Associations between social health factors, cognitive activity and neurostructural markers for brain health – A systematic literature review and meta-analysis

Social health factors (e.g., social activities or social support) and cognitive activity engagement have been associated with dementia risk, but their neural substrates have not been well established. This systematic review and meta-analysis summarizes the available evidence regarding the association between these factors and cerebral macro- and microstructure. A comprehensive literature search was conducted in various databases, following predefined criteria. Heterogeneity, risk of publication bias and overall certainty of evidence were assessed using standardized scales and, whenever appropriate, random effects meta-analysis was conducted. Of 6,715 identified articles, 43 were included. Overall, consistency of findings was low and methodological heterogeneity high for all outcomes. However, in some studies cognitive and social activities were positively associated with total brain, global and cortical grey matter and hippocampal volume as well as white matter microstructural integrity. Furthermore, structural social network characteristics (e.g., social network size) were associated with regional grey matter volumes, while functional social network characteristics (e.g., social support) were additionally associated with total brain volume. Meta-analyses revealed small but significant partial correlations between cognitive and social activities and hippocampal (three studies; n=892; rz=.07) and white matter hyperintensity volume (three studies; n=2934; rz=-.04). More prospective studies are needed to assess temporal associations.


Introduction
Due to a continuously ageing world-population, the number of people living with dementia is projected to rise from approximately 50 million to date to 82 million by 2030 and 152 million by 2050 (World Health Organization, 2018). Alongside identifying disease-modifying drugs, much focus is nowadays placed on risk reduction by acting on modifiable risk factors (Steyaert et al., 2020;World Health Organization, 2018). The Lancet Commission on Dementia Prevention, Intervention and Care (Livingston et al., 2020) estimated that a substantial proportion of all new dementia cases could theoretically be prevented by tackling common modifiable risk factors.
Higher engagement in cognitive and social leisure activities has been associated with a lower risk for dementia or cognitive decline (Duffner et al., 2022;Iizuka et al., 2019;Sajeev et al., 2016;Stern and Munn, 2010;Yates et al., 2016;Yu et al., 2020). Such activities include a high cognitive or social component and are mainly conducted for the sake of pleasure or out of curiosity, such as reading, playing an instrument or meeting up with family or friends. In a similar vein, both structural (e.g., larger social network size) and functional (e.g., more social support) social network characteristics have been associated with better cognitive performance and lower risk for developing dementia in some studies (Kuiper et al., 2015;Lenart-Bugla et al., 2022;Marseglia et al., 2023;Penninkilampi et al., 2018;Samtani et al., 2022). The umbrella term of 'social health' has been introduced to capture an array of concepts reflecting human capacities to engage in social relationships, including the engagement in social activities, as well as structural and functional social network characteristics (Vernooij-Dassen et al., 2022a).
While evidence is accumulating for the role of such social health factors and cognitive activity as potential buffers against cognitive decline and dementia, biological evidence concerning their underlying neural macro-and micro-structural correlates has not been well established (Bielak and Gow, 2022). Such correlates include volumetric and damage markers on magnetic resonance imaging (MRI) as well as white matter microstructural properties measured by diffusion tensor imaging (DTI). Knowledge about these neurostructural underpinnings may aid in further exploring the potential role of social health factors and cognitive activity in curbing age-related brain changes.
Thus far, one systematic review has investigated the relationship between cognitive and social leisure activities and structural brain markers, focusing on observational studies including healthy participants aged 60 years and older (Anatürk et al., 2018). It concluded that higher self-reported levels of cognitive and social activity engagement are associated with larger regional grey matter, larger global white matter, and smaller white matter hyperintensity volumes.
Though both social contact and cognitive activity engagement have been recommended as potential targets for dementia risk reduction, it is not clear in what period of life interventions may be most beneficial (Livingston et al., 2020). Hence, there is a need to update and extend previous work to those of younger age (<60 years). Furthermore, in consideration of the multifaceted nature of the social health domain, there is a need to assess a broader range of factors beyond social activity engagement, including structural and functional aspects (e.g., social support). Therefore, the current systematic review and meta-analysis aims to summarize the best available evidence regarding the association between social health factors (social activity engagement, structural and functional social network characteristics) as well as cognitive activity engagement and the brain's macro-and microstructure across the entire adult age span in cognitively healthy individuals.

Methods
This systematic review was conducted in line with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) and Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines (Page et al., 2021;Stroup et al., 2000). A full study protocol is available for this review and can be consulted on the PROSPERO international register of systematic reviews (Record number: CRD42020193278).

Data sources and search strategy
A literature search was performed in the PsycINFO, MEDLINE and CINAHL databases, as well as the Cochrane Central Register of Controlled Trials, and included all articles from inception to date of abstract extraction on July 18th 2022. A keyword profile was created based on search terms related to (1) cognitive leisure activities, (2) social leisure activities, (3) structural or (4) functional social network characteristics and (5) brain structure. The keywords were identified through a pilot literature search. The full keyword profile used for the search (in EBSCOhost) is included in Supplementary Table 1. Applied filters consisted of a restriction to human studies with participants older than 18 years, English language, and publication in peer-reviewed journals.

