Introduction

Cardiovascular disease (CVD) remains a major global health challenge, contributing significantly to morbidity and mortality worldwide. Physical activity (PA) is a well-established independent predictor of cardiovascular health and a modifiable risk factor for CVD1,2. However, population-based studies indicate that approximately 27.5% of adults fail to meet recommended guidelines of PA, with substantial implications for population health2.

Arterial stiffness, particularly measured by carotid-femoral pulse wave velocity (cfPWV), has emerged as a key indicator of vascular aging and a critical predictor of future cardiovascular risk. cfPWV has been shown to provide incremental predictive value for CVD and all-cause mortality beyond traditional risk factors, such as blood pressure3,4. However, its clinical application is often constrained by the need for specialized equipment and trained personnel, limiting its widespread use in routine clinical practice. In contrast, estimated pulse wave velocity (ePWV) offers a promising, non-invasive alternative that has shown predictive accuracy comparable to cfPWV5. Recent studies have highlighted a significant association between ePWV and cardiovascular events, positioning it as a potential widely accessible biomarker for cardiovascular risk assessment6,7,8. Notably, a large European cohort study demonstrated that ePWV is independently associated with all-cause mortality, cardiovascular mortality, and composite cardiovascular outcomes, extending predictive capabilities beyond traditional risk models such as the Systematic Coronary Risk Evaluation (SCORE) and the Framingham Risk Score (FRS)9.

While emerging evidence suggests a link between ePWV and cardiovascular outcomes, the mechanistic pathways through which PA influences cardiovascular health remain underexplored. To date, no study has comprehensively examined whether ePWV mediates the relationship between physical inactivity and CVD risk. The present study uniquely employs causal mediation analysis (CMA)10 to investigate whether ePWV mediates the effect of physical inactivity on CVD risk. By utilizing the potential outcome framework, we aim to estimate the causal effect of exposure-induced mediators and identify the indirect pathway through which PA influences cardiovascular risk via ePWV.

Methods

Study design and participants

This 7 year follow-up cohort study was conducted as part of the China PEACE Million Persons Project (MPP), details of which have been previously described11,12. The study targeted a sub-cohort of community-dwelling adults in southern China, specifically from six locations in Jiangxi Province, in 2016. Eligible participants were adults aged 35–75 years who had resided in the study area for at least six months in the preceding year. The baseline survey was conducted in 2016, focusing on the high-risk CVD population. Annual follow-up questionnaires were administered by trained medical personnel to track CVD incidence through 2023.

This study received ethical approval from the Institutional Review Board of Fuwai Hospital, Chinese Academy of Medical Sciences (No. 2014-74). All procedures were conducted in accordance with the principles of the Declaration of Helsinki, and written informed consent was obtained from each participant.

Data collection

Trained healthcare professionals conducted face-to-face interviews to collect data on participants’ demographic characteristics, socioeconomic status, and comorbidities. Following enrollment, each participant underwent a thorough physical examination. Blood pressure was measured twice on the right arm while seated, and the average value was recorded. Anthropometric measurements were collected following standardized protocols, and body mass index (BMI) was calculated as weight divided by height squared. Lipid profiles, including total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C), were assessed, along with fasting blood glucose (FBG), according to established protocols11.

Hypertension was defined by meeting any of the following criteria: (1) a clinically confirmed diagnosis; (2) systolic blood pressure (SBP) ≥ 140 mmHg and/or diastolic blood pressure (DBP) ≥ 90 mmHg; or (3) ongoing antihypertensive pharmacotherapy. Diabetes mellitus was diagnosed based on either: (1) a physician-diagnosed history; (2) FBG ≥ 7.0 mmol/L; or (3) current use of glucose-lowering agents.

