Abstract

Background:

Whether links between adulthood stressful life events (SLEs) and heart disease are comparable across countries—and whether they follow a dose–response pattern—remains uncertain.

Methods:

We harmonized exposure, outcome, covariates, and model specifications across three nationally representative aging cohorts in China (CHARLS), the United States (HRS), and the United Kingdom (ELSA). Cox models using attained age as the time scale estimated the association between adulthood SLEs (any vs. none) and incident heart disease; we also assessed dose–response (0, 1, 2, ≥3 events). Robustness was evaluated using alternative exposure definitions, an alternative event-date specification, multiple imputation, Fine–Gray competing-risk models (where mortality was available), alternative handling of physical activity, exclusion of events occurring within the first 1 or 2 years of follow-up, and an alternative time scale. Exploratory subgroup analyses with multiplicative interaction terms probed potential effect modification.

Results:

The analysis included 11,240 (CHARLS), 13,099 (HRS), and 3,390 (ELSA) participants and documented 2,074, 2,124, and 678 incident events over 84.2, 95.3, and 26.7 thousand person-years, respectively. Compared with no SLEs, reporting at least one SLE was associated with a higher risk of incident heart disease: adjusted HR 1.20 (95% CI 1.09–1.31) in CHARLS, 1.23 (1.11–1.36) in HRS, and 1.53 (1.27–1.85) in ELSA. Under the original six-item definition, risk increased across SLE-count categories in HRS and ELSA, whereas in CHARLS, risk was elevated in the 1- and 2-event categories but did not show a clear further increase in the ≥3 category. Most sensitivity analyses yielded similar estimates; however, associations were attenuated under alternative exposure definitions excluding health- and injury-related items, particularly in HRS and ELSA. Effect modification was limited to HRS, where associations were stronger in women and in participants without hypertension or diabetes.

Conclusion:

Across three national cohorts, adulthood SLEs were associated with a small-to-moderate increase in incident heart disease, although the graded association varied across cohorts. The present study adds cross-nationally comparable evidence on adulthood adversity and heart disease risk under a harmonized analytic framework.

1 Introduction

The prevalence of cardiovascular diseases (CVDs) continues to rise in parallel with global population aging and now represents a major cause of mortality and disability. Clinically, various forms of heart disease account for well over half of all CVD-related deaths, making them the predominant contributors to the global cardiovascular burden (1). In addition to classical risk factors such as hypertension, diabetes, and dyslipidemia, psychosocial stress is increasingly recognized as a significant contributor to cardiovascular disease. It exerts its effects through multiple biological and behavioral pathways, including activation of the hypothalamic–pituitary–adrenal (HPA) axis and the sympathetic–adrenal–medullary (SAM) system, promotion of chronic inflammation and endothelial dysfunction, increased metabolic load, and clustering of adverse health behaviors such as smoking, alcohol consumption, and physical inactivity (2–6). Together, these pathways accelerate the development of atherosclerosis and increase the risk of adverse cardiovascular events (5, 7, 8). However, the measurement of “stress” remains inconsistent across population-based studies. Compared with general perceived stress scales, adulthood stress life events, which are based on discrete life events, may better reflect real-life stressors and facilitate harmonization and comparability across cohorts.

Adulthood stressful life events (SLEs) are defined as major life changes occurring during adulthood that carry significant threat or personal significance. These events include both chronic or structural stressors—such as unemployment and financial strain—and acute or catastrophic stressors such as bereavement, serious accidents, or physical assault (9, 10). Existing evidence generally indicates that stress-related exposures are associated with an elevated risk of coronary heart disease (CHD) and other cardiovascular outcomes. For instance, job strain has been linked to an elevated risk of CHD (11), and psychosocial stress has shown an independent association with acute myocardial infarction (12). The short-term risk of acute myocardial infarction or atrial fibrillation increases significantly following bereavement (13, 14). Moreover, cumulative exposure to multiple types of stressful life events across the life course has been positively associated with a higher likelihood of being diagnosed with heart disease or stroke in later life (15). However, heterogeneity in the measurement and definition of adulthood stress events across countries, cultures, and survey systems limits the comparability of effect sizes and their external generalizability. Furthermore, evidence on the dose–response relationship of cumulative stress exposure remains insufficient; it is still unclear whether an increasing number of stressful life events corresponds to a consistent gradient of cardiovascular risk. These limitations highlight the need for more rigorous and replicable estimates under standardized, cross-national frameworks.

