Nonalcoholic fatty liver disease (NAFLD) is the most prevalent chronic liver disease worldwide, which has been linked to numerous cardio-metabolic dysfunctions, notably type 2 diabetes mellitus and cardiovascular disease.^1-4 Although the pathogenesis of NAFLD remains incompletely elucidated, it has been recognized as the intricate consequence of an amalgamation of social, environmental, and genetic risk factors.^5,6

However, prior epidemiological research on NAFLD mainly focused on the biological determinants, with limited attention towards its socioeconomic causes. Among the several components of socioeconomic status (SES), educational attainment emerges as a relatively important and enduring feature. Therefore, education level has served as a pivotal metric in many studies as a surrogate for the overall socioeconomic construct.^7,8 Lower education level has been repeatedly reported as a risk factor of metabolic diseases. Likewise, both Fedeli et al.⁹ and Jia et al.¹⁰ have reported that lower educational attainment is associated with higher risk and unfavorable outcomes of NAFLD. However, it is noteworthy that studies exploring the effect of education on NAFLD in developing countries remain relatively scarce at present.¹¹

In recent decades, the prevalence of NAFLD has exhibited a concerning upward trend in low- and middle-income countries.¹² For instance, in China, the prevalence of NAFLD has already reached 29.2% in 2018.⁴ Nevertheless, knowledge regarding the impact of SES on NAFLD and its progression is disproportionately limited. So far, no study has comprehensively evaluated the association between SES, NAFLD, and advanced liver fibrosis in the Chinese population.

In this study, we aimed to estimate the effect of education level on NAFLD and liver fibrosis using data from a multicenter, population-based study in China. Moreover, considering the potential gender-specific nature of NAFLD pathogenesis,^7,13 we would stratify the analysis by gender in this study.

1. Participants

The REACTION (Risk Evaluation of cAncers in Chinese diabeTic Individuals: A lONgitudinal) study was a cross-sectional, nationwide, population-based cohort study in China. The detailed design was reported elsewhere.¹⁴ The current study was conducted across 25 centers in mainland China with varied geographic regions and inconsistent economic development stages. No restriction on gender or ethnicity was applied in the enrollment stage. Eligible participants were identified from the local medical registration records and approached by trained community workers. Finally, a total of 259,657 individuals were recruited.¹⁵ The study was approved by the Institutional Review Board at Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (IRB number: 2011-14). All participants have provided written informed consent. All work was carried out in compliance with the Ethical Principles for Medical Research Involving Human Subjects outlined in the Helsinki Declaration in 1975 (revised in 2000).

2. Data collection

Information on complete medical history, lifestyle factors, education levels and marital status were obtained from a structured and validated questionnaire. Participants were asked to choose their highest educational attainment from the set options, which were further categorized into four levels: primary school or below, middle school, high school and college or above according to their responses. A food frequency questionnaire was used to collect diet habits during the previous 12 months, and the International Physical Activity Questionnaire (IPAQ short questionnaire) was used to record data of daily physical activity.¹⁶

Measurements of height, body weight and waist circumference (WC) were taken under a strict protocol by trained stuff, with participants wearing light clothes and no shoes. Body mass index (BMI) was calculated as the weight in kilograms divided by height in meters squared and WC was measured with tape positioned at the middle of the lowest rib and the superior border of the iliac crest.

Blood samples were collected after an overnight fast of at least 10 hours. Tests of alanine transaminase (ALT), aspartate transaminase (AST), and gamma-glutamyl transferase (GGT) were conducted in the qualified laboratory of local medical institutions. For the measurement of total cholesterol, low density lipoprotein cholesterol, high density lipoprotein cholesterol, and triglycerides (TG), blood samples were shipped at 0℃ to 8℃ to the central laboratory at Ruijin Hospital and examined with an ARCHITECT ci16200 autoanalyzer (Abbott Laboratories, Abbott Park, IL, USA).

