The data were extracted from that collected during the Kailuan study (trial registration number, ChiCTR-TNC-1100148; trial registration site, http://www.chictr.org.cn/index.aspx). This is a community-based prospective cohort study being performed in Kailuan, Tangshan City, northern China. Detailed information regarding the study design and procedures of the Kailuan study has been published previously (16). Briefly, a total of 101,510 participants (81,110 men and 20,400 women, aged 18–98 years) were recruited for the initial survey, and they completed questionnaires and underwent physical examinations and laboratory testing in the 11 hospitals in the Kailuan community between June 2006 and October 2007. Re-examinations were conducted every 2 years until December 31, 2019 or death.

Of the participants recruited for the initial survey, we excluded 1,219 for whom baseline ultrasonographic data were not available, 24,338 for whom information regarding alcohol intake was not available or who consumed an excessive amount of alcohol (defined as an alcohol intake ≥30 g/day for men and ≥20 g/day for women), 3,160 who were positive for hepatitis B surface antigen (HBsAg) or for whom this information was not available, 125 with liver cirrhosis, 366 who had previously experienced ischemic or hemorrhagic stroke, myocardial infarction, or cancer, and 450 for whom detailed data were not available. Thus, 71,852 participants who were recruited at baseline were included in the present study (Figure 1).

Experienced radiologists who were blinded to the clinical presentation and laboratory findings of each participant performed abdominal ultrasonography. A high-resolution B-mode topographical ultrasound system with a 3.5 MHz probe (Acuson X300, Siemens, Munich, Germany) was used.

NAFLD was diagnosed on the basis of the presence of at least two of the following abnormal findings on abdominal ultrasonography, according to the criteria for the assessment and management of NAFLD in China (17): (1) increased echogenicity of the liver near-field region with deep attenuation of the ultrasound signal; (2) hyperechogenicity of liver tissue (“bright liver”), compared with the echogenicity of the kidney cortex; and (3) vascular blurring. NAFLD was diagnosed in the presence of fatty liver and in the absence of excessive alcohol consumption or other causes of chronic liver disease (18).

The severity of NAFLD was defined ultrasonographically, as previously described (19), as mild (diffuse increase in fine echoes from the liver parenchyma), moderate (diffuse increase in fine echoes, with impaired visualization of the intrahepatic vessel borders and diaphragm), or severe (diffuse increase in fine echoes, with no visualization possible of the intrahepatic vessel borders or diaphragm).

Blood samples were obtained from the anterior elbow of the participants under fasting conditions (at least 8 h) in the morning. Fasting blood glucose (FBG) concentration was measured using the hexokinase/glucose-6-phosphate dehydrogenase method (Mind Bioengineering Co. Ltd, Shanghai, China). DM was diagnosed if participants had an FBG of ≥7.00 mmol/L, self-reported a physician diagnosis, or self-reported the use of antidiabetic medication. Pre-DM was defined using an FBG of 5.60–6.99 mmol/L for participants who were not taking antidiabetic medication. Participants without pre-DM or DM were defined as having normal NGR.

The primary outcome of the present study was an incident CVE, including myocardial infarction (MI), hemorrhagic stroke (HS), and ischemic stroke (IS). The CVEs were identified by ICD 10th edition codes, as described previously (16). Information regarding incident CVEs was gathered at biennial interviews and from annual discharge lists, which were reviewed and adjudicated by a committee consisting of three experienced blinded physicians from the 11 Kailuan hospitals. Information regarding mortality was collected from the provincial vital statistics offices and reviewed by clinicians. In line with the World Health Organization's Multinational Monitoring of Trends and Determinants in Cardiovascular Disease (MONICA) criteria (20), MI was diagnosed on the basis of clinical symptoms, high circulating cardiac enzyme activities, or biomarker concentrations, as well as electrocardiography. Stroke was diagnosed according to the World Health Organization criteria (21) on the basis of neurological signs, clinical symptoms, neurological images (computed tomography or magnetic resonance imaging), and other diagnostic reports, as described previously (16).

