The exposome of healthy and accelerated aging across 40 countries

Abstract

Protective and risk factors can drive healthy or accelerated aging, with distinct environments modulating their effects. The impact of the exposome-the combined physical and social exposures experienced throughout life-on accelerated aging remains unknown. We assessed delayed and accelerated aging in 161,981 participants from 40 countries (45.09% female; mean age, 67.06; s.d., 9.85) by measuring biobehavioral age gaps (BBAGs), defined as the difference between estimated age from protective and risk factors and chronological age, in cross-sectional and longitudinal designs. BBAGs predicted chronological age, followed by regional and exposomal factor analyses, linked to accelerated aging. Europe led in healthy aging, while Egypt and South Africa showed the greatest acceleration; Asia and Latin America fell in between (Cliff's delta (δd) = 0.15-0.52; all P < 0.0001). Accelerated aging was more evident in eastern and southern Europe; globally, it was also associated with lower income (δd = 0.48-0.56, P < 1 × 10^-15). Exposomal factors of accelerated aging include physical (air quality), social (socioeconomic and gender inequality, migration) and sociopolitical (representation, party freedom, suffrage, elections and democracy) determinants (all Cohen's d (d) > 0.37, P < 0.0001). BBAGs predicted future functional (r (Pearson correlation) = -0.33, P < 1 × 10^-15, d = 0.70) and cognitive declines (r = -0.22, P < 1 × 10^-15, d = 0.44), and larger BBAGs (P < 0.0001, d = 1.55). Healthy and accelerated aging are influenced by physical, social and sociopolitical exposomes, with considerable disparities across nations.

Extended Data Fig. 1. Age distribution by countries

The figure illustrates the age distribution across countries for three different datasets. a) Age distribution for the cross-sectional dataset, which includes a total of 161,981 participants. b) Age distribution for the cross-sectional subsample consisting of 102,725 participants, with additional predictors available for analysis. c) Age distribution for the longitudinal dataset, where the same group of 21,631 participants is observed across two waves (Wave 1 and Wave 2), highlighting the distribution of age by country in each wave.

Extended Data Fig. 2. Supplementary analysis on the cross-sectional subsample

a) Feature importance, assessed via mean decrease in impurity (MDI), enabled the prediction of chronological age using biobehavioral factors. The sample size for the analyses reported in this figure included 102,725 individuals. Goodness-of-fit and feature importance metrics are provided. (b) MDI facilitated the characterization of groups with more delayed (left panel) and more accelerated (right panel) aging. Goodness-of-fit and feature importance metrics are provided. (c) Average BBAGs distribution by continent. The color bar indicates younger (blue) and older (red) BBAGs. (d) BBAGs comparisons by continent (left panel) and by European regions (right panel). (e) BBAGs comparisons between low- and high-income countries, based on gross national income (GNI) and gross domestic product (GDP) indicators. (f) Linear regression models were used to assess the relationship between BBAGs and all exposomes, as well as combined social, physical, and sociopolitical exposomes. (g) Linear regression models were also used to assess the associations between BBAGs and individual social (gender equality, migration, and structural equality) and physical exposomes (air quality). (h) Linear regression models were used to examine the association between BBAGs and individual sociopolitical exposomes (democracy indicators). All P-values reported in Panels f, g, and h were < 1e-15. The maps were created in Python using the Plotly library (https://plotly.com/python/maps/). The authors also designed all other illustrations and icons from scratch using GIMP (https://www.gimp.org), a free design tool.

Extended Data Fig. 3. Sensitivity analyses of exposome effects

The 3D plots display the density of exposomes and BBAGs. The sample size for the analyses reported in this figure included 161,981 individuals. (a) Linear regression models tracked the relationship between BBAGs and combined social exposomes, controlling for physical and sociopolitical exposomes. (b) Linear regression models assessed the role of individual social exposomes (gender equality, migration, and structural equality) and physical exposomes (air quality) on BBAGs, controlling for sociopolitical exposomes. (c) Linear regression models examined the association between BBAGs and individual sociopolitical exposomes (democracy indicators), controlling for social and physical exposomes. We used a one-sided F-test to evaluate the statistical significance of the linear regression model, testing whether the model explains a significantly greater proportion of variance than expected by chance. (d) Linear fit between BBAGs and all exposomes, excluding zones with sparse exposome density, and combined social, physical, and sociopolitical exposomes. (e) Linear regression models assessing the relationship between BAGs and individual social (gender equality, migration, and structural equality) and physical exposomes (air quality), excluding zones with sparse exposome density. (f) Linear regression models examining the association between BBAGs and individual sociopolitical exposomes (democracy indicators), excluding zones with sparse exposome density. All P-values reported in Panels (a) to (f) were < 1e-15.

