Outcomes of subjects who are lean, overweight or obese with nonalcoholic fatty liver disease: A cohort study in China

Study design and participants

The Kailuan cohort (Chinese Clinical Trial registry number: ChiCTR‐TNRC‐11001489) is a prospective cohort established in the Kailuan community in Tangshan, Hebei Province, China, to explore risk factors for noncommunicable diseases in the Chinese population. We enrolled 101,510 participants (81,110 men and 20,400 women, aged 18–98 years) who completed a baseline survey from 2006 to 2007. In addition, we excluded subjects with (1) hepatitis B surface antigen‐positive or missing data (n = 5170); (2) excessive alcohol consumption (>30 g ethanol/day for men and >20 g ethanol/day for women) or missing data (n = 21,196); (3) liver cirrhosis (n = 85); (4) malignancy history (n = 292); (5) ultrasound examination or BMI data missing (n = 838); (6) implausible or extreme values of BMI and weight (BMI < 15.0 kg/m² or >50.0 kg/m² or weight < 30 kg) (n = 22). Finally, 73,907 subjects were enrolled (Figure S1). We performed this study according to the tenets of the Declaration of Helsinki and obtained approval from the Ethics Committee of the Kailuan General Hospital and the Institute of Basic Medicine Sciences Chinese Academy of Medical Sciences. In addition, we obtained written informed consent from all participants.

Definition of NAFLD

NAFLD was diagnosed based on hepatic steatosis on abdominal ultrasonography scan (HD‐15; Philips, the Netherlands) and absence of secondary causes of liver‐fat accumulation according to the American Association for the Study of Liver Diseases Practice Guidance.^{[ 20 ]} As we have excluded secondary causes of liver‐fat accumulation, including hepatitis B virus infection and excessive alcohol consumption, in the study population, participants who met the criteria of fatty liver were considered to have NAFLD.

Outcome variables

The outcomes included all‐cause and liver‐related mortality, CVD, and digestive system and obesity‐related cancers. Liver‐related mortality was defined as death associated with chronic liver disease and cirrhosis, viral hepatitis, hepatobiliary malignancies, and expanded liver diagnosis, including esophageal varices, hepatic failure, toxic liver disease, other inflammatory liver diseases, and other diseases of the liver^{[ 21 ]}; CVD included myocardial infarction, ischemic stroke, and hemorrhagic stroke; digestive system cancers included colorectal, liver, pancreatic, gastric, biliary, small intestine, and esophageal cancer; obesity‐related cancers included colorectal, liver, pancreatic, gastric, biliary, kidney, and esophageal cancer.^{[ 7} , ^{22 ]} We gathered death information from provincial vital statistics offices in Hebei province, China. Cancer and CVD cases were identified by self‐reported information through questionnaires or by annual linkage with the local vital statistics data, the Tangshan medical insurance system, or the Kailuan Social Security Information System. An expert panel then reconfirmed the diagnosis by checking discharge summaries from the hospitals where participants were diagnosed. Clinical experts validated cancer diagnosis through a thorough review of medical records, including pathological, imaging, and blood biomarker findings. The diagnosis of myocardial infarction was determined by the patient's clinical symptoms, electrocardiogram, and dynamic myocardial enzyme changes following the World Health Organization's Multinational Monitoring of Trends and Determinants in Cardiovascular Disease criteria. Stroke was diagnosed based on neurologic signs, clinical symptoms, and neuroimaging tests, including computed tomographic or magnetic resonance imaging, in line with the World Health Organization criteria.^{[ 23 ]} In addition, according to the International Classification of Diseases, 10th revision, we coded and validated cancer, CVD, and liver‐related death cases by qualified coders (Table S1). Follow‐up ended at the date of death, CVD, cancer diagnosis, or information collection termination of death (December 31, 2019), cancer (December 31, 2017), and CVD (December 31, 2019). Death causes were collected before December 31, 2016, so follow‐up ended at the date of death or information collection termination of death (December 31, 2016) when analyzing the risk of liver‐related death.