Study selection
Extracted studies were screened for broad eligibility by two independent reviewers (L.D. and N.D.J.), who then compared and discussed their results. In case of discrepancy, discussion with a third reviewer (S. K. or K.D.) took place. Full texts were then retrieved and examined for final eligibility. Reference lists of included articles and reviews were scrutinized for additional relevant studies. The risk of bias in the individual studies was assessed by two independent reviewers (L.D. and N.D. J.). In case of disagreement, results were discussed with a third reviewer (S.K. or K.D.). We used the Newcastle-Ottawa Scale (NOS; Wells et al., 2000) for the bias-assessment of prospective studies. For cross-sectional studies, an adapted version of the NOS was used (Deckers et al., 2017; theoretical range 0-9). The study quality of randomized trials was evaluated using the National Institutes of Health Study Quality Assessment Tool (National Institutes of Health NIH, 2013).

Inclusion criteria and study eligibility 2.3.1. Study design
Randomized controlled trials (RCTs) and other experimental or quasi-experimental studies, prospective cohort studies and crosssectional studies were included.

Types of participants
Studies including adult participants (i.e., older than 18 years) without a diagnosis of dementia and free from other neurological or psychiatric disorders at baseline measurement were eligible.

Types of exposures/interventions
Three clusters of predictors were included in this review: (1) cognitive leisure activities (CA) and social leisure activities (SA), (2) structural social network characteristics, and (3) functional social network characteristics. CA were defined as any kind of activity performed for the sake of pleasure that requires the individual to be mentally involved, with a relatively low degree of social involvement (e.g., doing crossword puzzles, reading or playing an instrument). SA were those activities that serve the purpose of connecting with or getting in contact with others (e. g., social clubs or visiting friends and family). Both studies assessing joint (CA + SA) and individual (CA or SA) associations with brain markers were eligible. Structural social network characteristics included social network size and complexity. Functional social network characteristics included perceived instrumental or emotional support from someone's social network and subjective feelings of loneliness. Both ratings of individual activities/social network characteristics as well as overall ratings/summary scores were included. Studies investigating the association between cognitive training of specific neuropsychological functions (e.g., memory training) and brain structure were excluded, as were studies on highly selective or specialized groups such as experts and professionals (e.g., professional chess-players).

Types of outcomes
Potential studies used structural imaging including MRI, Computed tomography (CT) and DTI and explicitly reported measures of direct associations between CA/SA, social network characteristics and brain structure. Eligible measures were volumetric-(e.g., total and regional grey matter volume), or cortical thickness-measures, diffusion characteristics (e.g., fractional anisotropy, a measure of the microstructural integrity of white matter tracts) or damage markers (e.g., cerebral small vessel disease or atrophy ratings). Both whole brain and regional measures were eligible.

Data extraction
Data were collected using a standardized extraction protocol (Supplementary Table 2). Information derived included demographic information about study participants, methodological information (e.g., study design), information about predictors and outcomes (e.g., brain regions of interest). In this study, age ranges were clustered into young adulthood (younger than 40 years), midlife (40-64 years) and late life (65 years and older), for reasons of interpretability and comparability. For studies assessing CA/SA, we reclassified activities if necessary based on the classification made by Seider et al. (2016) Supplementary  Table 3a). This was done in light of considerable between-study heterogeneity concerning the classification of an activity as primarily cognitive or social. Analyses were labeled as 'Region of Interest' (ROI) if target regions were predefined, and labeled as 'whole brain analysis' if global measures (e.g., total grey and white matter volumes) were provided. Studies using an exploratory (data driven) approach were labeled 'exploratory'. DTI studies were classified based on the image processing technique used, such as tract-based spatial statistics (TBSS) or tractography.

Grading of evidence
To provide an overall level of certainty of the available evidence, we used the existing Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach (Atkins et al., 2004). Briefly, GRADE provides a framework for assessing level of certainty at an outcome level, based on various criteria. In line with GRADE, certainty of outcomes was graded from very low to high, on the basis of risk of bias of individual studies (outlined in Section 2.1), consistency of findings, construct validity, estimate precision and publication bias. Consistency of findings was graded (based on the number of studies reporting a particular finding) as high (>80%), moderate (60-80%) or low (<60%). For associations reported by less than three studies, no conclusion was drawn. Results of all included studies are based on their fully adjusted statistical models (if available/reported).

