Study population

This study combines secondary dataset analyses and Mendelian randomization analyses. The National Health and Nutrition Examination Survey (NHANES) is a continuous series of nationwide surveys conducted every two years to collect representative cross-sectional data related to the health of the general US population. The NHANES program, employing a complex multi-stage probabilistic sampling design, involves household interviews, physical examinations, and laboratory tests on a selected sample of the non-institutionalized US population. Detailed study design, methods, and data collection can be found on the NHANES website. This analysis utilized publicly available data downloaded from the NHANES official website.

This study utilized data from the NHANES 1999–2010 cycles because phenotypic age data, used to measure the level of biological aging, could only be calculated through original clinical biomarkers available in these cycles. Thus, the study included 62160 participants, excluding those missing phenotypic age data (n=23112), under 20 years of age (n=10513), missing serum cotinine levels (n=9607), active smokers (n=9049), and missing other covariates (n=1810). A final cohort of 8069 eligible participants underwent weighted analysis to represent the national sample. Figure 1 illustrates the participant selection process.

Figure 1

Participant selection and Mendelian randomization assumptions in the study of secondhand smoke and biological aging

Secondhand smoke exposure levels

To assess the level of SHS exposure, we used serum cotinine data from NHANES. Cotinine, the primary metabolite of nicotine in the body, serves as an effective biomarker for assessing smoke exposure due to its serum half-life being eight times that of nicotine¹⁴. Serum cotinine levels were determined using Isotope Dilution High-Performance Liquid Chromatography Atmospheric Pressure Chemical Ionization Tandem Mass Spectrometry (ID HPLC-APCI MS/MS). Detailed laboratory information can be found on the official website¹⁵. Active smokers, defined as those who have smoked >100 cigarettes in their lifetime¹⁶, were excluded to better assess SHS exposure. Based on previous studies¹⁷, SHS exposure levels were classified into no exposure (<0.05 ng/mL), low exposure (0.05–0.99 ng/mL), and high exposure (≥1 ng/mL).

Measurement of biological aging

Following prior research¹⁸, phenotypic age was calculated using 9 biomarkers, including actual age, albumin, creatinine, glucose, C-reactive protein, lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase, and white blood cell count. The calculation of phenotypic age was based on data from the NHANES III cohort and has been proven to be a reliable predictor of mortality, reflecting the state of biological aging more accurately than actual age¹⁹.

To further demonstrate the validity of phenotypic age, we assessed its concordance with chronological age in our study population. As shown in Supplementary file Figure 1, phenotypic age was strongly correlated with chronological age (Pearson’s r=0.92, p<2.2×10^-16), supporting its use as a surrogate marker of biological aging.

Covariates

Following previous research²⁰, the following covariates were included: age, gender, race, education level, family poverty-income ratio, body mass index, alcohol consumption, hypertension, diabetes, and stroke history. Historical data were self-reported.

Race/ethnicity was categorized as Mexican American, Non-Hispanic White, Non-Hispanic Black, Other Hispanic, and Other (including multiracial). Education level was grouped into three categories: <high school, high school graduate, and >high school. BMI (kg/m²) was calculated from measured weight and height. PIR, as an indicator of socioeconomic status, was defined as the ratio of total household income to the federal poverty threshold determined by the US Census Bureau.

Smoking status was classified based on whether participants had smoked at least 100 cigarettes in their lifetime. Alcohol consumption was defined as having consumed alcohol at least 12 times in any one year. Hypertension, diabetes, and stroke were determined based on self-reported physician diagnoses; participants reporting a history of any of these conditions were classified as positive.

Statistical analysis

Continuous variables were expressed as mean ± standard deviation (SD) or median and interquartile range (IQR), and categorical variables as frequencies (n) and percentages (%). The Kolmogorov-Smirnov test was used to assess normal distribution. Depending on the characteristics of the variables, analysis of variance, non-parametric tests, or chi-squared tests were used to compare differences between groups. Weighted multifactorial linear regression was used to evaluate the relationship between SHS exposure and biological aging, using three models: Model 1 with no adjustments; Model 2 adjusted for gender and age; and Model 3 fully adjusted for age, gender, race, education level, family poverty-income ratio, body mass index, alcohol consumption, hypertension and stroke history. Generalized additive models (GAM) assessed the nonlinear relationship between SHS exposure and biological aging; if a nonlinear relationship was observed, two-stage linear regression and recursive methods calculated the effect threshold. Restricted cubic splines visualized the relationship between SHS exposure and biological aging, with four nodes set. Due to the skewed distribution of serum cotinine levels, log transformation was necessary for enhancing the statistical efficiency of the restricted cubic splines analysis. Non-linearity was tested using a Wald test, which jointly evaluates whether the coefficients of the spline terms (excluding the linear component) are equal to zero. Stratified linear regression models conducted subgroup analyses, and likelihood ratio tests evaluated the effect across subgroups. Based on prior evidence linking SHS exposure to increased diabetes risk, and recognizing the role of diabetes in promoting biological aging, we conducted mediation analysis to examine whether diabetes mediates the effect of SHS exposure on biological aging. All statistical analyses were performed using R 4.3.0. A two-sided p<0.05 was considered statistically significant.

