A nomogram for estimating the probability of nonalcoholic fatty liver disease in a Chinese population: A retrospective cohort study.

Studies have showed that dyslipidemia is closely related to nonalcoholic fatty liver disease (NAFLD). However, less attention has been paid to the relationship between early dyslipidemia and long-term risk of NAFLD. Therefore, we aimed to develop a simple-to-use nomogram to predict early dyslipidemia and long-term risk of NAFLD onset.A retrospective cohort study including 3621 employees (including retirees) from 7 companies was conducted between 2012 and 2019. Anthropometric, potential laboratory parameters and abdominal ultrasound were performed at baseline and after a 5-year follow-up. Cox proportional hazards model was used to determine predictors for NAFLD onset. The effects of lipids, age, body mass index (BMI), and serum uric acid (UA) on NAFLD were evaluated with the use of Kaplan-Meier curves (log-rank test). A nomogram was developed based on the Cox proportional hazard model and a 2-piecewise linear regression model. The accuracy of model was evaluated according to the area under the receiver operating characteristic curves.A total of 1545 subjects were included in the final analysis. The mean follow-up time was 52 ± 6.6 months. Of the total subjects, 77.61% were male and 22.39% were female. The mean age at the time of initial visit was 45.21 ± 11.20 years. Five hundred fifty-five subjects (35.92% of all subjects) were finally diagnosed with NAFLD. Variables in the nomogram included age, BMI, triglycerides, high-density lipoprotein, low-density lipoprotein, and UA. The accuracy of the nomogram for predicting 5-year cumulative occurrence of NAFLD was 0.8135 (95% confidence interval: 0.7921-0.8349), and the sensitivity and specificity were 0.8108 and 0.6960, respectively.The combination of age, BMI, triglycerides, high-density lipoprotein, low-density lipoprotein, and UA translated into a nomogram can reliably estimate the incidence of NAFLD within 5 years. It may serve as a decision support tool to determine whether to intervene at an early stage.

Introduction

Nonalcoholic fatty liver disease (NAFLD) is an important issue of global public health in the 21st century.[ Multiple studies have shown that the global prevalence of NAFLD is estimated to be 6% to 35%, with a median of 20% for the general public.[ The disease affects 15% to 20% of adults in China and continues to increase due to the prevalence of overweight and obesity in the Chinese population.[ NAFLD covers a range of liver diseases, ranging from simple steatosis to non-alcoholic steatohepatitis (NASH), with varying degrees of fibrosis, which may eventually develop into cirrhosis.[ Simple steatosis is considered benign and progresses slowly over the years, while NASH may develop into cirrhosis and hepatocellular carcinoma.[ Although the effects of NAFLD on cardiovascular diseases are controversial,[ the incidence of NAFLD may significantly increase the risk of T2DM and chronic kidney disease.[

Therefore, it is of utmost importance to identify those who may further develop NAFLD. Early intervention in these “at risk” patients may prevent the development of NAFLD through education on healthy diet, physical activity, weight loss, and prevention of weight gain.[

Dyslipidemia is a well-documented influencing factor of NAFLD, which is usually characterized by elevated levels of low-density lipoprotein (LDL) cholesterol and triglycerides (TG) or decreased levels of high-density lipoprotein (HDL).[ Excess fat accumulates in hepatocytes in the form of lipid droplets, which are coated by several structural proteins that may be involved in the pathophysiology of liver diseases.[ This intrahepatic lipid accumulation is caused by abnormal lipid metabolism, such as increased systemic lipolysis, increased hepatic free fatty acid uptake and very LDL synthesis, and decreased free fatty acid oxidation and TG export.[ Serum uric acid (UA), an end product of purine metabolism in the liver. Recently, some evidence suggests that elevated UA is also associated with the development or progression of NAFLD.[

Previous studies have shown that age, body mass index (BMI), TG, total cholesterol (TC), HDL, LDL, and other factors are closely related to NAFLD.[ However, most traditional factor analyses based on cross-sectional studies revealed risk factors for NAFLD.[ Previous cross-sectional studies have not been able to determine whether dyslipidemia is a cause or a consequence of NAFLD. In addition, some studies have shown that lipid abnormalities were not detected before NAFLD.[ In particular, there are conflicting data on the role of individual lipids in promoting hepatic fat accumulation.[

