Abstract
Background
Cardiometabolic multimorbidity (CMM) is a major global health burden associated with increased morbidity and mortality. The oxidative balance score (OBS), a composite measure of dietary and lifestyle factors related to oxidative balance, has not been systematically evaluated in relation to CMM and mortality.
Methods
We analyzed 11,365 NHANES participants. Logistic regression, Cox proportional hazards models, and restricted cubic splines were used to evaluate associations of OBS with CMM and mortality. Mediation analysis further evaluated inflammatory and insulin resistance indicators as potential mediators among non-CMM participants. Machine learning identified key OBS components for predicting outcomes.
Results
Among 11,365 participants (median age, 45 years; 52.55% male), higher OBS was associated with more favorable sociodemographic characteristics and a lower prevalence of chronic conditions. OBS was lower in participants with CMM and decreased further with a greater number of comorbidities. CMM was associated with higher mortality, whereas each 5-point increase in OBS was associated with lower odds of CMM (OR = 0.96) and reduced risks of all-cause mortality (HR = 0.85), with similar inverse associations observed for cardiovascular and non-cardiovascular mortality. These associations remained robust among non-CMM individuals but were attenuated among those with CMM. Mediation analysis suggested partial mediation of the association between OBS and mortality through inflammatory and insulin resistance pathways. In machine learning analysis, LightGBM achieved the best predictive performance (AUC: 0.817 and 0.849), with body mass index, physical activity, and vitamin intake identified as key predictors.
Conclusion
Higher OBS was associated with lower CMM prevalence and reduced mortality risk among non-CMM participants, partially through inflammatory and insulin resistance pathways, underscoring its potential relevance in cardiometabolic risk assessment.
Graphical Abstract
Supplementary information
The online version contains supplementary material available at 10.1186/s12967-026-08291-y.
Keywords: Oxidative balance score, Cardiometabolic multimorbidity, Inflammation, Insulin resistance, Machine learning, Mortality
Introduction
With the rapid progression of global population aging, the prevalence of multimorbidity has surged, posing a substantial threat to public health [1]. This growing burden not only elevates the risk of disability and mortality but also strains healthcare resources and diminishes quality of life [2, 3]. Cardiometabolic multimorbidity (CMM), defined as the co-occurrence of at least two cardiometabolic conditions including hypertension, cardiovascular disease (CVD), stroke, or diabetes mellitus (DM), is among the most prevalent and adverse forms of chronic disease clustering [4, 5]. Individuals with CMM have been reported to exhibit a 3.7–6.9-fold higher risk of all-cause mortality and a shortened life expectancy by 12–15 years from age 60, compared to disease-free individuals [4]. Given its substantial health burden, identifying modifiable risk factors for CMM is critical for prevention and risk management.
Oxidative imbalance has been implicated in both cardiovascular and metabolic disorders [6, 7]. To better quantify redox balance at the population level, the oxidative balance score (OBS) integrates prooxidant and antioxidant-related dietary and lifestyle factors into a composite measure of overall oxidative status [8]. Unlike single biomarkers that reflect transient physiological states, OBS captures longer-term exposure patterns, providing a more stable indicator relevant to chronic disease risk [9, 10]. Previous studies have linked lower OBS levels to increased risks of various chronic conditions, including atherosclerosis, metabolic syndrome, chronic kidney disease, and certain malignancies [11]. However, the association between OBS and CMM remains unclear, particularly regarding its prognostic value and the mechanisms that may explain these associations.
To address this gap, we examined the association of OBS with CMM and evaluated its relationship with mortality stratified by CMM status. In addition, inflammatory biomarkers and insulin resistance (IR) indicators were analyzed to assess their mediating roles in the association between OBS and mortality. Beyond conventional analyses, multiple machine learning algorithms were applied to evaluate mortality prediction using OBS and clinical covariates while quantifying the relative contributions of individual OBS components to CMM and mortality.
Method
Study population
We conducted both cross-sectional and prospective analyses using data from nine continuous cycles (2001–2018) of the National Health and Nutrition Examination Survey (NHANES). NHANES is a nationally representative survey of the non-institutionalized U.S. population conducted by the National Center for Health Statistics (NCHS), employing a complex, multistage probability sampling design [12].
A total of 91,351 participants from NHANES were initially considered. We excluded participants based on the following criteria: (1) age < 20 years (n = 41,150); (2) missing information necessary to define CMM (n = 30,410); (3) incomplete data for calculating the OBS (n = 7,427); (4) missing survival status or follow-up information (n = 33); and (5) missing key covariates (n = 966). Following the application of these exclusion criteria, 11,365 participants remained for analysis, representing approximately 137.6 million non-institutionalized U.S. adults. Among them, 8,792 were classified as non-CMM and 2,573 as CMM participants (Figure S1).
OBS and CMM
The OBS integrates prooxidant and antioxidant-related dietary and lifestyle factors based on prior evidence suggesting their relevance to oxidative stress. In this study, we adopted the scoring methodology proposed by Zhang et al., incorporating a total of 20 variables, including 16 dietary nutrients and 4 lifestyle components [10]. Dietary nutrients were calculated as the average of two 24-hour dietary recall interviews, while lifestyle factors were assessed through questionnaires and laboratory measurements. The intraclass correlation coefficient for the dietary nutrient component was 0.546, indicating moderate reproducibility between the two recalls. The final OBS was derived by summing the scores of individual components based on the scoring criteria shown in Table S1 and was categorized into quartiles. Correlations among OBS components are illustrated in Figure S2, and principal component analysis was further conducted to assess potential redundancy among components (Figure S3). The distribution of antioxidant and prooxidant components across OBS quartiles is presented in Table S2.
CMM was defined as the coexistence of two or more chronic cardiometabolic conditions, including CVD (congestive heart failure, heart attack, coronary heart disease, or angina), stroke, DM (including prediabetes), and hypertension [4]. The presence of CVD and stroke was determined based on self-reported physician diagnoses. DM was defined as self-reported diagnosis, use of antidiabetic medications, fasting plasma glucose (FPG) ≥126 mg/dL, or hemoglobin A1c (HbA1c) ≥6.5%, while prediabetes was defined as FPG 100–125 mg/dL or HbA1c 5.7–6.4%. Hypertension was identified by a self-reported diagnosis, current use of antihypertensive medication, or measured average systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg based on three seated readings.
Potential confounders
Potential confounders included demographic characteristics, lifestyle factors, and comorbidities. Age was recorded in years, and sex was categorized as male or female. Race was classified into non-Hispanic White, Mexican American, non-Hispanic Black, and other races. Education attainment was grouped into <high school, high school, and ≥college. Income was assessed using the poverty income ratio (PIR) and categorized as low income (PIR < 1.3), middle income (1.3 ≤ PIR <3.5), and high income (PIR ≥ 3.5). Marriage was divided into never married, married or living together, and separated, divorced, or widowed. Hyperlipidemia was defined as meeting any of the following criteria: total cholesterol > 200 mg/dL, low-density lipoprotein cholesterol (LDL-C) >130 mg/dL, triglycerides > 150 mg/dL, high-density lipoprotein cholesterol (HDL-C) <40 mg/dL in men or <50 mg/dL in women, or current use of antihyperlipidemic medications. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [13]. White blood cell (WBC) was also included as a potential covariate.
Furthermore, to explore potential mediation pathways, we included systemic inflammatory indexes as mediators, including the neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammation index (SII), systemic inflammation response index (SIRI), and aggregate index of systemic inflammation (AISI). In addition, indicators of IR indicators were also incorporated as mediators, including the triglyceride-glucose (TyG), TyG–body mass index (TyG-BMI), and TyG–waist circumference (TyG-WC) index. The formulas for calculating these indices are as follows:
Inflammatory indexes:
 |
 |
 |
 |
IR indicators:
 |
 |
 |
L, Lymphocytes; M, Monocytes; N, Neutrophils; PLT, Platelets; BMI, body mass index; WC, waist circumference.
