Abstract
Background
Social relations are important for health, particularly at older ages. We examined the salience of frequency of social contacts and marital status for cognitive ageing trajectories over 21 years, from midlife to early old age.
Methods
Data are from the Whitehall II cohort study, including 4290 men and 1776 women aged 35–55 years at baseline (1985–88). Frequency of social contacts and marital status were measured in 1985–88 and 1989–90. Assessment of cognitive function on five occasions (1991–94, 1997–99, 2003–04, 2007–09 and 2012–13) included the following tests: short-term memory, inductive reasoning, verbal fluency (phonemic and semantic) and a combined global score. Cognitive trajectories over the study period were analysed using longitudinal latent growth class analyses, and the associations of these latent classes (trajectory memberships) with social relations were analysed using multinominal logistic regression.
Results
More frequent social contacts [relative risk (RRR) 0.96, 95% confidence interval (CI) 0.94 – 0.98] and being married (RRR 0.70, 95% CI 0.58 – 0.84) were associated with lower probability of being on a low rather than high cognitive performance trajectory over the subsequent 21 years. These associations persisted after adjustment for covariates. Of the sub-tests, social relations variables had the strongest association with phonemic fluency (RRR 0.95, 95% CI 0.94 – 0.97 for frequent contact; RRR 0.59, 95% CI 0.48 – 0.71 for being married).
Conclusions
More frequent social contacts and having a spouse were associated with more favourable cognitive ageing trajectories. Further studies are needed to examine whether interventions designed to improve social connections affect cognitive ageing.
Keywords: Cognitive ageing, longitudinal, cohort study, Public sector, Social network
Key Messages
Baseline cognitive performance was not strongly associated with subsequent rate of cognitive decline.
Frequent social contacts and being married were associated with better cognitive performance trajectory over time.
The associations between social relations and cognitive performance trajectories were independent of age, gender, socioeconomic position, behavioural risks and cardiometabolic factors.
Background
People with dense, good and supportive social relations are healthier and they live longer than those without such relations.1–3 The effects of social relations on health are attributable to multiple pathways, including stress related cardiometabolic changes and health risk behaviours.4 Both better structural (the size, frequency of contacts and structure of network, e.g. marital status) and functional (frequency or amount of social support) aspects of social relations have been associated with better cognitive performance5 and slower cognitive decline.6,7 More frequent social contacts have also been associated with reduced risk of dementia.6
Although there are a number of studies that have examined the association between social support (functional aspects of social relations) and cognitive performance using cross-sectional8–10 and longitudinal designs,8 only two time points of cognitive performance measures have typically been used. This is a limitation due to the fluctuation in cognitive performance, which could be especially marked at older age. Moreover, most longitudinal studies have used linear mixed-effects models,11,12 which models correlated repeated measures with random effects, to allow individuals differences in both cognitive scores at baseline and rates of cognitive decline. However, this strategy does not take into account the possibility that certain groups of individuals may have different developmental trajectories. An alternative approach, namely latent class modelling (a semi-parametric specification of mixed modelling)13 allows the underlying, unobserved characteristics of groups of individuals to be taken into account, with identification of homogeneous subgroups within the study population.
To address these limitations, the aim of the current study was to examine, using latent class modelling, the association between structural aspects of social relations, i.e. frequency of social contact and marital status, with cognitive performance. We used five assessments of cognitive performance spanning 21 years, and examined the associations with specific cognitive abilities.
Methods
Study sample
Participants were from the ongoing Whitehall II Study,14,15 which originally included 10 308 London-based civil servants from 20 civil service departments, who were 35–55 years of age at study baseline (phase 1: data collection took place between 1985 and 1988). Data from phase 1 (baseline, 1985–88), phase 2 (1989–90) and five follow-up phases 3 (1991–94), 5 (1997–99), 7 (2003–04), 9 (2007–09) and 11 (2011–13) were used in this study. All participants who provided data at baseline and at the second follow-up phase (n = 7578), and at any subsequent follow-up phases (n = 6072; 4290 men, 1776 women) were included in the analyses. Ethical approval for the Whitehall II Study was obtained from the University College London Medical School Committee on the ethics of human research, and informed consent was obtained from all study participants.