3. Definition of covariates

According to the 2008 Physical Activity Guidelines for Americans, healthy physical activity was defined as a moderate-intensity exercise of more than 150 min/wk or a vigorous-intensity exercise of more than 75 min/wk, or moderate and vigorous physical activity of more than 150 min/wk.¹⁷ Dietary score was calculated defined on the intake of five elements: high consumption of fruits and vegetables (≥4.5 cups/day), fish (≥ two 100-g serving/wk), cereal grains (≥ three 28-g serving/day), soy food (soy protein ≥25 g/day), and low consumption of sugar-sweetened beverages (≤450 kcal/wk). Each element was assigned a score of 1, and a healthy diet was defined as a summed dietary score of 4.¹⁸ Drinking and smoking status were binarily classified as current smoker (yes/no) and current drinker (yes/no) according to the information obtained from the questionnaires. Marital status was recorded as married/single, divorced, separated or widowed.

General obesity was defined as a BMI greater than 25 kg/m² according to the World Health Organization recommendation for Asian individuals,¹⁹ and central obesity was defined as WC ≥90 cm for men or WC ≥80 cm for women under the guidelines for Asians.²⁰ Dyslipidemia was defined as TG ≥2.26 mmol/L or total cholesterol ≥6.22 mmol/L or low density lipoprotein cholesterol ≥4.14 mmol/L or high density lipoprotein cholesterol <1.04 mmol/L.²¹ Diabetes was diagnosed as fasting plasma glucose ≥7.0 mmol/L, or 2-hour plasma glucose ≥11.1 mmol/L after a 75 g glucose load, or hemoglobin A1c ≥6.5%, or a previous diagnosis of diabetes.²²

4. Assessment of NAFLD and high probability of liver fibrosis

Outcomes included the risk of NAFLD in the general population and a high probability of fibrosis in patients with NAFLD.

Fatty liver index (FLI) is a reliable predictive tool in the diagnosis of NAFLD.²³ NAFLD was diagnosed when the FLI exceeded 60.²⁴ The formula for FLI calculation is as below:

FLI = (e^{0.953*loge (TG) + 0.139*BMI + 0.718*loge (GGT) + 0.053*WC - 15.745}) / (1 + e^{0.953*loge (TG) + 0.139*BMI + 0.718*loge (GGT) + 0.053* WC - 15.745}) ×100

The probability of liver fibrosis was assessed by the AST to ALT ratio, AST to ALT ratio above 1.4 is considered as high probability of fibrosis.²⁵

5. Statistical analysis

All the analyses were conducted separately for men and women. Baseline characteristics of the participants were listed in Table 1. Continuous variables were normally distributed, and were presented as mean±standard deviation, and categorical variables were presented as number (percentage). Differences between groups were examined with one-way analysis of variance for continuous variables and the chi-square test for categorical variables.

Table 1 . Baseline Characteristics of the Participants According to Education Level in Men and Women.

