Early-life determinants of cardiometabolic outcomes and accelerated biological ageing in Colombia

Study population and characteristics

Participants’ characteristics stratified by gender are presented in Table 1. In both samples (overall and biomarkers subsample), study participants were on average 69.1 (S.D. 7.05) years old and 56% were women. Most of the participants were Mestizo (42-44%), followed by White participants (27-29%). The study population had a high prevalence of low childhood socioeconomic position (73-79%), indicating an extensive experience of early life disadvantage within this generation.

ACEs were also prevalent. Only 28.6% of the entire sample reported experiencing no ACEs, meaning over 70% had faced at least one significant adversity during childhood. The most commonly reported ACEs included scarcity of food (25-28%), poor childhood health (11-12%), and domestic violence (16-19%). A gender difference emerged in the experience of domestic violence, with women consistently reporting higher prevalence (19.2% overall sample; 18.4% in the biomarker subsample) compared to men (15% overall sample; 15.9% biomarker subsample). While other ACEs showed minor fluctuations between sexes, this particular difference was pronounced. The cumulative burden of ACEs also highlighted the widespread nature of these experiences, with a considerable proportion reporting multiple ACEs: two ACEs (16-20%), three ACEs (8-11%) and four or more ACEs (2-4%). Participants reporting more than one ACE were more likely to be women and non-white (p<0.001). Women more frequently reported emotional abuse and domestic violence compared to males. Men were more likely to report poor health status and scarcity of food. By ethnic groups, Indigenous and Afro-Colombian women had higher ACEs scores than White women and men.

Cardiometabolic outcomes

Analysis of adult cardiometabolic outcomes showed significant gender differences. Hypertension was highly prevalent, affecting over half of the population (52%). Women showed a higher prevalence of hypertension (57.6% overall; 59.5% biomarker subsample) compared to men (52.6% overall; 59.6% biomarker subsample). Diabetes was also more prevalent among women than men (18% vs 13.6% overall sample). The most significant gender difference was observed in obesity, where women consistently showed a substantial higher prevalence (38.4% overall; 30.4% biomarker subsample) compared to men (27% overall; 18.2% biomarker subsample).

Supplementary Table 4 reports odds ratios and prevalence ratios for the association between ACEs and cardiometabolic outcomes in the overall sample. In adjusted models for demographic factors and low childhood SEP, women exposed to emotional abuse showed a strong and significant association with increased odds/prevalence of CVD (OR=1.68, PR=1.51, both p<0.001). This was not observed in men or for other cardiometabolic outcomes in either gender. Similar to emotional abuse, childhood maltreatment showed a strong and significant association with increased risk of CVD in women (OR=1.55, PR=1.44, both p<0.001), but this was not observed for other outcomes. Associations with CVD in men were less strong (OR=1.24, PR=1.20, p=0.020 and p=0.018 respectively).

Forced childhood migration was associated with increased risk of diabetes (OR=1.60, PR=1.47, p=0.005 and p=0.003 respectively), CVD (OR=1.55, PR=1.46, p=0.012 and p=0.006 respectively), and hypertension (OR=1.43, PR=1.20, p=0.008 and p=0.002 respectively) only among men. Scarcity of food was associated with increased risk of CVD in women (OR=1.44, PR=1.34, both p<0.001), but not with other outcomes. Poor childhood health was associated with increased risk of CVD (OR=1.66, PR=1.50, both p<0.001) in women and hypertension (OR=1.16, PR=1.09, p=0.027 and p=0.013) in men.

Supplementary Table 4 also illustrates the dose-response relationship between the cumulative adverse childhood experiences and the risk of cardiometabolic diseases, particularly strong among women. Tests for trend show that as the number of ACEs increases, the risks of cardiovascular diseases (p=<0.001), diabetes (women p=0.002 and men p=0.006) and hypertension (women p=0.005 and men p=0.0002) also increase. No significant dose-response association was observed between ACEs and obesity in either gender. In general, the confidence intervals for women were narrower than those for men.

Biological age acceleration

We observed strong correlations between biological age and chronological age (KDMAge r = 0.83), which was expected since chronological age is a component of the KDM clock (Supplementary Table 3). Supplementary Table 5 reports the associations between adverse childhood experiences and biological ageing. Among all exposures, only forced childhood migration was associated with accelerated aging, with a stronger association in women. In the overall population, forced childhood migration was associated with 1.52-year higher biological age (∆KDMAge β =1.52, 95% CI = 0.57, 2.47, p=0.002). The association was substantially stronger in women (β =2.67, 95% CI = 1.24 – 4.10), indicating that women exposed to forced migration were on average, 2.67 years biologically older than their chronological age. Forced childhood migration was also associated with higher odds of accelerated ageing (OR=1.69, 95% CI = 1.09 – 2.48, p=0.019). These findings, suggest that exposures to early-life forced migration in conflict settings has a significant long-term impact on biological ageing, with greater vulnerability observed among women.

