We identified 96,963 adults aged 18–90 years who underwent routine health examinations at the Third Xiangya Hospital of Central South University in Changsha between August 2017 and July 2021. The population consisted of a mix of urban and rural residents. After excluding participants with incomplete information, prior cardiovascular disease, and malignancy and those not meeting the definition of NAFLD, 27,738 NAFLD subjects remained for the cross-sectional study 1 analyses of the risk associated with arterial stiffness. We further excluded participants with no carotid vascular ultrasonography, leaving a final sample of 14,323 NAFLD participants for the cross-sectional study 2 analyses of the risk associated with carotid plaque (Figure 1). Informed consent and the protocol of the overall physical examination were reviewed and approved by the institutional review board at the Third Xiangya Hospital (No. 2018-S393).

All participants completed a computerized National Physical Examination Questionnaire (10, 11). Personal details (demographic characteristics, health-related habits, family history, current medication information from pill bottles, previous medical diagnoses, etc.) were recorded according to standard protocols. The assessment and definitions of lifestyle factors, hypertension and diabetes are detailed in the Online Appendix.

The detailed methods used for physical examinations and laboratory evaluations have been previously described (11). Briefly, body weight, height, waist circumference (WC), heart rate and blood pressure were measured. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²).

After the participants fasted overnight for at least 8 h, blood samples were collected and immediately processed and analyzed for alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin, total bilirubin, uric acid, creatinine, glucose, lipids and platelet count at the clinical laboratory of the Third Xiangya Hospital. The sample analysis was performed in accordance with the manufacturer’s specifications. FIB-4 was used as a noninvasive and accurate marker of fibrosis based on the following formula: FIB-4 index = age × AST(U/L)/platelet count (×10⁹/L) √ ALT U / L (12). The estimated glomerular filtration rate (eGFR) was used as an index of renal disease based on the modification of diet in renal disease formula for Chinese subjects: eGFR = 175 × Scr^−1.234 × age^−0.179 [if female, ×0.79] (13).

NAFLD was defined as the presence of hepatic steatosis without excessive alcohol consumption (≥30 g/d in men and ≥20 g/d in women) or concomitant liver disease (14). The method used for detecting hepatic steatosis was hepatic ultrasound (Logiq 9, GE Medical System, Milwaukee, WI, United States). Positive abdominal ultrasound findings included the following 5 criteria: (1) parenchymal brightness, (2) liver-to-kidney contrast, (3) deep beam attenuation, (4) bright vessel walls and (5) gallbladder wall definition (15). Subjects with at least two abnormal findings were diagnosed with hepatic steatosis (16).

In accordance with the Working Group on Obesity in China (17), BMIs of 18.5–23.9 kg/m² were defined as normal weight; BMIs of 24.0–27.9 kg/m² were defined as overweight and ≥28.0 kg/m² were defined as obesity. The MetS definition in the National Cholesterol Education Program-Adult Treatment Panel III International Diabetes Federation (IDF) criteria (18) included elevated triglyceride (TG) levels (≥1.69 mmol/L) or the use of lipid-lowering drugs, low high-density lipoprotein (HDL) cholesterol levels (<1.03 mmol/L in men and <1.29 mmol/L in women), elevated systolic blood pressure (≥130 mmHg) or diastolic blood pressure (≥85 mmHg) or the use of antihypertensive drugs, and elevated blood glucose levels (≥5.6 mmol/L) or the use of any medications for diabetes (insulin or oral glucose-lowering medications). WC was not included because of collinearity with BMI. Individuals with one or fewer of these components were deemed metabolically healthy (MH), with two or more being deemed metabolically unhealthy (MU) according to the Framingham Heart Study (19).

Based on the combination of BMI categories and metabolic health status, the clinical phenotypes of NAFLD were then categorized into 6 groups: normal weight-MH, normal weight-MU, overweight-MH, overweight-MU, obese-MH and obese-MU.

