Risk and protective factors of healthy cognitive ageing across diverse global cohorts and causal effect of education

Risk and protective factors of healthy cognitive ageing across diverse global cohorts and causal effect of education | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Risk and protective factors of healthy cognitive ageing across diverse global cohorts and causal effect of education Heidi Johansen-Berg, Bonnie Yin Ka Lam, Vincent Mok, Henry Brodaty, and 25 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7195000/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The global rise in cognitive impairment calls for preventive strategies through early identification of risk and protective factors in the community healthy elderlies that take into account cultural and geographic diversity. This study investigates how risk and protective factors influence cognitive functioning and depressive symptoms of older adults across six diverse cohorts (n=1,636) from Europe, Asia, and Australia. We found that younger age at baseline and longer education covary with better baseline cognitive function, with a marginal average effect of education beyond individual, geographical, and birth cohort differences. Harnessing multimodal brain MRI, we find that this relationship is mediated by normalised grey matter, with a statistically significant pooled effect across cohorts. Using a natural experiment, we then establish the causal effect of education on cognition six decades later. By including underrepresented populations and by generalising findings, this research extends the evidence base beyond dominant Western-focused research norms, underscoring a call for inclusive and equitable access to education to enhance lifelong cognitive trajectories. Biological sciences/Neuroscience/Cognitive ageing Social science/Education Health sciences/Neurology/Neurological disorders/Dementia/Alzheimer's disease risk factors cognition depressive symptoms healthy aging neuroimaging canonical correlation analysis meta-analysis multi-cohort Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The proportion of the global population aged 60 and over is projected to reach 22% by 2050 (1). During this period, while global life expectancy is expected to increase by five years, healthy life expectancy is projected to rise by less than three years (2). According to the Organisation for Economic Co-operation and Development (OECD), this growing gap may drive healthcare costs to as much as 12% of the gross domestic product (3). Faced with increasing population ageing, there is a growing need to maintain brain health and cognition globally. Brain health can be impacted by multiple factors, including age, sex, education, cardiovascular and metabolic risk factors (e.g. hypertension, diabetes), lifestyle (e.g. exercise, smoking), diet, socioeconomic factors, air pollution, genetics, and culture (4-8). To develop better and cost-effective health policies fostering healthy cognitive brain health at the global level, it is of paramount importance to identify what are accessible and modifiable factors with potential causal effects beyond cultural, geographic, and background differences. While there is existing evidence of risk and protective factors for brain health, most previous neuroimaging studies are dominated by homogenous, predominantly White participants. This is because they are most often from high-income, western countries with predominantly European ancestry (9), such as the UK Biobank (10), Lifebrain (11), The Alzheimer's Disease Neuroimaging Initiative (ADNI) (12), The Lifespan Human Connectome Project (HCP) in Aging (13), and The Open Access Series of Imaging Studies (OASIS) (14). With very few exceptions (e.g. the ENIGMA consortium (15), most studies do not capture the full ethnic, geographical and cultural diversity of the global population (16). Because population characteristics as well as healthcare systems, education, lifestyle, diet, and management of vascular risk factors vary across different regions around the world, the limited demographic and cultural variability within predominantly European ancestry datasets limits the degree to which research findings can be generalised to a much more diverse global population (16-18). To better understand human brain health and allow greater generalisability of study findings, research programmes must include participants from diverse global backgrounds. In this work, we aimed to create a more diverse representation of study participants by combining existing cohorts from Europe, Asia and Australia. Particular challenges arise as large-scale, open-access neuroimaging datasets become a key global research resource that increasingly dominates the literature. While these offer powerful opportunities to advance our understanding, the biases present in those datasets also bring important limitations (16, 17). When associations found in one cohort are used to derive models that are tested on other cohorts, predictive power will be greatest when participants have a similar geographical, ethnic and cultural background to the original cohort. Consequently, the dominating models may be disproportionately influenced by particular cohort characteristics, thereby limiting generalisability of findings. In addition to the analytic approach, a few other examples related to research methods, including subject recruitment, and the use of research instruments can present inequities and have marked downstream effects on generalisability of the findings. For example, research tests are mainly available in English which creates a critical barrier in non-English speaking regions; and individuals with religious hair coverings are often excluded from MRI scanning (16, 17). Therefore, an obvious gap in the current research is the lack of cohort diversity that addresses demographic and cultural differences. In this work, we first aimed to identify how several risk and protective factors relate to longitudinal changes in cognition and depressive symptoms across diverse global cohorts. To do this, we used six unique cohorts spanning six regions (United Kingdom, Germany, Sweden, Hong Kong Special Administrative Region, Singapore, and Australia) and three continents (Europe, Asia, and Australia) with diverse cultures, ethnicities, socioeconomic factors, and healthcare systems, in addition to within-cohort differences. We aimed to identify covariation modes common across all six global cohorts, relating individual differences in demographic, risk and protective factors, and lifestyle factors, with individual differences in cognition and depressive symptoms, independent of cohort-specific differences. We hypothesise that different risk and protective factors are associated with distinct cognitive domains or depressive symptoms, and there may be at least one set of covariation modes that is common across all six cohorts. While it is equally interesting to understand the differences between cohorts, given the variations between cohorts, we consider it to be more meaningful as a first step to understand the similarities as well as marginal average effects across cohorts to derive healthcare policies that can be generalised across the globe. We then aimed to identify whether a precise set of neuroimaging biomarkers assessing distinct features of brain health mediate the covariation of risk and protective factors with cognition and depressive symptoms across the six global cohorts. Last, given the need for causal insight to draw better policies, we leveraged a natural experiment and a quasi-experimental method to examine the causal effect of education on long-term cognitive health. Results Variabilities in ageing cohorts from Europe, Asia, and Australia Data from individuals who were 60 years old or above were retrospectively collected from six cohorts with global geographical variation from Europe, Asia, and Australia. These included:1) a British cohort (Whitehall-II Imaging Sub-study) (19); 2) a German cohort (the Berlin Aging Study II; BASE-II) (20), 3) a Swedish cohort (Betula project) (21), which are all part of the Lifebrain consortium(http://www.lifebrain.uio.no/) (11); 4) a Hong Kong Chinese cohort (The Chinese University of Hong Kong – Risk Index for Subclinical brain lesions in Hong Kong; CU-RISK) (22), 5) a Singaporean Chinese cohort (Singapore Longitudinal Aging Brain Study; SLABS) (23); and 6) an Australian cohort (Sydney Memory and Aging Study; MAS) cohort (24). Participants within each cohort were primarily of the same national origin and predominantly from the same ethnic group (Supplementary Table 1). Based on inclusion and exclusion criteria (e.g. multimodal MRI at baseline, at least two timepoints of longitudinal cognitive assessment; see Methods for detailed information), a total of 1,636 subjects from the 6 global cohorts were included in the study. Risk and protective factors (age at baseline, sex, years of education, body mass index, frequency of exercise, smoking, presence of diabetes mellitus, hypertension, measure of systolic blood pressure and fasting blood glucose), cognition & depressive symptoms for baseline and follow-up, and neuroimaging measures were available across all 6 cohorts. The mean age across all cohorts was 71.3 ± 6.2 years old, the mean education years was 12.0 ± 4.9, and mean follow-up period was 3.4 ± 1.2 years. Comparison of cohort characteristics were shown using the analysis of covariances (ANCOVAs), controlling for time intervals between assessments ( Table 1 ). For example, there were significant differences in age of the cohorts, with the Australian cohort being the oldest (78.82 ± 4.38 years) and the British cohort being the youngest (68.2 ± 5.22 years); and significant differences in duration of education, with the Hong Kong Chinese cohort having the shortest education duration (8.08 ± 4.91 years) and the British the longest (14.87 ± 3.42 years) (see Table 1 ). While cognitive scores at baseline and follow-up were similar in most cohorts, there was a significant decline in general cognition score over time in the Australian cohort (baseline cognition: 28.26 ± 1.53; follow-up cognition: 27.70 ± 2.37; p<0.001, Cohen’s D = 0.26) and in the Hong Kong Chinese cohort (baseline cognition: 27.78 ± 2.61; follow-up cognition: 27.38 ± 3.12; p<0.001, Cohen’s D = 0.17) (see Supplementary Table 2). Overall, these cohort differences are expected because our study conducts a retrospective conjoint analysis of data that had been collected for other reasons using study-specific inclusion and exclusion criteria, hence resulting in variation in demographic and clinical characteristics ( Figure 1 ). Therefore, subsequent analyses focused on identifying common relationships across the six global cohorts whilst explicitly considering the hierarchical structure of the data. Common covariation modes link risk and protective factors with cognition and depressive symptoms Here we aimed to investigate whether sets of risk and protective factors covaried with cognition and depressive symptoms in a common, shared fashion across all six cohorts, above and beyond between-cohort differences. To test this hypothesis a grand canonical correlation analysis (CCA) was performed including all six cohorts. CCA is a symmetric, cross-decomposition method that characterises covariation modes between two sets of variables: broadly, risk factors and cognitive variables (25). It characterises pairs of latent projections (one per set of variables) that are maximally correlated with each other. Because age, years of education, other risk and protective factors, cognitive tests scores and depressive symptoms were collected separately for each cohort, all data collected were harmonised and individual scores within each cohort were converted to z-scores to respect the hierarchical structure of the data (see Methods). Furthermore, to perform valid statistical inferences while respecting dependencies given this hierarchical structure of the data (26), CCA was carried out with block-aware permutation testing only allowing subjects’ permutation within-cohort but not between-cohorts (27). The input to the CCA included two sets of variables: 1) risk and protective factors (ten variables): age at baseline, sex, years of education, fasting blood glucose level, body mass index, systolic blood pressure, hypertension, diabetes mellitus, smoking, and exercise; and 2) longitudinal cognition and depressive symptoms across two time-points (TPs; median follow-up of 3.43 years) (four variables): general cognition at TP1, difference score in cognition between TP2 and TP1, depressive symptoms at TP1 and difference score in depressive symptoms between TP2 and TP1. The raw difference scores were used without baseline adjustment to provide an unbiased estimate between the causal effect of education on the change in cognition (28). The time interval between TP1 and TP2 was included as a confound of no interest in the CCA. The grand CCA showed three shared, statistically significant modes of covariation ( Figure 2 , top panel), linking specific patterns of risk and protective factors with specific patterns of cognition and depression. These inferences were adjusted for the hierarchical structure of the study design - namely sampling across distinct cohorts - and the effects were consistent across all cohorts as shown by their covariation slopes ( Figure 2 , bottom panel). The 1st mode of covariation (ρ = 0.39; FWE-corr p = 0.001; n = 1,636; Figure 2a) linked younger age and longer education with better baseline cognitive performance ( Figure 3b). To further understand the strength of associations separately in each cohort, the cross-loadings of each cohort are plotted separately (Figure 3c) . These results showed that even though there is variability in cohort characteristics, and how the strength of covariation between the sets of risk & protective factors and sets of cognitive measures & depressive symptoms may differ between cohorts, the modes of covariation are significant, and consistent across the six cohorts. This pattern did not change when assessing the conditional unique effects using a multivariate regression model. A caveat of these results is that the current approach cannot distinguish whether age acts as an effect of the older generation with a more deprived childhood or as a biological ageing process. However, a key implication of these findings is that while age is not modifiable, education can be made a universally accessible modifiable factor. The 2nd mode of CCA covariation (ρ = 0.13, FWE-corr p = 0.002; Figure 2b ) linked lower body mass index, with more cognitive decline over time ( Supplementary Figure 1a & 1b ). The 3rd mode (ρ = 0.12, FWE-corr p = 0.021; Figure 2c ) linked less regular exercise with more depressive symptoms at baseline ( Supplementary Figure 1c & 1d ). Due to the relatively lower canonical correlations observed in modes 2 and 3 (ρ = 0.13 & ρ = 0.12, respectively), interpretation would be less meaningful (29). Therefore, the remaining of the manuscript will focus only on mode 1 as it is more effective to target the common and strongest associations among cohorts. Next, as a control test, we aimed to assess whether at the population-level there was evidence of an average treatment effect of education on baseline cognition alone whilst taking into account the fact that the effect of education may vary by cohort. To do this, first we fitted a multilevel mixed-effect model with random cohort slopes for education, birth cohort fixed-effects, and adjusting for all other covariates. This allows to estimate the effect of education while allowing it to vary between cohorts for the purpose of generalisation beyond the cohorts used here, whilst controlling for unobserved generational differences between birth cohorts, and conditioning on all other covariates. Model comparison showed that a mixed-effect model is superior to a fixed-effect model. Model outputs showed a small but non-zero variation in the effect of education across cohorts as well as a negative slope–intercept correlation. Of interest the model also showed a negative slope–intercept correlation, meaning that the effect of education is stronger at lower levels of cognition. Second, we estimated marginal (average) effects for education by averaging over all covariates using G-computation. This allows to estimate not conditional but the marginal effect of education beyond inter-individual differences in covariates. This analysis showed a positive and statistically significant marginal effect of education on baseline cognition (average treatment effect (ATE) = 0.06; p < 0.0001; n = 1,636), showing how a one-year increase in education would result in 0.06 standard deviation increase in baseline cognition as assessed via the Mini-Mental State Examination (MMSE) at age 70. However, a key limitation of these estimates is the absence of early-life or adulthood socioeconomic status measures which are likely to confound the relationship between education and cognition (30). Unfortunately, this is due to the multi-cohort nature of the study, e.g. classifications of socioeconomic status cannot easily be harmonised across cohorts. To overcome this problem, we carried out a sensitivity analysis aimed at establishing how strong an unmeasured confounder would have to be to explain away the observed effect. On average, each of our cohort shows a range of 4 years between those subjects with highest and lowest education. We, therefore, tested how strong should an unobserved variable be to wipe out the effect of education on cognition here observed. To do this we computed an E-value: the minimum strength of association (on a risk ratio scale) that unmeasured confounder(s) would need to have to fully explain away the effect observed, conditional on the covariates. This analysis showed that fully explain away the observed effect an unmeasured confounder would need a risk ratio greater to 1.26 with both education and cognition and across all cohorts, and whilst accounting for generational effects. This suggests that the observed effect of education has moderate robustness to confounding. The results did not change when adjusting also for depressive symptoms. Furthermore, when repeating these analyses using CCA cognitive scores (instead of the baseline cognition), the results did not change qualitatively (ATE = 0.05; p = 0.0006; n = 1,636; E-value = 1.24). Together, these findings show a marginal effect of education that is unlikely to be driven by unmeasured confounder(s) and that is generalisable beyond cohort characteristics. Between-cohorts meta-analysis of multimodal brain mediators Next, we sought to establish whether, consistently across all six global cohorts, candidate MRI markers of brain health mediated the link of longer education and younger age with better cognitive performance (CCA mode 1). Based on previous studies, we hypothesised four neuroimaging markers that have consistently shown sensitivity in detecting individual differences in ageing populations: normalised grey matter volume (31, 32); normalised hippocampal volume (32-34); white matter hyperintensity volume (31, 35-37) and white matter mean diffusivity (22, 37-40). However, by necessity this study involved multi-site MRI data which were acquired using different MRI hardware and with different acquisition protocols. MRI variability is still not a fully resolved issue in multi-site cohort studies and, therefore, we decided not to harmonise MRI data between cohorts. Instead, we used gold-standard neuroimaging processing pipelines (see Methods) and performed MRI statistical analyses separately for each cohort, then combined results across cohorts using a meta-analytic approach. The meta-analysis showed that for CCA mode 1, baseline normalised grey matter volume mediated the association with a small but statistically significant pooled effect across cohorts (Z=3.11; I² = 0%; p = 0.002; Figure 4 ). This shows that across cohorts, greater grey matter volume was a significant brain health mediator of the relationship between longer childhood education, younger age at baseline and greater cognition, independent of geographical or cultural differences between cohorts. The normalised hippocampal volume was also a significant mediator in CCA Mode 1 but did not survive Bonferroni correction ( p = 0.05/4 = 0.0125). A natural experiment to estimate the long-term causal effect of education Studying the causal effect of education on long-term cognition may be challenging due to endogeneity problems. Unobserved factors, such as socioeconomic factors, family environment, innate abilities, or secular trends, may influence both the time spent in education and cognitive development, therefore, biasing