CoChat
Sign Up
Cited by 836Dementia and Cognitive Impairment ResearchCognitive Abilities and TestingAging and Gerontology Research

Timing of onset of cognitive decline: results from Whitehall II prospective cohort study

Open access
Singh-Manoux, A., Kivimaki, M., Glymour, M. M., Elbaz, A., Berr, C., Ebmeier, K. P., Ferrie, J. E., Dugravot, A.·2012·BMJ
DOI 10.1136/bmj.d7622PMC PMC3281313
Open this paper in CoChat — free accountOpen PDF

Abstract

OBJECTIVES: To estimate 10 year decline in cognitive function from longitudinal data in a middle aged cohort and to examine whether age cohorts can be compared with cross sectional data to infer the effect of age on cognitive decline. DESIGN: Prospective cohort study. At study inception in 1985-8, there were 10,308 participants, representing a recruitment rate of 73%. SETTING: Civil service departments in London, United Kingdom. PARTICIPANTS: 5198 men and 2192 women, aged 45-70 at the beginning of cognitive testing in 1997-9. MAIN OUTCOME MEASURE: Tests of memory, reasoning, vocabulary, and phonemic and semantic fluency, assessed three times over 10 years. RESULTS: All cognitive scores, except vocabulary, declined in all five age categories (age 45-49, 50-54, 55-59, 60-64, and 65-70 at baseline), with evidence of faster decline in older people. In men, the 10 year decline, shown as change/range of test × 100, in reasoning was -3.6% (95% confidence interval -4.1% to -3.0%) in those aged 45-49 at baseline and -9.6% (-10.6% to -8.6%) in those aged 65-70. In women, the corresponding decline was -3.6% (-4.6% to -2.7%) and -7.4% (-9.1% to -5.7%). Comparisons of longitudinal and cross sectional effects of age suggest that the latter overestimate decline in women because of cohort differences in education. For example, in women aged 45-49 the longitudinal analysis showed reasoning to have declined by -3.6% (-4.5% to -2.8%) but the cross sectional effects suggested a decline of -11.4% (-14.0% to -8.9%). CONCLUSIONS: Cognitive decline is already evident in middle age (age 45-49).

AI snapshot

Written from the paper’s full text.

TL;DR

In the Whitehall II prospective cohort, cognitive scores except vocabulary declined over 10 years in all age groups from 45–49 to 65–70, with faster decline at older ages. Cross-sectional age comparisons overestimated cognitive decline in women, probably because of cohort differences in education.

Key findings

  • All cognitive scores except vocabulary declined in all five baseline age categories (45–49, 50–54, 55–59, 60–64, 65–70), with evidence of faster decline in older people. In men, 10-year decline in reasoning was −3.6% (95% CI −4.1% to −3.0%) in those aged 45–49 and −9.6% (−10.6% to −8.6%) in those aged 65–70; in women, the corresponding declines were −3.6% (−4.6% to −2.7%) and −7.4% (−9.1% to −5.7%).
  • Longitudinal and cross-sectional estimates diverged, especially in women: in women aged 45–49, reasoning declined by −3.6% (−4.5% to −2.8%) longitudinally, but cross-sectional effects suggested a decline of −11.4% (−14.0% to −8.9%).
  • Adjustment for education did not change longitudinal estimates much but substantially reduced cross-sectional estimates of ageing, consistent with cohort differences in education. For women aged 45–49 at baseline, 34% had less than secondary school education, 28% secondary school, and 40% a university degree; for women aged 55–59, the corresponding figures were 58%, 25%, and 17%.
  • Vocabulary was little influenced by age; significant declines were seen for all tests except vocabulary in all age categories in both men and women.

Methods

  • Prospective cohort study using Whitehall II data; cognitive testing was introduced at phase 5 (1997–9) and repeated at phases 7 (2002–4) and 9 (2007–9). Analysis included 7390 participants (5198 men, 2192 women) aged 45–70 at baseline; 4675 (63%) had complete data at all three phases.
  • Cognitive battery included the Alice Heim 4-I reasoning test, a 20-word free recall memory test, phonemic and semantic fluency tests, and the Mill Hill vocabulary test. Results were expressed as percent change (change/range of test × 100).
  • Linear mixed models with random intercept and slope estimated 10-year decline, including age cohort × time interaction terms and stratification by sex; cross-sectional linear regression at baseline was compared with longitudinal estimates. Analyses used SAS Proc Mixed version 9.1.

