Deep divefrom Episode 8Brain & mood27 October 202619 min

What You Eat at Forty-Five and Your Memory at Seventy-Five

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Diet Quality from Midlife to Later Life Relates to Late-Life Brain Health and Verbal Memory in the S G seventy Cohort (bioRxiv preprint)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. Today, one paper about what people in Singapore ate in their forties, and what their brains and memories looked like thirty years later.

Host 2The paper is by Eric Ng, Jun Lai, Woon-Puay Koh, Juan Helen Zhou and colleagues at the National University of Singapore and the Agency for Science, Technology and Research. It was posted to bioRxiv on the twenty-eighth of September twenty twenty-six, so it is a preprint and has not been peer reviewed. They followed eight hundred and ninety-two Chinese Singaporeans whose diets were recorded up to three times across about twenty-eight years, then gave them brain scans and memory tests in their seventies.

Host 1And the headline?

Host 2People who ate closer to a blood-pressure diet called DASH, in midlife or in their sixties, remembered word lists slightly better in their seventies. Some features of their brain scans also looked a bit healthier, and those features account for roughly an eighth of the memory link. It is an association in an observational cohort, not a test of whether changing your diet changes your brain.

Why this field exists

Host 1Start with why anyone studies this. Why diet and the ageing brain?

Host 2Dementia has no cure, so a lot of effort goes into risk factors people can change. As established background, the Lancet Commission on dementia now lists fourteen of them, including high blood pressure, diabetes, physical inactivity and obesity. Diet is not on that list as a separate item, but it feeds into several of the items that are.

Host 1Where does DASH come from? It sounds like a diet with a marketing department.

Host 2It came out of a feeding trial funded by the United States National Institutes of Health in the nineteen-nineties, again as background. DASH stands for Dietary Approaches to Stop Hypertension: plenty of fruit, vegetables, whole grains, nuts, legumes and low-fat dairy, with less red and processed meat, sugary drinks and salt. It was designed to lower blood pressure, and in trials it did that; the brain came later.

Host 1So the logic is: blood vessels feed the brain, DASH helps blood vessels, maybe DASH helps the brain.

Host 2That is the logic. The brain uses a large share of the body's blood supply, and damage to the small vessels deep in the brain is one of the common routes to memory problems and vascular dementia.

What was known before

Host 1What did we already know before this paper?

Host 2Many cohort studies find that people with healthier diet scores have better cognition and lower dementia rates later, and this paper cites a recent study of about one hundred and fifty-nine thousand people linking midlife DASH adherence to late-life cognition. This same Singapore cohort had already linked DASH to global cognition and cognitive impairment. So the association itself is not new.

Host 1And from trials?

Host 2Trials have been less encouraging, which is background, not from this paper. The best-known one, the MIND diet trial published in twenty twenty-three, ran for three years in older Americans and found no clear cognitive difference compared with a control diet, although both groups lost weight and both improved. Three years in your seventies may simply be too short and too late, which is exactly why long-running cohorts like this one are interesting.

The open question

Host 1So what is the open question?

Host 2The authors name three gaps. Most studies measure diet once, often in later life, so they cannot tell a midlife effect from a late-life one. Studies with long follow-up usually look only at dementia diagnoses, without brain scans or tests broken down by type of thinking.

Host 1And the third?

Host 2Nobody knows the biological route. If diet is linked to memory, is it through brain volume, through small-vessel damage, or through something like inflammation, and can you show the brain change sits between the diet and the memory? That is what they set out to test.

What this paper did

Host 1Walk me through who these people are.

Host 2The Singapore Chinese Health Study recruited sixty-three thousand two hundred and fifty-seven adults between nineteen ninety-three and nineteen ninety-eight. Between twenty twenty-one and twenty twenty-five, one thousand one hundred and sixty-five of them, now aged sixty-eight to eighty-two, came back for detailed ageing assessments, and that group is called S G seventy. This analysis uses the eight hundred and ninety-two with usable scans, tests and diet records; four hundred and eighty-nine are women.

