Association of Hepatic Steatosis and Metabolic Dysfunction With Incident Cardiovascular Disease: A Nationwide Cohort Study

Background/purpose: Metabolic dysfunction (MD)-associated fatty liver disease (MAFLD) is new clearer nomenclature with positive diagnostic criteria on the basis of hepatic steatosis and MD, involving exclusion criteria for nonalcoholic fatty liver disease (NAFLD). Methods: This retrospective cohort study included 333,389 participants who received health examination between 2009 and 2010 from the Korean National Health Insurance Service database. Hepatic steatosis was dened using the Korean National Health and Nutrition Examination Survey-derived NAFLD scoring system. Participants were followed over 5-6 years from the index date of health examination to 31 December 2015 for cardiovascular disease (CVD). The Cox proportional hazards regression was adopted to determine adjusted hazard ratio (aHR) with 95% condence interval (CI) for CVD according to the presence of hepatic steatosis, MD, and MAFLD. Results: This study included 179,437 men and 153,952 women with a median age of 57 years. Hepatic steatosis with MD (aHR, 2.00; 95% CI, 1.89-2.13) and without MD (aHR, 1.30; 95% CI, 1.10-1.54) signicantly increased the risk of CVD compared to no steatosis without MD (reference). However, steatosis revealed no signicant difference in the risk of CVD compared to no steatosis among participants with one metabolic dysfunction (aHR, 1.09; 95% CI, 0.91-1.30). In participants with steatosis, the presence of one and ≥ 2 metabolic dysfunctions had aHR of 1.25 (95% CI, 0.87-1.79) and 1.71 (95% CI, 1.22-2.41) compared to no MD, respectively. Conclusion: MAFLD was signicantly associated with increased CVD risk and had better predictive performance for CVD risk compared to hepatic steatosis or metabolic dysfunction alone. Health Nutrition Survey nonalcoholic disease MAFLD, metabolic dysfunction-associated fatty liver disease; PY, aHR, condence interval; MD, metabolic dysfunction; hepatic steatosis


Introduction
Nonalcoholic fatty liver disease (NAFLD) was coined three decades ago by Ludwig and colleagues and de ned as hepatic steatosis with the exclusion of other competing etiologies of chronic liver diseases. [1,2] Despite the various preventive and therapeutic approaches, the prevalence of NAFLD continues to increase with signi cant impacts on clinical medicine and economic burden to the society that nearly one billion people are being affected. [3] Considering growing knowledges on the heterogeneity of the patients with NAFLD with respect to its major drivers, including sociodemographic, clinical, and biological characteristics, an international panel of experts has suggested metabolic dysfunction (MD)-associated fatty liver disease (MAFLD) as a more articulate term that leads the research community to the renewal of the nomenclature to accelerate the development of new therapeutic approaches. [4] In response to unmet needs in this eld, the diagnosis of MAFLD can now be made using positive criteria on the basis of hepatic steatosis (HS) irrespective of alcohol consumption and concomitant liver diseases, with 3 subphenotypes including i) overweight or obesity, ii) lean or normal weight, and iii) type 2 diabetes mellitus (DM). [5] Lean or normal weight MAFLD is diagnosed if at least two MDs coexist, which includes abnormalities in waist circumference (WC), blood pressure (BP), triglycerides (TG), high-density lipoprotein (HDL)-cholesterol, glucose levels, insulin resistance, and high-sensitivity C-reactive protein as proposed in the Adult Treatment Panel III. [6,7] Currently, ultrasound, vibration-controlled transient elastography, computed tomography, magnetic resonance imaging and spectroscopy, and serum biomarker-based models with available data on the diagnostic and prognostic performances, such as fatty liver index (FLI), are recommended for the detection of HS. [8,9] Recently, we have developed the Korean National Health and Nutrition Examination Survey (KNHANES)derived NAFLD scoring system (K-NAFLD) based on patient characteristics and serum biomarkers, which represents metabolic risk and insulin resistance. [10] Given the close relationship between NAFLD, MD, and cardiovascular disease (CVD), there is an unmet need for updates on the association of MAFLD with CVD. [11,12] Herein, CVD risk was assessed according to the presence of HS, MD, and MAFLD using a Korean nationwide cohort.

