Effects of Metabolism-Related Indicators on Nonalcoholic Fatty Liver Disease in Nonobese Population Based on the National Health and Nutrition Examination Survey

,


Introduction
Nonalcoholic fatty liver disease (NAFLD) is a metabolic dysfunction related to liver disease characterized by the excessive deposition of fat in the liver (≥5%) [1,2].Due to lifestyle changes, the incidence of NAFLD has increased signifcantly over the past few decades [3].NAFLD usually presents as obesity or overweight, but NAFLD also occurs in nonobese subjects with similar pathologic severity as obese NAFLD patients [4].Studies showed that NAFLD in the nonobese population accounts for 5%-20% of the total prevalence, including Asia (38.6%),Europe (51.3%), and America (56.6%) [5][6][7][8].NAFLD includes nonalcoholic simple fatty liver, nonalcoholic steatohepatitis and its associated cirrhosis, and hepatocellular carcinoma [9].Nonalcoholic fatty liver disease has become the second leading cause of liver transplantation in the United States [10,11].In the future, NAFLD may become a major cause of end-stage liver disease, seriously afecting public health globally [12].
Previous studies revealed that obesity is a critical factor in the development and progression of NAFLD [13,14].NAFLD is often neglected in the nonobese population, and there is no clear defnition of "nonobese-NAFLD."Weight is not a diagnostic criterion for NAFLD, and multiple factors cause the occurrence of NAFLD.Terefore, it is inaccurate to describe this disease with nonobese NAFLD.So, we use NAFLD in nonobese individuals to describe this disease in the paper [15].Tere is a lack of research on nonobese patients.Te risk factors and clinical characteristics of NAFLD in the nonobese population remain unclear, though BMI, advanced age, and lipid levels may be involved [16,17].Te pathogenesis of NAFLD in nonobese individuals is not fully understood.It may be related to metabolic dysfunction (e.g., insulin resistance and hyperandrogenemia), dietary habits (e.g., sugary drinks and high-fat diet), gut microbiota changes, cytokines (e.g., IL-1 and IL-6), genetic predisposition, and other changes [18][19][20].Furthermore, some studies have found that nonobese patients with NAFLD have a higher risk of cardiovascular disease, type 2 diabetes mellitus (T2DM), and hepatocellular carcinoma [21][22][23].Past researchers have claimed that early weight loss or dietary modifcation in nonobese patients with NAFLD can improve or even eliminate steatosis [24][25][26].Terefore, more studies are needed to conduct early diagnosis and intervention of this disease.Tis will have important implications for the prevention of NAFLD-related end-stage disease and death [27].
NAFLD is associated with metabolic disorders, and the liver is a key factor in metabolic abnormalities [21,28].It is generally accepted that, like obese NAFLD, subjects with NAFLD in the nonobese population have altered glycolipid metabolism and metabolic profles [29,30].A study found that several lipid metabolism-related protein markers have a high diagnostic value for NAFLD in the nonobese population by proteomic profling of plasma in nonobese subjects with or without NAFLD [31].NAFLD in the nonobese population may have a steatosis-like phenotype, characterized by impaired lipogenesis, hypertriglyceridemia, and hepatic steatosis [32].NAFLD in the nonobese subjects tended to have less metabolic disturbances than obese NAFLD subjects.However, NAFLD in the nonobese population was associated with a higher risk of metabolic disease than obese NAFLD [33].Terefore, it is of great signifcance to further study the correlation between metabolism and the risk of NAFLD in nonobese people.
Metabolic disorders are a signifcant risk factor afecting NAFLD development [34,35].We collected data from the National Health and Nutrition Examination Survey (NHANES) during 2017-2018 to discover the role of metabolic-related indicators coexposure in NAFLD and nonobese individuals using WQS and BKMR models.Terefore, the aim of this study was to identify metabolically relevant indicators associated with the development of NAFLD in nonobese individuals for reducing the incidence of NAFLD.

Study Sample.
NHANES is a cross-sectional survey of the health status of the United States population performed by the National Center for Health Statistics of the Centers for Disease Control and Prevention.Te study randomly selected about 5,000 people each year who live in counties in the US to represent national health.Te entire survey included a structured interview, followed by standardized health assessments at mobile examination centers (MECs), which included questionnaires, physical examinations, and laboratory tests.Here, we used the data collected from 2017 to 2018.Written informed consent was obtained from all participants, and the protocol was approved by the National Center for Health Statistics.

