Background
Population ageing and related cognitive decline are global issues implying increased costs for governments, communities, families and individuals [
1]. The WHO estimates that around 55 million people have dementia worldwide, with nearly 10 million new cases every year. The total number of people with dementia and severe cognitive impairment is projected to reach 78 million in 2030 and 139 million in 2050 (see:
https://www.who.int/news-room/fact-sheets/detail/dementia). Dementia is a progressive and severely disabling condition often requiring intensive formal and informal home and/or institutional care [
2]. It also tends to cluster with other diseases increasing the risk of unplanned hospitalisation, longer in-hospital stay and re-admissions, as well as functional impairment [
3,
4]. The WHO estimates that the total global societal cost of dementia was US $ 1.3 trillion in 2019 but these costs are expected to surpass US$ 2.8 trillion by 2030, with half of them attributed to informal care (see:
https://www.who.int/news-room/fact-sheets/detail/dementia). Understanding prevention and protection mechanisms that can minimise the risk of dementia and postpone its onset in an ageing population is key to reduce pressure on health care systems and welfare institutions, as well as to improve the quality of life of both families and caregivers.
Although complex factors may lead to individual transitions from normal cognition to dementia, research has recently shown that lifestyle-related factors, such as physical inactivity, tobacco use, unhealthy diet and the harmful use of alcohol, as well as several cardiovascular and metabolic conditions, such as hypertension and diabetes, increase the risk of cognitive impairment [
5]. Recent studies have suggested that this risk greatly depends on the social context in which individuals are embedded [
6,
7]. The modulation function of social conditions includes the prominent role of social networks, i.e., direct and indirect contacts between individuals in which information, attitudes and norms are shared [
8‐
10]. For instance, a recent study showed that the network size and density, as well as the presence of weak ties (i.e., social bridging), moderate the association between brain atrophy and cognitive function, while marriage/cohabitation (i.e., social bonding) moderate the association between perceived stress and cognitive function [
11].
Therefore, complex social and psychological factors and their underlying biological mechanisms could affect the risk of dementia, as well as on its prevention and protection. One of the most convincing hypotheses is that social activity and social engagement of older adults may promote neuro-protection and compensation, including the beneficial effect of physical exercise on neuro-degeneration [
12]. For instance, previous research has shown that high level of social engagement and larger social networks are associated with better glucose regulation in adults without diabetes and better diabetes self-management that reduce the risk of dementia, thereby indicating a possible pathway that connects social relationships and cognitive abilities [
13].
Previous sociological research has identified various structural and functional aspects of social relationships that may have either direct or indirect effects on cognitive decline among older adults [
14‐
17]. Structural aspects typically include the individual network size (e.g., the number of frequent contacts, including family members, friends and acquaintances) and social activity (e.g., voluntary work, participation to community organisations, social clubs, neighbourhood associations) [
8]. Functional aspects of social relationships typically include sources of social support, including material help and emotional support [
18], and subjective perception of social integration against loneliness and social isolation [
19]. For instance, rich and diverse social relationships can allow individuals to access information instrumental for better prevention and protection [
6,
20]. Social relationships can also convey material and emotional support that can increase the capabilities of individuals to face critical events and processes related to ageing [
12,
21]. These aspects shape the ‘personal community’ [
22] of older adults conveying material and emotional resources that typically follow ego-specific social and spatial stratification and segmentation [
23‐
26]. Along these spatial and social fault lines, various ‘social foci’ exist that determine individual heterogeneity of opportunities and constraints [
27,
28], with potentially important implications on cognitive processes. For instance, loneliness and objective and subjective social isolation have a detrimental impact on the mental and physical health of older adults, which exceeds that of smoking 15 cigarettes per day or obesity [
29,
30].
While research on this intersection of social factors and cognitive decline is growing, findings are still controversial especially regarding the effect of different types of social relationships and the accuracy of estimations of causal relationships between social and cognitive factors [
31]. In a meta-analysis including relevant longitudinal cohort studies published until 2012, Kuiper and colleagues [
32] found that despite heterogeneity in study design and measures, multiple aspects of social relationships were associated with cognitive decline. However, due to various sources of possible bias in measurements and estimates, these statistical associations should be interpreted with caution. For instance, due to reverse causality between social and cognitive factors, the authors of the meta-analysis concluded that more careful study design was needed to assess findings more systematically and disentangle various sources of complexity.