Characteristic	Primary school or below	Middle school	High school	College or above	p-value
Men
No. of patients	13,236	21,492	16,934	10,511
Age, yr	63.0±9.9	57.7±9.5	57.2±9.7	59.2±10.9	<0.001
Body mass index, kg/m2	24.3±3.7	24.9±3.5	24.9±3.6	25.0±3.3	<0.001
General obesity	6,786 (51.3)	12,937 (60.2)	10,332 (61.0)	6,568 (62.5)	<0.001
Waist circumference, cm	84.9±10.5	86.9±9.6	87.5±9.2	88.2±8.9	<0.001
Central obesity	4,345 (32.8)	8,478 (39.4)	6,971 (41.2)	4,556 (43.3)	<0.001
Current smoker	4,404 (34.4)	8,047 (38.5)	5,578 (33.8)	2,554 (25.0)	<0.001
Current drinker	949 (7.6)	1,740 (8.6)	1,415 (8.9)	859 (8.7)	<0.001
Healthy physical activity	1,021 (8.2)	2,532 (12.2)	2,642 (16.0)	2,210 (21.5)	<0.001
Healthy diet	5,325 (52.6)	10,248 (57.6)	9,152 (62.9)	6,049 (65.8)	<0.001
Diabetes	3,261 (25.6)	5,771 (27.7)	4,854 (29.8)	3,256 (31.8)	<0.001
Dyslipidemia	5,215 (39.4)	10,384 (48.3)	8,851 (52.3)	5,876 (55.9)	<0.001
Married	12,212 (92.6)	20,694 (96.6)	16,225 (96.2)	10,136 (96.7)	<0.001
NAFLD	2,105 (15.9)	4,773 (22.2)	3,863 (22.8)	2,335 (22.2)	<0.001
Women
No. of patients	53,291	49,688	39,866	11,820
Age, yr	60.8±9.6	55.3±8.6	54.4±8.0	54.6±9.8	<0.001
Body mass index, kg/m2	24.9±3.8	24.7±3.7	24.2±3.5	23.9±3.4	<0.001
General obesity	30,822 (57.8)	27,422 (55.2)	19,776 (49.6)	5,255 (44.5)	<0.001
Waist circumference, cm	84.3±10.2	83.2±9.6	81.6±9.3	80.9±9.3	<0.001
Central obesity	36,258 (68.0)	31,785 (64.0)	22,969 (57.6)	6,432 (54.4)	<0.001
Current smoker	721 (1.4)	645 (1.3)	451 (1.2)	84 (0.7)	<0.001
Current drinker	363 (0.7)	321 (0.7)	272 (0.7)	117 (1.0)	<0.001
Healthy physical activity	4,133 (8.1)	6,413 (13.3)	6,056 (15.5)	2,099 (18.1)	<0.001
Healthy diet	21,738 (52.5)	25,094 (59.4)	23,131 (66.1)	7,049 (67.3)	<0.001
Diabetes	14,099 (27.5)	11,088 (22.9)	7,929 (20.4)	2,259 (19.5)	<0.001
Dyslipidemia	21,625 (40.6)	20,117 (40.5)	16,208 (40.7)	4,859 (41.1)	<0.001
Married	45,750 (86.1)	45,041 (90.9)	36,034 (90.8)	10,618 (90.1)	<0.001
NAFLD	7,904 (14.8)	6,282 (12.6)	3,908 (9.8)	1,003 (8.5)	<0.001

Data are presented as mean±SD or number (%)..

NAFLD, nonalcoholic fatty liver disease..

Multivariable logistic regressions were used to examine the association of education level (categorical variable) with NAFLD risk among general population, and that with high probability of fibrosis among NAFLD patients. Model 1 was only adjusted for age. Model 2 was further adjusted for current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no) and healthy diet (yes/no). Diabetes and dyslipidemia were further adjusted as dichotomous variables in model 3. In models 4 and 5, BMI and WC were additionally adjusted, respectively. Odds ratios (ORs) and 95% confidence intervals (95% CIs) were calculated for NAFLD and high probability of liver fibrosis in both genders.

In the current study, a gender difference in the association between education level and NAFLD was observed. To explore the underlying mediators, a two-step mediation analysis was conducted. Firstly, a Spearman correlation test was applied to determine the correlation coefficients between education level and the potential risk factors. Variables of significant correlation with education level were selected. Secondly, mediation effects of the selected variables were further quantified with a causal mediation effect model. PROC CAUSALMED procedure in SAS software was employed, in which education level was treated as a categorical variable. A p-value <0.05 was considered statistically significant.

All the statistical tests were two-sided. All the statistical analyses were completed with SAS (version 9.4, Institute, Inc., Cary, NC, USA).

The flowchart of the current study is shown in Supplementary Fig. 1. After excluding those with missing information on education level (n=6,167), 253,490 individuals remained in the cohort. Among them, those diagnosed with viral or autoimmune hepatitis, or with a history of liver cirrhosis (n=5,859), or with excessive alcohol consumption (more than 20 g daily for women and 30 g daily for men) (n=21,989),²⁶ or those with missing information on the variables in FLI formula (BMI, WC, TG, and GGT) were excluded (n=8,804). Overall, a total of 216,838 participants were included in the analysis of NAFLD. Among the 32,173 NAFLD patients, 440 patients were further excluded due to missing information of either ALT or AST, leaving 31,733 patients eligible for the analysis of fibrosis.

The characteristics of the participants according to their education levels by gender are shown in Table 1. The overall prevalence of NAFLD in men and women were 21.0% and 12.4%, respectively (Supplementary Table 1). Men with higher education level were more likely to be obese, while conversely, better-educated women were leaner (Supplementary Fig. 2). Additionally, individuals with higher education levels were more likely to adhere to a healthy diet and engage in physical activity in both genders. However, no obvious difference was detected across education levels with respect to smoking and drinking status. Baseline characteristics according to education level among NAFLD patients are displayed in Supplementary Table 2. Notably, no significant differences were observed in obesity indices across various education levels among NAFLD patients.

