Cognitive stimulation in the workplace, plasma proteins, and risk of dementia: three analyses of population cohort studies
BMJ 2021; 374 doi: https://doi.org/10.1136/bmj.n1804 (Published 19 August 2021) Cite this as: BMJ 2021;374:n1804
- Mika Kivimäki, professor123,
- Keenan A Walker, tenure-track investigator45,
- Jaana Pentti, senior statistician236,
- Solja T Nyberg, postdoctoral fellow2,
- Nina Mars, postdoctoral fellow7,
- Jussi Vahtera, professor emeritus68,
- Sakari B Suominen, professor69,
- Tea Lallukka, professor2,
- Ossi Rahkonen, professor emeritus2,
- Olli Pietiläinen, postdoctoral fellow2,
- Aki Koskinen, statistician3,
- Ari Väänänen, research professor3,
- Jatinderpal K Kalsi, project manager1,
- Marcel Goldberg, professor emeritus1011,
- Marie Zins, director1011,
- Lars Alfredsson, professor1213,
- Peter J M Westerholm, professor emeritus14,
- Anders Knutsson, professor emeritus15,
- Töres Theorell, professor emeritus16,
- Jenni Ervasti, senior research fellow3,
- Tuula Oksanen, professor17,
- Pyry N Sipilä, postdoctoral fellow2,
- Adam G Tabak, postdoctoral fellow118,
- Jane E Ferrie, senior research fellow119,
- Stephen A Williams, director20,
- Gill Livingston, professor2122,
- Rebecca F Gottesman, professor4,
- Archana Singh-Manoux, research professor111,
- Henrik Zetterberg, professor2324,
- Joni V Lindbohm, postdoctoral fellow12
- Correspondence to: M Kivimäki m.kivimaki@ucl.ac.uk
- Accepted 14 July 2021
Abstract
Objectives To examine the association between cognitively stimulating work and subsequent risk of dementia and to identify protein pathways for this association.
Design Multicohort study with three sets of analyses.
Setting United Kingdom, Europe, and the United States.
Participants Three associations were examined: cognitive stimulation and dementia risk in 107 896 participants from seven population based prospective cohort studies from the IPD-Work consortium (individual participant data meta-analysis in working populations); cognitive stimulation and proteins in a random sample of 2261 participants from one cohort study; and proteins and dementia risk in 13 656 participants from two cohort studies.
Main outcome measures Cognitive stimulation was measured at baseline using standard questionnaire instruments on active versus passive jobs and at baseline and over time using a job exposure matrix indicator. 4953 proteins in plasma samples were scanned. Follow-up of incident dementia varied between 13.7 to 30.1 years depending on the cohort. People with dementia were identified through linked electronic health records and repeated clinical examinations.
Results During 1.8 million person years at risk, 1143 people with dementia were recorded. The risk of dementia was found to be lower for participants with high compared with low cognitive stimulation at work (crude incidence of dementia per 10 000 person years 4.8 in the high stimulation group and 7.3 in the low stimulation group, age and sex adjusted hazard ratio 0.77, 95% confidence interval 0.65 to 0.92, heterogeneity in cohort specific estimates I2=0%, P=0.99). This association was robust to additional adjustment for education, risk factors for dementia in adulthood (smoking, heavy alcohol consumption, physical inactivity, job strain, obesity, hypertension, and prevalent diabetes at baseline), and cardiometabolic diseases (diabetes, coronary heart disease, stroke) before dementia diagnosis (fully adjusted hazard ratio 0.82, 95% confidence interval 0.68 to 0.98). The risk of dementia was also observed during the first 10 years of follow-up (hazard ratio 0.60, 95% confidence interval 0.37 to 0.95) and from year 10 onwards (0.79, 0.66 to 0.95) and replicated using a repeated job exposure matrix indicator of cognitive stimulation (hazard ratio per 1 standard deviation increase 0.77, 95% confidence interval 0.69 to 0.86). In analysis controlling for multiple testing, higher cognitive stimulation at work was associated with lower levels of proteins that inhibit central nervous system axonogenesis and synaptogenesis: slit homologue 2 (SLIT2, fully adjusted β −0.34, P<0.001), carbohydrate sulfotransferase 12 (CHSTC, fully adjusted β −0.33, P<0.001), and peptidyl-glycine α-amidating monooxygenase (AMD, fully adjusted β −0.32, P<0.001). These proteins were associated with increased dementia risk, with the fully adjusted hazard ratio per 1 SD being 1.16 (95% confidence interval 1.05 to 1.28) for SLIT2, 1.13 (1.00 to 1.27) for CHSTC, and 1.04 (0.97 to 1.13) for AMD.
