Older Japanese adults who cooked weekly had lower dementia rates — with a large asterisk

Japan has the demographic problem the rest of the world is heading towards, and it also has an unusual research asset: a public long-term care insurance system whose certification records capture cognitive impairment severe enough to require care. That makes dementia an outcome you can follow administratively, across a whole population, without waiting for a clinical diagnosis to be recorded.

A cohort study published online on 24 March used that record to ask a narrow question: does cooking at home track with later dementia [s1]?

What was measured

Participants came from the Japan Gerontological Evaluation Study, a population-based cohort. Cooking frequency and self-rated cooking skills were assessed in a baseline survey, and 10,978 participants were followed for six years [s1]. Dementia incidence was ascertained through the public long-term care insurance system, which captures functionally significant cognitive impairment requiring care — not milder impairment, and not undiagnosed cases [s1].

Over follow-up, 1,195 dementia cases were recorded [s1].

The analysis compared high cooking frequency, defined as at least once a week, against low frequency, less than once a week [s1]. Because people who cook often differ systematically from people who do not, the authors used propensity score matching on demographic, socioeconomic and health-related factors, matching within sex [s1]. That produced 1,347 matched male pairs and 321 matched female pairs [s1]. Death was treated as a competing event using Fine-Gray models — an important choice in an elderly cohort, where people who die cannot go on to develop dementia [s1].

What was found

The subdistribution hazard ratio for high versus low cooking frequency was 0.77 (95% CI 0.61 to 0.98) in men and 0.73 (95% CI 0.54 to 0.98) in women [s1].

Both estimates cross close to 1.0 at their upper bound. In men, the confidence interval runs from a 39% lower risk to a 2% lower risk; in women, from 46% lower to 2% lower. The direction is consistent across sexes, but the precision at the null end is thin. This is a signal, not a settled effect size.

The subgroup result is the one likeliest to be over-read. Among participants who rated their cooking skills as low, the association was much stronger: SHR 0.33 (95% CI 0.13 to 0.84) [s1]. That confidence interval spans a factor of more than six. It is compatible with a large effect and with a modest one, and it comes from a subgroup of an already-matched sample. Subgroup findings with wide intervals are the standard place where cohort studies generate claims that later fail to replicate.

The problem the design cannot solve

Reverse causation is the central difficulty here, and it is not a technicality.

Cooking is a complex instrumental activity of daily living. Planning a meal, sequencing steps, judging timing and handling a stove all draw on exactly the executive functions that decline first in preclinical dementia — often years before a long-term care certification. A person in the early, undiagnosed phase of cognitive decline may cook less because of that decline. In a study that measures cooking at baseline and dementia later, that would produce the observed association without cooking protecting anyone.

Propensity matching on demographics, socioeconomic status and health does not fix this, because the confounder is the outcome itself in an earlier and unmeasured form. The authors' own conclusion is worded carefully — that creating an environment where people can cook when older may be important for dementia prevention [s1] — and "may" is carrying the weight.

The low-skill subgroup finding is interesting partly because it cuts slightly against the reverse-causation story: if cooking frequency were purely a marker of preserved cognition, you would not obviously expect the strongest association among people who cook badly. That is an argument, not evidence, and the interval around it is too wide to lean on.

What it does and does not support

It supports the claim that, in one large Japanese cohort, weekly home cooking at baseline was associated with lower six-year incidence of dementia requiring long-term care, with the association robust enough to survive propensity matching and a competing-risk model [s1].

It does not establish that taking up cooking reduces dementia risk. No trial has been done, and the observational design cannot distinguish cooking as a cause from cooking as an early symptom. Anyone reading this as an instruction to start cooking is reading past what the study measured.

What to watch

Whether the finding replicates in cohorts with repeated measures of cooking frequency over time — which would let researchers see whether cooking declines before the dementia trajectory begins, the single most informative test available short of a trial. Japan's long-term care records make that a tractable study to run.

This article is informational and does not constitute medical advice.

Sources

• [s1] Tani Y, Fujiwara T, Kondo K., "Home cooking, cooking skills and dementia requiring long-term care: a population-based cohort study in Japan," Journal of Epidemiology and Community Health, published online 24 March 2026. https://doi.org/10.1136/jech-2025-225139