NAFLD Defnition.
After excluding hepatitis B or C virus infection and signifcant alcohol consumption, NAFLD was defned as a CAP of ≥274 dB/m [36,37].Liver ultrasound transient elastography is a noninvasive technique used to objectively assess liver fbrosis and steatosis.In the 2017-2018 survey, technicians performed liver ultrasound transient elastography examinations in participants by using the FibroScan model 502 V2 Touch (Echosens).Exams were considered complete if participants fasted for at least 3 hours prior to the exam, there were 10 or more complete LSM, and the liver stifness IQR/median was <30%.CAP values ranged from 100 to 400 dB/m, with higher values indicating higher amounts of fat in the liver.Te device can record CAP as an evaluation index of hepatic fat deposition, and for steatosis of ≥34%, the area under the receiver operating characteristic curve (AUROC) is 0.80, with a sensitivity and specifcity of 79% and 71% [38,39].For quality assurance, NHANES health technicians completed a two-day training program with survey staf and an expert FibroScan technician.

Study Design.
For this analysis, patients aged ≥18 years with controlled attenuation parameter (CAP) of ≥274 dB/m and body mass index (BMI) of <25 kg/m 2 were selected.Patients with hepatitis B or C virus infection, and excessive alcohol consumption (defned as >21 standard drinks per week in males and >14 standard drinks per week in females) were excluded.Among the 9254 patients who participated in NHANES during 2017-2018, we excluded 8350 participants.Te exclusion criteria were as follows: (1) participants without available MEC examination information (n � 550), those aged <18 years (n � 3171), and those with a BMI of ≥25 kg/m 2 or missing BMI (n � 4062); (2) participants with evidence of viral hepatitis B and C (n � 20), those without alcohol intake information (n � 117), and those with signifcant alcohol intake (n � 173); (3) participants with physical limitations for the liver ultrasound transient elastography (n � 48); and (4) participants with missing covariates (n � 210).Finally, we enrolled 904 patients in our study (Figure 1).Based on a self-report, we assessed health conditions such as hypertension, diabetes, smoking, and alcohol consumption."Did you smoke 100 and more cigarettes in your lifetime?"and "Are you currently smoking cigarettes?" Nonsmokers were classifed as those who replied "no" to question 1; ex-smokers were classifed as those who said "yes" to question 1 but "not at all" to question 2; and current smokers were classifed as those who replied "yes" to question 1 and "every day" or "someday" to question 2.An interview was used to acquire the participants' history of hypertension and T2DM.Hypertension was diagnosed based on the information provided by a doctor or other healthcare professional.A history of T2DM was considered for those with a self-reported history of diabetes or HbA1c above 6.4%.Alcohol consumption of >3 alcoholic drinks a day for men and >2 alcoholic drinks a day for women was considered excessive drinking.Hepatitis C virus (HCV) infection was diagnosed by hepatitis C antibody or RNA, while viral hepatitis B virus (HBV) infection was diagnosed by hepatitis B surface antigen.

Statistical Methods.
Categorical variables were expressed as numbers and proportions, and continuous variables were expressed as mean ± standard error (SE).Comparisons between non-NAFLD and NAFLD were performed by using the Rao-Scott chi-square test or t-tests.Te multivariate logistic regression models were used to explore the relationship between the metabolism-related indicators and NAFLD in nonobese patients.Te BKMR and WQS regression models were used to identify the associations of metabolism-related indicators with NAFLD in nonobese patients.BKMR (R package BKMR) is characterized by the exposure-response function modeling and facilitates the visualization of the efect of a single or combined exposure.WQS regression (R package gWQS) integrated the metabolism-related indicators into one index.Te contribution of a single metabolism-related indicator level was weighted according to its relevance to the overall association with the outcome.Te weights were constrained to sum to 1, with higher numbers indicating a larger contribution.Te International Journal of Clinical Practice associations between the metabolism-related indicators and NAFLD in nonobese patients were analyzed by WQS regression.
To construct the nomogram, factors with signifcant predictive value were utilized in the multivariate analysis.By using the R caret package, 904 NAFLD in nonobese patients were randomized into two cohorts, a development cohort of 633 participants and a validation cohort of 271 participants with a ratio of 7.5 : 2.5, which reached the theoretical ratio of 3 : 1. Tis increased the robustness and dependability of our results.Te validation of the nomogram was conducted by using the AUROC, calibration curve, and decision curve analysis (DCA).1000 bootstrap resamples were applied to the AUC value and calibration curve.P < 0.05 was considered statistically signifcant.Data analyses were performed using SAS software (version 9.4) and R software (version 4.1.4). 1 show the characteristics of the study population of 904 US adults, including 788 non-NAFLD participants and 116 NAFLD participants.Tere was no difference in biological sex between NAFLD and non-NAFLD.Te diference in age distribution was statistically signifcant between the two groups.Te age of the NAFLD group was 57.70 ± 1.59 years (41.27 ± 0.88 years for non-NAFLD).Diabetes and hypertension distribution was signifcantly diferent between the two groups.Although diabetes or hypertension was more common among non-NAFLD participants, most 116 NAFLD participants sufered from both diseases (72/116 and 93/116).In nonobesepopulation, there may be a potential association between NAFLD, diabetes, and hypertension.Metabolic-related indicators such as TC, TG, HDL-c, HbA1c, uric acid, BMI, and WHR were statistically signifcant between the NAFLD and non-NAFLD groups (P < 0.05), and indicator levels were higher in NAFLD than in non-NAFLD participants.