Limitations

  • The paper states that longitudinal data can underestimate age effects because of practice effects and selective sample retention; non-response was higher in older individuals, though the overall response rate was 80% (7390/9250) among those invited to cognitive testing.
  • The paper states that Whitehall II is not representative of the general population: participants were mostly white collar workers in relatively stable employment and two thirds were men, so results could underestimate cognitive decline at the population level.
  • The paper notes that adjustment for education may not capture all cohort-related effects and may overcorrect by attributing shared variance between age and education to education. Evident from the text: cognitive testing began at age 45, so the study cannot determine whether decline begins before age 45.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1136/bmj.d7622

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 15 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    There was a cross sectional association between age at baseline and all cognitive scores (all P<0.001), except vocabulary in men (P=0.13).

    “There was a cross sectional association between age at baseline, categorised into five age cohorts of five years each spanning 45-70 years, and all cognitive scores (all P<0.001), except vocabulary in men (P=0.13)”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
  • Findingown data

    In men aged 65-70 at baseline, the 10-year decline in reasoning was -9.6% (95% CI -10.6% to -8.6%).

    “and -9.6% (-10.6% to -8.6%) in those aged 65-70.”
    • Also stated in: § Results · plain
      “while in those aged 65-70 it was -9.6% (-10.6% to -8.6%)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.
  • Findingown data

    In women aged 65-70 at baseline, the 10-year decline in reasoning was -7.4% (95% CI -9.1% to -5.7%).

    “and -7.4% (-9.1% to -5.7%).”
    • Also stated in: § Results · plain
      “and -7.4% (-9.1% to -5.7%) in those aged 65-70.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.
  • Findingown data

    In women aged 45-49 at baseline, the 10-year decline in reasoning was -3.6% (95% CI -4.6% to -2.7%).

    “In women, the corresponding decline was -3.6% (-4.6% to -2.7%)”
    • Also stated in: § Results · plain
      “In women, the corresponding decline was -3.6% (-4.6% to -2.7%) in those aged 45-49”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.
  • Findingown data

    There were considerable cohort differences in education in women.

    “There were considerable cohort differences in education in women.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 12
  • Classified as probably not a substantial claim. 15

  • May combine two claims.

    Findingown data

    In women aged 45-49, longitudinal analysis showed reasoning declined by -3.6% (95% CI -4.5% to -2.8%) but cross sectional effects suggested a decline of -11.4% (95% CI -14.0% to -8.9%).

    “the longitudinal analysis in women shows reasoning to have declined by -3.6% (-4.5% to -2.8%) in those aged 45-49 but the cross sectional effects suggest a decline of -11.4% (-14.0% to -8.9%).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.

  • May combine two claims.

    Findingown data

    All cognitive scores, except vocabulary, declined in all five age categories.

    “All cognitive scores, except vocabulary, declined in all five age categories (age 45-49, 50-54, 55-59, 60-64, and 65-70 at baseline)”
    • Also stated in: § Discussion · plain
      “average performance in all cognitive domains, except vocabulary, declined over follow-up in all age groups, even in those aged 45-49.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.

  • May combine two claims.

    Findingown data

    Adjustment for education did not change the longitudinal estimates much but substantially reduced the cross sectional estimates of the effect of ageing on cognitive decline.

    “Adjustment for education did not change the longitudinal estimates much but substantially reduced the cross sectional estimates of the effect of ageing on cognitive decline.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.
    • Rests on an assumption: The cognitive tests are valid measures of cognitive function in this middle-aged population.
    • Rests on an assumption: Cognitive decline over the 10-year follow-up is linear in time within each age cohort.

  • Classified as probably not a substantial claim.

    Premisecitation

    Clinicopathological studies show good correlation between neuropathology and the severity of cognitive decline.

    “Clinicopathological studies show good correlation between neuropathology and the severity of cognitive decline, [9][10][11]”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [9][10][11]

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    There is a finely graded inverse association between age and cognitive performance.

    “There is a finely graded inverse association between age and cognitive performance, [3][4][5]”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3][4][5]

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Emerging consensus on the long gestation period of dementia suggests that adults aged under 60 are likely to experience age related cognitive decline.

    “Emerging consensus on the long gestation period of dementia 14 15 also suggests that adults aged under 60 are likely to experience age related cognitive decline.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites 14 15

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    As non-response was not substantial (overall response rate 80%), major bias is unlikely.