Study population
The Korean National Health Insurance Service (NHIS)-National Health Screening Cohort (NHIS-HEALS) provided the study population. The NHIS of the Republic of Korea provides mandatory health insurance for all Koreans with an enrollment of nearly 98% covering almost all types of healthcare services. [13] The biannual health screening examinations of the NHIS include anthropometric measurements, laboratory examinations, medical history, prescriptions, and self-reported questionnaires for claims purposes. The NHIS-HEALS cohort is composed of approximately 500,000 participants aged between 40 and 79 years, which is about 10% of the total population. [14] This study was conducted in accordance with the STROBE guidelines for cohort studies. The institutional review board of Seoul National University Hospital approved the present study (E-2002-040-1099) in accordance with relevant guidelines and terms.
The requirement for informed consents was waived as the NHIS data were anonymized according to strict con dentiality guidelines.
[10] K-NAFLD score<−3.285 and >0.884 ruled out and in HS, respectively, as described previously. HS was de ned as K-NAFLD score>0.884, whereas NAFLD required additional exclusion criteria, such as no alcohol consumption and no chronic viral hepatitis B and C de ned using the international Classi cation of Diseases 10th Revision (ICD-10) code of B18, as described previously. [15] The FLI was calculated by (e 0.953 ×Log e (TG)+0.139×BMI +0.718×Log e (γ-glutamyl transpeptidase; γ-GT)+0.053×WC−15.745)/(1+e 0.953 ×Log e (TG)+0.139×BMI+0.718×Log e (γ-GT)+0.053×WC−15.745) ×100. [16] MAFLD was diagnosed when a participant had either i) BMI≥23 kg/m 2 , ii) FSG≥126 mg/dL and/or prescription of anti-diabetic drugs, or iii) ≥2 other MDs, on the top of underlying HS in accordance with the international expert consensus statement. [5] WC≥90 cm for men and ≥80 cm for women, BP≥130/85 mmHg or speci c drug treatment, TG≥150 mg/dL or speci c drug treatment, HDL-cholesterol<40 mg/dL for men and <50 mg/dL for women, and prediabetes (FSG between 100 to 125 mg/dL) constituted the components of the MD. Lipid accumulation product (LAP), an indicator for the severity of HS, was calculated by (WC−65)×TG for men and (WC−58)×TG for women, respectively. [17] LAP was evaluated in dichotomous groups strati ed by HSS and HS index (HSI; 8×ALT/aspartate aminotransferase ratio+BMI(+2, if DM; +2 if female) to support the association between the severity of HS and the HSS (Supplementary Table 1). [18] Key variables The following covariates were considered key variables for adjustments in multivariate analyses: age, sex, insurance premium (upper half and lower half), BMI (kg/m 2 ), FSG (mg/dL), TC (mg/dL), ALT (IU/L), smoking (never, previous, and current), alcohol consumption (never, 1-4 days/week, and ≥5 days/week), physical activity [moderate-to-vigorous physical activity (MVPA)≤1 time/week and >1 time/week], and the Charlson comorbidity index (CCI). The CCI was calculated in accordance with a previous study. [19] Incidence of CVD The attending physicians were required to insert the ICD-10 code for the primary disease for each patient upon hospitalization. CVD was operatively de ned as 2 or more days of hospitalization due to coronary heart disease (ICD-10 code: I20-I25) or stroke (ICD-10 code: I60-I69), which were adopted by the American Heart Association guidelines. [20] The validity of the operational de nition using the NHIS database was described elsewhere. [21] Starting from 1st Jan 2009, all participants were followed for CVD and death, whichever occurred earlier, till 31st Dec 2015.