This study integrates data from three nationally representative aging cohorts—CHARLS (16), HRS (17), and ELSA (18). Under a harmonized analytic framework, SLEs were defined as the primary exposure, with event count also considered. Incident heart disease was treated as the main outcome, and Cox proportional hazards (PH) models using attained age as the time scale were applied to estimate comparable, stratified, and covariate-adjusted associations. Multiple prespecified sensitivity analyses were conducted to assess the robustness of the association between adult SLEs and incident heart disease. This study design aims to enhance both internal consistency and external generalizability. If robust and dose-dependent associations are observed, the findings could support the integration of stress assessment and intervention into primary cardiovascular prevention and risk stratification strategies for older adults across diverse populations. In addition, the results may inform future research on more refined characterization and intervention targeting of stress exposures.

2 Materials and methods2.1 Study population

We analyzed three nationally representative aging cohorts: CHARLS, HRS, and ELSA. These three nationally representative aging cohorts have harmonized data resources available through the Gateway to Global Aging Data (19), which facilitates cross-national comparative research; prior studies have also used these cohorts in cross-national analyses (20, 21). To minimize cross-study heterogeneity and enable comparability, analyses were restricted to harmonized periods: CHARLS waves 1–5 (2011–2020), HRS waves 10–15 (2010–2020), and ELSA waves 5–10 (2010–2020). Eligible respondents were aged ≥ 45 years, free of heart disease at baseline, and had information on SLEs and covariates. Baseline was defined as the first wave with SLEs ascertained; covariates were taken from the same wave. In CHARLS, physical activity used a missing-indicator approach, whereas complete-case data were required for other covariates in the primary analysis. Participants with no follow-up time after baseline were excluded (Figure 1).

Flow of participant inclusion for CHARLS, HRS, and ELSA.

2.2 Exposure assessment

Adulthood SLEs were assessed using six predefined self-reported items: unemployment, asset poverty, death of a child, death of a spouse/partner, life-threatening illness or accident, and physical attack/injury. Detailed item-level harmonization across CHARLS, HRS, and ELSA is provided in Supplementary Table S1. The primary exposure was defined as a binary variable indicating the presence of any adulthood SLE (≥1 vs. 0). Participants were coded as exposed if any item was endorsed and as unexposed only if all six items were reported as No.

2.3 Outcome ascertainment

The primary outcome was incident heart disease, defined as a new self-reported, physician-diagnosed heart condition during follow-up, including heart attack, angina, coronary heart disease, heart failure, or other heart problems. Participants with heart disease at baseline were excluded. For the primary analysis, the event date was defined as the midpoint between the last interview without reported heart disease and the first interview at which heart disease was reported. In a sensitivity analysis, the event date was alternatively defined as the date of the first interview at which heart disease was reported.

2.4 Covariates

A priori covariates were harmonized across cohorts: sex (male/female), marital status (married or cohabiting vs. living alone), education (below high school, high school, college, or above), physical activity, smoking status, drinking status, hypertension, and diabetes. Because attained age was used as the underlying time scale in the primary Cox models, baseline age was not additionally adjusted. In CHARLS, physical activity had substantial missingness and was handled using a missing-indicator approach (active/inactive/missing) rather than multiple imputation. Body mass index (BMI) was excluded from primary models owing to substantial missingness in CHARLS and HRS and the absence of baseline BMI in ELSA; it was considered only in a sensitivity analysis restricted to participants with non-missing BMI.