    “As non-response was not substantial (the overall response rate was 80% (7390/9250) in those invited to cognitive testing), however, major bias is unlikely.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Longitudinal data are known to underestimate the effect of age because of practice effects and selective sample retention.

    “Longitudinal data are known to underestimate the effect of age because of practice effects 22 and are subject to selective sample retention.23”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [22]

      Not checked in the free run.

    • cites [23]

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisenone given

    Cross sectional data might not accurately distinguish longitudinal change from secular change (cohort effects).

    “Cross sectional data might not accurately distinguish longitudinal change-that is, the effect of chronological age-from secular change, also called “cohort” effects.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The mini-mental state examination was not used in the current analysis because of ceiling effects.

    “The mini-mental state examination, 17 although included in the test battery, was not used in the current analysis because of ceiling effects (that is, simple tests leading to little variability in scores).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Sex specific cohort differences in educational attainment seen in this sample reflect global trends in educational opportunities across the 20th century.

    “The sex specific cohort differences in educational attainment seen in this sample, however, reflect global trends in educational opportunities across the 20th century.24 25”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites 24 25

      Not checked in the free run.

  • May combine two claims.

    Study design and conductown data

    In this study, non-response was higher in older individuals, affecting both cross sectional and longitudinal estimates.

    “In our study non-response was higher in the older individuals, affecting both cross sectional as well and longitudinal estimates.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The participants are mostly white collar workers in relatively stable employment and two thirds are men, implying that the results could underestimate cognitive decline at the population level.

    “The participants are mostly white collar workers in relatively stable employment and two thirds are men, implying that our results could underestimate cognitive decline at the population level.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    There is emerging consensus that what is good for our hearts is also good for our heads.

    “There is emerging consensus that “what is good for our hearts is also good for our heads,” 36 37”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites 36 37

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The analysis was based on 7390 (5198 men and 2192 women) of the 9250 individuals still in the study.

    “Our analysis was based on 7390 (5198 men and 2192 women) of the 9250 individuals still in the study.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    The exposure, time over 10 years, is uniformly applicable to all age cohorts in the analysis.

    “The exposure, in this case time over 10 years, is uniformly applicable to all age cohorts in the analysis.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    The obvious advantage of using longitudinal analysis is that the estimates of cognitive decline are based on changes within an individual, and these are unlikely to be affected by error arising from natural variation between individuals.

    “The obvious advantage of using longitudinal analysis is that the estimates of cognitive decline are based on changes within an individual, and these are unlikely to be affected by error arising from natural variation between individuals.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisenone given

    One challenge in small scale studies set up to investigate cognitive ageing is that they might be based on a selected sample of individuals.

    “One challenge in small scale studies set up to investigate “cognitive ageing” is that they might be based on a selected sample of individuals.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Of the 10,308 participants at study inception, 306 had died and 752 had withdrawn before the start of cognitive data collection in 1997-9.

    “Of the 10 308 participants at study inception (1985-8), 306 had died and 752 had withdrawn from the study before the start of the cognitive data collection in 1997-9.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Those included in the analyses were more educated and younger than those not included.

    “Those included in the analyses were more educated (29.5% v 20.8% had a university degree, P<0.001) and were younger”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The results for vocabulary were expected as it is known to be little influenced by age.

    “The results for vocabulary were expected as it is known to be little influenced by age. 4”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4]

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning

    These estimates reflect both the effects of chronological ageing and cohort effects, each produced by distinct underlying mechanisms.

    “These estimates, however, reflect both the effects of chronological ageing and cohort effects, each produced by distinct underlying mechanisms.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Of the 7390 individuals included in the analysis, complete data were available for 4675 (63%) at all three phases, 1613 at two phases, and 1102 at only one phase.

    “Of the 7390 individuals included in the analysis, we had complete data for 4675 (63%) at all three phases, 1613 at two phases, and 1102 at only one phase.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Cohort effects are produced when a birth cohort enters adult life with long lasting effects on cognitive function because of childhood experiences like quality of education, nutrition, socioeconomic circumstances, etc.

    “Cohort effects are produced when a birth cohort enters adult life with long lasting effects on cognitive function because of childhood experiences like quality of education, nutrition, socioeconomic circumstances, etc. [31][32][33]”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [31][32][33]

      Not checked in the free run.

24 of 24

  • The paper relies on this without support.

    Missing cognitive data are missing at random, so linear mixed models yield unbiased estimates.

    “Linear mixed models 21 were used to estimate cognitive decline over follow-up.This method uses all available data over follow-up, takes into account the fact that repeated measures on the same individual are correlated with each other, and can handle missing data.”