Structural aspects of social relations
Self-reported frequency of social contacts from phase 2 (1989/90), and marital status were used as structural measures of social relations. Frequency of social contacts score (0–28) was adapted from the Berkman/Syme social network index,16 and is the sum of all the items assessing structural aspects of social contacts (i.e. questions on the frequency of contacts with relatives, friends and colleagues, the frequency of participation in social or religious activities and the total number of relatives or friends seen once a month or more) and reflects amount and frequency of social connectedness. The index measures the network structure (how many peoples are there in the individual’s social network) and network interactions (how frequently the individual is in contact with people in their social network). Marital status was dichotomized as married/cohabiting vs unmarried (including never married, separated, divorced or widowed).
Cognitive performance
Cognitive testing was introduced to the Whitehall study midway through phase 3 (1991/94). Consequently, cognitive data are available only for 40% of the participants at phase 3 but for the entire sample at phases 5, 7, 9 and 11. The cognitive test battery comprised four cognitive tests to assess different cognitive abilities and was administered at five clinical examinations over 21 years (1991/94 to 2012/13), as follows.
The Alice Heim 4-I (AH4-I)17 is composed of a series of 65 verbal and mathematical reasoning items of increasing difficulty, to be completed in 10 min.
Short-term verbal memory was assessed with a 20-word free recall test. Participants were presented with a list of 20 one- or two-syllable words at 2-s intervals and then had to recall them in writing in 2 min.
There were two tests of verbal fluency.18 Participants were asked to recall in writing as many ‘S’ words (phonemic fluency) and as many animal names as they could (semantic fluency) in 1 min.
The four cognitive tests were combined to create a global cognitive z-score [mean 0, standard deviation (SD) = 1], to minimize problems due to measurement error on individual tests. First, the raw scores from each test were standardized to a z-score based on the phase 5 mean value and SD, and then these z scores were averaged to obtain the global cognitive performance scores. Similar procedure has been used in previous studies.11
Covariates
Age, sex, ethnicity (White, non-White) and socioeconomic status, measured as occupational position (low, intermediate, high) were reported at the study baseline and were used as covariates in all analyses. Occupational position has been shown to be a broad marker of socioeconomic status in the Whitehall II study, as it has been associated with salary, educational level and the level of responsibility at work.15,19 To control for potential confounding and mediating factors, we included an inflammatory marker, level of C-reactive protein (CRP); for cardiometabolic risk factors, we used diastolic and systolic blood pressure, low-density lipoprotein (LDL) cholesterol and fasting glucose; and for behavioural risks, we included alcohol consumption and body mass index. All of these factors have been associated with cognitive function or cognitive decline previously.12,20–26 The methods used for measuring these variables have been reported in detail elsewhere.14
Statistical analysis
Trajectories of cognitive performance were defined using group-based trajectory models (GBTM) that identify classes of individuals (trajectory groups) with a similar trajectory over time (a special case of latent class analyses). GBTM is increasingly being applied to clinical research to map the developmental course of disease and to identify the number, shape and size of different (latent) trajectory groups in the data. We used Bayesian information criteria (BIC) and Akaike’s information criterion (AIC) to determine the optimal number of trajectories: lower absolute values correspond to better fit. We hypothesized a priori that there would be two to three latent trajectories, as suggested by previous research.27,28 In addition to global cognitive score, the change in each individual cognitive performance test score (memory, inductive reasoning, and phonemic and semantic fluency) were analysed using GBTM. The latent class modelling is a semi-parametric specification of mixed modelling, which approximates the underlying continuous distribution with a discrete distribution.13 The underlying theory of the latent class modelling posits that individual behaviour depends on observable attributes and on latent heterogeneity that varies with factors that are unobserved by the analyst, and heterogeneity is analysed through a model of discrete parameter variation. Each individual was classified as being a member of a given trajectory group based on a posterior classification index for each individual, i.e. the mean probability of being assigned to the given class. The method includes participants with data from any of the data collection phases.