the estimation of the causal effect of education on cognition. To overcome this problem and to test for a causal effect of education, we took advantage of a natural experiment, exposure to the 1957-1958 “Asian influenza” pandemic during school age, as an exogenous shock that resulted in variation in educational experience independent of individual characteristics. The “Asian influenza” pandemic was first reported in April 1957 in Hong Kong (41). It emerged unexpectedly affecting 10% of the population, with a particularly pronounced impact on school-aged children (42) who experienced the highest influenza attack rates (69%–78%; much higher than adults, 19%–24%) (43) and yet the lowest excess mortality rates (42), hence generating sharp predictions for long-term effects. Crucially, in 1957 Hong Kong, primary and secondary education were neither universal nor guaranteed, with children enrolling in primary and secondary schools respectively at age 6 and 12, and with strong gender differences and societal norms. Previous research demonstrated how perinatal exposure to the 1918 “Spanish” influenza pandemic in the U.S. led to significant reductions in educational attainment and long-term health sequelae (44). Here, we leveraged the sudden exposure to the “Asian influenza” pandemic during the critical transition between primary-to-secondary school period as an exogenous instrument that influenced education attainment. We hypothesised that pandemic exposure in children at age 12, which is during the critical primary-to-secondary school transition, led to reduced educational attainment compared to other birth cohorts, which, in turn, resulted in dose-response negative effects on cognitive health six decades later. Because of gender differences in societal norms and labour market opportunities during the study period (Hong Kong, 1957-58), and because of marked gender differences in educational attainment present in the data, we focused only on males. To test our hypothesis, we used an instrumental variable (IV) approach (45), addressing both the endogeneity problem and allowing us to test for the causal effect of education drop on cognitive health at around 70 years old. First, as hypothesised, a marked decrease in years of education was present in the birth cohort of males aged 12 in 1957 when compared to nearby birth cohorts and females, who showed no reduction in education years. This was apparent even in the raw data, with this group dropping from an average of 10 years of education seen in adjacent cohorts, to an average of just under 7 years in the cohort aged 12 in 1957 ( Figure 5 ). This difference showed a significant and large negative effect size (Cohen’s d = 3; p = 0.003, assessed via permutation testing of general linear models whilst adjusting for date of birth). Curtailment of education for military service cannot be the reason for this variation as Hong Kong has never implemented military conscription. Second, the 1957 influenza pandemic itself is unlikely to directly influence long-term cognitive health outcomes six decades later, except through its impact on educational attainment: (i) Short-term health effects of exposure in childhood (e.g. respiratory illness) were unlikely to have persistent direct effects on cognitive health sex decades later and, even if they did, they would not be specific to the cohort of interest aged 12 in 1957. (ii) Exposure to pandemics is known to reduce educational attainment (44). (iii) Economic and social pressures from the pandemic likely influenced household decisions regarding children's schooling. Thus, both instrumental variable assumptions — the strong first stage and exclusion restriction — are satisfied. The IV analysis revealed a positive and significant causal effect of years of education on general cognitive abilities assessed at around 70 years of age, hence six decades later. Specifically, the first-stage IV statistics confirmed the relevance of the instrument (F = 9.44, p = 0.003, adjusted R-squared=0.07). Then, using exposure to the 1957 influenza pandemic in males aged 12 as an instrument for the endogenous treatment, years of education, the IV estimate indicated that each additional year of education caused a 1.15% increase (~0.2 standard deviation change) in cognitive functioning, above and beyond the effect of age. The results were robust to covarying for linear and quadratic trends in birth year, age-adjusted time of assessment, and to rank-inverse transforming the data (IV Estimate: 0.33; 95% CI: 0.09–0.88-unit, p = 0.021; n=110). All second-stage IV robustness estimator tests (LIML, TSLS, Fuller), as well as alternative tests (Anderson-Rubin, and Conditional Likelihood Ratio tests), supported the validity of the findings providing similar results with confidence intervals for the causal effect entirely above zero. Further analyses confirmed no direct effect of the instrument on the outcome when adjusting for treatment (significant) and confounds, hence providing further support for the exclusion restriction. Crucially, the same IV analysis did not show any significant effects of education on cardiovascular (high blood pressure, second-stage IV p = 0.58) or metabolic health (diagnosis of diabetes mellitus, second-stage IV p = 0.28). These findings highlight the long-term causal role of education in supporting healthy cognitive functioning in old age, even after accounting for aging and potential endogeneity issues. Discussion Here we identified how modifiable lifestyle factors relate to healthy cognitive ageing across diverse global cohorts spanning six regions (United Kingdom, Germany, Sweden, Hong Kong Special Administrative Region, Singapore, and Australia) and three continents (Europe, Asia, and Australia), characterised by diverse cultures, ethnicities, socioeconomic factors, and healthcare systems, in addition to between-cohort differences. We first identified globally shared patterns of covariation and highlighted the marginal effect of education on cognitive healthy ageing, measured by cognitive scores and depressive symptoms. By harnessing non-invasive multimodal brain MRI, we then identified brain structure that mediates the risk and protective factors and longitudinal changes in cognition and depressive symptoms across diverse global cohorts. Last, by leveraging a natural experiment, we showed the causal effect of education highlighting the critical importance of education as a modifiable factor to improve brain healthy ageing globally. By incorporating diverse global cohorts that reflect a wide range of ethnic and cultural backgrounds, focusing on accessible and modifiable lifestyle factors, and ensuring that findings are robust and generalisable to broader, mixed populations, this work meaningfully advances diversity in brain health research. One key motivation for this work was to identify modifiable lifestyle factors that can be targeted in global public health policies to improve healthy brain ageing. To do this, and to explore the complex picture of human behaviour and population diversity across geographical contexts, we leveraged a large pool of risk and protective factors across six diverse global cohorts ( Figure 1 ), together with longitudinal assessments of cognition and depressive symptoms. We found three, statistically significant CCA modes of covariation, independent of underlying geographic and cultural differences ( Figure 2 ). The strongest mode of covariation linked younger age at baseline and longer education with better baseline cognitive performance ( Figure 3 ). Because age is a non-modifiable factor, follow-up analyses focused on education whilst adjusting for birth cohort fixed effects to control for unobserved generational differences. This showed evidence for a statistically significant marginal treatment effect on the CCA cognitive outcomes, holding other variables constant. This means that, on average across the mixed populations, each additional year of education is associated with an average increase of 1.06% in cognitive outcomes beyond the dominant effects of birth cohorts, sex, smoking status, cardiovascular and metabolic health, or other background characteristics. This is highly relevant for policy making and public health because, unlike conditional regression effects - which apply to individuals with specific characteristics - here marginal effects reflect the average, causal impact of education across the global populations, beyond background characteristics. One important note though is that greater years of education was mainly associated with better baseline cognitive scores but not the change in cognition. The association between higher years of education is associated with higher levels of cognition in early adulthood is well established, as least in previous studies from WEIRD (Western, Educated, Industrialised, Rich and Democratic countries). However, whether education is associated with the rate of change in cognition is highly debated and empirical evidence from longitudinal meta-analysis did not reveal a consistent and substantial association (46). Some suggested that education in early adulthood protects the brain from cognitive decline at old age (47) while some longitudinal studies showed that education is not related to individual differences in rates of cognitive change (48) (49) (50) (51). A mega-analysis from longitudinal cohorts across 33 countries (mainly Western countries) showed education was associated with better memory, larger intracranial volume and slightly larger brain region relating to memory but it did not protect against rate of cognitive decline (9). While this current study and the large large-scale longitudinal cohort have overlapping cohorts (the three European cohorts included in this study are part of the Lifebrain consortium (11), this study included two Asian cohorts and one Australian cohort, thus, provided validation across independent cohorts with substantial variability in demographics, cultural contexts, and even healthcare systems. These findings underscore the importance to support the role of education across global populations, and that education was mainly associated with better baseline cognitive scores but not the change in cognition, and most importantly, this association holds beyond western-centric research evidence. A second objective of the study was to leverage multi-site neuroimaging to identify brain metrics that mediate between risk and protective factors and longitudinal changes in cognition and depressive symptoms across diverse global cohorts. Because MRI variability is still not a fully resolved issue in multi-site cohort studies, instead of harmonising MRI data between cohorts, we opted for performing causal mediation analysis separately within-cohort followed by pooling across-cohorts by feeding these estimates of indirect mediation effect into a meta-analysis. Among the neuroimaging variables investigated (normalised grey matter volume, normalised hippocampal volume, white matter hyperintensities volume, and white matter mean diffusivities), there is a statistically significant pooled effect for normalised grey matter volume (and for normalised hippocampal volume, although the latter did not survive Bonferroni correction across MRI candidates tested). These findings show that normalised grey matter volume was a common, significant mediator ( Figure 4 ) of the relationship between younger age and longer education with healthier cognitive trajectories. Our findings echo previous studies exploring brain mediators between ageing and cognition (34, 52-54), but crucially they also extend beyond the existing literature. First, by combining datasets from three continents, these findings offer improved generalisability of scientific insight across diverse populations. Second, by using difference scores (without adjustment to baseline) to investigate longitudinal changes, this allowed us to provide an unbiased view between the effect of education on change in cognition and depressive symptoms. Third, by employing a multi-site MRI approach and demonstrating statistically significant pooled effects across cohorts, the results presented here are robust also against variability that could otherwise be ascribed to differences in MRI protocols or potential incomplete harmonisation of MRI imaging data. Together, these neuroimaging findings highlight a potential modifiable pathway that can be targeted to promote healthier trajectories of cognitive and mental health across diverse populations. Because the effect of education can be confounded by the effect of secular improvements in the socioeconomic context across time as well as by endogeneity biases, a third objective of the study was to establish the causal effect of education beyond these issues. To do this we used instrumental variable analysis to take advantage of the quasi-random exposure to the 1957-1958 “Asian influenza” pandemic during school age in Hong Kong as an exogenous shock that resulted in variation in educational experience independent of individual characteristics. Although greater education is often linked with better brain and cognitive health in ageing (55), this association is often confounded by early-life socioeconomic factors, health, and genetics (56, 57). Here, using exposure to the pandemic influenza at age 12 as an instrument, this approach allowed us to overcome these endogeneity problems. Leveraging a model previously designed to study the long-run consequence of exposure to the 1918 influenza pandemic (44), our results showed the protective, causal effect of education on cognitive health six decades later ( Figure 5 ), even after accounting for smooth linear and quadratic trends in birth cohort and age at assessment (relative to birth year). Supporting our results, Davies et al. used a natural experiment exploiting national changes in school policy to show that longer education causally reduces the risk of cardiovascular and metabolic diseases, and mortality (57). These findings strengthen the case for education—a modifiable lifestyle factor—as a true protective factor to support healthy cognitive aging. These results, therefore, have significant implications for informing equitable healthcare policies. A clear limitation of this work that must be addressed in future studies is that our cohorts were opportunistically selected based on data availability. While they cover diverse populations across three continents, several world geographies and ethnicities are not represented. This is particularly the case for low- and middle-income countries, in which two-thirds of the world's population aged 60 years or above reside (1). Furthermore, among the cohorts that we assessed, participants were primarily of the same national origin and/or predominantly from the same self-reported ethnic group. However, this information was not explicitly collected in all cohorts and/or, if available, it was not standardised. This limitation highlights a broader need for guidelines, practices, and standards for data collection and harmonisation across global cohorts. Common and international standards, relatable to those already in use by other international organisations, e.g. the OECD, would help the neuroimaging community to move forward and directly address questions with real-world public health implications. Conclusions By emphasizing diversity in study design, research objectives, and analytical approaches, and by focusing on accessible modifiable factors, this work takes an initial step towards addressing the gap left by traditional Western-focused research norms. This work provides important evidence for the causal role of education in lifelong cognitive health. These findings should contribute to evidence-based policy making, with the aim of ultimately improving healthy brain ageing globally while reducing health disparities. Methods Cohort samples Data from individuals who were 60 years old or above were retrospectively collected from six cohorts with geographical variations. The community cohorts included:1) a British cohort (Whitehall-II Imaging Sub-study) (19); 2) a German cohort (the Berlin Aging Study II; BASE-II) (20), 3) a Swedish cohort (Betula project) (21), which are all part of the Lifebrain consortium(http://www.lifebrain.uio.no/(11)). In addition, we included 4) a Hong Kong Chinese cohort (The Chinese University of Hong Kong – Risk Index for Subclinical brain lesions in Hong Kong; CU-RISK) (22), 5) a Singaporean Chinese cohort (Singapore Longitudinal Aging Brain Study; SLABS) (23); and 6) an Australian cohort (Sydney Memory and Aging Study; MAS) cohort (24). Participants in each cohort were primarily of the same national origin and predominantly from the same ethnic-racial group. For more details on cohort information, please see Supplementary Table 1. The work described has been carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) and has been approved by the Joint Chinese University of Hong Kong-New Territories East Cluster ethics committee (CREC Ref. No. 2022.119). The inclusion criteria were i) community-dwelling individuals aged over 60 years old; ii) who had demographic information and clinical data on risk and protective factors; iii) at least two-time points of measurement on cognition and depressive symptoms, preferably with 3-4 years of assessment interval; iii) with at least one-time point of MRI with T1-weighted, fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted sequences available. The exclusion criteria were i) subjects with stroke, dementia, and other neurological diseases at baseline; ii) those without longitudinal cognitive and depressive symptom measures; and iii) those without MRI data available. Details of neuropsychological assessments and depressive symptoms measured for each cohort are presented in Supplementary Table 4. While not all cohorts captured all cognitive domains, there was variation in the precise measures used, and therefore, some harmonisation of measures was required (see data harmonisation section below). Risk and protective factors For demographic information, age at baseline, sex, years of education, ethnicity, country of origin, body mass index, and socioeconomic status were included. For vascular risk factors, information on the pulse, systolic blood pressure, diastolic blood pressure, fasting blood glucose, glycated haemoglobin, triglyceride level, high-density-lipoprotein level, low-density-lipoprotein level, total cholesterol level, hypertension, diabetes mellitus, and hyperlipidaemia were collected. Regarding behavioural and lifestyle factors, physical activities, functional dependence, smoking habits, and information on alcohol consumption were collected. Motor function data were captured using the functioning timed walk and 10-meter Walk Test. Information on intelligence quotient was collected. Assessments on cognition and depressive symptoms For cognitive data, the Mini-Mental State Examination (MMSE) or Montreal Cognitive Assessment (MoCA) was used to measure general cognition. In addition, specific cognitive domains including attention and processing speed, executive function, memory, visuospatial function, and language were considered if available. The German version of the MMSE used in the BASE-II study was indicated with a special note, please refer to Supplementary Table 4. To assess depressive symptoms, data were captured using the Center for Epidemiologic Studies Depression Scale (CES-D) (cutoff at ≥ 16) (36) or the Geriatric Depression Scale (GDS) (cutoff at ≥ 5) (58). In addition to the baseline measures (TP1) in cognition and depressive symptoms, the follow-up measures (TP2) were collected. To assess longitudinal change in cognition and depressive symptoms, the difference scores between TP2 and TP1 were used (28). The dates of assessment of all tests were recorded and the time interval between baseline and longitudinal assessments were calculated. Although data on specific cognitive domains were requested, not all cohorts collected data relating to the same domain or there was no common neuropsychological test to represent specific cognitive domains that can be harmonised across all six cohorts. Similarly, for motor function and intelligence quotient, not all cohorts obtained this measure (see Supplementary Table 4). Therefore, only variables with complete data that can be harmonised across cohorts were used in the canonical correlation analysis (see Supplementary Table 5a-5c). Missing data were excluded from subsequent analyses. Magnetic resonance imaging (MRI) MRI data were collected and processed for the six cohorts separately. MRIs from Whitehall II were acquired at the University of Oxford using two scanners: a 3T Siemens Magnetom Verio scanner and a 3T Siemens Prisma scanner. MRI