Plasma alanine aminotransferase (ALT) activity and the plasma total bilirubin (TBIL), creatinine (CR), uric acid (UA), triglyceride (TG), total cholesterol (TC), high-density lipoprotein-cholesterol (HDL-C), and low-density lipoprotein-cholesterol (LDL-C) concentrations were measured in the central laboratory of Kailuan hospital using an auto-analyzer (Hitachi 747; Hitachi, Tokyo, Japan). Hepatitis B surface antigen (HBsAg) was assayed using an enzyme-linked immunosorbent assay (Shanghai Kehua Bio-Engineering, Shanghai, China). Plasma high-sensitivity C-reactive protein concentration was measured using a high-sensitivity particle-enhanced immunonephelometric assay (Cias Latex CRP-H, Kanto Chemical Co Inc, Tokyo, Japan). Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation (22). Hyperlipidemia was defined using a compatible medical history, a fasting total cholesterol (TC) concentration ≥5.18 mmol/L, or a triglyceride (TG) concentration ≥1.7 mmol/L.

Blood pressure (BP) was measured by well-trained operators between 07:00 and 09:00 on the day of the physical examination. Smoking, caffeine consumption, and physical activity were prohibited for 30 min prior to the measurement and the participants were seated quietly for at least 5 min beforehand. Mercury sphygmomanometers were used to measure BP three times at intervals of 1–2 min, and the mean value was obtained. To avoid measurement bias, the sphygmomanometers were regularly checked and calibrated. The heart rate, height, and weight of the participants were measured by trained field workers during the surveys. Body mass index (BMI) was calculated as weight (kg) divided by the square of height (m²).

Information regarding sociodemographic parameters (e.g., birth date and sex), lifestyle (e.g., smoking status, alcohol consumption, and physical activity) and previous medical history (e.g., hypertension, diabetes, and the use of antihypertensive agents) was collected by trained interviewers using questionnaires, as described previously (16). Smokers were defined as participants who had a history of smoking or who were current smokers, and those who had a history of alcohol consumption or who currently consumed alcohol were defined asdrinkers. Physical activity was defined as engaging in aerobic exercise ≥3 times/week, for ≥30 min/session.

Statistical analyses were performed using SAS version 9.4 (SAS Institute, Inc, Cary, NC, USA) and R, version 3.5.2 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are presented as mean ± standard deviation (SD) or median with interquartile range, and categorical variables are described as number or percentage. The Kolmogorov-Smirnov test was used to identify the distributions of the variables.

One-way ANOVA, the Kruskal-Wallis test or the chi-square test was used to compare the data according to the glucose control status of the participants, as appropriate. The event-free survival rates of the groups were determined using the Kaplan-Meier method and compared using the log-rank test. Univariate and multivariate cox proportional hazard regression models were used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for incident cardiovascular disease for each group. Model 1 was unadjusted. Model 2 was adjusted for age, sex, BMI (≥30, 25–29.9, 18.5–24.9, or <18.5 kg/m²), physical activity (<3 or ≥3 times/week), and smoking status (never a smoker or a smoker). Model 3 was further adjusted for hyperlipidemia, the use of lipid-lowering medication, HDL-C, systolic blood pressure (SBP), high-sensitivity C-reactive protein (hsCRP), and eGFR. We conducted trend tests in the Cox proportional hazard regression models, with the median value of each group being included as a continuous variable.

We calculated the C-statistic and ΔC-statistic to evaluate the efficiency of the models and the incremental value of adding the combination of NAFLD and glucose control status to the original model. Sensitivity analyses were performed by excluding all the participants who experienced a CVE within the first year of follow-up to assess the robustness of the results. All the statistical tests were two-sided and P < 0.05 was considered to represent statistical significance.

The 71,852 participants were allocated to the NGR (n = 50,820), pre-DM (13,990), and DM (7,042) groups and the baseline characteristics of the sample are shown in Table 1. Participants with pre-DM or DM were more likely to be older; to have a large waist circumference, high BMI, high blood pressure, and high FBG; to be smokers and drinkers; to have high lipid (TC, LDL-C, and TG), creatinine, TBIL, and HsCRP concentrations; to have high AST activity; and to have low eGFR (all P < 0.001) than the NGR group. Moreover, the percentages of participants with hypertension or hyperlipidemia, and of those taking antihypertensive or antidiabetic medication, increased from the NGR to the pre-DM and the DM group (P < 0.001).