2.5 Statistical analysis

Baseline characteristics by SLE exposure were summarized as mean ± SD for continuous variables and n (column %) for categorical variables; between-group differences were evaluated using t-tests or χ² tests, as appropriate.

2.5.1 Primary analysis

For cohort-specific associations between adulthood SLEs and incident heart disease, we fit Cox proportional hazards models with attained age as the underlying time scale. Person-time accrued from the baseline interview to incident heart disease, death, loss to follow-up, or administrative censoring, whichever occurred first. We estimated four hierarchical models: Crude (unadjusted); Model 1 adjusted for sex, marital status, and education; Model 2 additionally adjusted for smoking status, drinking status, and physical activity; and Model 3 further adjusted for hypertension and diabetes. The reference group was participants reporting no adulthood SLEs.

2.5.2 Dose–response analysis

Among participants with complete information on all six adulthood SLE items, we additionally examined the number of adulthood SLEs as a count-based exposure, categorized as 0, 1, 2, and ≥3. Hazard ratios were estimated using Cox proportional hazards models with the 0-event category as the reference. P for trend was obtained from a Wald test of the ordinal SLE count term in the Cox model. For comparability, alternative count-based analyses using the five-item and four-item definitions were also conducted within the same six-item complete-case sample.

2.5.3 Missing data

Missingness in variables included in the original multiple imputation framework is summarized in Supplementary Table S2. Missing covariate data were handled using multiple imputation by chained equations under the missing-at-random assumption, generating 10 imputed datasets. Only covariates with missing values were imputed; the main exposure, outcome variable, and time-to-event/censoring variables were not imputed. The imputation model included the exposure, outcome indicator, and all covariates used in the fully adjusted model, as well as time-to-event/censoring information to improve prediction of missing values. Estimates were combined using Rubin's rules. In CHARLS, physical activity was not imputed because it was handled using a missing-indicator approach in the main analysis.

2.5.4 Proportional hazards diagnostics and robustness checks

We assessed the PH assumption using scaled Schoenfeld residuals (global and covariate-specific tests). When the global test indicated departures for selected covariates (not the exposure), we conducted robustness checks using: (i) stratified Cox models via strata() for the offending covariate(s); (ii) models with time-varying coefficients implemented as interactions with log(time); and (iii) models with cluster-robust (sandwich) standard errors at the individual level. Across these specifications, the exposure estimate was materially similar to the primary model (Supplementary Table S3).

2.5.5 Secondary and sensitivity analyses

Several secondary and sensitivity analyses were conducted to assess the robustness of the findings. First, to evaluate whether the observed associations were influenced by health- or injury-related items, we additionally examined alternative adulthood SLE composite definitions by excluding life-threatening illness/accident from the original six-item definition (five-item definition), and then by further excluding physical attack/injury (four-item definition), in both binary and count-based analyses. Second, to assess the impact of event-date specification, we redefined the event date as the date of the first interview at which heart disease was reported, rather than the midpoint between the last negative and first positive interviews. Third, multiple imputation analyses were performed by repeating Model 3 on the imputed datasets. Fourth, in CHARLS and HRS, where mortality data were available, competing-risk analyses were performed using Fine–Gray subdistribution hazard models with follow-up time as the time scale, treating all-cause mortality as the competing event and additionally adjusting for baseline age. Fifth, we reestimated Model 3 using follow-up time as the time scale, with additional adjustment for baseline age. Sixth, we additionally adjusted for BMI in the subset with non-missing baseline BMI (CHARLS and HRS only). Seventh, to assess the influence of physical activity handling, we conducted sensitivity analyses using three alternative strategies: no adjustment for physical activity, restriction to participants with non-missing physical activity, and multiple imputation in which physical activity was imputed. Eighth, to reduce potential reverse causation, we repeated the fully adjusted analyses after excluding incident heart disease events occurring within the first 1 and first 2 years of follow-up, respectively. Finally, exploratory subgroup and interaction analyses were conducted for age (≤65 vs. >65 years), sex, marital status, education, physical activity, smoking, drinking, hypertension, and diabetes by adding multiplicative interaction terms to Model 3 and reporting P values for interaction. No adjustment for multiplicity was applied.