Exposure variable and mediator

PA status, assessed in the 2016 survey, served as the exposure variable. Participants were asked to report the average daily time spent on occupational, leisure, transportation-related, and household activities over the past year. Total weekly PA (MET-min/week) was calculated by summing domain-specific products of time and MET values. MET assignments followed Li et al.’s population-specific framework13, which adapts established compendia to Chinese activity patterns. Based on established classifications, participants were categorized into three PA levels: “high” (> 3000 MET-min/week), “moderate” (600–3000 MET-min/week), and “low” (< 600 MET-min/week)14,15.

ePWV was used as the mediator in 2018, calculated using the participant’s age and mean blood pressure (MBP) from the 2018 follow-up measurement. Since this study specifically targets a high-risk CVD population, the following formula was used to calculate ePWV5,16: ePWV = 9.587 − (0.402 × age) + [4.560 × 0.001 × (age2)] − [2.621 × 0.00001 × (age2) × MBP] + (3.176 × 0.001 × age × MBP) − (1.832 × 0.01 × MBP). MBP was calculated as DBP + 0.4 (SBP − DBP)17.

In accordance with expert consensus recommendations, ePWV was categorized into two groups: “<10 m/s” and “≥10 m/s"18. For sensitivity analyses, CMA was also conducted considering ePWV as a continuous variable.

Outcome definition

The primary outcome of this study was the incidence of CVD over a 5 year follow-up period, from 2018 to 2023. After the baseline assessment, participants were followed for CVD events via annual clinic visits or telephone follow-ups. During these interactions, study personnel systematically inquired about all medical events, including hospitalizations.

CVD was defined as the first occurrence of any of the following events: myocardial infarction, angina, heart failure, arrhythmias, valvular heart disease, coronary heart disease, stroke, or death due to cardiovascular causes. Cardiovascular death was classified according to the International Classification of Diseases, 10th Revision (ICD-10), specifically codes for heart diseases (I00–I09, I11, I13, I20–I51) and cerebrovascular diseases (I60–I69).

Statistical analysis

Continuous variables are presented as means ± standard deviation (SD), and categorical variables are expressed as percentages. Differences between groups were assessed using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. Kaplan-Meier survival curves were generated to evaluate the cumulative incidence of CVD across different PA levels, with comparisons made using log-rank tests, stratified by ePWV. Multivariate Cox proportional hazards models were employed to examine the associations between ePWV and CVD, as well as between PA level and CVD, reporting hazard ratio (HR) with corresponding 95% confidence interval (CI). Cox proportional hazards models were rigorously evaluated for proportional hazards assumptions using Schoenfeld residuals (P > 0.05 for all models). Three hierarchical adjustment models were specified: The Crude Model was unadjusted. Model 1 was adjusted for sex (male, female) and age group (< 65, ≥ 65 years). Model 2 was additionally adjusted for marital status (single/separated, coupled), education attainment (< high school, ≥high school), occupation (farmer: yes, no), annual household income (< 50,000, ≥ 50,000 CNY, unclear), smoking status (current smoker: yes, no), alcohol consumption (current drinker: yes, no), sedentary time (< 4, ≥ 4 h/day), PA level (high, moderate, low), BMI level (< 28, ≥ 28 kg/m2), baseline pulse pressure (continuous), hypertension (yes, no), diabetes mellitus (yes, no), FBG (continuous), TC (continuous), and HDL-C (continuous). When analyzing PA and CVD associations or causal mediation effects, Model 2 excluded PA level to avoid overadjustment. Figure 1 presents the hypothesized causal diagram for this study. To explore whether ePWV mediated the relationship between PA (the exposure variable) and CVD (the outcome), we performed CMA, using the “CMAVerse” package in R using the regression-based approach19. Binary mediators (ePWV level) were modeled using logistic regression (logit link), continuous mediators (continuous ePWV) with linear regression (identity link), and time-to-event outcomes with Cox proportional hazards regression. All models were adjusted for pre-specified baseline confounders. Effect estimates were obtained using bootstrap resampling with 500 iterations. In this framework, the exposure variable was PA level from the 2016 survey, the mediator was ePWV from the 2018 survey, and the outcome was CVD incidence from 2018 to 2023. The analysis decomposed the total effect (TE) of PA on CVD into the natural direct effect (NDE) and the natural indirect effect (NIE) via ePWV. A non-zero NIE suggests that the exposure affects the mediator, which in turn influences the outcome, indicating a mediation Additionally, we calculated the proportion mediated (PM), which quantifies the extent to which the NIE explains the total effect. For models using continuous ePWV as a mediator, linear regression was applied to the mediator regression model. Missing data were imputed using mean substitution for continuous variables and mode substitution for categorical variables. Details of missing data before imputation are presented in Supplementary Table S1. A complete case analysis was conducted for sensitivity analysis. To assess the impact of unmeasured covariates on the estimates, we calculated the mediational E-values for the NIE and NDE. These values represent the minimum association strength required for an unmeasured confounder with the mediator and outcome to explain the mediation effect under the measured covariates. Given minimal missingness (< 2.5% for all variables) and predominantly categorical data, mean/mode imputation was selected over multiple imputation to avoid unnecessary model complexity while maintaining statistical stability, consistent with guidelines for low-missingness scenarios. All statistical analyses were performed using R software version 4.3.1, with statistical significance set at P < 0.05.