Conclusions The risk of dementia in old age was found to be lower in people with cognitively stimulating jobs than in those with non-stimulating jobs. The findings that cognitive stimulation is associated with lower levels of plasma proteins that potentially inhibit axonogenesis and synaptogenesis and increase the risk of dementia might provide clues to underlying biological mechanisms.
Introduction
Cognitive stimulation has been hypothesised to help preserve cognitive function and decrease the risk of dementia in old age.1 Studies to date that have focused on cognitive stimulation in adulthood, however, have been small and with insufficient control for confounding factors and have failed to produce compelling evidence of benefits.23 Trials, typically based on relatively small sample sizes and short term interventions, have reported inconsistent results.456 Most of the recent long term cohort studies have suggested that leisure time cognitive activity does not reduce the risk of dementia either.789 A progressive reduction occurs in cognitive activities before dementia onset, but this seems to be because the gradual onset of dementia results in inactivity rather than lower cognitive activity being related to dementia.78
It is unclear whether the reason for modest findings is that the decrease in brain plasticity with age prevents cognitive activities across adult life from conferring protection against dementia, or, in the case of interventions, that the cognitive stimulation studied has not been intensive or engaging enough to preserve cognitive function. Work roles might be useful in the study of these associations. Exposure to cognitive stimulation at work can extend over decades and amount to tens of thousands of hours, so this stimulation lasts considerably longer than cognitive interventions or, typically, cognitively stimulating hobbies. According to the demand-control model, cognitively stimulating “active” jobs include demanding tasks and high job decision latitude (also known as job control).1011 Non-stimulating “passive” jobs are those with low demands and lack of job control. The combination of high demands and low control, in turn, characterises stressful work or job strain, which might be a risk rather than a protective factor for dementia.1213
In this report from the IPD-Work consortium (individual participant data meta-analysis in working populations),14 a large ongoing multicohort study of work and health, we examined the association between cognitively stimulating work and subsequent risk of dementia, controlling for established dementia risk factors and job strain. To identify potential biological pathways for this association, we performed a data driven analysis of 4953 plasma proteins, measured using slow off-rate modified aptamers.15 Plasma proteins are relevant targets for mechanistic study because they are affected by environmental exposures and serve many functions in regeneration, degeneration, and disease.151617
Methods
Study populations
Established at the Four Centres meeting in London in 2008, the IPD-Work consortium is a collaborative research project of 13 European cohort studies, which aims to estimate associations between work related factors and chronic diseases, disability, and mortality. To meet this goal, the consortium uses predefined exposure definitions (to minimise selective reporting) and large pooled datasets (to allow confirmation of findings across subgroups and, in the case of null findings, to show and publish absence of associations convincingly).
Seven of the 13 eligible cohort studies had relevant data to examine the association between cognitive stimulation at work and dementia incidence (analysis 1, see supplementary figure): the Finnish Public Sector study (FPS); GAZEL study, France; Health and Social Support (HeSSup) study, Helsinki Health Study (HHS), and Still Working study, Finland; the Whitehall II study, UK; and the Works, Lipids, and Fibrinogen (WOLF) Stockholm and Norrland studies, Sweden (fig 1). These studies comprised 107 896 men and women who were free of dementia at baseline (1986 to 2002), when cognitive stimulation and covariates were assessed. Follow-up for dementia was through to end of 2017.