Outcome measures
The primary outcome of this study was all-cause mortality. Secondary outcomes included cardiovascular or non-cardiovascular mortality. Mortality status and causes of death were ascertained via linkage to the National Death Index (NDI), maintained by NCHS. Causes of death were classified using the 10th Revision of the International Classification of Diseases (ICD-10). Follow-up time, calculated in months, was defined as the interval from the date of the NHANES interview to the date of death or the end of follow-up (December 31, 2019), whichever occurred first.
Statistical analysis
The complex sampling design of NHANES was accounted for by incorporating sample weights, clustering, and stratification to obtain nationally representative estimates. Categorical variables were presented as weighted percentages, and continuous variables were expressed as weighted medians with interquartile ranges. Group differences were assessed using survey-weighted chi-square tests for categorical variables, the Wilcoxon rank-sum test for two-group comparisons of continuous variables, and the Kruskal-Wallis test for comparisons involving more than two groups. Standardized mean differences (SMDs) were also calculated to quantify the magnitude of between-group differences.
To assess the association between OBS and CMM, multivariable logistic regression models were constructed. Multivariable Cox models were used to assess the association of CMM status with mortality and to evaluate the associations between OBS and mortality in the overall cohort and stratified by CMM status. For both logistic and Cox models, three progressively adjusted models were applied: Model 1 was unadjusted; Model 2 was adjusted for age and sex; and Model 3 was further adjusted for race, marriage, income, education, hyperlipidemia, eGFR, and WBC. The proportional hazards assumption was evaluated using Schoenfeld residuals, and no evidence of violation was observed (Table S3). For survival analyses, Kaplan–Meier survival curves were plotted across OBS quartiles, and the log-rank test was used to compare survival differences between groups. A restricted cubic spline (RCS) model was further applied to explore the potential nonlinear association between continuous OBS scores and mortality risk, with knots placed at the 10th, 50th, and 90th percentiles of the OBS distribution. The RCS analyses were adjusted for the same set of covariates as Model 3.
To further examine the association between OBS and mortality, subgroup analyses were conducted according to CMM status, as well as the number and specific types of CMM conditions. In the non-CMM participants, analyses were stratified by sex, age, race, education, marital status, and income. Sensitivity analyses included alternative OBS definitions (Day 1, Day 2, Dietary nutrients-only, and Lifestyle components-only scores) and a leave-one-component-out approach to evaluate the influence of individual components on the associations. Additional analyses were primarily conducted in the non-CMM participants and included excluding accidental mortality, excluding participants with baseline CVD, further adjusting for medication use, and conducting separate analyses in individuals without chronic diseases or with a single cardiometabolic condition. Additional analyses restricting the sample to the 2001–2012 survey cycles and excluding deaths within the first 24 months of follow-up were conducted in both the non-CMM and overall participants.
Mediation analysis was conducted to evaluate whether inflammatory markers and IR indicators mediated the association between OBS and mortality. Indirect effects were estimated using 1,000 bootstrap resamples. All mediation models were adjusted for the same covariates as Model 3.
Several machine learning algorithms, including LightGBM, XGBoost, Random Forest, and Logistic Regression, were used to model the association between OBS and CMM in the overall population and to predict all-cause mortality among non-CMM participants. The dataset was randomly split into a training set (67%) and a testing set (33%). Predictors included age, sex, and the components of the OBS. Multicollinearity was assessed using variance inflation factors (VIF). Hyperparameters were optimized using random search with ten-fold cross-validation, and the optimal parameter set was selected based on cross-validated AUC. Model performance was evaluated in the testing set using ROC curves and multiple metrics. The best-performing model was selected and compared with the conventional logistic regression model. SHapley Additive exPlanations (SHAP) values were then used to interpret the relative importance of OBS components in the selected model.
All statistical analyses were performed using R (version 4.4.1), and figures were generated using R (version 4.4.1) and GraphPad Prism (version 9.0). A two-sided p-value < 0.05 was considered statistically significant.
Result
Among 11,365 participants, the median age was 45 years (33–58), with 5,875 males (50.81%), 5,674 non-Hispanic Whites (72.69%), 1,716 Mexican Americans (6.94%), 2,073 non-Hispanic Blacks (9.63%), and 1,902 other races (10.74%). Stratification by OBS showed that higher OBS groups were associated with higher proportions of greater educational attainment, more participants who were married or living together, and higher income. Additionally, the prevalence of hyperlipidemia, DM, hypertension, CVD, and stroke progressively decreased as OBS increased (all p < 0.05). SMD suggested varying degrees of imbalance across OBS quartiles, with larger differences observed for smoking, education, and income (Table 1). When stratified by the presence of CMM, CMM participants exhibited significantly lower OBS. Moreover, OBS declined progressively with an increasing number of comorbidities. Regarding OBS components, CMM participants had lower levels of vitamin B12, niacin, total folate, calcium, magnesium, copper, total fat, physical activity, alcohol, and cotinine, alongside higher BMI, and these patterns were consistent regardless of the number of comorbidities (all p < 0.05) (Table S4).
Table 1.
Baseline characteristics of all participants categorized by oxidative balance score quartiles
| Variable | Total (n = 11365) |
OBS Q1 (<15) (n = 2485) |
OBS Q2 (15 to <21) (n = 2868) |
OBS Q3 (21 to <26) (n = 2693) |
OBS Q4 (≥26) (n = 3319) |
SMD (Q4 vs. Q1) |
P |
|---|---|---|---|---|---|---|---|
| Age, Year | 45 (33 - 58) | 44 (31 - 57) | 45 (33 - 59) | 45 (32 - 57) | 46 (34 - 58) | 0.05 | 0.053 |