Extended Data Fig. 4. Longitudinal analysis excluding the healthy aging factor assessed for association

The sample size for the analyses reported in this figure included 21,631 individuals. (a) Linear regression models tracked the relationships between recalculated BBAGs in wave 1 (excluding the specific healthy aging factor used as the outcome) and healthy aging factors (cognition, functional ability, and well-being) in wave 2. Recalculated BBAGs in wave 1 significantly predicted poorer cognition, functional ability, and well-being in wave 2. All P-values reported were < 1e-15.

Extended Data Fig. 5. Longitudinal analysis on the South African dataset

The sample size for the analyses reported in this figure included 5,431 individuals. (a) Linear regression models were used to assess the relationships between BBAGs in wave 1 and healthy aging factors (walking and memory) in wave 2. BBAGs in wave 1 significantly predicted poorer cognition, functional ability, and well-being in wave 2. (b) Linear regression models tracked the relationships between BBAGs in wave 1 as significant predictors of BBAGs in wave 2. All P-values reported were < 1e-15.

Extended Data Fig. 6. Validation using epidemiological metrics and meta-analysis on the cross-sectional subsample and the South African dataset

(a) Odds ratios and attributable risk for the cross-sectional subsample. Results showed that BBAGs are significantly associated with poorer functional ability and cognition. Analyses reported in this panel included 101,870 individuals for cognition, 99,588 for functional ability, and 101,618 individuals for well-being. (b) Relative risk and Attributable risk for the South African dataset. BBAGs in wave 1 were significantly associated with declines in walking and memory, despite the effects on walking ability showing an uncertain confidence interval. Analyses reported in this panel included 3,868 individuals. (c) Meta-analysis on cross-sectional subsample for cognition, functional ability, and wellbeing using common-effects and random-effects models. Larger BBAGs are linked to poorer outcomes in all domains, with high heterogeneity across countries. All P-values reported in this panel were < 0.01. We used Cochran’s Q test to assess heterogeneity across studies, reporting the associated p-value, degrees of freedom, and I² statistic. (d) Summary of cross-sectional results, showing the average of common- and random-effects models by income level (low or high) based on gross national income (GNI) and gross domestic product (GDP) classifications. Color bar indicates effect sizes (small: blue and large: red). Accelerated aging is more strongly linked to poorer healthy aging outcomes—cognition, functional ability, and well-being—in low-income countries compared to high-income countries for both classifications (GNI: p = 4.05e-9, r = 0.82 and GDP: p = 8.35e-4, r = 0.47). The maps were created in Python using the Plotly library (https://plotly.com/python/maps/). The authors also designed all other illustrations and icons from scratch using GIMP (https://www.gimp.org), a free design tool.

Extended Data Fig. 7. Behavioral age and BBAGs calculation without imputation

(a) Feature importance analyses enabled the prediction of chronological age using biobehavioral factors. The sample size for the analyses reported in this figure included 148,188 individuals. Goodness-of-fit and feature importance metrics are provided. (b) Mean decrease in impurity (MDI) facilitated the characterization of groups with more delayed (left panel) and more accelerated (right panel) aging. Goodness-of-fit and feature importance metrics are provided.

Extended Data Fig. 8. Behavioral age and BBAGs modulation by sex

The sample size for the analyses reported in this figure was 148,188 individuals. Panels (a) and (c): Feature importance analyses enabled the prediction of chronological age using biobehavioral factors in females and males, respectively. Goodness-of-fit and feature importance metrics are provided. Panels (b) and (d): Feature importance analyses characterized groups with more delayed (left panel) and more accelerated (right panel) aging in females and males, respectively. Goodness-of-fit and feature importance metrics are provided. Panels (e), (f), and (g): Linear regression models tracked the relationship between BBAGs and various social exposomes across sex: Panel (e): Combined social exposomes. Panel (f): Social and physical exposomes. Panel (g): Sociopolitical exposomes. All P-values reported were < 1e-15.

Fig 1.. Study design and analysis pipeline.