Here, we first aimed to update the previous meta-analysis by extending it to all relevant publications from 2012 to 2020. Second, we performed a cumulative meta-analysis that allowed us to assess the temporal evolution of the statistical estimates performed in all studies, including those reported in the previous meta-analysis. This was key to provide a more informative picture of the robustness of measurements and methodologies used in this growing field of area (+80% of articles from 2012-2020 compared to the previous period). Improving methods and measurements also increases our capability of assessing causal relationships between social and non-social factors, thus improving the quality of research design and measurements. We also need to understand whether certain direct or indirect interventions on social factors could be effective to either postpone or reduce the effect of cognitive decline for the general public.
Methods
This systematic review and meta-analysis was pre-registered and the review protocol can be accessed at
https://www.crd.york.ac.uk/prospero/ (ID: CRD42019130667). Reporting followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 updated guidelines [
33].
Systematic search and study selection
We performed a systematic database search from Scopus and Web of Science (WOS) on 12 February 2019 using the same keywords and search design of the previous systematic review for all publications from 2012 to 2018. On 29 December 2020, the search was extended to all publications until 2020 by using the same search strings (see
Supplementary Material, Section 1).
A total of 16,502 entries were initially selected resulting in 10,460 unique articles. Two members of our team independently screened the titles and abstracts giving 175 eligible articles. Any disagreements were resolved in consensus meetings. Persistent disagreements were resolved by a final decision being made by two additional authors. Following criteria used in the previous systematic review [
32], articles were included if they: (i) were peer reviewed; (ii) reported an association between social relationships measured at baseline and the follow-up in a quantitative way; (iii) included a longitudinal prospective cohort study design conducted on the general population. Only articles published in English, German or French were included. Studies considering dementia as outcome were excluded. Note that we considered studies relying on samples including subjects living independently and community-dwelling.
Data collection, items, risk of bias
The same two authors involved in the screening, then independently extracted data used in the study, i.e., population characteristics, timing of follow-up, measurement of social relationships, measurement of cognitive functioning, statistical methods and results. Whenever possible, estimates adjusted for potential confounding factors were used for the meta-analysis. For the sake of consistency with the previous meta-analysis, we considered the following potential confounding factors: (1) age; (2) presence of depressive symptoms; (3) alcohol use; (4) education; (5) baseline cognition; and (6) physical functioning. This last included at least one or a combination of the following: (i) physical activity; (ii) functional disability; (iii) chronic diseases, such as traumatic brain injuries, cardiovascular disease or cerebrovascular accident.
The methodological quality of the included studies was assessed independently by the two reviewers who had screened all entries with the Quality of Prognosis Studies in Systematic Reviews (QUIPS) tool [
34]. The QUIPS tool includes six domains of possible bias that should be considered whenever evaluating the validity and bias in prognostic factors, each presented with prompting items and considerations. As regards the participation domain, we considered whether series of participants were consecutive and if participation was adequate compared to the initial number of recruited individuals. We evaluated study attrition according to data completeness with reference to the outcome and lack of differences between sample and dropout. We then assessed the validity of methods and the completeness of data to measure social relationships and cognitive abilities. We also considered whether the assessment of cognitive abilities and social relationships of participants was performed separately by different interviewers. We included measurement and inclusion in the analysis of any relevant confounding factors. We also included an item about minimisation of reverse causality by assessing whether the analysis was adjusted for baseline cognitive function or subjects with cognitive impairments or dementia were excluded at baseline. Finally, the statistical analysis and reporting domain were assessed for the risk of over-fitting [
35].
Disagreements were resolved in consensus meetings. Again, to make quality evaluation consistent with the previous meta-analysis, the reviewers adopted the same tool used in the previous systematic review [
32].
Statistical analysis
We performed a meta-analysis using a random effects model to estimate the pooled estimates. We used the DerSimonian and Laird method to estimate the between-study variance components [
36]. We assessed statistical heterogeneity among studies by using the
Q-test based on the chi-squared statistics and quantified the proportion of total variation contributed by between-study variance through the
I2 statistic [
37]. We combined all selected papers from this study with those used in the meta-analysis in the previous systematic review [
32]. Additionally, we performed a cumulative meta-analysis to map any temporal changes in the magnitude and significance of estimates for the association between social relationships and cognitive decline.
We then followed the previous review and stratified our statistical analysis by considering three categories of social relationships: (i) structural aspects; (ii) functional aspects; (iii) a combination of the two. Structural aspects of social relationships refer to the structure of social networks and social activities, such as the size, frequency and heterogeneity of social contacts [
38‐
41]. Functional aspects of social relationships refer to sources support and social integration [
42‐
45]. Finally, the combination between the two included composite indices of social network characteristics, social capital and social engagement [
46,
47]. Whenever social relationship factors were given as categorical variables, we dichotomized them so that the lowest category (poor social relationships) was tested against the other categories combined. We then used the odds ratio based on the new two-by-two table [
32].