To explore biological plausibility in a data driven analysis (analysis 2, see supplementary figure), we assessed 4953 proteins in plasma from a random sample of 2261 participants in one of the IPD-Work cohorts (the Whitehall II study)15 and examined associations between cognitive stimulation and plasma proteins (fig 1). Cognitive stimulation at work was assessed in 1991-93, and blood samples for the assessment of plasma protein were taken in 1997-99. The participants were free of dementia at the assessment of cognitive stimulation and plasma proteins.
In analysis 3 (see supplementary figure), we examined associations between proteins and dementia. As a complementary dataset to Whitehall II with follow-up for dementia until October 2019, we included a non-IPD-Work cohort, the multiethnic Atherosclerosis Risk in Communities (ARIC) study with 11 395 participants (fig 1).18 In ARIC, blood samples for protein analyses were drawn in 1993-95 and dementia follow-up was through to end of 2017. Cognitive stimulation at work was not measured in ARIC.
Table 1 summarises the study designs, participant numbers, and outcome ascertainment methods. Supplementary etables 1-8 show the characteristics of the cohorts.
Measurement of cognitive stimulation at work
We used predefined protocols to assess indicators of cognitive stimulation at work: job demands and job control. A description of the self-administered multi-item measures of these characteristics in each participating IPD-Work study and data harmonisation are available elsewhere and in the supplementary file (pp 13-14).19 Briefly, to minimise investigator bias, we validated and harmonised the job demand and control measures across participating cohort studies before extracting data for dementia, with investigators masked to outcome information. Questions in the job demands and job control scales had Likert-type response formats. We computed mean response scores for job demand items and for job control items for each participant. High job demand was defined as having a job demand score that was greater than the study specific median score; we defined high job control as having a job control score that was higher than or equal to the study specific median score. These categorisations are the originals and most used. We used the dichotomised measures to construct three categories of cognitive stimulation at work along the active-passive work dimension of Karasek’s demand-control model1011 in which both high demands and high control indicate higher stimulation. We defined low cognitive stimulation at work as low demands and low control, medium cognitive stimulation as high control and low demands or high demands and low control, and high cognitive stimulation as high demands and high control. The supplementary file (pp 13-14, efigure 1) provides a detailed description of the demand-control model.
To obtain an alternative continuous longitudinal measure unaffected by individual response style for supplementary analyses, we used a job exposure matrix indicator of cognitive stimulation that captures any changes of job and changes in level of stimulation over time. This measure was available for participants in the largest cohort study, the FPS. Data on participants’ occupational titles were obtained from Statistics Finland at baseline in 2000 and at follow-up in 2005, 2010, and 2015 and were categorised according to the International Standard Classification of Occupations (ISCO-88) by the International Labour Organisation (www.ilo.org/public/english/bureau/stat/isco/isco88/index.htm, accessed 3 July 2021). Using the three digit ISCO codes, we identified a total of 87 different occupations with at least 20 participants to determine the job axis for the job exposure matrix. The exposure axis was constructed by calculating the proportion of workers with high cognitive stimulation at work based on all responses within the same occupation. This occupation based exposure value was then assigned to each participant in the occupation (ranging from 0% for booking clerks to 62% for city mayors and other top administrators) in 2000, 2005, 2010, and 2015 or until the participant retired or was censored owing to dementia or death.
Covariates
Covariates were established dementia risk factors across the life course, as defined by the 2020 Lancet Commission on Dementia.1 These included participants’ age and sex, obtained from national or employers’ registers or self-reported. Education, a measure of cognitive stimulation in childhood, was self-reported (GAZEL, HHS, Still Working, Whitehall, WOLF, HeSSup, ARIC), or obtained from national registers (FPS).20 Education was categorised into low (primary or lower secondary), intermediate (higher secondary), and high (tertiary qualification, college, or university) levels.