Poor childhood health was associated with higher biological age acceleration in later lifer (∆KDMAge β=0.53, 95% CI =0.04 – 1.01). In contrast, food scarcity during childhood showed a protective effect among men (β=-0.35, 95% CI= 0.73 – 0.03), suggesting lower biological age acceleration, although the association was not statistically significant. Similarly, cumulative childhood adversity in men was associated with lower biological age acceleration (β=-0.44; 95% CI -0.86 – 0.02). This unexpected finding may reflect unmeasured resilience pathways or suggesting a potential adaptative or selective-survivorship pattern or some type of bias; therefore, but this pattern requires cautious interpretation. No associations were observed between emotional abuse and accelerated biological aging. Adjusting for age, gender, ethnicity and low childhood SEP produced minimal changes in effects sizes or significant values, indicating that these factors were not major confounders.

Limitations of the study

There are several limitations that need to be considered. First, an important limitation is the use of a deterministic approach to model biological aging. Contemporary geroscience views aging as a non-linear, stochastic process. While the KDM provides mean estimates suitable for population-level analyses, it may not capture stochastics bursts of accelerated aging at the individual level. Future longitudinal studies may better capture these dynamics. Second, the study was based on a retrospective investigation of ACEs and other risk factors, i.e., the design is limited in its ability to establish causal associations and we may not have been able to account for all confounders. Third there was limited ACEs assessment in SABE. For example, there was no information on other vital ACEs such as physical and sexual child abuse, parental mental illness, unemployment, and racism. Fourth, SABE did not collect data on the context of exposure, duration, and severity. Fifth, self-reporting by older adults may introduce bias in ACEs recall or measurement. This could be particularly true for adversities like emotional abuse and domestic violence, which were still broadly socially accepted in the study context when adversities occurred in our population from 1955 to 1965. While people with self-reported diseases like cardiometabolic diseases can be biased in their recall of past ACEs, this is unlikely for people showing accelerated ageing based on biomarker measurements. A sixth concern is that the study relies on self-reported diagnoses for conditions such as hypertension and diabetes. While self-reports are commonly used in large epidemiological studies when clinical measurements are unavailable, they may result in underreporting of these conditions and could introduce bias.

A final concern is whether a customised version of the BA predictors can accurately compute the Klemera and Doubal index. Other studies have encountered this limitation due to data availability or the absence of biomarkers included in the original list. A number of studies using simplified biological ageing estimates, or different combinations of biomarkers to estimate BA suggests that these can be as reliable as traditional methods [46, 54–59]. Despite the extensive developments in quantifying BA rates, a universally accepted method for estimating BA is not currently available. In addition, studies in biological ageing show different methods of data collection, data availability, and association assumptions in different settings.

Study design and participants

We used data from the Health, Well-Being, and Ageing Study (SABE-Colombia, first wave), which employed multistage random cluster sampling across 240 municipalities. Of 36,153 eligible older adults, 23,694 were surveyed (66% response rate). The present study included 18,044 participants with complete ACEs and self-reported health data during household interviews (Supplementary Figure 1). A subsample of 4,092 individuals from the full cohort, randomly sampled from five major cities (Barranquilla, Bogotá, Cali, Medellín, and Bucaramanga) was selected for biomarker analysis (Supplementary Figure 2).

Each participant provided written informed consent, and a local ethics committee approved the study. Blood samples were collected using a vacutainer tube and stored at–80° C following standardized protocols. Further information on the representativeness of the sample, survey implementation, data collection, quality control, and ethical approval is provided elsewhere [60, 61]. The study excluded institutionalized older people or those with significant cognitive impairment. Data on social determinants, physical function, behaviors, self-reported health status, and blood samples were collected in 2015. Questionnaires were filled by interviewers, and blood pressure, anthropometric measurements and blood samples were taken at each respondent's residence.

Adverse childhood experiences

We adapted a list of ACEs from World Health Organization (WHO) ACE-IQ questionnaire [62]. As the original SABE study did not capture all ACEs, we supplemented it with emotional abuse and forced migration available in other SABE’s questionnaires. Five ACEs were included: emotional abuse, scarcity of food, poor childhood health, domestic violence, and childhood migration. All responses were dichotomized (‘yes/one or more times’ vs. ‘no/never’) to indicate exposure. Forced childhood migration was defined as a positive response to having being ‘displaced due to armed conflict’ and based on reported date of forced migration prior to age 15.

