by Simone Salemme
For our paper of the month for March, we have selected Zhang Y, Liu Y, Li Y, et al. Coffee and Tea Intake, Dementia Risk, and Cognitive Function. JAMA. 2026 Feb 9. doi:10.1001/jama.2025.27259.
Coffee and tea are among the most widely consumed beverages worldwide, yet evidence linking them to cognitive health has remained inconsistent—often because studies do not clearly separate caffeinated from decaffeinated coffee, and because outcomes range from subjective symptoms to clinically diagnosed dementia. In this large prospective analysis, the authors examined whether long-term intake of caffeinated coffee, decaffeinated coffee, and tea is associated with incident dementia and with both subjective and objective cognitive outcomes.
The study leveraged two long-running US cohorts: the Nurses’ Health Study (86,606 women; dietary data 1980–2023) and the Health Professionals Follow-up Study 45,215 men; dietary data 1986–2023). Participants were free of cancer, Parkinson’s disease, and dementia at baseline. Dietary intake was assessed every 2–4 years using validated food frequency questionnaires and modelled as a cumulative average to better capture long-term habits. Dementia was identified through death records and self-reported physician diagnoses, with medical record confirmation when available. Subjective cognitive decline was assessed using cohort-specific questionnaires (cases defined as a score ≥3), while objective cognitive function was available only in the Nurses’ Health Study via telephone-based neuropsychological testing, including the Telephone Interview for Cognitive Status (TICS) and a global cognition composite.
Across 131,821 participants followed for up to 43 years (median 36.8), 11,033 incident dementia cases were documented. After multivariable adjustment and pooling results across cohorts, higher caffeinated coffee intake was associated with a lower risk of dementia (hazard ratio 0.82 comparing the highest vs lowest quartile; 141 vs 330 cases per 100,000 person-years) and a lower prevalence of subjective cognitive decline (prevalence ratio 0.85; 7.8% vs 9.5%). Tea intake showed a similar pattern: compared with the lowest tertile, the highest tertile was associated with lower dementia risk (hazard ratio 0.86) and a lower prevalence of subjective cognitive decline (prevalence ratio 0.86). In contrast, decaffeinated coffee was not associated with lower dementia risk, and—interestingly—was associated with a higher prevalence of subjective cognitive decline in pooled analyses (prevalence ratio 1.16 comparing highest vs lowest tertile).
Objective cognitive outcomes (available only in the Nurses’ Health Study) were directionally consistent but modest in magnitude. Participants in the highest vs lowest quartile of caffeinated coffee intake had a slightly higher TICS score (mean difference 0.11 points), while the association with the global cognition composite did not reach conventional statistical significance. The authors appropriately frame these differences as small at an individual level, and more interpretable as population-level shifts rather than clinically meaningful change for a given person.
A key contribution of the study is the dose–response analysis, which suggested a non-linear (threshold/plateau) association. The lowest dementia risk was observed at moderate consumption levels—approximately 2–3 cups/day of caffeinated coffee, 1–2 cups/day of tea, or around 300 mg/day of caffeine—with no clear additional advantage at higher intakes. Notably, similar non-linear patterns were reported for subjective cognitive decline and objective cognitive performance, reinforcing the notion of a “moderate intake” sweet spot.
Overall, these findings suggest that long-term moderate consumption of caffeinated coffee and tea is associated with lower dementia risk and modestly more favourable cognitive outcomes, whereas decaffeinated coffee does not show the same signal in this dataset. As with all observational evidence, residual confounding and reverse causality cannot be fully excluded, and dementia ascertainment still leaves room for misclassification. The study’s scale, repeated exposure assessment, and consistent dose–response pattern make it an important contribution to population-level brain health discourse, while its effect sizes and design underline why these results should be communicated as associative rather than causal.