Mendelian randomization

The MR study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology Using MR (STROBE-MR) guidelines²¹. Summary data from genome-wide association studies (GWAS) for SHS exposure and biological aging selected suitable SNPs as instrumental variables (IVs) for MR analysis to explore the causal relationship. The application of IVs in MR analysis relies on satisfying three critical assumptions: 1) the chosen IVs exhibit a robust association with the exposure of interest, 2) there is no confounding of the IVs with factors influencing the outcome apart from the exposure; and 3) the selected IVs solely influence the outcome through the exposure²². Since the workplace is a primary source of SHS exposure, data on workplace SHS exposure from the UK Biobank (GWAS ID: ukb-d-22611_1) involving 90168 participants and biological aging data from a 2021 GWAS study (GWAS ID: ebi-a-GCST90014304) with 6148 participants were analyzed. All participants were of European ancestry and provided informed consent, with no overlap between datasets.

SNPs closely associated with exposure (p<1×10^-5) were selected, and linkage disequilibrium (LD) among SNPs was removed (r²<0.001, clumping distance=10000 kb). When encountering genetic variations in LD, the variant with the lowest p-value related to the exposure factor was selected. To ensure the IV’s strong association with exposure, SNPs with an F-statistic >10 were selected to exclude weak instrumental variables. The inverse-variance weighted (IVW, random effect) method served as the primary analysis method to assess the correlation between exposure and outcomes, providing accurate estimates when IVs are valid²³. We further implemented three supplementary methods: 1) MR-Egger regression, which allows for directional pleiotropy adjustment through its intercept term; 2) Weighted Median estimator, providing valid estimates when up to 50% of instruments are invalid; and 3) Simple Median estimator as an additional pleiotropy-robust approach. Results are presented as β-values with standard errors (SE). Cochrane’s Q test evaluated heterogeneity. Potential horizontal pleiotropy was assessed using the intercept of MR-Egger regression.

Baseline characteristics according to levels of secondhand smoke exposure

The baseline characteristics categorized by different levels of SHS exposure are summarized in Table 1, with all data being appropriately weighted. The study involved 8069 participants, among whom 4947 reported no exposure to SHS, 2408 low exposure, and 714 high exposure. Notably, increased levels of SHS exposure correlated with being younger, male, of Non-Hispanic Black ethnicity, a low level of education, low family poverty-income ratio, high body mass index, and consuming alcohol (p<0.001). Furthermore, elevated exposure levels were significantly associated with hypertension (p=0.009). To illustrate biological aging more effectively, we computed both the phenotypic age and the phenotypic age residual (phenotypic age minus actual age), revealing that higher SHS exposure levels corresponded to a reduced phenotypic age and a smaller phenotypic age residual. This indicates accelerated biological aging (p<0.001) (Table 1).

Multifactorial linear regression analysis

We conducted weighted multifactorial linear regression analysis to investigate the relationship between SHS exposure and biological aging, employing three different models for this analysis. Model 1 applied no adjustments; Model 2 adjusted for gender and age; and Model 3 was fully adjusted for all covariates. As depicted in Table 2, Model 1 (unadjusted) showed that, compared to individuals with no exposure, low exposure to SHS was associated with a 0.95-year acceleration in biological aging (β=0.95; 95% CI: 0.59–1.31), and high exposure resulted in a 1.68-year increase (β=1.68; 95% CI: 1.11–2.24). Model 2, adjusting for gender and age, indicated that low exposure accelerated biological aging by 1.07 years (β=1.07, 95% CI: 0.71–1.44), and high exposure by 1.85 years (β=1.85; 95% CI: 1.28–2.42). In the fully adjusted Model 3, low exposure was linked to a 0.37-year acceleration in biological aging (β=0.37; 95% CI: 0.04–0.70), and high exposure by 0.76 years (β=0.76; 95% CI: 0.23–1.29) (Table 2). To further test the robustness of our findings, we conducted a sensitivity analysis by additionally adjusting for physical activity, given its importance as a potential covariate of interest. The results remained consistent: compared with the non-exposed group, low SHS exposure was associated with a 0.37-year increase (β=0.37; 95% CI: 0.04–0.70) in biological aging, and high exposure with a 0.74-year increase (β=0.74; 95% CI: 0.21–1.27) (Supplementary file Table 1), confirming the stability of our main findings.

Threshold effects and non-linear associations

To further examine the non-linear association between SHS exposure and biological aging, we utilized a generalized additive model (GAM) for two-stage linear regression and applied recursive methods for determining the effect threshold²⁴. Employing restricted cubic splines, we examined the relationship between log-transformed serum cotinine concentrations (a proxy for SHS exposure) and biological aging. The results revealed a U-shaped relationship between log serum cotinine concentration and biological aging (p for nonlinearity <0.001), with a log-critical threshold identified at -1.53 (Figure 2). Two-stage linear regression showed that, for log serum cotinine concentrations below -1.53, each unit increase in the log-transformed cotinine concentration was associated with a 1.89-year delay in biological aging (β= -1.89; 95% CI: -2.79 – -1.00); conversely, above -1.53, each unit increase was linked to a 0.38-year acceleration in aging (β=0.38; 95% CI: 0.20–0.57) (Supplementary file Table 2).