Biomarkers

We included 11 out of 17 markers representing cardiometabolic function for the quantification of individual biological ages, including triglycerides (mg/dL), glycated haemoglobin (HbA1C) (%), low-density lipoprotein (LDL) cholesterol (mg/dL), high-density lipoprotein (HDL) cholesterol (mg/dL), non-HDL cholesterol (mg/dL), total cholesterol (mg/dL), systolic blood pressure (mmHg), arm circumference (cm), calf circumference (cm), body mass index (BMI) and A Body Shape Index (ABSI) [63]. ABSI is a measure of body shape that estimates risk of related abdominal obesity; while BMI only considers heigh and weight, ABSI also incorporates waist circumference to better account for the distribution of body fat, particularly visceral around the abdomen, which is a known risk factor for mortality [63]. The measurement protocols for these markers have been previously described [64, 65]. Diastolic blood pressure, pulse pressure, waist circumference, knee height, 30-Second Sit-to-Stand Test, and handgrip strength were excluded because they had missing values between 30%-49%. Height and weight were excluded to avoid multicollinearity with body mass index BMI and ABSI. Markers with skewed distributions were log-transformed (HbA1C and triglycerides).

Biological age

The Klemera-Doubal Method for Biological Age (KDMAge) is reported in Supplementary Table 1 and was calculated using the BioAge R package [66, 67]. KDMAge predicts individuals’ biological age and provides insights into accelerated aging. KDMAge employs a relatively simple statistical model to predict BA based on a limited set of organ functions and inflammatory markers. KDMAge has been demonstrated to be a reliable predictor of mortality [68] and provides stable results for practical assessment of the BA [55]. In addition, as it provides a more accessible method for quantifying BA, it is one of the most cross-validated biological age indicators [52, 55, 56, 67, 69, 70].

For our linear regression models, we utilized ∆KDMAge residuals, where negative values indicate slower aging and positive values indicate faster aging. For our logistic regression, the continuous KDMAge variable was dichotomized to classify individuals into “KDMAge acceleration” or “KDMAge non-acceleration”. Participants with positive KDM Age values > chronological age were classified as “KDMAge acceleration”; while participants with negative or zero values < chronological age were classified as “KDMAge non-acceleration”. Supplementary Table 2 describes the models used for the KDMAge.

We validated KDM implementation by: i) replicating NHANES-derived parameters from Levine [71], This means that our implementation of KDMA biological age algorithm was correct by reproducing the same parameters previously derived from NHANES data; ii) confirming age-biomarker correlations in our data were within ranges consistent with prior KDM application (r=0.73-0.89), iii) verifying that the distribution of biological age and its correlations with chronological age were comparable to those reported in studies relative to prior studies using older populations [22, 53, 56, 72]; and iv) conducting a sensitivity analyses using alternative biomarkers combination separately for women, men, and the overall population. For each specification, KDM parameters were recalibrated and biological age acceleration recomputed; across models, the magnitude and direction of associations remained unchanged.

Diseases in adulthood

SABE provides data on outcomes from self-reported diagnoses of cardiovascular diseases (CVD), hypertension (HTA), and diabetes. CVD includes self-reported episodes of heart attack, angina, and thrombosis. BMI was calculated based on physical measurements of height and weight (kg / [height (m)]²) and used a cut-off to indicate obesity (BMI >=30).

Covariates

The covariates included in the analyses were demographic factors (age, sex, and ethnicity) and low childhood socioeconomic position. These variables were self-reported at wave 1, at the same time as the biological samples and measures were taken. The ethnic groups of the participants were Mestizo (mixed European and Amerindian heritage), white, Afro-Colombian (including raizales, palenqueros, and Afro-descended populations), indigenous, and others. Low childhood socioeconomic position was assessed using a question about socioeconomic position before 15 years old. Responses were categorized as high, intermediate, or low, and then dichotomised into “Yes” (low) and “No” (high or intermediate). Supplementary Figure 3 presents the Direct Acyclic Graph (DAG) illustrating the role of confounders and competing exposures in the relationship between ACEs, biological ageing, and health outcomes.

Statistical analysis

To study ACE exposures, we first examined each ACE individually and then calculated an overall ACEs score [18] by summing the number of single adversities.

To examine the associations between single and cumulative ACEs with four diseases, we conducted logistic regression and Poisson regression analyses stratified by gender. We fitted two sets of models for each ACE operationalisation: a crude model and a model adjusted for age, ethnicity and low childhood SEP. To test the trend with the number of ACEs, we performed a Cochran Armitage test. Using both methods, odds ratios and prevalence ratios, provided a more complete understanding of the associations and ensured that results were not biased by method-specific limitations.

Associations between ACEs and biological age measures were estimated using (1) linear regression (continuous ∆KDMAge residual) and (2) logistic regression (dichotomous KDMAge acceleration as dependent variable). Results are interpretable as effect sizes on Beta coefficients and odds ratios, accompanied by 95% confidence intervals and p-values. A two-sided p-value of < 0.05 was considered statistically significant. Several sensitivity analyses were conducted by comparing results with published biological age estimates and by replicating the same models separately in the training and testing datasets.

All analyses were performed in R Studio version 4.1.2 (The R Foundation for Statistical Computing, Vienna, Austria). R packages used in the analysis included compareGroups to validate correlations, and the ggbetweenstats package [73] and the Sjplot package for regression modelling [74].