (0.82-0.95)
PREDIMED 2018	7,447	Mediterranean diet + olive oil or nuts	Cardiovascular events	0.69 and 0.72
Cochrane 2020	56,675	Reducing saturated fat	Events, death	Events 0.83; death 0.96 (null)
DASH 1997	459	Fruit, veg, low-fat dairy pattern	Blood pressure	−5.5/−3.0 mm Hg
POUNDS LOST 2009	811	Four macronutrient ratios	Weight at 2 years	No difference
Hall 2019	20	Ultra-processed vs unprocessed	Calories eaten	+508 kcal/day
Liu 2022	139	8-hour eating window	Weight at 1 year	No difference
Generation 100 (2020) — exercise, for comparison	1,567	Supervised training twice weekly for 5 years, against already-active controls following activity guidelines	All-cause mortality	No significant difference

Look at the fourth column. Blood pressure, calories, and body weight are surrogates — plausible stand-ins for health, not health itself. Only two trials here measured hard clinical events, and only one found fewer deaths.

How to read those numbers

The Result column mixes two kinds of number, and they are not comparable.

Ratios — 0.88, 0.69, 0.83, 0.96. These divide the rate of events in the treated group by the rate in the control group. 1.0 means no effect. Below 1.0 means fewer events, above means more. So 0.88 is a 12% lower rate, and 0.96 is a 4% lower rate that, as it happens, is not distinguishable from no effect at all.

Absolute differences — −5.5/−3.0 mm Hg, +508 kcal/day. These are measured in real units, and here 0 means no effect. Nothing in the column signals the switch, so it is worth naming: 0.88 and −5.5 are different species of number.

The bracketed range is the 95% confidence interval — the span of values compatible with the data. When it crosses 1.0, as saturated fat’s 0.90-1.03 does, the result cannot be distinguished from no effect. SSaSS’s 0.82-0.95 does not cross, so the true effect is plausibly as large as 18% or as small as 5%, but almost certainly real.

And a ratio hides how big the effect actually is. This is where most health writing misleads, including when the number is accurate. SSaSS’s death rate fell from 44.61 to 39.28 per 1,000 person-years. That is the same fact stated three ways:

a 12% lower rate of death — the relative version, and the one that gets headlines
5.3 fewer deaths per 1,000 people per year — the absolute version
roughly 190 person-years of using the salt substitute per death avoided

All three are true. They feel nothing alike. Whenever you meet a percentage reduction with no absolute figure beside it, the absolute figure is the one being withheld.

One caveat specific to that first row, since this article leads with it: in SSaSS stroke was the primary endpoint, and death from any cause was secondary. Secondary outcomes carry a higher risk of turning up positive by chance, so they warrant more caution than a primary result. Here the death finding moves in step with both the primary stroke result (0.86) and major cardiovascular events (0.87), which is what makes it credible. It is still a secondary outcome, and that belongs in any honest description of it.

The last row is not a diet trial, and it is there as a reference point. Generation 100 randomised 1,567 Norwegian adults aged 70-77 to five years of supervised moderate or high-intensity training, against a control group told to follow national activity guidelines. All-cause mortality was the primary endpoint — which almost no lifestyle trial can afford to attempt. Five-year mortality was 4.5% in the combined training groups and 4.7% in controls: no significant difference. The high-intensity arm showed a non-significant trend (HR 0.63, 0.33-1.20).

That is emphatically not a finding that exercise does not work, and the reason is the control group. Those people were not sitting still. They were told to follow national activity guidelines, 80% already reported medium or high physical activity at baseline, and they ended up doing more of their activity at high intensity than the moderate-training arm did. The trial compared structured supervised training against already-active older adults training on their own. It was never exercise versus no exercise. Add only 72 deaths across five years, and there was very little contrast left for it to detect.