Data synthesis and meta-analysis
Where appropriate, outcome measures were pooled by means of random effects meta-analysis. Partial correlations between social health factors, cognitive activity and brain markers, adjusted for age, sex and total intracranial volume, were used as measure of effect size. If these minimally adjusted partial correlations were not available, fullyadjusted correlations were used. Alternatively, t-values of betacoefficients were transformed into partial correlation coefficients and their 95% confidence intervals (see Aloe and Thompson, 2013 for formulas used). Corresponding authors were contacted for additional data, if necessary, via email and received two reminders in case of nonresponse.
Heterogeneity between studies was assessed by I 2 values and Cochran's Q (Cochran, 1954;Higgins et al., 2003), using two-sided tests with an alpha-level of.05. Risk for small study bias was assessed visually using Funnel plots and Egger's tests for asymmetry (Egger et al., 1997). Pooling was possible if partial correlations for at least three studies were available. Studies reporting multiple estimates (e.g., midlife and late life CA/SA, or separate results for men and women) were combined based on a method described by Borenstein, chapter 24) et al. (2009) to yield pooled estimates. Partial correlations of one study were estimated based on the mean differences reported by the authors (Hafsteinsdottir et al., 2012), according to formulas suggested by Pustejovsky (2014). All analyses were conducted in Stata (version 17;StataCorp, 2021).

Search results
The search yielded 6715 unique articles, of which 74 were included for full-text review. Fig. 1 depicts a flow diagram of the study selection process. Three studies were identified through cross-referencing (Bickart et al., 2011;Gidicsin et al., 2015;Wirth et al., 2014). A total number of 43 studies were eventually included in the review (Tables 1a, 1b, 1c). Results of the risk of bias assessment of individual studies can be found in Supplementary Tables 4a-c. All studies were of sufficient quality to be included in the review.

Participant characteristics
Across study designs, the mean age of participants ranged from 20.0 to 85.2 years. However, only some studies (n = 16) included participants younger than 65 years on average (Tables 1a,1b,1c). Notably, studies investigating structural social network characteristics included somewhat younger participants as compared to studies assessing functional ones or CA and/or SA (weighted mean ages 58.3 vs. 63.1 and 64.7, respectively). Overall, 53.1% of participants were female, with similar proportions when assessed per predictor.

Data synthesis
Twenty-five authors were contacted for additional information. Of these, 13 responded and six eventually provided data. Due to the low number of studies specifically investigating one leisure activity domain, meta-analyses for CA/SA were performed on their combined association with a particular brain marker. This was possible for studies investigating hippocampal and white matter hyperintensity volume. Additionally, for functional social network characteristics, pooling was possible for studies which investigated the association between loneliness and hippocampal volume. Meta-analysis could not be conducted for any other association. Fisher-z transformed partial correlations for statistically significant associations are still presented in Tables 2a, 2b, and 2c, wherever calculation thereof was possible.
In the subsequent sections, information will be presented separately for (1) CA/SA, (2) structural and (3) functional social network

Table 1a
Characteristics of included studies investigating the association between SA/CA and brain structure. FLAIR= fluid attenuated inversion recovery; FSE= fast spin echo; FSPGR= fast spoiled gradient echo; IQR=interquartile range; MPRAGE=magnetization-prepared rapid gradient echo; MRI= magnetic resonance imaging; PD= proton density; RCT=randomized controlled trial; SD=standard deviation; SPGR= spoiled gradient recalled; a Current= report of activities as they are carried out at the time of the interview; early=retrospective assessment of activities carried out in young adulthood; midlife=retrospective assessment of activities carried out in midlife; past=retrospective assessment of activities carried out in the past, without individual assessment of life-stages (excluding current engagement); lifetime= assessment of activities carried out throughout the lifespan, without individual assessment of life-stages (including current engagement); b Prospectiveactivities assessed at five timepoints across 15 yearsone MRI measure; c Prospectiveactivities assessed at two timepoints across 7.6 yearsone MRI measure; d 308 without missing information about leisure activities; e For Hafsteinsdottir et al., only subgroup descriptives were available. The overall mean and SD have thus been calculated (Higgins et al., 2022); f Subsample with available DTI measures (total n = 442); g Prospectiveactivities assessed at two timepoints across 3 yearstwo MRI measures; h Prospectiveactivities assessed at baselinetwo MRI measures across 2-3 years; i Composite measure of life-stage specific experiences, which are partly measured retrospectively, including education (young adulthood), occupational complexity (midlife) and current (i. e. late-life) CA/SA engagement; j Prospectiveactivities assessed at baselinetwo MRI measures across 3 years; k Prospectiveactivities assessed at baselinetwo MRI measures across 2.5 years characteristics. For each of the three, results will be summarized per brain marker as outcome. Within each subsection, significant findings will be presented according to study sample size (large to small), whenever possible. For CA and SA, studies assessing their joint (CA+SA) or individual (CA or SA) association with a brain marker will be presented separately. Detailed results of individual studies can be found in Tables 2a, 2b, and 2c for CA/SA, structural and functional social network characteristics, respectively. A global summary of main findings is included in Table 3.