Figure 2

Nonlinear association between log-transformed serum cotinine and biological aging

Subgroup and mediation analyses

In subgroup analyses that included variables such as age, gender, ethnicity, education level, BMI, hypertension, diabetes, and alcohol consumption, we found that low exposure to SHS remained significantly associated with biological aging in males, Non-Hispanic Whites, individuals with a high school education, those with a BMI <30 kg/m², and alcohol consumers. High exposure was significantly linked to biological aging in individuals under 60 years, females, Non-Hispanic Whites, those with an education level beyond high school, individuals with a BMI <25 kg/m², and those with hypertension, diabetes, and alcohol consumption habits. Moreover, significant interactions were observed for age (p for interaction=0.023), gender (p for interaction=0.011), and diabetes status (p for interaction=0.014), indicating that the association between SHS exposure and biological aging may be modified by these factors (Supplementary file Table 3).

Through subgroup analysis, we observed that the association between SHS exposure and biological aging was amplified among individuals with diabetes (Supplementary file Table 3). Therefore, we used a mediation effect model to explore whether the association between SHS exposure and biological aging was mediated by diabetes. After adjusting for all covariates and applying weighting, results indicated a significant direct effect of SHS exposure on biological aging (β=0.38; 95% CI: 0.09–0.68, p=0.010) and an indirect effect through the impact of SHS exposure levels on diabetes, ultimately affecting biological aging (β=0.17; 95% CI: 0.00–0.40, p=0.046). The mediation proportion attributed to diabetes was 31.25%, establishing a partial mediation model (Figure 3).

Figure 3

Mediation effect of diabetes in the association between SHS exposure and biological aging

Mendelian randomization analysis

In our MR analysis, we specifically focused on workplace SHS exposure as it represents a major source of environmental smoke exposure for adults. While residential exposure constitutes another important source of SHS, our MR design intentionally utilized genetic variants associated with workplace exposure to ensure biological plausibility of the instrumental variables and minimize confounding from shared household environments. It should be emphasized that this MR analysis specifically addresses workplace exposure. The IVW model revealed that genetically predicted workplace SHS exposure was associated with a 3.05-year acceleration in biological aging (β=3.05; 95% CI: 0.24–5.85, p=0.033) (Figure 4A). The scatter plot shows that the results of merging MR models such as MR Egger, Simple Median, and Weighted Median are basically consistent, indicating the robustness of the results (Figure 4B). MR-Egger analysis showed no horizontal pleiotropy (intercept p=0.352), though this method requires satisfaction of the Instrument Strength Independent of Direct Effect (InSIDE) assumption, and that the strength of SNP-exposure associations is independent of their pleiotropic effects. Cochrane’s Q test indicated no heterogeneity (p=0.817), supporting the robustness of these workplace-specific findings.

Figure 4

Mendelian randomization analysis of workplace SHS exposure and biological aging

Strengths and limitations

The strengths of this study lie in its novel exploration of SHS exposure’s effect on biological aging, unveiling an association with accelerated aging and providing fresh epidemiological insights for reducing SHS exposure and delaying aging. The use of nationally representative data with a substantial sample size strengthens our conclusions. By analyzing the threshold effect, we determined a U-shaped relationship between serum cotinine concentration and biological aging, establishing an optimal concentration for aging delay. This finding lays down the groundwork for mechanistic studies and further population segmentation. Particularly, our analysis highlights the relevance of this relationship in the diabetic population. Through two-sample MR, we corroborated the causal link between workplace SHS exposure and biological aging, supplementing cross-sectional studies with genetic evidence.

However, this study’s limitations include the inability to account for all potential covariates due to data constraints and to explore alternative biological aging measures comprehensively. Our MR analysis also has several limitations. First, the Mendelian randomization analysis relied on genetic instruments derived from European ancestry GWAS datasets for both workplace SHS exposure and phenotypic age acceleration. While we minimized weak instrument bias by selecting SNPs with strong associations and performed sensitivity analyses to address pleiotropy, the lack of multi-ethnic GWAS summary statistics limits the generalizability of our causal estimates to non-European populations. Differences in linkage disequilibrium patterns, allele frequencies, and gene-environment interactions across ancestries may influence the validity of genetic instruments in diverse groups. Second, sex-stratified analyses were precluded by the absence of sex-specific GWAS data for phenotypic age acceleration. Third, while MR-Egger regression and Cochran’s Q test indicated no significant horizontal pleiotropy or heterogeneity, residual confounding from unmeasured pleiotropic pathways cannot be entirely ruled out. These limitations underscore the need for future large-scale GWAS efforts in underrepresented populations to strengthen causal inference and improve the external validity of MR findings in aging research.