The exercise-versus-nothing trial does not exist and never will, because randomising people to years of inactivity is neither ethical nor practical. That question is answered observationally, and there the signal is enormous. Arem’s 2015 pooled analysis followed 661,137 people for a median 14.2 years and recorded 116,686 deaths. Measured against people doing no leisure-time activity at all, mortality ran 20% lower in those who exercised but fell short of the recommended minimum, 31% lower at one to two times the minimum, and 39% lower at three to five times, where the benefit plateaus. That evidence is observational and carries a real reverse-causality problem — ill people exercise less, which flatters the active groups — but the effect is consistent across cohorts and far too large to dismiss.

So the two results answer different questions. Whether to exercise has a large and consistent answer. How much and at what intensity, once you are already active, is where Generation 100 looked and found nothing separable. That distinction is the point this table exists to make: when a lifestyle intervention is randomised against a genuinely fair comparator and followed all the way to death rather than to a biomarker, clean answers are rare. Diet is not uniquely short of good evidence. Almost everything is.

The salt result

The Salt Substitute and Stroke Study randomised 600 villages in rural China to either regular salt or a substitute of 75% sodium chloride and 25% potassium chloride. Over 4.74 years:

Stroke: rate ratio 0.86 (95% CI 0.77-0.96)
Major cardiovascular events: 0.87 (0.80-0.94)
Death from any cause: 0.88 (0.82-0.95)

In absolute terms, deaths fell from 44.61 to 39.28 per 1,000 person-years — about 5.3 fewer deaths per 1,000 people per year. Serious adverse events from high potassium did not increase significantly.

Now the limits, because they are large. Participants were high risk: 72.6% had already had a stroke and 88.4% had hypertension. They were consuming about 12.2 g of salt per day, far above typical intake in most countries. And their sodium came from home cooking, so changing the salt jar changed nearly everything. In a country where most sodium arrives pre-added inside packaged food, swapping your table salt accomplishes much less.

One safety point that matters more than the result. Potassium-enriched salt is genuinely dangerous for some people — those with chronic kidney disease, or taking potassium-sparing diuretics, ACE inhibitors, or angiotensin receptor blockers. The trial excluded people at high risk of high potassium, so its clean safety record does not transfer to them. This is a talk-to-your-doctor item, not a buy-it-on-the-way-home item.

The Mediterranean diet, and what it actually showed

PREDIMED is the most cited diet trial in existence and one of the most overstated. It assigned 7,447 Spanish adults aged 55-80 at high cardiovascular risk to a Mediterranean diet supplemented with either extra-virgin olive oil (at least 4 tablespoons daily) or 30 g of mixed nuts daily, against advice to eat a low-fat diet. No calorie restriction.

Major cardiovascular events occurred in 3.8% of the olive oil group, 3.4% of the nut group, and 4.4% of controls — hazard ratios of 0.69 and 0.72. That is a relative reduction of about 30% and an absolute reduction of roughly 0.6 to 1.0 percentage points over 4.8 years.

Four things usually go missing:

It did not measure longevity. The endpoint was heart attack, stroke, or cardiovascular death combined. Not all-cause mortality.

It did not really test the whole Mediterranean pattern. Most participants were already eating close to it. Compared with controls, the intervention groups increased fish by 0.3 weekly servings and legumes by 0.4. The real difference between groups was the free olive oil and nuts.

It was stopped early on a planned interim analysis, which tends to make effects look bigger than they are.

The two foods that worked were donated by the industries that sell them — olive oil from Hojiblanca and Patrimonio Comunal Olivarero, walnuts from the California Walnut Commission, almonds from Borges, hazelnuts from Morella Nuts. The authors state that no sponsor influenced the design, analysis, or reporting, and there is no reason to think otherwise. It still belongs in the citation.

Processing seems to matter on its own

Hall’s 2019 trial is tiny — 20 people — but unusually well controlled. Participants lived on a metabolic ward for 28 days and ate either ultra-processed or unprocessed food for two weeks each, in random order, eating as much as they wanted. Crucially, the two diets were matched for calories presented, energy density, macronutrients, sugar, sodium, and fibre.

People ate 508 ± 106 kcal more per day on the ultra-processed diet. They gained 0.9 kg on it and lost 0.9 kg on the unprocessed one. They also ate faster.