Statistical analysis
Continuous and categorical variables were presented with median (interquartile range; IQR) and frequency (%), respectively. Between-group difference was evaluated using the chi-square test and independent t-test for categorical and continuous variables, respectively. The incidence rate of CVD was evaluated by crude rate/1,000 person-year (PY). The Multivariate Cox proportional hazards regression model was applied for evaluation of adjusted hazard ratio (aHR) with 95% con dence interval (CI) for CVD according to the HS. Considering potential bias due to other cause before the follow-up investigation, 1-and 2-year of latent periods were washed out for sensitivity analyses. Subgroup analyses were conducted based on the key variables. For incident CVD in the overall follow-up period, receiver operating characteristic contrast estimation was carried out for NAFLD, HS, and MAFLD. A P value of less than 0.05 was considered statistically signi cant. All data collection, mining, and statistical analyses were performed using SAS Enterprise Guide 7.1 (SAS Institute, Cary, USA).

Subjects characteristics
The descriptive characteristics of the study participants are shown in Table 1. There were 179,437 men (53.8%) and 153,952 women (46.2%) with a median age of 57 (IQR, 51-64), a total follow-up period of 1,850,704 PY, and a total CVD event of 32,411 cases. The median BMI and WC were 23.9 kg/m 2 (IQR, 22.0-25.8) and 82 cm (IQR, 76-87), respectively. Among entire study population, 46,319 participants (13.9%) had HS. Compared to the participants with no HS (n=287,070), the participants with HS (n=46,319) revealed signi cantly higher proportion of men, smoking, alcohol consumption, chronic viral hepatitis, hypertension, type 2 DM, dyslipidemia, and CCI≥2, indicating that the HS group had a higher cardiometabolic risk than the non-HS group (Supplementary Table 2).

Associations of HS, MD, and MAFLD with CVD risk
We rst examined the association of HS with CVD using the HSS and FLI (Supplementary Table 3). HS was in direct proportion to CVD risk since both 4th quartile groups of HSS and FLI showed highly increased CVD risk compared to 1st quartile groups. The discriminative performance of the HSS in the prediction of CVD was more effective than that of the FLI in terms of the CVD incidence rates. Subsequently, the study population was strati ed into 4 groups, including very low, low, moderate, and high HSS. The median HSS of the participants with no ruled in nor ruled out for HS were −1.690, which was set as the cut-off for strati cation into low and moderate HSS groups. The incidence rate of CVD was lowest in very low HSS, followed by low, moderate, and high HSS ( Table 2). Among participants with no MD, the aHRs were 1.30 (95% CI, 1.17-1.44), 1.24 (95% CI, 1.01-1.52), and 1.27 (95% CI, 0.78-2.06) for low, moderate, and high HSS, respectively. In addition, HS was found to signi cantly increase CVD risk (aHR, 1.17; 95% CI, 1.08-1.27) only among participants with ≥2 MDs, who corresponded to the MAFLD group, suggesting that signi cantly increased CVD risk in overall population might be highly attributed to participants with ≥2 MDs. Subgroup analysis of the effect of HS on incident CVD among subjects with MD showed that a signi cant trend of increased CVD risk by HS severity was observed in all subgroups except for CCI=0 (Table 3). To con rm whether the prognostic impact of MAFLD on incident CVD is consistent across various cardiovascular risk factors, subgroup analyses were carried out after stratifying the study population into MAFLD and non-MAFLD (Fig. 2). Differing from the effect of HS on incident CVD shown in Table 3, MAFLD signi cantly increased CVD risk only in the age≥65 years, male, past or current smoker, BMI<25 kg/m 2 , and CCI≥1 subgroups.
When examining the combined effect of the HSS and MD on CVD risk, both the HSS and MD additively increased the risk of CVD (Fig. 3). aHR of the HSS>0.884 and ≥2 MDs group (i.e., the MAFLD group), was 2.22 (95% CI, 2.07-2.38) compared to the HSS<−3.285 and no MD group (i.e., the reference group) (Supplementary Table 4). Considering the validity of the FLI in the re ection of HS, we strati ed the participants according to the IQR of the FLI. In accordance with the HSS, the 4th quartile of the FLI with ≥2 MDs had the highest aHR of 2.23.