As previously reported (11), baPWV was measured with an automatic waveform analyzer (BP-203 RPE III, Omron Health Medical, Dalian, China). After a minimum rest of 5 min in the supine position, 4 cuffs were wrapped around the extremities (upper arms and ankles) and then connected to the plethysmography sensor (volume pulse form) and an oscillometric pressure sensor. Pressure waveforms were recorded at both the brachial and tibial arteries to assess the transmission time between the initial increases in these waves. The measurements were performed twice, and the mean of the left- and right-side baPWV values were calculated. Moreover, substantial side differences in the baPWV of more than 10 m/s indicated problems with measurement, and the measurement should be repeated. The highest baPWV quartile (>1,590 cm/s) was classified as increased baPWV (arterial stiffness) (20).

As previously reported (21), carotid plaque was assessed using carotid artery sonography (Siemens AcusonSequoiaTM512 Ultrasound System, Mountain View, CA, United States) with a 9 MHz linear array transducer. Experienced sonographers performed carotid examinations, including bilateral visualization of the common, internal, and external carotid arteries. Carotid plaque was defined as a focal wall thickening of at least 0.5 mm or 50% of the surrounding CIMT that encroached into the arterial lumen or a focal region with CIMT greater than 1.5 mm that protruded into any carotid segment (22).

Descriptive characteristics are presented as the mean ± SD or median (interquartile range) for continuous variables and as the number (percentage) for categorical variables. The covariates between different groups were compared using a t test or the Mann–Whitney U test for continuous variables and the chi-square test for categorical variables. Multivariate logistic regression models were performed to assess the association of baPWV (the highest quartile versus other quartiles) and carotid plaques (presence versus absence) with (1) the 6 clinical phenotypes above; (2) the 3 BMI categories (normal weight, overweight and obese); (3) the 4 different categories of metabolic abnormalities (0, 1, 2 and ≥3 indices); and (4) individual metabolic abnormality risk factors (elevated blood pressure, triglyceridemia, hyperglycemia levels and low HDL-cholesterol). In addition, multiple linear regression analysis was used to evaluate the associations between baPWV (defined as a continuous variable) and the 6 clinical phenotypes. Potential covariates were adjusted for age, sex, demographic characteristics, lifestyles, WC, heart rate, albumin, total bilirubin, FIB-4, uric acid and eGFR.

Additional sensitivity analysis was performed to replicate our main findings with stricter definitions of metabolically healthy status, including the WC criterion (cutoff points of ≥90 cm for men and ≥85 cm for women) for the definition and defining metabolically healthy participants as having none of the five possible metabolic abnormality risk factors.

A p-value <0.05 was considered statistically significant. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC), and graphs were drawn by GraphPad Prism version 6.00 (GraphPad Software, La Jolla California, United States).

The cross-sectional analysis included 27,738 overall participants (mean age: 49.8 years; 80.3% were male) with complete baPWV information in cross-sectional study sample I. A total of 14,323 individuals (mean age: 60 years; 76.7% were male) with complete carotid plaque information were included in cross-sectional study sample II.

Of the 27,738 NAFLD participants, 13.2%, 55.1% and 31.7% fell into the normal weight, overweight and obese categories, respectively (Figure 2A). In the normal weight group, the MH subjects accounted for a higher percentage than did the MU subjects (56.3% vs. 43.7%), whereas MU status was predominant in the overweight and obese groups (63.5% vs. 36.5%; 74.8% vs. 25.2%) (Figure 2B).

Within each BMI strata, MU individuals were older, had a higher proportion of poor diet scores, had a higher prevalence of diabetes and hypertension, and had higher WC, heart rate, blood pressure, and levels of fasting glucose, triglycerides, liver enzymes and uric acid than MH individuals. In contrast, a greater proportion of MH participants had a university degree, a worker occupation and a short sleep duration, and they had significantly greater levels of HDL-C and total bilirubin than participants with MU (Table 1).

