Food and the Brain: What the Evidence Actually Says

That 45% figure is real, it's from the world's most rigorous standing review of dementia prevention evidence, and it's the single most misused statistic in this entire strand when it gets picked up by anyone less careful than the Commission that produced it. So before anything else, here's exactly what it means and what it doesn't.

What "potentially preventable" actually means

That is a population-attributable-fraction estimate. In plain terms: if every one of the 14 risk factors below were removed from an entire population, across everyone's lifetime, the Commission's modelling suggests dementia cases would fall by roughly 45% across that population. It is not a claim that any individual person who addresses these factors has a 45% chance of avoiding dementia, and it is not a claim that dementia is "curable" or that existing cognitive decline can be undone. It's a statement about where the modifiable risk sits, in aggregate, across a lifetime, at the level of a whole population.

The 14 factors the Commission names are: low education, hearing loss, hypertension, smoking, obesity, depression, physical inactivity, diabetes, excessive alcohol consumption, traumatic brain injury, air pollution, social isolation, untreated vision loss (added in the 2024 update), and high LDL cholesterol (also added in 2024).

Diet is not, itself, on that list. This matters, and it would be dishonest to pretend otherwise. What is on the list — and this is the actual chain of reasoning, not a workaround — is obesity, diabetes, hypertension and high LDL cholesterol, four conditions where diet is one of the best-established levers available to a person managing their own risk.

What the largest cohort studies actually found

Three major observational studies looked specifically at ultra-processed food and brain health outcomes. All three are large, all three are peer-reviewed, and all three are exactly what they are — associations from cohort data, not experiments, and every one of them is confounded to some degree by income, education and lifestyle factors that also affect dementia risk. Here's what each one actually found, with the real numbers.

It's worth pausing on what makes these three studies collectively more persuasive than any one of them alone. They come from three different research groups, working with three different cohorts, on two different continents — a US cohort of over-55s, a Brazilian cohort followed into cognitive testing, and the long-running Framingham Heart Study in the United States. They used different measures of outcome: incident dementia diagnosis in one case, rate of cognitive test decline in another, incident Alzheimer's specifically in the third. And they still point in the same direction. That kind of convergence across independent groups and independent methods is exactly the pattern epidemiologists look for before treating an association as worth taking seriously — it's also, importantly, still not the same thing as a randomised trial, and none of these three studies claim to be one.

The Framingham finding adds a genuinely interesting wrinkle rather than simply repeating the other two: the association only showed up clearly in participants under 68 at baseline, not in the older group. That's not a weakness in the data to explain away — it's a real finding, and one honest reading is that diet-related risk may compound earlier in life, well before any diagnosis, rather than acting as a late switch that flips back on its own. It's also a reminder that "ultra-processed food is associated with dementia risk" is not a uniform finding across every age group studied; the real evidence is more textured than a single headline number.

Read across all three and the pattern is consistent — higher ultra-processed food intake tracks with worse cognitive outcomes in every cohort that looked — but consistency across observational studies is not the same thing as proof. None of these studies randomised anyone's diet. All three are vulnerable to the same confounders: people who eat less ultra-processed food also tend, on average, to have higher incomes, more years of education, and other lifestyle differences that independently affect brain health. The studies' authors know this, adjust for what they can measure, and cannot adjust for what they can't. That caveat isn't a footnote here — it's the correct way to read every number above.

A named hypothesis, correctly labelled

"Type 3 diabetes" is a genuinely live research question, worth knowing about, and it is not a settled diagnosis. Both things are true at once. Any piece of writing that presents it as an established medical fact is overstating the evidence; any piece that dismisses it outright is ignoring a real and ongoing line of research. This strand does neither — it names the hypothesis, names where it stands with the bodies that would recognise it if the evidence supported doing so, and stops there.

What we do not know — and two claims worth correcting outright

Popular writing about diet and dementia tends to accumulate two claims that sound compelling and don't survive contact with the evidence. Both circulate widely enough that correcting them directly, rather than quietly avoiding them, builds more trust than staying silent would.

Claim: "Dementia is not found in populations eating a natural diet." This is false, and it's worth being direct about that. The Indianapolis–Ibadan Dementia Project, a long-running comparative study, found an annual dementia incidence of 1.35% among Yoruba communities in Nigeria, against 3.24% among African-Americans in Indianapolis studied by the same research team. That's a real and interesting difference, and diet is one of several plausible contributing factors researchers have examined. But 1.35% is not zero. No human population studied anywhere has been documented with zero dementia. The honest version of this finding is "meaningfully lower incidence in one population than another, for reasons that likely include diet among other factors" — not "immune."

Claim: "Brain damage can be undone within 3–12 months." There is no credible evidence for this, and this strand deliberately avoids language that implies existing brain damage can be undone, for a reason. The claim traces largely to the Bredesen protocol, which rests on a small number of uncontrolled case series that, per published clinical critique, lack a proper methods section, lack defined inclusion criteria, and report no data on people who didn't respond to the intervention. Contrast that with the FINGER trial, a properly randomised controlled study, which found that a structured multidomain lifestyle intervention — diet, exercise, cognitive training, vascular risk management — helped at-risk older adults improve or maintain their cognitive performance over the trial period. That is a genuinely strong result. It is a finding about prevention and maintenance, not about undoing existing brain damage, and the two are not the same claim. This strand uses "preventable" throughout, and nothing stronger, because the evidence doesn't support anything stronger.

The steel-man

Every study cited above is observational, and that is a real limitation, not a technicality to be waved past. Income, education, physical activity, social connection and general health-seeking behaviour all correlate with both diet quality and dementia risk, and no amount of statistical adjustment fully separates them. Someone who can afford to buy less ultra-processed food is also, on average, more likely to have had more years of education, better access to healthcare, and lower lifetime stress — any of which could independently lower dementia risk regardless of what's on their plate. Nobody has run, or could ethically run, a randomised trial that assigns people a diet for forty years and waits to see who develops dementia. What we have is the best evidence that kind of question can realistically produce, and it points in a consistent direction across three large, independent cohorts — which is meaningful, but it is association, not proof, and every claim above has said so for a reason.

It's also worth stating plainly that dementia has always existed, everywhere, including in populations eating diets nothing like the modern British one — the Yoruba data above makes that point on its own. Diet is one modifiable factor among fourteen the Lancet Commission names, not a single master cause, and treating it as the whole story would misrepresent the Commission's own conclusions as badly as ignoring diet altogether would.

The reader's takeaway

There's no dietary rule that guarantees anything about dementia risk, and this piece makes no such promise. What the evidence supports is narrower and still genuinely useful: shifting the balance of what you eat away from ultra-processed food and toward unprocessed or minimally processed food is associated, in the largest cohort studies available, with lower dementia risk and slower cognitive decline. That's not a guarantee, and not a claim that any specific ingredient this strand has named is dangerous. It's a direction, backed by the best evidence we have, stated exactly as strongly as that evidence justifies and no further.

For the legal and economic mechanisms behind why ultra-processed food is often the cheaper option on a UK shelf, see The Five Percent Rule and The Filled Dairy Loop. For the practical version of this strand's advice, see How to Read a Food Label. Start at The Adulteration Files for the full picture and the evidence-grading system this piece uses throughout.