Sensitivity analyses on the associations of HS and MAFLD with CVD risk
To support the aforementioned effects of HS and MAFLD on incident CVD, 1-2 years of latent periods were washed out to exclude potentially uncontrolled factors. After 1-year wash out, HS lost its signi cant association with the risk of incident CVD, whereas the association of MAFLD with incident CVD risk remained signi cant (aHR 1.05; 95% CI, 1.00-1.10; Supplementary Table 5). After washing out 2 years of latent period, similar results were found for HS and MAFLD, supporting that MAFLD is an independent predictor of incident CVD. We also adopted the HSI as a proxy for HS and NAFLD and rea rmed that MAFLD was signi cantly associated with increased CVD risk (aHR, 1.15; 95% CI, 1.11-1.19) in agreement with the ndings based on the HSS and FLI (Supplementary Table 6).

Comparison of NAFLD and HS with and without MD in the prediction of CVD risk
According to the different probability levels of NAFLD ranged by the HSS cutoffs, the incidence rates of CVD were 20.0, 20.5, and 25.2 in the very low, low-to-moderate, and high probability NAFLD subgroups, respectively. (Table 4) In terms of the different probability levels of HS, the incidence rates of CVD were 12.0, 18.6, and 22.1 in the very low, low-to-moderate, and high probability HS subgroups, respectively. Furthermore, strati cation of the participants with HS into MD=0, =1, and ≥2 revealed that the no MD group had the lowest incidence rate (9.0) despite the presence of HS. Of the HS subgroups, the 1 MD group also showed a relatively lower incidence rate (14.9) compared to NAFLD with low-to-moderate probability (20.5) or HS with low-to-moderate probability (18.6), suggesting that the presence of MD may be more effective in the prediction of CVD risk compared to the presence of NAFLD or HS. Moreover, MAFLD was more predictive of overall CVD incidence in the receiver operating characteristic contrast estimation compared to NAFLD without statistical signi cance (P=0.060; Supplementary Table 7).

Associations of the different MAFLD subphenotypes with CVD risk
When stratifying MAFLD into three subphenotypes, DM-negative overweight/obese, DM-negative normal weight, and DM-positive subphenotypes showed signi cantly increased CVD risk in all unadjusted and adjusted models (Table 5). However, after further adjusting for the number of MD, only DM-MAFLD subtype signi cantly showed an increased risk of CVD despite the attenuation of statistical signi cance (aHR, 1.18; 95% CI, 1.10-1.26), but both DM-negative overweight/obese and DM-negative normal weight subphenotypes lost their statistical signi cance (Table 5).

