Two coffee-and-dementia analyses landed three days apart. They agree on the dose.
A 131,821-person cohort followed for up to 43 years and a dose-response meta-analysis of 450,000 people converge on roughly two to three cups — and diverge sharply above it.
| Group | Value (per 100,000 person-years) |
|---|---|
| Highest quartile | 141 |
| Lowest quartile | 330 |
Two analyses of coffee, tea and dementia published within three days of each other this month reach conclusions that look opposed in their summaries and are largely compatible in their numbers. Reading them together is a useful exercise in what nutrition epidemiology can and cannot settle.
The cohort study
The larger of the two, in JAMA, drew on two long-running US cohorts: 86,606 women in the Nurses' Health Study with data from 1980 to 2023, and 45,215 men in the Health Professionals Follow-up Study with data from 1986 to 2023, all free of cancer, Parkinson disease and dementia at entry [s1]. Diet was collected every two to four years with validated food frequency questionnaires [s1].
Among 131,821 participants followed for up to 43 years — median 36.8 years, IQR 28 to 42 — there were 11,033 incident dementia cases [s1]. That length of follow-up is the study's principal asset. Reverse causation is the standing objection to coffee-and-dementia studies, because people in the prodromal phase of dementia change what they consume; four decades of repeated measurement does not eliminate the problem but pushes the exposure a long way ahead of the outcome.
After adjustment and pooling across cohorts, participants in the highest quartile of caffeinated coffee intake had lower dementia risk than those in the lowest: 141 versus 330 cases per 100,000 person-years, hazard ratio 0.82 (95% CI 0.76–0.89) [s1]. Subjective cognitive decline was less prevalent as well, at 7.8% versus 9.5%, prevalence ratio 0.85 (0.78–0.93) [s1].
Objective cognitive testing, available only in the Nurses' Health Study, showed a smaller and partly non-significant picture. The highest quartile had a higher mean Telephone Interview for Cognitive Status score (mean difference 0.11, 95% CI 0.01 to 0.21) on a scale running 0 to 41, and a higher mean global cognition score (mean difference 0.02, 95% CI −0.01 to 0.04) that did not reach statistical significance, at P = .06 [s1].
A difference of 0.11 points on a 41-point scale is not a change anyone would notice. It is consistent with the dementia finding and it is not, on its own, evidence of a meaningful cognitive benefit.
Tea showed similar associations to caffeinated coffee [s1]. Decaffeinated coffee was not associated with lower dementia risk or better cognitive performance [s1] — which is the study's most informative internal comparison, because decaf differs from caffeinated coffee mainly in the caffeine.
The dose-response analysis was nonlinear, and the most pronounced differences appeared at roughly 2 to 3 cups a day of caffeinated coffee or 1 to 2 cups a day of tea [s1].
The meta-analysis
The second paper pooled the published cohort literature rather than analysing individuals. It searched PubMed and EMBASE to 9 December 2025, included cohort and cohort-nested case-control studies in populations free of chronic disease and without prior dementia, and assessed quality with the ROBINS-E tool [s2]. Ten studies met criteria, with more than 450,000 participants at baseline and a mean follow-up of 11.5 years [s2].
For coffee, the pooled dose-response curve was U-shaped, with the lowest risk between 2 and 3 cups a day — about 300 to 450 mL [s2]. For Alzheimer's dementia specifically, risk showed no difference up to 3 cups a day and then began to increase [s2]. The authors' stated conclusion is that moderate coffee consumption does not appear to affect dementia risk, while high consumption of 3 or more cups a day could increase the risk of all-cause dementia and Alzheimer's dementia [s2].
Tea behaved differently: a progressive, linear decrease in all-cause dementia risk with increasing consumption, with comparable results for all tea types and for green tea alone [s2]. For Alzheimer's dementia, no further decrease appeared above about one cup a day [s2].
Where they agree and where they do not
Both analyses locate the same optimum for coffee — 2 to 3 cups [s1][s2] — and both find tea associated with lower risk [s1][s2]. Neither finds a benefit that keeps growing with dose.
The divergence is entirely above 3 cups. The cohort study's highest quartile carried the lowest risk [s1]; the meta-analysis found the curve turning back upward there [s2]. Two structural differences plausibly account for that. The JAMA analysis compares quartiles of a population's actual distribution, so its "highest quartile" is not necessarily a very high absolute intake, and it does not report a harm threshold. The meta-analysis models the pooled curve across studies whose populations, cup sizes and confounder adjustments differ, which is where nonlinear pooling is most fragile.
The framings also differ in emphasis in a way the data do not fully justify on either side. "Coffee associated with lower dementia risk" and "moderate coffee does not appear to affect dementia risk" are both defensible descriptions of a shallow curve.
The limits that apply to both
Neither study can establish cause. Both rely on self-reported intake, in the cohort's case by questionnaire every two to four years [s1]. Coffee and tea consumption travel with education, smoking, alcohol, sleep and physical activity, and no adjustment fully removes that. The meta-analysis pooled only ten studies [s2], and the cohorts in the JAMA paper are US health professionals — a group whose diet, income and health care are not representative of the general population [s1].
There is also no trial. Nothing in either paper tells a reader whether changing their intake would change their risk, and neither set of authors claims it does.
What to watch
The decaffeinated comparison in the JAMA study [s1] is the thread most likely to be pulled next, because it points at caffeine rather than at the other several hundred compounds in coffee. If subsequent work reproduces a caffeine-specific signal that decaf does not share, the hypothesis narrows usefully. If it does not, the association stays where it has been for two decades: consistent, modest and unexplained.
Sources
- [s1] Coffee and tea intake, dementia risk, and cognitive function. JAMA, 9 February 2026. https://doi.org/10.1001/jama.2025.27259
- [s2] Coffee and tea consumption and risk of dementia: a dose-response meta-analysis of cohort and cohort-nested case-control studies. Journal of Epidemiology and Population Health, 12 February 2026. https://doi.org/10.1016/j.jeph.2026.203168
Sources
- Coffee and tea intake, dementia risk, and cognitive function — JAMA , February 9, 2026
- Coffee and tea consumption and risk of dementia: a dose-response meta-analysis of cohort and cohort-nested case-control studies — Journal of Epidemiology and Population Health , February 12, 2026
More on
A sense of purpose tracks with living longer and less dementia. The catch is causation.
Two large cohorts link purpose in life to lower mortality and lower Alzheimer's risk. Whether purpose protects health, or health sustains purpose, is what the studies cannot fully separate.
CTE in 2026: what the new evidence established, and what it still cannot
A 614-donor autopsy study tied advanced CTE to dementia independently of other brain disease. A separate imaging report in three former athletes hints at a test for the living. The caveats on both are heavy.
Cognitive reserve is real. What you can do about it is a smaller, honest list.
Autopsy studies show educated brains carry the same disease but express it as dementia later. That is the concept's strength — and the reason it resists being turned into a to-do list.
Crossword-doers decline more slowly. That is not the same as crosswords working.
The link between puzzles and a sharper old age is real in the data and largely the wrong way round: the earliest changes of dementia quietly reduce the very activities that later look protective.