The analysis used the database of National Health and Nutrition Examination Survey (NHANES) which is a program of studies conducted by the Centers for Disease Control and Prevention (CDC). It is designed to assess the health and nutritional status of adults and children in the United States. The database consists of data collected from interviews and physical examinations and is open to the public (26).

In our analysis, we focused on two key mortality outcomes: all-cause mortality defined as the deaths from any cause, and CVD mortality specifically referring to the deaths caused by heart disease and cerebrovascular disease (27).

We included 80,312 participants in total from 8 cycles of survey (2003–2004 to 2017–2018) of NHANES. According to the adult age criteria utilized in other studies (28), we only included the participants of 20 years old and above (n = 44,790). In consideration of the data integrity, 6,160 participants were excluded due to missing data on the calculation of Fatty Liver Index (FLI) and 118 participants were excluded due to missing data on mortality status. To further eliminate the liver diseases potentially caused by viral hepatitis and alcohol consumption, 229 with HBV positive, 638 with HCV positive and 4,947 participants with significant alcohol consumption were also excluded. Finally, there were 32,698 participants eligible for our analysis. More details about the participants screening process are shown in Figure 1.

Although as the gold standard of NAFLD diagnosis (29), liver biopsy is not widely used in the first-line screening due to its invasive nature (30). Numerous non-invasive methods, e.g., serum marker and ultrasound are introduced to screen NAFLD. As an algorithm, fatty liver index has been in use in practice for over 17 years and demonstrated the accuracy comparable with other diagnostic methods (31, 32).

Fatty liver index was calculated as following formula:

Where TG (mg/dL), GGT (U/L), BMI (kg/m²), and WC (cm) stand for triglycerides, gamma-glutamyl transferase, body mass index and waist circumference, respectively.

The FLI values (33, 34) ranged from 0 to 100 and were categorized into three levels: <30 no steatosis, 30–59 in intermediate state and ≥ 60 with hepatic steatosis (34, 35). Therefore, in our analysis, it was diagnosed as NAFLD when FLI ≥ 30 if other known causes of liver diseases could be ruled out (36).

The information of follow-up and mortality could be obtained from the Public-Use Linked Mortality Liles where the data through December 31, 2019 linked to the participants in NHANES. By referring to the codes of the International Statistical Classification of Diseases and Related Health Problems, we were able to figure out the cause of death.

Numerous studies have underscored the relationship between NAFLD and mortality, with this association often being influenced by some demographic characteristics. Therefore, to further explore the potential association, some covariates were identified and included in our analysis.

The covariates included age (years: ≤ 39, 40–59, ≥60), sex, race and ethnicity (Mexican American, non-Hispanic Black, non-Hispanic White, and other including other Hispanic, other non-Hispanic, and non-Hispanic multiple races), education level (<High school, High school and Some college or above), smoking status, daily eating frequency (DEF) which was measured by the frequency of eating occasion during day 1 and day 2 (37), Body mass index (BMI: <30.0, ≥30.0), diabetes, hypertension, physical activity measured by Metabolic equivalent (MET) intensity score (VLPA: <150 MET-min/week, LPA: 150–960 MET-min/week, MPA: 961–1800 MET-min/week and HPA: >1800 MET-min/week), alanine aminotransferase (ALT in U/L), aspartate aminotransferase (AST in U/L), high-density lipoprotein (HDL in mmol/L) and low-density lipoprotein (LDL in mmol/L).

In this study, we aimed to explore the long-term increased risk associated with NAFLD by performing a survival analysis on NHANES data during 2003–2018. The effect size was measured using hazard ratio (HR) by comparing those with NAFLD to those without.

According to the official guidance by NHANES, survey weights were utilized in NHANES to account for the complex survey design, including oversampling, survey non-response and post-stratification adjustment (38). To combine two or more 2-year cycles of the continuous NHANES data, the appropriate weights should be constructed to ensure the estimates accurately represented each survey cycle.

For demographic characteristics, the summary for categorical variables was presented by frequency with percentage and medians with interquartile range (IQR) for continuous variables given their non-normal distribution. The comparison between NAFLD and Non-NAFLD was performed by using rank sum test, e.g., Kruskal-Wallis rank test for continuous variables and Chi-square test for categorical variables, with corresponding p values reported. In the survival analysis, we applied weighted Cox proportional hazards regression model to examine the link between NAFLD and all-cause mortality and CVDs mortality. The results were presented as HR with 95% Cis and Kaplan–Meier survival curves. Additionally, forest plots were also presented to detect the heterogeneity of this association across various demographic characteristics.

To detect the dose–response relationship between SUA and all-cause and CVDs mortality, a restricted cubic spline with four knots was employed. To ensure the robustness of our outcomes, the sensitivity analyses were also conducted: (1) only considered the participants with hepatic steatosis (FLI ≥ 60) as NAFLD to perform survival analysis; (2) excluded the participants who died within the first 2 years of follow-up to exclude the potential mortality by competing risk, e.g., cancer, etc.

Multiple imputation method was applied to impute the covariates with missing values. All statistical analyses were conducted by using R software (version 4.3.2). The statistical significance was indicated by two-sided p value less than 0.05.