Fig. 1
Fig. 1
Full size image

Directed acyclic graph (DAG). Depicting causal relationships between PA, ePWV, and CVD risk, adjusting for potential confounders.

Results

Baseline characteristics

A total of 10,055 participants were included in the study, with the participants enrollment process illustrated in Fig. 2. Among the 10,055 participants, the mean age was 57.25 years (SD = 9.38), and 60.39% were female. During the follow-up period starting in 2018, 562 participants experienced incident CVD. Stroke (n = 265) and ischemic heart disease (n = 144, including angina, myocardial infarction, and coronary heart disease) were the most frequent events, collectively accounting for the majority of CVD endpoints—a finding consistent with the recognized role of arterial stiffness as a predictor of cerebrovascular and coronary diseases. The median observation period was 1772 days (IQR = 1372–1900 days), with a CVD incidence rate of 1.29/100 person-years. At baseline, PA levels were classified as “high” (74.17%, n = 7458), “moderate” (24.19%, n = 2432), and “low” (1.64%, n = 165), with detailed characteristics presented in Table 1.

Fig. 2
Fig. 2
Full size image

Participants enrollment and inclusion flowchart.

Table 1 Characteristics of the participants at baseline (n = 10,055).
Full size table

Association between ePWV or PA and CVD risk, PA and ePWV

The association between ePWV and CVD risk is detailed in Table 2. After adjusting for all covariates and PA level, each 1 m/s increase in ePWV was associated with a 24.2% increase in CVD risk (HR = 1.242; 95% CI 1.154, 1.337). Furthermore, an ePWV ≥ 10 m/s was significantly correlated with an increased CVD risk (HR = 1.673; 95% CI 1.319, 2.124). After adjusting for all covariates, participants in the moderate PA level group had a 1.236-fold higher CVD risk compared to the high PA level group (HR = 1.236; 95% CI 1.024, 1.492), while those in the low PA level group had 2.087 times the CVD risk of the high PA level group (HR = 2.087; 95% CI 1.348, 3.231), as shown in Supplementary Table S2. Additionally, lower PA levels were associated with higher ePWV, as shown in Supplementary Table S3. Kaplan-Meier survival curves (Supplementary Fig. S1) showed that participants with low and moderate PA levels had a higher cumulative CVD incidence compared to those with high PA levels. Additionally, individuals with ePWV ≥ 10 m/s exhibited a higher cumulative incidence of CVD than those with ePWV < 10 m/s.

Table 2 Association between the ePWV and CVD incidence (n = 10,055).
Full size table

Mediation analysis

CMA revealed that insufficient PA (both low and moderate levels) was associated with an increased risk of incident CVD. The high ePWV group (≥ 10 m/s) mediated 10.6% of this association (NIE: HR = 1.025; 95% CI 1.013, 1.041). Similarly, a positive NIE via the high ePWV group showed a mediation effect of 11.2% between moderate PA levels and an increased risk of CVD compared to high PA levels (NIE: HR = 1.022; 95% CI 1.011, 1.038). In contrast, low PA levels mediated 8.3% of the increased CVD risk through the high ePWV group compared to the high PA level group (NIE: HR = 1.048; 95% CI 1.011, 1.095) (Table 3).

