At Raised Tau, a Low-Inflammation Diet Tracked 29% Less Dementia

Somewhere tonight a woman in her sixties is standing in her kitchen after visiting her mother in a memory unit, reading about blood tests that can now estimate, years ahead of any symptom, whether the same proteins are accumulating in her own brain. The question she is turning over is not whether vegetables are good for her. It is narrower and colder than that: if the biology has already started, does anything she puts on a plate still change where it goes.

That is the question a new cohort analysis in JAMA Network Open was built around, and it is what makes the study more interesting than the headline it generated. Most diet-and-dementia research asks whether eating well lowers risk across a general population. This one asked whether it still does among people whose blood already shows the markers of Alzheimer disease pathology.

The analysis draws on the Swedish National Study on Aging and Care in Kungsholmen, a population-based cohort that enrolled adults aged 60 and older in a central Stockholm district between March 2001 and August 2004 and examined them up to six times, with follow-up running to somewhere between February 2016 and November 2019. This analysis included 1,865 participants who were free of dementia at baseline. Their mean age at entry was 70.5 years, with a standard deviation of 9.3, and 1,125 of them, or 60.3 percent, were women.

Diet was not measured once and assumed to hold. Adherence to three dietary patterns was tracked repeatedly over six years: the Alternate Mediterranean Diet, the Alternative Healthy Eating Index, and the reversed Empirical Dietary Inflammatory Index, a score on which a higher value means a diet with lower inflammatory potential. Against those the researchers set baseline serum concentrations of three blood markers: phosphorylated tau at threonine 217, a relatively specific signal of Alzheimer disease pathology; neurofilament light chain, which rises with neuronal injury of many kinds; and glial fibrillary acidic protein, which reflects the reaction of the brain's support cells.

The primary outcome was all-cause dementia, identified through clinical diagnosis, medical records and death certificates, with Alzheimer-related dementia as a secondary outcome. Risk was modelled with adjusted Cox regression, and the authors also estimated ten-year dementia probabilities and restricted mean time lost to dementia. Over a mean follow-up of 8.4 years, ranging from under a month to 15.9 years, 240 participants developed dementia.

Among participants whose biomarkers were already elevated, the anti-inflammatory pattern was the one that held. Each one-standard-deviation increase in adherence to the reversed Empirical Dietary Inflammatory Index was associated with a hazard ratio of 0.71 (95% CI, 0.58 to 0.88) among those with elevated p-tau217; 0.79 (95% CI, 0.66 to 0.95) among those with elevated neurofilament light chain; and 0.73 (95% CI, 0.60 to 0.89) among those with elevated glial fibrillary acidic protein. Findings for Alzheimer-related dementia were similar.

The other two patterns behaved differently, and this is the part the coverage compressed. Associations between the Alternate Mediterranean Diet or the Alternative Healthy Eating Index and lower dementia risk were generally found only among participants whose biomarker levels were lower, meaning among those whose brains showed less evidence that the process had begun. Read closely, the study does not report that three healthy diets all protect people with Alzheimer pathology. It reports that one of the three tracked lower risk in that group, in this cohort, and that the two more familiar ones did not.

Set that beside the trial evidence and a real tension appears. In 2015, a prospective study of 923 adults aged 58 to 98, followed an average of 4.5 years, reported that people in the highest tertile of MIND diet scores had a hazard ratio of 0.47 (95% CI, 0.26 to 0.76) for incident Alzheimer's disease, and those in the second tertile 0.65 (95% CI, 0.44 to 0.98) — rates 53 and 35 percent lower than the bottom tertile. Numbers of that size are why the MIND diet entered ordinary conversation.

Then it was tested directly. A two-site randomized trial published in the New England Journal of Medicine in 2023 enrolled 604 cognitively unimpaired adults who had a family history of dementia, a body-mass index above 25 and a suboptimal diet, assigning 301 to the MIND diet and 303 to a control diet, both with mild caloric restriction, for three years. Global cognition improved in both arms, by 0.205 standardized units in the MIND group and 0.170 in the control group. The difference between them was 0.035 standardized units, with a confidence interval running from -0.022 to 0.092 and a P value of 0.23. Changes in white-matter hyperintensities, hippocampal volume and total gray- and white-matter volume were similar in the two groups.

A trial that finds nothing is not the same as a trial that finds there is nothing there. Both arms received counselling, weight-loss support and mild caloric restriction, which makes the comparison one against an actively improved diet rather than against ordinary eating. Three years is short against a disease process that develops over decades. And participants were recruited on the basis of a suboptimal baseline diet and then coached, so the contrast a cohort study observes across decades of settled habit is not the contrast a three-year trial can manufacture.

What the Swedish analysis adds is not a larger number but a better-specified question. If part of what diet does runs through inflammatory processes rather than through the amyloid-and-tau cascade itself, then a score built to capture inflammatory potential would be expected to behave differently in people whose pathology is already underway than a general-purpose healthy-eating index does. That is roughly what was observed. It is a mechanism worth testing, not a conclusion reached.

It also reframes what the earlier null result means. The randomized trial asked whether a particular dietary pattern, taught for three years to people at family risk, moves cognitive test scores. The Kungsholmen cohort asked whether decades of dietary habit track the arrival of dementia in people whose blood already shows the process under way. Those are different questions, and the second is closer to the one being asked in the kitchen.

The honest read is that this study is more useful for what it narrows than for what it promises. Its headline — healthy diets are linked to less dementia — has been true and unhelpful for a decade. The finding worth keeping is the asymmetry: in a cohort where Alzheimer pathology was already measurable in blood, the score built around inflammatory potential tracked lower dementia rates while the two familiar healthy-eating indices did not, and that split is a clue about mechanism rather than another endorsement of vegetables. It sits uneasily beside the one randomized test the field has run, which found a 0.035-standard-unit difference in global cognition and could not exclude zero. Those two results are less in contradiction than in different rooms: one watched decades of settled habit in people whose biology had already turned, the other coached a three-year diet change in people whose had not. The 29 percent figure is best held as a rate observed in 1,865 Swedes followed 8.4 years, not as a number that transfers to a plate tonight. What the field does not have is a randomized trial in people with elevated biomarkers, and until it does, this is the most interesting unproven idea in dietary dementia prevention.