from Betula was acquired from Umeå University using a 3T Discovery MR750 scanner (General Electric, Milwaukee, USA). In BASE-II, a 3T Siemens Tim Trio scanner was used (Siemens, Erlangen, Germany). MRI from CU-RISK was performed at the Prince of Wales Hospital Hong Kong with a 3T Philips Achieva TX (Philips Medical Systems, Best, The Netherlands). MRI from SLABS was acquired at the Duke-NUS (which is an independent school of the National University of Singapore) using a 3T Siemens Tim Trio scanner (Siemens, Erlangen, Germany). MRI from MAS was acquired from the University of New South Wales using a Philips 3T Achieva Quasar Dual scanner (Philips Medical Systems, Best, The Netherlands). All cohorts collected T1-weighted and diffusion-weighted images. The FLAIR images at waves 5 & 6 of the Swedish cohort were available but inaccessible at the time of analyses, hence, white matter hyperintensity volume was not assessed in this cohort. Some cohorts additionally collected other MRI modalities that are not considered here (see Supplementary Table 6). MRI analysis In brief, the T1-weighted, FLAIR, and diffusion-weighted sequences that matched the baseline clinical and cognitive assessments were used for analyses. The MRI data were processed separately for the six cohorts. All images were de-identified before analyses and processed using FSL v6.0 tools (59). T1-weighted scans were pre-processed with bias-field correction, brain extraction using non-linear registration-based masking, tissue-type segmentation, and subcortical segmentation. The total brain, grey matter were estimated using FSL-Sienax (60). Sienax estimated total brain tissue and grey matter volumes from a single image, normalised for skull size. In brief, the brain was extracted, then the brain and skull images were used to estimate the scaling between the subject's image and standard space. Further, issue segmentation was performed to estimate the brain tissue volume, and multiplies this by the estimated scaling factor, thus, reducing head-size-related variability between subjects. The hippocampal volume was generated using FSL-FIRST and corrected for head size (61). FLAIR images were brain extracted and processed with bias-field correction using FSL-FAST (62). We used FSL-BIANCA (63). A subset of 20 manually segmented WMH images (from the CU-RISK study) were provided as the training data for all the cohorts. During image visual inspection, for subjects with moderate WMH that affects T1 image tissue segmentation (e.g. WMH misclassified as grey matter), a WMH lesion mask generated from FSL-BIANCA was used to fill the lesion before tissue segmentation was done. For DWI images, data were pre-processed to correct for susceptibility-induced distortions using topup (64), and for eddy currents as well as subject head movements using eddy (65). For cohorts without the reverse phase-encoding direction, we applied Synb0-DISCO v2.0 (66), a tool which synthesises an “undistorted” b0 image based on the geometry of the given structural T1-weighted scans. The UK Biobank FA template was used as the group template (10). Mean diffusivity (MD), which is the average MD within the skeleton, was obtained using DTIFIT and TBSS (TBSS v 1.2 (67). Data harmonisation Clinical, and data on cognition and depressive symptoms were collected separately from each cohort. All data were harmonised, and individual test scores were converted to z-scores within each cohort for further analysis. In brief, we categorised neuropsychological tests based on the respective domains measured. For example, for general cognition, either the Mini-mental state examination (MMSE) or Montreal cognitive assessment (MoCA) was used. Since only one cohort used the MoCA while the other five cohorts used MMSE, the MoCA score of that cohort was converted to MMSE (68). These raw test scores were converted to z-scores and used in the measure for “general cognition”. In addition, specific cognitive domains including attention and processing speed, executive function, memory, visuospatial function, and animal fluency were requested if available. However, specific cognitive domains were not included in the CCA models as there was no cognitive domain that was able to be harmonised across all six cohorts. The neuropsychological tests used for general cognition and specific cognitive domains in each cohort were summarised (see Supplementary Table 4). For depressive symptoms, the Center for Epidemiological Studies-Depression Scale and Geriatric Depression Scale were the two main tests used across all 6 cohorts. It is impossible to completely harmonise the two tests, but the test score was converted to a z-score within the cohort for analysis instead. Harmonisation of the different neuropsychological tests was based on common practice and with reference to a previous publication (69). For detailed information on the specific tests and cut-offs used in each measure specific to the cohort, please refer to Supplementary Table 5. This study involves multi-site MRI data which were acquired using different MRI machines and with different acquisition protocols (see Supplementary Table 6). MRI variability is still not a fully resolved issue in multi-site cohort studies. We did not attempt to harmonise the MRI data collected across cohorts. Instead, we used a standard neuroimaging analysis pipeline and performed MRI analysis for each cohort individually (instead of combining imaging data across cohorts) when assessing statistical associations. The advantage of adopting such a per cohort analysis approach is that it allows us to investigate how neuroimaging measures contribute directly and specifically in each cohort, and to avoid issues arising from potential incomplete harmonisation of MRI imaging data. Causal mediation analysis was then performed separately for each cohort, testing whether MRI candidate markers (M) mediate the effect between risk and protective factors (X) and cognition and depression (Y). Causal mediation outputs per each cohort (coefficient of indirect effect) were fed into a meta-analysis to test our hypothesis of neuroimaging mediation at the between-cohorts level. Statistical analyses Canonical correlation analysis (CCA) Canonical correlation analysis (CCA) is a multivariate approach to identify covariation between two sets of variables. In this study, our first aim is to investigate the underlying link between risk & protective factors and cognition & depressive symptoms using a CCA with permutation inference (25, 27). The input data included two sets of variables – 1) risk and protective factors (ten variables – age at baseline, sex, years of education, fasting blood glucose level, body mass index, systolic blood pressure, hypertension, diabetes mellitus, smoking, and exercise); and 2) cognition and depressive symptoms measures (four variables - general cognition at TP1, difference in general cognition between TP2 and TP1, depressive symptoms at TP1, and difference in depressive symptoms between TP2 and TP1. In the CCA model, we investigated the associations between risk & protective factors and cognition & depressive symptoms, without the use of neuroimaging variables. However, only subjects with complete clinical data, longitudinal data on cognition and depressive symptoms, as well as MRI data were included in the CCA analysis. The intention here was to match the group of subjects being used in the CCA and subsequent mediation analyses using MRI data such that there is a more accurate assessment of the mediating effect of neuroimaging markers in the covariation derived from the CCA. All variables were converted to z-score within each cohort and all data from the six cohorts were used in the CCA, which effectively adjusted for cohort differences. The time interval between baseline assessments and follow-up assessments was included as covariates. Here CCA would allow to identify latent modes of covariation which link the two sets of variables across the six cohorts, while quantifying linear combinations of risk and protective factors that maximally correlate with linear combinations of cognition and depression scores. These linear combinations are known as CCA variate pairs. Significance of CCA modes was determined by nonparametric inference testing with 1,000 permutations allowing shuffling of subjects only within - but not between - cohorts to take into account the cluster structure of the cohorts (k = 6). Multiple comparisons across CCA modes were corrected via family-wise error rate correction (FWE-corr). CCA cross-loadings, quantifying the involvement of each original variable in the CCA variate pairs, were then calculated for all CCA modes deemed significant at FWE-corr p < 0.05 (27, 70). Furthermore, to estimate unbiased CCA loadings, we used a multilevel mixed-effect model with random slopes, hence allowing effects to vary between cohorts for the purpose of generalisation. We also used birth cohort fixed effects to control for unobserved generational differences. Age cohort fixed effects controlling for life-cycle differences were not used however as birth year and age at assessment are highly correlated. Last, for the purpose of generalizability again, we estimated CCA cross-loadings as marginal (average) treatment effects by marginalising over the distribution of the covariates across the global cohorts. Marginal effects were calculated as average, partial derivatives of the regression equation via the R port of Stata's ‘⁠margins⁠’ command (71). Bonferroni-Holm correction was used to control for familywise error rate, the probability of at least one type I error across predictors (30). Mediation analysis To assess the mediating effect of neuroimaging markers in the covariation of risk and protective factors with cognition and depressive symptoms, a mediation analysis was performed separately for each cohort. The CCA variate of risk and protective factors was used as the independent variable while the CCA variate of cognition and depressive symptoms was used as the dependent variable. Based on prior evidence, we hypothesise that the brain mediators for changes in cognition and depressive symptoms will be underpinned by distinct brain regions such as the grey matter volume (31, 32), hippocampal volume (32-34), white matter hyperintensity volume (31, 35-37), and white matter mean diffusivity (22, 37-40). As we use the nonparametric percentile bootstrap method in mediation analysis (which uses random resampling with replacement to estimate a population parameter), the outputs are the direct effect, indirect effect (used in the meta-analysis), total effect, and proportion mediated (72). The time interval between baseline assessments and follow-up assessments was included as covariates. The R 'mediation' package was used for mediation analysis. Meta-analysis A meta-analysis was conducted to obtain comprehensive summary statistics of the neuroimaging mediation effects across all six global cohorts. Separately for each CCA mode deemed significant, the indirect effect estimates from the mediation analysis linking between the risk and protective factors (X) with cognition and depressive symptoms (Y) via brain MRI markers (M) in CCA mode 1 were used in the meta-causal mediation analysis. Heterogeneity was evaluated using Cochrane’s Q test and I 2 statistics. If there was a significant heterogeneity (p 50%), a random effect model was used in the meta-analysis. Otherwise, a fixed-effect model was applied. A Bonferroni correction was applied to adjust for multiple testing across neuroimaging markers: p-value = 0.05 / 4 neuroimaging markers = Bonferroni p-value = 0.0125. The meta-analysis was performed using the R package 'metafor'. Instrumental Variable (IV) analysis An IV analysis was conducted to address possible endogeneity problems in the relationship between the endogenous variable (education) and the outcome (cognition) and to assess causality. The IV was applied only to the Hong Kong cohort as this was the only cohort where education was neither universal nor guaranteed in 1957 and with sufficient sample size to conduct the analysis. (In 1957 Singapore education was also neither universal nor guaranteed. However, the Singapore cohort did not have enough subjects in the birth cohort of interest, e.g. subjects aged 12 in 1957 n = 3). Exposure was defined as being 12 year old in 1957 hence - contrary to ‘graduation year’ - this is exogenous and only determined by birth year. The outcome model for the IV analysis was designed after Almond (2006) (44) with exposure controlled for smooth linear and quadratic trends in birth cohort and age-adjusted time of assessment (relative to birth year). This approach assumes that, in absence of the pandemic, the birth cohort of interest would have followed the same smooth trajectory in outcome, e.g. cognitive scores. This assumption is justified by the known smooth relationship between age and change in cognition. The IV analysis was performed using the R package 'ivmodel'. Declarations Declaration of interest H J-B is or has been an advisory board member or consultant to Biogen, Eisai, Eli Lilly, Medicines Australia, Roche and Skin2Neuron. He is a Medical/Clinical Advisory Board member for Montefiore Homes and Cranbrook Care. MWLC is an advisor to Oura Health and Quantactions. PSS was a paid member of Advisory Panels for Biogen and Roche Australia in 2020 and 2021. All other co-authors declare no conflicts of interest. Data availability statement The data and scripts that support the findings of this study are available on request from the corresponding authors, H.J-B and P.S. Raw data that belongs to other cohorts may only be available upon request from the cohort Principal Investigator. Acknowledgements BYKL was supported by the Lee Hysan Postdoctoral Fellowship in Clinical Neurosciences. MWLC was supported by the following sources: Yong Loo Lin School of Medicine (National University of Singapore, Singapore), The Lee Foundation (Singapore), National Medical Research Council Singapore (STaR May2019-001), (STaR/0013/2013) (STaR/0004/2008), Biomedical Research Council, Singapore (04/1/36/19/372). The Sydney Memory and Ageing Study was supported by three National Health & Medical Research Council (NHMRC) Program Grants (ID No. ID350833, ID568969, and APP1093083). MRI scans were supported by NHMRC Project Grants (510175 and 1025243) and an ARC Discovery Project Grant (DP0774213) and John Holden Family Foundation. JHZ was supported by the following sources: Singapore National Medical Research Council (NMRC/OFLCG19May-0035, NMRC/CIRG/1485/2018, NMRC/CSA-SI/0007/2016, NMRC/MOH-00707-01, NMRC/CG/435 M009/2017-NUH/NUHS, CIRG21nov-0007and HLCA23Feb-0004), RIE2020 AME Programmatic Fundf rom A*STAR, Singapore (No. A20G8b0102), Ministry of Education (MOE-T2EP40120-0007&. T2EP2-0223-0025, MOE-T2EP20220-0001), and Yong Loo Lin School of Medicine Research Core Funding, National University of Singapore, Singapore. H J-B was supported by a Wellcome Trust PRF (222446/Z/21/Z), LG was supported by an Alzheimer’s Association Grant (AARF-21-846366). This work was supported by the NIHR Oxford Health Biomedical Research Centre (NIHR203316 ) . The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. The Oxford Centre for Integrative Neuroimaging was supported by core funding from the Wellcome Trust (203139/Z/16/Z and 203139/A/16/Z). The Whitehall II Imaging Sub-study was supported by the UK Medical Research Council (MRC) grants “Predicting MRI abnormalities with longitudinal data of the Whitehall II Sub-study” (G1001354; PI KPE; Clinical Trials. gov Identifier: NCT03335696), the HDH Wills 1965 Charitable Trust (Nr: 1117747, PI: K.P.E), and the European Commission Horizon 2020 grant “Lifebrain” (732592, Co-PI KPE). The Betula study was supported by a grant to L.N. from the Knut and Alice Wallenberg (KAW) Foundation (Sweden).This article uses data from the Berlin Aging Study II (BASE-II). BASE-II was supported by the German Federal Ministry of Education and Research under grant numbers #01UW0808; #16SV5536K, #16SV5537, #16SV5538, #16SV5837, #01GL1716A, and #01GL1716B. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. References Organisation WH. Ageing and health 2022 [Available from: https://www.who.int/news-room/fact-sheets/detail/ageing-and-health. (IHME) IfHMaE. Global Burden of Disease 2021: Findings from the GBD 2021 Study. Seattle, WA: IHME; 2024. OECD. Fiscal Sustainability of Health Systems: How to Finance More Resilient Health Systems When Money Is Tight? Paris: OECD Publishing; 2024. Baumgart M, Snyder HM, Carrillo MC, Fazio S, Kim H, Johns H. Summary of the evidence on modifiable risk factors for cognitive decline and dementia: a population-based perspective. Alzheimer's & Dementia. 2015;11(6):718-26. Franz CE, Hatton SN, Elman JA, Warren T, Gillespie NA, Whitsel NA, et al. Lifestyle and the aging brain: interactive effects of modifiable lifestyle behaviors and cognitive ability in men from midlife to old age. Neurobiol Aging. 2021;108:80-9. Manuello J, Min J, McCarthy P, Alfaro-Almagro F, Lee S, Smith S, et al. The effects of genetic and modifiable risk factors on brain regions vulnerable to ageing and disease. Nature Communications. 2024;15(1):2576. Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet. 2020. Livingston G, Huntley J, Liu KY, Costafreda SG, Selbæk G, Alladi S, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. The Lancet. 2024;404(10452):572-628. Fjell AM, Røgeberg O, Sørensen Ø, Amlien IK, Bartrés-Faz D, Brandmaier AM, et al. Reevaluating the Role of Education in Cognitive Decline and Brain Aging: Insights from Large-Scale Longitudinal Cohorts across 33 Countries. medRxiv. 2025. Miller KL, Alfaro-Almagro F, Bangerter NK, Thomas DL, Yacoub E, Xu J, et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature neuroscience. 2016;19(11):1523-36. Walhovd KB, Fjell AM, Westerhausen R, Nyberg L, Ebmeier KP, Lindenberger U, et al. Healthy minds 0–100 years: Optimising the use of European brain imaging cohorts (“Lifebrain”). European Psychiatry. 2018;50:47-56. Petersen RC, Aisen PS, Beckett LA, Donohue MC, Gamst AC, Harvey DJ, et al. Alzheimer's Disease Neuroimaging Initiative (ADNI): clinical characterization. Neurology. 2010;74(3):201-9. Bookheimer SY, Salat DH, Terpstra M, Ances BM, Barch DM, Buckner RL, et al. The Lifespan Human Connectome Project in Aging: An overview. Neuroimage. 2019;185:335-48. Kurdi B, Lozano S, Banaji MR. Introducing the Open Affective Standardized Image Set (OASIS). Behav Res Methods. 2017;49(2):457-70. Thompson PM, Stein JL, Medland SE, Hibar DP, Vasquez AA, Renteria ME, et al. The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav. 2014;8(2):153-82. Ricard JA, Parker TC, Dhamala E, Kwasa J, Allsop A, Holmes AJ. Confronting racially exclusionary practices in the acquisition and analyses of neuroimaging data. Nature Neuroscience. 2023;26(1):4-11. Cardenas-Iniguez C, Gonzalez MR. Recommendations for the responsible use and communication of race and ethnicity in neuroimaging research. Nature Neuroscience. 2024;27(4):615-28. Henrich J, Heine SJ, Norenzayan A. The weirdest people in the world? Behavioral and Brain Sciences. 2010;33(2-3):61-83. Filippini N, Zsoldos E, Haapakoski R, Sexton CE, Mahmood A, Allan CL, et al. Study protocol: the Whitehall II imaging sub-study. BMC psychiatry. 2014;14(1):1-16. Bertram L, Böckenhoff A, Demuth I, Düzel S, Eckardt R, Li S-C, et al. Cohort Profile: The Berlin Aging Study II (BASE-II)†. International Journal of Epidemiology. 2014;43(3):703-12. Nilsson L-G, BÄCkman L, Erngrund K, Nyberg L, Adolfsson R, Bucht G, et al. The betula prospective cohort study: Memory, health, and aging. Aging, Neuropsychology, and Cognition. 1997;4(1):1-32. Lam BYK, Leung K, Yiu B, Zhao L, Biesbroek J, Au L, et al. Peak width of skeletonized mean diffusivity and its association with age-related cognitive alterations and vascular risk factors. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring 2019;11:721-9. Chee MWL, Chen KHM, Zheng H, Chan KPL, Isaac V, Sim SKY, et al. Cognitive Function and Brain Structure Correlations In Healthy Elderly East Asians. NeuroImage. 2009;47:S175. Sachdev PS, Brodaty H, Reppermund S, Kochan NA, Trollor JN, Draper B, et al. The Sydney Memory and Ageing Study (MAS): methodology and baseline medical and neuropsychiatric characteristics of an elderly epidemiological non-demented cohort of Australians aged 70–90 years. International psychogeriatrics. 2010;22(8):1248-64. Smith SM, Nichols TE, Vidaurre D, Winkler AM, Behrens TE, Glasser MF, et al. A positive-negative mode of population covariation links brain connectivity, demographics and behavior. Nature neuroscience. 2015;18(11):1565-7. Winkler AM, Webster MA, Vidaurre D, Nichols TE, Smith SM. Multi-level block permutation. Neuroimage. 