Association of estimated pulse wave velocity with non-alcoholic fatty liver disease in multiple cohorts

Summary

Diagnosing non-alcoholic fatty liver disease (NAFLD) via ultrasound faces cost and accessibility challenges. We assessed whether estimated pulse wave velocity (ePWV), a measure of arterial stiffness, could serve as an accessible diagnostic marker for NAFLD. In two large cohorts (NAGALA, n = 14,251; NHANES, n = 12,767), higher ePWV independently predicted NAFLD, with top-quartile participants showing significantly higher odds (OR = 1.85, 1.40–2.45 for NAGALA; OR = 3.05, 1.67–5.59 for NHANES). A non-linear relationship was observed: Below the inflection point, each 1 m/s increase in ePWV raised NAFLD risk by 33.6% (NAGALA) and 57% (NHANES), plateauing thereafter. These findings demonstrate that ePWV is independently associated with increased NAFLD risk, supporting its potential utility for early risk stratification.

Graphical abstract

Highlights

• •ePWV independently predicts incident NAFLD, with a robust nonlinear relationship
• •ePWV is a cost-effective, non-invasive screening indicator for clinical use
• •ePWV-NAFLD link modified by age and exercise via intrinsic and behavioral factors

Health sciences

Introduction

Globally, non-alcoholic fatty liver disease (NAFLD) is the most widespread chronic liver disorder, affecting approximately one-quarter of the global population.^1,2 NAFLD progresses from steatosis to cirrhosis and is linked to cardiovascular and metabolic diseases.^3,4 Early-stage NAFLD is potentially reversible. Enhanced detection strategies combined with public-health interventions promoting healthy diets and regular physical activity may mitigate its projected disease burden.^5,6 Currently, the screening and diagnosis of NAFLD primarily depend on liver imaging (e.g., ultrasonography) or serum biomarkers (e.g., transaminases). However, the widespread application of these approaches in large-scale population screening is limited by factors such as cost, accessibility, and suboptimal sensitivity or specificity.⁷ Consequently, there is an immediate requirement to identify alternative markers that are easily accessible, economical, and mechanistically linked to core pathophysiological processes in NAFLD.

Arterial stiffness serves as an early and integrative indicator of vascular impairment and a robust predictor of cardiovascular outcomes.⁸ Estimated pulse wave velocity (ePWV) offers a practical and accessible measure of arterial stiffness based on routinely available clinical data.⁹ Unlike device-dependent methods such as carotid-femoral PWV, ePWV is cost-effective, readily standardized, and exhibits a strong correlation with directly measured values,⁸ making it a feasible tool for broader clinical use. From a pathophysiological perspective, both NAFLD and arterial stiffness share underlying mechanisms driven by insulin resistance, oxidative stress, and chronic low-grade inflammation.^10,11 This suggests that ePWV-detected vascular dysfunction may reflect systemic metabolic dysregulation and underlying hepatic steatosis. While previous studies have established a correlation between ePWV and metabolic dysfunction-associated steatotic liver disease (MASLD) in US young adults, this finding was exclusively derived from a single cohort aged under 45 years.¹² Therefore, the relationship between ePWV and NAFLD still needs to be further clarified.

To address these gaps, we performed a pooled analysis of two independent, nationwide cohorts—the NAGALA study and NHANES. This dual-cohort design provided internal and external validation, significantly strengthening the robustness and generalizability of our findings regarding this association.

Results

Study subjects and characteristics

A total of 14,251 apparently healthy adults undergoing routine health screening were included, with a median age of 42 years (interquartile range [IQR] 36–50) and 48.0% being female. Baseline characteristics stratified by ePWV quartiles are summarized in Table 1 (Q1: 4.62–6.21; Q2: 6.22–6.87; Q3: 6.88–7.79; Q4: 7.80–16.22). The prevalence of NAFLD progressively increased across ascending ePWV quartiles (Q1: 5.23%, Q2: 19.27%, Q3: 33.50%, Q4: 40.00%). Higher ePWV quartiles were associated with significantly higher body mass index (BMI), waist circumference (WC), liver enzyme levels, glycemic parameters, blood pressure (BP), and all lipid measures except high-density lipoprotein cholesterol (HDL-C). These groups also showed higher proportions of current smokers and alcohol consumers. Moreover, there was a progressive rise in male participants across ePWV quartiles, accompanied by a corresponding decline in female participants. Table S1 displays the initial characteristics of all participants based on ePWV quartiles in the NHANES cohort. Among 12,767 participants (mean age 54.0 ± 17.8 years; 50.4% male), 5,733 (44.9%) were identified with NAFLD. Higher ePWV quartiles were associated with older age, increased adiposity (higher BMI and WC), poorer cardiometabolic profiles (elevated BP, glucose, hemoglobin A1c [HbA1c], triglycerides, total cholesterol, gamma-glutamyl transferase [GGT], and fatty liver index [FLI], with lower HDL-C), and reduced physical activity. These groups also showed distinct behavioral patterns: a lower proportion of never-drinkers, a higher share of former drinkers, and less frequent current smoking. The increase in ePWV was linked to a significant rise in NAFLD prevalence, with the third quartile having the highest occurrence (Table S1).

