Mostly plants, even after the blood test comes back
A Swedish cohort asked whether diet still matters once the blood says Alzheimer-type change has begun. It does — and the method matters as much as the answer.
Michael Pollan opened In Defense of Food with seven words — “Eat food. Not too much. Mostly plants.”[6] — and eighteen years on, the most interesting dementia paper of the winter has quietly extended his sentence: even after your blood says the pathology has started.
A Swedish group published a cohort study in JAMA Network Open in June that I keep returning to, partly for the finding and partly for the method[1]. They followed 1,865 Stockholm residents aged sixty and over, none with dementia at the start, for up to fifteen years. At baseline they drew blood and measured three markers that have changed the field in the last five years: p-tau217, the closest thing we have to an Alzheimer-specific blood test; neurofilament light, a marker of neuronal injury; and GFAP, a marker of astrocyte activation. Separately, they scored everyone’s diet three times over six years against three patterns — Mediterranean, the Harvard Alternative Healthy Eating Index, and an anti-inflammatory index. Then they waited. Two hundred and forty people developed dementia.
The question that matters is not whether healthy eaters get less dementia; we have known that from observational data since Scarmeas showed it in New York in 2009[4]. The question is whether diet still helps once the biology is already under way. It did. Among participants whose p-tau217 was already elevated, each standard-deviation improvement in the anti-inflammatory diet score was associated with a 29 per cent lower hazard of dementia, and reductions of about a fifth to a quarter were seen for the other two markers. Converted into something a patient can feel: over ten years, the healthier eaters in the high-p-tau217 group kept roughly eight to eleven more months dementia-free than the poorer eaters. That is not a cure. It is most of a year of recognising your grandchildren, and I would take it.
The sceptic’s turn
Because the record will be tested later. This is observational, diet was self-reported, and the high-biomarker group was older, sicker and less educated — residual confounding is a certainty, not a possibility. The paper’s neatest storyline, that the anti-inflammatory pattern helps the already-affected while the Mediterranean pattern helps the not-yet-affected, rests on interaction tests that mostly fail the authors’ own multiple-comparison threshold; I would not build a clinic on it. And the cohort is affluent, university-educated, urban Stockholm. The populations carrying the heaviest dementia burden, including Aboriginal and Torres Strait Islander communities, are simply not in this evidence base, and every “precision prevention” claim built on it inherits that gap.
Why it still matters
Because of what the accompanying commentary by Charisis and Scarmeas argues about trial design[2]. Randomised diet trials for cognition have mostly been null, and their explanation is plausible: the trials enrolled healthy people with little pathology to buffer, then looked for a signal in two or three years. Blood biomarkers now let us phenotype who is actually on the Alzheimer continuum while they are still well. That addresses the reverse-causality problem in observational work — pathology begins twenty years before symptoms and may itself change what people eat — and it lets the next generation of trials enrol the people most likely to benefit. The same authors’ Nature Reviews Neurology roadmap says as much[3], and a 2026 Nature Medicine paper showing a plasma p-tau217 “clock” can estimate symptom onset to within three to four years makes the enrolment strategy concrete[5].
A blood biomarker is not an imaging finding. What imaging measures well is the cardiometabolic substrate that every one of these diets acts on.
This is also where healthspan medicine has to keep its discipline. A blood biomarker is not an imaging finding, and nothing I do on a scanner detects p-tau217. What imaging does measure well is the cardiometabolic substrate — visceral fat, muscle, coronary calcium, liver fat — that the Healthy Eating Index was designed around and that runs through every one of these dietary patterns. The defensible offer to a patient is not “we can stop Alzheimer’s”; it is “we can see the risk you can change, and the evidence says changing it still counts even late.”
Eat food. Not too much. Mostly plants. Even after the blood test comes back.
References
- [1] Mrhar A, Carballo-Casla A, Grande G, et al. Diet quality and dementia risk in older adults with Alzheimer pathology. JAMA Netw Open. 2026;9(6):e2620254. doi:10.1001/jamanetworkopen.2026.20254
- [2] Charisis S, Scarmeas N. Diet and dementia risk in individuals with prevalent neuropathology (invited commentary). JAMA Netw Open. 2026;9(6):e2620261. doi:10.1001/jamanetworkopen.2026.20261
- [3] Charisis S, Yannakoulia M, Scarmeas N. Diets to promote healthy brain ageing. Nat Rev Neurol. 2025;21(1):5–16. doi:10.1038/s41582-024-01036-9
- [4] Scarmeas N, Luchsinger JA, Schupf N, et al. Physical activity, diet, and risk of Alzheimer disease. JAMA. 2009;302(6):627–637. doi:10.1001/jama.2009.1144
- [5] Petersen KK, Milà-Alomà M, Li Y, et al. Predicting onset of symptomatic Alzheimer’s disease with plasma p-tau217 clocks. Nat Med. 2026;32(3):1085–1094. doi:10.1038/s41591-026-04206-y
- [6] Pollan M. In Defense of Food: An Eater’s Manifesto. Penguin Press, 2008 (opening line).
Sources verified via PubMed and the publisher's site, 5 September 2026.
Connect with me on LinkedIn · Instagram · drlisasorger.com · lisa@drlisasorger.com
All views expressed here are my own personal opinions and are not medical advice. General information only — not clinical or financial advice. Discuss diet, memory concerns or any biomarker testing with your own doctor. myradiologist.ai · ABN 29 692 758 115 · ACN 692 758 115
← Back to writing