The risk estimates of NAFLD across different education levels are depicted in Table 2. Using men with primary school education as the reference group, the ORs (95% CIs) for individuals with middle school, high school and college education were 1.17 (1.07 to 1.27), 1.26 (1.15 to 1.38) and 1.28 (1.16 to 1.41), respectively in model 4 when BMI was adjusted. However, after adjusting for WC, the positive association between education level and NAFLD risk in men diminished, with an OR (95% CI) of 1.03 (0.94 to 1.13) for those with high school education and 0.94 (0.86 to 1.04) for those with college education. In contrast, the association between education and NAFLD in women consistently followed a negative pattern across all five statistical models. After accounting for lifestyle factors and BMI (model 4), the estimates of NAFLD for women with college education was 0.94 (95% CI, 0.89 to 0.99). And when WC was adjusted instead of BMI, the OR (95% CI) changed to 0.88 (0.80 to 0.97).

Table 2 . The Association between Education Level and NAFLD in Men and Women.

Gender	Primary school or below	Middle school			High school			College or above
		OR (95% CI)	p-value		OR (95% CI)	p-value		OR (95% CI)	p-value
Men
No. of cases (%)	2,105 (15.9)	4,773 (22.2)			3,863 (22.8)			2,335 (22.2)
Model 1	Ref	1.31 (1.24–1.39)	<0.001		1.34 (1.26–1.42)	<0.001		1.36 (1.27–1.45)	<0.001
Model 2	Ref	1.25 (1.17–1.33)	<0.001		1.24 (1.16–1.33)	<0.001		1.26 (1.17–1.36)	<0.001
Model 3	Ref	1.23 (1.15–1.31)	<0.001		1.21 (1.13–1.30)	<0.001		1.23 (1.14–1.32)	<0.001
Model 4	Ref	1.17 (1.07–1.27)	<0.001		1.26 (1.15–1.38)	<0.001		1.28 (1.16–1.41)	<0.001
Model 5	Ref	1.06 (0.97–1.15)	0.219		1.03 (0.94–1.13)	0.517		0.94 (0.86–1.04)	0.175
Women
No. of cases (%)	7,904 (14.8)	6,282 (12.6)			3,908 (9.8)			1,003 (8.5)
Model 1	Ref	0.98 (0.95–1.02)	<0.001		0.76 (0.73–0.79)	<0.001		0.63 (0.59–0.68)	<0.001
Model 2	Ref	0.95 (0.91–0.99)	0.014		0.72 (0.68–0.75)	<0.001		0.62 (0.57–0.67)	<0.001
Model 3	Ref	0.94 (0.90–0.98)	0.004		0.71 (0.68–0.74)	<0.001		0.61 (0.56–0.65)	<0.001
Model 4	Ref	1.02 (0.97–1.08)	0.389		0.98 (0.89–1.07)	0.587		0.94 (0.89–0.99)	0.042
Model 5	Ref	1.01 (0.96–1.07)	0.609		0.93 (0.88–0.99)	0.017		0.88 (0.80–0.97)	0.008

NAFLD, nonalcoholic fatty liver disease; OR, odds ratio; CI, confidence interval..

Model 1: adjusted for age (yr); Model 2: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), and healthy diet (yes/no); Model 3: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no) and dyslipidemia (yes/no); Model 4: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no), dyslipidemia (yes/no) and body mass index (kg/m²); Model 5: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no), dyslipidemia (yes/no) and waist circumference (cm)..

The association of education level with high probability of fibrosis among NAFLD patients is presented in Table 3. Unlike the results observed in the NAFLD analysis, the association between education level and fibrosis exhibited a uniform pattern across both genders. In comparison to individuals with primary school education or below, the risk of fibrosis significantly decreased as education level increased, regardless of gender. Male and female NAFLD patients with college education demonstrated a 39% and 31% lower risk of fibrosis, respectively after adjusting for BMI. Furthermore, similar results were observed when WC was included as an alternative adjustment of BMI. Men and women with college education had a significantly lower risk of fibrosis (men: OR, 0.61; 95% CI, 0.51 to 0.73 and women: OR, 0.68; 95% CI, 0.57 to 0.81). Sensitivity analysis taking marital status into account did not substantially change the above findings (Supplementary Tables 3 and 4).