We used odds ratios to represent the risk of developing cognitive impairment among people with poor social relationships compared to people with better social relationships. We interpreted hazard ratios as odds ratios. Given that studies mostly reported results with standardized and un-standardized coefficients from linear regression models, we converted these to odds ratios, as suggested by previous research [
48]. Whenever in the original article any information for calculation of odds ratios and 95% confidence intervals were missing, we contacted the authors for any additional information.
When multiple articles were based on the same database, we selected results based on the following criteria: (i) an estimate for the meta-analysis; (ii) determinant measured as a composite measure of social relationship factors, or most compatible with the other studies; (iii) outcome measures such as global cognitive functioning, or most compatible with the other studies; (iv) longest follow-up duration; and (v) largest sample size.
We examined the heterogeneity sources by conducting stratified analyses for structural, functional and combined factors. We included the following characteristics: (i) year of publication (before 2006, 2007-2011, 2012-2018, after 2019); (ii) inclusion in the previous review [
32]; (iii) geographic area (i.e., Asia, Europe, America); (iv) sample size (≤687, 688-1635, 1636-3413, >3413); (v) follow-up duration (≤3, 4-5, 6-9, >9); (vi) average age of baseline participants (≤65, 66-74, ≥75); (vii) outcome (i.e., cognitive function, cognitive decline); (viii) type of outcome (i.e., continuous, dichotomous), as reported in each study; (ix) social relationship measurement (i.e., low social activity, small social network size); (x) selected confounding factors (i.e., age, depression, alcohol consumption and physical activity) adjusted for.
We assessed publication bias by visual inspection of funnel-plots for asymmetry and through the Egger’s test for asymmetry [
49].
Discussion
Summary
Research on the association between social relationships and cognitive decline in older adults has recently increased in terms of numbers of publications, as well as qualitatively with larger and more international samples. Our results confirmed the effect of structural and functional aspects, as well as of their combination for cognitive decline: consistent with the previous meta-review [
32], poorer social relationships predicted cognitive decline.
However, our results confirmed that there was still a considerable level of heterogeneity in the estimation of these statistical effects. After carefully examining this heterogeneity via sub-group analysis, we found that the most probable root-causes were certain methodological differences in social and cognitive variable measurements, the geographic characteristics of sampled populations and the duration of the follow-up study.
By means of a cumulative meta-analysis, we found that the precision and accuracy of estimations increased with a progressive reduction of 95% confidence intervals for all aspects of social relationships. However, this could be due to the increased sample sizes rather than precise variable measurements. Indeed, while studies before 2013 were based on data from local experimental design, authors of studies performed after 2013 increasingly relied on larger representative national surveys. Furthermore, note that 50% of studies did not control for alcohol use, about 40% for depression and about 50% for physical activity, thus not sufficiently controlling for relevant confounding factors. This may have led to an over-estimation of the association, which should be assessed in future reviews.
Additionally, we found other important methodological novelties in more recent studies. In some studies, structural equation modelling was used rather than regression analyses to test linear causal relationships among variables, thus simultaneously accounting for measurement errors [
15,
64,
65]. This modelling technique is key to estimate causal mechanisms more accurately, especially in contexts in which reverse causality and varying possible causal paths need to be jointly assessed. Our test suggests that reverse causality between social relationships and cognitive decline has also been more carefully assessed in the most updated research. However, besides using more analytical statistical models, future research should also try to follow more robust sampling selection procedures, e.g., excluding participants with cognitive impairment or dementia at baseline. This would help minimising reverse causality issues and improve the robustness of results.
Strengths and limitations
Our study has several strengths, which should be underlined. First, we extended and updated a previous systematic review [
32], thus consolidating the systematic review approach in an interdisciplinary area where social scientists, geriatrics, neuro-epidemiologists and other experts are increasingly collaborating. Second, while the previous review addressed the study of cognitive decline and ageing on social relationships [
32], our review presented a cumulative meta-analysis on the entire field, revealing that studies have achieved better OR estimates and progressively reduced the 95% confidence intervals for all aspects of social relationships, probably due to increased sample sizes.
