Data availability statement

The HRS data used in this study are publicly available at (https://hrs.isr.umich.edu/about). The NHANES data are accessed from the National Health and Nutritional Examination Survey (https://wwwn.cdc.gov/nchs/nhanes/) and NHANES 2019 Public-Use Linked Mortality File (https://www.cdc.gov/nchs/linked-data/mortality-files).

Study population

The HRS is a nationally representative longitudinal survey of approximately 20,000 U.S. adults aged 50 and older, initiated in 1992 and conducted biennially. Details of the multistage area probability sampling method used to recruit a nationally representative sample of noninstitutionalized U.S. adults have been reported previously [15]. For this study, we use the data from the waves 2016, 2018, and 2020 because biomarker data from the Venous Blood Study (VBS) were only available in wave 2016. Participants were eligible for inclusion if they met the following criteria: 1) no self-reported diagnosis of HF or cancer at baseline in 2016; 2) complete data to calculate AL at baseline; 3) complete information on the key covariates, including age, sex, race/ethnicity, education, family income, marital status, smoking status, drinking status, diabetes, and hypertension. A total of 3,814 eligible participants were included at baseline in the HRS cohort. After excluding participants with missing covariate data, 2,951 participants were included in the fully adjusted analyses (Figure S1).

The NHANES is an official representative survey designed to assess the health and nutritional status of the civilian and noninstitutionalized U.S. population. It uses a complex, multistage probability sampling design and is conducted in a 2-year cycle by the official institute, the National Center for Health Statistics (NCHS) [16–18]. For this analysis, we included data from seven continuous NHANES cycles: 1999–2000, 2001–2002, 2003–2004, 2005–2006, 2007–2008, 2009–2010, and 2015–2016. The waves in 2011–2012 and 2013–2014 were deleted due to missing data on C-reactive protein (CRP) for AL calculation. Participants from these cycles were linked to mortality outcomes through the data from the 2019 Linked Mortality File. The inclusion criteria were as follows: 1) age ≥ 25; 2) complete data on AL; 3) with HF; 4) complete data on key covariates, including age, sex, race/ethnicity, education, family income, marital status, smoking status, drinking status, serum creatinine, coronary heart disease, diabetes, and taking antihypertensive medication. Participants with missing figures for mortality status or survival time were excluded. A total of 1,200 individuals with HF were included in the NHANES cohort at baseline. After exclusion of participants with missing covariate data, 568 participants were included in the fully adjusted analyses (Figure S2).

Measurement

HF in the HRS was identified using the self-reported responses to the survey question: “Has a doctor ever told you that you have congestive heart failure?”. In NHANES, HF was determined by the question “Ever told had congestive heart failure?”.

The AL was constructed using the same set of physiological indicators in both the HRS and NHANES cohorts. In NHANES studies and as established in the formal studies on AL, the AL was deprived from ten physiological biomarkers representing three major biological system: CRP for the immune system; systolic and diastolic BP and resting pulse for the cardiovascular system; and body mass index (BMI), total cholesterol (TC), high density lipoprotein cholesterol (HDL-C), glycosylated hemoglobin (HbA1c), albumin, and triglycerides (TG) for the metabolic system [12, 19, 20]. However, TG is a weak indicator of AL in prior studies and was therefore excluded in the present analysis [21–23]. The AL in this study included nine biomarkers: systolic BP, diastolic BP, resting pulse, BMI, TC, HDL-C, HbA1c, albumin, and CRP [24]. Each biomarker was assigned a score of 1 if the value exceeded clinical cut-off thresholds, and 0 otherwise. The total AL score ranged from 0 to 9. Participants were classified into three groups: low (0–2), medium (3), and high AL (≥ 4), based on score distribution and prior studies [23]. The clinical cut-off of AL thresholds was selected based on widely accepted standards from previous literature and applied consistently across both datasets [24]. Specifically, the following cut-off values were used: systolic BP ≥ 140 mmHg, diastolic BP ≥ 90 mmHg, TC ≥ 240 mg/dL, HDL-C < 40 mg/dL, CRP ≥ 0.3 mg/dL, HbA1c ≥ 6.4%, albumin < 3.8 mg/dL, resting pulse ≥ 90 bpm, and BMI ≥ 30 kg/m².

Covariates included age, sex (male, female), race/ethnicity (white, others), education (< 12 years, 12 years, and > 12 years), marital status (married, separate), family income (< $14,999; $15,000–24,999; $25,000–49,999; ≥ $50,000), smoking statue (yes, no), drinking statue (yes, no), serum creatinine, coronary heart disease (yes, no), diabetes (yes, no) and taking antihypertensive medication (yes, no).