  • The paper relies on this without support.

    The cognitive tests are valid measures of cognitive function in this middle-aged population.

    “The cognitive test battery, administered at three clinical examinations over 10 years (phases 5, 7, and 9), was chosen to provide a comprehensive assessment of cognitive function and be appropriate for this population composed of individuals younger than in most studies on cognitive ageing.”

  • The paper relies on this without support.

    Cognitive decline over the 10-year follow-up is linear in time within each age cohort.

    “The models included terms for time (age in years, divided by 10 so the coefficients yield effects of decline over 10 years), age cohorts (age at baseline in five year bands) to adjust for secular effects, and an interaction term between age cohort and time, separately in men and women.”
3 of 3

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Decreased cognitive performance begins around midlife from 45 years and increases with age due to structural and functional brain changes.

    “The decreased cognitive performance begins around midlife from 45 years [173] and increases with age due to structural and functional changes in the brain (for instance, regional brain volumes and integrity of the white matter, reduction in the fluidity of brain membranes, and changes in lipid composition) [163,174,175,176].”

    From Alzheimer's Disease: Treatment Strategies and Their Limitations. (2022) · cites this paper as [173]

  • Cross-sectional analysis overestimated age-associated cognitive decline compared with longitudinal design.

    “For instance, in comparison with longitudinal design, cross-sectional analysis was found to overestimate age-associated cognitive decline (Singh-Manoux et al, 2012).”

    From Microglia Priming with Aging and Stress (2017) · cites this paper as Singh-Manoux et al, 2012

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Two values given for reasoning decline in younger men

    Minor

    The results report a 10-year reasoning decline of -3.6% for men aged 45-49 at baseline, but the concluding summary reports -3.5% for the same group.

  • Conflicting secondary-school education percentage for younger women

    Minor

    The results state that 28% of women aged 45-49 had completed secondary school, while the table footnote gives 26% for the same group. The results' three education percentages for these women also total 102%.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe: http://www.bmj.com/subscribe

Cited by

Sorted by citations

Related

Authors
Singh-Manoux, A., Kivimaki, M., Glymour, M. M., Elbaz, A., Berr, C., Ebmeier, K. P., Ferrie, J. E., & Dugravot, A.
Affiliations
  • Harvard University (US)
  • Inserm (FR)
  • Warneford Hospital (GB)
  • University of Bristol (GB)
ORCID
Venue
BMJ
Publisher
BMJ
Published
2012-01-05
Volume
344
Issue
jan04 4
Pages
d7622-d7622
ISSN
  • 0007-1447 (ISSN-L)
  • 0959-8138
  • 1468-5833
DOI
10.1136/bmj.d7622
PMC
PMC3281313
OpenAlex
W2168922851
Cited by
836
References
41
Citations by year
2026: 45 · 2025: 39 · 2024: 66 · 2023: 67 · 2022: 52 · 2021: 81 · 2020: 70 · 2019: 64 · 2018: 44 · 2017: 54 · 2016: 51 · 2015: 45 · 2014: 52 · 2013: 55 · 2012: 46
FWCI
44.75
Type
journal article
Language
en
Open access
Licence
cc-by-nc
Locations
  • BMJ (publishedVersion, cc-by-nc)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • Digital Access to Scholarship at Harvard (DASH) (Harvard University) (submittedVersion, cc-by)
  • HAL (submittedVersion, other-oa)
  • UCL Discovery (University College London) (submittedVersion)
Funders
  • British Heart Foundation (R01 AG013196, R01AG034454, RG/07/008/23674)
  • Bupa Foundation
  • European Science Foundation
  • Academy of Finland (R01 AG034454)
  • National Institutes of Health
  • Medical Research Council (R01AG034454, G0902037)
  • National Institute on Aging (R01AG034454, R01AG013196)
  • National Heart, Lung, and Blood Institute (R01HL036310)
Topics
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Abilities and Testing
  • Social Sciences › Psychology › Neuropsychology and Physiological Psychology › Aging and Gerontology Research
Keywords
cognitive decline, cognitive aging, middle age (human life stage), age cohort effects, sex differences, reasoning, memory, vocabulary, fluency, Whitehall II cohort, public sector employees
MeSH terms
Adult, Aged, Aging, Cognition Disorders, Educational Status, Epidemiologic Methods, Female, Humans, London, Male, Middle Aged, Neuropsychological Tests, Age of Onset

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review