Cognitive/social activity and total brain volume
Overall, studies reported small positive associations between CA/SA (jointly or individually) and total brain volume with low consistency and low certainty based on GRADE criteria (see Table 3). More specifically, two studies investigated the joint association between CA+SA and total brain volume. Cross-sectional analyses of the Age Gene/Environment Susceptibility Reykjavik Study (AGES-Reykjavik; n = 4303) found that those in the highest CA+SA quartile had significantly larger total brain volumes as compared to those in the lower quartiles (Hafsteinsdottir et al., 2012). A small longitudinal study (Valenzuela et al., 2008, n = 79)

Table 1b
Characteristics of included studies investigating the association between structural social network characteristics and brain micro and macro structure. Abbreviations: DTI= diffusion-tensor imaging; MDEFT= modified driven equilibrium fourier transform; MPRAGE=magnetization-prepared rapid gradient echo; MRI= magnetic resonance imaging; SD= standard deviation; SNS= social network size; a Current= assessment of network characteristics as they occur at the time of the interview or self-report; b Sample for analyses of online social network size (number of social media friends); c Sample for analyses of real-world social network size

Table 1c
Characteristics of included studies investigating the association functional social network characteristics and brain micro and macro structure.

Table 2a
Findings of studies investigating the association between CA/SA and brain micro and macro structure. CA/SA TBV b Regional GM volume a,g Regional WM volume a,g  reported no significant results at baseline or three-year follow up.
Four studies assessed the individual association between CA or SA and total brain volume. One cross-sectional study (n = 348) found that people who reported higher SA had higher total brain volumes (James et al., 2012). Moreover, one randomized trial (Mortimer et al., 2012) found that a 40-week SA intervention, consisting of three weekly group meetings and discussions, was associated with a significant increase in total brain volume compared to a control group (n per group=30). Two further studies were non-significant (Bennett et al., 2006, n = 106;Foubert-Samier et al., 2012, n = 331).

Global grey matter
Studies investigating the joint and individual associations between CA/SA and global GM volumes reported small positive associations with low consistency and low overall certainty (see Table 3). Three crosssectional studies investigated the joint association between CA+SA and global GM volume (Gow et al., 2012;Hafsteinsdottir et al., 2012;Vaughan et al., 2014). In the large study by Hafensteinsdottir et al. (2012;n = 4303), the highest activity quartile had higher global GM volumes as compared to the lower quartiles. Two studies were non-significant (Gow et al., 2012, n = 691;Vaughan et al., 2014, n = 393).
One cross-sectional (James et al., 2012) and two longitudinal studies (Anatürk et al., 2020(Anatürk et al., , 2021 separately assessed CA or SA in relation to global GM volume. Using data of the Whitehall-II study (n = 574), one study serially assessed CA over a period of 15 years in relation to global and voxel-wise GM volume (Anatürk et al., 2020). While they did not find an association between baseline CA or SA and GM volume, associations between change in CA over time and global GM volume were significant. Another study (n = 348) found a positive association between SA and global GM volume in a sample containing both former lead factory workers and control subjects (James et al., 2012). One large study by Anatürk n=7, 152) et al. (2021) assessed trajectories of individual CA's or SA's across 7.6 years and global GM volume at follow-up and was non-significant.
3.7.2. Regional grey matter 3.7.2.1. Cognitive/social activity and hippocampal (HC) volume. Overall, studies assessing CA/SA (jointly and individually) in association with HC volume suggested small positive associations, but with low consistency and low certainty (see Table 3). More specifically, of the eight studies investigating the joint association between CA+SA and HC volume, six were cross-sectional (Arenaza-Urquijo et al., 2017;Gidicsin et al., 2015;Moored et al., 2020;Schultz et al., 2015;Suo et al., 2012;Vemuri et al., 2012) and two were longitudinal studies (Valenzuela et al., 2008;Vemuri et al., 2016). Some of these studies suggested positive cross-sectional associations with HC volume ( Ten studies investigated the individual association between CA or SA and HC volume (eight cross-sectional studies; Bennett et al., 2006;Bittner et al., 2019;James et al., 2012;Jeon et al., 2020;Schultz et al., 2015;Shen et al., 2022;Seider et al., 2016;Wirth et al., 2014 and two randomized trials;Bellander et al., 2016;Greenberg et al., 2019). Shen et al. (2022) found that a social isolation composite measure including social inactivity (less vs. more than weekly engagement), contact frequency and living arrangement (living alone vs. with others) was negatively related to HC volume in 32,263 individuals of the UK Biobank cohort. Two further cross-sectional studies found that higher SA was associated with larger HC volumes (Bennett et al., 2006, n = 106;Bittner et al., 2019 n = 549). In one more study, higher CA, but not SA, was positively associated with HC volume (Seider et al., 2016, n = 65). Lastly, one randomized trial reported a larger increase in HC volume in response to a language learning intervention (Bellander et al., 2016, n = 56). Five studies did not find significant results (Greenberg et  Partial correlations from three studies, representing 892 participants, could be obtained for meta-analysis Moored et al., 2020;Vemuri et al., 2012). All three reported non-significant associations themselves. In contrast, pooled analysis revealed a small but significant overall partial correlation between CA+SA and HC volume (r z =0.07; 95%CI=0.01-0.14; p = .033; Fig. 2), while there was no inter-study heterogeneity (Q=0.27; p = .87; I 2 = 0.0%). There were no signs of small-study bias based on visual inspection of the Funnel plot ↑= positive association between specific predictor and MRI outcome; ↓= negative association between specific predictor and MRI outcome; Abbreviations: AM= amygdala; CA= cognitive activity; N.s.= not significant; GM= grey matter; HC= hippocampus; LEQ= Lifetime of Experiences Questionnaire; OFC= orbitofrontal cortex; PFC=prefrontal cortex; SA=social activity; SNS= social network size; TBV= total brain volume; TIV= total intracranial volume; VW= voxel-wise; WB= whole brain; WM= white matter; WMH= white matter hyperintensity; a Voxel-based morphometry; b Whole brain analysis; c Tract-based spatial statistics (TBSS); d Quadratic coefficients (initial increase and subsequent leveling) of CA significantly associated with total GM volume; e Region of interest (ROI) analysis; f Tractography; g Exploratory (data driven) analysis; h MRI measure controlled for total intracranial volume (TIV); i Association between CA+SA and left superior frontal gyrus (orbital part) only significant in model that is additionally adjusted for hearing difficulty; j Association between CA+SA and right caudate nucleus only significant in model that is additionally adjusted for activities of daily living; k Presented as percentage of TIV; l Engagement in leisure activities was associated with normal appearing white matter in age and sex adjusted model; m SA assessed as part of a social isolation measure, alongside questions about living arrangement; n Baseline, 3-year follow up and change were assessed; o Measured as part of a multimodal marker including HC volume, cortical thickness and FDG-PET APOE-e4 carrier status L.A. Duffner et al. (see Supplementary Figure 1) and the Egger's test (p = .107). Fig. 3.