Statistical analysis

Continuous variables were expressed as means ± standard deviation or median (interquartile range), while categorical variables were expressed as counts and percentages. Normally distributed continuous variables were compared using independent t-tests or ANOVA, whereas non-normally distributed variables were compared using Mann–Whitney tests or Kruskal-Wallis tests, as appropriate. Categorical variables were compared using Chi-square tests. Logistic regression was used to evaluate the association between AL and the incidence of HF using the HRS data, and the Cox proportional hazards regression model was used to assess the association between AL and the risk of cardiovascular and all-cause mortality using the data from the NHANES and HRS.

Logistic regression results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). The Cox proportional hazards regression results were reported as hazard ratios (HRs) with 95% CIs. For both datasets, three models were constructed: Model 1 was unadjusted; Model 2 adjusted for demographic characteristics (age, sex, education level, marital status, race/ethnicity, and income); Model 3 further adjusted for health behaviors and clinical factors (smoking, drinking, serum creatinine, coronary heart disease, antihypertensive medication, and diabetes). Tests for trend were performed by modeling ordered AL categories as ordinal variables in the regression analyses. Interaction analyses were conducted by including multiplicative interaction terms between AL and subgroup variables in the regression models. To evaluate the incremental predictive value of AL, we constructed a baseline line prediction model based on variables included in the American Heart Association PREVENT equation, including age, sex, systolic BP, TC, HDL-C, BMI, smoking status, diabetes, and estimated glomerular filtration rate (eGFR) [25]. AL was then added to the baseline model to assess its additional predictive contribution. Model discrimination was evaluated using the C-statistic (area under the receiver operating characteristic curve, AUC). Differences in AUCs between models were compared using DeLong’s test with bootstrap resampling. Reclassification performance was further assessed using continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI).

In a sensitivity analysis, participants who died within 2 years of follow-up were excluded. To make full use of all the information in the sample, multiple imputation (MI) methods have been performed in another sensitivity analysis. Although there were no changes to the method and laboratory for AL-related measurements, we conducted another sensitivity analysis to test the potential effects of measurement changes over time by dividing the cohort into three groups (NHANES 1999–2004, NHANES 2005–2010, NHANES 2015–2016). Additional sensitivity analyses were conducted after excluding participants with baseline coronary heart disease to further evaluate the robustness of the observed associations. Furthermore, stratified analysis was performed by selected variables: age (< 65 vs. ≥ 65 years old), sex (male vs. female), race (white vs. others), marital status (married vs. single), taking antihypertensive medication (yes vs. no), and diabetes (yes vs. no). In addition, Kaplan–Meier survival analyses stratified by AL groups were conducted for both CVD and all-cause mortality. We assessed the incremental predictive value of AL by comparing base models (containing demographic, socioeconomic, health behavior, and clinical covariates) to extended models that included AL in both cohorts. Multicollinearity among covariates was assessed using variance inflation factors (VIFs), and no significant multicollinearity was detected (all VIFs < 5). The proportional hazards assumption was evaluated using Schoenfeld residuals. Statistical analyses were performed using R 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria). A two-sided P value of < 0.05 was considered statistically significant.

Association between AL and HF in the general population

About 31.8% of participants were classified in the low AL group. During approximately 4 years of follow-up, a total of 151 incident HF cases (9.93 per 1,000 person-years) were identified in the HRS cohort. The incidences of HF were 2.74%, 4.06%, and 5.99% in participants with low, moderate, and high AL groups, respectively.

Each unit increase in AL score was associated with a 49% higher risk of HF (OR = 1.49; 95% CI: 1.21–1.82, P < 0.001). Compared to the low AL group, participants in the high AL group had a significantly higher risk of HF (OR = 2.17; 95% CI: 1.46–3.21, P < 0.001), and the medium AL group did not show statistically higher risk of HF (OR = 1.44; 95% CI: 0.94–2.21, P = 0.090). After adjusting for other covariates, consistently, each unit increase in AL was associated with a 30% increase in the risk of HF (OR = 1.30; 95% CI: 1.13–1.49, P < 0.001). Compared to the low AL group, a significantly higher risk of HF (OR = 2.07; 95% CI: 1.29–3.32, P = 0.002) remained in the high AL group, but not for the medium AL group (OR = 1.26; 95% CI: 0.75–2.11, P = 0.391) (Table 2).

Association between AL and mortality among patients with HF

Only 39.8% of patients were classified in the low AL group. Over a median follow-up of 5.92 years (Q1–Q3: 3.17–11.08 years), 324 (27%) cardiovascular deaths and 773 (64.4%) all-cause deaths were observed. The cardiovascular mortality rates among HF patients in low, moderate, and high AL groups were 25.65%, 25.28%, and 29.67%; and 61.23%, 64.68% and 67.99% for all-cause mortality rates, respectively.