| Sex, Male | 5875 (50.81) | 1359 (52.55) | 1494 (52.40) | 1377 (49.32) | 1645 (49.67) | 0.06 | 0.153 |
| Race | <0.001 | ||||||
| Non-Hispanic White | 5674 (72.69) | 1142 (67.56) | 1389 (71.51) | 1351 (72.30) | 1792 (76.99) | 0.21 | |
| Mexican-American | 1716 (6.94) | 347 (6.78) | 432 (6.45) | 415 (7.20) | 522 (7.19) | 0.02 | |
| Non-Hispanic Black | 2073 (9.63) | 649 (15.26) | 572 (11.05) | 455 (8.90) | 397 (5.70) | 0.32 | |
| Other races | 1902 (10.74) | 347 (10.40) | 475 (10.99) | 472 (11.60) | 608 (10.12) | 0.01 | |
| Smoking | <0.001 | ||||||
| Never | 6180 (54.37) | 1102 (43.31) | 1514 (51.82) | 1496 (55.88) | 2068 (61.90) | 0.38 | |
| Former | 2956 (25.86) | 613 (23.40) | 747 (25.68) | 719 (26.06) | 877 (27.34) | 0.09 | |
| Now | 2229 (19.76) | 770 (33.29) | 607 (22.51) | 478 (18.06) | 374 (10.76) | 0.56 | |
| Drinking | <0.001 | ||||||
| Never | 1214 (8.88) | 281 (10.07) | 304 (8.90) | 270 (9.18) | 359 (9.37) | 0.02 | |
| Former | 1682 (12.32) | 441 (16.91) | 451 (13.05) | 361 (11.88) | 429 (11.37) | 0.16 | |
| Now | 7803 (73.85) | 1598 (73.02) | 1953 (78.05) | 1896 (78.94) | 2356 (79.26) | 0.15 | |
| Education | <0.001 | ||||||
| <High school | 2201 (12.60) | 673 (19.64) | 613 (13.66) | 459 (11.29) | 456 (8.50) | 0.32 | |
| High school | 2574 (22.77) | 699 (30.82) | 711 (25.92) | 610 (22.68) | 554 (15.56) | 0.37 | |
| ≥College | 6590 (64.63) | 1113 (49.54) | 1544 (60.42) | 1624 (66.04) | 2309 (75.94) | 0.57 | |
| Marriage | <0.001 | ||||||
| Never | 2011 (17.52) | 493 (20.22) | 487 (17.12) | 470 (17.81) | 561 (15.95) | 0.11 | |
| Married or living together | 7198 (66.48) | 1455 (60.48) | 1786 (65.16) | 1723 (66.78) | 2234 (70.92) | 0.22 | |
| Separated, divorced, or widowed | 2156 (16.00) | 537 (19.30) | 595 (17.73) | 500 (15.41) | 524 (13.13) | 0.17 | |
| Income | <0.001 | ||||||
| Low income | 3002 (17.92) | 870 (27.16) | 782 (18.67) | 636 (15.75) | 714 (13.39) | 0.35 | |
| Middle income | 4335 (36.05) | 1006 (40.30) | 1170 (38.46) | 1022 (36.59) | 1137 (31.24) | 0.19 | |
| High income | 4028 (46.02) | 609 (32.54) | 916 (42.86) | 1035 (47.67) | 1468 (55.37) | 0.47 | |
| eGFR, mL/min/1.73 m2 | 96.81 (82.36 - 110.27) | 97.89 (82.59 - 112.35) | 95.81 (80.82 - 109.78) | 97.85 (82.86 - 109.89) | 96.05 (82.88 - 109.48) | 0.05 | 0.037 |
| WBC, 109/L | 6.40 (5.40 -7.70) | 6.80 (5.60 - 8.30) | 6.50 (5.50 - 7.80) | 6.40 (5.40 - 7.80) | 6.10 (5.10 - 7.20) | 0.21 | <0.001 |
| NLR | 1.93 (1.47 - 2.53) | 1.95 (1.47 - 2.63) | 1.94 (1.48 - 2.56) | 1.93 (1.47 - 2.50) | 1.89 (1.47 - 2.47) | 0.05 | 0.128 |
| SII | 462.00 (337.78 - 648.38) | 487.20 (348.25 - 699.21) | 468.46 (345.00 - 653.44) | 462.68 (338.00 - 643.38) | 441.60 (327.86 - 615.57) | 0.10 | <0.001 |
| SIRI | 0.98 (0.68 - 1.40) | 1.04 (0.70 - 1.52) | 1.00 (0.69 - 1.44) | 0.97 (0.67 - 1.39) | 0.92 (0.65 - 1.32) | 0.10 | <0.001 |
| AISI | 233.70 (155.03 - 360.06) | 261.51 (166.25 - 405.60) | 239.95 (162.21 - 371.74) | 233.80 (154.38 - 355.73) | 217.78 (145.88 - 331.27) | 0.13 | <0.001 |
| Tyg index | 8.53 (8.12 - 8.98) | 8.62 (8.20 - 9.05) | 8.60 (8.16 - 9.02) | 8.51 (8.13 - 9.00) | 8.42 (8.03 - 8.89) | 0.15 | <0.001 |
| Tyg-BMI index | 236.93 (199.07 - 282.03) | 250.85 (214.74 - 294.69) | 245.36 (207.18 - 289.17) | 236.78 (199.98 - 283.68) | 221.96 (189.41 - 263.39) | 0.22 | <0.001 |
| Tyg-WC index | 835.04 (715.49 - 953.46) | 868.00 (753.32 - 975.90) | 858.58 (735.81 - 980.23) | 835.42 (718.54 - 955.91) | 794.04 (683.34 - 909.61) | 0.20 | <0.001 |
| Hyperlipidemia | 8128 (70.22) | 1892 (75.17) | 2090 (71.68) | 1934 (71.05) | 2212 (65.47) | 0.21 | <0.001 |
| Anti-hypertension medication | 3256 (24.67) | 788 (26.33) | 889 (27.62) | 755 (24.33) | 824 (21.68) | 0.11 | <0.001 |
| Anti-hyperlipidemia medication | 2024 (16.23) | 454 (16.14) | 555 (17.55) | 482 (15.66) | 533 (15.72) | 0.01 | 0.415 |
| Anti-DM medication | 1023 (6.76) | 276 (8.36) | 273 (7.44) | 242 (6.24) | 232 (5.65) | 0.11 | 0.005 |
| DM, Including prediabetes | 3601 (26.51) | 861 (28.49) | 968 (28.78) | 861 (26.87) | 911 (23.33) | 0.12 | <0.001 |
| Hypertension | 4460 (34.62) | 1129 (39.05) | 1186 (37.61) | 1004 (33.86) | 1141 (30.25) | 0.19 | <0.001 |
| CVD | 825 (5.74) | 245 (7.41) | 228 (6.66) | 179 (4.90) | 173 (4.66) | 0.12 | 0.001 |
| Stroke | 313 (2.14) | 99 (2.77) | 89 (2.71) | 70 (2.00) | 55 (1.42) | 0.09 | 0.008 |
| CMM | 2573 (17.98) | 677 (20.45) | 688 (19.85) | 588 (17.98) | 620 (15.07) | 0.14 | <0.001 |
| CMM counts | <0.001 | ||||||
| Non-CMM | 8792 (82.02) | 1808 (79.55) | 2180 (80.15) | 2105 (82.02) | 2699 (84.93) | 0.14 | |
| Two CMMs | 2048 (14.50) | 513 (15.95) | 536 (15.61) | 480 (15.27) | 519 (12.20) | 0.11 | |
| Three CMMs | 465 (3.12) | 140 (3.99) | 140 (3.79) | 93 (2.48) | 92 (2.58) | 0.08 | |
| Four CMMs | 60 (0.36) | 24 (0.51) | 12 (0.45) | 15 (0.23) | 9 (0.30) | 0.03 |
Continuous variables expressed as median (interquartile range); categorical variables expressed as n (weighted %). SMD values < 0.1 were considered to be indicative of a good balance
Abbreviations: AISI, Aggregate index of systemic inflammation; CMM, Cardiometabolic multimorbidity; CVD, Cardiovascular disease; DM, Diabetes mellitus; eGFR, Estimated glomerular filtration rate; NLR, Neutrophil-to-lymphocyte ratio; OBS, Oxidative balance score; PIR, Poverty income ratio; SII, Systemic immune-inflammation index; SIRI, Systemic inflammation response index; SMD, Standardized mean difference; WBC, White blood cell
During a median follow-up of 112 (60–167) months, a total of 1,228 all-cause deaths occurred, including 372 cardiovascular deaths and 856 non-cardiovascular deaths. Cox proportional hazards models indicated that CMM participants had significantly higher mortality risk, which remained robust after adjustment for covariates (Table S5). Logistic regression analyses in the overall population demonstrated that higher OBS was independently associated with a significantly lower likelihood of CMM. Specifically, in the unadjusted model, each 5-point increment in OBS was associated with a 9% reduction in the odds of CMM, which slightly attenuated but remained significant after full adjustment. When OBS was categorized into quartiles, a decreasing trend in CMM risk was observed. Compared with Q1, Q2 and Q3 showed no significant differences in the fully adjusted model, while participants in Q4 exhibited a 12% lower risk of CMM (OR = 0.88, 95% CI: 0.78–0.99, p = 0.031). Although the effect estimates attenuated after covariate adjustment, the inverse association between higher OBS and CMM remained statistically significant (Table S6).