(a) The dataset includes 161,981 participants from aging, health, and well-being surveys across LAC, Europe, Asia (China, South Korea, Israel, and India), and Africa (Egypt and South Africa) (sample sizes by country in parentheses). (b) Chronological age was predicted using harmonized risk and protective factors. (c) A Gradient Boosting regression model, trained on 90% of the dataset with 10-fold cross-validation, estimated chronological age based on these factors. (d) Biobehavioral age gaps (BBAGs) were computed as the difference between predicted and chronological age (BBAG > 0: accelerated aging; BBAG < 0: delayed aging), adjusted for regression-to-the-mean bias. (e) Feature importance was assessed via mean decrease in impurity (MDI), and BBAG distributions were compared across continents and income groups. (f) BBAGs were tested cross-sectionally by different exposomal factors including macro-level social (gender equality, migration, structural inequality), physical (air quality), and sociopolitical (democracy indicators) variables. (g) Longitudinal analyses evaluated BBAGs as predictors of future aging trajectories, examining their association with changes in cognition, functional ability, and well-being over time. (h) Validation analyses included epidemiological metrics (odds ratios and risk ratios) and meta-analyses to assess the validity of associations by income level (low vs. high, based on GNI and GDP classifications). Supplementary Table 7 details subsample datasets. The maps were created in Python using the Plotly library (https://plotly.com/python/maps/). The authors also designed all other illustrations and icons from scratch using GIMP (https://www.gimp.org), a free design tool.

Fig. 2.. Cross-sectional results on BBAGs and multiple associations.

(a) Feature importance, assessed via mean decrease in impurity (MDI), enabled the prediction of chronological age using biobehavioral factors. The sample size for the analyses reported in this figure included 161,981 individuals. Goodness-of-fit and feature importance metrics are provided. (b) MDI facilitated the characterization of groups with more delayed (left panel) and more accelerated (right panel) aging. Goodness-of-fit and feature importance metrics are provided. (c) Average BBAGs distribution by continent. The color bar indicates younger (blue) and older (red) BBAGs. (d) BBAGs comparisons by continent (left panel) and by European regions (right panel). (e) BBAGs comparisons between low- and high-income countries, based on gross national income (GNI) and gross domestic product (GDP) indicators. (f) Linear regression models were used to assess the interaction between BBAGs and all exposomal factors, as well as the combined effects of social, physical, and sociopolitical exposomes. (g) Linear regression models were applied to examine the associations between BBAGs and individual social (gender equality, migration, and structural equality) and physical exposomal factors (air quality). (h) Linear regression models were used to assess the relationship between BBAGs and individual sociopolitical exposomal factors (democracy indicators). All P-values reported in Panels f, g and h were < 1e-15. Extended Fig. 2 and 3 present additional results.

Fig. 3.. Longitudinal results on healthy aging outcomes.

(a) Linear regression models tracking the relationships between BBAGs in wave 1 and healthy aging factors (cognition, functional ability, well-being) in wave 2. The sample size for the analyses reported in this figure included 21,631 individuals. BBAGs in wave 1 significantly predicted adverse outcomes in cognition, functional ability, and well-being in wave 2. (b) Linear regression models assessing BBAGs in wave 1 as significant predictors of BBAGs in wave 2. Results excluding the healthy aging factor assessed for association and those from the South African dataset are shown in Extended Fig. 4 and 5, respectively. All P-values reported in Panels A and B were < 1e-15.

Fig. 4.. Validation via epidemiological metrics and meta-analysis.

(a) Odds ratios and attributable risk for the cross-sectional analysis. Results showed that BBAGs are significantly associated with poorer functional ability and cognition. Cognition results were based on 161,564 participants, and functional ability results on 157,825 participants. (b) Relative risk and attributable risk for the longitudinal analysis. BBAGs in wave 1 were significantly associated with declines in functional ability, cognition, and well-being. Cognition, functional ability, and well-being results were based on 20,597 participants. (c) Meta-analysis of cross-sectional data for cognition and functional ability using common-effects and random-effects models. Larger BBAGs were associated with poorer outcomes in both domains, with high heterogeneity across countries. All P-values reported in the meta-analysis of cross-sectional data were < 0.01. (d) Meta-analysis of longitudinal data for cognition, functional ability, and well-being using common-effects and random-effects models. Increased BBAGs predicted declines across all outcomes, particularly for functional ability, with high heterogeneity. All P-values reported in the meta-analysis of longitudinal data were < 0.01. (c-d) We used Cochran’s Q test to assess heterogeneity across studies, reporting the associated p-value, degrees of freedom, and I² statistic. (e) Summary of cross-sectional and longitudinal results, showing the average effect sizes from common- and random-effects models by income level (low or high), based on gross national income (GNI) and gross domestic product (GDP) classifications. The color bar represents effect sizes (blue: small; red: large). Accelerated aging was more strongly associated with worse healthy aging outcomes—cognition, functional ability, and well-being—in low-income countries across LAC, Asia (China, South Korea, and India), and Egypt. Extended Fig. 6 and Supplementary Tables 1–3 provide complementary results. The maps were created in Python using the Plotly library (https://plotly.com/python/maps/). The authors also designed all other illustrations and icons from scratch using GIMP (https://www.gimp.org), a free design tool.