2015;123:253-68. Salvan P, Wassenaar T, Wheatley C, Beale N, Cottaar M, Papp D, et al. Multimodal Imaging Brain Markers in Early Adolescence Are Linked with a Physically Active Lifestyle. The Journal of Neuroscience. 2021;41(5):1092. Glymour MM, Weuve J, Berkman LF, Kawachi I, Robins JM. When is baseline adjustment useful in analyses of change? An example with education and cognitive change. American journal of epidemiology. 2005;162(3): 267-78. Meloun M, Militký J. 4 - Statistical analysis of multivariate data. In: Meloun M, Militký J, editors. Statistical Data Analysis: Woodhead Publishing India; 2011. p. 151-403. Holm S. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics. 1979;6(2):65-70. Wang R, Fratiglioni L, Kalpouzos G, Lövdén M, Laukka EJ, Bronge L, et al. Mixed brain lesions mediate the association between cardiovascular risk burden and cognitive decline in old age: A population-based study. Alzheimer's & Dementia. 2017;13(3):247-56. Binnewies J, Nawijn L, Brandmaier AM, Baaré WFC, Boraxbekk C-J, Demnitz N, et al. Lifestyle-related risk factors and their cumulative associations with hippocampal and total grey matter volume across the adult lifespan: A pooled analysis in the European Lifebrain consortium. Brain Research Bulletin. 2023;200:110692. Santos MAO, Bezerra LS, Carvalho A, Brainer-Lima AM. Global hippocampal atrophy in major depressive disorder: a meta-analysis of magnetic resonance imaging studies. Trends Psychiatry Psychother. 2018;40(4):369-78. Cumplido-Mayoral I, Brugulat-Serrat A, Sánchez-Benavides G, González-Escalante A, Anastasi F, Milà-Alomà M, et al. The mediating role of neuroimaging-derived biological brain age in the association between risk factors for dementia and cognitive decline in middle-aged and older individuals without cognitive impairment: a cohort study. The Lancet Healthy Longevity. 2024;5(4):e276-e86. Wirth M, Haase CM, Villeneuve S, Vogel J, Jagust WJ. Neuroprotective pathways: lifestyle activity, brain pathology, and cognition in cognitively normal older adults. Neurobiology of Aging. 2014;35(8):1873-82. Adams S, Conner S, Himali JJ, Beiser A, Vasan RS, Seshadri S, et al. Vascular risk factor burden and new-onset depression in the community. Prev Med. 2018;111:348-50. Shen J, Tozer DJ, Markus HS, Tay J. Network Efficiency Mediates the Relationship Between Vascular Burden and Cognitive Impairment: A Diffusion Tensor Imaging Study in UK Biobank. Stroke. 2020;51(6):1682-9. Marebwa BK, Adams RJ, Magwood GS, Basilakos A, Mueller M, Rorden C, et al. Cardiovascular Risk Factors and Brain Health: Impact on Long-Range Cortical Connections and Cognitive Performance. J Am Heart Assoc. 2018;7(23):e010054. Blöchl M, Schaare HL, Kumral D, Gaebler M, Nestler S, Villringer A. Vascular risk factors, white matter microstructure, and depressive symptoms: a longitudinal analysis in the UK Biobank. Psychol Med. 2024;54(1):125-35. Hoagey DA, Lazarus LTT, Rodrigue KM, Kennedy KM. The effect of vascular health factors on white matter microstructure mediates age-related differences in executive function performance. Cortex. 2021;141:403-20. Honigsbaum M. Revisiting the 1957 and 1968 influenza pandemics. The Lancet. 2020;395(10240):1824-6. Viboud C, Simonsen L, Fuentes R, Flores J, Miller MA, Chowell G. Global Mortality Impact of the 1957-1959 Influenza Pandemic. J Infect Dis. 2016;213(5):738-45. Jordan WSJ, Denny FWJ, Badger GF, Curtiss C, Dingle JH, Oseasohn R, et al. A study of illness in a group of Cleveland families: XVII. The occurrence of asian influenza 123. American Journal of Epidemiology. 1958;68(2):190-212. Almond D. Is the 1918 Influenza Pandemic Over? Long-term Effects of In Utero Influenza Exposure in the Post-1940 U.S. Population. Journal of Political Economy. 2006;114(4):672-712. Marinescu IE, Lawlor PN, Kording KP. Quasi-experimental causality in neuroscience and behavioural research. Nature Human Behaviour. 2018;2(12):891-8. Seblova D, Berggren R, Lövdén M. Education and age-related decline in cognitive performance: Systematic review and meta-analysis of longitudinal cohort studies. Ageing Research Reviews. 2020;58:101005. Chan MY, Han L, Carreno CA, Zhang Z, Rodriguez RM, LaRose M, et al. Long-term prognosis and educational determinants of brain network decline in older adult individuals. Nature Aging. 2021;1(11):1053-67. Lövdén M, Fratiglioni L, Glymour MM, Lindenberger U, Tucker-Drob EM. Education and Cognitive Functioning Across the Life Span. Psychol Sci Public Interest. 2020;21(1):6-41. Zahodne LB, Glymour MM, Sparks C, Bontempo D, Dixon RA, MacDonald SW, et al. Education does not slow cognitive decline with aging: 12-year evidence from the victoria longitudinal study. J Int Neuropsychol Soc. 2011;17(6):1039-46. Sepulcre J. College education as a modulator of the aging brain. Nature Aging. 2021;1(11):980-1. Nyberg L, Magnussen F, Lundquist A, Baaré W, Bartrés-Faz D, Bertram L, et al. Educational attainment does not influence brain aging. Proceedings of the National Academy of Sciences. 2021;118(18):e2101644118. Smith ET, Hennessee JP, Wig GS, Frank S, Gonzalez H, Bacci J, et al. Longitudinal changes in gray matter correspond to changes in cognition across the lifespan: implications for theories of cognition. Neurobiology of Aging. 2023;129:1-14. Ferreira D, Molina Y, Machado A, Westman E, Wahlund LO, Nieto A, et al. Cognitive decline is mediated by gray matter changes during middle age. Neurobiol Aging. 2014;35(5):1086-94. Ballester PL, Suh JS, Ho NCW, Liang L, Hassel S, Strother SC, et al. Gray matter volume drives the brain age gap in schizophrenia: a SHAP study. Schizophrenia. 2023;9(1):3. Brayne C, Ince PG, Keage HA, McKeith IG, Matthews FE, Polvikoski T, et al. Education, the brain and dementia: neuroprotection or compensation? Brain. 2010;133(Pt 8):2210-6. Nguyen TT, Tchetgen Tchetgen EJ, Kawachi I, Gilman SE, Walter S, Liu SY, et al. Instrumental variable approaches to identifying the causal effect of educational attainment on dementia risk. Ann Epidemiol. 2016;26(1):71-6.e1-3. Davies NM, Dickson M, Davey Smith G, Windmeijer F, van den Berg GJ. The causal effects of education on adult health, mortality and income: evidence from Mendelian randomization and the raising of the school leaving age. Int J Epidemiol. 2023;52(6):1878-86. Shin C, Park MH, Lee S-H, Ko Y-H, Kim Y-K, Han K-M, et al. Usefulness of the 15-item geriatric depression scale (GDS-15) for classifying minor and major depressive disorders among community-dwelling elders. Journal of Affective Disorders. 2019;259:370-5. Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM. FSL. NeuroImage. 2012;62(2):782-90. Smith S, Zhang Y, Jenkinson M, Chen J, Matthews P, Federico A, et al. Accurate, robust and automated longitudinal and cross-sectional brain change analysis. NeuroImage. 2002;17(1):479-89. Patenaude B, Smith SM, Kennedy DN, Jenkinson M. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage. 2011;56(3):907-22. Zhang Y, Brady M, Smith S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. Medical Imaging, IEEE Transactions on. 2001;20(1):45-57. Griffanti L, Zamboni G, Khan A, Li L, Bonifacio G, Sundaresan V, et al. BIANCA (Brain Intensity AbNormality Classification Algorithm): A new tool for automated segmentation of white matter hyperintensities. Neuroimage. 2016;141:191-205. Andersson JL, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage. 2003;20(2):870-88. Andersson JL, Graham MS, Zsoldos E, Sotiropoulos SN. Incorporating outlier detection and replacement into a non-parametric framework for movement and distortion correction of diffusion MR images. Neuroimage. 2016;141:556-72. Schilling KG, Blaber J, Hansen C, Cai L, Rogers B, Anderson AW, et al. Distortion correction of diffusion weighted MRI without reverse phase-encoding scans or field-maps. PLOS ONE. 2020;15(7):e0236418. Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, et al. Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data. NeuroImage. 2006;31(4):1487-505. Trzepacz PT, Hochstetler H, Wang S, Walker B, Saykin AJ, Initiative AsDN. Relationship between the Montreal Cognitive Assessment and Mini-mental State Examination for assessment of mild cognitive impairment in older adults. BMC geriatrics. 2015;15:1-9. Lo JW, Crawford JD, Desmond DW, Godefroy O, Jokinen H, Mahinrad S, et al. Profile of and risk factors for poststroke cognitive impairment in diverse ethnoregional groups. Neurology. 2019;93(24):e2257-e71. Winkler AM, Renaud O, Smith SM, Nichols TE. Permutation inference for canonical correlation analysis. NeuroImage. 2020;220:117065. Greene WH. Economic Analysis. 7th Edition ed. Boston: Pearson; 2011. Vanderweele TJ. Explanation in causal inference: Methods for mediation and interaction.: Oxford University Press.; 2015. Table Table 1 is available in the Supplementary Files section Additional Declarations There is NO Competing Interest. Supplementary Files TABLESRiskprotectivefactorshealthycognitiveageing6cohorts20250723.docx Table 1 SupplementarytablesfigureRiskprotectivefactorshealthycognitiveageing6cohorts20250723.docx Supplementary file Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7195000","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":493140475,"identity":"8f699f78-ef08-496f-a246-980bfac4c951","order_by":0,"name":"Heidi Johansen-Berg","email":"data:image/png;base64,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","orcid":"","institution":"University of Oxford","correspondingAuthor":true,"prefix":"","firstName":"Heidi","middleName":"","lastName":"Johansen-Berg","suffix":""},{"id":493140476,"identity":"25faee3b-758c-4f37-bfc8-da33ef62b0d6","order_by":1,"name":"Bonnie Yin Ka Lam","email":"","orcid":"https://orcid.org/0000-0002-1656-2433","institution":"The Chinese University of Hong Kong","correspondingAuthor":false,"prefix":"","firstName":"Bonnie","middleName":"Yin Ka","lastName":"Lam","suffix":""},{"id":493140477,"identity":"11c2258b-3975-4049-9ced-3d96c016a3da","order_by":2,"name":"Vincent Mok","email":"","orcid":"https://orcid.org/0000-0002-8102-8835","institution":"The Chinese University of Hong Kong","correspondingAuthor":false,"prefix":"","firstName":"Vincent","middleName":"","lastName":"Mok","suffix":""},{"id":493140478,"identity":"d6089a8d-26f1-4b89-b1b6-339cf7ff8684","order_by":3,"name":"Henry Brodaty","email":"","orcid":"https://orcid.org/0000-0001-9487-6617","institution":"Centre for Healthy Brain Ageing","correspondingAuthor":false,"prefix":"","firstName":"Henry","middleName":"","lastName":"Brodaty","suffix":""},{"id":493140479,"identity":"a5aee9da-d029-4f7f-9414-26b92448fb18","order_by":4,"name":"Yuan Cai","email":"","orcid":"","institution":"Gerald Choa Neuroscience Institute, Lau Tat-cheun Research Center of Brain Degenerative Diseases in Chinese, Li Ka Shing Institute of Health Sciences, Division of Neurology, Department of Medicine a","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Cai","suffix":""},{"id":493140480,"identity":"2c324bfd-5bb1-4c6d-8501-635623c9c79a","order_by":5,"name":"Michael Wei Liang Chee","email":"","orcid":"https://orcid.org/0000-0002-6087-0548","institution":"National University of Singapore","correspondingAuthor":false,"prefix":"","firstName":"Michael","middleName":"Wei Liang","lastName":"Chee","suffix":""},{"id":493140481,"identity":"e02c8edf-87e7-47b7-a9a4-6ead7e2a101d","order_by":6,"name":"Ilja Demuth","email":"","orcid":"https://orcid.org/0000-0002-4340-2523","institution":"Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin","correspondingAuthor":false,"prefix":"","firstName":"Ilja","middleName":"","lastName":"Demuth","suffix":""},{"id":493140482,"identity":"beeffd80-4684-41b3-af48-c2c7ce99b8f3","order_by":7,"name":"Johanna Drewelies","email":"","orcid":"https://orcid.org/0000-0002-3774-2169","institution":"Max Planck Institute for Human Development","correspondingAuthor":false,"prefix":"","firstName":"Johanna","middleName":"","lastName":"Drewelies","suffix":""},{"id":493140483,"identity":"34657332-a3d2-4724-9755-8a8a08d7ea23","order_by":8,"name":"Sandra Duezel ","email":"","orcid":"","institution":"Center for Lifespan Psychology, Max Planck Institute for Human Development, Berlin, Germany; Friede Springer Cardiovascular Prevention Center at Charité, Berlin, Germany","correspondingAuthor":false,"prefix":"","firstName":"Sandra","middleName":"","lastName":"Duezel ","suffix":""},{"id":493140484,"identity":"18199899-2c2b-4e5b-9472-e16229e71f74","order_by":9,"name":"Klaus Ebmeier","email":"","orcid":"https://orcid.org/0000-0002-5190-7038","institution":"University of Oxford","correspondingAuthor":false,"prefix":"","firstName":"Klaus","middleName":"","lastName":"Ebmeier","suffix":""},{"id":493140485,"identity":"f9cc78f7-4378-4c44-83bb-78e36fae55ba","order_by":10,"name":"Denis Gerstorf","email":"","orcid":"","institution":"Humboldt University Berlin","correspondingAuthor":false,"prefix":"","firstName":"Denis","middleName":"","lastName":"Gerstorf","suffix":""},{"id":493140486,"identity":"75aea543-2f1e-4bc6-ae7b-3c7e111f88bc","order_by":11,"name":"Ludovica Griffanti ","email":"","orcid":"","institution":"Oxford Centre for Integrative Neuroimaging, FMRIB Center, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK；Oxford Centre for Integrative Neuroimaging (Centre fo","correspondingAuthor":false,"prefix":"","firstName":"Ludovica","middleName":"","lastName":"Griffanti ","suffix":""},{"id":493140487,"identity":"176ca6f5-3205-456c-b022-418023858f71","order_by":12,"name":"Vincent Ming Ho Hui ","email":"","orcid":"","institution":"Gerald Choa Neuroscience Institute, Lau Tat-cheun Research Center of Brain Degenerative Diseases in Chinese, Li Ka Shing Institute of Health Sciences, Division of Neurology, Department of Medicine a","correspondingAuthor":false,"prefix":"","firstName":"Vincent","middleName":"Ming Ho","lastName":"Hui ","suffix":""},{"id":493140488,"identity":"f505d72f-7d56-4d40-8f0b-42b1b26b283a","order_by":13,"name":"Jiyang Jiang","email":"","orcid":"https://orcid.org/0000-0002-2147-6302","institution":"Centre for Healthy Brain Ageing","correspondingAuthor":false,"prefix":"","firstName":"Jiyang","middleName":"","lastName":"Jiang","suffix":""},{"id":493140489,"identity":"7f57f6f9-39c3-4767-a4c6-989ebe6e11b7","order_by":14,"name":"Nicole Kochan","email":"","orcid":"","institution":"Centre for Healthy Brain Ageing, University of New South Wales","correspondingAuthor":false,"prefix":"","firstName":"Nicole","middleName":"","lastName":"Kochan","suffix":""},{"id":493140490,"identity":"1fac90c9-e477-43fd-887e-7d2fd3f08b77","order_by":15,"name":"Simone Kühn","email":"","orcid":"https://orcid.org/0000-0001-6823-7969","institution":"Max Planck Institute for Human Development","correspondingAuthor":false,"prefix":"","firstName":"Simone","middleName":"","lastName":"Kühn","suffix":""},{"id":493140491,"identity":"6311644a-f97e-41c3-8feb-62d4ea19e78b","order_by":16,"name":"Ruth Leong","email":"","orcid":"https://orcid.org/0000-0001-5567-2152","institution":"NUS Yong Loo Lin School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Ruth","middleName":"","lastName":"Leong","suffix":""},{"id":493140492,"identity":"24d3fe11-3a0e-4cdb-b12b-78c0dad69cc6","order_by":17,"name":"Ulman Lindenberger","email":"","orcid":"","institution":"Max Planck Institute for Human Development","correspondingAuthor":false,"prefix":"","firstName":"Ulman","middleName":"","lastName":"Lindenberger","suffix":""},{"id":493140493,"identity":"feba29d6-ada0-47a5-a212-ea1826d18e5e","order_by":18,"name":"Eric Kwun Kei Ng ","email":"","orcid":"","institution":"Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore, Singapore.","correspondingAuthor":false,"prefix":"","firstName":"Eric","middleName":"Kwun Kei","lastName":"Ng ","suffix":""},{"id":493140494,"identity":"33a43e05-e719-454c-85a9-b12da556e576","order_by":19,"name":"Thomas Nichols","email":"","orcid":"https://orcid.org/0000-0002-4516-5103","institution":"Oxford University","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Nichols","suffix":""},{"id":493140495,"identity":"cffb0362-6bef-45ec-adad-1af10b55c1ba","order_by":20,"name":"Lars Nyberg","email":"","orcid":"https://orcid.org/0000-0002-3367-1746","institution":"Umeå Center for Functional Brain Imaging (UFBI), Umeå University, S-90187 Umeå","correspondingAuthor":false,"prefix":"","firstName":"Lars","middleName":"","lastName":"Nyberg","suffix":""},{"id":493140496,"identity":"8103f5a2-5732-4004-817b-09ff177b9ab7","order_by":21,"name":"Perminder Sachdev","email":"","orcid":"https://orcid.org/0000-0002-9595-3220","institution":"University of New South Wales","correspondingAuthor":false,"prefix":"","firstName":"Perminder","middleName":"","lastName":"Sachdev","suffix":""},{"id":493140497,"identity":"35e989c9-c6c6-4a54-9d38-717a29b6ac0f","order_by":22,"name":"Sana Suri","email":"","orcid":"","institution":"University of Oxford","correspondingAuthor":false,"prefix":"","firstName":"Sana","middleName":"","lastName":"Suri","suffix":""},{"id":493140498,"identity":"031a1f5c-f972-4e4f-ac12-acc35b3e31c9","order_by":23,"name":"Jize Wei ","email":"","orcid":"","institution":"Gerald Choa Neuroscience Institute, Lau Tat-cheun Research Center of Brain Degenerative Diseases in Chinese, Li Ka Shing Institute of Health Sciences, Division of Neurology, Department of Medicine a","correspondingAuthor":false,"prefix":"","firstName":"Jize","middleName":"","lastName":"Wei ","suffix":""},{"id":493140499,"identity":"aac756fb-5dc0-42e9-88c6-7a4483d2efc0","order_by":24,"name":"Wei Wen","email":"","orcid":"https://orcid.org/0000-0003-2753-3870","institution":"University of New South Wales","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Wen","suffix":""},{"id":493140500,"identity":"aa42a3f7-7aef-407b-bfac-da9dd61f122f","order_by":25,"name":"Huijing Zheng ","email":"","orcid":"","institution":"Gerald Choa Neuroscience Institute, Lau Tat-cheun Research Center of Brain Degenerative Diseases in Chinese, Li Ka Shing Institute of Health Sciences, Division of Neurology, Department of Medicine a","correspondingAuthor":false,"prefix":"","firstName":"Huijing","middleName":"","lastName":"Zheng ","suffix":""},{"id":493140501,"identity":"62882e2b-57c5-4070-a40d-e9ae19e68a54","order_by":26,"name":"Juan Helen Zhou","email":"","orcid":"https://orcid.org/0000-0002-0180-8648","institution":"National University of Singapore","correspondingAuthor":false,"prefix":"","firstName":"Juan","middleName":"Helen","lastName":"Zhou","suffix":""},{"id":493140502,"identity":"6b072893-6dcd-4257-a9b2-32c77aef66eb","order_by":27,"name":"Ho Ko","email":"","orcid":"https://orcid.org/0000-0002-0254-3274","institution":"The Chinese University of Hong Kong","correspondingAuthor":false,"prefix":"","firstName":"Ho","middleName":"","lastName":"Ko","suffix":""},{"id":493140503,"identity":"b8c57017-13c8-408b-a66e-8f190659e32c","order_by":28,"name":"Piergiorgio Salvan","email":"","orcid":"","institution":"University of Oxford","correspondingAuthor":false,"prefix":"","firstName":"Piergiorgio","middleName":"","lastName":"Salvan","suffix":""}],"badges":[],"createdAt":"2025-07-23 10:05:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7195000/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7195000/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88038150,"identity":"923ba5c6-a1ba-469a-923d-d03add72d7c7","added_by":"auto","created_at":"2025-07-31 16:37:22","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":303179,"visible":true,"origin":"","legend":"\u003cp\u003eLeft: Box plots of variability of clinical variables (including age, education, general cognition at baseline and follow-up).\u003c/p\u003e\n\u003cp\u003eRight: Scatterplots showed the association between age and cognition at baseline and differences in cognition (time-point 2 minus time-point 1).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/2fa06bc45aca990f6958f89b.png"},{"id":88037341,"identity":"f13b518e-5799-4cdb-a932-f34512c57b93","added_by":"auto","created_at":"2025-07-31 16:29:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":408130,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSignificant modes of covariation link risk and protective factors with cognition and depressive symptoms across multiple cohorts\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe top panel shows the scatterplots of the significant modes of covariation linking the risk and protective factors with cognition and depressive symptoms in mode 1 (2a); mode 2 (2b) and mode 3 (2c) in all subjects. The bottom panel shows the scatterplots grouped by the different cohorts in each mode respectively.