Cognitive/social activity and amygdala volume.
Overall, studies reported small associations between CA or SA (but not CA+SA) and amygdala volume, while consistency was low and certainty very low (Table 3). More specifically, three cross-sectional studies assessing the joint association between CA+SA and amygdala volume did not find significant results (Moored et al., 2020, n = 90; Schultz et al., 2015,

Table 2b
Findings of studies investigating the association between structural social network characteristics and brain micro and macro structure.  Findings of studies investigating the association between functional social network characteristics and brain micro and macro structure. Abbreviations: AM= amygdala; GM= grey matter; HC= hippocampus; MRI= magnetic resonance imaging; N.s.= not significant; PFC=prefrontal cortex; SNS=social network size; TBV= total brain volume; WMH=white matter hyperintensity; a MRI measure controlled for total intracranial volume; b Region of interest (ROI) analysis; c Exploratory (data driven) analysis; d Voxel-based morphometry; e Whole-brain analysis; f Analysis method not reported g Multimorbidity score included cancer, coronary heart disease, heart failure, chronic obstructive pulmonary disease, chronic kidney disease, diabetes mellitus type 2, stroke and Parkinson's disease.

Table 3
Summary of main findings including GRADE certainty ratings.

Cognitive/social activity and regional cortical grey matter.
Generally, studies suggested small positive associations between CA/SA (jointly and individually) and regional cortical GM volume with high consistency and moderate overall certainty (see Table 3). These studies employed both ROI and data driven (exploratory) analyses. Three crosssectional studies assessed the joint association between CA+SA and regional cortical GM volume (Arenaza-Urquijo et al., 2017;Foubert-Samier et al., 2012;Seider et al., 2016). One of them found positive associations with GM volumes in all four lobes (Seider et al., 2016, n = 65), one in frontal and temporal areas specifically (Arenaza-Urquijo et al., 2017, n = 45) and one was non-significant (Foubert-Samier et al., 2012, n = 331).

Cognitive/social activity and global white matter volume
Only one study reported small positive associations between CA+SA (but not CA or SA) and global WM volume. Consistency and overall certainty of findings were thus graded low and very low, respectively (see Table 3). More specifically, only in the abovementioned study by Hafsteinsdottir and colleagues (2012, n = 4303), those reporting the highest level of CA+SA had larger global WM volumes as compared to those reporting lower levels. Three further studies assessing the joint association between CA+SA and global WM volume were nonsignificant (Foubert-Samier et al., 2012, n = 331;Gow et al., 2012, n = 691;Vaughan et al., 2014, n = 393), as were both studies assessing individual associations (Anatürk et al., 2021, n = 7152;James et al., 2012, n = 348).