In addition, less attention has been paid to the relationship between early dyslipidemia and the long-term risk of NAFLD in general population. Therefore, in order to prevent NAFLD, more attention should be paid to patients with dyslipidemia at an early stage. Our goal was to develop a simple-to-use nomogram to predict early dyslipidemia and long-term risk of NAFLD onset. These tools will help provide early evidence for NAFLD and thus effectively prevent NAFLD.

Materials and methods

Study population

This was a retrospective observational cohort study involving 3621 employees (including retirees) from 7 companies who regularly underwent annual health examination at the Health Examination Center of Guangdong Integrated Chinese and Western Medicine Hospital. Data were collected between 2012 and 2019. In the cohort study, 2274 subjects without NAFLD were initially recruited. Exclusion criteria at baseline included:

alcoholism (male>140 g/w, female >70 g/w);

subjects with incomplete clinical data or lost to follow-up;

impaired fasting glucose (>110 mg/dL), impaired glucose tolerance or known type 2 diabetes;

medical history, including viral hepatitis, autoimmune hepatitis, and other known chronic liver diseases;

hypertension, defined as blood pressure (≧140/90 mm Hg) or hypertension under treatment. Eventually, 1545 subjects were included in the analysis (Fig. 1). The study protocol was conducted in accordance with the Declaration of Helsinki and all data used in this study were approved by the institutional review board of Guangdong Provincial Hospital of Integrated Traditional Chinese and Western Medicine.

Figure 1

The flow chart of population selection. In total, 3621 employees (including retirees) were recruited from the Foshan city of Guangdong Province. Among them, 1347 participants with nonalcoholic fatty liver disease were excluded at baseline. After excluding 729 participants who met at least 1 of the exclusion criteria. the final analysis sample of 1545 participants was obtained.

Data collection

The trained staff would deliver the standardized spreadsheet to collect the general information, such as age, gender, height, weight, systolic blood pressure (SBP), diastolic blood pressure (DBP), medical history, etc BMI was calculated as weight/height2. Data collection in subjects included laboratory evaluation of TG, TC, HDL, LDL, fasting blood-glucose (FBG), Alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), UA, and γ-glutamyl transpeptidase (γ-GT), total bile acid (TBA). The assays above were conducted on an Olympus AU2700 biochemical auto-analyzer (Olympus, Japan). NAFLD was diagnosed by abdominal ultrasonography (ACUSON X700 ultrasound system, Siemens, Japan).

Definition

The diagnostic criteria for NAFLD were in accordance with the Chinese Association of Liver Diseases.[ In general, NAFLD was diagnosed as at least 2 of 3 abnormalities, namely diffuse hyperechoicity of the liver relative to the spleen and kidney, attenuated ultrasound beams and poor visibility of intrahepatic structural details.

Follow-Up

All participants were followed up every 9 to 15 months at the Department of Health Examination Center. Within each follow-up, all participants received laboratory tests (such as TG, TC, HDL, LDL, FBG, ALT, AST, ALP, UA, γ-GT, TBA) and abdominal ultrasonography. The cut-off time was December 2019.