A two-sided P < 0.05 was considered statistically significant. Analyses were conducted in R, version 4.5.1.

3 Results3.1 Study population

A total of 11,240 (CHARLS), 13,099 (HRS), and 3,390 (ELSA) participants were included. Across cohorts, individuals reporting ≥1 adulthood SLE were older and more likely to live alone. In CHARLS and ELSA, exposed participants had higher prevalences of hypertension at baseline. Educational attainment was consistently lower among exposed participants in all three cohorts. Physical inactivity was more common in the exposed group in HRS and ELSA (Table 1).

VariablesAdulthood stressful life eventsP-valueTotalUnexposedExposedCHARLS11,2405,3215,919Age, year59.5 ± 9.857.8 ± 8.461.1 ± 10.6<0.001Sex, Male5,417 (48.2)2,566 (48.2)2,851 (48.2)0.952Marital status<0.001 Married or living with a partner9,580 (85.2)5,216 (98.0)4,364 (73.7) Living alone1,660 (14.8)105 (2.0)1,555 (26.3)Education level<0.001 Below high school10,168 (90.5)4,724 (88.8)5,444 (92) High school944 (8.4)526 (9.9)418 (7.1) College or above128 (1.1)71 (1.3)57 (1.0)Physical activity<0.001 Inactive1,535 (13.7)693 (13.0)842 (14.2) Active3,141 (27.9)1,633 (30.7)1,508 (25.5) NA6,564 (58.4)2,995 (56.3)3,569 (60.3)Smoking status0.243 No7,679 (68.3)3,664 (68.9)4,015 (67.8) Yes3,561 (31.7)1,657 (31.1)1,904 (32.2)Drinking status0.538 No7,466 (66.4)3,519 (66.1)3,947 (66.7) Yes3,774 (33.6)1,802 (33.9)1,972 (33.3) Hypertension2,546 (22.7)1,144 (21.5)1,402 (23.7)0.006 Diabetes573 (5.1)279 (5.2)294 (5)0.506HRS13,0993,9749,125Age, year66.3 ± 11.064.7 ± 9.667.0 ± 11.5<0.001Sex, male5,016 (38.3)1,687 (42.5)3,329 (36.5)<0.001Marital status<0.001 Married or living with a partner7,528 (57.5)3,277 (82.5)4,251 (46.6) Living alone5,571 (42.5)697 (17.5)4,874 (53.4)Education level<0.001 Below high school2,543 (19.4)426 (10.7)2,117 (23.2) High school7,658 (58.5)2,320 (58.4)5,338 (58.5) College or above2,898 (22.1)1,228 (30.9)1,670 (18.3)Physical activity<0.001 Inactive2,671 (20.4)528 (13.3)2,143 (23.5) Active10,428 (79.6)3,446 (86.7)6,982 (76.5)Smoking status<0.001 No11,074 (84.5)3,551 (89.4)7,523 (82.4) Yes2,025 (15.5)423 (10.6)1,602 (17.6)Drinking status<0.001 No5,724 (43.7)1,448 (36.4)4,276 (46.9) Yes7,375 (56.3)2,526 (63.6)4,849 (53.1) Hypertension7,049 (53.8)1,926 (48.5)5,123 (56.1)<0.001 Diabetes2,475 (18.9)598 (15)1,877 (20.6)<0.001ELSA3,3901,2262,164Age, year65.3 ± 9.061.0 ± 5.867.8 ± 9.5<0.001Sex, male1,444 (42.6)576 (47)868 (40.1)<0.001Marital status<0.001 Married or living with a partner2,110 (62.2)1,049 (85.6)1,061 (49.0) Living alone1,280 (37.8)177 (14.4)1,103 (51.0)Education level<0.001 Below high school1,016 (30.0)219 (17.9)797 (36.8) High school1,712 (50.5)692 (56.4)1,020 (47.1) College or above662 (19.5)315 (25.7)347 (16.0)Physical activity<0.001 Inactive482 (14.2)71 (5.8)411 (19.0) Active2,908 (85.8)1,155 (94.2)1,753 (81.0)Smoking status<0.001 No2,909 (85.8)1,088 (88.7)1,821 (84.1) Yes481 (14.2)138 (11.3)343 (15.9)Drinking status<0.001 No436 (12.9)79 (6.4)357 (16.5) Yes2,954 (87.1)1,147 (93.6)1,807 (83.5) Hypertension1,394 (41.1)400 (32.6)994 (45.9)<0.001 Diabetes346 (10.2)78 (6.4)268 (12.4)<0.001