In the overall study population, Kaplan–Meier survival analyses demonstrated significant differences in survival across OBS groups, with participants in the Q4 exhibiting the greatest survival probability (all Log-rank p < 0.05) (Figure S4). RCS analyses further confirmed significant dose-dependent inverse associations between OBS and the risks of all-cause, cardiovascular, and non-cardiovascular mortality in all participants (P for overall < 0.05, P for nonlinear > 0.05) (Figs. 1A–C). Consistently, Cox regression analyses in the overall population revealed that higher OBS was associated with a lower risk of mortality. In the unadjusted model, each 5-point increase in OBS was associated with an 18–19% reduction in the risk of all-cause, cardiovascular, and non-cardiovascular mortality (all p < 0.001). After full adjustment, these associations remained significant, with each 5-point increase in OBS was associated with lower risks of all-cause (HR = 0.85, 95% CI: 0.81–0.89, p < 0.001), cardiovascular (HR = 0.85, 95% CI: 0.77–0.95, p = 0.003), and non-cardiovascular (HR = 0.85, 95% CI: 0.81–0.90, p < 0.001) mortality (Table 2). Similarly, when OBS was categorized into quartiles, mortality risk decreased progressively with higher OBS levels. Compared with Q1, participants in Q4 had substantially lower mortality risks in the unadjusted models (all p < 0.001). After full adjustment, these associations were attenuated but remained significant, with Q4 showing a 44% lower risk of all-cause mortality (HR = 0.56, 95% CI: 0.47–0.67), 45% lower cardiovascular mortality (HR = 0.55, 95% CI: 0.39–0.78), and 43% lower non-cardiovascular mortality (HR = 0.57, 95% CI: 0.45–0.72) (all p < 0.001).
Fig. 1.
Restricted cubic spline analysis of the associations between OBS and the risk of all-cause, cardiovascular, and non-cardiovascular mortality among different participant groups. (A–C) all participants; (D–F) non-CMM participants; (G–I) CMM participants. The model adjusted for sex, age, race, marriage, income, education, hyperlipidemia, eGFR, and WBC. Abbreviations: CI, Confidence interval; CMM, cardiometabolic multimorbidity; DM, diabetes mellitus; eGFR, Estimated glomerular filtration rate; HR, Hazard ratio; OBS, oxidative balance score; WBC, White blood cell
Table 2.
Association of OBS with all-cause, cardiovascular, and non-cardiovascular mortality risk in all participants
| Variables | n (%) | Model 1 | Model 2 | Model 3 | |||
|---|---|---|---|---|---|---|---|
| HR (95% CI) | P | HR (95% CI) | P | HR (95% CI) | P | ||
| All-cause mortality | |||||||
| OBS (per 5 points) | 1228 (7.58) | 0.82 (0.78 - 0.86) | <0.001 | 0.81 (0.77 - 0.85) | <0.001 | 0.85 (0.81 - 0.89) | <0.001 |
| Q1 | 374 (11.59) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 349 (8.89) | 0.83 (0.70 - 0.97) | 0.021 | 0.75 (0.64 - 0.88) | <0.001 | 0.82 (0.70 - 0.96) | 0.015 |
| Q3 | 265 (6.73) | 0.64 (0.52 - 0.78) | <0.001 | 0.62 (0.51 - 0.74) | <0.001 | 0.72 (0.60 - 0.87) | <0.001 |
| Q4 | 240 (4.79) | 0.48 (0.39 - 0.58) | <0.001 | 0.46 (0.38 - 0.55) | <0.001 | 0.56 (0.47 - 0.67) | <0.001 |
| Cardiovascular mortality | |||||||
| OBS (per 5 points) | 372 (2.13) | 0.81 (0.74 - 0.90) | <0.001 | 0.81 (0.73 - 0.89) | <0.001 | 0.85 (0.77 - 0.95) | 0.003 |
| Q1 | 125 (3.57) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 97 (2.31) | 0.70 (0.53 - 0.94) | 0.019 | 0.63 (0.48 - 0.84) | 0.001 | 0.68 (0.51 - 0.91) | 0.008 |
| Q3 | 78 (1.75) | 0.54 (0.38 - 0.76) | <0.001 | 0.53 (0.38 - 0.74) | <0.001 | 0.61 (0.43 - 0.86) | 0.005 |
| Q4 | 72 (1.40) | 0.46 (0.33 - 0.65) | <0.001 | 0.46 (0.33 - 0.64) | <0.001 | 0.55 (0.39 - 0.78) | <0.001 |
| Non-cardiovascular mortality | |||||||
| OBS (per 5 points) | 856 (5.45) | 0.82 (0.77 - 0.87) | <0.001 | 0.81 (0.76 - 0.85) | <0.001 | 0.85 (0.81 - 0.90) | <0.001 |
| Q1 | 249 (8.08) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 252 (6.58) | 0.88 (0.71 - 1.09) | 0.250 | 0.81 (0.65 - 1.00) | 0.048 | 0.88 (0.71 - 1.10) | 0.271 |
| Q3 | 187 (4.99) | 0.68 (0.54 - 0.86) | 0.001 | 0.65 (0.53 - 0.81) | <0.001 | 0.77 (0.62 - 0.95) | 0.015 |
| Q4 | 168 (3.39) | 0.48 (0.38 - 0.62) | <0.001 | 0.46 (0.37 - 0.58) | <0.001 | 0.57 (0.45 - 0.72) | <0.001 |
Model 1 was unadjusted
Model 2 adjusted for sex and age
Model 3 adjusted for sex, age, race, marriage, income, education, hyperlipidemia, eGFR, and WBC
Abbreviations: CI, Confidence interval; CMM, Cardiometabolic multimorbidity; DM, Diabetes mellitus; eGFR, Estimated glomerular filtration rate; HR, Hazard ratio; OBS, Oxidative balance score; WBC, White blood cell
Given the strong inverse association observed in the overall population, we further examined whether this relationship persisted among participants with and without CMM. Among participants with CMM, Kaplan–Meier curves showed significantly different survival distributions for all-cause and non-cardiovascular mortality across OBS groups. However, no significant difference in cardiovascular mortality was observed (Log-rank p > 0.05) (Figure S5). In the non-CMM group, higher OBS levels were associated with better survival across all mortality endpoints (all Log-rank p < 0.05) (Figure S6). RCS analyses indicated no significant association between OBS and mortality outcomes in CMM participants (P for overall > 0.05), but significant dose-dependent inverse associations in non-CMM participants (P for overall < 0.05, P for nonlinear > 0.05) (Figs. 1D–I). Among non-CMM participants, higher OBS was consistently associated with lower risks of all-cause, cardiovascular, and non-cardiovascular mortality across both unadjusted and fully adjusted models (Table 3). In Model 1, each 5-point increase in OBS was associated with mortality risk (all p < 0.001) and these associations remained robust in Model 3 (HRs = 0.82, 0.82, and 0.82; all p < 0.001). Compared with Q1, participants in Q4 exhibited substantially reduced risks in Model 1 (HR = 0.49, 0.46, and 0.50, respectively; all p < 0.001), which remained significant after full adjustment (HR = 0.49, 0.47, and 0.51; all p < 0.001). In contrast, among CMM participants, the inverse associations between OBS and mortality were weaker and largely attenuated after full adjustment (Table S7). Significant effect modification by CMM status was observed in continuous analyses (Table S8), whereas interaction tests based on OBS quartiles were less consistent, likely reflecting reduced statistical power.
Table 3.