Risk factors in adulthood included smoking (current versus former or never smoker), alcohol consumption (heavy (>14 units/week in women, >21 units/week in men versus other; one unit is approximately equivalent to 10 g of ethanol), physical activity (high versus low), obesity (body mass index ≥30 versus <30), hypertension (yes versus no), and prevalent diabetes (yes versus no).212223242526 Job strain (yes versus no) was measured with questions from the validated job content or demand-control questionnaire,19 and, as in other IPD-Work studies, was defined as high demands and low control, with all other combinations denoting no job strain.1427
Because people with cardiometabolic disease have increased risk of dementia, covariates also included diabetes (diagnostic codes E10 and E11 in the international classification of diseases, 10th revision; ICD-10), coronary heart disease (non-fatal myocardial infarction, ICD-10 codes I21–I22, or coronary death recorded as ICD-10 I20–I25), and stroke (ICD-10 codes I60, I61, I63, I64), including both prevalent cases at baseline and incident cases between baseline and before dementia diagnosis, assessed using linked records.
When possible we used additional covariates, available for specific cohorts only. Analyses of a longitudinal job exposure matrix indicator of cognitive stimulation (analysis 1) were based on the FPS study, with data on several additional dementia risk factors: social isolation (yes or no), depression (ICD-10 F32, F33), traumatic brain injury (ICD-10 S06), and atrial fibrillation (ICD-10 I48); the last three were identified from records of hospital admissions before dementia diagnosis. In protein analyses (analyses 2 and 3), we additionally adjusted effect estimates for ethnicity (white versus non-white) because more than 20% of the ARIC participants were non-white. In protein analyses based on the Whitehall II subcohort, APOE genotype (0, 1, or 2 of ε4 risk alleles), a strong predictor of dementia, was an additional covariate. Further information about additional covariates is available in the supplementary file (pp 14-15, etables 9-11).
Measurement of plasma proteins
Proteins were analysed using the SomaScan version 4 assay (Somalogic, CO).15 The analyses used plasma EDTA samples collected in 1997-99 (Whitehall II) or 1993-95 (ARIC) and stored at−80°C. Earlier studies and the supplementary file (p 16) describe in detail the performance of the SomaScan assay and the modified aptamer binding.282930 Briefly, the assay uses a mix of thousands of slow off-rate modified aptamers (SOMAmers). The aptamers bind to proteins in participants’ plasma samples and the specificity is ensured by a two step process analogous to a conventional immunoassay. The specificity of the aptamer reagents is good and has been tested in several ways.15 Median intra-assay and inter-assay coefficients of variation for SomaScan version 4 are about 5%, and assay sensitivity is comparable to that of typical immunoassays, with a median lower limit of detection in the femtomolar range.28
Follow-up for dementia
For the seven IPD-Work studies, including Whitehall II and its subcohort, we extracted data on dementia status at follow-up from hospital admissions records and death registries with any mention of dementia in the diagnosis, with or without reimbursements for medical treatment for dementia (Anatomical Therapeutic Chemical code N06D; table 1).20 Electronic records included exact date of diagnosis, death, or entitlement to reimbursement, and follow-up duration was measured as the difference between date of baseline examination and date of diagnosis, death, or entitlement to reimbursement. Although ascertainment of dementia from electronic health records underestimates prevalence, it has been shown to be a valid method in the study of associations between risk factors and dementia.313233 ICD-10 codes for all cause dementia were F00, F01, F02, F03, G30, and G31, with earlier ICD codes converted to ICD-10 codes (supplementary file p16). Codes F00 and G30 were used to define Alzheimer’s disease.
In the ARIC study, adjudicated people with dementia were primarily identified using data from ARIC clinic examinations conducted at visits 5 (2011-13) and 6 (2016-17). This included a neuropsychological battery using standardised protocols, with scores converted to z scores to assess change over time. Dementia was identified based on important decline on a serial cognitive battery, poor current test performance on a comprehensive neuropsychological battery, and impairments in activities of daily living based on informant rating on the clinical dementia rating scale and the functional activities questionnaire. A committee of clinicians then adjudicated participants with suspected dementia based on available cognitive and functional data. For participants who could not attend visits 5 or 6, those with dementia were identified through telephone cognitive assessment, plus surveillance of hospital discharge and death certificate codes related to dementia as well as screening during annual and semi-annual follow-up calls. Dementia suspected in participants who had died was identified through informant interviews.