Cognitive/social activity and regional white matter volume
One cross-sectional study assessed the joint relationship between CA+SA and regional WM volume but did not report significant results (Foubert-Samier et al., 2012, n = 331). The cross-sectional study by James et al. (2012, n=348) found a positive association between SA and corpus callosum volume. Given the small number of studies (<3), no assessment of consistency or overall certainty was made.

Cognitive/social activity and white matter microstructure
Studies reported small positive associations between CA, SA (jointly and individually) and white matter microstructural integrity, with low consistency and low overall certainty of findings (see Table 3). Of two cross-sectional studies assessing CA+SA in association with WM microstructure (Arfanakis et al., 2016;Gow et al., 2012), one study found better regional white matter structural integrity (Arfanakis et al., 2016, n = 379) in relation to higher CA+SA levels. The other study reported no significant findings (Gow et al., 2012, n = 691).

Structural social network characteristics and MRI measures
No studies reported on the association between structural social network characteristics and total brain, global and regional WM and global grey matter volumes or WM hyperintensities.

Structural social network characteristics and regional grey matter
Studies reported small positive associations between structural social network characteristics (social network size and complexity) and regional GM volumes, but with low consistency and very low certainty (see Table 3). The nine included studies used both ROI and data driven (exploratory) approaches (Bickart et al., 2011;Bittner et al., 2019;Kanai et al., 2012;Noonan et al., 2018;Powell et al., 2012;Sharifian et al., 2022;Sherman et al., 2016;Turel et al., 2018;Von der Heide et al., 2014).
One study with 549 participants reported that scores on the Social Integration Index (Berkman et al., 2004; composite score of SA and social network characteristics), were positively related to GM volume in the left HC as well as cortical folding in the right ventrolateral prefrontal cortex (Bittner et al., 2019). Another study found that online (number of social media friends) but not offline social network size was related to volume of cortical structures in a sample of 125 younger adults (mean age= 23.3 years; Kanai et al., 2012). In two further smaller studies, social network size and complexity were related to thickness (Bickart et al., 2011) and volume (Noonan et al., 2018) of various cortical areas. Five studies did not find significant correlations (Powell et al., 2012;Sharifian et al., 2022;Sherman et al., 2016;Turel et al., 2018;Von der Heide et al., 2014).

Structural social network characteristics and white matter microstructure
Both cross-sectional studies assessing structural social network characteristics in relation to WM microstructure reported significant findings (Molesworth et al., 2014;Noonan et al., 2018). More specifically, in both studies larger social network size was associated with better white matter microstructural integrity (Molesworth et al., 2014, n = 155;Noonan et al., 2018, n = 18). In one of the studies, this association was also found for social network complexity (Molesworth et al., 2014). Given the small number of studies, no assessment of consistency or certainty was conducted.

Functional social network characteristics and MRI measures
No studies reported on associations between functional social network characteristics and global and regional WM volumes.

Functional social network characteristics and total brain volume
Overall, studies found small positive associations between functional social network characteristics (feelings of loneliness and perceived social support) and total brain volume with high consistency and moderate certainty (see Table 3). Of the three large studies (two cross-sectional; Salinas et al., 2021Salinas et al., , 2022one longitudinal;van der Velpen et al., 2021), one (n = 3737) found that higher levels of perceived social support were associated with larger total brain volume at baseline (van der Velpen et al., 2021). It was also associated with better preservation of total brain volume at follow-up. No such correlations were found for loneliness.
Assessing specific aspects of social support, another large study (n = 2171) found that the availability of a supportive listener, availability of somebody to give advice, and satisfaction with the frequency of social contacts were positively related to total brain volume (Salinas et al., 2021). Another study found that in 1611 people of the Framingham Study cohorts, loneliness, as measured by a single item of the Center for Epidemiological Studies Depression scale (CES-D; Radloff, 1977), was associated with lower total brain volumes (Salinas et al.,Fig. 4. Forest plot of studies assessing the relationship between loneliness and hippocampal volume. 2022).

Functional social network characteristics and white matter microstructure
One cohort study investigating the cross-sectional and longitudinal relationships between loneliness, social support and WM microstructure, found that higher perceived social support at baseline was associated with higher baseline microstructural WM integrity, globally (van der Velpen et al., 2021). There was no association with loneliness. No assessment of consistency or overall certainty was conducted.