The MU individuals had an obviously higher prevalence of increased baPWV and carotid plaque than the MH individuals across all BMI categories. The extent of subclinical atherosclerosis was similar among MH individuals across BMI categories. For MU individuals, the extension was relatively higher for normal-weight individuals than for overweight and obese individuals (37.2% vs. 32.2% vs. 28.2%; 51.6% vs. 48.9% vs. 49.7%) (Figures 2C,D). A possible reason is that lean NAFLD subjects tend to be older than non-lean NAFLD subjects (52.2 vs. 49.5 years).

Increased baPWV and the presence of carotid plaque were available for the analysis in cross-sectional study sample I and sample II, respectively.

First, we evaluated the association between subclinical atherosclerosis and clinical phenotypes (6 subgroups) after adjusting for potential confounding factors (Table 2). The associations, which were defined according to the highest baPWV quartiles and carotid plaque, were evaluated using multiple logistic regression analysis. With the normal weight-MH subgroup as a reference, among MU subjects, normal weight, overweight and obesity were all significantly associated with a higher risk for increased baPWV and carotid plaque, and the adjusted odds ratio of risk was similar across BMI categories; however, among MH individuals, overweight and obesity showed a much lower but marginally significant risk of increased baPWV but were not associated with carotid plaque. Additionally, we conducted linear regression analysis for baPWV, which was defined as a continuous variable. Similar patterns were found regarding the relationship of subclinical atherosclerosis and MAFLD clinical phenotypes.

Second, to further explore whether the increased risk was mediated by metabolic abnormalities but not by increased BMI, we performed separate analyses to further assess the association between subclinical atherosclerosis and body size phenotypes after adjustment for confounding variables (Figures 3A,B,D,E). Whereas increased baPWV and carotid plaque increased with the increase in the number of metabolic abnormalities, there was no increase in the incidence of subclinical disease across BMI categories.

Finally, the adjusted risk for subclinical disease was also calculated according to individual and joint metabolic risk factors. Both the presence of elevated blood pressure and elevated glucose levels were revealed to be obviously significant covariates in predicting subclinical disease. Elevated triglyceride levels and decreased HDL levels were borderline or not significantly associated with subclinical disease (Figures 3C,F).

Using a strict definition of metabolic health led to a slight reinforcement of the magnitude of the associations. However, this did not generally affect the trend of these associations (Supplementary Table S1 and Supplementary Figure S1).

We report here, for the first time, evidence of an association between clinical phenotypes and subclinical atherosclerosis in light of two different indices. We found a wide spectrum of clinical phenotypes in the NAFLD population. The poor metabolic profile was noticeable in overweight and obese individuals. Compared with metabolically healthy subjects, metabolically unhealthy subjects had a higher burden of subclinical atherosclerosis regardless of their BMI phenotypes. Unhealthy metabolic status, but not BMI-based body size, was the main contributor to the associations of clinical phenotypes with subclinical atherosclerosis. An increasing number of comorbid MetS traits were linked with a higher risk in this cohort. Among the MetS components, elevated blood pressure and blood glucose levels were stronger risk factors associated with the prevalence of subclinical diseases. Our results held true not only when we used a clinical definition of metabolic healthy status (presence of one or fewer of the 4 metabolic risk factors) but also when we used a more stringent definition (presence of none of the 5 cardiometabolic risk factors).

The presence of carotid plaque was validated as an excellent marker of atherosclerotic lesions and structural abnormalities, reflecting generalized atherosclerosis. Increased baPWV is a reliable index of early vascular functional stiffness. A growing body of studies has reported that NAFLD is associated with the two markers of preclinical atherosclerosis, notably independent of established CVD risk factors (8, 23). However, these previous studies have several limitations. First, the most frequently included populations were hospital-based or outpatient cohorts, and some patients were diagnosed with liver biopsy, which resulted in a relatively small sample and led to selection bias. Hence, the results might lead to an overestimation of the effect (24–28). Second, the subjects among whom this link was observed were simply NAFLD vs. non-NAFLD. Thus, the findings can be interpreted that the NAFLD population could be recognized as having a high risk of atherosclerotic CVD. However, it is important to determine how risk discrimination can be performed within the NAFLD population. In our study, the metabolic profile was the greatest contributor to the relationship of subclinical atherosclerosis with different clinical phenotypes. This was illustrated by the effect of risk in subclinical atherosclerosis being concentrated among subjects in the metabolic abnormality subgroups, suggesting a major role for metabolic health in comparison with BMI ranges. Our findings highlight the message that precise metabolic phenotyping assessment beyond BMI could enable proper classification of NAFLD patients.