Baseline characteristics by exposure to adulthood stressful life events in CHARLS, HRS, and ELSA.

Values are mean ± SD or n (%). In CHARLS, physical activity includes an explicit “NA” category representing missing responses.

Bold values indicate the cohort labels and the corresponding total, unexposed, and exposed sample sizes for each cohort.

3.2 Follow-up and events

Over 84.2, 95.3, and 26.7 thousand person-years of follow-up in CHARLS, HRS, and ELSA, the median follow-up was 8.9 years (IQR 5.5–9.0), 8.9 (4.8–9.8), and 8.1 (4.9–11.4), respectively. During follow-up, we observed 2,074, 2,124, and 678 incident heart-disease events in the three cohorts, respectively.

3.3 Association between adulthood SLEs and incident heart disease

Exposure to any adulthood SLE was associated with a higher hazard of incident heart disease in all cohorts. In fully adjusted models (Model 3), the HRs were 1.20 (95% CI: 1.09–1.31) in CHARLS, 1.23 (95% CI: 1.11–1.36) in HRS, and 1.53 (95% CI: 1.27–1.85) in ELSA (all P < 0.001). Effect estimates were largely stable from crude to adjusted models in CHARLS, attenuated modestly in HRS, and strengthened slightly in ELSA (Table 2).

VariableCrude modelModel 1Model 2Model 3HR (95% CI)P-valueHR (95% CI)P-valueHR (95% CI)P-valueHR (95% CI)P-valueCHARLS Unexposed1(Ref)1(Ref)1(Ref)1(Ref) Exposed1.16 (1.06–1.26)0.0011.20 (1.09–1.31)<0.0011.20 (1.09–1.31)<0.0011.20 (1.09–1.31)<0.001HRS Unexposed1(Ref)1(Ref)1(Ref)1(Ref) Exposed1.29 (1.17–1.42)<0.0011.28 (1.15–1.42)<0.0011.25 (1.13–1.39)<0.0011.23 (1.11–1.36)<0.001ELSA Unexposed1(Ref)1(Ref)1(Ref)1(Ref) Exposed1.50 (1.25–1.80)<0.0011.61 (1.33–1.94)<0.0011.58 (1.31–1.90)<0.0011.53 (1.27–1.85)<0.001

Association between adulthood SLEs and incident heart disease.

Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards models with attained age as the time scale. Model 1: Adjusted for sex, marital status, and education; Model 2: Model 1 + smoking, drinking, and physical activity; Model 3: Model 2 + hypertension and diabetes. In CHARLS, physical activity includes an explicit “NA” (missing) category.

3.4 Dose–response analysis

In dose–response analyses (Table 3), hazards increased with the number of events in HRS (P for trend < 0.001) and ELSA (P for trend = 0.004). In CHARLS, risk was elevated in the 1- and 2-event categories but did not show a clear further increase in the ≥3 category (P for trend < 0.001).

VariableEvents/total, nCrude modelModel 1Model 2Model 3HR (95% CI)P-valueHR (95% CI)P-valueHR (95% CI)P-valueHR (95% CI)P-valueCHARLS0