Our cohort study was conducted in the Chongming District, Shanghai and the detailed information about study design, eligibility criteria, and sampling has been described previously (20). In brief, a total of 9930 participants received a baseline survey from 2011 to 2012 and 7707 participants completed the follow-up survey in 2014. In our present study, 3577 subjects who had complete baseline and follow-up information were included. Those individuals with diabetes at baseline (n=771), a history of known liver disease including viral or autoimmune hepatitis, liver cancer, or cirrhosis (n=35), abusing alcohol (alcohol consumption >140 g/week in men or >70 g/week in women, n=75), or missing information of fatty liver (n=6) were excluded. Finally, 2690 participants were included for this analysis. Our prospective cohort study was approved by the Ethical Committee of Zhongshan Hospital, Fudan University, and each participant was provided with a written informed consent.

Standard questionnaires were employed to obtain the information about demographic characteristics, lifestyles, history of diseases and medication on site conducted by trained investigators. Body weight and height were obtained in light clothes and bare feet to the nearest 0.1 kg and 0.1 cm, respectively. Body mass index (BMI) was derived from weight in kilograms divided by square of height in meters. Waist circumference (WC) was measured at the level of umbilicus in a standing position. Blood pressure was measured on non-dominant arm at a seated position, three times consecutively with 1-min rest and 10-min interval using an automated electronic sphygmomanometer (OMRON Model HEM-752 FUZZY’ Omron Co., Dalian, China). The average value of three readings was used. Current smokers were defined as participants regularly consuming cigarettes (duration> 6 months) right before the survey. Former smokers were defined as participants with a history of cigarettes consuming for longer than 6 months and having quitted smoking at the time of survey. Similarly, current drinkers were defined as participants regularly consuming alcohol (duration > 6 months) right before the survey. Former drinkers were defined as participants with a history of alcohol consuming for longer than 6 months and having quitted drinking at the time of survey.

Blood samplings were done two times, one at baseline and another at the 3.5-year follow-up. Fasting venous blood samples were collected after at least 10-h fasting. Serum triglyceride (TG), total cholesterol (TC), low density lipoprotein cholesterol (LDL-C), high density lipoprotein cholesterol (HDL-C), alanine aminotransferase (ALT), alanine aminotransferase (AST), gamma-glutamyl transpeptidase (GGT) were measured on the auto analyser (Modular E170, Roche).

A 75g oral glucose tolerance test (OGTT) was conducted and blood samples at 0h and 2h after glucose load were collected. Fasting blood glucose (FBG) and 2-h post-load glucose levels were measured using glucose oxidase method on an auto analyser (Modular P800, Roche). Serum insulin was measured by an electrochemiluminescence assay (Modular E170, Roche). The homeostasis model assessment of insulin resistance index (HOMA_IR) was calculated as fasting insulin (μIU/ml) × fasting glucose (mmol/L)/22.5. Glycated hemoglobin (HbA1c) was measured by high-performance liquid chromatography. According to American Diabetes Association 2010 criteria, diabetes mellitus was defined as 1) self-reported doctor-diagnosed diabetes or taking antidiabetic medications, and/or 2) FBG levels ≥ 7.0 mmol/L and/or, 3) 2h post-load glucose levels ≥ 11.1 mmol/L, and/or 4) HbA1c concentration ≥ 6.5% (48mmol/mol). In the absence of unequivocal hyperglycemia, diagnosis requires two abnormal test results from the same sample or in two separate test samples.

NAFLD was diagnosed by ultrasonography with exclusion of a history of known liver diseases. Liver ultrasonography was operated by two specialists who were blinded to clinical data using a high-resolution B-mode tomographic ultrasound system (Esaote Biomedica SpA, Italy) equipped with a 3.5-MHz probe. Fatty liver was defined as the presence of at least two of the following three findings: 1) diffusely increased echogenicity of the liver relative to kidney; 2) ultrasound beam attenuation; 3) poor visualization of intrahepatic structures. The definitions for NAFLD development were absence of NAFLD at baseline and presence of NAFLD at the end of follow-up, NAFLD remission presence of NAFLD at baseline and absence of NAFLD at the end of follow-up, consistent non-NAFLD absence of NAFLD at baseline till the end of follow-up and sustained NAFLD presence of NAFLD at baseline till the end of follow-up. Non-invasive NAFLD scores was used to assess the non-alcoholic steatohepatitis (NASH). Gholam’s model was calculated as 2.627 * ln AST + 2.13 for diabetics, with a cut-off for predicting NASH of 8.22 (21, 22).

Normally distributed continuous variables were presented as means with standard deviations (SDs), whereas skewed distributed continuous variables were presented as geometrical median and interquartile range. Continuous variables were compared by student t tests and one-way analysis of variance (ANOVA), whereas skewed distributed variables were compared by Mann Whitney U and Kruskal Wallis tests. Categorical variables were expressed as proportions and compared across groups using chi-square tests or fisher exact test. The unadjusted and multivariate adjusted logistic regression analyses were performed to investigate the odds ratios of new development and remission of NAFLD on the risk of incident diabetes. In the NAFLD remission group, logistic regression analysis was further performed to compare the risk of incident diabetes between subjects with or without steatohepatitis at baseline. Statistical analyses were performed on SPSS version 26 (IBM Corp., Armonk, NY). A two-sided p value less than 0.05 was considered as statistical significance.