Statistical analysis
We analysed the data in three parts (table 1). Cognitive stimulation at work was treated as a three level categorical variable, with low stimulation as the reference. The longitudinal job exposure matrix indicator of cognitive stimulation was analysed as a continuous exposure variable. In all three analyses, in multivariable models we adjusted effect estimates for age, sex, established dementia risk factors in childhood (education) and adulthood (smoking, heavy alcohol consumption, physical inactivity, job strain, obesity, hypertension, prevalent diabetes), and cardiometabolic diseases before dementia (ie, prevalent and incident diabetes, coronary heart disease, and stroke, treated as time dependent covariates). Before analysis, we imputed missing values in covariates using multiple imputation by substantive model compatible fully conditional specification.34 With data from all study variables, cohort indicator, and baseline year, 30 imputed datasets were created for the pooled dataset, Whitehall II random sample, and ARIC cohort.
Analysis 1 (cognitive stimulation-dementia)
To examine associations of cognitively stimulating work with dementia risk, we followed each participant from the date of cognitive stimulation assessment to the first record of dementia, death, or the end of follow-up. We used a two stage approach, including study specific analyses with Cox regression in the first stage and pooling the study specific estimates with random effects meta-analysis in the second. To estimate heterogeneity among the study specific estimates, we calculated I2 (higher values denoting greater heterogeneity).
To examine the robustness of the findings, we performed separate analyses for men and women, younger (<60 years at baseline) and older participants (≥60 years at baseline), and an alternative, repeatedly measured job exposure matrix indicator of cognitive stimulation at work, entered into the model as a time dependent exposure. We also performed the analysis separately for incident dementia during the first 10 years of follow-up (when assessment of cognitively stimulating work might become inaccurate in the preclinical or prodromal stage of dementia) and incident dementia from year 10 onwards in those without a dementia diagnosis at year 10. The assumption in analyses separating the assessment of cognitively stimulating work and dementia diagnosis by at least 10 years is that the cognitive stimulation-dementia association is less likely to be biased owing to reverse causation. To examine whether the association between cognitive stimulation and dementia depended on the method of dementia ascertainment, we stratified analyses by ascertainment method with the following categories: primary and specialised medical care (ascertainment based on self-report, prescription records, hospital admission, and death records) versus specialised medical care only (hospital admissions and death records). We also examined the association of cognitive stimulation with early onset (diagnosis aged <65) and late onset (diagnosis aged ≥65) dementia, Alzheimer’s disease, and non-Alzheimer’s disease dementia.
To address potential survival bias, we conducted a Fine and Gray competing risk analysis, with dementia and death as outcomes.35 In addition to the standard set of covariates (ie, age, sex, education, risk factors in adulthood, and cardiometabolic disease before dementia), we included cohort indicator as a covariate in multivariable adjusted analyses based on pooled data. Further covariates, available for analyses of the longitudinal job exposure matrix indicator of cognitive stimulation, included social isolation, depression, traumatic brain injury, and atrial fibrillation at baseline.
Analysis 2 (cognitive stimulation-proteins)
The distribution of many of the plasma proteins was skewed. We applied rank based inverse normal transformation to achieve normal distributions. Logistic regression was used to study the associations of proteins with high versus low cognitive stimulation and medium versus low cognitive stimulation and were expressed as odds ratios per 1 standard deviation higher level of protein. Tests of statistical significance were corrected for multiple comparison using the Bonferroni method for 4953 tests (P<1.0×10−5). In addition to adjustments for the standard set of covariates, we adjusted the effect estimates for ethnicity and APOE genotype.
Analysis 3 (proteins-dementia)
Cox proportional hazards models were used to study associations between proteins that survived Bonferroni correction in analysis 2 and dementia. In the two cohorts, hazard ratios were computed for 1 standard deviation higher protein level, adjusting for age, sex, ethnicity, and the standard set of covariates. We conducted a Fine and Gray competing risk analysis to address potential survival bias.35 Cohort specific effect estimates were pooled using fixed effect meta-analysis.