In this study, a total of 32,698 participants consisting of 2,752 diagnosed with NAFLD and 29,946 without NAFLD were included. The comparison of baseline characteristics between NAFLD and Non-NAFLD (Table 1) showed that a higher percentage of 40–59 age, male, race of Black and White, education level at high school, former or current smokers, ≤ 3.5 and 3.6–4.9 of daily eating frequency, BMI ≥ 30.0, hypertension in NAFLD group compared with Non-NAFLD. The values of ALT, AST, HDL and Serum uric acid of NAFLD were significantly larger than Non-NAFLD.

Based on the three categories by FLI values, the HRs of intermediate state and hepatic steatosis were estimated with no steatosis as the reference level from model 1 to model 3 shown in Table 2. The nonlinear association was found between FLI and all-cause mortality and CVDs mortality in Figure 2. The risk of all-cause mortality was increasing with larger FLI values while the risk of CVDs mortality was the lowest around 30 of FLI and increasing as less or larger FLI.

The comparable mortality rates were noted between the participants of NAFLD and Non-NAFLD. Specifically, the rates of all-cause mortality were 11.3% vs. 11.7% and the rates of CVDs mortality were 3.96% vs. 3.75%, respectively.

The survival analysis was conducted using a weighted Cox proportional hazards regression model. Model 1 was a crude model. In model 2 and 3, except the variable indicating NAFLD or not, for multivariable adjustment, age and sex were included in model 2 and all the significant variables including age, sex, race and ethnicity, education level, smoking status, daily eating frequency, BMI, diabetes, hypertension, serum uric acid group, ALT, AST, HDL, and serum uric acid value were included in model 3. In terms of all-cause mortality, all the HRs of model 1 to model 3 were consistently larger than 1 with statistical significance indicating increased risk in the participants with NAFLD compared to Non-NAFLD. The trend was mirrored in the analysis of CVDs mortality, suggesting NAFLD was associated with an elevated risk of dying from all-cause and cardiovascular causes detailed in Table 3 (39).

The Kaplan–Meier curves derived from model 1 are depicted in Figure 3. Notably, the curve of CVDs mortality appeared flatter than all-cause mortality due to fewer events cumulated during the follow-up.

To further detect any variability in the impact of NAFLD on mortality based on different baseline characteristics (40, 41), the subgroup analyses were performed on all-cause mortality and CVDs mortality, respectively, by including the risk factors of sex, age, race and ethnicity, education level, daily eating frequency, BMI, hypertension, physical activity, serum uric acid with 3 and 4 tertiles as illustrated in Figure 4, 5.

Regarding the endpoint of all-cause mortality, the statistical significance indicating NAFLD as a risk factor of mortality was observed among the subpopulation of female, aged during 40–59, other race, education level of college or above, 5 or more of daily eating frequence, non-hypertension, physical activity of MPA and HPA. An interesting finding was that as the serum uric acid levels increased, the risk of mortality reached the bottom at tertile 2 before rising again, suggesting a need for further investigation into the relationship. It is also worth noting that in the participants with diabetes or hypertension, HRs were less than 1, although without statistical significance. It might be attributed to its complicated mix of mortality causes and excess mortality irrelated to CVDs occurring in Non-NAFLD group.

For CVDs mortality, with the exception of high school of education level, daily eating frequency (≤ 3.5) and physical activity (VLPA), most HRs supported the trend or conclusion that NAFLD elevated the risk of mortality. Especially among the participants of diabetes and hypertension, the results were consistent among all the sub-levels.

To detect the dose–response relationship between SUA and all-cause and CVDs mortality among the participants with NAFLD, we constructed three models by using restricted cubic spline depicted in Figure 6.

The median follow-up duration was 7.25 years with 310 all-cause mortality and 109 CVDs mortality recorded among the participants with NAFLD (42). The HRs and 95% CIs were listed in Table 4. Model 1 is a crude model, model 2 included age and sex as covariates and model 3 included all significant covariates in Table 1. In model 3, an increase of 1 mg/dL in serum uric acid was associated with an 18% increased risk of all-cause mortality [HR (95%): 1.18 (1.07, 1.30)]. For CVDs mortality, while no significant results were observed across various serum uric acid levels relative to the reference level, an increment of 1 mg/dL in serum uric acid was associated with an 22% increased risk of CVDs mortality [HR (95%): 1.22 (1.03, 1.43)].

In the sensitivity analyses, we applied a more rigorous criterion to define NAFLD, i.e., identifying only those with hepatic steatosis (FLI ≥ 60) as NAFLD. Survival analyses were then performed on model 1–3 with their corresponding 95% CIs presented in Table 5. Consistent with the results of Table 3, HRs were all greater than 1 with statistical significance, thereby reinforcing the conclusion that NAFLD was a risk factor for both all-cause mortality and CVDs mortality.

Furthermore, to mitigate the impact of potential mortality due to competitive risks, the participants who died within the first 2 years of follow-up were exclude from the analysis. The resulting HRs and 95% CIs were detailed in Table 6. The results remained consistent with the primary analysis and NAFLD demonstrated a substantial risk factor for all-cause mortality and CVDs mortality.