We used multinomial ordinal regression analysis and expressed the results as relative rate ratios (RRR) and their 95% confidence intervals (CI), where all three trajectory groups were analysed together. First we tested proportional odds assumption using the likelihood ratio chi-square test, and ‘high’ trajectory was set as an outcome reference group against which we tested the association with social relations variables. The estimates were adjusted for (i) age and sex, (ii) additionally for socioeconomic status and (iii) additionally for cardiometabolic risk factors (including CRP) and health behaviours. All the analyses were performed using STATA 13.1 statistical package.
Results
When compared with those who dropped out from the original sample, participants included in the study sample were more likely to be White (79% vs 62%; P < 0.001), men (81% vs 70%; P < 0.001), slightly younger (mean age in years: 44.7 vs 45.8, P < 0.001) and from a higher employment position (85% vs 60%; P < 0.001). In addition, individuals who were included into the study sample had a higher frequency of social contacts (10 vs 9.3 P < 0.001), were more likely to be married or cohabiting (79% vs 73%; P < 0.001) and their blood glucose levels were lower (5.5 mmol/l vs 6.5 mmol/l; P < 0.001) than those who dropped out from the study.
A three-trajectory solution of global cognitive function scores with non-linear trajectories yielded better fit (BIC = −18 848.93/AIC = −18 788.28) than two-trajectory (BIC = −21 667.40/AIC = −21 626.97) or linear solutions (BIC = −19 139.29/AIC = −19 102.90). The four-class solution was slightly better than the three-class model (BIC = −17 932.12, AIC = −17 883.60), but one of the classes would have been relatively small (12%), and thus we chose the three–factor solution with more even distribution of the participants (Supplementary Table 1S, available as Supplementary data at IJE online). Figure 1 shows that each trajectory had a slightly declining shape over time. Predicted probabilities of group membership totalled 20% with ‘low’, 49% with ‘intermediate’ and 31% with ‘high’ global cognitive function trajectory. Trajectories of cognitive performance subscales are presented in Supplementary Figures 1–4, available as Supplementary data at IJE online. There were clear differences in the level (intercept) of the cognitive performance trajectories. There were also small differences in the slope between the trajectory groups; the steepest decline was in the ‘low’ and the mildest decline in the ‘high’ group. Similar pattern was found in all performance subscales; the biggest decline was found in the low-performance group.
Figure 1.
Trajectories of global cognitive performance score over five study phases (from 1991/94 to 2012/13 (n = 6072).
Description of the study sample according to global cognitive performance trajectory membership is shown in Table 1. There were clear differences in cognitive function trajectory memberships between various demographic categories, health behaviour categories and most cardiometabolic risk factors. Participants in the ‘high’ cognitive trajectory group were younger at baseline and more likely to be male, White and have high socioeconomic status. The mean BMI, systolic blood pressure, fasting glucose, CRP and alcohol consumption were lower in the ‘high’ group.
Table 1.
