One question about your health was linked to dementia years later.

Meta-analysis of cohort studies. 143,925 people. self-rated health once at (or near) baseline; cognition re-tested at the latest follow-up, 3 to 27 years later depending on the study.

Many studies, added up. People were watched, not told what to do. A clue, not proof.

Who: Adults in 15 aging cohorts from 38 countries (CHARLS China, CLS Australia, CRELES Costa Rica, ELSA England, HRS USA, JSTAR Japan, LASA Netherlands, LBC1936 Scotland, LBLS USA, MHAS Mexico, MIDUS USA, SATSA Sweden, SHARE 27 European countries + Israel, WLS graduate and sibling samples USA); mean baseline age 53-76 by study. n = 143,925 is the sum of the Table 1 study sizes (abstract: 'over 140,000'); analysed n per model is not given in the main text. Dementia analysis in 10 studies, cognitive impairment in 6

• Thinking tests later13 of 15studies agreed
• Lower dementia odds6 of 10linked, not proven
• Lower impairment oddsunclearafter adjustment

More to know, from our read of the paper:

• Observational: linked, not proven. Authors say they cannot rule out reverse causation in the cross-sectional models (people with poorer thinking may rate their health lower)
• Effects are small: fully adjusted, 1 SD better self-rated health went with about 0.02 SD better cognition at follow-up (B = 0.022). Say 'small but steady', never 'much sharper'
• Cognitive impairment (test score below MMSE/TICS/MoCA cut-off): significant unadjusted (B = -0.115) but NOT after full adjustment (B = -0.058, p = 0.123) -> 'no clear link after adjustment'. (Discussion sentence says 'the meta-analytic summary was also null for the unadjusted and fully adjusted models', which contradicts the Results p = 0.001 unadjusted; use the Results numbers)
• Per-study counts ('14 of 15', '13 of 15', '6 of 10', '5 studies') come from the unadjusted per-study models, which were the preregistered primary tests; adjusted models were the robustness check
• Dementia diagnosis mostly self- or informant-reported (8 of 10 studies) or ICD; self-rated health is a single self-reported item; chronic conditions are self-reported

Researchers ran the same preregistered analysis in 15 long-running aging studies: over 140,000 people from 38 countries. People who rated their own health higher did better on thinking tests 3 to 27 years later (13 of 15 studies), and had lower odds of dementia, even after accounting for age, education, depression and diseases like diabetes and heart problems. The links were small, and the result for milder impairment was unclear after adjustment. Observational: linked, not proven.

We read the abstract and the full text of this paper and checked every number on this page against it (17 facts, checked on 2026-10-29). The study design is named above. Cohort and other observational studies are described as linked, not proven. Found an error? Email theis@onestudyaday.com.