Covariates at baseline

We conducted questionnaire surveys to collect sociodemographic characteristics (age, sex, education level), lifestyle factors (smoking, alcohol intake, physical activity), and medical history. Alcohol consumption >0 g ethanol/day was defined as drinkers. We measured each subject's height and weight and calculated BMI using weight divided by height squared. Waist circumference was measured, and central obesity was defined as waist circumference ≥90 cm for men and ≥85 cm for women. Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg, medication history, or self‐reported hypertension history. In addition, we measured the serum glucose, total cholesterol, triglycerides (TGs), low‐density lipoprotein (LDL), high‐density lipoprotein (HDL), alanine aminotransferase (ALT), and C‐reactive protein (CRP) of each subject by using an auto‐analyzer (Hitachi 747; Hitachi, Tokyo, Japan) at the central laboratory of Kailuan General Hospital. We defined diabetes as fasting blood glucose concentration ≥7.0 mmol/L, history of medication, or self‐reported history. The enzyme‐linked immunosorbent assay was applied to test hepatitis B surface antigen (Shanghai Kehua Bio‐Engineering).

Statistical analyses

Characteristics are presented as median (interquartile range [IQR]) or mean ± SD for continuous variables and number (percentage) for categorical variables. We compared baseline characteristics among different groups by analysis of variance or Kruskal‐Wallis test for continuous variables and the Pearson chi‐square test for categorical variables.

We used the restricted cubic spline method to evaluate whether BMI ≥ 23.0 kg/m², recommended for the diagnosis of Asian adults who are overweight,^{[ 24 ]} was suitable for classification when exploring the outcomes for NAFLD by different BMI states in the Chinese population. We used the multivariable‐adjusted restricted cubic splines models, which were adjusted for sex, age groups, education level, smoking status, alcohol drinking status, physical activity, diabetes, hypertension, central obesity, TG, LDL, HDL, CRP, and ALT, to investigate the dose–response relationship between BMI and outcomes, with three knots, representing the fifth, 50th, and 95th percentiles of BMI, and BMI of 23.0 kg/m² as a reference point. We categorized the subjects into four groups: subjects with LN (NAFLD with BMI < 23 kg/m²), subjects with overweight or obese NAFLD (ON; NAFLD with BMI ≥ 23 kg/m²), subjects who were lean but without NAFLD (LNN), and subjects who were overweight or obese without NAFLD (ONN).

We used Kaplan‐Meier analysis to estimate the cumulative incidence of events and compared the incidence difference by log‐rank test. Compared to the LNN group, we used the traditional Cox regression model to estimate hazard ratios (HRs) and 95% CIs of death for the LN, ON, and ONN groups. The Fine and Gray models were used to estimate the risk difference in CVD, digestive system cancers, obesity‐related cancers, site‐specific cancers, and liver‐related death between LNN and the other three groups by adjusting the competing risk of death. We conducted three models: model 1 adjusted for age and sex, model 2 further adjusted for education level, smoking status, alcohol drinking status, and physical activity; and model 3 further adjusted for diabetes, hypertension, central obesity, TG, LDL, HDL, CRP, and ALT. Pairwise comparisons were also conducted. We also performed subgroup analyses by dividing patients with LN into subgroups according to the differential characteristic between those who developed adverse outcomes and those who did not.

To test potential interaction, we conducted stratified analysis by age, sex, diabetes, hypertension, smoking, drinking, and central obesity. In addition, sensitivity analyses were performed to assess the robustness of the results by excluding participants who died, had incident CVD, and developed cancers within the 3 years of follow‐up. We also excluded participants with a history of stroke, ischemic heart disease, diabetes, or smoking at baseline to minimize the potential reverse causation due to the effects of preexisting conditions or smoking on baseline BMI. All data were analyzed using SAS software, version 9.4 (SAS Institute, Cary, NC), with two‐sided tests and p < 0.05 as statistically significant.

Baseline characteristics of the cohort

Our final cohort included 73,907 subjects (Figure S1). Baseline characteristics of study participants are summarized in Table 1. The prevalence of NAFLD was 31.4%. Participants with LN were the oldest among the four groups. The prevalence of hypertension and diabetes and diastolic blood pressure, systolic blood pressure, fasting blood glucose, total cholesterol, TG, CRP, and ALT were highest in subjects with ON and lowest in non‐NAFLD groups.