Table 1.

Baseline characteristics of four groups

Variables	Total	ePWV quartile	–	–	–	p value
–	–	Q1 (4.62–6.21)	Q2 (6.22–6.87)	Q3 (6.88–7.79)	Q4 (7.80–16.22)	–
No.	14251	3506	3587	3590	3568	–
Age, years	43.5 ± 8.9	36.9 ± 5.5	39.7 ± 6.0	44.1 ± 6.7	53.2 ± 7.3	<0.001
Sex	–	–	–	–	–	<0.001
Female	6840 (48.0)	2523 (72)	1646 (45.9)	1336 (37.2)	1335 (37.4)	–
Male	7411 (52.0)	983 (28)	1941 (54.1)	2254 (62.8)	2233 (62.6)	–
BMI, kg/m2	22.1 ± 3.1	20.3 ± 2.3	21.8 ± 2.8	22.9 ± 3.2	23.3 ± 3.2	<0.001
WC, cm	76.2 ± 9.1	70.5 ± 6.9	75.3 ± 8.2	78.7 ± 8.8	80.2 ± 9.1	<0.001
ALT, U/L	16.0 (12.0, 23.0)	14.0 (11.0, 18.0)	16.0 (12.0, 22.0)	18.0 (14.0, 26.0)	19.0 (14.0, 25.0)	<0.001
AST, U/L	18.2 ± 8.7	16.1 ± 6.0	17.7 ± 11.8	19.2 ± 8.0	19.8 ± 7.3	<0.001
GGT, U/L	15.0 (11.0, 21.0)	12.0 (10.0, 15.0)	14.0 (11.0, 20.0)	16.0 (12.0, 24.0)	17.0 (13.0, 26.0)	<0.001
HDL-C, mmol/L	1.5 ± 0.4	1.6 ± 0.4	1.5 ± 0.4	1.4 ± 0.4	1.4 ± 0.4	<0.001
TC, mmol/L	5.1 ± 0.9	4.8 ± 0.8	5.0 ± 0.8	5.3 ± 0.8	5.5 ± 0.9	<0.001
TG, mmol/L	0.7 (0.5, 1.1)	0.5 (0.4, 0.7)	0.7 (0.5, 1.0)	0.8 (0.6, 1.2)	0.9 (0.7, 1.4)	<0.001
FPG, mmol/L	5.1 ± 0.4	4.9 ± 0.4	5.1 ± 0.4	5.2 ± 0.4	5.3 ± 0.4	<0.001
HbA1c, %	5.2 ± 0.3	5.1 ± 0.3	5.1 ± 0.3	5.2 ± 0.3	5.3 ± 0.3	<0.001
SBP, mmHg	113.9 ± 14.8	98.8 ± 7.3	110.4 ± 7.8	118.1 ± 10.2	128.1 ± 14.3	<0.001
DBP, mmHg	71.1 ± 10.4	60.0 ± 5.1	68.5 ± 5.0	74.4 ± 6.7	81.4 ± 9.4	<0.001
Exercise habits	2470 (17.3)	524 (14.9)	568 (15.8)	588 (16.4)	790 (22.1)	<0.001
Drinking status	–	–	–	–	–	<0.001
No or small	11805 (82.8)	3175 (90.6)	3023 (84.3)	2900 (80.8)	2707 (75.9)	–
Light	1758 (12.3)	273 (7.8)	429 (12)	492 (13.7)	564 (15.8)	–
Moderate	688 (4.8)	58 (1.7)	135 (3.8)	198 (5.5)	297 (8.3)	–
Smoking status	–	–	–	–	–	<0.001
No	8746 (61.4)	2535 (72.3)	2222 (61.9)	2026 (56.4)	1963 (55)	–
Past	2559 (18.0)	386 (11)	565 (15.8)	703 (19.6)	905 (25.4)	–
Current	2946 (20.7)	585 (16.7)	800 (22.3)	861 (24)	700 (19.6)	–
NAFLD	2507 (17.6)	131 (5.23)	483 (19.27)	890 (33.50)	1003 (40.00)	<0.001

Results were presented as mean (SD), medians (interquartile range), or n (%).

ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; DBP, diastolic blood pressure; ePWV, estimated pulse wave velocity; FPG, fasting plasma glucose; GGT, gamma-glutamyl transferase; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; NAFLD, non-alcoholic fatty liver disease; Q1–Q4, quartile 1–4; SBP, systolic blood pressure; TC, total cholesterol; TG, triglyceride; WC, waist circumference.

Multivariate analysis of the ePWV-NAFLD association

Analyses employing multivariable logistic methods demonstrated a stable and substantial positive relationship between ePWV and NAFLD (Table 2). Although sequential covariate adjustment attenuated the magnitude of the association, it remained statistically significant across higher ePWV strata. After full adjustment for all included covariates, every 1 m/s rise in ePWV corresponded to an 11% increased likelihood of NAFLD (Model 3). When examined by ePWV quartiles, a clear gradient in NAFLD risk was observed: Relative to the first quartile (Q1), the adjusted odds ratios (ORs) were 1.43 for Q2 (95% confidence intervals [CIs]: 1.13–1.82), 1.83 (95% CI: 1.44–2.32) for Q3, and 1.85 (95% CI: 1.40–2.45) for Q4. All of these results are statistically significant. In the NHANES sample, a consistent positive correlation was found between increased ePWV and the likelihood of NAFLD. One-unit rise in ePWV correlated with an unadjusted OR of 1.07 (95% CI: 1.06–1.09, p < 0.001), which remained significant in fully adjusted models (Model 3: OR 1.12, 95% CI: 1.02–1.22, p = 0.017). In comparison to the lowest quartile (Q1), individuals in the higher ePWV quartiles (Q2–Q4) had progressively higher adjusted ORs, between 1.99 and 3.08 (all p < 0.005), and statistical analysis revealed a significant dose-response trend across quartiles (P for trend <0.05) (Table S2).

Table 2.

Investigating the association between ePWV and NAFLD using logistic regression

Variable	Unadjusted model		Model 1		Model 2		Model 3
	OR (95% CI)	p value	OR (95% CI)	p value	OR (95% CI)	p value	OR (95% CI)	p value
ePWV	1.54 (1.49–1.59)	<0.001	1.15 (1.07–1.23)	<0.001	1.16 (1.09–1.24)	<0.001	1.11 (1.03–1.20)	0.004
ePWV(Quartile)	–	–	–	–	–	–	–	–
Quartile 1	1(Ref)	–	1(Ref)	–	1(Ref)	–	1(Ref)	–
Quartile 2	4.01 (3.29–4.89)	<0.001	1.65 (1.32–2.05)	<0.001	1.64 (1.31–2.04)	<0.001	1.43 (1.13–1.82)	0.003
Quartile 3	8.49 (7.02–10.27)	<0.001	2.26 (1.81–2.81)	<0.001	2.27 (1.82–2.83)	<0.001	1.83 (1.44–2.32)	<0.001
Quartile 4	10.07 (8.34–12.17)	<0.001	2.33 (1.81–3.00)	<0.001	2.41 (1.87–3.12)	<0.001	1.85 (1.40–2.45)	<0.001
Trend.test	–	<0.001	–	<0.001	–	<0.001	–	<0.001

Model 1 adjusted for age, gender, BMI, and WC.

Model 2 adjusted for age, gender, BMI, WC, exercise habits, smoking, and drinking status.

Model 3 adjusted for age, gender, BMI, WC, exercise habits, smoking and drinking status, ALT, AST, GGT, TG, HDL-C, TC, FPG, and HbA1c.

Nonlinear relationship of ePWV with NAFLD risk

The nonlinear association of ePWV with NAFLD risk was visualized using an restricted cubic spline (RCS) model (Figure 1). In the NAGALA cohort, a significant nonlinear, inverted L-shaped association was observed (P for nonlinearity <0.001), with an inflection point at 7.018 m/s determined by threshold regression. Below this point, each 1 m/s increase in ePWV was associated with 33.6% higher odds of NAFLD (OR = 1.336, 95% CI: 1.012–1.763, p = 0.04; Table 3); above it, the association plateaued. A similar nonlinear pattern was replicated in the NHANES cohort (P for nonlinearity <0.001), where the inflection occurred at 8.602 m/s. Below this threshold, a 57% higher probability was observed for each 1 m/s increment (OR = 1.57, 95% CI: 1.22–2.02, p < 0.001). The effect became less strong and was no longer significant when it was measured above it (OR = 1.13, 95% CI: 0.99–1.29, p = 0.062) (Figure S1; Table S3).

Figure 1.

Dose-response curves illustrate the connection between ePWV and NAFLD in the general public

Table 3.