Summary of findings
This systematic literature review and meta-analysis comprehensively summarized the available literature on the association between social health factors, CA and brain macro-and microstructure. Studies were predominantly observational and had a cross-sectional design, with a high degree of diversity between measures for predictors and outcomes. Indirectness of exposure, as part of the GRADE certainty assessment, was thus rated high for all outcomes. Moreover, relatively few studies assessed social health factors or CA in early adulthood or midlife in relation to late-life brain outcomes, making it difficult to reach a conclusion for the life-stage specificity of associations. Generally, only few studies were available on structural and functional social network characteristics.
Overall, consistency of findings was low and certainty was graded as low or very low for most outcomes. Certainty of studies assessing CA/SA and regional cortical GM volume and those investigating functional social network characteristics and total brain volume was graded as moderate. By and large, studies suggest that social health factors and CA may be associated positively with regional GM (CA/SA, structural and functional social network characteristics) and total brain volumes (CA/ SA and functional social network characteristics). They may also be negatively associated with WM hyperintensities (CA/SA). Meta-analysis suggested a modest positive correlation between CA/SA and hippocampal volume and a negative association with WM hyperintensities. As for WM macro-and microstructural measures in general, findings are suggestive for positive associations, but further study is required because of low numbers of studies. While results were generally mixed, studies either reported associations in the hypothesized direction or null-findings; there were no studies which reported opposite associations (e.g., higher CA/SA engagement in association with lower brain volumes).

Neurobiological underpinnings
Findings of this systematic review and meta-analysis may align with the heuristic concepts of brain reserve and brain maintenance, which both aim to explain inter-individual differences in cognition in light of similar neuropathological burden (Stern et al., 2018). Specifically, brain reserve refers to the neural profile of a person (e.g., higher neuronal capital), which may be protective as there is 'more to lose' before cognitive symptoms appear. While brain reserve is generally considered a fixed construct, life experiences such as stimulation through CA/SA are hypothesized to buffer age-related decline in brain reserve by the means of brain maintenance and better brain connectivity (DeJong et al., 2022). Neural mechanisms by which CA/SA potentially affect brain maintenance are suggested by animal models and may include the prevention of neuronal shrinkage and neuronal death (Milgram et al., 2006). Moreover, experiences such as CA/SA may exert a more direct effect on the brain by affecting neurogenesis (Lledo et al., 2006;Milgram et al., 2006), axonal and synaptic sprouting (Trachtenberg et al., 2002), dendrogenesis (Hickmott and Ethell, 2006) and myelination (Imfeld et al., 2009). Studies using rodents and human participants also support the role of larger and more diverse social networks for both myelin integrity and neuro-inflammation, whereas more social contact may decrease neuro-inflammation which in turn preserves WM integrity (Hermes et al., 2006;Karelina et al., 2009;Molesworth et al., 2014). In line with this, in one study of 776 young adults (age 18-27 years) lower levels of loneliness were associated with better WM integrity in various frontal, parietal and temporal regions (Nakagawa et al., 2015). Moreover, one review supports possible direct effects that lifestyle factors such as CA/SA may have on AD-pathology, including a slowing of amyloid-beta deposition in addition to compensatory mechanisms (Arenaza-Urquijo et al., 2015).

Exploration of potential pathways
While for the sake of this review, CA/SA, structural and functional social network characteristics have been assessed separately, they are obviously related. In practice, larger and more diverse social networks may offer more opportunities for engaging in social activities and may also offer more opportunities for receiving instrumental and emotional social support (Berkman et al., 2000). Unfortunately, none of the studies of this review included both CA/SA and structural or functional social network characteristics or controlled for their mutual influences. Further studies are thus needed to assess their interconnectedness in relation to brain markers. Such studies should also explore a larger chain of causation through additional physiological pathways, such as the structural and functional brain connectome and a "social buffering" of the hypothalamic-pituitary-adrenocortical (HPA) system (Hennessy et al., 2009).
Brain regions identified in this review may also be implicated in pathological ageing, including medial temporal (including the HC) and frontal brain areas. Various studies have found that differences in predominantly WM microstructural measures partially mediated the association between CA/SA and cognition (Arfanakis et al., 2016;Köhncke et al., 2016;Wirth et al., 2014). Likewise, one study found that social network size moderated the association between cortical thickness in areas related to Alzheimer's disease (as specified by Dickerson et al., 2009) and cognition (Sharifian et al., 2022). However, while reporting positive associations between CA/SA and brain structure as well as cognition, others could not establish a mediating or moderating link between the three factors (Vaughan et al., 2014). Regardless of the presence of moderation or mediation by structural brain changes, predictors included in this review may also more directly affect cognitive resilience against brain damage (Stern et al., 2018). In this regard, CA/SA and structural social network characteristics have also been suggested to moderate the association between other modifiable and non-modifiable risk factors and dementia (Dekhtyar et al., 2019;Marseglia et al., 2019). Social health factors may thus be potential targets for healthy cognitive ageing over and above their direct association with structural brain markers.