Post hoc analyses
To examine the effect of cognitive stimulation across the life course, we created a life course measure of cognitive stimulation by combining education and cognitive stimulation at work into a single variable, including the categories low education-low cognitive stimulation (reference), low education-high cognitive stimulation, high education-low cognitive stimulation, and high education-high cognitive stimulation. Here the low category included the low and medium categories of the original education and cognitive stimulation measures. We used Cox regression to assess the age, sex, and cohort adjusted association between cognitive stimulation across the life course and incident dementia.
We used SAS (version 9.4) to analyse associations in each cohort and in the pooled data in analysis 1, Stata (version 16.1) for multiple imputations, meta-analyses combining cohort specific estimates, and protein analyses. The supplementary file (pp 26-33) shows the statistical code.
Patient and public involvement
This is a secondary analysis of pre-existing datasets. No patients were involved in setting the present research question, setting the outcome measures, or developing plans for recruitment, design, or implementation of the study. No patients were asked to advise on the interpretation, but we have received feedback from a patient reviewer of The BMJ. The dissemination plan targets a wide audience, including members of the public, patients, health professionals, and experts in the specialty through various channels: written communication, events and conferences, networks, and social media.
Results
Overall, 2488 (2.3%) of 110 394 eligible participants who had missing data for cognitive stimulation or incident dementia and 10 (<0.1%) with a diagnosis of Alzheimer’s disease or other dementia before the study baseline were excluded from analyses. Of 107 896 participants included in the cognitive stimulation-dementia analysis (table 2), 45 080 (41.8%) were men and 62 816 (58.2%) were women, with a mean age of 44.6 (SD 9.5) years, at baseline. Of the participants, 29 243 (27.1%) had low cognitive stimulation at work, 50 724 (47.0%) had medium stimulation, and 27 929 (25.9%) had high stimulation.
Cognitive stimulation-dementia risk
Mean follow-up for dementia varied between 13.7 and 30.1 years depending on the cohort and was 16.7 (SD 4.9) years in the total sample. During 1 801 863 person years at risk, 1143 participants had a diagnosis of dementia between the ages of 42 and 93 years, mean 71.2 (SD 7.9) years (supplementary efigure 2). Cumulative hazards of dementia by age and level of cognitive stimulation at baseline showed a separation in dementia occurrence between high and low stimulation groups such that the cumulative incidence of dementia seen in participants who had high cognitive stimulation at work was observed at a younger age in those who had low stimulation at work (fig 2). For example, the cumulative incidence at age 75 in the high stimulation group (3.1%) was already observed at age 74 in the low stimulation group, and the incidence at age 80 (8.8%) was already observed at age 78.3, respectively. No difference in dementia was observed between participants with medium and low cognitive stimulation.
Meta-analysis of the seven cohort studies confirmed the association between high cognitive stimulation and lower risk of dementia in later life (fig 3). Incidence of dementia per 10 000 person years was 7.3 in the low cognitive stimulation group and 4.8 in the high stimulation group, the corresponding age and sex adjusted hazard ratio being 0.77 (95% confidence interval 0.65 to 0.92). This finding showed no significant heterogeneity across the cohorts (I2=0%, P=0.99) and the pattern of results was similar in the five cohorts with more than 25 participants with dementia (supplementary efigure 3). The association between higher cognitive stimulation and lower dementia risk was slightly weaker than that between education and dementia (hazard ratio 0.66, 95% confidence interval 0.55 to 0.79, supplementary etable 12), but the association was statistically significant and robust in analyses stratified by sex, age, and method of dementia ascertainment, before and after exclusion of participants with early onset dementia to minimise reverse causation, and before and after adjustment for established dementia risk factors in childhood and adulthood and the competing risk of death. The association was not mediated by prevalent or incident cardiometabolic diseases, although these diseases were related to increased dementia risk (age and sex adjusted hazard ratio 1.45, 95% confidence interval 1.23 to 1.71 for diabetes, 1.36, 1.08 to 1.72 for coronary heart disease, and 1.90, 1.51 to 2.38 for stroke). There was an indication that the cognitive stimulation-dementia association was attributable to Alzheimer’s diseases rather than to other dementias (P=0.06).
Πρόσφατα Σχόλια