Exploration of the threshold effect in the link between ePWV and NAFLD

ePWV	Adjusted model
	OR(95% CI)	p value
<7.018	1.336 (1.012–1.763)	0.04
≥7.018	1.082 (0.983–1.192)	0.11
Log likelihood ratio test	–	<0.001

OR, odds ratio; CI, confidence interval; ePWV, estimated pulse wave velocity; NAFLD, non-alcoholic fatty liver disease.

Adjusted for adjusted for age, sex, BMI, WC, habits of exercise, smoking status, drinking status, ALT, AST, GGT, TG, HDL-C, TC, FPG, and HbA1c.

Sensitivity analysis

Table S4 illustrates how ePWV is associated with NAFLD across various moderate-risk demographic subgroups in the NAGALA cohort. Sensitivity analyses performed across three distinct populations, with full adjustment for Model 3 covariates, produced results comparable with the main analysis, thereby strengthening the study’s findings.

Subgroup and interaction analyses

Analyses of subgroups and interactions were performed to evaluate the consistency of the ePWV-NAFLD relationship across important demographic and clinical categories. In the NAGALA cohort, the effects were significantly modified by age and physical activity level (P for interaction <0.05), but not by gender, BMI, smoking status, or alcohol consumption (P for interaction >0.05) (Table 4). In the NHANES cohort, the association was significantly modified by age (P for interaction = 0.008), with the strongest effect observed in adults aged 18–44 years (OR = 1.57, 95% CI: 1.15–2.14), and by BMI (P for interaction = 0.007), although subgroup estimates were imprecise. Current smokers exhibited a pronounced association (OR = 1.85, 95% CI: 1.35–2.54, p < 0.001), while no significant effect modification was detected for sex or physical activity (Table S5).

Table 4.

Analysis of subgroups regarding the link between ePWV and NAFLD in the NAGALA cohort

Subgroup	Adjusted OR (95% CI)	p value	P interaction
Age (years)
18–44	1.42 (1.24, 1.63)	<0.001	<0.001
45–59	1.12 (1.01, 1.23)	0.032	–
≥60	0.90 (0.65, 1.25)	0.539	–
Sex
Women	1.01 (0.88, 1.16)	0.866	0.396
Men	1.14 (1.04, 1.25)	0.004	–
BMI (kg/m2)
<25	1.11 (1.01, 1.21)	0.030	0.247
≥25	1.14 (1.00, 1.29)	0.048	–
Habits of exercise
No	1.15 (1.06, 1.25)	0.001	0.041
Yes	0.95 (0.79, 1.14)	0.581	–
Drinking status
No	1.13 (1.02, 1.25)	0.024	0.593
Light	1.18 (1.01, 1.37)	0.033	–
Moderate	1.01 (0.86, 1.18)	0.913	–
Smoking status
No	1.06 (0.98, 1.16)	0.153	0.334
Former	1.37 (1.11, 1.69)	0.003	–
Current	1.30 (0.97, 1.73)	0.080	–

BMI, body mass index; CI, confidence interval; OR, odds ratios.

Adjusted for age, gender, BMI, WC, exercise habits, smoking and drinking status, ALT, AST, GGT, TG, HDL-C, TC, FPG, and HbA1c.

Mediating effect analysis

In two independent cohorts, mediation analysis identified the triglyceride-glucose (TyG) index—calculated as Ln[fasting triglycerides (mg/dL) × fasting glucose (mg/dL)/2]—as a partial mediator of the ePWV-NAFLD association. In the NAGALA cohort (Figure S2A), the indirect effect via TyG was 0.0022, corresponding to a mediated proportion of 27.85% (p < 0.001; total effect = 0.0058), indicating that TyG explained approximately one-quarter of the observed association. In the NHANES cohort, the indirect effect was similarly 0.0022, with a larger mediated proportion of 33.67% (p = 0.004; total effect = 0.0066), suggesting roughly one-third mediation. A substantial direct effect persisted in both cohorts, consistent with partial mediation; thus, TyG contributes meaningfully but does not fully account for the ePWV-NAFLD link (Figure S2B).