Statistical analysis

Continuous variables were expressed as mean standard deviation or median inter-quartile range, while categorical values were expressed using relative frequencies and proportions. Comparisons of parameters between 2 different groups were conducted with the Student t test and the Mann–Whitney U test for continuous variables with or without normal distribution and with the Chi-squared test for categorical variables. Univariate and multivariate analyses were then performed to define independent factors associated with NAFLD. Besides, we used generalized additive model (GAM) to identify the non-linear relationship between continuous variable and NAFLD according to smooth curve fitting. If the non-linear correlation was observed, a 2-piecewise linear regression model was performed to calculate the threshold effect of continuous variable on NAFLD in terms of the smoothing plot. We used a likelihood ratio test comparing the 1-line linear regression model with a 2-piecewise linear model, when the likelihood ratio test was statistically significant (P < .05), a 2-piecewise linear model was used to analyze the relationship between the continuous variables and NAFLD. The effects of lipids, age, BMI, and UA categories on NAFLD were evaluated with the use of Kaplan–Meier curves (log-rank test). Cox proportional hazards model and a 2-piecewise linear regression were performed to estimate the odds ratio (OR) and 95% confidence interval (CI). The nomogram was developed based on Cox proportional hazards model and a 2-piecewise linear regression model, which allowed us to obtain NAFLD probability estimations. An automated stepwise variable selection method performed on 500 bootstrap samples was used to choose the final model. An area under the receiver operator characteristic (ROC) curve was used as a measure of the diagnostic accuracy. All of the analyses were performed with the statistical software packages R (http://www.R-project.org, The R Foundation) and EmpowerStats (http://www.empowerstats.com, X&Y Solutions, Inc., Boston, MA). A 2-tailed P value < .05 was considered statistically significant.

Results

Demographic and clinical characteristics

A total of 1545 subjects were included in the final analysis. The mean follow-up time was 52 ± 6.6 months. Of the total subjects, 77.61% were male and 22.39% were female. The mean age at the time of initial visit was 45.21 ± 11.20 years. Five hundred fifty-five subjects (35.92% of all subjects) were finally diagnosed with NAFLD. Table 1 presented a summary of patient anthropometric, clinical, and laboratory characteristics. Participants with NAFLD had higher BMI, SBP, and DBP. Serum levels of TC, TG, LDL, FBG, ALP, ALT, AST, UA, and γ-GT were significantly higher in participants with NAFLD than in those without NAFLD (all P < .05). However, Serum levels of HDL were significantly lower in participants with NAFLD compared to those without NAFLD (all P < .001), and levels of serum TBA were found to be non-associated with the presence of NAFLD (P = .160).

Table 1

Demographic and clinical characteristics of subjects.

Variables	All	Non-NAFLD	NAFLD	P-value
N	1545	990	555
Gender, male, %	1199 (77.61%)	722 (72.93%)	477 (85.95%)	<.001
Age, yr	45.21 ± 11.20	44.75 ± 11.80	46.05 ± 10.01	.029
ALT, U/L	21.00 (16.00–32.00)	19.00 (14.00–26.00)	29.00 (21.00–42.75)	<.001
AST, U/L	21.00 (18.00–25.00)	20.00 (17.00–24.00)	23.00 (19.00–28.00)	<.001
ALP, mmol/L	91.39 ± 18.00	89.68 ± 17.90	94.43 ± 17.79	<.001
UA, μmol/L	413.18 ± 99.17	390.53 ± 93.82	453.57 ± 95.64	<.001
TC, mmol/L	5.58 ± 1.15	5.44 ± 1.03	5.83 ± 1.31	<.001
HDL, mmol/L	1.44 ± 0.33	1.51 ± 0.33	1.31 ± 0.27	<.001
LDL, mmol/L	3.39 ± 0.87	3.32 ± 0.84	3.52 ± 0.92	<.001
TBA, mmol/L	3.64 (2.50–4.98)	3.75 (2.70–4.93)	3.47 (2.12–5.00)	.160
TG, mmol/L	1.55 (1.08–2.45)	1.32 (0.95–1.85)	2.25 (1.57–3.45)	<.001
γ-GT, mmol/L	30.00 (21.00–46.00)	25.00 (19.00–36.00)	41.00 (29.00–63.00)	<.001
FBG, mmol/L	5.38 ± 1.42	5.16 ± 1.03	5.77 ± 1.86	<.001
DBP, mm Hg	121.28 ± 13.42	118.46 ± 12.89	126.31 ± 12.87	<.001
SBP, mm Hg	75.03 ± 9.05	72.84 ± 8.27	78.93 ± 9.07	<.001
BMI, kg/m2	24.32 ± 2.77	23.28 ± 2.41	26.18 ± 2.39	<.001