Descriptive characteristics of the participants by trajectories of global cognitive performance (n = 6072) at the end of follow-up (phase 7)
| Trajectory of cognitive performance |
||||
|---|---|---|---|---|
| Low n = 1174 | Intermediate n = 3030 | High n = 1868 | P-value for heterogeneity | |
| n (%) | n (%) | n (%) | ||
| Sex: male | 619 (14%) | 2, 284 (53%) | 1, 391 (33%) | |
| female | 555 (31%) | 746 (42%) | 477 (27%) | <0.001 |
| Socioeconomic status: high | 79 (4%) | 965 (47%) | 980 (49%) | |
| intermediate | 494 (16%) | 1701 (56%) | 846 (28%) | |
| low | 601 (50%) | 364 (36%) | 42 (4%) | <0.001 |
| Ethnicity: White | 802 (15%) | 2874 (52%) | 1852 (33%) | |
| other | 371 (69%) | 151 (28%) | 16 (3%) | <0.001 |
| Mean (SE) | Mean (SE) | Mean (SE) | ||
| Age (years) | 64.2 (0.2) | 61.5 (0.1) | 58.5 (0.1) | <0.001 |
| Body mass index (kg/m2) | 27.4 (0.1) | 26.6 (0.1) | 26.3 (0.1) | <0.001 |
| Alcohol consumption past 1 month (range 1 > 2/day to 6) | 3.7 (0.0) | 2.9 (0.0) | 2.6 (0.0) | <0.001 |
| C-reactive protein (mg/l) | 3.3 (0.2) | 2.4 (0.1) | 2.2 (0.1) | <0.001 |
| Fasting glucose (mmol/l) | 5.6 (0.0) | 5.4 (0.0) | 5.4 (0.0) | <0.001 |
| LDL-cholesterol (mmol/l) | 3.4 (0.0) | 3.5 (0.0) | 3.5 (0.0) | 0.007 |
| Diastolic blood pressure (mmHg) | 69.8 (0.4) | 70.9 (0.2) | 70.9 (0.3) | 0.012 |
| Systolic blood pressure (mmHg) | 126.5 (0.6) | 125.3 (0.3) | 123.4 (0.4) | <0.001 |
SE, standard error.
The age-, sex-, ethnicity- and socioeconomic status-adjusted associations between structural social relations (network density and marital status) and cognitive function trajectory group memberships are presented in Table 2, first showing the risk ratio for membership in the ‘low’ cognitive trajectory and then showing the risk ratio for membership in the ‘intermediate’ cognitive trajectory, setting ‘high’ trajectory membership as a reference group in multinomial regression analyses. Higher frequency in social contacts and being married were both associated with less likelihood of being in the ‘low’ or ‘intermediate’ global cognitive function trajectory as compared with being on the ‘high’ trajectory. These associations were robust to adjustments for health behaviours, CRP and cardiometabolic factors. The corresponding analysis, in which individual cognitive function trajectory memberships were tested as outcomes, are presented in Table 3. Both social relations variables seemed to have a stronger association with fluency measures than with inductive reasoning. The trajectory of short-term memory performance was not associated either of the social relations measures.
Table 2.
Multinomial logistic regression analysis for the associations of social relations dimensions at phase 2 with trajectories of global cognitive performance from phase 3 until phase 11. Figures are relative risk ratios (RRR) and 95% confidence intervals (95% CI)
| Trajectory of cognitive performance |
||
|---|---|---|
| Social relations dimensions | ‘Low (ref., vs high’) | ‘Intermediate vs high’ |
| RRR (95% CI) | RRR (95% CI) | |
| Frequency of social contacts (one point increase on social network scale) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.96 (0.94 – 0.98) | 0.98 (0.96 – 0.99) |
| Adjusted additionally for health behaviours a | 0.96 (0.93 – 0.98) | 0.98 (0.96 – 0.99) |
| Adjusted additionally for cardiometabolic factorsb | 0.96 (0.93 – 0.99) | 0.98 (0.96 – 1.00) |
| Marital status (married) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.70 (0.58 – 0.84) | 0.77 (0.66 – 0.89) |
| Adjusted additionally for health behaviours a | 0.75 (0.59 – 0.96) | 0.88 (0.74 – 1.04) |
| Adjusted additionally for cardiometabolic factorsb | 0.76 (0.58 – 1.00) | 0.88 (0.74 – 1.06) |
aAlcohol consumption and BMI.
bC-reactive protein, glucose, LDL-cholesterol, diastolic and systolic blood pressure.
Table 3.