Matching all those nutrients did not cancel the effect, which suggests something about the food itself — texture, structure, how fast it goes down — drives overeating. Twenty people over two weeks measuring calories is not evidence about mortality. But it is the cleanest available demonstration that processing is not just a proxy for bad nutrients.

The macronutrients, one at a time

Before the individual verdicts, the finding that should reframe the whole question.

POUNDS LOST randomised 811 adults to four diets spanning fat from 20% to 40%, protein from 15% to 25%, and carbohydrate from 35% to 65% of calories, all calorie-reduced, for two years. Weight loss at two years: 3.0 versus 3.6 kg for protein, 3.3 versus 3.3 kg for fat, 2.9 versus 3.4 kg for carbohydrate. Every comparison had P > 0.20. Hunger and satisfaction were also the same.

One variable did predict weight loss: attendance at counselling sessions, worth 0.2 kg per session attended. In a trial built to find macronutrient effects, the thing that mattered was showing up.

Carbohydrate

The largest analysis, Seidelmann’s 2018 study, followed 15,428 US adults for a median of 25 years, then pooled them with seven other cohorts for 432,179 people and 40,181 deaths. Mortality was lowest at 50-55% of calories from carbohydrate, and rose at both extremes: hazard ratio 1.20 below 40%, and 1.23 above 70%.

But the substitution analysis is the finding worth remembering:

Carbohydrate replaced by animal fat or protein: 1.18 (1.08-1.29)
Carbohydrate replaced by plant fat or protein: 0.82 (0.78-0.87)

Two people eating identical carbohydrate percentages had opposite risk profiles depending on what filled the gap. The ratio was not the variable. The food was.

On carbohydrate quality, the WHO commissioned Reynolds’s 2019 review, pooling 185 prospective studies and 58 trials. Highest versus lowest fibre intake came with 15-30% lower all-cause mortality, cardiovascular mortality, coronary heart disease, stroke, type 2 diabetes, and colorectal cancer. Benefit was greatest at 25-29 g of fibre per day, and probably continues above it. Most adults eat under 20 g.

The same review found something inconvenient for a large popular literature: when carbohydrate quality was measured by glycaemic index or glycaemic load instead of fibre and whole grains, the protective signal mostly disappeared, and the evidence was graded low to very low certainty.

Fibre findings for mortality remain observational, and people who eat lots of fibre differ in many other ways. The randomised evidence covers weight, blood pressure, and cholesterol only.

Fat

Here is the most honest sentence available about dietary fat, from the 2020 Cochrane review of 15 trials and 56,675 people: reducing saturated fat cut cardiovascular events by 17% (RR 0.83, 95% CI 0.70-0.98) and had little or no effect on dying.

All-cause mortality: 0.96 (0.90-1.03). Cardiovascular mortality: 0.95 (0.80-1.12).

Both halves get dropped, depending on who is talking. People arguing saturated fat is deadly cite the event reduction and omit the mortality null. People arguing saturated fat is harmless cite the mortality null and omit the event reduction. The trials support both statements simultaneously.

In practical terms the event reduction is real but modest: 56 people need to reduce saturated fat intake for about four years for one to avoid a cardiovascular event. And the mortality confidence interval, 0.90 to 1.03, does not rule out a small benefit — “no effect” overstates the evidence too.

A dose-response did appear: the more saturated fat came down, and the more serum cholesterol fell with it, the greater the reduction in events. Replacing it with polyunsaturated fat or with carbohydrate worked about equally well. Evidence on replacing it with monounsaturated fat or protein was too thin to judge.

Worth noting what PREDIMED implies about total fat: its control group was the low-fat arm, and it did worse than two energy-unrestricted, generously fatty Mediterranean diets. Total fat is not the useful axis.

Protein

Protein is where the evidence is genuinely contradictory, and where you should be most suspicious of confident advice.