After adjusting for other covariates, each unit increase in AL score was significantly associated with a 18% higher risk of cardiovascular mortality (HR = 1.18; 95% CI: 1.05–1.31, P = 0.007). Patients with HF in the high AL group had doubled the risk of CVD death compared to the low AL group (HR = 2.03; 95% CI: 1.37–3.03, P < 0.001), while the medium AL group was not significant (HR = 1.30; 95% CI: 0.82–2.07, P = 0.261). After controlling for other covariates, a per-unit increase in AL was associated with a 15% increased risk of all-cause mortality (HR = 1.15; 95% CI: 1.06–1.24, P = 0.001). Compared to the low AL group, the high AL group was significantly associated with higher mortality (HR = 1.70; 95% CI: 1.30–2.22, P < 0.001), and the medium AL group was also significant (HR = 1.35; 95% CI: 1.01–1.79, P = 0.041) (Table 3 and Figure 1).

Display full size

Figure 1. Kaplan–Meier curves for cardiovascular disease (CVD). (A) death and all-cause death (B) by allostatic load (AL) group in the NHANES cohort. Numbers at risk and events are shown below the curves. Hazard ratios (HRs) were estimated using Cox proportional hazards models adjusted for covariates (age, sex, education level, marital status, race/ethnicity, income, smoking, drinking, serum creatinine, coronary heart disease, antihypertensive medication, and diabetes).

Predictive value of AL

In the HRS cohort, adding baseline AL to the HF model modestly improved discrimination for incident HF, with the C-statistic increasing from 0.711 (95% CI: 0.669, 0.753) to 0.720 (95% CI: 0.678, 0.761), yielding a difference of 1.15% (95% CI: –0.73%, 3.03%). The NRI was 0.142 (95% CI: –0.031, 0.316), and the IDI was 0.0016 (95% CI: –0.0022, 0.0053). The extended model incorporating AL showed a higher C-statistic for CVD mortality, increasing from 0.689 (95% CI: 0.654, 0.723) to 0.709 (95% CI: 0.674, 0.743), with a difference of 1.95% (95% CI: 0.36%, 3.89%). For all-cause mortality, the C-statistic rose from 0.689 (95% CI: 0.668, 0.714) to 0.699 (95% CI: 0.679, 0.725), with a difference of 1.00% (95% CI: 0.26%, 2.16%) (Table 4).

Stratified analysis and sensitivity analysis

We conducted stratified analyses to examine whether the association between high AL and adverse outcomes differed by subgroup and showed similar results (Figure 2 and Figure 3). Sensitivity analyses excluding participants who died within 2 years of follow-up (Table S2), imputation of missing data (Table S3 and Table S4), or grouping the NHANES data into three cohorts (1999–2004, 2005–2010, and 2015–2016) showed similar results (Table S5). The subgroup analysis findings should be considered exploratory because of the relatively limited sample size and statistical power. The associations between AL and study outcomes remained generally consistent after excluding participants with baseline coronary heart disease (Table S6).

Display full size

Figure 2. Subgroup analyses of the associations of a unit increase in allostatic load (AL) with incident heart failure in the HRS cohort (2016–2020). ORs and 95% CIs were obtained from logistic regression models with MI. Models were adjusted for all covariates (age, sex, education level, marital status, race/ethnicity, income, smoking, drinking, serum creatinine, coronary heart disease, antihypertensive medication, and diabetes). P values represent within-group effects; P for interaction tests cross-stratum effect modification. OR: odds ratio; HRS: Health and Retirement Study; CI: confidence interval; HF: heart failure.

Display full size

Figure 3. Subgroup analyses of mortality risk per 1-unit increase for cardiovascular disease (CVD). (A) death and all-cause death (B) in allostatic load (AL) among participants with heart failure in the NHANES cohort (1999–2016). HRs and 95% CIs were estimated from Cox proportional hazards models. Models were adjusted for all covariates (age, sex, education level, marital status, race/ethnicity, income, smoking, drinking, serum creatinine, coronary heart disease, antihypertensive medication, and diabetes). P values represent within-stratum associations; P for interaction indicates heterogeneity across subgroups. HR: hazard ratio; CI: confidence interval.

Conclusion

In conclusion, elevated AL is a significant predictor of both HF incidence and subsequent cardiovascular and all-cause mortality. AL captures the cumulative burden of chronic physiological stress and may serve as an integrative biomarker for early identification of at-risk individuals. Future research should test targeted interventions to reduce AL and improve HF prevention and management.