Factors associated with healthy aging in Latin American populations

Abstract

Latin American populations may present patterns of sociodemographic, ethnic and cultural diversity that can defy current universal models of healthy aging. The potential combination of risk factors that influence aging across populations in Latin American and Caribbean (LAC) countries is unknown. Compared to other regions where classical factors such as age and sex drive healthy aging, higher disparity-related factors and between-country variability could influence healthy aging in LAC countries. We investigated the combined impact of social determinants of health (SDH), lifestyle factors, cardiometabolic factors, mental health symptoms and demographics (age, sex) on healthy aging (cognition and functional ability) across LAC countries with different levels of socioeconomic development using cross-sectional and longitudinal machine learning models (n = 44,394 participants). Risk factors associated with social and health disparities, including SDH (β > 0.3), mental health (β > 0.6) and cardiometabolic risks (β > 0.22), significantly influenced healthy aging more than age and sex (with null or smaller effects: β < 0.2). These heterogeneous patterns were more pronounced in low-income to middle-income LAC countries compared to high-income LAC countries (cross-sectional comparisons), and in an upper-income to middle-income LAC country, Costa Rica, compared to China, a non-upper-income to middle-income LAC country (longitudinal comparisons). These inequity-associated and region-specific patterns inform national risk assessments of healthy aging in LAC countries and regionally tailored public health interventions.

Fig. 1. Methodological framework.

a, General design of the study describing the countries included in the cross-sectional analyses (Chile, Uruguay, Colombia and Ecuador) and in the longitudinal analyses (Costa Rica and China). b, Database selection in the cross-sectional (n = 31,680 participants) and longitudinal (n = 9,258 participants) studies (total n = 40,938 participants). c, Imputation procedures. d, Risk factors, including demographics, SDH, health status (cardiometabolic factors and falls), mental health symptoms and lifestyle risk factors. e, Outcomes: cognition and functional ability. f, Multimethod approach, including different regressions (linear regression, elastic net, LASSO, ridge regression). g, Multicollinearity between risk factors that justified the selection of ridge regression as the adequate model. h, Bayesian optimization to find the best hyperparameters for ridge regression. i, Regression step used with the ridge regression. j, Regression report. k, The multimethod results revealed high consistency across methods using goodness-of-fit metrics (R², Cohen’s F₂, MSE and RMSE) and a high coherence in the weight and ranking of risk factors of healthy aging (β estimates). In k, the upper panel presents the multimethod findings related to cognitive performance. The lower panel displays the functional ability data across all countries in Latin America. Risk factors: demographics, SDH, health, lifestyle, mental health symptoms and country.

Fig. 2. Cross-sectional results of cognition and functional ability in Latin America.

a,b, Results for cognition (a) and functional ability (b) for each country, all collapsed countries, as well as LA-HICs (Chile and Uruguay) and LA-LMICs (Ecuador and Colombia). The R², Cohen’s F₂, MSE and RMSE are reported for each model. The feature importance and their statistical significance are also provided for each model. β estimates were used to assess the weight of each feature in the models. The risk factors are: demographics, SDH, lifestyle, health status (cardiometabolic factors and falls), mental health symptoms and country. The two-sided P value of a Student’s t-statistic was calculated for the β values of the regression. *P < 0.05; **P < 0.01. No asterisk means not significant.

Fig. 3. Longitudinal risk factors of cognition and functional ability in an LA-UMIC (Costa Rica) and a non-LA-UMIC (China).

a,b, Longitudinal risk factors of cognition (a) and functional ability (b) were grouped into the following factors: demographics; SDH; lifestyle; health status (cardiometabolic factors and falls); mental health symptoms; and country. Features were ordered from most to least influential in the regression. The feature importance ranks in the regression model for cognition and functional ability are highlighted, accompanied by their statistical significance. Feature importance is represented by the radius of the circles and accentuated by the intensity of the color. The bottom parts of both panels show the countries’ comparison analyses (violin plots) used to test differences in the weight of significant risk factors (β estimates) of cognition and functional ability (n = 9,258). Ten iterations of the results were conducted to obtain ten β estimates for each risk factor, providing the minimum variance for performing group comparisons, which was analyzed with a two-sided Mann–Whitney U-test with Bonferroni correction. The specific values of the violin plots (minimum, maximum, center, 25th and 75th quartiles, inferior and superior whiskers) are provided in Extended Data Table 6 (for cognition) and Extended Data Table 7 (for functional ability). *P < 1.00 × 10⁻² ≤5.00 × 10⁻²; **P < 1.00 × 10⁻³ ≤ 1.00 × 10⁻²; ***P < 1.00 × 10⁻⁴ ≤ 1.00 × 10⁻³.

Extended Data Fig. 1. Correlations between factors and outcomes associated with healthy aging.

Matrix of correlation between factors and outcomes (cognition and functional ability) of healthy aging.

Extended Data Fig. 2. Longitudinal results of factors associated with cognition and functional ability.

Longitudinal predictors of cognition and functional ability. Predictors of each wave of cognition (MMSE scores, panel A) and functional ability (Barthel scores, panel B) are shown for the LA-UMIC (Costa Rica) and the Asian UMIC (China). All panels demonstrated the significance of the predictors assessed in the initial wave for the outcomes evaluated in each wave (wave 1 and wave 2). The risk factors were grouped into fivefold categories: demographics, social determinants of health, lifestyle, health status, mental health symptoms, and country. The features were ordered from most to least influential in the regression. The feature importance is represented by the radius of the circles and accentuated by the intensity of the color.