Multinomial logistic regression analysis for the associations of social relations dimensions at phase 2 with trajectories of individual cognitive performance tests from phase 3 until phase 11. Figures are relative risk ratios (RRR) and 95% confidence intervals (95% CI)
| Trajectory of cognitive performance |
||
|---|---|---|
| Social relations dimensions | ‘Low vs high’ | ‘Intermediate vs high’ |
| RRR (95% CI) | RRR (95% CI) | |
| Memory | ||
| Frequency of social contacts (one point increase on social network scale) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.98 (0.95 – 1.00) | 1.00 (0.98 – 1.03) |
| Adjusted additionally for health behaviours | 0.98 (0.95 – 1.01) | 1.01 (0.98 – 1.01) |
| Adjusted additionally for cardiometabolic factors | 0.99 (0.96 – 1.02) | 1.01 (0.98 – 1.04) |
| Marital status (married) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.80 (0.64 – 1.00) | 0.86 (0.69 – 1.07) |
| Adjusted additionally for health behaviours | 0.84 (0.65 – 1.08) | 0.88 (0.70 – 1.12) |
| Adjusted additionally for cardiometabolic factors | 0.88 (0.67 – 1.16) | 0.92 (0.72 – 1.18) |
| Inductive reasoning | ||
| Frequency of social contacts (one point increase on social network scale) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.97 (0.95 – 0.98) | 0.99 (0.97 – 1.00) |
| Adjusted additionally for health behaviours | 0.95 (0.92 – 0.98) | 0.99 (0.98 – 1.01) |
| Adjusted additionally for cardiometabolic factors | 0.95 (0.92 – 0.98) | 0.99 (0.97 – 1.01) |
| Marital status (married) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.80 (0.67 – 0.97) | 0.80 (0.70 – 0.92) |
| Adjusted additionally for health behaviours | 0.88 (0.69 – 1.14) | 0.98 (0.84 – 1.16) |
| Adjusted additionally for cardiometabolic factors | 0.90 (0.68 – 1.19) | 0.98 (0.82 – 1.16) |
| Phonemic fluency | ||
| Frequency of social contacts (one point increase on social network scale) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.95 (0.94 – 0.97) | 0.98 (0.96 – 0.99) |
| Adjusted additionally for health behaviours | 0.95 (0.93 – 0.98) | 0.98 (0.96 – 1.00) |
| Adjusted additionally for cardiometabolic factors | 0.96 (0.94 – 0.99) | 0.99 (0.97 – 1.01) |
| Marital status (married) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.59 (0.48 – 0.71) | 0.75 (0.62 – 0.89) |
| Adjusted additionally for health behaviours | 0.64 (0.51 – 0.80) | 0.77 (0.63 – 0.94) |
| Adjusted additionally for cardiometabolic factors | 0.60 (0.47 – 0.77) | 0.78 (0.64 – 0.97) |
| Semantic fluency | ||
| Frequency of social contacts (one point increase on social network scale) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.96 (0.94 – 0.98) | 0.98 (0.96 – 0.99) |
| Adjusted additionally for health behaviours | 0.96 (0.94 – 0.98) | 0.98 (0.96 – 1.00) |
| Adjusted additionally for cardiometabolic factors | 0.96 (0.94 – 0.99) | 0.99 (0.97 – 1.01) |
| Marital status (married) | ||
| Adjusted for age, sex, ethnicity and socioeconomic status | 0.71 (0.59 – 0.86) | 0.77 (0.66 – 0.91) |
| Adjusted additionally for health behaviours | 0.79 (0.63 – 1.00) | 0.89 (0.74 – 1.06) |
| Adjusted additionally for cardiometabolic factors | 0.81 (0.63 – 1.04) | 0.89 (0.74 – 1.08) |
We additionally tested the associations using mixed models with frequency of social contacts and marital status as predictors (separate analyses) and overall cognitive performance over all study phases as an outcome in analyses adjusted for age and sex. Both more frequent social contacts (B = 0.01, z = 4.27, P < 0.001) and being married (B = 0.06, z = 3.02, P = 0.002) were associated with better cognitive function over study periods, and frequency of social contacts * time interaction (z = −2.39, P = 0.017) was statistically significant. This interaction remained statistically significant (z = 2.63, P = 0.010) when adjusted additionally for socioeconomic status, alcohol consumption, body mass index, CRP, diastolic and systolic blood pressure, LDL cholesterol and fasting glucose. Both the frequency of social contacts (z = 3.45, P = 0.001) and being married (z = 2.42, P = 0.016) were associated with better cognitive function over study periods when added in to the regression model simultaneously, but no interaction between them was found (z = 1.56, P = 0.119).