Association of OBS with all-cause, cardiovascular, and non-cardiovascular mortality risk in non-CMM participants
| Variables | n (%) | Model 1 | Model 2 | Model 3 | |||
|---|---|---|---|---|---|---|---|
| HR (95% CI) | P | HR (95% CI) | P | HR (95% CI) | P | ||
| All-cause mortality | |||||||
| OBS (per 5 points) | 634 (5.11) | 0.82 (0.76 - 0.88) | <0.001 | 0.78 (0.73 - 0.83) | <0.001 | 0.82 (0.76 - 0.87) | <0.001 |
| Q1 | 185 (8.12) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 186 (6.01) | 0.80 (0.65 - 0.99) | 0.042 | 0.71 (0.57 - 0.89) | 0.003 | 0.76 (0.61 - 0.96) | 0.022 |
| Q3 | 129 (4.39) | 0.60 (0.45 - 0.80) | <0.001 | 0.53 (0.41 - 0.70) | <0.001 | 0.61 (0.47 - 0.80) | <0.001 |
| Q4 | 134 (3.29) | 0.49 (0.38 - 0.62) | <0.001 | 0.42 (0.33 - 0.54) | <0.001 | 0.49 (0.38 - 0.64) | <0.001 |
| Cardiovascular mortality | |||||||
| OBS (per 5 points) | 161 (1.16) | 0.82 (0.69 - 0.97) | 0.020 | 0.79 (0.66 - 0.93) | 0.006 | 0.82 (0.76 - 0.92) | <0.001 |
| Q1 | 57 (2.15) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 43 (1.17) | 0.60 (0.38 - 0.94) | 0.025 | 0.52 (0.34 - 0.80) | 0.003 | 0.59 (0.38 - 0.90) | 0.015 |
| Q3 | 30 (0.88) | 0.46 (0.27 - 0.77) | 0.003 | 0.41 (0.26 - 0.66) | <0.001 | 0.49 (0.30 - 0.80) | 0.004 |
| Q4 | 31 (0.79) | 0.46 (0.25 - 0.84) | 0.012 | 0.40 (0.23 - 0.72) | 0.002 | 0.47 (0.26 - 0.86) | 0.013 |
| Non-cardiovascular mortality | |||||||
| OBS (per 5 points) | 473 (3.95) | 0.82 (0.76 - 0.88) | <0.001 | 0.78 (0.72 - 0.84) | <0.001 | 0.82 (0.75 - 0.89) | <0.001 |
| Q1 | 128 (5.97) | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Q2 | 143 (4.84) | 0.87 (0.67 - 1.15) | 0.335 | 0.78 (0.58 - 1.05) | 0.098 | 0.84 (0.61 - 1.14) | 0.250 |
| Q3 | 99 (3.51) | 0.65 (0.46 - 0.91) | 0.012 | 0.58 (0.42 - 0.80) | 0.001 | 0.66 (0.48 - 0.92) | 0.015 |
| Q4 | 103 (2.50) | 0.50 (0.37 - 0.67) | <0.001 | 0.43 (0.31 - 0.58) | <0.001 | 0.51 (0.36 - 0.71) | <0.001 |
Model 1 was unadjusted
Model 2 adjusted for sex and age
Model 3 adjusted for sex, age, race, marriage, income, education, hyperlipidemia, eGFR, and WBC
Abbreviations: CI, Confidence interval; CMM, Cardiometabolic multimorbidity; DM, Diabetes mellitus; eGFR, Estimated glomerular filtration rate; HR, Hazard ratio; OBS, Oxidative balance score; WBC, White blood cell
Subgroup analyses revealed significant interactions between OBS and all-cause and non-cardiovascular mortality risks by education, and between OBS and cardiovascular mortality risk by race (Tables S9–S11), underscoring the importance of considering sociodemographic factors. No significant interaction was observed according to CMM number or type across mortality outcomes (all P for interaction > 0.05) (Table S12). Sensitivity analyses were conducted to assess the robustness of the findings. Across multiple alternative definitions and component exclusions, the inverse association between OBS and mortality remained robust and consistent (Table S13). Similarly, the associations remained largely consistent in direction and magnitude across additional sensitivity analyses, including restriction to the 2001 to 2012 cycles, exclusion of events within the first 24 months of follow-up, exclusion of participants with baseline CVD, further adjustment for medication use, and stratification of non-CMM participants by chronic disease status (Tables S14–S22).
OBS showed modest inverse correlations with these inflammatory and IR markers, supporting their consideration as potential mediators (Figure S7). Given the established links, we further explored potential mediation pathways. Mediation analyses in non-CMM participants indicated that the associations between OBS and risks of all-cause, cardiovascular, and non-cardiovascular mortality were partially mediated through both inflammatory and IR pathways. Specifically, the indirect effect of OBS through AISI accounted for 8.86%, 5.78%, and 10.31% of the total effects for the respective outcomes (Fig. 2). Comparable mediation proportions were observed for the other inflammatory markers (Figure S8A-C). In addition, several IR indicators also mediated a modest but significant proportion of the association between OBS and mortality outcomes, supporting a partial mediating role of IR in these relationships (Figure S8D-F).
Fig. 2.
Mediation analysis of AISI in the associations between OBS and mortality outcomes among non-CMM participants. The model adjusted for sex, age, race, marriage, income, education, hyperlipidemia, and eGFR. Abbreviation: AISI, Aggregate index of systemic inflammation; CMM, cardiometabolic multimorbidity; eGFR, Estimated glomerular filtration rate; OBS, oxidative balance score
To assess multicollinearity among variables, VIFs were calculated for all features, with all values below 8, indicating no significant multicollinearity (Table S23). Four ML models incorporating OBS components and clinical features (sex and age) were trained and evaluated on a testing set using ROC curve analysis. The final hyperparameters adopted for each model are listed in Table S24. Among these, the LightGBM model demonstrated the best performance for CMM, with an AUC of 0.817, F1 score of 0.457, and recall of 0.384 (Fig. 3A, Table S25). For predicting all-cause mortality risk in the non-CMM population, the LightGBM model also performed well, achieving an AUC of 0.849, F1 score of 0.372, and recall of 0.254 (Fig. 3B, Table S26). However, when compared with the fully adjusted logistic regression models, the improvements in AUC achieved by LightGBM were modest and not statistically significant (Tables S27–S28).
Fig. 3.
Performance evaluation and SHAP-based feature interpretation of machine learning models predicting CMM in all participants and all-cause mortality in non-CMM populations. (A) ROC curves for model in predicting CMM in all participants; (B) ROC curves for model in predicting all-cause mortality in non-CMM participants; (C) SHAP summary plot for model in predicting CMM in all participants; (D) SHAP summary plot for model in predicting all-cause mortality in non-CMM participants. Abbreviations: ATE: Alpha-tocopherol equivalent; AUC, Area under curve; CMM, cardiometabolic multimorbidity; ROC, Receiver operating characteristic; SHAP, SHapley additive exPlanations
SHAP value analysis of OBS components in the LightGBM models highlighted the contributions of modifiable exposures to predicted CMM and mortality outcomes. In the CMM risk model, BMI and niacin intake were the components contributing most strongly to higher predicted CMM risk, while higher levels of physical activity contributed negatively, indicating lower predicted risk (Fig. 3C). In the mortality risk model for non-CMM participants, greater intake of vitamin E and vitamin B2, along with higher physical activity, contributed to lower predicted mortality risk, emphasizing the potential contribution of modifiable antioxidant intake and lifestyle behaviors to the observed risk patterns (Fig. 3D). It is important to note that these SHAP-based contributions reflect the associations learned by the model and do not imply causal effects of individual components on actual disease or mortality risk.
Discussion
This study comprehensively examined the associations of OBS with CMM and mortality outcomes, and further explored the potential mediating roles of inflammation and IR in the association between OBS and mortality. Higher OBS was associated with lower odds of CMM and lower mortality, with stronger associations observed among non-CMM participants. RCS analyses indicated a linear inverse relationship between OBS and mortality risk. Mediation analysis suggested partial mediation by inflammation and IR. Additionally, a LightGBM model showed good predictive performance for CMM onset and all-cause mortality in non-CMM participants, and SHAP value-based variable importance ranking identified BMI, vitamin intake, and physical activity as key factors associated with CMM development and mortality risk. Overall, these findings highlight OBS as a composite indicator reflecting dietary and lifestyle exposures related to antioxidant–prooxidant balance and underscore the potential importance of these modifiable factors in CMM and mortality.