Results of univariate and multivariate analysis for risk factors of NAFLD

A univariate analysis was conducted to evaluate the associations between exposure factors and NAFLD. The results of univariate analysis showed that the levels of TG, TC, LDL, HDL, FBG, γ-GT, ALT, AST, UA, TC, LDL, TG, and ALP were significantly associated with NAFLD. However, TBA was not associated with NAFLD. After adjusting for gender, age, ALT, AST, ALP, UA, TC, HDL, LDL, TG, BMI, SBP, DBP, FBG, and γ-GT, Cox proportional hazards model showed that the risk factors for NAFLD included: ALT (HR = 1.043, 95CI%: 1.030–1.055, P < .001), AST (HR = 0.952, 95CI%: 0.931–0.975, P < .001), UA (HR = 1.908, 95CI%: 1.469–2.479, P < .001), TC (HR = 1.426, 95CI%: 1.162–1.751, P = .001), γ-GT (HR = 1.007, 95CI%: 1.003–1.011, P = .002), FBG (HR = 1.206, 95CI%: 1.076–1.350, P < .001), BMI (HR = 1.46, 95CI%: 1.357–1.583, P < .001). However, HDL was an independent protective factor for NAFLD (HR = 0.405, 95CI%: 0.311–0.527, P < .001) (Table 2).

Table 2

The results of univariate and multivariate analysis.

	Univariate analysis		Multivariate analysis
Exposures	HR (95%CI)	P-value	HR (95%CI)	P-value
Gender
male	1.0
female	0.44 (0.33, 0.58)	<.0001
Age, yr	1.01 (1.00, 1.02)	.0292
ALT, U/L	1.04 (1.03, 1.05)	<.0001	1.043 (1.030, 1.055)	<.001
AST, U/L	1.03 (1.02, 1.04)	<.0001	0.952 (0.931, 0.975)	<.001
ALP, mmol/L	1.01 (1.01, 1.02)	<.0001
UA, μmol/L	1.01 (1.01, 1.01)	<.0001	1.908 (1.469, 2.479)	<.001
TC, mmol/L	1.36 (1.23, 1.50)	<.0001	1.426 (1.162, 1.751)	.001
HDL, mmol/L	0.10 (0.07, 0.16)	<.0001	0.405 (0.311, 0.527)	<.001
LDL, mmol/L	1.31 (1.16, 1.48)	<.0001
TBA, mmol/L	1.00 (0.98, 1.02)	.8316
TG, mmol/L	1.63 (1.49, 1.79)	<.0001
γ-GT, mmol/L	1.02 (1.01, 1.02)	<.0001	1.007 (1.003, 1.011)	.002
FBG, mmol/L	1.50 (1.33, 1.68)	<.0001	1.206 (1.076, 1.350)	.001
DBP, mm Hg	1.05 (1.04, 1.06)	<.0001
SBP, mm Hg	1.09 (1.07, 1.10)	<.0001
BMI, kg/m2	1.74 (1.63, 1.85)	<.0001	1.46 (1.357, 1.583)	<.001

Generalized additive model analysis of risk factors for NAFLD

It was essential to analyze the nonlinear relationship between continuous variables and NAFLD, We then entered all of continuous variable (AGE, ALT, AST, ALP, UA, TC, HDL, LDL, ALB, TG, γ-GT, FGB, DBP, SBP, and BMI) into a generalized additive modelGAM. After adjusting for the aforementioned covariates, there was a non-linear relationship between age, UA, TG, HDL, LDL, BMI, and NAFLD (all the likelihood ratio tests<0.05), while the relationship between γ-GT, ALT, AST, FBG, TC, and NAFLD was a linear relationship (all P < .05). A 2-piecewise linear regression analysis revealed that age<47 (HR: 1.04, 95%CI: 1.01–1.07, P = .008); UA<500mmol/L (HR = 1.0043, 95%CI: 1.002–1.007, P = .0002); TG<2.7mmol/L (HR = 2.129, 95%CI: 1.705–2.659, P<.0001); LDL≧2.67 (HR = 1.50, 95% CI: 1.31–1.77, P < .0001); BMI<24 (HR = 2.70, 95%CI: 2.24–3.25, P < .001) and BMI≧24 (HR = 1.80, 95%CI: 1.42–2.28, P < .001) were the independent risk factors of NAFLD, while HDL≧1.14 (HR = 0.07, 95%CI: 0.04–0.12, P < .0001). Age≧47 (HR = 0.973, 95%CI: 0.997–0.0239, P = .0239) and LDL<2.67 (HR = 0.61, 95% CI: 0.38–0.97, P = .0384) were independent protective factors of NAFLD (Fig. 2, Table 3).