During a median follow-up period of 13.01 (0.64) years, 6,037 CVEs occurred (1,385 MIs and 4,652 strokes). Compared with participants without NAFLD, those with NAFLD had a 1.123-fold higher risk (95% CI 1.059–1.191) of CVEs after adjustment for confounding factors in the three models (Supplementary Table S1). As shown in Figure 2A, Kaplan-Meier analysis showed that participants with NAFLD had a lower event-free survival rate than those without (P < 0.001).

The participants were also allocated to four groups (no NAFLD, mild NAFLD, moderate NAFLD, or severe NAFLD) according to the severity of their NAFLD. Relative to participants without NAFLD, the HRs for participants with mild or moderate NAFLD were 1.449 (95% CI 1.365–1.539) and 1.694 (95% CI 1.568–1.831), respectively. Moreover, those with severe NAFLD were at a 1.957-fold higher risk of a CVE. This association remained significant after adjustment for potential confounders in the three models (all P < 0.001, Supplementary Table S1). Kaplan-Meier analysis showed that participants with more severe NAFLD were more likely to experience an event (P < 0.001, Figure 2B).

The prevalence of CVEs in participants with pre-DM or DM group during the study were 9.2 and 16.4%, respectively. Univariate Cox regression analysis demonstrated that pre-DM and DM were associated with 1.333-fold (95% CI 1.251–1.420) and 2.605-fold (95% CI 2.438–2.784) higher risks of CVEs, respectively, compared with the NGR group. The adjustment for confounding factors in the three models only slightly attenuated these relationships (Supplementary Table S2).

Furthermore, we assessed the risk of CVE according to both the presence of NAFLD and glucose metabolism status. Univariate Cox proportional hazard regression analyses revealed that NAFLD was associated with a higher risk of CVE across the glucose metabolism status compared with the NGR and non-fatty liver group (HR 1.243, 95% CI 1.225–1.262, P for trend < 0.001, Table 2). Notably, NAFLD plus pre-DM and NAFLD plus DM were associated with 1.809- and 3.003-fold higher risks of CVEs, respectively (both P < 0.001). Adjustment for potential confounding factors in the three models did not abrogate these associations, with an adjusted HR of 1.267 (95% CI 1.151–1.395, P < 0.001) for the pre-DM plus NAFLD group and an adjusted HR of 1.829 (95% CI 1.666–2.008, P < 0.001) for the DM plus NAFLD group (Table 2). The risks of incident CVEs also remained significant after the exclusion of participants who experienced events within the first year of follow-up (Supplementary Table S3).

Kaplan-Meier analysis with the log-rank test demonstrated that there was a significant difference in the cumulative event-free survival rates among the participants under different glucose metabolism status. Participants with DM were the most likely to experience an event (P < 0.001, Figure 2C). We also evaluated the prognostic utility of combining NAFLD and glucose metabolism status, and found that participants with both NAFLD and DM had the lowest cumulative event-free survival rate, followed by NAFLD plus pre-DM, compared with the reference group (non-NAFLD and NGR group, all comparisons P < 0.01, Figure 2D).

In the predictive model comprising conventional risk factors, the C-statistic was 0.7331 (95% CI 0.7274–0.7388, Table 3). The addition of the presence of NAFLD to this model improved the C-statistic [ΔC-statistic 0.0008 (0.0004–0.0013), P < 0.001], and the addition of glucose metabolism status to the original model also significantly improved the C-statistic to 0.7396 (0.7340–0.7452) [ΔC-statistic 0.0065 (0.0051–0.0078), P < 0.001]. Finally, the incorporation of both the NAFLD and glucose metabolism status into the original model further increased the C-statistic to 0.7398 (0.7341–0.7454) [ΔC-statistic 0.0066 (0.0053–0.0080), P < 0.001].