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/4d6ee437c84635c5bf66ea62.png"},{"id":88037340,"identity":"5d59a978-0ac9-4353-9f8c-29a5c2b8cc37","added_by":"auto","created_at":"2025-07-31 16:29:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":454362,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVariable contributions (cross-loadings) of significant modes of covariation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe first row shows the crossing loadings of risk and protective factors in all 3 modes combined (3a), for mode 1 only (the strongest mode) (3b), and for mode 1 only but cross-loadings presented separately for each cohort (3c). The second row shows the cross loadings of cognitive and depressive symptoms in all 3 modes combined (3a), for mode 1 only (the strongest mode) (3b), and for mode 1 only but cross-loadings presented separately for each cohort (3c). Cross-loadings outside of the black circle indicates a positive loading while those inside of the black circle refers a negative loading.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/fa999c030008c54972e6511a.png"},{"id":88037338,"identity":"aa3251e3-8398-4165-8b13-f947c672e475","added_by":"auto","created_at":"2025-07-31 16:29:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":273104,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMeta-analysis of neuroimaging mediation effect (indirect effect) between greater education and younger age with better baseline cognition in CCA mode 1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eForest plots of the indirect effect sizes of the normalised grey matter volume, normalised hippocampal volume; normalised white matter hyperintensity volume, and mean diffusivity were illustrated. Meta-analyses showed that normalised grey matter volume was a significant mediator in CCA Mode 1, after Bonferroni correction (p = 0.05/4 = 0.0125). The size of the black box is determined by the inverse variance of the sample size and variation of the indirect effect size.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/f16177d72a42e68ac18b3ce4.png"},{"id":88037342,"identity":"ab5c4772-8e8c-4da3-a62f-1bfe025a50ef","added_by":"auto","created_at":"2025-07-31 16:29:23","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":195196,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLeveraging exposure to 1957 influenza pandemic to establish the long-term causal effect of education on cognition.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTop panel: Number of female and male students with education (left); the years of education of female and male (middle); cognitive score of those students after six decades (right)\u003c/p\u003e\n\u003cp\u003eBottom panel: Education (z-score) of female and male students (left); cognition (z-score) of those students after six decades (middle); IV estimates of effect of education on cognition (right).\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/581041fec0b947a66541e200.png"},{"id":91686541,"identity":"c3144b6e-495e-4fd9-a8b4-687ed11c0327","added_by":"auto","created_at":"2025-09-19 07:49:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2323249,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/6f9fb195-8277-422c-be65-63c57dcc9983.pdf"},{"id":88037336,"identity":"c8544af1-b3cb-408f-9039-48ea92984965","added_by":"auto","created_at":"2025-07-31 16:29:22","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":32464,"visible":true,"origin":"","legend":"Table 1","description":"","filename":"TABLESRiskprotectivefactorshealthycognitiveageing6cohorts20250723.docx","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/85aa297df0feadf972c91672.docx"},{"id":88037339,"identity":"b004d5ae-61f5-4d81-b5bf-ff8406c43294","added_by":"auto","created_at":"2025-07-31 16:29:22","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1559796,"visible":true,"origin":"","legend":"Supplementary file","description":"","filename":"SupplementarytablesfigureRiskprotectivefactorshealthycognitiveageing6cohorts20250723.docx","url":"https://assets-eu.researchsquare.com/files/rs-7195000/v1/937ae5cd19c738381dbf6f90.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Risk and protective factors of healthy cognitive ageing across diverse global cohorts and causal effect of education","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe proportion of the global population aged 60 and over is projected to reach 22% by 2050 (1). During this period, while global life expectancy is expected to increase by five years, \u003cem\u003ehealthy\u003c/em\u003e life expectancy is projected to rise by less than three years (2). According to the Organisation for Economic Co-operation and Development (OECD), this growing gap may drive healthcare costs to as much as 12% of the gross domestic product\u0026nbsp;(3).\u003c/p\u003e\n\u003cp\u003eFaced with increasing population ageing, there is a growing need to maintain brain health and cognition globally. Brain health can be impacted by multiple factors, including age, sex, education, cardiovascular and metabolic risk factors (e.g. hypertension, diabetes), lifestyle (e.g. exercise, smoking), diet, socioeconomic factors, air pollution, genetics, and culture (4-8). To develop better and cost-effective health policies fostering healthy cognitive brain health at the global level, it is of paramount importance to identify what are accessible and modifiable factors with potential causal effects beyond cultural, geographic, and background differences.\u003c/p\u003e\n\u003cp\u003eWhile there is existing evidence of risk and protective factors for brain health, most previous neuroimaging studies are dominated by homogenous, predominantly White participants. This is because they are most often from high-income, western countries with predominantly European ancestry (9), such as the UK Biobank (10), Lifebrain (11), The Alzheimer's Disease Neuroimaging Initiative (ADNI) (12), The Lifespan Human Connectome Project (HCP) in Aging (13), and The Open Access Series of Imaging Studies (OASIS) (14). With very few exceptions (e.g. the ENIGMA consortium (15), most studies do not capture the full ethnic, geographical and cultural diversity of the global population (16). Because population characteristics as well as healthcare systems, education, lifestyle, diet, and management of vascular risk factors vary across different regions around the world, the limited demographic and cultural variability within predominantly European ancestry datasets limits the degree to which research findings can be generalised to a much more diverse global population (16-18). To better understand human brain health and allow greater generalisability of study findings, research programmes must include participants from diverse global backgrounds. In this work, we aimed to create a more diverse representation of study participants by combining existing cohorts from Europe, Asia and Australia.\u003c/p\u003e\n\u003cp\u003eParticular challenges arise as large-scale, open-access neuroimaging datasets become a key global research resource that increasingly dominates the literature. While these offer powerful opportunities to advance our understanding, the biases present in those datasets also bring important limitations (16, 17). When associations found in one cohort are used to derive models that are tested on other cohorts, predictive power will be greatest when participants have a similar geographical, ethnic and cultural background to the original cohort. Consequently, the dominating models may be disproportionately influenced by particular cohort characteristics, thereby limiting generalisability of findings. In addition to the analytic approach, a few other examples related to research methods, including subject recruitment, and the use of research instruments can present inequities and have marked downstream effects on generalisability of the findings. For example, research tests are mainly available in English which creates a critical barrier in non-English speaking regions; and individuals with religious hair coverings are often excluded from MRI scanning (16, 17). Therefore, an obvious gap in the current research is the lack of cohort diversity that addresses demographic and cultural differences.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn this work, we first aimed to identify how several risk and protective factors relate to longitudinal changes in cognition and depressive symptoms across diverse global cohorts. To do this, we used six unique cohorts spanning six regions (United Kingdom, Germany, Sweden, Hong Kong Special Administrative Region, Singapore, and Australia) and three continents (Europe, Asia, and Australia) with diverse cultures, ethnicities, socioeconomic factors, and healthcare systems, in addition to within-cohort differences. We aimed to identify covariation modes common across all six global cohorts, relating individual differences in demographic, risk and protective factors, and lifestyle factors, with individual differences in cognition and depressive symptoms, independent of cohort-specific differences. We hypothesise that different risk and protective factors are associated with distinct cognitive domains or depressive symptoms, and there may be at least one set of covariation modes that is common across all six cohorts. While it is equally interesting to understand the differences between cohorts, given the variations between cohorts, we consider it to be more meaningful as a first step to understand the similarities as well as marginal average effects across cohorts to derive healthcare policies that can be generalised across the globe. We then aimed to identify whether a precise set of neuroimaging biomarkers assessing distinct features of brain health mediate the covariation of risk and protective factors with cognition and depressive symptoms across the six global cohorts. Last, given the need for causal insight to draw better policies, we leveraged a natural experiment and a quasi-experimental method to examine the causal effect of education on long-term cognitive health.\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eVariabilities in ageing cohorts from Europe, Asia, and Australia\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from individuals who were 60 years old or above were retrospectively collected from six cohorts with global geographical variation from Europe, Asia, and Australia. These included:1) a British cohort (Whitehall-II Imaging Sub-study) (19); 2) a German cohort (the Berlin Aging Study II; BASE-II) (20), 3) a Swedish cohort (Betula project) (21), which are all part of the Lifebrain consortium(http://www.lifebrain.uio.no/) (11); 4) a Hong Kong Chinese cohort (The Chinese University of Hong Kong – Risk Index for Subclinical brain lesions in Hong Kong; CU-RISK) (22), 5) a Singaporean Chinese cohort (Singapore Longitudinal Aging Brain Study; SLABS) (23); and 6) an Australian cohort (Sydney Memory and Aging Study; MAS) cohort (24). Participants within each cohort were primarily of the same national origin and predominantly from the same ethnic group (Supplementary Table 1). Based on inclusion and exclusion criteria (e.g. multimodal MRI at baseline, at least two timepoints of longitudinal cognitive assessment; see Methods for detailed information), a total of 1,636 subjects from the 6 global cohorts were included in the study.\u003c/p\u003e\n\u003cp\u003eRisk and protective factors (age at baseline, sex, years of education, body mass index, frequency of exercise, smoking, presence of diabetes mellitus, hypertension, measure of systolic blood pressure and fasting blood glucose), cognition \u0026amp; depressive symptoms for baseline and follow-up, and neuroimaging measures were available across all 6 cohorts. The mean age across all cohorts was 71.3 ± 6.2 years old, the mean education years was 12.0 ± 4.9, and mean follow-up period was 3.4 ± 1.2 years. Comparison of cohort characteristics were shown using the analysis of covariances (ANCOVAs), controlling for time intervals between assessments (\u003cstrong\u003eTable 1\u003c/strong\u003e). For example, there were significant differences in age of the cohorts, with the Australian cohort being the oldest (78.82 ± 4.38 years) and the British cohort being the youngest (68.2 ± 5.22 years); and significant differences in duration of education, with the Hong Kong Chinese cohort having the shortest education duration (8.08 ± 4.91 years) and the British the longest (14.87 ± 3.42 years) (see \u003cstrong\u003eTable 1\u003c/strong\u003e). While cognitive scores at baseline and follow-up were similar in most cohorts, there was a significant decline in general cognition score over time in the Australian cohort (baseline cognition: 28.26 ± 1.53; follow-up cognition: 27.70 ± 2.37; p\u0026lt;0.001, Cohen’s D = 0.26) and in the Hong Kong Chinese cohort (baseline cognition: 27.78 ± 2.61; follow-up cognition: 27.38 ± 3.12; p\u0026lt;0.001, Cohen’s D = 0.17) (see Supplementary Table 2). Overall, these cohort differences are expected because our study conducts a retrospective conjoint analysis of data that had been collected for other reasons using study-specific inclusion and exclusion criteria, hence resulting in variation in demographic and clinical characteristics (\u003cstrong\u003eFigure 1\u003c/strong\u003e). Therefore, subsequent analyses focused on identifying \u003cem\u003ecommon\u003c/em\u003e relationships across the six global cohorts whilst explicitly considering the hierarchical structure of the data.\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Supplementary Table 1 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Table 1 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Supplementary Table 2 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt;Insert Figure 1 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCommon covariation modes link risk and protective factors with cognition and depressive symptoms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHere we aimed to investigate whether sets of risk and protective factors covaried with cognition and depressive symptoms in a common, shared fashion across all six cohorts, above and beyond between-cohort differences. To test this hypothesis a grand canonical correlation analysis (CCA) was performed including all six cohorts. CCA is a symmetric, cross-decomposition method that characterises covariation modes between two sets of variables: broadly, risk factors and cognitive variables (25). It characterises pairs of latent projections (one per set of variables) that are maximally correlated with each other. Because age, years of education, other risk and protective factors, cognitive tests scores and depressive symptoms were collected separately for each cohort, all data collected were harmonised and individual scores within each cohort were converted to z-scores to respect the hierarchical structure of the data (see Methods). Furthermore, to perform valid statistical inferences while respecting dependencies given this hierarchical structure of the data (26), CCA was carried out with block-aware permutation testing only allowing subjects’ permutation within-cohort but not between-cohorts (27). The input to the CCA included two sets of variables: 1) risk and protective factors (ten variables): age at baseline, sex, years of education, fasting blood glucose level, body mass index, systolic blood pressure, hypertension, diabetes mellitus, smoking, and exercise; and 2) longitudinal cognition and depressive symptoms across two time-points (TPs; median follow-up of 3.43 years) (four variables): general cognition at TP1, difference score in cognition between TP2 and TP1, depressive symptoms at TP1 and difference score in depressive symptoms between TP2 and TP1. The raw difference scores were used without baseline adjustment to provide an unbiased estimate between the causal effect of education on the change in cognition\u0026nbsp;(28). The time interval between TP1 and TP2 was included as a confound of no interest in the CCA.\u003c/p\u003e\n\u003cp\u003eThe grand CCA showed three shared, statistically significant modes of covariation (\u003cstrong\u003eFigure 2\u003c/strong\u003e, top panel), linking specific patterns of risk and protective factors with specific patterns of cognition and depression. These inferences were adjusted for the hierarchical structure of the study design - namely sampling across distinct cohorts - and the effects were consistent across all cohorts as shown by their covariation slopes (\u003cstrong\u003eFigure 2\u003c/strong\u003e, bottom panel).\u003c/p\u003e\n\u003cp\u003eThe 1st mode of covariation (ρ = 0.39; FWE-corr p = 0.001; n = 1,636; \u003cstrong\u003eFigure 2a)\u0026nbsp;\u003c/strong\u003elinked younger age and longer education with better baseline cognitive performance (\u003cstrong\u003eFigure 3b).\u0026nbsp;\u003c/strong\u003eTo further understand the strength of associations separately in each cohort,\u0026nbsp;the cross-loadings of each cohort are plotted separately \u003cstrong\u003e(Figure 3c)\u003c/strong\u003e. These results showed that even though there is variability in cohort characteristics, and how the strength of covariation between the sets of risk \u0026amp; protective factors and sets of cognitive measures \u0026amp; depressive symptoms may differ between cohorts, the modes of covariation are significant, and consistent across the six cohorts. This pattern did not change when assessing the conditional \u003cem\u003eunique\u003c/em\u003e effects using a multivariate regression model. A caveat of these results is that the current approach cannot distinguish whether age acts as an effect of the older generation with a more deprived childhood or as a biological ageing process. However, a key implication of these findings is that while age is not modifiable, education can be made a universally accessible modifiable factor.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe 2nd mode of CCA covariation (ρ = 0.13, FWE-corr p = 0.002; \u003cstrong\u003eFigure 2b\u003c/strong\u003e) linked lower body mass index, with more cognitive decline over time (\u003cstrong\u003eSupplementary Figure 1a \u0026amp; 1b\u003c/strong\u003e). The 3rd mode (ρ = 0.12, FWE-corr p = 0.021; \u003cstrong\u003eFigure 2c\u003c/strong\u003e) linked less regular exercise with more depressive symptoms at baseline (\u003cstrong\u003eSupplementary Figure 1c \u0026amp; 1d\u003c/strong\u003e). Due to the relatively lower canonical correlations observed in modes 2 and 3 (ρ = 0.13 \u0026amp; ρ = 0.12, respectively), interpretation would be less meaningful (29). Therefore, the remaining of the manuscript will focus only on mode 1 as it is more effective to target the common and strongest associations among cohorts.