Host 1Sixty-three thousand down to under nine hundred. That is a lot of people missing.

Host 2It is, and we will come back to it. Some of the original participants have died or did not return, and the S G seventy group is the youngest slice of the original cohort, people who were forty-four to fifty-five at recruitment, with an average age of forty-seven and a half.

Host 1How was diet measured?

Host 2Three times. At recruitment, an interviewer went through a one hundred and sixty-five item food questionnaire about the past year. Around twenty fourteen to twenty sixteen, average age sixty-seven, and again in S G seventy, average age seventy-four, they used a shorter screener with twenty-one food and drink items.

Host 1Two different questionnaires. Can you compare them?

Host 2The authors built the DASH score from food items harmonised across both, and they say the memory result held up. Two compromises are worth knowing: all dairy counted as a stand-in for low-fat dairy, and the salt component was dropped entirely because the shorter screener could not capture it. So this is DASH without the sodium part, which for a blood-pressure diet is a notable omission.

Host 1What did the scores look like?

Host 2Average DASH scores barely moved over the decades: about twenty-one at every visit, with a spread of roughly four and a half points. So most of the action is in differences between people, not in the population changing its diet.

Host 1Then the brain side. What did they test?

Host 2Nine tests covering seven kinds of thinking: processing speed, attention, visuospatial ability, executive function, visual memory, verbal memory and language. Verbal memory was the Rey Auditory Verbal Learning Test, where you hear a list of words and recall it several times and again after a delay. Every score was put on a scale where fifty is average and ten points is one standard deviation.

Host 1And the scanner?

Host 2A three tesla Siemens MRI scanner at the National University of Singapore, collecting three kinds of image. One measured the size and shape of grey and white matter, one picked up white matter hyperintensities, and one, diffusion imaging, estimated something called free water.

Host 1Explain hyperintensities first.

Host 2They are small bright patches on a particular type of scan, and they mostly reflect damage to tiny blood vessels in the white matter. Instead of adding them all up, this team divided the white matter into thirty-six zones and used a statistical method to group them into seven regional patterns, because patches in different places may have different causes.

Host 1And free water?

Host 2Diffusion imaging tracks how water molecules move, and free water is the fraction that moves unconstrained, outside cells, rather than hemmed in by tissue. Higher free water in grey matter is read as a sign of fluid changes linked to inflammation around blood vessels, or subtle tissue loss. It is a newer marker, and the authors say it has barely been used in diet research.

Host 1How did they connect all of this statistically?

Host 2Regression models, with three levels of adjustment. The first adjusted for energy intake, age, sex, handedness, dialect group and education; the second added smoking, alcohol, physical activity, body mass index, sleep and medical conditions measured at the same time as the diet; the third added the same health factors measured in late life.

Host 1And the brain maps?

Host 2For the scans they tested every point in the brain separately and only kept clusters that survived a correction for doing thousands of tests at once. For the seven thinking domains and seven hyperintensity patterns, they corrected for testing seven things each time.

Host 1Then the mediation part.

Host 2Mediation asks whether part of the diet-to-memory link runs through a brain measure: diet relates to the brain feature, and the brain feature relates to memory. They estimated that indirect path with five thousand bootstrap resamples, which gives a confidence interval without leaning on strict assumptions.

Host 1And the diet trajectories?

Host 2They used a method called latent class growth analysis to sort people into groups by how their DASH score moved across the three visits. It found six groups: consistently low, low-decreasing, low-increasing, consistently average, average-increasing, and consistently high.

Results

Host 1Results. What came out on thinking?

Host 2Of the seven domains, only verbal memory held up after correction for multiple tests. Higher DASH scores in midlife, in the sixties, and at the time of testing were each linked to better word-list memory.

Host 1Only one out of seven. Is that a disappointment or a finding?

Host 2The authors treat it as a finding: verbal memory is known to be sensitive to cardiovascular and metabolic health and is one of the first abilities to slip before clinical impairment. But it also means processing speed, attention, executive function and the rest showed nothing robust, and you should hear that as clearly as the positive result.