Dose–response relationship between BMI and risk of death, CVD, and cancers for subjects with and without NAFLD

Using multivariable‐adjusted cubic spline analyses, we found a U‐shaped association between BMI and all‐cause death in subjects with NAFLD (P _overall < 0.001, P _nonlinear < 0.001; Figure 1A1) and without NAFLD (P _overall < 0.001, P _nonlinear < 0.001; Figure 1A2), with higher mortality risks in the subjects with lower or higher BMI. Similar trends were also found in liver‐related death (Figure 1B).

The relationship between BMI and CVD differed between subjects with and without NAFLD. CVD risk presented a significant linear association in the NAFLD group (P _overall = 0.012, P _linear = 0.013; Figure 1C1), with the lowest risk in those with BMI < 23 kg/m²; however, CVD risk showed a rapid increase first and then a slightly decreasing trend in the non‐NAFLD group (P _overall < 0.001, P _nonlinear = 0.003; Figure 1C2).

Although results for digestive system cancers (Figure 1D) and obesity‐related cancers (Figure 1E) were not significant, a slightly increased risk in subjects with BMI < 23 kg/m² in NAFLD was observed (P _overall > 0.05, P _nonlinear > 0.05).

Association between LN and ON with death

Over a median follow‐up of 13.1 (IQR, 12.8–13.2) years, 8913 deaths occurred. The mortality rate in the LN group was higher than in the other three groups (log‐rank test, p < 0.001; Figure 2A; Table S2).

Using the LNN group as reference, LN showed a significantly higher risk of death (adjusted HR, 1.30; 95% CI, 1.14–1.49) whereas the ON (adjusted HR, 0.86; 95% CI, 0.80–0.91) and ONN (adjusted HR, 0.84; 95% CI, 0.80–0.89) groups appeared to have a decreased probability of death (Table 2). Due to the U‐shaped association between BMI and death in NAFLD and non‐NAFLD groups, we further divided the participants into three groups by BMI: <23 kg/m² (lean), 23 to 35 kg/m² (overweight to class I obesity), and ≥35 kg/m² (class II+ obesity). The protective association was only found in those with overweight to class I obesity in both NAFLD and non‐NAFLD groups (Table S3), with HRs of 0.85 (95% CI, 0.80–0.91) and 0.84 (95% CI, 0.80–0.89), respectively.

After adjusting the covariates, a pairwise comparison (Table 3) showed that the mortality risk in the LN group increased by 52% (95% CI, 33%–74%), 54% (95% CI, 35%–76%), and 30% (95% CI, 14%–49%) compared to the ON, ONN, and LNN groups, respectively. Compared to subjects who were lean, pairwise comparisons found that subjects who were overweight or obese decreased the death risk by 16% (95% CI, 11%–20%) in the non‐NAFLD population; however, the risk reduction increased to 34% (95% CI, 25%–42%) in the NAFLD population.

The LN group had the highest incidence of liver‐related death (log‐rank test, p = 0.053; Figure 2B; Table S2). After adjusting covariates, LN showed a significantly higher risk of liver‐related death compared to the LNN group (Table 2). Pairwise comparison (Table 3) showed that the LN group had a higher risk of liver‐related death than the ON and LNN groups, with HRs of 2.77 (95% CI, 1.23–6.24) for LN versus ON and 2.67 (95% CI, 1.18–6.03) for LN versus LNN.

Association between LN and ON with CVD

A total of 5618 incident CVDs developed after a median of 13.1 (IQR, 12.8–13.2) years of follow‐up. The ON group had the highest incidence of CVD, followed by the ONN group (log‐rank test, p < 0.001; Figure 2C; Table S2). After adjusting covariates, the trend remained (Table 2). Compared to the LNN group, the ON group increased CVD risk by 39% (95% CI, 27%–52%), followed by the ONN group by 19% (95% CI, 10%–28%). The increased hazard in the LN group was only observed in the univariate model (HR, 1.28; 95% CI, 1.05–1.57) and the lifestyle factors‐adjusted model (HR, 1.30; 95% CI, 1.06–1.59). However, there was no significant difference between LN and LNN groups (HR, 1.07; 95% CI, 0.87–1.32) after further adjusting for metabolism‐related factors.