Table 3 . The Association between Education Level and Liver Fibrosis in Men and Women with NAFLD.

Gender	Primary school or below	Middle school			High school			College or above
		OR (95% CI)	p-value		OR (95% CI)	p-value		OR (95% CI)	p-value
Men
No. of cases (%)	562 (27.0)	876 (18.5)			641 (26.2)			371 (15.1)
Model 1	Ref	0.76 (0.67–0.86)	<0.001		0.69 (0.61–0.79)	<0.001		0.61 (0.53–0.71)	<0.001
Model 2	Ref	0.76 (0.66–0.89)	<0.001		0.67 (0.57–0.78)	<0.001		0.59 (0.50–0.71)	<0.001
Model 3	Ref	0.77 (0.66–0.89)	<0.001		0.69 (0.59–0.81)	<0.001		0.61 (0.51–0.73)	<0.001
Model 4	Ref	0.77 (0.67–0.89)	<0.001		0.69 (0.59–0.81)	<0.001		0.61 (0.51–0.73)	<0.001
Model 5	Ref	0.77 (0.66–0.89)	<0.001		0.69 (0.59–0.81)	<0.001		0.61 (0.51–0.73)	<0.001
Women
No. of cases	2,752 (35.5)	1,480 (23.9)			804 (20.9)			231 (23.4)
Model 1	Ref	0.70 (0.65–0.76)	<0.001		0.62 (0.56–0.68)	<0.001		0.65 (0.55–0.76)	<0.001
Model 2	Ref	0.70 (0.64–0.76)	<0.001		0.62 (0.56–0.69)	<0.001		0.66 (0.56–0.79)	<0.001
Model 3	Ref	0.69 (0.64–0.76)	<0.001		0.63 (0.57–0.71)	<0.001		0.68 (0.57–0.80)	<0.001
Model 4	Ref	0.70 (0.64–0.76)	<0.001		0.64 (0.58–0.71)	<0.001		0.69 (0.58–0.82)	<0.001
Model 5	Ref	0.70 (0.64–0.76)	<0.001		0.64 (0.58–0.72)	<0.001		0.68 (0.57–0.81)	<0.001

NAFLD, nonalcoholic fatty liver disease; OR, odds ratio; CI, confidence interval..

Model 1: adjusted for age (yr); Model 2: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), and healthy diet (yes/no); Model 3: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no) and dyslipidemia (yes/no); Model 4: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no), dyslipidemia (yes/no) and body mass index (kg/m²); Model 5: adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no), dyslipidemia (yes/no) and waist circumference (cm)..

Results of the Spearmen correlation analysis are presented in Supplementary Table 5. Causal mediation analysis was employed to investigate potential mediating factors, aiming to elucidate the gender disparity observed in the relationship between education and NAFLD.

As presented in Table 4, the proportion mediated by general obesity in the education-NAFLD association was 51.00% in men and 48.58% in women, and for central obesity, these proportions were 68.04% and 32.58%, respectively. However, no mediation effect of obesity was detected in the association between education and high probability of fibrosis (mediation effect of general and central obesity: p=0.468 and p=0.995 in men; p=0.065 and p=0.221 in women) (Supplementary Table 6).

Table 4 . Mediation Analysis between Education Level and NAFLD Risk in Men and Women.

Obesity	Estimate	95% CI	SE	p-value
General obesity
Men
Total effect of education	0.0081	0.0044 to 0.0119	0.0019	<0.001
Direct effect of education	0.0040	0.0005 to 0.0075	0.0018	0.025
Indirect effect of education through general obesity	0.0042	0.0027 to 0.0056	0.0007	<0.001
Percentage mediated	51.00		12.0049	<0.001
Women
Total effect of education	–0.0180	–0.0201 to –0.0160	0.0010	<0.001
Direct effect of education	–0.0093	–0.0112 to –0.0073	0.0010	<0.001
Indirect effect of education through general obesity	–0.0088	–0.0094 to –0.0081	0.0003	<0.001
Percentage mediated	48.58		2.8598	<0.001
Central obesity
Men
Total effect of education	0.0197	0.0165 to 0.0229	0.0016	<0.001
Direct effect of education	0.0063	0.0035 to 0.0091	0.0014	<0.001
Indirect effect of education through central obesity	0.0134	0.0118 to 0.0150	0.0008	<0.001
Percentage mediated	68.04			<0.001
Women
Total effect of education	–0.0180	–0.0201 to –0.0160	0.0010	<0.001
Direct effect of education	–0.0122	–0.0141 to –0.0102	0.0010	<0.001
Indirect effect of education through general obesity	–0.0059	–0.0064 to –0.0053	0.0003	<0.001
Percentage mediated	32.58			<0.001