Another important point is that our review included a bias risk analysis showing that methodological problems of these studies concern especially the weak control on certain confounding factors, including alcohol use, depression and physical activity, as well as on the lack of important detail on the participation rate and outcome assessors. These recurrent deficiencies must be solved in future research in order to improve the quality of findings assessment and stimulate cumulativeness and systematic comparisons.
As in the previous review, we had to face certain methodological challenges. On the one hand, we confirmed the significant heterogeneity between studies previously reported [
32], which required meta-regression and subgroup analysis. On the other hand, we still detected possible publication bias for all three aspects of social relationships, which led us to conclude that estimates may well have been over-estimated. The prospective registration of observational studies and initiatives by journals and associations to increase data sharing and open data, occurring now in many other research areas [
69], could improve methodological standards and quality of study design in this field.
Understanding potential mechanisms responsible for the complex associations between social relationships and cognitive decline requires to tackle complex pathways [
70,
71]. On the one hand, social relationships can be instrumental for accessing relevant information for prevention and protection, stimulating intellectual and social engagement, increasing well-being and avoiding social isolation [
31,
72,
73]. Indeed, social relationships express the full spectrum of lifestyles, including behaviours and norms that can lead to healthy or unhealthy outcomes [
8]. For instance, recent research on the development of chronic diseases has shown that richer social networks can lower the speed of disease by improving prevention and protection [
74]. Given that the chronic diseases is often associated with dementia incident [
75], it is probable that this could be one pathway connecting social networks to cognitive decline. On the other hand, social networks convey a variety of emotional and material resources to individuals, whose complementarity or substitution effects are often difficult to estimate [
31].
Besides certain interesting recommendations from the previous review [
32], we suggest focusing on a ‘complexity hypothesis’. This is because social relationships are part of a complex social infrastructure linking individuals to a potential set of material and emotional resources related to cognition. It is likely that conventional measurements of social networks, such as network size and frequency of contacts, only partially reflect the complexity of personal networks. For instance, with data from the Longitudinal Aging Study Amsterdam (LASA), including 2,959 Dutch participants aged 54 to 85 at baseline in 1992 and six follow-ups covering a time span of twenty years, [
20] showed that a reduction in network complexity was detrimental for cognitive functioning, neither explained by size of the network nor by simple presence of specific relationship type. The importance of non-redundant ties and the variety and heterogeneity of contacts [
27] on health has been highlighted in a variety of studies [
76,
77]. Being connected with non-redundant ties, all possibly with different information, skills and lifestyles, could be instrumental for older adults to access a greater variety of cognitive stimuli, obtain relevant information and achieve help and support [
78,
79].
This is also linked to the so-called “focus theory”, which postulates that social relationships are more likely to form between individuals sharing certain ‘focused activities’ (e.g., community life, voluntary organisations) [
26]. Given that ageing implies transitions and changes (e.g., retirement, widowhood and informal caregiving), we should expect that new ‘focused activities’ in later life could greatly affect egocentric networks of older adults imposing important re-configurations [
80,
81]. As correctly suggested by [
26], more careful attention to changes in network boundaries during later life and the inclusion of space as a social environment where most of these ‘focused activities’ take place, could improve our understanding of the effect of these network changes on cognitive decline of older adults [
26].
Conclusions and further research
This review has updated a previous meta-analysis on the effect of social relationships on cognitive decline in longitudinal cohort studies performed in 2016. We confirmed previous evidence on the importance of multiple aspects of poor social relationships, including structural, functional and a combination of these factors, to predict cognitive decline. Our cumulative meta-analysis would indicate that the precision of estimations has increased, at least since 2006, probably due to the increased sample sizes of these studies. However, deficiencies and problems persist, especially in study design (e.g., missing information on dropouts) and measurements.
Our results suggest that future research should consider the complexity of social factors associated with cognitive decline more carefully by improving measurements, especially reconstructing personal networks with data on alters’ alters so that the variety and heterogeneity of contacts can be estimated more precisely, including network boundaries and redundancy [
26]. While this can be difficult for longitudinal cohort studies, less conservative and more explorative studies on new, ad-hoc samples, including experimental research, could help to test the accuracy of more complex network measurements, thus providing insights on how to incorporate these measurements in longitudinal cohort studies.
Finally, there is a need for research exploring the interplay between social networks, chronic diseases in adult life and cognitive decline in older age. This could also improve our understanding of the possible short and long-term impacts of the COVID-19 pandemic on new clinical conditions. The COVID-19 pandemic has hit especially hardly people in lower socio-economic strata of the population. Given that its consequences will probably increase health inequalities [
4], we need more information on how individuals react to these changing social conditions and have adapted to external shocks [
82].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.