Further pairwise comparison (Table 3) showed that the ON group had a higher risk than the others, with HRs being 1.39 (95% CI, 1.27–1.52) for ON versus LNN, 1.30 (95% CI, 1.06–1.58) for ON versus LN, and 1.17 (95% CI, 1.10–1.25) for ON versus ONN. The individuals who were overweight or obese had an increased risk of 30% (95% CI, 6%–58%) for CVD in the NAFLD group but 19% (95% CI, 10%–28%) in the non‐NAFLD group.

Association between LN and ON with cancers

Over a median follow‐up of 11.0 (IQR, 10.7–11.2) years, 851 incident digestive system cancers and 926 incident obesity‐related cancers developed. There was a significantly different risk of digestive system cancers among the four groups (log‐rank test, p < 0.001; Figure 2D; Table S2). Compared to the LNN group, only the LN group showed a significantly elevated digestive system cancer risk (adjusted HR, 2.12; 95% CI, 1.46–3.08); however, all three groups had increased obesity‐related cancers, with adjusted HRs increasing from 1.21 (95% CI, 1.01–1.46) in the ONN group to 1.30 (95% CI, 1.05–1.61) in the ON group and 2.04 (95% CI, 1.40–2.96) in the LN group (Table 2). Site‐specific analysis indicated a higher hazard of colorectal cancer in the LN group (HR, 2.20; 95% CI, 1.11–4.34) and the ON group (HR, 1.65; 95% CI, 1.13–2.41) than in the LNN group. In addition, subjects with LN had a significantly higher risk of esophagus cancer (HR, 6.46; 95% CI, 3.00–13.92). There were no cases in the LN group of kidney and small intestine cancer. An increased hazard was observed in the ON group (Table 4).

Further pairwise comparison (Table 3) showed that LN had the highest risks of digestive system cancers or obesity‐related cancers. NAFLD had higher risks than non‐NAFLD in subjects who were lean; however, no significant association was observed in subjects who were overweight or obese. In addition, a lower risk of obesity‐related cancers was observed in the LNN group compared to the ONN group.

Associations among the LN group and adverse outcomes

Baseline characteristics of subjects with LN with and without adverse outcomes are presented in Table S4. Subjects with LN who died or who had CVD, digestive system cancers, and obesity‐related cancers were older and had a higher prevalence of hypertension than those who did not have adverse outcomes. Therefore, we analyzed this group based on hypertension. The relationship among subjects with LN and adverse outcomes is shown in Table 5. Patients with LN with hypertension had a higher risk of death (HR, 1.62; 95% CI, 1.37–1.92), liver‐related death (HR, 2.92; 95% CI, 1.01–8.42), CVD (HR, 1.31; 95% CI, 1.01–1.70), digestive system cancers (HR, 2.63; 95% CI,1.67–4.15), and obesity‐related cancers (HR, 2.56; 95% CI, 1.62–4.03); those without hypertension had a similar risk as those with LNN.

Stratified analyses

Stratified analyses by sex, age, central obesity, diabetes, hypertension, drinking, and smoking are shown in Tables S5‐1 to S5‐5. Age was observed to modify the association between NAFLD status and death (p for interaction = 0.012). Compared to the LNN group, the increased mortality risk in the LN group was more prominent in the younger population, with HR reaching up to 2.57 (95% CI, 1.59–4.18) in those aged < 45 years. The relationship between NAFLD status and CVD was also observed to be modified by sex (p for interaction = 0.014), central obesity (p for interaction = 0.062), and smoking status (p for interaction = 0.043). Men or those with central obesity had a higher liver‐related death risk in the LN group than in the LNN group. Men, those with central obesity, or current smokers had a higher CVD risk in the ON and ONN groups compared to the LNN group. No other interactions were found.

Sensitivity analyses

Conclusions remained consistent after excluding the outcomes within 3 years of follow‐up (Table S6) and the subjects with a history of stroke, ischemic heart disease, diabetes, or smoking at baseline (Table S7).