Direct model comparisons between ePWV and its constituent variables

In the NAGALA cohort, the area under the curve (AUC) was 0.855 (95% CI: 0.847–0.863). In the NHANES cohort, the AUC was 0.728 (95% CI: 0.719–0.737). In addition, we evaluated the predictive performance of ePWV against three baseline models (age alone, BP alone, and the age-BP combined model) using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). In the NAGALA cohort, ePWV alone showed significantly better reclassification and discrimination than age alone (NRI = 0.3474, IDI = 0.0105; both p < 0.001), BP alone (NRI = 0.2950, IDI = 0.0032; both p < 0.001), and the age-BP combined model (NRI = 0.2925, IDI = 0.0022; both p < 0.05). In the NHANES cohort, ePWV alone outperformed age alone (NRI = 0.0512, p = 0.003; IDI = 0.0003, p = 0.027) and BP alone (NRI = 0.1668, p < 0.001); however, the IDI for the comparison with BP alone was slightly negative (−0.0009, p = 0.016). When compared with the age-BP combined model, ePWV alone yielded a positive NRI (0.0572, p < 0.001) but a negative IDI (−0.0012, p < 0.001), although both IDI values were extremely close to zero. Collectively, these findings indicate that ePWV consistently outperforms age and BP individually, but provides little to no discriminative improvement beyond their combination, particularly in the NHANES cohort (Figure S3; Table S6). The flowchart detailing the step-by-step selection process of study participants is presented in Figure 2.

Figure 2.

Study participant selection flowchart

Discussion

Although arterial stiffness is a recognized metabolic risk factor, its association with NAFLD via ePWV remains incompletely understood. This first multi-cohort study demonstrates that higher ePWV independently predicts increased NAFLD risk with a robust nonlinear dose-response relationship, consistent across sensitivity and subgroup analyses. While the use of FLI in NHANES may introduce circular associations due to overlap with metabolic variables, validation in the NAGALA cohort using ultrasound-based NAFLD diagnosis supports the robustness of the findings, though residual confounding cannot be fully excluded. Taken together, ePWV may serve as a practical, non-invasive tool to aid early NAFLD detection in clinical and public health practice.

Increased arterial stiffness acts both as a sensitive early marker and a significant contributing factor to a range of vascular disorders.¹³ It reflects and drives pathological processes such as atherosclerotic plaque formation, vascular calcification, chronic low-grade inflammation, and the progressive functional decline associated with vascular aging, thereby representing a critical common pathway in systemic vascular disease.^14,15 Substantial evidence links increased arterial stiffness to a spectrum of cardiometabolic and organ-specific diseases. This association extends beyond traditional risk factors, suggesting that arterial stiffness may represent a shared pathophysiological pathway contributing to the pathogenesis of heart failure, diabetic complications, renal impairment, stroke, and dementia.^16,17,18,19 Building upon this pathophysiological framework, various research efforts have explored the connection between ePWV and MASLD outcomes among American populations, highlighting the potential clinical relevance of this marker in relation to liver disease. Within the U.S. population, ePWV has been associated with MASLD both in terms of disease risk and prognosis. Research conducted by Li and colleagues discovered a connection between the occurrence of MASLD and ePWV, noting that the link intensifies with higher doses.¹² Beyond prevalence, elevated ePWV has also been identified as a prognostic marker for mortality from all causes and cardiovascular disease in patients suffering from MASLD.²⁰ Our study bridges this knowledge gap by demonstrating through a combined analysis of two independent cohorts, that higher ePWV is prospectively associated with an increased risk of developing NAFLD. This result is consistent with the pathophysiological premise linking arterial stiffness to metabolic disorders and provides the needed longitudinal validation that strengthens the evidence base for ePWV as a relevant marker in NAFLD risk stratification.

Notably, the correlation between ePWV and NAFLD followed a non-linear, “rising-plateau” curve, which was consistently observed across both cohorts. This non-linear pattern may be understood through a biphasic physiological model. At moderate levels of arterial stiffening, initial increases in ePWV might reflect a compensatory hemodynamic adaptation aimed at preserving vital organ perfusion, thereby partially mitigating cardiovascular risk.²¹ However, once arterial stiffness surpasses a critical threshold, the cumulative detrimental effects—including significantly impaired vascular compliance, sustained elevation in systolic pressure, and compromised coronary perfusion—likely overwhelm these adaptive mechanisms.²² This transition from compensation to decompensation may explain the sharp rise in mortality risk beyond this inflection point. Similar threshold phenomena have been reported in other populations,²³ highlighting that the correlation between arterial stiffness and clinical outcomes is modulated by variables such as age, comorbidities, and underlying vascular health. Additionally, while threshold analysis identified an inflection point, this value was empirically derived from data-driven model fitting rather than being prespecified based on known physiological transitions in arterial stiffness. The biological and clinical relevance of this specific threshold remains unclear, and it is not possible to determine whether it reflects a true physiological inflection point or an artifact of model specification. Although the nonlinear relationship was consistently observed across two independent cohorts, the precise threshold value warrants further validation in diverse populations before it can be considered clinically actionable. Therefore, these findings should be interpreted as hypothesis-generating rather than definitive.