Figure 2

The relationship between age, uric acid (UA), high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglycerides (TG), body mass index (BMI), and nonalcoholic fatty liver disease (NAFLD). A non-linear relationship between them was detected after adjusting for gender, age, alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, UA, total cholesterol, HDL, LDL, TG, BMI, systolic blood pressure, diastolic blood pressure, fasting blood-glucose, and γ-glutamyl transpeptidase. (A) age and NAFLD; (B)UA and NAFLD; (C)TG and NAFLD; (D)HDL and NAFLD; (E)LDL and NAFLD; (F)BMI and NAFLD.

Table 3

The results of 2-piecewise linear regression model.

			95%CI
Variables	Inflection point	Effect size (HR)	Lower	Upper	P-value
Age, yr	<47	1.04	1.01	1.07	.0080
	≧47	0.973	0.950	0.997	.0239
UA, μmol/L	<500	0.997	0.9929	1.0014	.1895
	≧500	1.004	1.0021	1.0066	.0002
TG, mmol/L	<2.70	2.129	1.705	2.659	<.0001
	≧2.70	0.959	0.861	1.070	.4549
HDL, mmol/L	<1.14	0.72	0.15	3.40	.682
	≧1.14	0.07	0.04	0.12	<.0001
LDL, mmol/L	<2.67	0.61	0.38	0.97	.0384
	≧2.67	1.50	1.31	1.77	<.0001
BMI, kg/m2	<24	2.70	2.24	3.25	<.0001
	≧24	1.80	1.42	2.28	<.0001

Kaplan–Meier curves analysis for 5-year cumulative incidence of NAFLD

On the basis of a 2-piecewise linear regression model analysis, we found that the optimal inflection points of age, UA, HDL, LDL, TG, BMI was 47 years, 500 μmol/L, 1.14 mmol/L, 2.67 mmol/L, 2.7mmol/L and 24 kg/m2 respectively. Therefore, according to the optimal Inflection point, patients were classified into the different categories. The cumulative hazards of NAFLD were significantly higher in the age≧47 years, UA≧500 μmol/L, TG≧2.7mmol/L, HDL<1.14mmol/L, LDL≧2.67mmol/L and BMI ≧24 kg/m2 groups (P < 0.001 for all; Fig. 3).

Figure 3

Kaplan–Meier curves of the cumulative incidence rate of nonalcoholic fatty liver disease stratified by the following categories: age (years)<47 or ≧47 (Fig. 2A); UA (μmol/L)<500 or≧500 (Fig. 2B); triglycerides (mmol/L)<2.7 or ≧2.7 (Fig. 2C); high-density lipoprotein (mmol/L)<1.14 or ≧1.14 (Fig. 2D); low-density lipoprotein (mmol/L)<2.27 or ≧2.27 (Fig. 2E); body mass index (kg/m2)<24 or ≧24 (Fig. 2F).

Predictive accuracy of nomogram for NAFLD

Risk factors based on the Cox proportional hazard model and a 2 piecewise linear regression analysis, a nomogram for evaluating the risk of NAFLD was developed for participants. The final proposed model included age, UA, ALT, LDL, TG, and BMI. Six independent predictors with the best performance were incorporated into our model and presented in a nomogram (Fig. 4). The nomogram is used by locating the position on each predictor variable scale according to the value of each participant. Each scale location has a corresponding prognostic point (top axis). The point value of each predictor variable is determined first, and then the total point value (located on the bottom axis) is calculated. Where a straight line is drawn to determine the probability of having NAFLD. Calibration curves were drawn to determine whether the predicted and observed probabilities of NAFLD were consistent. Bootstrap resampling (500 resamples) was used for this plot. The accuracy of the nomogram for predicting 5-year cumulative occurrence of NAFLD was 0.8135 (95% CI: 0.7921–0.8349), and the sensitivity and specificity were 0.8108 and 0.6960, respectively (Figs. 5 and 6).