Two large meta-analyses disagree about the direction of the association between total protein intake and all-cause mortality:

	Participants	Total protein	Plant protein
Naghshi 2020	715,128	0.94 (0.89-0.99) — lower risk	0.92 (0.87-0.97)
Chen 2020	350,452	1.05 (1.01-1.10) — higher risk	0.93 (0.87-0.99)

They agree on plant protein and disagree on total protein. These should not be averaged into a comfortable middle; the disagreement is the finding. Naghshi estimated that an extra 3% of daily energy from plant protein was associated with 5% lower mortality.

Chen’s analysis adds a distinction that undercuts the phrase “plant protein” altogether. Protein from grains and potatoes showed no association with mortality. Protein from legumes, nuts, vegetables, and fruits was associated with lower mortality. Lumping them together hides the effect.

Both are observational, both report effects in the 1.02-1.12 range that confounding can plausibly manufacture, and plant protein intake is a reliable marker of being wealthier, more educated, and less likely to smoke.

The gap nobody bridges: neither analysis measured muscle mass, strength, or physical function. Higher protein intake is widely recommended for older adults precisely to protect against muscle loss — an outcome these studies never recorded. There is no reason to assume the protein intake that minimises mortality associations in a cohort is the same number that best preserves the ability to stand up from a chair at 80. Anyone quoting a single protein target for older adults is filling that gap with assumption.

On protein sources, the IARC’s 2015 evaluation reviewed over 800 studies and classified processed meat as carcinogenic to humans (Group 1), with an 18% higher colorectal cancer risk per 50 g eaten daily, and red meat as probably carcinogenic (Group 2A), at 17% per 100 g daily if the association is causal — which IARC did not conclude.

One clarification does a lot of work here. Group 1 describes how strong the evidence is, not how big the risk is. Press coverage putting bacon in the same bracket as cigarettes confused those two things badly. IARC’s own position was that individual risk “remains small” and the significance is population-level, given how many people eat processed meat. Positive associations turned up in only about half the individual cohorts.

Meal timing

Time-restricted eating usually gets studied in a way that confuses timing with eating less, because shrinking your eating window tends to shrink your intake.

Liu’s 2022 trial separated them. 139 adults with obesity were randomised to an 8-hour window (08:00 to 16:00) plus calorie restriction, or the same calorie restriction with no time limit, for 12 months. Both groups got identical calorie targets, identical macronutrient targets, and identical counselling.

Weight loss: −8.0 kg with time restriction, −6.3 kg without. Net difference −1.8 kg, confidence interval −4.0 to 0.4, P = 0.11. Body fat, waist, blood pressure, and metabolic markers showed no meaningful differences either.

With calories matched, the clock added nothing detectable. Two caveats pull in opposite directions: participants already ate within about 10 hours 23 minutes at baseline, so the trial only really tested a two-hour narrowing; and a confidence interval reaching −4.0 kg does not exclude a real modest benefit.

If a shorter eating window helps you eat less, it is a useful tool. The evidence does not support it as a metabolic intervention independent of intake.

Now the meal plan, with a warning

Here is what I could not find while researching this article: a single trial that randomised people to a specific daily or weekly meal schedule and measured whether they lived longer. It does not exist.

What the evidence supports lives at three levels — dietary patterns, specific food swaps, and nutrient targets with dose-response data. It says nothing about how many meals to eat, when to eat them, whether breakfast matters, how to split macronutrients across a day, or what to rotate week to week.

So a meal plan is not an evidence-based object. It is an implementation device: one arbitrary arrangement of food that happens to hit targets which do have evidence behind them. Its value is that structure helps people follow through — and following through is the variable that keeps outperforming composition in actual trials. Any meal plan presented as scientifically optimal is claiming a precision that the underlying research cannot deliver.

With that said, you asked for a plan, and refusing to give one would be its own kind of dishonesty. Here is the reasoning made explicit.