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026lt;Insert Figure 2 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt;Insert Figure 3 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Supplementary Table 3 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Supplementary Figure 1 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003eNext, as a control test, we aimed to assess whether at the population-level there was evidence of an average treatment effect of education on baseline cognition alone whilst taking into account the fact that the effect of education may vary by cohort. To do this, first we fitted a multilevel mixed-effect model with random cohort slopes for education, birth cohort fixed-effects, and adjusting for all other covariates. This allows to estimate the effect of education while allowing it to vary between cohorts for the purpose of generalisation beyond the cohorts used here, whilst controlling for unobserved generational differences between birth cohorts, and conditioning on all other covariates. Model comparison showed that a mixed-effect model is superior to a fixed-effect model. Model outputs showed a small but non-zero variation in the effect of education across cohorts as well as a negative slope–intercept correlation. Of interest the model also showed a negative slope–intercept correlation, meaning that the effect of education is stronger at lower levels of cognition. Second, we estimated \u003cem\u003emarginal\u003c/em\u003e (average) effects for education by averaging over all covariates using G-computation. This allows to estimate not conditional but the marginal effect of education beyond inter-individual differences in covariates. This analysis showed a positive and statistically significant marginal effect of education on baseline cognition (average treatment effect (ATE) = 0.06; p \u0026lt; 0.0001; n = 1,636), showing how a one-year increase in education would result in 0.06 standard deviation increase in baseline cognition as assessed via the Mini-Mental State Examination (MMSE) at age 70. However, a key limitation of these estimates is the absence of early-life or adulthood socioeconomic status measures which are likely to confound the relationship between education and cognition (30). Unfortunately, this is due to the multi-cohort nature of the study, e.g. classifications of socioeconomic status cannot easily be harmonised across cohorts. To overcome this problem, we carried out a sensitivity analysis aimed at establishing how strong an unmeasured confounder would have to be to explain away the observed effect. On average, each of our cohort shows a range of 4 years between those subjects with highest and lowest education. We, therefore, tested how strong should an unobserved variable be to wipe out the effect of education on cognition here observed. To do this we computed an E-value: the minimum strength of association (on a risk ratio scale) that unmeasured confounder(s) would need to have to fully explain away the effect observed, conditional on the covariates. This analysis showed that fully explain away the observed effect an unmeasured confounder would need a risk ratio greater to 1.26 with both education and cognition and across all cohorts, and whilst accounting for generational effects. This suggests that the observed effect of education has moderate robustness to confounding. The results did not change when adjusting also for depressive symptoms. Furthermore, when repeating these analyses using CCA cognitive scores (instead of the baseline cognition), the results did not change qualitatively (ATE = 0.05; p = 0.0006; n = 1,636; E-value = 1.24). Together, these findings show a marginal effect of education that is unlikely to be driven by unmeasured confounder(s) and that is generalisable beyond cohort characteristics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBetween-cohorts meta-analysis of multimodal brain mediators\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNext, we sought to establish whether,\u0026nbsp;consistently across all six global cohorts, candidate MRI markers of brain health mediated the link of longer education and younger age with better cognitive performance (CCA mode 1). Based on previous studies, we hypothesised four neuroimaging markers that have consistently shown sensitivity in detecting individual differences in ageing populations: normalised grey matter volume (31, 32); normalised hippocampal volume (32-34); white matter hyperintensity volume (31, 35-37) and white matter mean diffusivity (22, 37-40). However, by necessity this study involved multi-site MRI data which were acquired using different MRI hardware and with different acquisition protocols. MRI variability is still not a fully resolved issue in multi-site cohort studies and, therefore, we decided not to harmonise MRI data between cohorts. Instead, we used gold-standard neuroimaging processing pipelines (see Methods) and performed MRI statistical analyses separately for each cohort, then combined results across cohorts using a meta-analytic approach.\u003c/p\u003e\n\u003cp\u003eThe meta-analysis showed that for CCA mode 1, baseline normalised grey matter volume mediated the association with a small but statistically significant pooled effect across cohorts (Z=3.11; I² = 0%; p = 0.002; \u003cstrong\u003eFigure 4\u003c/strong\u003e). This shows that across cohorts, greater grey matter volume was a significant brain health mediator of the relationship between longer childhood education, younger age at baseline and greater cognition, independent of geographical or cultural differences between cohorts. The normalised hippocampal volume was also a significant mediator in CCA Mode 1 but did not survive\u0026nbsp;Bonferroni correction (\u003cem\u003ep\u003c/em\u003e = 0.05/4 = 0.0125).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026lt; Insert Figure 4 about here\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA natural experiment to estimate the long-term causal effect of education\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eStudying the causal effect of education on long-term cognition may be challenging due to endogeneity problems. Unobserved factors, such as socioeconomic factors, family environment, innate abilities, or secular trends, may influence both the time spent in education and cognitive development, therefore, biasing the estimation of the causal effect of education on cognition. To overcome this problem and to test for a \u003cem\u003ecausal\u003c/em\u003e effect of education, we took advantage of a natural experiment, exposure to the 1957-1958 “Asian influenza” pandemic during school age, as an exogenous shock that resulted in variation in educational experience independent of individual characteristics.\u003c/p\u003e\n\u003cp\u003eThe “Asian influenza” pandemic was first reported in April 1957 in Hong Kong (41). It emerged unexpectedly affecting 10% of the population, with a particularly pronounced impact on school-aged children (42) who experienced the highest influenza attack rates (69%–78%; much higher than adults, 19%–24%) (43) and yet the lowest excess mortality rates (42), hence generating sharp predictions for long-term effects. Crucially, in 1957 Hong Kong, primary and secondary education were neither universal nor guaranteed, with children enrolling in primary and secondary schools respectively at age 6 and 12, and with strong gender differences and societal norms.\u003c/p\u003e\n\u003cp\u003ePrevious research demonstrated how perinatal exposure to the 1918 “Spanish” influenza pandemic in the U.S. led to significant reductions in educational attainment and long-term health sequelae (44). Here, we leveraged the sudden exposure to the “Asian influenza” pandemic during the critical transition between primary-to-secondary school period as an exogenous \u003cem\u003einstrument\u003c/em\u003e that influenced education attainment. We hypothesised that pandemic exposure in children at age 12, which is during the critical primary-to-secondary school transition, led to reduced educational attainment compared to other birth cohorts, which, in turn, resulted in dose-response negative effects on cognitive health six decades later. Because of gender differences in societal norms and labour market opportunities during the study period (Hong Kong, 1957-58), and because of marked gender differences in educational attainment present in the data, we focused only on males. To test our hypothesis, we used an instrumental variable (IV) approach (45), addressing both the endogeneity problem and allowing us to test for the causal effect of education drop on cognitive health at around 70 years old.\u003c/p\u003e\n\u003cp\u003eFirst, as hypothesised, a marked decrease in years of education was present in the birth cohort of males aged 12 in 1957 when compared to nearby birth cohorts and females, who showed no reduction in education years. This was apparent even in the raw data, with this group dropping from an average of 10 years of education seen in adjacent cohorts, to an average of just under 7 years in the cohort aged 12 in 1957 (\u003cstrong\u003eFigure 5\u003c/strong\u003e). \u0026nbsp;This difference showed a significant and large negative effect size (Cohen’s d = 3; p = 0.003, assessed via permutation testing of general linear models whilst adjusting for date of birth). Curtailment of education for military service cannot be the reason for this variation as Hong Kong has never implemented military conscription.\u003c/p\u003e\n\u003cp\u003eSecond, the 1957 influenza pandemic itself is unlikely to directly influence long-term cognitive health outcomes six decades later, except through its impact on educational attainment: (i) Short-term health effects of exposure in childhood (e.g. respiratory illness) were unlikely to have persistent direct effects on cognitive health sex decades later and, even if they did, they would not be specific to the cohort of interest aged 12 in 1957. (ii) Exposure to pandemics is known to reduce educational attainment (44).\u0026nbsp;(iii) Economic and social pressures from the pandemic likely influenced household decisions regarding children's schooling. Thus, both instrumental variable assumptions — the strong first stage and exclusion restriction — are satisfied.\u003cbr\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe IV analysis revealed a positive and significant causal effect of years of education on general cognitive abilities assessed at around 70 years of age, hence six decades later. Specifically, the first-stage IV statistics confirmed the relevance of the instrument (F = 9.44, p = 0.003, adjusted R-squared=0.07). Then, using exposure to the 1957 influenza pandemic in males aged 12 as an instrument for the endogenous treatment, years of education, the IV estimate indicated that each additional year of education caused a 1.15% increase (~0.2 standard deviation change) in cognitive functioning, above and beyond the effect of age. The results were robust to covarying for linear and quadratic trends in birth year, age-adjusted time of assessment, and to rank-inverse transforming the data (IV Estimate: 0.33; 95% CI: 0.09–0.88-unit, p = 0.021; n=110). All second-stage IV robustness estimator tests (LIML, TSLS, Fuller), as well as alternative tests (Anderson-Rubin, and Conditional Likelihood Ratio tests), supported the validity of the findings providing similar results with confidence intervals for the causal effect entirely above zero. Further analyses confirmed no direct effect of the instrument on the outcome when adjusting for treatment (significant) and confounds, hence providing further support for the exclusion restriction. Crucially, the same IV analysis did not show any significant effects of education on cardiovascular (high blood pressure, second-stage IV p = 0.58) or metabolic health (diagnosis of diabetes mellitus, second-stage IV p = 0.28). These findings highlight the long-term causal role of education in supporting healthy cognitive functioning in old age, even after accounting for aging and potential endogeneity issues.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026lt;Insert Figure 5 about here\u0026gt;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eHere we identified how modifiable lifestyle factors relate to healthy cognitive ageing across diverse global cohorts spanning six regions (United Kingdom, Germany, Sweden, Hong Kong Special Administrative Region, Singapore, and Australia) and three continents (Europe, Asia, and Australia), characterised by diverse cultures, ethnicities, socioeconomic factors, and healthcare systems, in addition to between-cohort differences. We first identified globally shared patterns of covariation and highlighted the marginal effect of education on cognitive healthy ageing, measured by cognitive scores and depressive symptoms. By harnessing non-invasive multimodal brain MRI, we then identified brain structure that mediates the risk and protective factors and longitudinal changes in cognition and depressive symptoms across diverse global cohorts. Last, by leveraging a natural experiment, we showed the causal effect of education highlighting the critical importance of education as a modifiable factor to improve brain healthy ageing globally. By incorporating diverse global cohorts that reflect a wide range of ethnic and cultural backgrounds, focusing on accessible and modifiable lifestyle factors, and ensuring that findings are robust and generalisable to broader, mixed populations, this work meaningfully advances diversity in brain health research.\u003c/p\u003e\n\u003cp\u003eOne key motivation for this work was to identify modifiable lifestyle factors that can be targeted in global public health policies to improve healthy brain ageing. To do this, and to explore the complex picture of human behaviour and population diversity across geographical contexts, we leveraged a large pool of risk and protective factors across six diverse global cohorts (\u003cstrong\u003eFigure 1\u003c/strong\u003e), together with longitudinal assessments of cognition and depressive symptoms. We found three, statistically significant CCA modes of covariation, independent of underlying geographic and cultural differences (\u003cstrong\u003eFigure 2\u003c/strong\u003e). The strongest mode of covariation linked younger age at baseline and longer education with better baseline cognitive performance (\u003cstrong\u003eFigure 3\u003c/strong\u003e). Because age is a non-modifiable factor, follow-up analyses focused on education whilst adjusting for birth cohort fixed effects to control for unobserved generational differences. This showed evidence for a statistically significant \u003cem\u003emarginal\u003c/em\u003e treatment effect on the CCA cognitive outcomes, holding other variables constant. This means that, on average across the mixed populations, each additional year of education is associated with an average increase of 1.06% in cognitive outcomes beyond the dominant effects of birth cohorts, sex, smoking status, cardiovascular and metabolic health, or other background characteristics. This is highly relevant for policy making and public health because, unlike conditional regression effects - which apply to individuals with specific characteristics - here marginal effects reflect the average, causal impact of education across the global populations, beyond background characteristics.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOne important note though is that greater years of education was mainly associated with better baseline cognitive scores but not the change in cognition. The association between higher years of education is associated with higher levels of cognition in early adulthood is well established, as least in previous studies from WEIRD (Western, Educated, Industrialised, Rich and Democratic countries). However, whether education is associated with the rate of change in cognition is highly debated and empirical evidence from longitudinal meta-analysis did not reveal a consistent and substantial association (46). Some suggested that education in early adulthood protects the brain from cognitive decline at old age (47) while some longitudinal studies showed that education is not related to individual differences in rates of cognitive change\u0026nbsp;(48)\u0026nbsp;(49)\u0026nbsp;(50)\u0026nbsp;(51). A mega-analysis from longitudinal cohorts across 33 countries (mainly Western countries) showed education was associated with better memory, larger intracranial volume and slightly larger brain region relating to memory but it did not protect against rate of cognitive decline\u0026nbsp;(9). While this current study and the large large-scale longitudinal cohort have overlapping cohorts (the three European cohorts included in this study are part of the Lifebrain consortium\u0026nbsp;(11), this study included two Asian cohorts and one Australian cohort, thus, provided validation across independent cohorts with substantial variability in demographics, cultural contexts, and even healthcare systems. These findings underscore the importance to support the role of education across global populations, and that education was mainly associated with better baseline cognitive scores but not the change in cognition, and most importantly, this association holds beyond western-centric research evidence.\u003c/p\u003e\n\u003cp\u003eA second objective of the study was to leverage multi-site neuroimaging to identify brain metrics that mediate between risk and protective factors and longitudinal changes in cognition and depressive symptoms across diverse global cohorts. Because MRI variability is still not a fully resolved issue in multi-site cohort studies, instead of harmonising MRI data between cohorts, we opted for performing causal mediation analysis separately within-cohort followed by pooling across-cohorts by feeding these estimates of indirect mediation effect into a meta-analysis. Among the neuroimaging variables investigated (normalised grey matter volume, normalised hippocampal volume, white matter hyperintensities volume, and white matter mean diffusivities), there is a statistically significant pooled effect for normalised grey matter volume (and for normalised hippocampal volume, although the latter did not survive Bonferroni correction across MRI candidates tested). These findings show that normalised grey matter volume was a common, significant mediator (\u003cstrong\u003eFigure 4\u003c/strong\u003e) of the relationship between younger age and longer education with healthier cognitive trajectories. Our findings echo previous studies exploring brain mediators between ageing and cognition (34, 52-54), but crucially they also extend beyond the existing literature. First, by combining datasets from three continents, these findings offer improved generalisability of scientific insight across diverse populations. Second, by using difference scores (without adjustment to baseline) to investigate longitudinal changes, this allowed us to provide an unbiased view between the effect of education on change in cognition and depressive symptoms. Third, by employing a multi-site MRI approach and demonstrating statistically significant pooled effects across cohorts, the results presented here are robust also against variability that could otherwise be ascribed to differences in MRI protocols or potential incomplete harmonisation of MRI imaging data. Together, these neuroimaging findings highlight a potential modifiable pathway that can be targeted to promote healthier trajectories of cognitive and mental health across diverse populations.\u003c/p\u003e\n\u003cp\u003eBecause the effect of education can be confounded by the effect of secular improvements in the socioeconomic context across time as well as by endogeneity biases, a third objective of the study was to establish the causal effect of education beyond these issues. To do this we used instrumental variable analysis to take advantage of the quasi-random exposure to the 1957-1958 “Asian influenza” pandemic during school age in Hong Kong as an exogenous shock that resulted in variation in educational experience independent of individual characteristics. Although greater education is often linked with better brain and cognitive health in ageing (55), this association is often confounded by early-life socioeconomic factors, health, and genetics (56, 57). Here, using exposure to the pandemic influenza at age 12 as an instrument, this approach allowed us to overcome these endogeneity problems. Leveraging a model previously designed to study the long-run consequence of exposure to the 1918 influenza pandemic\u0026nbsp;(44), our results showed the protective, causal effect of education on cognitive health six decades later (\u003cstrong\u003eFigure 5\u003c/strong\u003e), even after accounting for smooth linear and quadratic trends in birth cohort and age at assessment (relative to birth year). Supporting our results, Davies et al. used a natural experiment exploiting national changes in school policy to show that longer education causally reduces the risk of cardiovascular and metabolic diseases, and mortality\u0026nbsp;(57). These findings strengthen the case for education—a modifiable lifestyle factor—as a true protective factor to support healthy cognitive aging. These results, therefore, have significant implications for informing equitable healthcare policies.\u003c/p\u003e\n\u003cp\u003eA clear limitation of this work that must be addressed in future studies is that our cohorts were opportunistically selected based on data availability. While they cover diverse populations across three continents, several world geographies and ethnicities are not represented. This is particularly the case for low- and middle-income countries, in which two-thirds of the world's population aged 60 years or above reside (1). Furthermore, among the cohorts that we assessed, participants were primarily of the same national origin and/or predominantly from the same self-reported ethnic group. However, this information was not explicitly collected in all cohorts and/or, if available, it was not standardised. This limitation highlights a broader need for guidelines, practices, and standards for data collection and harmonisation across global cohorts. Common and international standards, relatable to those already in use by other international organisations, e.g. the OECD, would help the neuroimaging community to move forward and directly address questions with real-world public health implications.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eBy emphasizing diversity in study design, research objectives, and analytical approaches, and by focusing on accessible modifiable factors, this work takes an initial step towards addressing the gap left by traditional Western-focused research norms. This work provides important evidence for the causal role of education in lifelong cognitive health. These findings should contribute to evidence-based policy making, with the aim of ultimately improving healthy brain ageing globally while reducing health disparities.