Discussion
Three trajectory groups of global cognitive function were identified among middle-aged men and women, with up to five repeated measurements of cognitive performance over a maximum follow-up of 21 years. There were clear differences in intercepts of cognitive performance between trajectory groups, but relatively small differences in the steepness or shape of the slopes. Thus, the baseline level of cognitive performance did not have a very strong association with the rate of cognitive decline. Of the participants, 20% belonged to the ‘low’, 49% to the ‘intermediate’ and 31% to the ‘high’ global cognitive function trajectory group. Similar trajectories have also been previously identified.28 Our findings showed that having more frequent social contacts and being married were both associated with a better cognitive performance trajectory over time. The associations were independent of demographic, behavioural and cardiometabolic factors. Of the individual cognitive performance trajectories, all but short-term memory performance were better in people with more frequent social contacts and among those who were married. Taken together, the findings suggest that less social contacts and living without a partner predict poorer cognitive performance between midlife and old age. Particularly availability of other people, as indicated by frequent social contacts or living with a partner, seems to be a factor associated with cognitive decline.
Current findings are in line with previous studies where structural aspects of social relations have been associated with better average levels of cognitive performance and slower cognitive decline.6,7 Our results also show that decline in different cognitive abilities is similar, which is also in line with previous findings.29 Associations between structural aspects of social relations and various cognitive domains were similar, indicating that lack of social contacts is a risk factor for general cognitive ageing. We did find a stronger association between structural aspects of social relations and verbal ability. Married people have been shown to have more frequent contact with their social networks,30 and this engagement may directly strengthen cognitive ability through repeated practice and refinement of communication, which could explain this finding.
Our study adds new insight by using a relatively large dataset and repeat measures of cognitive performance, starting in midlife. Our approach to modelling risk factors for cognitive function, using long-term trajectory modelling of cognitive performance using the GBTM, is new. In previous studies, cognitive performance has been often examined using cross-sectional data, shorter follow-up time or analysing cognitive decline with mixed modelling11,12 that models correlated repeated measures with random effects, to allow individuals differences in both cognitive scores at baseline and rates of cognitive decline. This strategy does not take into account the possibility that certain categories of individuals may have different developmental trajectories. GBTM, a semi-parametric specification of mixed modelling,13 allows the underlying, unobserved characteristics of groups of individuals to be taken into account, with identification of homogeneous subgroups within the study population.
The large sample size, long follow-up period and multiple waves of cognitive assessment strengthen confidence in the results. Several known confounding factors were included in the analysis. Combining four tests into a single measure of global cognition reduces measurement error. All participants were from a sample of basically white-collar employees, and that restricts the generalizability of our results. However, the cohort covers a wide socioeconomic range, with a large difference in full-time salary between the highest and lowest occupational grade. The measures of social relationships were self-reported, so the information may be biased by respondents’ personality traits.31 The shape of the trajectories suggests that there were practice effects at the second time point for all three trajectories. As the effect appeared to be of a similar magnitude in all trajectories, this was unlikely to have affected our results. Subjective perceptions of the social environment, derived from a well-validated questionnaire, are relevant indictors of social relationships and have been shown to associate with various health outcomes.32–34
However, as this study was an observational study, further studies are needed to examine whether interventions designed to improve possibilities for social connections would affect cognition favourably. To form a comprehensive picture of the role of social relations in cognitive performance, future studies also should examine the roles of both structural and functional aspects of social relations.
We studied only the associations between midlife social relationships and subsequent cognitive performance development, to reduce the risk of reverse causality affecting our results. In future, the impact of social relationship changes during the time from middle age to old age on cognitive performance should be investigated. Our results emphasize the importance of structural aspects of social relations in relation to cognitive function.