Host 1Did the midlife diet matter on its own, or is it just that people who ate well at forty-five still eat well at seventy-five?

Host 2That is the right worry. When they put all three diet scores into one model, midlife DASH still related to verbal memory on its own, more than two decades later. That points toward the midlife period mattering in itself, although it does not rule out everything that tends to travel with a good diet.

Host 1How big is the effect?

Host 2Modest, and the authors call the mediation effects modest themselves. The scatter plots show a gentle upward slope through a wide cloud of people; this is a shift in averages across hundreds of people, not something you would see in one person.

Host 1Now the brain scans.

Host 2Higher midlife DASH went with more white matter volume in the corpus callosum, which links the two hemispheres, and in inferior frontal and parietotemporal regions. Those areas contain long association tracts such as the arcuate fasciculus and superior longitudinal fasciculus, which are involved in language processing.

Host 1Which would fit with verbal memory.

Host 2It fits, and the authors lean on that fit. Grey matter results were scattered and did not survive the stricter adjustments, and the authors say they should be read cautiously.

Host 1What about the vessel damage, the hyperintensities?

Host 2Higher DASH in the sixties went with fewer hyperintensities in the white matter around the deep basal ganglia, and more weakly in a zone near the front of the ventricles. The deep basal ganglia result survived all three adjustment levels; the anterior one did not.

Host 1And free water?

Host 2Higher DASH in the sixties went with lower free water in the cortex, in parts of the left frontal and right occipital lobes. The same link appeared, more widely, between diet and free water measured at the same visit in late life. There was nothing in deep grey matter or in white matter tracts.

Host 1So which of these brain features actually tracked memory?

Host 2Only two: more white matter volume and lower cortical free water went with better verbal memory. Fewer hyperintensities, despite tracking diet, did not relate to memory here, which the authors flag as unexpected.

Host 1And the mediation numbers.

Host 2White matter volume accounted for twelve point three percent of the link between midlife diet and late-life memory, with a confidence interval that clearly excludes zero. Cortical free water accounted for twelve point five percent of the link between diet in the sixties and memory. In both cases most of the link, nearly nine-tenths, runs through something these measures did not capture, or is not causal at all.

Host 1Now the part people will want to hear. Can you fix it later?

Host 2The trajectory groups are the closest the paper gets to that. People who started low in midlife and improved remembered better than the consistently low group, but still worse than the consistently high group, and they had less white matter and more free water than the high group.

Host 1And the ones who started average?

Host 2The average-increasing group also remembered better than the consistently low group, and on verbal memory and cortical free water they were statistically indistinguishable from the consistently high group. They still had less white matter volume than the high group, which the authors read as a midlife window that had already closed for that measure.

Host 1How solid are those group comparisons?

Host 2Some are borderline. The low-increasing group's memory advantage over the consistently low group had a p value of point zero three four, the white matter difference for the average-increasing group was only a trend, and the free water difference sat right at the edge after correction. "Statistically indistinguishable" from the high group also means the test could not separate them, not that they were proven equal.

Host 1Did anything change the results when they pushed on them?

Host 2They repeated the analyses without one hundred and four people who had signs of cognitive impairment or depressive symptoms, leaving seven hundred and eighty-eight, and the key findings held. The verbal memory, white matter, deep basal ganglia and free water results survived all three adjustment levels; grey matter and the anterior hyperintensity result did not.

Host 1Before we get to what it all means, a short break.

A word from our sponsor

Host 1Do you have something that you know you should have dealt with years ago, but you would rather deal with it on the morning itself?

Host 1The Hastrup Review of Science is brought to you by dry shampoo. A powder in a can, applied at the roots, and entirely unconcerned with your long-term trajectory.

Host 2I used it this morning instead of washing. My midlife adherence was poor and my late-life intervention was rushed, and the result was statistically indistinguishable from effort.