Several previous studies have reported an association between clinical phenotypes and the prognosis of NAFLD. In Gut, Younes et al. (29) reported a multicenter study comparing the long-term prognosis, including the onset of diabetes and long-term CVD events, between lean and non-lean NAFLD populations. They found that hepatic and non-hepatic clinical complications of the two subgroups did not differ significantly over a decade of follow-up. In addition, they found that 77.5% of the lean NAFLD patients stayed at a normal weight at the end follow-up, which implied that the longitudinal progression to obesity did not contribute to the disease outcomes in such lean subjects. Another study (30) investigated the effect of metabolic abnormalities on the risk of hepatic prognosis among patients with NAFLD over a 9 years period. Among lean NAFLD patients with one or zero metabolic risk factors, the future risk is very low. Therefore, these studies challenged that a BMI-driven strategy for screening NAFLD subjects for cardiovascular or liver damage outcomes can be misleading and should be reevaluated as metabolic derangements.

Although BMI is an acceptable marker for overall adiposity, it cannot distinguish between muscle and fat and does not capture assessments of body-fat distribution, such as visceral and subcutaneous adipose tissue (31). Lean NAFLD is a common phenotype among Asian people, especially as they seem to have higher amounts of central fat deposition than White people and develop NAFLD and metabolic complications within a low BMI range (32, 33). Findings from body composition studies showed that increased BMI was a poor surrogate for increased visceral adipose tissue and intrahepatic fat, both of which are strongly associated with NAFLD and cardiometabolic risk factors (34–37). In contrast, several studies revealed that WC was a reliable anthropometric index of visceral adiposity. Additionally, it was recommended that health professionals should be trained to properly perform this simple measurement and consider it an important “vital sign” in clinical practice. Therefore, importantly, BMI might be a suboptimal parameter for evaluating the association between obesity and subclinical atherosclerosis in the NAFLD population. Further studies that include precise quantification of different fat distributions could help to resolve this issue.

In this study, MetS itself or components of metabolic syndrome were consistent correlates of subclinical atherosclerosis. A complicated and dynamic interaction between a multitude of factors, including genetic, epigenetic, nutrition and dietary factors, lifestyle factors, and gut microbiota, is likely to shape individual metabolic profiles (1). In addition, NAFLD and MetS share a bidirectional association as a cause and a consequence (38). NAFLD is accompanied by impaired insulin-mediated suppression of hepatic glucose production, leading to liver steatosis, hyperglycemia and dyslipidemia. In turn, selective hepatic insulin resistance is thought to result in the development of both hepatic and peripheral metabolic dysfunction. Several types of pathophysiological crosstalk between NAFLD and MetS, including insulin resistance, abnormal lipoprotein metabolism, chronic low-grade inflammation, lipotoxicity and excessive oxidative stress, exert a proatherogenic effect on blood vessels (7). Similarly, Lee et al. (39) and Wang et al. (40) present evidence from large cohort studies from Korea and China, respectively, on the utility of the MAFLD definition for incident CVD and the risk of all-cause deaths. As expected from the positive definition (with a set of accompanying metabolic abnormalities), it is appealing to cover more patients with CVD and death under the umbrella term MAFLD compared with NAFLD. However, a dissociation between NAFLD and insulin resistance/dyslipidemia/CVD is present. NAFLD patients with a specific “liver-genetic” background have a lower risk of CVD, which may partially explain why a subgroup of patients with NAFLD are metabolically healthy (41). In addition, NAFLD impacts metabolism and CVD risk through hepatokines, such as fetuin-A, ANGPTL3 (angiopoietin-related protein 3), FGF21 (fibroblast growth factor 21), SHBG (sex hormone-binding globulin), selenoprotein P, fetuin-B and follistatin (42).