Discussion
Recently, an international consensus of experts proposed MAFLD as new clearer nomenclature with positive diagnostic criteria to surmount the limitations on an operational de nition of NAFLD. [5,22] In the current study, both HS and MD were unfavorable prognosticators of incident CVD. In particular, the composite term, MAFLD conferred the most highly elevated risk of CVD in general population. The incidence rate of CVD over the follow-up period of 5-6 years was 7.3 among participants without HS and MD, which reached up to 23.0 in participants with MAFLD. These results indicate the better performance of MAFLD in the prediction of CVD risk, circumventing the complicated use of both NAFLD and metabolic abnormalities.
The major differences between NAFLD and MAFLD are i) a set of new positive criteria for the diagnosis of MAFLD, regardless of other concomitant liver diseases and alcohol consumption, and ii) the inclusion of MDs in the diagnostic criteria. A previous study demonstrated that MD does not independently increase CVD risk among US population with NAFLD.
[23] However, in the current study, CVD risk varied among different MAFLD subphenotypes. Speci cally, nondiabetic MAFLD subphenotypes showed no signi cant associations with CVD risk, whereas diabetic MAFLD subphenotype demonstrated signi cantly increased risk of CVD compared to the subjects without HS, suggesting that MD may be an independent predictor of incident CVD in nondiabetic participants with HS.
Despite growing evidence on the association of either NAFLD or MD with CVD, it may be di cult to evaluate the prognostic impact of hepatic and metabolic risk factors on CVD risk in a uni ed and precise manner. In addition, con icting results exist regarding the prognostic impact of NAFLD on the risk of CVD, as con rmed by the meta-analysis of 16 observational studies.
[24] Recently, we developed the K-NAFLD score and con rmed its signi cant associations with metabolic risks and insulin resistance. [10] This scoring system allows for an operational de nition of NAFLD and its application in the nationwide population-based cohort. Furthermore, CVD risk in subjects with HS should be analyzed separately in different subgroups with varying degree of metabolic risk despite clear evidence on the link between NAFLD and cardiometabolic risk. With an emerging concept of MAFLD, metabolic risk and HS seem to be now evaluable precisely in a concurrent manner.
In the current study, the HSS which was also found to be associated with HS severity, is in direct proportion to the risk of CVD even after adjustments for sociodemographic characteristics (age, sex, and insurance premium), anthropometric measurements (BMI and WC), serologic characteristics (FSG, TC, and ALT), lifestyle behaviors (smoking, alcohol consumption, and physical activity), and the CCI, suggesting that the HSS may be an independent predictor of incident CVD. Even participants with a low HSS had signi cantly increased CVD risk compared to those with a very low HSS. In contrast, no signi cant increase in CVD risk was observed in participants with 1 MD compared to no MD among participants with HS, which may reinforce the international consensus on the de nition of MAFLD as HS with ≥2 MDs. However, it should be validated in future large-scale con rmatory studies.
From our perspective, better performance of MAFLD versus NAFLD in the prediction of CVD risk might be attributed to the inclusion of participants with alcohol consumption and chronic viral hepatitis in the MAFLD group. Moreover, our results indicated that participants with HS had 3.0% and 1.0% higher proportion of alcohol consumption≥5 days/week and chronic viral hepatitis, respectively, than those without HS. Repeated binge drinking or excessive cumulative alcohol consumption has been con rmed to have deleterious cardiovascular health consequences by increasing the incidence of CVD and related mortality. [ Although our ndings con rmed the deleterious effects of HS and MD on CVD risk, the current study had several inherent limitations that need to be considered when interpreting the results. First, the homeostasis model assessment for insulin resistance and high-sensitivity C-reactive protein level, which are integral components for the diagnostic criteria of MAFLD, were not available in the NHIS database. Second, the presence and severity of HS were de ned using the HSS, FLI, HSI, and LAP due to the nature of a large-scale cohort in which radiological and histological data were not available. Therefore, further prospective studies are warranted to validate our results using more accurate assessment tools of HS, such as imaging or biopsy. Finally, the current study adopted the ethnicity-speci c (i.e., Asian) criteria for overweight (≥23 kg/m 2 ), obesity (≥25 kg/m 2 ), and waist circumference (≥90 cm for men and ≥80 cm for women). Thus, further studies using Caucasian populations with different criteria for obesity are needed to con rm our ndings.
With these caveats in mind, our ndings indicate that HS and MD are signi cantly associated with increased CVD risk. The bene ts of applying MAFLD rather than NAFLD in the prediction of CVD risk include positive inclusion criteria and better discriminative ability in the strati cation of individuals at different CVD risks. The incidence rate of CVD for MAFLD was 23.0, which was over 3 folds of normal population without HS and MD, suggesting that MAFLD needs to be considered an important risk factor for CVD, as well as a new etiologic subtype of CVD.

Funding
Not applicable.

Competing interests
The authors declare no competing interests.

Ethics approval
The institutional review board of Seoul National University Hospital approved the present study (E-2002-040-1099) in accordance with relevant guidelines and terms.

Consent to participate
The requirement for informed consents was waived as the NHIS data were anonymized according to strict con dentiality guidelines.

Consent to publication
Not applicable.

Availability of data and material
Data used in the presents study were achieved from the NHIS. Researchers can apply for the data from the NHIS (https://nhiss.nhis.or.kr) on research purposes.

Code availability
Codes used in this study can be obtained from the corresponding author under reasonable request.  aHR, adjusted hazard ratio; CI, con dence interval. * P<0.05. ** P<0.01. *** P<0.001. smoking, alcohol consumption, physical activity, and the Charlson comorbidity index. g Adjusted for factors involved in model C and the number of metabolic dysfunctions. MAFLD de ned using the Korean National Health and Nutrition Survey nonalcoholic fatty liver disease hepatic steatosis score. Acronyms: MAFLD, metabolic dysfunction-associated fatty liver disease; DM, diabetes mellitus; PY, person-year; HR, hazard ratio; aHR, adjusted hazard ratio; CI, con dence interval. * P<0.05. ** P<0.01. *** P<0.001.