A need for further research
Due to the cross-sectional design of most included studies, causal links cannot be inferred. More specifically, the possibility cannot be excluded that ageing-related changes in brain macro-and microstructure may actually bring about unfavorable changes in CA/SA and social network characteristics. Evidence from a large longitudinal study assessing dementia risk over a period of 25 years suggests that reverse causation may account for the positive associations between CA/SA and incident dementia found by studies with limited follow up times (Floud et al., 2021). Another study investigated changes in cognition in association with levels of loneliness using cross-legged panel models in the Lothian Birth Cohort 1936 (Okely and Deary, 2019). While worse cognition at baseline predicted unfavorable changes in loneliness, the opposite was not the case. It is likely that such reverse causality may also be a threat to studies assessing structural brain markers. Such directionality may also align with the "social brain hypothesis" of the relationship between social behaviors and brain characteristics. It suggests that in primates, the increasing complexity of social interaction (e.g., larger and more complex social networks) is reflected in larger brain sizes (Dunbar, 1998) and that increasing cognitive demands imposed by social interaction may underlie this relationship. Overall, randomized trials or studies with longer periods of follow (preferably more than 10 years) are indispensable for reducing the possibility of reverse causality, in addition to the inclusion of a measure of baseline cognition as covariate in future analyses. The results of the three randomized trials already included in this review are mixed, which aligns with the few available MRI results of ongoing multi-component lifestyle intervention trials (Moon et al., 2022;Stephen et al., 2019Stephen et al., , 2020. Results from further ongoing clinical trials are indispensable for drawing firmer conclusions, including both multi-component interventions and (sub-) studies assessing intervention effects of CA and SA specifically.
The strength of association may also potentially differ by the period in life at which stimulation by CA and SA occurs. As for the risk of cognitive decline and dementia, it is suggested that midlife engagement (before brain changes associated with pathological ageing occur) may be particularly beneficial (Chan et al., 2018;Staff et al., 2018). However, evidence concerning differences in brain structure in association with specifically midlife CA and SA based on studies included in the current review, is mixed. More specifically, of the six studies retrospectively assessing midlife CA and SA, only two found a significant association with structural brain markers in late life (Valenzuela et al., 2008;Wirth et al., 2014). Moreover, none of the studies included in this review assessed early and midlife structural and functional social network characteristics in relation to late life MRI markers, making this an important point for future studies. Furthermore, an extensive body of research has linked other factors with a strong cognitively stimulating component, such as education and occupational complexity to both brain structure and cognition (Stern et al., 2018;Neth et al., 2020). While educational attainment has been controlled for in most included studies, the role of occupational complexity as a potential confounder should be further explored in future research.
Finally, the present review was confined to studies assessing structural brain differences. However, structural brain changes typically occur relatively slowly in mid-life, making large sample sizes and long periods of follow-up a prerequisite to detect the subtle effects of social health factors and CA (Buyanova and Arsalidou, 2021). In some studies included in this review length of follow-up was as brief as 3 years, which may be too short to uncover associations with brain structure. In addition, structural measures alone may not have sufficient sensitivity to clearly observe associations. In clinical populations, measures of functional brain connectivity in combination with structural measures have been shown to be more sensitive to network changes than structural measures on their own (Dai et al., 2019). They may also correlate with cognitive outcomes, and even potentially mediate the relationship between social health factors, CA and cognition (Stumme et al., 2020). Moreover, it is unlikely that social health factors and CA affect only single brain structures such as the hippocampus or the amygdala. Network-based MRI methods that also consider structural and functional interconnections between different brain regions may therefore be more suitable and should be examined by future studies.

Strengths and limitations
This systematic literature review and meta-analysis has various strengths including a broad and carefully selected keyword profile, a literature search conducted in various databases and no restriction concerning publication dates. It also has limitations. As outlined above, it was not possible to clearly disentangle unique associations between CA or SA and brain markers. Likewise, due to non-response of corresponding authors or unavailability of data, only a limited number of studies could be included in the meta-analyses. Given that data of three or more studies was necessary to allow for pooling, meta-analysis was unfortunately not possible for the majority of brain markers. The number of studies assessing predictors other than CA/SA was low, and studies were mostly cross-sectional and showed a high degree of diversity between measures. The latter may stem from a general lack of consensus regarding definitions and operationalization of social health factors and a common conceptual framework (Vernooij-Dassen et al., 2022b). It is likely that this, along with the heterogeneity in MRI-outcomes has also resulted in the non-identification of some important studies, such as Nakagawa et al. (2015), who examined associations between loneliness and regional white matter density using DTI. Lastly, given that this review was confined to healthy individuals for conceptual reasons, extrapolation of the results to patient populations may not be possible. Given these potential limitations, we adopted a fairly conservative viewpoint regarding the overall interpretation of results.

Conclusions
Social health factors and CA were related to measures of brain macro-and microstructure, but with low consistency and high methodological variability. While this may provide some evidence for their potential role as brain resilience factors, further research is indispensable, especially on structural and functional social network characteristics, the risk for reverse causation, the potential interrelatedness of social health factors and the role of other (functional) brain mechanisms.

Declaration of Competing Interest
The authors declare to not have any conflicts of interest. L.A. Duffner et al.