Table 3 Mediation effect of ePWV levels (≥ 10/<10 m/s) on the association between PA levels and CVD risk (n = 10,055).
Full size table

Sensitivity analyses

Sensitivity analyses, presented in Table 4, further confirmed these findings by treating continuous ePWV as a mediator. The magnitude of each NIE was larger compared with the previous results, with a corresponding increase in the PM. These results indicated a more substantial NIE of ePWV through insufficient PA on CVD risk: For low/moderate PA levels, HR = 1.064 (95% CI 1.042, 1.093; PM = 25.5%); For moderate PA levels, HR = 1.060 (95% CI 1.038, 1.088; PM = 28.1%); For low PA levels, HR = 1.096 (95% CI 1.042, 1.167; PM = 15.9%). In the complete case analysis, the NIE at each PA level was consistent with the results from the imputed datasets, along with the corresponding PM, as detailed in Supplementary Table S4. Finally, Supplementary Table S5 presents the mediational E-values for unmeasured confounders in the relationship between ePWV, PA, and CVD risk. The E-values consistently exceeded the maximum NIE or NDE calculated for ePWV, further supporting the robustness of the CMA.

Table 4 Mediation effect of ePWV (continuous variable) on the association between PA levels and CVD risk (n = 10,055).
Full size table

Discussion

This study examined the relationship between PA-associated increases in ePWV and the risk of CVD, demonstrating that elevated ePWV is significantly correlated with increased CVD risk. Additionally, low and moderate PA levels were strongly associated with higher ePWV, with ePWV acting as a significant mediator in the relationship between insufficient PA and heightened CVD risk. Specifically, ePWV ≥ 10 m/s account for approximately 10.6% of the observed association between low/moderate PA levels and CVD risk. When ePWV was treated as a continuous mediator, it mediated 25.5% of the association between low/moderate PA levels and increased CVD risk.

These findings align with previous studies that have reported relationships between physical inactivity and CVD incidence20,21, as well as between ePWV and CVD occurrence22,23. However, the causal pathway by which insufficient PA contributes to increased ePWV and subsequent CVD risk has not been extensively explored from a mediation analysis perspective. By employing CMA, our study not only corroborates previous findings but also provides novel evidence that an increase in ePWV resulting from insufficient PA significantly elevates CVD risk in middle-aged and elderly populations. These findings underscore the potential value of enhancing PA as an effective strategy for preventing atherosclerosis and subsequent CVD events.

ePWV derived from age and MBP using validated formulas demonstrates good agreement with measured cfPWV, establishing its utility as a surrogate when direct cfPWV assessment is impractical5. Although the formula was initially developed in European cohorts, its applicability for risk prediction has been confirmed in Chinese populations. Analysis of 13,116 middle-aged and elderly participants in the China Health and Retirement Longitudinal Study (CHARLS) demonstrated that ePWV independently predicts all-cause mortality (adjusted HR = 2.32 per SD increase; 95% CI 2.20, 2.45)24. Multinational validation (US-HRS/UK-ELSA/China-CHARLS; n = 6458) confirmed ePWV’s consistent association with stroke outcomes across ethnicities25. ePWV exhibits predictive value and efficacy comparable to cfPWV for the composite cardiovascular endpoint, including cardiovascular death, nonfatal myocardial infarction, stroke, and hospitalization for ischemic heart disease5. However, although multiple studies have explored associations between ePWV and various diseases in Chinese populations, the quantitative correspondence between ePWV and cfPWV remains unelucidated. Future research should further evaluate the correlation between ePWV and cfPWV in Chinese cohorts and assess the predictive value of ePWV relative to cfPWV for disease outcomes.