Figure 4

Nomogram for predicting nonalcoholic fatty liver disease. Values for each variable are individually plotted and correspond to point values assigned from the point scale (top). These point values are then totaled and plotted on the total point scale (bottom), which is used to assign a corresponding value for risk of nonalcoholic fatty liver disease.

Figure 5

Receiver operating characteristic curves for the accuracy of the nonalcoholic fatty liver disease (NAFLD). Nomogram in patients with NAFLD. The overall predictive accuracy of the nomogram for NAFLD was 0.8135 (95%CI:0.7921–0.8349), and the sensitivity and specificity were 0.8108 and 0.6960, respectively. AUC = area under the curve, CI = confidence interval, ROC = receiver operating characteristic.

Figure 6

Calibration curves of the nomogram in the cohort. The black line indicates perfect prediction. The red line means the predictive performance of the nomogram. The blue region represents a 95% confidence interval of the nomogram.

Discussion

Studies have found that NAFLD is an important risk factor for advanced liver disease[ and non-liver-related diseases such as diabetes[ and cardiovascular diseases.[ Therefore, various attempts have been made to diagnose NAFLD.[ However, identification of high-risk individuals may be more valuable than diagnosis. If we can predict these high-risk subjects, we can prevent disease onset by improving lifestyle, such as improving nutrition and increasing exercise.[ To help meet the ongoing need to identify more useful, simple and practical predictors of NAFLD incidence in general practice, we conducted the current retrospective cohort study.

Our investigation demonstrated the prediction effects of metabolic markers for detection of NAFLD among a Chinese people. The research findings showed that the subjects with increased TG, HDL, UA, BMI, FBG, γ-GT, BMI, and age may have higher risks and cumulative incidence of NAFLD. A nomogram was constructed to evaluate 5-years cumulative incidence of NAFLD, consisting of variables of age, UA, LDL, HDL, TG, and BMI. The accuracy of the nomogram for predicting 5-year cumulative occurrence of NAFLD was 0.8135 (95% CI: 0.7921–0.8349), and the sensitivity and specificity were 0.8108 and 0.6960, respectively.

Variables of the constructed nomogram evaluation system were easily available in clinical practice. Namely, we provided a nomogram to identify which patient was more likely to develop NAFLD and possibly for appropriate prevention at an early stage. Our study may help identify high-risk subjects for specific preventive measures. The calibration results confirmed that the model can be used to detect NAFLD.