Host 1Dry shampoo is sold in most drugstores at ordinary prices. No food questionnaire, no diffusion scanner, no twenty-eight-year follow-up. The link is in the show notes.

Why it matters

Host 1Back to the paper. What is the principle here?

Host 2That diet at different ages may relate to different parts of the brain. In this data, midlife diet goes with white matter volume, a kind of structural reserve, while diet in the sixties goes with markers of small-vessel damage and inflammation-like fluid changes.

Host 1Who does that change anything for?

Host 2For researchers, it suggests that diet studies should pick their brain outcomes by age, and that free water is worth adding to the toolkit. For public health, it supports the idea that diet in midlife counts for decades, and that improving it later is still associated with something, at least for people starting from average.

Host 1And for someone listening who is sixty and eats badly?

Host 2The paper is consistent with improvement after midlife being worthwhile, which is also what the broader evidence on blood pressure and heart health already says. The authors point out that most people eat averagely rather than terribly or perfectly, which is why the average-increasing group is the practically relevant one.

Host 1What does it not mean?

Host 2It does not show that DASH prevents dementia; nobody here was followed to a dementia diagnosis. It does not show that diet causes the white matter or free water differences, because diet was not assigned and each brain was scanned only once. And it does not mean any single food matters, because this is a broad pattern score with the salt part missing.

Caveats

Host 1Then the caveats, all of them.

Host 2First, this is observational. People who ate more DASH-like diets in nineteen ninety-five likely differed in many ways, and adjusting for education, smoking, exercise and illness reduces that problem without removing it.

Host 1Second?

Host 2Survivorship. Out of sixty-three thousand people recruited, about eleven hundred came back in their seventies, and the eight hundred and ninety-two analysed were the ones healthy and willing enough to finish scans and tests. The authors argue this kind of bias tends to weaken associations rather than create them, which is a reasonable argument but not a guarantee.

Host 1Third?

Host 2Diet was self-reported, with a long questionnaire at the start and a twenty-one-item screener later, and the salt component was absent throughout. The authors acknowledge recall and social desirability bias.

Host 1Fourth?

Host 2Only one brain scan per person, in late life. So you cannot see whether white matter actually shrank more slowly in good eaters, or whether they simply had more of it to begin with, perhaps for reasons tied to education or childhood.

Host 1Anything else?

Host 2Selective reporting risk is moderate: seven thinking domains, seven hyperintensity patterns, many brain maps, three life stages, and only some survived. The mediation analysis was also limited to measures already found significant, which can flatter the result. And one small sign of the preprint stage: the participant table lists nearly forty-four percent of people as underweight at the second visit, which looks like a table error rather than a real finding.

Host 1Is the data available for anyone to check?

Host 2The analysis code is on GitHub. The data themselves are not public because of privacy law, but available from the authors on reasonable request.

What to watch next

Host 1What should we watch for next?

Host 2Repeated brain scans in the same people, which the authors themselves say are needed to test whether the age-specific patterns are real. Peer review of this preprint, including a check of that participant table.

Host 1And beyond this cohort?

Host 2Replication in other populations, since the authors argue the vascular angle may matter especially in Asian groups, where vascular dementia is more common. And, ultimately, longer trials starting earlier than the seventies, because the short late-life trials so far have not settled the question.

Bottom line

Host 1Bottom line?

Host 2In about nine hundred older Chinese Singaporeans, a healthier, DASH-like diet in midlife and in the sixties went with modestly better word-list memory in the seventies, and with a few healthier-looking brain features that account for about an eighth of that link. It is a careful, long-running observational study with one main cognitive result, several modest brain associations, and the usual limits of self-reported diet and a single scan.

Sign-off

Host 1That is the deep dive: Diet Quality from Midlife to Later Life Relates to Late-Life Brain Health and Verbal Memory in the S G seventy Cohort, by Ng, Koh, Zhou and colleagues, a preprint on bioRxiv. The link is in the show notes.

Host 1Every Friday this same feed carries the roundup episode: five new papers as trailers, and you pick which become deep dives. This has been the Hastrup Review of Science.