NAFLD, nonalcoholic fatty liver disease; CI, confidence interval; SE, standard error..

Adjusted for age (yr), current drinker (yes/no), current smoker (yes/no), healthy physical activity (yes/no), healthy diet (yes/no), diabetes (yes/no) and dyslipidemia (yes/no)..

In this nationwide, population-based cohort study, we have uncovered distinct gender-specific associations between education level and the risk of NAFLD. Specifically, a positive association was observed in men, while a negative association was detected in women. These associations were largely mediated by obesity. Furthermore, we revealed a consistent decline in the probability of liver fibrosis with increasing education levels for both genders. To our knowledge, this is the first comprehensive study to evaluate the association of education with NAFLD and related liver fibrosis in Chinese adults.

Among several SES indicators, education level stands out as the most stable and easily accessible metric. Therefore, it is widely used as the proxy of SES in epidemiological studies. Many studies have shown that lower education was associated with a higher risk of NAFLD occurrence and progression, primarily within Western populations.^27-29 In a NAHANES study, lower education level entailed a 35% reduced risk of NAFLD occurrence,⁷ and similar results were reported by a Greek study.²³ However, the associations between education and NAFLD risk in China were discordant across studies.^9,30,31

In statistical models not taking obesity indices into account, the association between education and NAFLD was positive in men and negative in women. A recent nomenclature change for NAFLD to metabolic dysfunction-associated steatotic liver disease was considered as well.^32,33 The associations between education and metabolic dysfunction-associated steatotic liver disease are presented in Supplementary Table 7, and the results aligned closely with our initial findings.

Nevertheless, the underlying reasons for the gender-specific associations remained less clear.³⁰ In males, the relationship between education and NAFLD underwent a significant change after adjusting for various confounding factors. Although a positive association between education and NAFLD was initially observed after adjusting for lifestyle factors and BMI, this trend shifted when WC was utilized as an alternative adjustment: men with higher education levels did not exhibit an increased risk of NAFLD with WC adjusted. Additionally, in women, some risk estimates lost statistical significance after adjusting for obesity indices. These findings allow us to further explore the potential mediating role of obesity.

Several previous studies have proposed obesity as the most well-defined risk factor of NAFLD in Western countries.^34,35 With the causal mediation model, we further verified that obesity is the major mediator in the education-NAFLD relationship in both genders. The gender-specific relationship between education and obesity may contribute to the divergent education-NAFLD association observed in men and women. Therefore, to mitigate the risk of NAFLD associated with educational inequalities, prioritizing effective obesity management is essential.³⁶

In developed countries, high education level is causally related to healthier lifestyles and more optimal body size.³⁷ In China, however, this educational advantage was accompanied by a higher prevalence of obesity among men.³⁸ This paradoxical observation could, in part, be attributed to the difference in urbanization and economic development levels.³⁹ Exploratory analysis stratified by gross domestic product level was conducted to test whether development discrepancy might contribute to our observations. The results showed distinct estimates of education’s impact on NAFLD risk between low and high gross domestic product groups (Supplementary Tables 8 and 9). The interaction effect of development stage and education on NAFLD risk deserves further exploration.