Throughout the life course, even short-term reductions in PA can accumulate, potentially increasing susceptibility to cardiovascular risk factors, non-fatal or fatal cardiovascular events, related complications, and premature mortality26,27. Physical inactivity is known to increase the expression and activity of vascular NADPH oxidase, leading to enhanced production of reactive oxygen species (ROS). In contrast to active lifestyles, prolonged sedentary behavior promotes endothelial dysfunction and accelerates atherosclerosis28. The primary cardiovascular benefits of PA, along with the underlying mechanisms, include the creation of a healthier metabolic environment, reduction in chronic systemic inflammation, and induction of beneficial adaptations at the vascular (anti-atherosclerotic effects) and cardiac levels (e.g., myocardial regeneration and protection)29.

The relationship between PA and CVD outcomes could involve pathways through cardiovascular risk factors, including body weight, hypertension, diabetes, and lipid profiles. PA prevents hypertension by modulating sympathetic nervous system activity, renin-angiotensin system function, sodium handling, and improving endothelial function30. While tools like the FRS and the SCORE incorporate blood pressure measurements, their predictive performance remains suboptimal due to the simplification of complex risk interactions9. This limitation has spurred research into vascular biomarkers that may better account for residual risk and improve personalized CVD risk assessment. Arteries, as the primary target organs in hypertension, also exhibit increased stiffness, which can precede and predict the onset of hypertension. Notably, ePWV has shown superior predictive value for future cardiovascular events compared to traditional risk assessment tools5.

PA is widely recognized as a cornerstone for enhancing cardiovascular health, effectively reducing CVD risk through various mechanisms. Within this framework, ePWV serves as a critical mediator, providing insights into vascular aging. Vascular aging is driven by factors such as oxidative stress, chronic inflammation, and cellular senescence31, with atherosclerosis being a key consequence32,33. These pathological processes are central to CVD pathogenesis and affect multiple organ systems34. Arterial stiffness, a hallmark of vascular aging, reflects structural and functional changes in arteries that occur with advancing age. Increased arterial stiffness is strongly linked to the progression of vascular diseases and serves as a key risk factor for CVD. Elevated stiffness can exacerbate hypertension, raise pulse pressure, decrease coronary perfusion pressure, increase left ventricular afterload, and induce myocardial remodeling35. Moreover, it amplifies pulsatile flow into the microvasculature of vital organs36, which can negatively impact cerebral and cardiac functions6,37. ePWV serves as a reliable marker for arterial stiffness, reflecting vascular abnormalities that predispose individuals to CVD. The PM quantifies the extent to which PA’s cardioprotective effects are mediated through arterial stiffness pathways. For individuals engaging in moderate PA, the observed PM of 0.112 (95% CI 0.036–0.446) suggests that between 3.6% and 44.6% of PA’s total cardiovascular benefit may be mediated by improved vascular health. This finding implies that arterial stiffness may serve as a measurable biological pathway through which PA confers protection. However, it is important to note that the relative contribution of this pathway varies across different subpopulations. These results further highlight ePWV’s potential as a mechanistic biomarker for understanding the physiological impacts of exercise.

The precise mechanisms through which PA influences arterial stiffness remain incompletely understood. One plausible mechanism is that PA improves endothelial function38, as both aerobic and resistance training have been shown to reduce oxidative stress and vascular inflammation39,40. This reduction may enhance the bioavailability of nitric oxide, a critical mediator of endothelial function that affects vascular compliance and distensibility41. Regular moderate PA has also been shown to decrease inflammatory markers and oxidative stress, offering protection to the vascular endothelium, which in turn may influence ePWV. Moreover, aerobic exercise has been found to reduce sympathetic nervous system activity in individuals with cardiovascular conditions42. Reduced sympathetic activity can alleviate vascular smooth muscle tone, enhancing arterial compliance and elasticity. In individuals with CVD, excessive sympathetic activation has been shown to impair endothelial function43, and reductions in muscle sympathetic nerve activity following aerobic training could be a key mechanism through which PA improves ePWV44.