We also tried to disentangle the effects of individual metabolic traits and found no correlation or only a weak effect of low HDL levels and hypertriglyceridemia on the risk of NAFLD atherosclerosis. Among the indices of lipid metabolism, only HDL cholesterol and triglyceride levels are used to define MetS. Nevertheless, total cholesterol and LDL cholesterol are risk factors for atherosclerotic lesions and structural abnormalities, which are not included in the definition.

This study adds an important facet to the NAFLD cardiometabolic risk factor research spectrum in which, within the heterogeneity of body shape phenotype, metabolic health is more closely related to subclinical atherosclerosis, above and beyond adiposity defined by BMI. Our findings add to evidence that metabolic health conferred important clues to cardiometabolic risk that BMI may not explain.

In regard to the risk stratification implications, the strategy is solely stratified by BMI, and lean subjects with NAFLD may be misclassified as low risk and “metabolically healthy.” Obese NAFLD subjects who are “metabolically healthy” will be misclassified as being at higher risk. Accordingly, screening at-risk CVD populations instead of considering obesity or lean NAFLD as particular subgroups and performing a comprehensive assessment of specific metabolic profiles should be recommended. In other words, a novel definition of MH may be better suited to stratify subjects based on cardiometabolic risk parameters. In regard to the intervention management implications, lifestyle interventions remain the cornerstone of treatment. Weight loss has been shown substantial benefits not only for intrahepatic fat loss but also for improvement in metabolic parameters of glucose control and insulin sensitivity, whether NAFLD with overweight or normal weight (43). Several diet regimens were recommended, such as Mediterranean diet, low-carbohydrate diet, a low-fat diet and intermittent fasting over a regular diet (9, 44, 45). High adherence to the Mediterranean diet is associated with better profile of MetS features in NAFLD subjects owing to its close association with CVD (46). Nonetheless, further research is needed for its contribution on the development of subclinical atherosclerosis. In addition, some evidence showed that drinking three or more cups of coffee daily, dietary intake of vitamins E and C and phenolic acids had a protective association with NAFLD and its metabolic factors (47–49).

The strengths of our study include many NAFLD participants; data on both structural and functional markers of subclinical disease; rich covariable adjustments encompassing sociodemographic factors and related CVD risk factors; and additional sensitivity analysis with strict MetS criteria, which added robustness to our findings.

Moreover, our results need to be interpreted with certain limitations. First, our cross-sectional design fails to determine causal relationships between clinical phenotypes and the development of subclinical atherosclerosis. Second, this study population consisted mostly of men, which potentially limits the generalizability of the results. Third, although we used two different versions of the MetS definition, impaired glucose tolerance and the inflammatory index, which are important components of new metabolic dysfunction criteria, were not included in this study (50). Fourth, recent data suggest that clinical phenotypes might be transient for a large proportion of individuals (19). Thus, the presence of phenotypes during one clinical examination could not capture the natural course of the variability within individuals, which remains controversial regarding the findings. Fifth, we chose only two markers of subclinical atherosclerosis, and other surrogate markers, such as coronary artery calcification (CAC) and brachial arterial flow-mediated dilation (FMD), were not included in our analysis. Sixth, we did not include information on emerging “nontraditional” CVD risk factors, such as decreased vitamin D and adiponectin levels, which may influence the relationship (51, 52). Lastly, carotid plaque indexes (such as carotid plaque burden and maximum carotid plaque thickness) are better predictors of cardiovascular disease risk than CIMT, and we did not obtain quantitative indicators of CIMT in our study (53).