A growing body of evidence suggests that higher OBS, reflecting a more favorable balance between antioxidant and prooxidant exposures, is associated with lower risks of several chronic diseases, including diabetes, hypertension, cardiovascular disease, and cancer [11, 14–16]. However, the potential role of OBS in CMM, a condition characterized by the coexistence of multiple cardiometabolic diseases, has not been systematically evaluated. In this study, higher OBS was associated with lower odds of CMM and reduced mortality risk, further supporting the potential role of dietary and lifestyle-related oxidative exposures in cardiometabolic health. Although the association with CMM was modest, CMM is a multifactorial condition influenced by numerous metabolic, behavioral, and clinical determinants, and OBS likely represents only one component of this complex disease process. Nevertheless, as a composite indicator reflecting modifiable dietary and lifestyle exposures, OBS may still provide useful information for identifying individuals with more favorable oxidative profiles. The stronger associations observed for mortality may reflect the cumulative impact of long-term oxidative and metabolic imbalance on overall health outcomes.
One important finding of this study is that the association between OBS and mortality differed by CMM status, with a stronger inverse association observed among individuals without CMM and an attenuated association among those with CMM. This pattern suggests that modifiable oxidative exposures may play a more prominent role during the earlier stages of cardiometabolic disease development. Once multiple metabolic conditions coexist and form a CMM state, the biological processes influencing health outcomes become increasingly complex [17]. Multiple pathological pathways may interact simultaneously, and a single composite indicator such as OBS may therefore be insufficient to fully capture this complexity [18, 19]. In contrast, before the onset of multimorbidity, OBS may more sensitively reflect oxidative exposure patterns and provide greater value for early risk identification and prevention. These findings may have important public health implications. Identifying individuals with lower OBS may help inform early lifestyle interventions, such as improving antioxidant-rich dietary intake, promoting regular physical activity, maintaining a healthy body weight, and reducing pro-oxidant behaviors, which may contribute to lowering the risk of CMM and related mortality [20–22]. In addition, stratifying populations according to disease stage may help account for population heterogeneity and support the development of more targeted prevention strategies [23]. For individuals at elevated risk but without established CMM, early interventions aimed at improving oxidative balance may provide greater long-term health benefits [24].
Several biological mechanisms may help explain the associations observed in this study. Oxidative stress plays a key role in the development of cardiometabolic diseases by promoting chronic inflammation, metabolic dysfunction, and IR [7, 25, 26]. Together, these processes contribute to disease initiation and progression. OBS is a composite index reflecting oxidative balance that integrates multiple dietary and lifestyle exposures with both prooxidant and antioxidant properties [9, 10]. As such, it captures the overall oxidative exposure pattern of individuals and reflects its potential influence on systemic redox regulation. Individuals with higher OBS typically have greater antioxidant intake and healthier lifestyle behaviors [9, 11]. These characteristics are generally associated with lower oxidative burden and stronger endogenous antioxidant defenses, which may partly explain the inverse associations observed between OBS and the risks of CMM and mortality. Consistent with this interpretation, previous studies have reported significant associations between OBS and several biomarkers related to oxidative stress [8, 27]. In the present study, mediation analyses further suggested that inflammatory processes and IR may represent potential pathways linking oxidative exposure patterns with mortality risk. Although the mediated proportions were modest, this finding is consistent with the multifactorial nature of cardiometabolic diseases, which usually arise from the interaction of multiple biological pathways rather than a single mechanism. Beyond inflammation and metabolic dysfunction, other oxidative stress–related processes may also contribute to these associations. These mechanisms may include mitochondrial dysfunction, oxidative DNA damage, and alterations in cellular signaling pathways [7, 28].
In addition to conventional statistical approaches, we explored the potential predictive value of OBS using machine learning models. A LightGBM model incorporating OBS components and demographic variables demonstrated slightly higher predictive performance than traditional regression models for predicting CMM in the overall population and all-cause mortality among participants without CMM. The improvement in predictive performance was modest, suggesting that OBS itself already captures a substantial amount of relevant lifestyle related information. Nevertheless, machine learning models offer advantages for identifying complex patterns within multidimensional data [29]. Using SHAP based interpretation, BMI, niacin intake, physical activity, and overall dietary antioxidant intake were identified as important contributors to the prediction of CMM and mortality risk. These findings further highlight the importance of modifiable lifestyle factors in cardiometabolic health.
Previous studies have shown that elevated BMI is associated with redox imbalance, chronic inflammation, and increased risks of CMM and mortality [30, 31]. Regular physical activity, in contrast, can enhance endogenous antioxidant defense systems and improve metabolic regulation [32]. Dietary antioxidants such as vitamin E and vitamin B2 can scavenge reactive oxygen species, inhibit lipid peroxidation, support mitochondrial function, and help maintain vascular endothelial integrity [33, 34]. However, it is important to recognize that certain antioxidants may exhibit bidirectional effects depending on dosage and metabolic context [35]. High doses of niacin, for example, may promote inflammatory responses through via its metabolite 4-pyridone-3-carboxamide [36], while excessive vitamin E supplementation has also been linked to adverse cardiovascular outcomes and bleeding risk in some studies [37]. For this reason, although OBS reflects the combined effects of multiple oxidative exposures, caution is needed when interpreting the roles of individual components. Variable importance identified in predictive models should not be interpreted as direct evidence of causality. Overall, OBS is a composite measure summarizing dietary and lifestyle factors related to oxidative balance. Our findings suggest that this profile may be informative for early risk assessment in cardiometabolic health. Further studies are needed to clarify the underlying biological pathways, determine optimal exposure levels, and evaluate whether interventions targeting oxidative balance can reduce the long-term risks of CMM and related mortality.
Limitation
Despite these findings, several limitations should be acknowledged. First, the OBS was calculated from a single baseline measurement based on self-reported dietary and lifestyle parameters, which may not fully capture long-term oxidative status or temporal fluctuations. OBS reflects estimated oxidative exposures rather than directly measured redox status, and the absence of biochemical oxidative stress markers limits its biological interpretation. Furthermore, although the overall sample size was large, some subgroups had relatively small sample sizes and limited numbers of events, which may have reduced statistical power to detect associations within these groups. Second, some cardiometabolic diagnoses were self-reported and lacked standardized clinical verification, introducing potential misclassification bias. Third, although our main models adjusted for a wide range of confounders, residual confounding due to treatment adherence, disease severity, or lifestyle changes after diagnosis cannot be completely ruled out. Additionally, deaths occurring before CMM diagnosis may have introduced survivor bias. Fourth, the association between OBS and CMM was based on cross-sectional analyses, and therefore the temporal sequence and causal relationship cannot be determined. Fifth, although both inflammatory markers and IR indicators showed modest mediating effects, the limited effect sizes suggest that additional biological pathways may also contribute to the association between OBS and mortality, warranting further investigation. Sixth, as a composite index, OBS does not allow for causal interpretation of individual components, and dose-dependent effects of antioxidants are not captured by the scoring system. Finally, the generalizability of our findings may be limited due to the population-specific construction of the OBS and the lack of external validation for our ML models, and require further research before widespread clinical application. Future prospective studies with repeated measurements and validation in independent cohorts are needed to strengthen causal inference and evaluate the clinical utility of OBS.