The present study included 2690 participants free of diabetes at baseline from 2011 to 2012, and followed up in 2014. Diabetes developed in 484 subjects (18.0%). The baseline characteristics of participants by incident diabetes at follow-up were shown in Table 1 . Participants who developed diabetes were older (p = 0.006), had higher BMI and WC, higher concentrations of TC, TG (all p < 0.0001) and LDL-C (p = 0.01) at baseline. The incidence of diabetes was 21.5% in subjects with presence of NAFLD at baseline and 16.1% in subjects without NAFLD at baseline (21.5% VS 16.1%, p < 0.0001).

Table 2 showed the change of NAFLD during 3.5-year follow-up. Of 1746 non-NAFLD subjects at baseline, 580 (33.2%) participants developed NAFLD and 1166 (66.8%) was consistently free of NAFLD throughout the follow-up. Of 944 NAFLD subjects at baseline, 150 participants (15.9%) had NAFLD remission and 794 (84.1%) participants had sustained NAFLD. We then investigated the association of NAFLD alteration and incident diabetes. 170 of 1166 (14.6%) participants with consistent non-NAFLD developed diabetes, whereas 184 of 794 (23.2%) participants with sustained NAFLD developed diabetes. In contrast, 111 of 580 (19.1%) subjects with NAFLD development developed diabetes, and 19 of 150 (12.7%) subjects with NAFLD remission developed diabetes.

Table 3 showed the baseline clinical and biochemical characteristics according to NAFLD alterations during 3.5-year follow-up. The subjects with consistent non-NAFLD were younger by age, had lower BMI, WC, blood pressure, plasma glucose, TG, and higher HDL-C, whereas sustained-NAFLD group was older and had higher BMI, WC, blood pressure, plasma glucose, insulin resistance, and more adverse lipid metabolism at baseline (all p < 0.0001). There were no significant differences in smoking or drinking status across four groups.

Then logistic regression analyses were performed to study the effect of NAFLD alteration on the risk of incident diabetes ( Figure 1 ). After adjustment for age, gender, smoking and drinking status, subjects with NAFLD development had a significantly higher risk for diabetes as compared with sustained non-NAFLD group (OR, 1.43; 95%CI, 1.10-1.86). The risk was not changed after further adjustment for BMI (OR, 1.36; 95%CI, 1.03-1.79) or WC (OR, 1.38; 95%CI, 1.05-1.81). After adjustment for age, gender, smoking and drinking status, subjects with remission of NAFLD had a significantly decreased risk for diabetes as compared with sustained NAFLD (OR, 0.48; 95%CI, 0.29-0.80). The decreased risk was not changed after further adjustment for BMI (OR, 0.49; 95%CI, 0.30-0.83) or WC (OR, 0.49; 95%CI, 0.29-0.82). Since change in NAFLD status is always accompanied with change of BMI or WC, and meanwhile BMI and WC have strong associations with incident diabetes, therefore we assessed the risk after adjustment for BMI change and WC change in the existing model, respectively. The results showed that the association of change of NAFLD status with incident diabetes was independent of the change of BMI and WC ( Figure 2 ).

In order to investigate what might contribute to the incidence of diabetes in subjects whose NAFLD remitted, we then calculated the Gholam’s model to assess their NAFLD severity at baseline. 27 of 150 (18.0%) subjects were identified with NASH at baseline. The incidence of diabetes in subjects with NASH at baseline was obviously higher than in those without NASH at baseline (25.9% VS 9.8%, p=0.048) ( Supplemental Table 1 ). In the age, gender adjusted- logistic model, presence of NASH at baseline increased risk of incident diabetes in participants with NAFLD remission (OR, 3.08; 95%CI, 1.05-8.99). After further adjustment for smoking and drinking status, and baseline BMI, the association persisted (OR, 3.03; 95%CI, 1.01-9.12) ( Table 4 ).

The strengths of the present study are as follows. First, we focused on change in NAFLD status as effects of alcohol consuming and other liver diseases were ruled out. We conducted a well-designed longitudinal cohort and reported the effect of NAFLD status change, including new development and remission of NAFLD on incident diabetes in a 3.5-year Chinese cohort population for the first time. Third, standardized collection of covariates allowed for adjustment for potential confounders. We also have some limitations. First, the study was performed in middle-aged and older Chinese population and cannot be generalized to adolescent or other ethnical populations. Second, NAFLD was determined by ultrasonography, which had limited sensitivity to detect low-level liver fat, limiting the generalizability of our study to earlier stages of NAFLD. NASH were assessed by non-invasive score instead of gold-standard hepatic biopsy. Third, diagnoses of diabetes and NAFLD were only made at baseline and the 3.5-year follow-up, so it might not differentiate which one developed first, and an annual screening for incident diabetes could be helpful.