Several studies have already reported the combination of biochemical and clinical markers to predict NAFLD. Hamaguchiet et al[ studied 3147 participants attending hospital health checkups and found that after an average of 414 days of follow-up, men and women who met the metabolic syndrome criteria at baseline were more likely to have NAFLD. Yu et al[ showed in a prospective study of 7000 employees in China that UA and serum hemoglobin had significantly high predictive value for NAFLD. In addition, Chang et al[ conducted a population-based study of 5237 healthy men and found that higher ALT levels within the reference range at baseline were independent risk factors for NAFLD onset. Zheng et al[ found that elevated serum UA levels were positively correlated with lean-NAFLD independent of other metabolic factors. The ROC curve for detection of mild fatty liver by ultrasonography was 0.735. Miyake T et al[ showed that BMI, ALT, TG, UA, and HbA1c as well as low HDL-C levels were risk factors for NAFLD, and the ROC predicted for NAFLD was 0.807 (0.790–0.824) in man and 0.748 (0.735–0.761) in woman, respectively. Another study[ showed that triglyceride to HDL cholesterol ratio had better predictive effects of onset of NAFLD than other lipid index. The ROC of TG/HDL-C ratio in the male was 0.70 (0.68–0.72) and 0.72 (0.70–0.75) in the female. The above research showed that dislipidemia, UA, BMI and so on were effective indexes for predicting NAFLD, which were similar to our results. However, those above were cross-sectional studies or cohort studies with short follow-up time, which cannot explain the long-term results. In addition, logistic or Cox proportional hazards model analysis was used to analyze the effect of dyslipidemia on NAFLD in most of studies above. However, in our study, we not only used logistic and Cox proportional hazards models to evaluate the relationship between dyslipidemia and NAFLD, but also used the generalized additive modelGAM to clarify the nonlinear relationship, and it was found that age, UA, TG, HDL, LDL, and BMI were non-linear correlated with NAFLD. On the basis of a 2-piecewise linear regression model analysis, we found that the optimal inflection point of age, UA, HDL, LDL, TG, BMI was 47 years, 500 μmol/L, 1.14mmol/L, 2.67mmol/L, 2.7mmol/L, and 24 kg/m2 respectively. And the cutoff values of predictive factors above were useful for clinicians in guiding subjects with dyslipidemia. Therefore, GAM has obvious advantages in dealing with non-linear relations and it can handle the non-parametric smoothing and will fit a regression spline to the data. The use of GAM will help us to better discover the real relationships between exposures and outcomes.

Recently, few studies have been conducted recently using a nomogram to establish a risk prediction model for the development of NAFLD.[ However, they also focus on cross-sectional study, which was hard to infer causality. Nomogram analysis in our study showed that age, UA, TG, LDL, HDL and BMI had better sensitivity and specificity in early prediction of 5-year cumulative incidence of NAFLD. The nomogram helps to meet our needs for visual tools and personalized prevention. Compared with traditional prediction models, nomograms have a visual digital interface, higher accuracy and easier understanding of risk prediction, and can be more directly applied to clinical decision-making. Nomograms are simple, rapid and inexpensive and can be used to effectively identify the risk of developing NAFLD.[ After diagnosis, medical interventions, lifestyle changes, diagnostic management and treatment can be initiated. According to the calculation results, prediction models of age, UA, LDL, HDL, TG, and BMI can be established for clinical practice.

However, some limitations need to be mentioned in this study. First, the study was conducted retrospectively in a single-center, so some inevitable bias must be noted. Second, abdominal ultrasonography is not the perfect gold standard examination for NAFLD and may be missed diagnosis. Third, this patient cohort was from 7 companies, thus these findings may not be applicable to the general population in the community. Fourth, during the follow-up, the patient's medication status was unclear, which would affect the final prediction results. Fifth, most of the subjects included were male, and therefore, gender factors may be a potential bias leading to the nomogram score developed.

Despite these limitations, our study produced some strengths. In particular, to limit heterogeneity in our study, we used the following methods. First, the selection of all participants was done by 2 independent researchers according to predefined inclusion and exclusion criteria and discrepancies were resolved through discussion between these reviewers. Second, in the existing literature on predicting the risk of NAFLD, most of the statistical methods used are logistic or Cox proportional hazard model analysis.[ However, in our study, we not only used logistic and Cox proportional hazard models, the generalized additive model was also used to clarify the non-linear relationship, which may reduce the heterogeneity caused by statistical methods.

Conclusions

The combination of age, BMI, TG, HDL, LDL, and UA translated into a nomogram can reliably estimate the incidence of NAFLD within 5 years. It may serve as a decision support tool to determine whether to intervene at an early stage.

Acknowledgments

We would like to thank the participants in this study for their cooperation. We also thank the academic and non-academic staff of Guangdong Provincial Hospital of Integrated Traditional Chinese and Western Medicine for their help.

Author contributions

Conceptualization: Weining Xie.

Data curation: Shengxin Chen.

Formal analysis: Weining Xie.

Investigation: Weining Xie.

Methodology: Weining Xie.

Project administration: Weining Xie.

Supervision: Weining Xie.

Validation: Weining Xie.

Writing – original draft: Shengxin Chen.

Writing – review & editing: Weining Xie.