The association between ePWV and NAFLD risk exhibited significant heterogeneity across lifestyle and demographic subgroups. Notably, in the NAGALA cohort, this relationship was modified by exercise habits, potentially reflecting the effectiveness of exercise intervention in reducing arterial stiffness in young men and improving insulin sensitivity.^24,25 In addition, the more pronounced association in younger participants may reflect a lower baseline burden of metabolic comorbidities, allowing the relative contribution of ePWV to be more readily discerned compared with older individuals, in whom cumulative exposure to multiple risk factors may attenuate the relative impact of any single marker. Regarding smoking status, the stronger association observed among former smokers may reflect the cumulative burden of prior smoking exposure, which is known to independently contribute to both arterial stiffness and hepatic steatosis. This finding may also be explained by synergistic biological interactions or residual confounding from unmeasured factors such as smoking intensity, duration, or other lifestyle variables. Given the exploratory nature of subgroup analyses, the potential for multiple testing, and the inherent limitations of residual confounding, these findings should be interpreted as hypothesis-generating rather than definitive.

The detailed processes that link NAFLD to arterial stiffness are not fully comprehended, but there are several possible explanations. Central to this link is insulin resistance, which acts as a shared driver, promoting both hepatic lipid accumulation and impaired arterial elasticity.²⁶ Mediation analysis identified TyG as a partial mediator of the ePWV-NAFLD association in both cohorts, suggesting that insulin resistance partly explains the observed link, which is consistent with previous studies. A key downstream effect is the activation of the renin-angiotensin-aldosterone system by chronic hyperglycemia and hyperinsulinemia.²⁷ This activation stimulates vascular smooth muscle cell proliferation and enhances collagen deposition in the arterial wall, processes that culminate in vascular remodeling, fibrosis, and increased stiffness.^28,29 Second, NAFLD-associated inflammation and oxidative stress directly impair vascular homeostasis, linking liver disease to arterial stiffness.^30,31 Systemic oxidative stress exacerbates endothelial dysfunction, stimulates vascular smooth muscle proliferation and migration, and degrades elastic fibers, culminating in irreversible arterial remodeling and increased stiffness.^32,33 Third, the connection between NAFLD and arterial stiffness might be directly linked to dysfunction in endocrine functions derived from adipose tissue.³⁴ Adiponectin deficiency impairs its anti-inflammatory and insulin-sensitizing actions, promoting vascular inflammation, fibrosis, and arterial stiffening.^35,36

Furthermore, our study directly compared ePWV against three alternative models: age alone, BP alone, and their combination. The seemingly contradictory findings between the two cohorts collectively underscore the value of ePWV as an integrated indicator. In the NAGALA cohort, ePWV outperformed all three models, demonstrating its potential as a superior composite metric. In the NHANES cohort, although ePWV did not exceed BP alone or the age-BP combined model, it performed comparably to these variables. Several factors may explain these discrepancies. First, the definition of NAFLD differed substantially: NAGALA used ultrasound, whereas NHANES relied on the FLI, a surrogate measure subject to misclassification that may have attenuated ePWV’s performance. Second, demographic characteristics varied markedly, including differences in age distribution, baseline risk profile, and comorbidity burden. In two substantial population-based samples from the NAGALA and NHANES cohorts, this research shows that a higher ePWV is closely linked to a greater risk of developing NAFLD, suggesting that ePWV may serve as a potential predictor for NAFLD. These findings further imply that interventions aimed at reducing arterial stiffness could offer meaningful benefits for the prevention and management of NAFLD, particularly among young and middle-aged adults.

Limitations of the study

This study has several limitations. First, the cross-sectional design precludes causal inference regarding the relationship between ePWV and NAFLD. Future studies employing longitudinal designs or Mendelian randomization approaches are warranted to elucidate the causal direction. Second, despite comprehensive adjustment, the possibility of residual confounding—an inherent limitation of observational studies—cannot be fully excluded. Key factors, including dietary habits, socioeconomic status, and genetic variants, were not available in the current datasets, and their potential impact on the observed associations should be considered. Future studies incorporating negative control outcomes or employing instrumental variable analyses may help further assess robustness against unmeasured confounding. Third, it is important to note that ePWV is not a direct physiological measurement of arterial stiffness but rather an estimated parameter derived primarily from age and mean arterial pressure. Both of these constituent variables are well-established risk factors for NAFLD; thus, the observed association between ePWV and NAFLD may partially reflect the influence of age and blood pressure rather than an independent relationship with arterial stiffness. We selected ePWV for its pragmatic value as a cost-effective, readily available screening tool that can be implemented in settings where direct arterial stiffness measurement is not feasible. Finally, the severity of NAFLD could not be effectively differentiated.