Conclusion
CMM poses a substantial and growing threat to population health. In this context, OBS provides a framework reflecting dietary and lifestyle exposures related to oxidative balance and may help identify individuals at elevated risk in the early stages of cardiometabolic disease. Incorporating OBS into health risk assessment could support early identification of high-risk individuals and inform lifestyle-based prevention strategies aimed at improving cardiometabolic health outcomes. Future studies are needed to further elucidate the underlying biological mechanisms, determine optimal exposure patterns, and evaluate whether OBS-informed approaches can contribute to improved cardiometabolic health.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Abbreviations
- ATE
Alpha-tocopherol equivalent
- AUC
Area under curve
- AISI
Aggregate index of systemic inflammation
- CI
Confidence interval
- CKD-EPI
Chronic kidney disease epidemiology collaboration
- CMM
Cardiometabolic multimorbidity
- CVD
Cardiovascular disease
- DM
Diabetes mellitus
- eGFR
Estimated glomerular filtration rate
- FPG
Fasting plasma glucose
- HbA1c
Hemoglobin A1c
- HDL-C
High-density lipoprotein cholesterol
- HR
Hazard ratio
- ICD
International classification of diseases
- LDL-C
Low-density lipoprotein cholesterol
- IR
insulin resistance
- NCHS
National center for health statistics
- NDI
National death index
- NHANES
National health and nutrition examination survey
- NLR
Neutrophil-to-lymphocyte ratio
- MET
Metabolic equivalent
- OBS
Oxidative balance score
- OR
Odds ratio
- PC
Principal Component
- PIR
Poverty income ratio
- RCS
Restricted cubic spline
- RE
Retinol equivalent
- ROC
Receiver operating characteristic
- ROS
Reactive oxygen species
- SII
Systemic immune-inflammation index
- SIRI
Systemic inflammation response index
- SMD
Standardized mean difference
- TyG index
Triglyceride–Glucose index
- TyG-BMI index
Triglyceride-Glucose-Body Mass Index
- TyG-WC index,
Triglyceride-Glucose-Waist Circumference index
- SHAP
SHapley additive exPlanations
- VIF
Variance inflation factors
- WBC
White blood cell.
Author contributions
TSY: Conceptualization, Visualization, Validation, Writing-original draft, Writing – review & editing; ZGM: Software implementation, Writing-review & editing; HT: Methodology development, Writing-review & editing; ZZX: Methodology development, Software implementation, Writing-review & editing; ZJB: Investigation, Resources, Writing-review & editing; ZJY: Validation, Writing-review & editing; LFQ: Validation, Writing-review & editing; CH: Writing-review & editing; LC: Writing-review & editing; XYC: Writing-review & editing; LZJ: Writing-review & editing; LQZ: Writing-review & editing; TT: Conceptualization, Methodology development, Supervision, Writing-review & editing; LQM: Conceptualization, Methodology development, Supervision, Writing-review & editing. All authors gave final approval of the manuscript and agree to be accountable for all aspects of the work to ensure its integrity and accuracy.
Funding
This work was supported by National Natural Science Foundation of China (No. 82270337, No. 82070356, No. 82470333); the Key Project of Hunan provincial science and technology innovation (No. 2024JK2119); the Science and technology innovation of 2030 major project topics of China (No.2023ZD0504200); the National Science Foundation of Hunan Province of China (No. 2024JJ6593, No. 2024JJ9198); Clinical Medical Technology Innovation Guidance Project of the Hunan Provincial Department of Science and Technology (NO. 2021SK53527).
Data availability
The datasets analyzed in this study are publicly available at https://wwwn.cdc.gov/nchs/nhanes/Default.aspx. Further datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request. The machine learning analysis code is available at: https://github.com/asdfvxfcz/CMM-and-All-cause-mortality-R.git.
Declarations
Ethics approval and consent to participate
This NHANES study was performed using public data from the NCHS and the NHANES. Prior to enrollment, all participants furnished informed consent. The study requires no further approval and follows ethical guidelines.
Consent to participate
All the authors participated in the study and made significant intellectual contributions to the manuscript.
Consent for publication
This manuscript is not currently under consideration for publication elsewhere, and the work reported will not be submitted for publication elsewhere until a final decision has been made as to its acceptability by the journal.
Conflicts of interest
All authors confirm that they have no conflicts of interest to declare, and none has received any personal fees.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Tao Tu, Email: tutaotodd@csu.edu.cn.
Qiming Liu, Email: qimingliu@csu.edu.cn.
References
- 1.GBD 2021 Diseases and Injuries Collaborators. Global incidence, prevalence, years lived with disability (YLDs), disability-adjusted life-years (DALYs), and healthy life expectancy (HALE) for 371 diseases and injuries in 204 countries and territories and 811 subnational locations, 1990-2021: a systematic analysis for the Global burden of disease study 2021. Lancet. 2024;403(10440):2133–61. 10.1016/S0140-6736(24)00757-8. [DOI] [PMC free article] [PubMed]
- 2.Prince MJ, Wu F, Guo Y, Gutierrez Robledo LM, O’Donnell M, Sullivan R, et al. The burden of disease in older people and implications for health policy and practice. Lancet. 2015;385(9967):549–62. 10.1016/S0140-6736(14)61347-7. [DOI] [PubMed] [Google Scholar]
- 3.Ni Y, Zhou Y, Kivimäki M, Cai Y, Carrillo-Larco RM, Xu X, et al. Socioeconomic inequalities in physical, psychological, and cognitive multimorbidity in middle-aged and older adults in 33 countries: a cross-sectional study. LancetThe Lancet Healthy Longevity. 2023;4(11):e618–28. 10.1016/S2666-7568(23)00195-2. [DOI] [PubMed]
- 4.Di Angelantonio E, Kaptoge S, Wormser D, Willeit P, Butterworth AS, Bansal N, et al. Association of cardiometabolic multimorbidity with mortality. JAMA. 2015;314(1):52–60. 10.1001/jama.2015.7008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Han Y, Hu Y, Yu C, Guo Y, Pei P, Yang L, et al. Lifestyle, cardiometabolic disease, and multimorbidity in a prospective Chinese study. Eur Heart J. 2021;42(34):3374–84. 10.1093/eurheartj/ehab413. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Luo J, Mills K, le Cessie S, Noordam R, van Heemst D. Ageing, age-related diseases and oxidative stress: what to do next? Ageing Res Rev. 2020;57:100982. 10.1016/j.arr.2019.100982. [DOI] [PubMed] [Google Scholar]
- 7.Forman HJ, Zhang H. Targeting oxidative stress in disease: promise and limitations of antioxidant therapy. Nat Rev Drug Discov. 2021;20(9):689–709. 10.1038/s41573-021-00233-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Hernández-Ruiz Á, García-Villanova B, Guerra-Hernández EJ, Carrión-García CJ, Amiano P, Sánchez MJ, et al. Oxidative balance scores (OBSs) integrating nutrient, food and lifestyle dimensions: development of the NutrientL-OBS and FoodL-OBS. Antioxid (Basel). 2022;11(2):300. 10.3390/antiox11020300. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Chen Y, Wu S, Liu H, Zhong Z, Bucci T, Wang Y, et al. Role of oxidative balance score in staging and mortality risk of cardiovascular-kidney-metabolic syndrome: insights from traditional and machine learning approaches. Redox Biol. 2025;81:103588. 10.1016/j.redox.2025.103588. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Zhang W, Peng SF, Chen L, Chen HM, Cheng XE, Tang YH. Association between the oxidative balance score and telomere length from the National health and nutrition examination survey 1999-2002. Oxidative Med Cellular Longevity. 2022;2022(1):1345071. 10.1155/2022/1345071. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Xu Z, Liu D, Zhai Y, Tang Y, Jiang L, Li L, et al. Association between the oxidative balance score and all-cause and cardiovascular mortality in patients with diabetes and prediabetes. Redox Biol. 2024;76:103327. 10.1016/j.redox.2024.103327. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Survey NHaNE. https://wwwn.cdc.gov/nchs/nhanes/Default.aspx.
