A total of 11,960 subway workers from Wuhan Metro Group Co., Ltd. were recruited and completed questionnaires between December 2018 and January 2019. In the present study, we excluded participants who had missing data on shift work (n = 518), had cancer (n = 8), without liver ultrasound examination (n = 2,321), or with insufficient data to diagnose MAFLD (n = 17). Thus, a total of 9,105 participants were included in the analyses. Detailed information on participant selection is depicted in Supplementary Figure S1.

All participants provided written informed consent, and this study was approved by the Wuhan Center for Disease Prevention and Control Ethics Committee (WHCDCIRB-K-2018042).

Shift work was defined as having any work schedule involving irregular working hours instead of a normal daytime work schedule (11). Participants were asked, “Does your work involve shift work?” (yes/no). The participants who answered “yes” were then asked to specify their shift work type (two-shift, three-shift, or four-shift) and the number of years they had been engaged in shift work. In the two-shift system, employees work either a day shift (08:00–17:00) or a night shift (20:00–05:00), 5 days a week, with a one-month shift cycle. In the three-shift system, employees work one of the three shifts: morning shift (06:00–14:00), evening shift (14:00–22:00), or night shift (22:00–06:00). Workers rotate through these shifts over a 3-week cycle, working 5 days a week. For example, an employee might work the morning shift in the first week, the evening shift in the second week, the night shift in the third week, and then repeat the cycle, starting with the morning shift again. In the four-shift system, employees work rotate through a morning shift (06:00–14:00), an evening shift (14:00–22:00), a night shift (22:00–06:00), and a 24-h rest shift. The shift cycle is 4 days, with each shift lasting one day followed by 3 days off. Temporary substitutions may occur in cases of staff shortages. Self-reported shift work duration was categorized as ≤3 years, 3–6 years, and >6 years.

Covariates, including age, sex, educational status (high school or below, university/college, or graduate student or above), smoking and drinking status (current, past, or never), dietary habits, sleep duration, sleep quality, and active exercise, were obtained by questionnaire. Participants who smoked at least one cigarette per day for the past 6 months were defined as current smokers. Participants who drank ≥1 time/week over the past half a year were defined as current drinkers. Active exercise was defined as engaging in moderate or moderate-vigorous intensity activities ≥150 min/week or engaging in vigorous-intensity activities ≥75 min/week (12). Body mass index was calculated by weight (kg) divided by height (m) squared. Hypertension was identified through a combination of self-reported medical history, ongoing treatment with antihypertensive drugs, or blood pressure ≥140/90 mm Hg (13). Dyslipidemia was diagnosed based on self-reported medical history, active lipid-lowering therapy, or specific lipid profile thresholds: total cholesterol ≥6.22 mmol/L, high-density lipoprotein cholesterol <1.04 mmol/L, low-density lipoprotein cholesterol ≥4.14 mmol/L, or triglycerides ≥2.26 mmol/L (14). Diabetes was ascertained by self-reported medical history, the presence of anti-diabetic medication, or fasting blood glucose ≥7.0 mmol/L (15). We grouped 26 types of jobs into four categories: technicians (including vehicle maintenance workers, telecommunications workers, signal workers, etc.), train drivers, service workers (including station attendants, ticketing administrators, etc.), and office workers (including managers, clerical support workers, etc.).

The abdominal ultrasound examination is conducted by experienced sonographers using high-resolution ultrasound machines. MAFLD was defined as abdominal ultrasonography-diagnosed fatty liver disease along with the presence of one of the following three criteria: overweight/obesity (BMI ≥ 23 kg/m² in Asians), presence of diabetes mellitus, or evidence of metabolic dysregulation (16). The metabolic dysregulation was defined as the presence of at least 2 of the following metabolic risk abnormalities: ① blood pressure ≥130/85 mmHg or receiving antihypertensives, ② plasma triglyceride ≥1.70 mmol/L or receiving specific drug treatment, ③ plasma high-density lipoprotein cholesterol <1.0 mmol/L for men and <1.3 mmol/L for women or receiving specific drug treatment, ④ prediabetes defined as having fasting glucose 5.6 to 6.9 mmol/L (16).

Continuous variables were presented as mean ± SD, and categorical variables were presented as frequency (percentage). Differences between groups were compared using student t-test for continuous variables and chi-square test for categorical variables.