The targets worth hitting

Each of these traces to something measured, with the strength of that measurement noted:

Target	Where it comes from	Strength
25-29 g fibre daily, more is likely better	Reynolds 2019	Meta-analysis; observational for mortality
~30 g mixed nuts daily	PREDIMED nut arm dose	Randomised, cardiovascular events
~4 tbsp extra-virgin olive oil daily	PREDIMED olive oil arm dose	Randomised, cardiovascular events
8-10 servings fruit and vegetables daily	DASH combination diet	Randomised, blood pressure
Less sodium; potassium-enriched salt if safe for you	SSaSS	Randomised, all-cause mortality
Minimise processed meat	IARC 2015	Hazard evaluation, observational
Mostly minimally processed food	Hall 2019	Small randomised trial, intake
Protein weighted toward legumes and nuts	Naghshi 2020, Chen 2020	Observational, internally contradictory
Roughly half your calories from carbohydrate, plant-sourced swaps	Seidelmann 2018	Observational

Everything more specific than this table is convention, not evidence.

One day that hits them

Calorie level is the part you have to set yourself, based on your body size, activity, and whether you are trying to lose, maintain, or gain. The arrangement below lands around 2,000-2,200 kcal.

Morning. Rolled oats cooked with milk or plain yogurt, 15 g walnuts, a large handful of berries. Roughly 8-10 g of fibre before you have done anything else, which is where most people’s fibre deficit is won or lost.

Midday. A large salad built on a cup of cooked lentils or chickpeas, with mixed vegetables, dressed with 2 tablespoons of olive oil and vinegar, plus wholegrain bread. Another 12-15 g of fibre.

Afternoon. The other 15 g of nuts and a piece of fruit.

Evening. Fish, or a bean and vegetable dish, with two servings of vegetables, a wholegrain or a potato, and the remaining 2 tablespoons of olive oil used in cooking.

That reaches 25-30 g of fibre, the PREDIMED nut and olive oil doses, and 8-10 servings of produce, without needing anything weighed precisely.

A word on the olive oil, since 4 tablespoons is around 450 calories. PREDIMED was deliberately not calorie-restricted and did not observe weight gain, but if you are eating in a deficit, that oil has to fit inside it rather than sit on top.

One week

The weekly layer exists mainly to keep variety and to place the foods that belong in a week rather than a day:

Fish: two or three meals.
Legumes: most days, in some form. This is the single highest-leverage habit in the whole plan, since it moves fibre and plant protein at once.
Poultry or eggs: a few meals.
Red meat: one modest portion, or none. The IARC dose-response is per 100 g daily, so occasional intake sits well below the exposures in that analysis.
Processed meat: treat as an occasional exception rather than a staple.
Nuts and olive oil: every day, not weekly.
Salt: cook with less, and consider a potassium-enriched substitute if your kidney function and medications allow it — confirm that first.

Notice what is absent: meal counts, eating windows, fasting days, macro splits per meal, and rules about eating after dark. Those were omitted because nothing supports them, not because they were forgotten.

What this does not cover

This article is about generally healthy adults trying to reduce long-term risk. It is not adequate for pregnancy or breastfeeding, chronic kidney disease, diabetes managed with insulin or sulfonylureas, heart failure, eating disorder history, diagnosed food allergies, or people on medications that interact with potassium or vitamin K. Several of the recommendations above change or reverse in those situations, and the potassium point could cause real harm.

Everything here also inherits the ceiling of its evidence. The mortality claims rest largely on observational data with confounding that cannot be adjusted away. The randomised trials mostly measured surrogates, ran for a few years at most, and studied particular populations — high-risk Spaniards, rural Chinese villagers, overweight Americans, twenty people on a metabolic ward.

The defensible summary

If you want the shortest honest version:

Eat mostly minimally processed food, built around vegetables, legumes, whole grains, nuts, and olive oil, with fish more often than red meat, processed meat rarely, and less salt. Get 25-30 g of fibre. Stop optimising ratios; the trials that varied them found nothing. Stop optimising the clock; the trial that controlled for calories found nothing.

Then find an arrangement of that you can actually keep doing, because across trial after trial, adherence beat every specification the researchers were testing.

The strongest single change with mortality evidence behind it is the least glamorous one in this entire article: use less salt, and consider replacing what you do use with a potassium-enriched substitute — after checking that it is safe for you.