- 13.Levey AS, Stevens LA, Schmid CH, Zhang Y, Castro AF 3rd, Feldman HI, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150(9):604–12. 10.7326/0003-4819-150-9-200905050-00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Chen K, Li S, Xie Z, Liu Y, Li Y, Mai J, et al. Association between oxidative balance score, systemic inflammatory response index, and cardiovascular disease risk: a cross-sectional analysis based on NHANES 2007–2018 data. Front Nutr. 2024;11:1374992. 10.3389/fnut.2024.1374992. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lai Q, Ye L, Luo J, Zhang C, Wu Q, Shao Y. The cross-sectional correlation between the oxidative balance score and cardiometabolic risk factors and its potential correlation with longitudinal mortality in patients with cardiometabolic risk factors. BMC Public Health. 2024;24(1):1452. 10.1186/s12889-024-18967-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Gao Q, Zhu X, Chen M, Xia R, Zhang Q. Association between oxidative balance score and all-cause and cancer-specific mortality among cancer survivors. Front Immunol. 2025;16:1541675. 10.3389/fimmu.2025.1541675. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Li D, Lin J, Yang H, Zhou L, Li Y, Xu Z, et al. Causal association of modifiable factors with cardiometabolic multimorbidity: an exposome-wide Mendelian randomization investigation. Cardiovasc Diabetol. 2025;24(1):241. 10.1186/s12933-025-02790-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Ornago AM, Gregorio C, Triolo F, Moore AZ, Marengoni A, Beridze G, et al. Shared and specific blood biomarkers for multimorbidity. Nat Med. 2026;32(2):736–45. 10.1038/s41591-025-04038-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Huang M, Fu R, Zhao X, Liu T, Li X, Jiang W. Life’s essential 8 and progression of cardiometabolic multimorbidity trajectory: a prospective study of Uk Biobank. Eur J Prev Cardiol. 2025: Online ahead of print. 10.1093/eurjpc/zwaf247. [DOI] [PubMed]
- 20.Aleksandrova K, Koelman L, Rodrigues CE. Dietary patterns and biomarkers of oxidative stress and inflammation: a systematic review of observational and intervention studies. Redox Biol. 2021;42:101869. 10.1016/j.redox.2021.101869. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Legaard GE, Feineis CS, Johansen MY, Hansen KB, Vaag AA, Larsen EL, et al. Effects of an exercise-based lifestyle intervention on systemic markers of oxidative stress and advanced glycation endproducts in persons with type 2 diabetes: secondary analysis of a randomised clinical trial. Free Radical Biol & Med. 2022;188:328–36. 10.1016/j.freeradbiomed.2022.06.013. [DOI] [PubMed] [Google Scholar]
- 22.Wang P, Gao X, Willett WC, Giovannucci EL. Socioeconomic status, diet, and behavioral factors and cardiometabolic diseases and mortality. JAMA Netw Open. 2024;7(12):e2451837–2451837. 10.1001/jamanetworkopen.2024.51837. [DOI] [PMC free article] [PubMed]
- 23.Yang L, Zhang Z, Zhang J, Miao J, Zhang H, Du Y, et al. Stage-specific risk factors of cardiometabolic multimorbidity: a systematic review and meta-analysis from incidence to mortality. Ageing Res Rev. 2026;114:102991. 10.1016/j.arr.2025.102991. [DOI] [PubMed] [Google Scholar]
- 24.Xu C, Cao Z. Cardiometabolic diseases, total mortality, and benefits of adherence to a healthy lifestyle: a 13-year prospective Uk Biobank study. J Transl Med. 2022;20(1):234. 10.1186/s12967-022-03439-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Xu X, Pang Y, Fan X. Mitochondria in oxidative stress, inflammation and aging: from mechanisms to therapeutic advances. Sig Transduct Target Ther. 2025;10(1):190. 10.1038/s41392-025-02253-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Fiorenza M, Onslev J, Henríquez-Olguín C, Persson KW, Hesselager SA, Jensen TE, et al. Reducing the mitochondrial oxidative burden alleviates lipid-induced muscle insulin resistance in humans. Sci Adv. 2024;10(44):eadq 4461. 10.1126/sciadv.adq4461. [DOI] [PMC free article] [PubMed]
- 27.Tylavsky FA, Han L, Sims Taylor LM, Mason WA, Carroll KN, Bush NR, et al. Oxidative balance score during pregnancy is associated with oxidative stress in the CANDLE study. Nutrients. 2022;14(11):2327. 10.3390/nu14112327. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Halliwell B. Understanding mechanisms of antioxidant action in health and disease. Nat Rev Mol Cell Biol. 2024;25(1):13–33. 10.1038/s41580-023-00645-4. [DOI] [PubMed] [Google Scholar]
- 29.Acosta JN, Falcone GJ, Rajpurkar P, Topol EJ. Multimodal biomedical AI. Nat Med. 2022;28(9):1773–84. 10.1038/s41591-022-01981-2. [DOI] [PubMed] [Google Scholar]
- 30.Anusruti A, Jansen E, Gào X, Xuan Y, Brenner H, Schöttker B. Longitudinal associations of body mass index, waist circumference, and waist-to-hip ratio with biomarkers of oxidative stress in older adults: results of a large cohort study. Obes Facts. 2020;13(1):66–76. 10.1159/000504711. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Świątkiewicz I, Wróblewski M, Nuszkiewicz J, Sutkowy P, Wróblewska J, Woźniak A. The role of oxidative stress enhanced by adiposity in cardiometabolic diseases. IJMS. 2023;24(7):6382. 10.3390/ijms24076382. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Schöttker B, Salem AA, Schwenk M, GÀo X, Jansen E, Brenner H. Relationship of physical activity at older age with biomarkers of oxidative stress: a large, population-based cohort study. Med Sci Sports Exercise. 2021;53(12):2528–35. 10.1249/MSS.0000000000002731. [DOI] [PubMed] [Google Scholar]
- 33.Chandimali N, Bak SG, Park EH, Lim H-J, Won Y-S, Kim E-K, et al. Free radicals and their impact on health and antioxidant defenses: a review. Cell Death Discov. 2025;11(1):19. 10.1038/s41420-024-02278-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Meulmeester FL, Luo J, Martens LG, Mills K, van Heemst D, Noordam R. Antioxidant Supplementation in oxidative stress-related diseases: what have we learned from studies on Alpha-tocopherol? Antioxid (Basel). 2022;11(12):2322. 10.3390/antiox11122322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Li B, Ming H, Qin S, Nice EC, Dong J, Du Z, et al. Redox regulation: mechanisms, biology and therapeutic targets in diseases. Sig Transduct Target Ther. 2025;10(1):72. 10.1038/s41392-024-02095-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Ferrell M, Wang Z, Anderson JT, Li XS, Witkowski M, DiDonato JA, et al. A terminal metabolite of niacin promotes vascular inflammation and contributes to cardiovascular disease risk. Nat Med. 2024;30(2):424–34. 10.1038/s41591-023-02793-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Kaye AD, Thomassen AS, Mashaw SA, MacDonald EM, Waguespack A, Hickey L, Singh A, Gungor D, Kallurkar A, Kaye AM, et al. Vitamin E (α-tocopherol): emerging clinical role and adverse risks of supplementation in adults. Cureus. 2025;17:e78679. 10.7759/cureus.78679. [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets analyzed in this study are publicly available at https://wwwn.cdc.gov/nchs/nhanes/Default.aspx. Further datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request. The machine learning analysis code is available at: https://github.com/asdfvxfcz/CMM-and-All-cause-mortality-R.git.