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eCohort samples\u003c/p\u003e\n\u003cp\u003eData from individuals who were 60 years old or above were retrospectively collected from six cohorts with geographical variations. The community cohorts included:1) a British cohort (Whitehall-II Imaging Sub-study) (19); 2) a German cohort (the Berlin Aging Study II; BASE-II) (20), 3) a Swedish cohort (Betula project) (21), which are all part of the Lifebrain consortium(http://www.lifebrain.uio.no/(11)). In addition, we included 4) a Hong Kong Chinese cohort (The Chinese University of Hong Kong \u0026ndash; Risk Index for Subclinical brain lesions in Hong Kong; CU-RISK) (22), 5) a Singaporean Chinese cohort (Singapore Longitudinal Aging Brain Study; SLABS) (23); and 6) an Australian cohort (Sydney Memory and Aging Study; MAS) cohort (24). Participants in each cohort were primarily of the same national origin and predominantly from the same ethnic-racial group. For more details on cohort information, please see Supplementary Table 1. The work described has been carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) and has been approved by the Joint Chinese University of Hong Kong-New Territories East Cluster ethics committee (CREC Ref. No. 2022.119).\u003c/p\u003e\n\u003cp\u003eThe inclusion criteria were i) community-dwelling individuals aged over 60 years old; ii) who had demographic information and clinical data on risk and protective factors; iii) at least two-time points of measurement on cognition and depressive symptoms, preferably with 3-4 years of assessment interval; iii) with at least one-time point of MRI with T1-weighted, fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted sequences available. The exclusion criteria were i) subjects with stroke, dementia, and other neurological diseases at baseline; ii) those without longitudinal cognitive and depressive symptom measures; and iii) those without MRI data available.\u003c/p\u003e\n\u003cp\u003eDetails of neuropsychological assessments and depressive symptoms measured for each cohort are presented in Supplementary Table 4. \u0026nbsp; While not all cohorts captured all cognitive domains, there was variation in the precise measures used, and therefore, some harmonisation of measures was required (see data harmonisation section below).\u003c/p\u003e\n\u003cp\u003eRisk and protective factors\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor demographic information, age at baseline, sex, years of education, ethnicity, country of origin, body mass index, and socioeconomic status were included. For vascular risk factors, information on the pulse, systolic blood pressure, diastolic blood pressure, fasting blood glucose, glycated haemoglobin, triglyceride level, high-density-lipoprotein level, low-density-lipoprotein level, total cholesterol level, hypertension, diabetes mellitus, and hyperlipidaemia were collected. Regarding behavioural and lifestyle factors, physical activities, functional dependence, smoking habits, and information on alcohol consumption were collected. Motor function data were captured using the functioning timed walk and 10-meter Walk Test. Information on intelligence quotient was collected.\u003c/p\u003e\n\u003cp\u003eAssessments on cognition and depressive symptoms\u003c/p\u003e\n\u003cp\u003eFor cognitive data, the Mini-Mental State Examination (MMSE) or Montreal Cognitive Assessment (MoCA) was used to measure general cognition. In addition, specific cognitive domains including attention and processing speed, executive function, memory, visuospatial function, and language were considered if available. The German version of the MMSE used in the BASE-II study was indicated with a special note, please refer to Supplementary Table 4. To assess depressive symptoms, data were captured using the Center for Epidemiologic Studies Depression Scale (CES-D) (cutoff at \u0026ge; 16) (36) or the Geriatric Depression Scale (GDS) (cutoff at \u0026ge; 5) (58). In addition to the baseline measures (TP1) in cognition and depressive symptoms, the follow-up measures (TP2) were collected. To assess longitudinal change in cognition and depressive symptoms, the difference scores between TP2 and TP1 were used\u0026nbsp;(28).\u0026nbsp;The dates of assessment of all tests were recorded and the time interval between baseline and longitudinal assessments were calculated.\u003c/p\u003e\n\u003cp\u003eAlthough data on specific cognitive domains were requested, not all cohorts collected data relating to the same domain or there was no common neuropsychological test to represent specific cognitive domains that can be harmonised across all six cohorts. Similarly, for motor function and\u0026nbsp;intelligence quotient, not all cohorts obtained this measure (see Supplementary Table 4). Therefore, only variables with complete data that can be harmonised across cohorts were used in the canonical correlation analysis (see Supplementary Table 5a-5c). Missing data were excluded from subsequent analyses.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMagnetic resonance imaging (MRI)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMRI data were collected and processed for the six cohorts separately. MRIs from Whitehall II were acquired at the University of Oxford using two scanners: a 3T Siemens Magnetom Verio scanner and a 3T Siemens Prisma scanner. MRI from Betula was acquired from Ume\u0026aring; University using a 3T Discovery MR750 scanner (General Electric, Milwaukee, USA). \u0026nbsp;In BASE-II, a 3T Siemens Tim Trio scanner was used (Siemens, Erlangen, Germany). MRI from CU-RISK was performed at the Prince of Wales Hospital Hong Kong with a 3T Philips Achieva TX (Philips Medical Systems, Best, The Netherlands). MRI from SLABS was acquired at the Duke-NUS (which is an independent school of the National University of Singapore) using a 3T Siemens Tim Trio scanner (Siemens, Erlangen, Germany). MRI from MAS was acquired from the University of New South Wales using a Philips 3T Achieva Quasar Dual scanner (Philips Medical Systems, Best, The Netherlands). All cohorts collected T1-weighted and diffusion-weighted images. The FLAIR images at waves 5 \u0026amp; 6 of the Swedish cohort were available but inaccessible at the time of analyses, hence, white matter hyperintensity volume was not assessed in this cohort. Some cohorts additionally collected other MRI modalities that are not considered here (see Supplementary Table 6).\u003c/p\u003e\n\u003cp\u003eMRI analysis\u003c/p\u003e\n\u003cp\u003eIn brief, the T1-weighted, FLAIR, and diffusion-weighted sequences that matched the baseline clinical and cognitive assessments were used for analyses. The MRI data were processed separately for the six cohorts. All images were de-identified before analyses and processed using FSL v6.0 tools (59). T1-weighted scans were pre-processed with bias-field correction, brain extraction using non-linear registration-based masking, tissue-type segmentation, and subcortical segmentation. The total brain, grey matter were estimated using FSL-Sienax (60). Sienax estimated total brain tissue and grey matter volumes from a single image, normalised for skull size. In brief, the brain was extracted, then the brain and skull images were used to estimate the scaling between the subject\u0026apos;s image and standard space. Further, issue segmentation was performed to estimate the brain tissue volume, and multiplies this by the estimated scaling factor, thus, reducing head-size-related variability between subjects. The hippocampal volume was generated using FSL-FIRST and corrected for head size (61).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFLAIR images were brain extracted and processed with bias-field correction using FSL-FAST (62). We used FSL-BIANCA (63). A subset of 20 manually segmented WMH images (from the CU-RISK study) were provided as the training data for all the cohorts. During image visual inspection, for subjects with moderate WMH that affects T1 image tissue segmentation (e.g. WMH misclassified as grey matter), a WMH lesion mask generated from FSL-BIANCA was used to fill the lesion before tissue segmentation was done.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor DWI images, data were pre-processed to correct for susceptibility-induced distortions using topup (64), and for eddy currents as well as subject head movements using eddy (65). For cohorts without the reverse phase-encoding direction, we applied Synb0-DISCO v2.0 (66), a tool which synthesises an \u0026ldquo;undistorted\u0026rdquo; b0 image based on the geometry of the given structural T1-weighted scans. The UK Biobank FA template was used as the group template (10). \u0026nbsp;Mean diffusivity (MD), which is the average MD within the skeleton, was obtained using DTIFIT and TBSS (TBSS v 1.2 (67).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData harmonisation\u003c/p\u003e\n\u003cp\u003eClinical, and data on cognition and depressive symptoms were collected separately from each cohort. All data were harmonised, and individual test scores were converted to z-scores within each cohort for further analysis. In brief, we categorised neuropsychological tests based on the respective domains measured. For example, for general cognition, either the Mini-mental state examination (MMSE) or Montreal cognitive assessment (MoCA) was used. Since only one cohort used the MoCA while the other five cohorts used MMSE, the MoCA score of that cohort was converted to MMSE (68). These raw test scores were converted to z-scores and used in the measure for \u0026ldquo;general cognition\u0026rdquo;. In addition, specific cognitive domains including attention and processing speed, executive function, memory, visuospatial function, and animal fluency were requested if available. However, specific cognitive domains were not included in the CCA models as there was no cognitive domain that was able to be harmonised across all six cohorts. The neuropsychological tests used for general cognition and specific cognitive domains in each cohort were summarised (see Supplementary Table 4).\u0026nbsp;\u0026nbsp;For depressive symptoms, the Center for Epidemiological Studies-Depression Scale and Geriatric Depression Scale were the two main tests used across all 6 cohorts. It is impossible to completely harmonise the two tests, but the test score was converted to a z-score within the cohort for analysis instead. Harmonisation of the different neuropsychological tests was based on common practice and with reference to a previous publication\u0026nbsp;(69).\u0026nbsp;For detailed information on the specific tests and cut-offs used in each measure specific to the cohort, please refer to Supplementary Table 5.\u003c/p\u003e\n\u003cp\u003eThis study involves multi-site MRI data which were acquired using different MRI machines and with different acquisition protocols (see Supplementary Table 6). MRI variability is still not a fully resolved issue in multi-site cohort studies. We did not attempt to harmonise the MRI data collected across cohorts. Instead, we used a standard neuroimaging analysis pipeline and performed MRI analysis for each cohort individually (instead of combining imaging data across cohorts) when assessing statistical associations. \u0026nbsp;The advantage of adopting such a per cohort analysis approach is that it allows us to investigate how neuroimaging measures contribute directly and specifically in each cohort, and to avoid issues arising from potential incomplete harmonisation of MRI imaging data. Causal mediation analysis was then performed separately for each cohort, testing whether MRI candidate markers (M) mediate the effect between risk and protective factors (X) and cognition and depression (Y). Causal mediation outputs per each cohort (coefficient of indirect effect) were fed into a meta-analysis to test our hypothesis of\u0026nbsp;neuroimaging mediation at the between-cohorts level.\u003c/p\u003e\n\u003cp\u003eStatistical analyses\u003c/p\u003e\n\u003cp\u003eCanonical correlation analysis (CCA)\u003c/p\u003e\n\u003cp\u003eCanonical correlation analysis (CCA) is a multivariate approach to identify covariation between two sets of variables. In this study, our first aim is to investigate the underlying link between risk \u0026amp; protective factors and cognition \u0026amp; depressive symptoms using a CCA with permutation inference (25, 27). The input data included two sets of variables \u0026ndash; 1) risk and protective factors (ten variables \u0026ndash; age at baseline, sex, years of education, fasting blood glucose level, body mass index, systolic blood pressure, hypertension, diabetes mellitus, smoking, and exercise); and 2) cognition and depressive symptoms measures (four variables - general cognition at TP1, difference in general cognition between TP2 and TP1, depressive symptoms at TP1, and difference in depressive symptoms between TP2 and TP1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn the CCA model, we investigated the associations between risk \u0026amp; protective factors and cognition \u0026amp; depressive symptoms, without the use of neuroimaging variables. However, only subjects with complete clinical data, longitudinal data on cognition and depressive symptoms, as well as MRI data were included in the CCA analysis. The intention here was to match the group of subjects being used in the CCA and subsequent mediation analyses using MRI data such that there is a more accurate assessment of the mediating effect of neuroimaging markers in the covariation derived from the CCA. All variables were converted to z-score within each cohort and all data from the six cohorts were used in the CCA, which effectively adjusted for cohort differences. The time interval between baseline assessments and follow-up assessments was included as covariates.\u003c/p\u003e\n\u003cp\u003eHere CCA would allow to identify latent modes of covariation which link the two sets of variables across the six cohorts, while quantifying linear combinations of risk and protective factors that maximally correlate with linear combinations of cognition and depression scores. These linear combinations are known as CCA variate pairs. Significance of CCA modes was determined by nonparametric inference testing with 1,000 permutations allowing shuffling of subjects only within - but not between - cohorts to take into account the cluster structure of the cohorts (k = 6). Multiple comparisons across CCA modes were corrected via family-wise error rate correction (FWE-corr). CCA cross-loadings, quantifying the involvement of each original variable in the CCA variate pairs, were then calculated for all CCA modes deemed significant at FWE-corr p \u0026lt; 0.05 (27, 70). \u0026nbsp;Furthermore, to estimate unbiased CCA loadings, we used a multilevel mixed-effect model with random slopes, hence allowing effects to vary between cohorts for the purpose of generalisation. We also used birth cohort fixed effects to control for unobserved generational differences. Age cohort fixed effects controlling for life-cycle differences were not used however as birth year and\u0026nbsp;age at assessment are highly correlated. Last, for the purpose of generalizability again, we estimated CCA cross-loadings as marginal (average) treatment effects by marginalising over the distribution of the covariates across the global cohorts. Marginal effects were calculated as average, partial derivatives of the regression equation via the R port of Stata\u0026apos;s \u0026lsquo;⁠margins⁠\u0026rsquo; command\u0026nbsp;(71).\u0026nbsp;Bonferroni-Holm correction was used to control for familywise error rate, the probability of at least one type I error across predictors\u0026nbsp;(30).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMediation analysis\u003c/p\u003e\n\u003cp\u003eTo assess the mediating effect of neuroimaging markers in the covariation of risk and protective factors with cognition and depressive symptoms, a mediation analysis was performed separately for each cohort. The CCA variate of risk and protective factors was used as the independent variable while the CCA variate of cognition and depressive symptoms was used as the dependent variable. \u0026nbsp;Based on prior evidence, we hypothesise that the brain mediators for changes in cognition and depressive symptoms will be underpinned by distinct brain regions such as the grey matter volume (31, 32), hippocampal volume (32-34), white matter hyperintensity volume (31, 35-37), and white matter mean diffusivity (22, 37-40). As we use the nonparametric percentile bootstrap method in mediation analysis (which uses random resampling with replacement to estimate a population parameter), the outputs are the direct effect, indirect effect (used in the meta-analysis), total effect, and proportion mediated (72). The time interval between baseline assessments and follow-up assessments was included as covariates. The R \u0026apos;mediation\u0026apos; package was used for mediation analysis.\u003c/p\u003e\n\u003cp\u003eMeta-analysis\u003c/p\u003e\n\u003cp\u003eA meta-analysis was conducted to obtain comprehensive summary statistics of the neuroimaging mediation effects across all six global cohorts. Separately for each CCA mode deemed significant, the indirect effect estimates from the mediation analysis linking between the risk and protective factors (X) with cognition and depressive symptoms (Y) via brain MRI markers (M) in CCA mode 1 were used in the meta-causal mediation analysis. Heterogeneity was evaluated using Cochrane\u0026rsquo;s Q test and I\u003csup\u003e2\u003c/sup\u003e statistics. If there was a significant heterogeneity (p\u0026lt;0.05 or I\u003csup\u003e2\u003c/sup\u003e \u0026gt; 50%), a random effect model was used in the meta-analysis. Otherwise, a fixed-effect model was applied. A Bonferroni correction was applied to adjust for multiple testing across neuroimaging markers: p-value = 0.05 / 4 neuroimaging markers = Bonferroni p-value = 0.0125. The meta-analysis was performed using the R package \u0026apos;metafor\u0026apos;.\u003c/p\u003e\n\u003cp\u003eInstrumental Variable (IV) analysis\u003c/p\u003e\n\u003cp\u003eAn IV analysis was conducted to address possible endogeneity problems in the relationship between the endogenous variable (education) and the outcome (cognition) and to assess causality. The IV was applied only to the Hong Kong cohort as this was the only cohort where education was neither universal nor guaranteed in 1957 and with sufficient sample size to conduct the analysis. (In 1957 Singapore education was also neither universal nor guaranteed. However, the Singapore cohort did not have enough subjects in the birth cohort of interest, e.g. subjects aged 12 in 1957 n = 3). Exposure was defined as being 12 year old in 1957 hence - contrary to \u0026lsquo;graduation year\u0026rsquo; - this is exogenous and only determined by birth year. The outcome model for the IV analysis was designed after Almond (2006)\u0026nbsp;(44) with exposure controlled for smooth linear and quadratic trends in birth cohort and age-adjusted time of assessment (relative to birth year). This approach assumes that, in absence of the pandemic, the birth cohort of interest would have followed the same smooth trajectory in outcome, e.g. cognitive scores. This assumption is justified by the known smooth relationship between age and change in cognition. The IV analysis was performed using the R package \u0026apos;ivmodel\u0026apos;.\u003cstrong\u003e\u003cbr\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eDeclaration of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH J-B is or has been an advisory board member or consultant to Biogen, Eisai, Eli Lilly, Medicines Australia, Roche and Skin2Neuron. He is a Medical/Clinical Advisory Board member for Montefiore Homes and Cranbrook Care. MWLC is an advisor to Oura Health and Quantactions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePSS was a paid member of Advisory Panels for Biogen and Roche Australia in 2020 and 2021.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll other co-authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data and scripts that support the findings of this study are available on request from the corresponding authors, H.J-B and P.S. Raw data that belongs to other cohorts may only be available upon request from the cohort Principal Investigator.