Logistic regression models were used to assess the association of shift work with MAFLD. Model 1 adjusted for age, sex, educational status, smoking status, drinking status, sleep duration, sleep quality, dietary consumption of grain, beans or soy products, vegetables and fruits, milk or dairy products, meat, fish or seafood, and egg, and job category; model 2 further adjusted for hypertension, dyslipidemia, and diabetes. To explore the association between shift work and MAFLD independent of BMI, we additionally adjusted for BMI based on model 2. Missing data of continuous variables were imputed with the median value, and a dichotomous variable was created to flag the missingness, and an extra category was added to denote the missingness for categorical variables. Restricted cubic spline regression was employed to evaluate the potential nonlinear relationship between the shift work duration and MAFLD. As a previous study indicated that BMI may mediate the association between shift work and MAFLD (9), our study revealed that the association was substantially attenuated after additionally adjusting for BMI, which provided preliminary support for the plausible mediating role of BMI. Thus, we used the R package “mediation” (17) to estimate the potential mediation effect of BMI between shift work and MAFLD. The mediated proportion reflected the average mediation effect of BMI. To explore potential heterogeneity, stratified analyses were performed by sex (male or female), BMI [<24 or ≥24 kg/m² according to the standard classification specific for the Chinese population (18)], current smokers (yes or no), current drinkers (yes or no), sleep duration (≤7 or >7 h), hypertension (yes or no), dyslipidemia (yes or no), and job category (technicians, train drivers, service workers, or office workers). The interactions were tested by adding multiplicative interaction terms into the models. Sensitivity analysis was conducted by excluding participants with missing values.

Analyses were conducted using R software (version 4.2.2, R Core Team) or SAS program (version 9.4, SAS Institute).

Among the 9,105 participants eligible for the analyses, 2,590 had MAFLD. The mean age of the participants was 27.2 ± 4.2 years, and 20.0% were women. Subjects with MAFLD were older, more likely to be male, had higher prevalences of current smoking and drinking, with higher BMI and shorter sleep duration, and were more likely to have hypertension, dyslipidemia, and diabetes (Table 1). 70.4% of the participants were involved in shift work. Participants who worked in shifts were younger, more likely to be female, less likely to be smokers and drinkers, exercised inactively, had longer sleep duration, and were less likely to have dyslipidemia and diabetes (Supplementary Table S1).

As shown in Table 2, compared with individuals with no shift work, the risk of MAFLD decreased by 18% (OR: 0.82, 95% CI: 0.70, 0.95) for those with ≤3 years of shift work, and increased by 18% (OR: 1.18, 95% CI: 1.03, 1.36) and 49% (OR: 1.49, 95% CI: 1.29, 1.73) for those with >3–6 and >6 years of shift work, respectively, after adjustments for demographics and lifestyles. When additionally adjusting for hypertension, dyslipidemia, and diabetes, the ORs (95% CIs) were 0.80 (0.68, 0.94), 1.21 (1.04, 1.41), and 1.60 (1.37, 1.88), respectively. For every 1-year increase in shift work, the risk of MAFLD increased by 6% (OR: 1.06, 95% CI: 1.04, 1.08). When further adjusted for BMI, the association was slightly attenuated. Restricted cubic spline regression analysis showed a J-shaped relationship between the shift work duration and MAFLD (P_overall < 0.001, P_nonlinear = 0.002; Figure 1), with the likelihood of MAFLD substantially increased after 3 years of shift work.

Compared with participants with no shift work, individuals who worked two-shift, three-shift, and four-shift were associated with 13% (OR: 1.13 95% CI: 0.95, 1.34), 22% (OR: 1.22, 95% CI: 1.05, 1.42), and 21% (OR: 1.21, 95% CI: 1.02, 1.42) higher risk of MAFLD, respectively, after full adjustments for demographics, lifestyles, and preexisting chronic conditions. When further adjusted for BMI, the association was attenuated. In the sensitivity analyses, excluding participants with missing values did not materially change the results (Supplementary Table S2).

As an established risk factor for MAFLD, we observed that BMI was significantly associated with shift work duration (β: 0.07, 95% CI: 0.05, 0.09), three-shift (β: 0.22, 95% CI: 0.02, 0.42), and four-shift (β: 0.25, 95% CI: 0.04, 0.46) systems. As shown in Figure 2, we found that BMI mediated a separate 48.5% (34.0, 64.0%), 42.9% (5.0, 99.4%), and 47.5% (4.4, 111.5%) of the association between shift work duration, three-shift system, four-shift system, and MAFLD, respectively.

In the analyses stratified by job category, significant heterogeneity was observed in the association between shift work duration and the risk of MAFLD prevalence (P_interaction = 0.02). For every 1-year increase in shift work, ORs (95% CIs) for maintenance workers, service workers, drivers, and office workers were 1.03 (1.00, 1.05), 1.15 (1.05, 1.25), 1.19 (1.09, 1.29), and 1.07 (1.04, 1.11), respectively (Supplementary Figure S2). Stratified analyses also suggested that the association between shift work duration and MAFLD was more pronounced among participants who slept≥7 h per night (P_interaction = 0.03). Other stratified analyses yielded no significant interactions (Supplementary Tables S3, S4).