Resource availability

Lead contact

Further information and resource requests should be directed to the lead contact, Xin-yang Zhang (949995161@qq.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •The data used in this study are accessible to researchers upon application. Requests for data access should be made directly to the NAGALA and NHANES.
• •This article does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

None.

Author contributions

J.-j.J.: Conceptualization, methodology, formal analysis, software, data curation, writing – original draft, supervision, and project administration. Y.Y.: investigation, data curation, writing – original draft, and writing – review and editing. X.-y.Z.: conceptualization, methodology, formal analysis, software, writing – original draft, writing – review and editing, supervision, and project administration. Y.L.: methodology, data curation, and writing – review and editing. F.Y.: writing – original draft and writing – review and editing. Y.-l.Q.: writing – original draft and writing – review and editing.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Depositeddata
NAGALA cohort	Okamura et al.2019; Dryad	https://doi.org/10.5061/dryad.8q0p192
NHANES 1999-2020	National Health and Nutrition Examination Survey	https://www.cdc.gov/nchs/nhanes/
Softwareandalgorithms
R (Version 4.2.3)	RStudio	http://www.R-project.org
Free Statistics software (Version 2.3)	Beijing FreeClinical Medical Technology Co., Ltd, Beijing, China	http://www.clinicalscientists.cn/freestatistics

Experimental model and study participant details

Human study participants

These investigations analyzed information from two independent cohorts, namely the NAGALA and the NHANES. In NAGALA cohort, comprehensive information on the study protocol and objectives of NAGALA has been released previously.³⁷ This ongoing cohort has enrolled individuals undergoing health examinations at Murakami Memorial Hospital since 1994. The study systematically collects and analyzes participant data to support early identification of chronic diseases and their risk factors, with significant public health relevance. It also provides valuable evidence for informing chronic disease prevention strategies and clinical management. The de-identified dataset is publicly available on Dryad and can be employed for secondary analyses under the specified access conditions.³⁷

Data from 20,944 participants enrolled in the NAGALA cohort before 2016 were obtained from the Dryad repository. Based on the study objectives, we excluded individuals meeting any of the following criteria: (1) diagnosed viral or alcoholic hepatitis, diabetes, or fasting plasma glucose >6.1 mmol/L at enrollment (n = 1,547); (2) baseline alcohol consumption ≥210 g/week for men or ≥140 g/week for female (n = 1,952); (3) use of any medication at enrollment (n = 2,321); (4) incomplete baseline data (n = 863); or (5) withdrawal from the study (n = 10). After excluding certain participants, 14,251 individuals were part of the final analysis, which is summarized in Figure 2A.

NAFLD was diagnosed by abdominal ultrasonography. To minimize bias, image acquisition and interpretation were performed independently: a certified technician performed the scan, and a gastroenterologist blinded to clinical data reviewed the images. Diagnosis followed standard ultrasonographic criteria for hepatic steatosis, including increased liver echogenicity, hepatorenal contrast, vascular blurring, and deep attenuation.³⁸

For our analysis, we relied on data from the NHANES spanning 1999 to 2020. The original sample consisted of 58,744 people aged 20 and over. Participants were sequentially excluded for the following reasons: missing FLI data (n = 34,719); excessive alcohol intake, which is more than one drink per day for women or more than two drinks per day for men (n = 7,147), and a positive test for hepatitis B surface antigen or hepatitis C antibody (n = 1,973); missing ePWVdata (n = 632); pregnancy (n = 398); and missing covariate data (n = 1,108). After exclusions, the final analytical cohort included 12,767 participants with NAFLD, as shown in Figure 2B.

Hepatic steatosis was assessed using the FLI, a validated non-invasive scoring system calculated from four anthropometric and biochemical parameters: BMI, WC, serum triglyceride level, and GGT activity. Based on its original validation, an FLI score ≥60 was applied as the threshold to identify probable NAFLD, following the exclusion of other etiologies of liver disease (including clinically significant alcohol consumption).

FLI is calculated using the formula below³⁹:

$$FLI=\left(e^{ˆ}L\left(1+e^{ˆ}L\right)\right)\times 100,whereL=0.953\times \mathit{ln}\left(triglycerides\left[mg/dL\right]\right)+0.139\times BMI\left[kg/m^2\right]+0.718\times \ln \left(\gamma -glutamyltransferase\left[U/L\right]\right)+0.053\times waistcircumference\left[cm\right]-15.745$$ (Equation 1)

Method details

Data collection and assessment

In the NAGALA cohort, following established protocols,³⁷ trained healthcare personnel conducted standardized anthropometric measurements, including height, weight, WC, systolic blood pressure (SBP), and diastolic blood pressure (DBP). Data on age, sex, alcohol consumption, smoking status, and physical activity were collected using structured questionnaires. The classification of alcohol consumption was none or minimal (fewer than 140 g per week for women and fewer than 210 g per week for men), light, or moderate. Smoking history was classified as never, former, or current smoker. To be classified as regular exercise, one must partake in physical activity at least once per week. Following an overnight fast of 8 h, venous blood samples were gathered. Serum biochemical indices, including aspartate aminotransferase (AST), alanine aminotransferase (ALT), GGT, fasting plasma glucose (FPG), HbA1c, triglycerides (TG), total cholesterol, and HDL-C, were assayed on an automated biochemistry analyzer.