Meanwhile, higher educational attainment was consistently associated with lower probability of fibrosis in both genders, even after adjusting for obesity. Moreover, we did not observe any significant mediation effect of obesity on fibrosis among NAFLD patients. This observation aligned with the findings from an NAHANES study.⁴⁰ Existing literature revealed that the progression from NAFLD to fibrosis is a tale of two “hits” theory.⁴¹ The second hits from non-alcoholic fatty liver to nonalcoholic steatohepatitis and advanced fibrosis is more complicated.^42,43 Our findings corroborated this concept, as obesity did not function as a mediator in fibrosis risk. Given the uncertain pathophysiology of fibrosis, currently, there is no effective method to cure hepatic fibrosis except for liver transplantation. Therefore, early detection and timely intervention are especially important. Individuals with higher levels of education usually have richer health knowledge, better access to health care and better adherence to the medical advice.⁴⁴ Therefore, they would be more likely to block the disease progression and have better outcomes.⁴⁵ However, considering NAFLD and fibrosis were detected simultaneously at the study entry, the current observation only indicated a negative association between education and fibrosis, and yet we could not ascertain whether low education could predict a worse outcome of NAFLD. Future prospective studies are warranted to determine the predictive value of educational attainment on the prognosis of NAFLD.

Strength of the study include a large sample size, a comprehensive assessment of physical and biochemical indices, as well as the appropriate statistical methods. However, the current study also has several limitations to be addressed. Firstly, information on education level was self-reported, thereby a recall bias might exist. Nevertheless, the validity and reliability of self-reported education levels have been confirmed before.⁴⁶ Secondly, NAFLD was diagnosed with FLI rather than ultrasound, and the high probability of fibrosis was determined with AST to ALT ratio score rather than liver biopsy. Despite these considerations, these formulas have been repeatedly validated and are widely accepted as diagnostic surrogates in epidemiological studies.⁴⁷ Thirdly, factors contributing to NAFLD and fibrosis, such as inflammation status and food insecurity, were not included in the analysis, which limited our ability to fully explain the observed associations. Lastly, causal relationships cannot be established through the current observational study, and further Mendelian randomization studies could provide more information.

Our results demonstrated that low education level was associated with higher risk of NAFLD among women, as well as high probability of liver fibrosis in both genders. Greater efforts are needed to address the educational disparities in NAFLD. As China is the biggest developing country in the world and now harbors the largest number of NAFLD patients, the findings of our study might also provide valuable insights for countries at similar development stages.

This work was supported by the National Natural Science Foundation of China (Grant No. 81930021, 81970728, 81970691, 81900741, 81870604, 82100916 and 21904084), Shanghai Municipal Human Resource Development Program for Outstanding Academic Leaders in Medical Disciplines (Grant No. 20XD1422800), Chinese Academy of Medical Sciences (Grant No. 2018PT32017, 2019PT330006), S Ministry of Science and Technology of China (Grant No. 2 2022YFC2505202), Clinical Research Plan of SHDC (Grant No. SHDC2020CR3064B), and Science and Technology Commission of Shanghai Municipality (Grant No. 19411964200 and 20Y11905100).

The authors thank all team members and participants from the REACTION study.

No potential conflict of interest relevant to this article was reported.

Study concept and design: Y.Y.Z., L.W., L.L. Data acquisition: Y.Y.Z., L.W., L.L., Y.N.H., Q.W., Y.F.Q., R.Y.H., L.X.S., Q.S., X.F.Y., L.Y., G.J.Q., X.L.T., G.C., S.Y.W., H.L., X.Y.W., C.Y.H., M.L., M.X., Y.X., T.G.W., Z.Y.Z., Z.N.G., G.X.W., F.X.S., X.J.G., Z.J.L., L.C., Q.L., Z.Y., Y.F.Z., C.L., Y.M.W., S.L.W., T.Y., H.C.D., L.L.C., T.Z.S., J.J.Z., Y.M.M., W.Q.W., G.N., Y.F.B., Y.H.C., J.L.L. Data analysis and interpretation: Y.Y.Z., L.W., L.L., J.L.L. Drafting of the manuscript: Y.Y.Z., L.W., L.L., J.L.L. Critical revision of the manuscript for important intellectual content: Y.F.B., Y.H.C., J.L.L. Study supervision: G.N., W.Q.W., Y.F.B. Statistical analysis: Y.Y.Z., L.W., L.L. Obtained funding: L.W., L.L., G.N., Y.F.B., J.L.L. Approval of final manuscript: all authors.

Supplementary materials can be accessed at https://doi.org/10.5009/gnl230220.