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBYKL was supported by the Lee Hysan Postdoctoral Fellowship in Clinical Neurosciences. MWLC was supported by the following sources: Yong Loo Lin School of Medicine (National University of Singapore, Singapore), The Lee Foundation (Singapore), National Medical Research Council Singapore (STaR May2019-001), (STaR/0013/2013) (STaR/0004/2008), Biomedical Research Council, Singapore (04/1/36/19/372). The Sydney Memory and Ageing Study was supported by three National Health \u0026amp; Medical Research Council (NHMRC) Program Grants (ID No. ID350833, ID568969, and APP1093083). MRI scans were supported by NHMRC Project Grants (510175 and 1025243) and an ARC Discovery Project Grant (DP0774213) and John Holden Family Foundation. JHZ was supported by the following sources: Singapore National Medical Research Council (NMRC/OFLCG19May-0035, NMRC/CIRG/1485/2018, NMRC/CSA-SI/0007/2016, NMRC/MOH-00707-01, NMRC/CG/435 M009/2017-NUH/NUHS, CIRG21nov-0007and HLCA23Feb-0004), RIE2020 AME Programmatic Fundf rom A*STAR, Singapore (No. A20G8b0102), Ministry of Education (MOE-T2EP40120-0007\u0026amp;. T2EP2-0223-0025, MOE-T2EP20220-0001), and Yong Loo Lin School of Medicine Research Core Funding, National University of Singapore, Singapore. H J-B was supported by a Wellcome Trust PRF (222446/Z/21/Z), LG was supported by an Alzheimer\u0026rsquo;s Association Grant (AARF-21-846366). This work was supported by the NIHR Oxford Health Biomedical Research Centre (NIHR203316\u003cstrong\u003e)\u003c/strong\u003e. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. The Oxford Centre for Integrative Neuroimaging was supported by core funding from the Wellcome Trust (203139/Z/16/Z and 203139/A/16/Z). The Whitehall II Imaging Sub-study was supported by the UK Medical Research Council (MRC) grants \u0026ldquo;Predicting MRI abnormalities with longitudinal data of the Whitehall II Sub-study\u0026rdquo; (G1001354; PI KPE; Clinical Trials. gov Identifier: NCT03335696), the HDH Wills 1965 Charitable Trust (Nr: 1117747, PI: K.P.E), and the European Commission Horizon 2020 grant \u0026ldquo;Lifebrain\u0026rdquo; (732592, Co-PI KPE). The Betula study was supported by a grant to L.N. from the Knut and Alice Wallenberg (KAW) Foundation (Sweden).This article uses data from the Berlin Aging Study II (BASE-II). BASE-II was supported by the German Federal Ministry of Education and Research under grant numbers #01UW0808; #16SV5536K, #16SV5537, #16SV5538, #16SV5837, #01GL1716A, and #01GL1716B. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eOrganisation WH. Ageing and health 2022 [Available from: https://www.who.int/news-room/fact-sheets/detail/ageing-and-health.\u003c/li\u003e\n \u003cli\u003e(IHME) IfHMaE. Global Burden of Disease 2021: Findings from the GBD 2021 Study. Seattle, WA: IHME; 2024.\u003c/li\u003e\n \u003cli\u003eOECD. Fiscal Sustainability of Health Systems: How to Finance More Resilient Health Systems When Money Is Tight? Paris: OECD Publishing; 2024.\u003c/li\u003e\n \u003cli\u003eBaumgart M, Snyder HM, Carrillo MC, Fazio S, Kim H, Johns H. Summary of the evidence on modifiable risk factors for cognitive decline and dementia: a population-based perspective. Alzheimer\u0026apos;s \u0026amp; Dementia. 2015;11(6):718-26.\u003c/li\u003e\n \u003cli\u003eFranz CE, Hatton SN, Elman JA, Warren T, Gillespie NA, Whitsel NA, et al. Lifestyle and the aging brain: interactive effects of modifiable lifestyle behaviors and cognitive ability in men from midlife to old age. Neurobiol Aging. 2021;108:80-9.\u003c/li\u003e\n \u003cli\u003eManuello J, Min J, McCarthy P, Alfaro-Almagro F, Lee S, Smith S, et al. The effects of genetic and modifiable risk factors on brain regions vulnerable to ageing and disease. Nature Communications. 2024;15(1):2576.\u003c/li\u003e\n \u003cli\u003eLivingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet. 2020.\u003c/li\u003e\n \u003cli\u003eLivingston G, Huntley J, Liu KY, Costafreda SG, Selb\u0026aelig;k G, Alladi S, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. The Lancet. 2024;404(10452):572-628.\u003c/li\u003e\n \u003cli\u003eFjell AM, R\u0026oslash;geberg O, S\u0026oslash;rensen \u0026Oslash;, Amlien IK, Bartr\u0026eacute;s-Faz D, Brandmaier AM, et al. Reevaluating the Role of Education in Cognitive Decline and Brain Aging: Insights from Large-Scale Longitudinal Cohorts across 33 Countries. medRxiv. 2025.\u003c/li\u003e\n \u003cli\u003eMiller KL, Alfaro-Almagro F, Bangerter NK, Thomas DL, Yacoub E, Xu J, et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature neuroscience. 2016;19(11):1523-36.\u003c/li\u003e\n \u003cli\u003eWalhovd KB, Fjell AM, Westerhausen R, Nyberg L, Ebmeier KP, Lindenberger U, et al. Healthy minds 0\u0026ndash;100 years: Optimising the use of European brain imaging cohorts (\u0026ldquo;Lifebrain\u0026rdquo;). European Psychiatry. 2018;50:47-56.\u003c/li\u003e\n \u003cli\u003ePetersen RC, Aisen PS, Beckett LA, Donohue MC, Gamst AC, Harvey DJ, et al. Alzheimer\u0026apos;s Disease Neuroimaging Initiative (ADNI): clinical characterization. Neurology. 2010;74(3):201-9.\u003c/li\u003e\n \u003cli\u003eBookheimer SY, Salat DH, Terpstra M, Ances BM, Barch DM, Buckner RL, et al. The Lifespan Human Connectome Project in Aging: An overview. Neuroimage. 2019;185:335-48.\u003c/li\u003e\n \u003cli\u003eKurdi B, Lozano S, Banaji MR. Introducing the Open Affective Standardized Image Set (OASIS). Behav Res Methods. 2017;49(2):457-70.\u003c/li\u003e\n \u003cli\u003eThompson PM, Stein JL, Medland SE, Hibar DP, Vasquez AA, Renteria ME, et al. The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav. 2014;8(2):153-82.\u003c/li\u003e\n \u003cli\u003eRicard JA, Parker TC, Dhamala E, Kwasa J, Allsop A, Holmes AJ. Confronting racially exclusionary practices in the acquisition and analyses of neuroimaging data. Nature Neuroscience. 2023;26(1):4-11.\u003c/li\u003e\n \u003cli\u003eCardenas-Iniguez C, Gonzalez MR. Recommendations for the responsible use and communication of race and ethnicity in neuroimaging research. Nature Neuroscience. 2024;27(4):615-28.\u003c/li\u003e\n \u003cli\u003eHenrich J, Heine SJ, Norenzayan A. The weirdest people in the world? Behavioral and Brain Sciences. 2010;33(2-3):61-83.\u003c/li\u003e\n \u003cli\u003eFilippini N, Zsoldos E, Haapakoski R, Sexton CE, Mahmood A, Allan CL, et al. Study protocol: the Whitehall II imaging sub-study. BMC psychiatry. 2014;14(1):1-16.\u003c/li\u003e\n \u003cli\u003eBertram L, B\u0026ouml;ckenhoff A, Demuth I, D\u0026uuml;zel S, Eckardt R, Li S-C, et al. Cohort Profile: The Berlin Aging Study II (BASE-II)\u0026dagger;. International Journal of Epidemiology. 2014;43(3):703-12.\u003c/li\u003e\n \u003cli\u003eNilsson L-G, B\u0026Auml;Ckman L, Erngrund K, Nyberg L, Adolfsson R, Bucht G, et al. The betula prospective cohort study: Memory, health, and aging. Aging, Neuropsychology, and Cognition. 1997;4(1):1-32.\u003c/li\u003e\n \u003cli\u003eLam BYK, Leung K, Yiu B, Zhao L, Biesbroek J, Au L, et al. Peak width of skeletonized mean diffusivity and its association with age-related cognitive alterations and vascular risk factors. Alzheimer\u0026apos;s \u0026amp; Dementia: Diagnosis, Assessment \u0026amp; Disease Monitoring 2019;11:721-9.\u003c/li\u003e\n \u003cli\u003eChee MWL, Chen KHM, Zheng H, Chan KPL, Isaac V, Sim SKY, et al. Cognitive Function and Brain Structure Correlations In Healthy Elderly East Asians. NeuroImage. 2009;47:S175.\u003c/li\u003e\n \u003cli\u003eSachdev PS, Brodaty H, Reppermund S, Kochan NA, Trollor JN, Draper B, et al. The Sydney Memory and Ageing Study (MAS): methodology and baseline medical and neuropsychiatric characteristics of an elderly epidemiological non-demented cohort of Australians aged 70\u0026ndash;90 years. International psychogeriatrics. 2010;22(8):1248-64.\u003c/li\u003e\n \u003cli\u003eSmith SM, Nichols TE, Vidaurre D, Winkler AM, Behrens TE, Glasser MF, et al. A positive-negative mode of population covariation links brain connectivity, demographics and behavior. Nature neuroscience. 2015;18(11):1565-7.\u003c/li\u003e\n \u003cli\u003eWinkler AM, Webster MA, Vidaurre D, Nichols TE, Smith SM. Multi-level block permutation. Neuroimage. 2015;123:253-68.\u003c/li\u003e\n \u003cli\u003eSalvan P, Wassenaar T, Wheatley C, Beale N, Cottaar M, Papp D, et al. Multimodal Imaging Brain Markers in Early Adolescence Are Linked with a Physically Active Lifestyle. The Journal of Neuroscience. 2021;41(5):1092.\u003c/li\u003e\n \u003cli\u003eGlymour MM, Weuve J, Berkman LF, Kawachi I, Robins JM. When is baseline adjustment useful in analyses of change? An example with education and cognitive change. American journal of epidemiology. 2005;162(3): 267-78.\u003c/li\u003e\n \u003cli\u003eMeloun M, Militk\u0026yacute; J. 4 - Statistical analysis of multivariate data. In: Meloun M, Militk\u0026yacute; J, editors. Statistical Data Analysis: Woodhead Publishing India; 2011. p. 151-403.\u003c/li\u003e\n \u003cli\u003eHolm S. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics. 1979;6(2):65-70.\u003c/li\u003e\n \u003cli\u003eWang R, Fratiglioni L, Kalpouzos G, L\u0026ouml;vd\u0026eacute;n M, Laukka EJ, Bronge L, et al. Mixed brain lesions mediate the association between cardiovascular risk burden and cognitive decline in old age: A population-based study. Alzheimer\u0026apos;s \u0026amp; Dementia. 2017;13(3):247-56.\u003c/li\u003e\n \u003cli\u003eBinnewies J, Nawijn L, Brandmaier AM, Baar\u0026eacute; WFC, Boraxbekk C-J, Demnitz N, et al. Lifestyle-related risk factors and their cumulative associations with hippocampal and total grey matter volume across the adult lifespan: A pooled analysis in the European Lifebrain consortium. Brain Research Bulletin. 2023;200:110692.\u003c/li\u003e\n \u003cli\u003eSantos MAO, Bezerra LS, Carvalho A, Brainer-Lima AM. Global hippocampal atrophy in major depressive disorder: a meta-analysis of magnetic resonance imaging studies. Trends Psychiatry Psychother. 2018;40(4):369-78.\u003c/li\u003e\n \u003cli\u003eCumplido-Mayoral I, Brugulat-Serrat A, S\u0026aacute;nchez-Benavides G, Gonz\u0026aacute;lez-Escalante A, Anastasi F, Mil\u0026agrave;-Alom\u0026agrave; M, et al. The mediating role of neuroimaging-derived biological brain age in the association between risk factors for dementia and cognitive decline in middle-aged and older individuals without cognitive impairment: a cohort study. The Lancet Healthy Longevity. 2024;5(4):e276-e86.\u003c/li\u003e\n \u003cli\u003eWirth M, Haase CM, Villeneuve S, Vogel J, Jagust WJ. Neuroprotective pathways: lifestyle activity, brain pathology, and cognition in cognitively normal older adults. Neurobiology of Aging. 2014;35(8):1873-82.\u003c/li\u003e\n \u003cli\u003eAdams S, Conner S, Himali JJ, Beiser A, Vasan RS, Seshadri S, et al. Vascular risk factor burden and new-onset depression in the community. Prev Med. 2018;111:348-50.\u003c/li\u003e\n \u003cli\u003eShen J, Tozer DJ, Markus HS, Tay J. Network Efficiency Mediates the Relationship Between Vascular Burden and Cognitive Impairment: A Diffusion Tensor Imaging Study in UK Biobank. Stroke. 2020;51(6):1682-9.\u003c/li\u003e\n \u003cli\u003eMarebwa BK, Adams RJ, Magwood GS, Basilakos A, Mueller M, Rorden C, et al. Cardiovascular Risk Factors and Brain Health: Impact on Long-Range Cortical Connections and Cognitive Performance. J Am Heart Assoc. 2018;7(23):e010054.\u003c/li\u003e\n \u003cli\u003eBl\u0026ouml;chl M, Schaare HL, Kumral D, Gaebler M, Nestler S, Villringer A. Vascular risk factors, white matter microstructure, and depressive symptoms: a longitudinal analysis in the UK Biobank. Psychol Med. 2024;54(1):125-35.\u003c/li\u003e\n \u003cli\u003eHoagey DA, Lazarus LTT, Rodrigue KM, Kennedy KM. The effect of vascular health factors on white matter microstructure mediates age-related differences in executive function performance. Cortex. 2021;141:403-20.\u003c/li\u003e\n \u003cli\u003eHonigsbaum M. Revisiting the 1957 and 1968 influenza pandemics. The Lancet. 2020;395(10240):1824-6.\u003c/li\u003e\n \u003cli\u003eViboud C, Simonsen L, Fuentes R, Flores J, Miller MA, Chowell G. Global Mortality Impact of the 1957-1959 Influenza Pandemic. J Infect Dis. 2016;213(5):738-45.\u003c/li\u003e\n \u003cli\u003eJordan WSJ, Denny FWJ, Badger GF, Curtiss C, Dingle JH, Oseasohn R, et al. A study of illness in a group of Cleveland families: XVII. The occurrence of asian influenza 123. American Journal of Epidemiology. 1958;68(2):190-212.\u003c/li\u003e\n \u003cli\u003eAlmond D. Is the 1918 Influenza Pandemic Over? Long-term Effects of In Utero Influenza Exposure in the Post-1940 U.S. Population. Journal of Political Economy. 2006;114(4):672-712.\u003c/li\u003e\n \u003cli\u003eMarinescu IE, Lawlor PN, Kording KP. Quasi-experimental causality in neuroscience and behavioural research. Nature Human Behaviour. 2018;2(12):891-8.\u003c/li\u003e\n \u003cli\u003eSeblova D, Berggren R, L\u0026ouml;vd\u0026eacute;n M. Education and age-related decline in cognitive performance: Systematic review and meta-analysis of longitudinal cohort studies. Ageing Research Reviews. 2020;58:101005.\u003c/li\u003e\n \u003cli\u003eChan MY, Han L, Carreno CA, Zhang Z, Rodriguez RM, LaRose M, et al. Long-term prognosis and educational determinants of brain network decline in older adult individuals. Nature Aging. 2021;1(11):1053-67.\u003c/li\u003e\n \u003cli\u003eL\u0026ouml;vd\u0026eacute;n M, Fratiglioni L, Glymour MM, Lindenberger U, Tucker-Drob EM. Education and Cognitive Functioning Across the Life Span. Psychol Sci Public Interest. 2020;21(1):6-41.\u003c/li\u003e\n \u003cli\u003eZahodne LB, Glymour MM, Sparks C, Bontempo D, Dixon RA, MacDonald SW, et al. Education does not slow cognitive decline with aging: 12-year evidence from the victoria longitudinal study. J Int Neuropsychol Soc. 2011;17(6):1039-46.\u003c/li\u003e\n \u003cli\u003eSepulcre J. College education as a modulator of the aging brain. Nature Aging. 2021;1(11):980-1.\u003c/li\u003e\n \u003cli\u003eNyberg L, Magnussen F, Lundquist A, Baar\u0026eacute; W, Bartr\u0026eacute;s-Faz D, Bertram L, et al. Educational attainment does not influence brain aging. Proceedings of the National Academy of Sciences. 2021;118(18):e2101644118.\u003c/li\u003e\n \u003cli\u003eSmith ET, Hennessee JP, Wig GS, Frank S, Gonzalez H, Bacci J, et al. Longitudinal changes in gray matter correspond to changes in cognition across the lifespan: implications for theories of cognition. Neurobiology of Aging. 2023;129:1-14.\u003c/li\u003e\n \u003cli\u003eFerreira D, Molina Y, Machado A, Westman E, Wahlund LO, Nieto A, et al. Cognitive decline is mediated by gray matter changes during middle age. Neurobiol Aging. 2014;35(5):1086-94.\u003c/li\u003e\n \u003cli\u003eBallester PL, Suh JS, Ho NCW, Liang L, Hassel S, Strother SC, et al. Gray matter volume drives the brain age gap in schizophrenia: a SHAP study. Schizophrenia. 2023;9(1):3.\u003c/li\u003e\n \u003cli\u003eBrayne C, Ince PG, Keage HA, McKeith IG, Matthews FE, Polvikoski T, et al. Education, the brain and dementia: neuroprotection or compensation? Brain. 2010;133(Pt 8):2210-6.\u003c/li\u003e\n \u003cli\u003eNguyen TT, Tchetgen Tchetgen EJ, Kawachi I, Gilman SE, Walter S, Liu SY, et al. Instrumental variable approaches to identifying the causal effect of educational attainment on dementia risk. Ann Epidemiol. 2016;26(1):71-6.e1-3.\u003c/li\u003e\n \u003cli\u003eDavies NM, Dickson M, Davey Smith G, Windmeijer F, van den Berg GJ. The causal effects of education on adult health, mortality and income: evidence from Mendelian randomization and the raising of the school leaving age. Int J Epidemiol. 2023;52(6):1878-86.\u003c/li\u003e\n \u003cli\u003eShin C, Park MH, Lee S-H, Ko Y-H, Kim Y-K, Han K-M, et al. Usefulness of the 15-item geriatric depression scale (GDS-15) for classifying minor and major depressive disorders among community-dwelling elders. Journal of Affective Disorders. 2019;259:370-5.\u003c/li\u003e\n \u003cli\u003eJenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM. FSL. NeuroImage. 2012;62(2):782-90.\u003c/li\u003e\n \u003cli\u003eSmith S, Zhang Y, Jenkinson M, Chen J, Matthews P, Federico A, et al. Accurate, robust and automated longitudinal and cross-sectional brain change analysis. NeuroImage. 2002;17(1):479-89.\u003c/li\u003e\n \u003cli\u003ePatenaude B, Smith SM, Kennedy DN, Jenkinson M. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage. 2011;56(3):907-22.\u003c/li\u003e\n \u003cli\u003eZhang Y, Brady M, Smith S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. Medical Imaging, IEEE Transactions on. 2001;20(1):45-57.\u003c/li\u003e\n \u003cli\u003eGriffanti L, Zamboni G, Khan A, Li L, Bonifacio G, Sundaresan V, et al. BIANCA (Brain Intensity AbNormality Classification Algorithm): A new tool for automated segmentation of white matter hyperintensities. Neuroimage. 2016;141:191-205.\u003c/li\u003e\n \u003cli\u003eAndersson JL, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage. 2003;20(2):870-88.\u003c/li\u003e\n \u003cli\u003eAndersson JL, Graham MS, Zsoldos E, Sotiropoulos SN. Incorporating outlier detection and replacement into a non-parametric framework for movement and distortion correction of diffusion MR images. Neuroimage. 2016;141:556-72.\u003c/li\u003e\n \u003cli\u003eSchilling KG, Blaber J, Hansen C, Cai L, Rogers B, Anderson AW, et al. Distortion correction of diffusion weighted MRI without reverse phase-encoding scans or field-maps. PLOS ONE. 2020;15(7):e0236418.\u003c/li\u003e\n \u003cli\u003eSmith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, et al. Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data. NeuroImage. 2006;31(4):1487-505.\u003c/li\u003e\n \u003cli\u003eTrzepacz PT, Hochstetler H, Wang S, Walker B, Saykin AJ, Initiative AsDN. Relationship between the Montreal Cognitive Assessment and Mini-mental State Examination for assessment of mild cognitive impairment in older adults. BMC geriatrics. 2015;15:1-9.\u003c/li\u003e\n \u003cli\u003eLo JW, Crawford JD, Desmond DW, Godefroy O, Jokinen H, Mahinrad S, et al. Profile of and risk factors for poststroke cognitive impairment in diverse ethnoregional groups. Neurology. 2019;93(24):e2257-e71.\u003c/li\u003e\n \u003cli\u003eWinkler AM, Renaud O, Smith SM, Nichols TE. Permutation inference for canonical correlation analysis. NeuroImage. 2020;220:117065.\u003c/li\u003e\n \u003cli\u003eGreene WH. Economic Analysis. 7th Edition ed. Boston: Pearson; 2011.\u003c/li\u003e\n \u003cli\u003eVanderweele TJ. Explanation in causal inference: Methods for mediation and interaction.: Oxford University Press.; 2015.\u003cstrong\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"risk factors, cognition, depressive symptoms, healthy aging, neuroimaging, canonical correlation analysis, meta-analysis, multi-cohort","lastPublishedDoi":"10.21203/rs.3.rs-7195000/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7195000/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"The global rise in cognitive impairment calls for preventive strategies through early identification of risk and protective factors in the community healthy elderlies that take into account cultural and geographic diversity. This study investigates how risk and protective factors influence cognitive functioning and depressive symptoms of older adults across six diverse cohorts (n=1,636) from Europe, Asia, and Australia. We found that younger age at baseline and longer education covary with better baseline cognitive function, with a marginal average effect of education beyond individual, geographical, and birth cohort differences. Harnessing multimodal brain MRI, we find that this relationship is mediated by normalised grey matter, with a statistically significant pooled effect across cohorts. Using a natural experiment, we then establish the causal effect of education on cognition six decades later. By including underrepresented populations and by generalising findings, this research extends the evidence base beyond dominant Western-focused research norms, underscoring a call for inclusive and equitable access to education to enhance lifelong cognitive trajectories.","manuscriptTitle":"Risk and protective factors of healthy cognitive ageing across diverse global cohorts and causal effect of education","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-31 16:29:18","doi":"10.21203/rs.3.rs-7195000/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f4d0316f-29a9-4859-ac4d-5efe4ccc93fa","owner":[],"postedDate":"July 31st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":52365704,"name":"Biological sciences/Neuroscience/Cognitive ageing"},{"id":52365705,"name":"Social science/Education"},{"id":52365706,"name":"Health sciences/Neurology/Neurological disorders/Dementia/Alzheimer's disease"}],"tags":[],"updatedAt":"2025-09-19T07:41:44+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-31 16:29:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7195000","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7195000","identity":"rs-7195000","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}