In the NHANES cohort, the covariates comprised age, gender, BMI, WC, smoking status, drinking status, physical activity and laboratory data. Age and WC were treated as continuous variables, while gender and BMI were categorical. BMI, calculated as weight/height² (kg/m²), was classified as normal (<25 kg/m²) or overweight (≥25 kg/m²). Smoking status was categorized as never (lifetime exposure <100 cigarettes), former (≥100 cigarettes, currently abstinent), or current (≥100 cigarettes, presently smoking). Alcohol consumption was categorized as: never (lifetime intake <12 drinks), light (average ≤1 drink/day for women or ≤2 drinks/day for men), or current (intake ≥12 times in the past year, or lifetime intake ≥12 drinks with no past-year consumption). Individuals were considered inactive if they engaged in less than 150 min of moderate-intensity physical activity per week, while those who did at least 150 min were considered active. Laboratory measurements comprised the liver enzymes ALT, AST, GGT, TC, TG, HDL-C, FPG and HbA1c.

Definition of ePWV

The formula was applied to determine ePWV in both groups. ePWV was derived using the following formula⁴⁰:

$$ePWV=9.587-0.402\times age+4.560\times 10-3\times {age}^2-2.621\times 10-5\times {age}^2\times MBP+3.176\times 10-3\times age\times MBP-1.832\times 10-2\times MBP$$ (Equation 2)

MBP was determined by applying the following formula:

$$MBP=DBP+0.4\times (SBP-DBP)$$ (Equation 3)

Quantification and statistical analysis

Continuous variables are presented as mean ± SD or median (IQR), based on distribution assessed via Q-Q plots. Group comparisons used one-way ANOVA (normal) or Kruskal-Wallis test (non-normal). Categorical variables are expressed as n (%) and compared using the chi-square test. The association between ePWV and NAFLD was assessed using multivariable logistic regression, with results reported as ORs and 95% CIs. Potential multicollinearity was evaluated prior to analysis using variance inflation factors. Three progressively adjusted models were fitted: Model 1 adjusted for age, sex, BMI, and waist circumference; Model 2 additionally included lifestyle factors (physical activity, smoking, and alcohol consumption); and Model 3 further adjusted for liver enzymes and metabolic parameters.

To evaluate the pattern of the dose-response curve and visualize potential non-linearity, RCS models were fitted, adjusting for all covariates included in Model 3. To evaluate potential threshold effects, a two-segment (piecewise) logistic regression model with smoothing was applied using the same adjustment set. Inflection points were identified via likelihood-ratio tests and validated through bootstrap resampling (1,000 repetitions). To examine the strength of the ePWV-NAFLD correlation, three sensitivity analyses were carried out using the fully adjusted Model 3 covariates in distinct subgroups: (1) young and middle-aged participants, (2) individuals with normal blood pressure, and (3) non-obese subjects. Moreover, stratified analyses were conducted by age, sex, BMI, physical activity, smoking status, and alcohol consumption. Likelihood-ratio tests were used to compare models across strata. Additionally, mediation analysis was performed using nonparametric bootstrapping (n = 1000) to examine both direct and indirect relationships and to quantify the magnitude of the mediating effects. To determine whether ePWV adds predictive value beyond its constituent variables, we performed pairwise model comparisons in each cohort: age alone versus ePWV alone, BP alone versus ePWV alone, and age plus BP combined versus ePWV alone. Model discrimination was assessed using the AUC with DeLong tests. Incremental reclassification was evaluated using the continuous NRI and the IDI. All analyses were conducted with R (4.2.3) and Free Software Foundation statistics software (version 2.3).^41,42 Statistical significance was determined by a two-sided p value of less than 0.05.

The exact sample size (n) for each analysis is provided in the Results section and in Figure 2. In all cases, n represents the number of individual participants. Statistical tests were not blinded, as the analyses were performed on de-identified, pre-existing cohort data. No data were excluded from the primary analyses. Sample size was determined by the available data from the NAGALA and NHANES cohorts; no a priori power calculation was performed.