Supplementary Data
Supplementary data are available at IJE online.
Funding
The Whitehall II Study is supported by grants from the Medical Research Council (K013351), the British Heart Foundation, the National Heart Blood and Lung Institute, National Institutes of Health (NIH) (R01HL036310), and the National Institute of Aging, National Institutes of Health (NIH) (R01AG013196 and R01AG034454). M.E. and M.V. are supported by the Academy of Finland (265977, 292824, 258598), A.S. by the Wellcome Trust (200163/Z/15/Z) and the UCL Hospitals Biomedical Research Centre and M.K. by the UK Medical Research Council (K013351), NordForsk, the Nordic Council of Ministers (grant 75021) and the Academy of Finland (311492). The funders were not involved in the design aor conduct of the study, the collection, management, analysis or interpretation of the data or the preparation, review or approval of the manuscript.
Conflict of interest: None declared.
Supplementary Material
References
- 1. Holt-Lunstad J, Smith TB, Baker M, et al. Loneliness and social isolation as risk factors for mortality: a meta-analytic review. Perspect Psychol Sci 2015;10:227–37. [DOI] [PubMed] [Google Scholar]
- 2. Holt-Lunstad J, Smith TB, Layton JB. Social relationships and mortality risk: a meta-analytic review. PLoS Med 2010;7:e1000316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Hakulinen C, Pulkki-Raback L, Jokela M, et al. Structural and functional aspects of social support as predictors of mental and physical health trajectories: Whitehall II cohort study. J Epidemiol Community Health 2016;70:710–15. [DOI] [PubMed] [Google Scholar]
- 4. Cacioppo JT, Hawkley LC, Crawford LE, et al. Loneliness and health: potential mechanisms. Psychosom Med 2002;64:407–17. [DOI] [PubMed] [Google Scholar]
- 5. Cacioppo JT, Hawkley LC. Perceived social isolation and cognition. Trends Cogn Sci 2009;13:447–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kuiper JS, Zuidersma M, Oude Voshaar RC, et al. Social relationships and risk of dementia: A systematic review and meta-analysis of longitudinal cohort studies. Ageing Res Rev 2015;22:39–57. [DOI] [PubMed] [Google Scholar]
- 7. Kuiper JS, Zuidersma M, Zuidema SU, et al. Social relationships and cognitive decline: a systematic review and meta-analysis of longitudinal cohort studies. Int J Epidemiol 2016;45:1169–206. [DOI] [PubMed] [Google Scholar]
- 8. Seeman TE, Lusignolo TM, Albert M, et al. Social relationships, social support, and patterns of cognitive aging in healthy, high-functioning older adults: MacArthur studies of successful aging. Health Psychol 2001;20:243–55. [DOI] [PubMed] [Google Scholar]
- 9. Eisele M, Zimmermann T, Kohler M, et al. Influence of social support on cognitive change and mortality in old age: results from the prospective multicentre cohort study AgeCoDe. BMC Geriatr 2012;12:9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Green A, Rebok G, Lyketsos C. Influence of social network characteristics on cognition and functional status with aging. Int J Geriatr Psychiatry 2008;23:972–78. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Liao J, Head J, Kumari M, et al. Negative aspects of close relationships as risk factors for cognitive aging. Am J Epidemiol 2014;180:1118–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Singh-Manoux A, Dugravot A, Brunner E, et al. Interleukin-6 and C-reactive protein as predictors of cognitive decline in late midlife. Neurology 2014;83:486–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Greene W, Hensher D. A latent class model for discrete choice analysis: contrasts with mixed logit. Transportation Res 2003;B37:681–98. [Google Scholar]
- 14. Marmot M, Brunner E. Cohort Profile: The Whitehall II study. Int J Epidemiol 2005;34:251–56. [DOI] [PubMed] [Google Scholar]
- 15. Marmot MG, Davey Smith G, Stansfeld S, et al. Health inequalities among British civil servants: the Whitehall II study. Lancet 1991;337:1387–93. [DOI] [PubMed] [Google Scholar]