Through E-value sensitivity analysis, the point estimates for the mediational E-value (NIE pathway) ranged from 1.172 to 1.420 (95% CI lower bounds: 1.116–1.251). This suggests that an unmeasured confounder would need to have associations with both PA and arterial stiffness (ePWV) of at least RR = 1.172–1.420 to completely explain away the mediated effect through ePWV. This range approaches the typical threshold for common confounders in cardiovascular research, indicating moderate sensitivity of the mediation pathway to unmeasured confounding. In contrast, the NDE-pathway demonstrated greater robustness, with point estimates ranging from 1.641 to 3.638 (CI lower bounds: 1.000–1.943). Notably, the E-value for the low-PA group reached 3.638 (CI lower bound: 1.943), surpassing the effect size of established strong confounders, thus supporting the robustness of PA’s direct cardioprotective effects. Importantly, there is no universal threshold for E-value interpretation; contextualizing E-values against domain-specific benchmarks is essential. Here, all NDE-pathway E-values exceeded typical thresholds for cardiovascular risk factors, while the NIE-pathway values approached critical levels. This highlights the need for future validation of vascular mediation mechanisms through approaches like Mendelian randomization or dynamic ePWV monitoring.

From a clinical perspective, maintaining adequate PA is critical for preventing and delaying the progression of atherosclerosis and preserving cardiovascular health. By adulthood, early stages of CVD, particularly atherosclerosis, often begin to develop, with cardiovascular risk factors becoming more pronounced as individuals approach middle age27. The most effective treatments for managing these conditions include dietary improvements and increased PA, alongside pharmacological interventions45. Therefore, incorporating ePWV monitoring into clinical practice could offer a valuable strategy for health maintenance in individuals with insufficient PA. Ensuring ePWV remains at lower levels is crucial for preserving cardiovascular health and mitigating CVD risk. From a public health standpoint, promoting PA and reducing sedentary behavior are essential strategies for preventing atherosclerosis and its associated cardiovascular complications. Public health initiatives should prioritize these goals to reduce the burden of CVD across populations.

This study offers a novel perspective on the indirect effects of PA on CVD risk through ePWV as a mediator. By exploring this previously unexamined pathway, it contributes to a better understanding of the role of vascular health in CVD prevention. Moreover, it highlights the importance of integrating non-invasive measures like ePWV into clinical risk assessments and public health strategies.

There are several limitations to consider. First, the study did not use a nationally representative sample, so the generalizability of the findings to different racial and ethnic groups, as well as other populations, should be cautious. Second, the study did not include direct measurements of cfPWV. While ePWV is not intended to replace cfPWV or serve as an interchangeable alternative46, it can act as a useful initial screening tool for CVD risk assessment. Third, as in other large-scale studies, data on several risk factors, including PA, sedentary time, tobacco smoking, and alcohol consumption, were self-reported, which may have introduced recall and social desirability biases. Moreover, although the PM provides insight into the role of arterial stiffness in mediating the cardioprotective effects of PA, its clinical interpretation should be approached with caution. The PM estimates suggest that a substantial portion of PA’s benefit may operate through slowing arterial aging, reinforcing the potential of ePWV as a monitoring biomarker for targeted interventions. However, we observed considerable variability and wide confidence intervals in the PM estimates, reflecting inherent methodological challenges. As a ratio of the natural indirect effect to the total effect (NIE/TE), the PM amplifies statistical uncertainty, particularly when effect sizes are modest or when covariates effectively capture the mediation pathways. This limitation is common in observational mediation studies and underscores the need for future randomized controlled trials, ideally incorporating serial ePWV measurements, to further validate and refine these mechanistic pathways. Furthermore, a substantial proportion of participants had incomplete data on PA, which may have reduced the sample size and introduced biases. Finally, the low prevalence of low-PA-level participants (< 2%) reflects our cohort’s rural composition. This limits precision for low-PA effects. Future work will oversample urban populations to address this gap.

In conclusion, this 7 year prospective cohort study provides robust evidence that elevated ePWV significantly increases CVD incidence risk and may help explain the relationship between physical activity and new-onset CVD in middle-aged and older adults. While ePWV shows promise for risk stratification research, its clinical application requires confirmation through interventional studies and external validation in multi-ethnic cohorts.