- 16. Berkman LF, Syme SL. Social networks, host resistance, and mortality: a nine-year follow-up study of Alameda County residents. Am J Epidemiol 1979;109:186–204. [DOI] [PubMed] [Google Scholar]
- 17. Heim AW, Watts KP, Simmond V, Walters A. AHA4 Group Test for General Intelligence. Nashville, TN: Nelson, 1970. [Google Scholar]
- 18. Borkowski JG, Benton AL, Spreen O. Word fluency and brain damage. Neurophysiologia 1967;5:135–40. [Google Scholar]
- 19. Elovainio M, Ferrie JE, Singh-Manoux A, et al. Socioeconomic differences in cardiometabolic factors: social causation or health-related selection? Evidence from the Whitehall II Cohort Study, 1991–2004. Am J Epidemiol 2011;174:779–89. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Britton A, Singh-Manoux A, Marmot M. Alcohol consumption and cognitive function in the Whitehall II Study. Am J Epidemiol 2004;160:240–47. [DOI] [PubMed] [Google Scholar]
- 21. Horvat P, Richards M, Kubinova R, et al. Alcohol consumption, drinking patterns, and cognitive function in older Eastern European adults. Neurology 2015;84:287–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Sabia S, Elbaz A, Britton A, et al. Alcohol consumption and cognitive decline in early old age. Neurology 2014;82:332–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Singh-Manoux A, Britton AR, Marmot M. Vascular disease and cognitive function: evidence from the Whitehall II Study. J Am Geriatr Soc 2003;51:1445–50. [DOI] [PubMed] [Google Scholar]
- 24. Singh-Manoux A, Marmot M. High blood pressure was associated with cognitive function in middle-age in the Whitehall II study. J Clin Epidemiol 2005;58:1308–15. [DOI] [PubMed] [Google Scholar]
- 25. Singh-Manoux A, Richards M, Marmot M. Leisure activities and cognitive function in middle age: evidence from the Whitehall II study. J EpidemiolCommunity Health 2003;57:907–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Tuligenga RH, Dugravot A, Tabak AG, et al. Midlife type 2 diabetes and poor glycaemic control as risk factors for cognitive decline in early old age: a post-hoc analysis of the Whitehall II cohort study. Lancet Diabetes Endocrinol 2014;2:228–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Hayden KM, Reed BR, Manly JJ, et al. Cognitive decline in the elderly: an analysis of population heterogeneity. Age Ageing 2011;40:684–89. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Marioni RE, Proust-Lima C, Amieva H, et al. Social activity, cognitive decline and dementia risk: a 20-year prospective cohort study. BMC Public Health 2015;15:1089. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Salthouse TA. Selective review of cognitive aging. J Int Neuropsych Soc 2010;16:754–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Sarkisian NG. Does singlehood isolate or integrate? Examining the link between marital status and ties to kin, friends, and neighbors. J Soc Pers Relationships 2015;33:361–84. [Google Scholar]
- 31. Stansfeld S, Marmot M. Deriving a survey measure of social support – the reliability and validity of the Close Persons Questionnaire. Soc Sci Med 1992;35:1027–35. [DOI] [PubMed] [Google Scholar]
- 32. Stringhini S, Berkman L, Dugravot A, et al. Socioeconomic status, structural and functional measures of social support, and mortality: The British Whitehall II Cohort Study, 1985–2009. Am J Epidemiol 2012;175:1275–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Stansfeld SA, Fuhrer R, Shipley MJ. Types of social support as predictors of psychiatric morbidity in a cohort of British Civil Servants (Whitehall II Study). Psychol Med 1998;28:881–92. [DOI] [PubMed] [Google Scholar]
- 34. Stafford M, McMunn A, Zaninotto P, et al. Positive and negative exchanges in social relationships as predictors of depression: evidence from the English Longitudinal Study of Aging. J Aging Health 2011;23:607–28. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.