Associations between Metabolism-Related Indicators and NAFLD in Nonobese Population. Table 2 presents logistic regression results to show the relationship between nine metabolic-related indicators and NAFLD in nonobese patients.
We found that the waist-to-hip ratio (2.264 (1.535-3.338))and TG (1.009 (1.005-1.013))were associated with the onset of NAFLD in the nonobese population.Higher levels of TG and WHR in nonobese people can increase the risk of developing NAFLD (2.264 and 1.0009-fold).Elevated HDL-c levels (1.016 (1.002-1.031))and HbA1c (1.296 (1.028-1.634))also increase the risk of NAFLD in nonobese people.In the model, we found that the risk of NAFLD increased by 29.6% with each increment unit of serum HbA1c.No statistical diference was observed with the other indicators.

Association of Metabolism-Related Indicators with NAFLD in a Nonobese Population using WQS regression model and BKMR model.
In the covariate-adjusted model, the WQS index was statistically signifcant (P < 0.05) and signifcantly associated with the occurrence of NAFLD in the nonobese population (OR: 5.789, 95% CI: 3.933-8.520).Te weighting of all WQS indices is shown in Figure 2. Te weighting of WHR (0.373) is the most important among all metabolic-related indicators.WHR is the main factor driving the occurrence of NAFLD in the nonobese population.After WHR, TC, TG, and BMI weights were higher in this population (0.162, 0.127, and 0.096, respectively).SBP had the lowest weight (0.003).
Te BKMR model analyzed the relationship between metabolic-related indicators and NAFLD in nonobese people.In the model, biological sex, age, smoking, race, hypertension, and T2DM were adjusted.Figure 3(a) shows the cumulative efect of metabolic-related indicators on NAFLD risk in nonobese people.Nine common metabolic-related markers were used to assess their association with NAFLD risk in nonobese people.Te results showed that the risk of NAFLD increased with increased exposure to metabolic-related markers.When other metabolites were fxed at the median concentration, each metabolite was analyzed in relation to NAFLD (Figure 3(b)).We found that WHR, TG, HDL-c, and HbA1c were positively associated with developing NAFLD.WHR (PIP = 1.0000) and TG (PIP = 1.0000) contributed the most to developing NAFLD (Figure 3(c)).WHR and TG increased from the 25th to the 75th percentile (other metabolic exposures remained fxed at the 75th percentile), and the risk of developing NAFLD increased in nonobese people (Figure 3(d)).Te bivariate exposureresponse function suggests a potential interaction between WHR and TG, synergistically furthering NAFLD in nonobese populations (Figure 4).2) mentioned above were introduced to construct the nomogram (Figure 5(a)).Te vertical line is drawn from the variable value to the vertical scale to calculate the number of points specifed for the variable value.Te points are added up for every variable.Te sum is calculated on the total numerical scale and projected vertically on the bottom axis to assess a person's risk of developing NAFLD in the nonobese population.Te predictive models' ability was evaluated by using the AUROC, which revealed that the combined model had the best AUROC value.In the training cohort, the AUROC value for this prediction model was 0.794 (95% CI: 0.761-0.825)(Figure 5(b)), with a sensitivity and specifcity of 84.0 and 61.6%, respectively.Te AUROC value for this prediction model in the validation cohort was 0.796 (95% CI: 0.743-0.843)(Figure 5(c)), with a sensitivity and specifcity of 82.9 and 66.9%, respectively.DCA shows a satisfactory positive net beneft for most threshold probability, either in the training or validation cohorts (Figures 5(d) and 5(e)).DCA indicates that the model has good potential clinical efects.According to the calibration curve for this dataset, there is a good agreement between the prediction and the actual results for NAFLD and non-NAFLD (Figures 5(f) and 5(g)).

Discussion
International Journal of Clinical Practice nine metabolic-related indicators in the general population on the occurrence of NAFLD in nonobese populations and established a clinical prediction model.For the logistic regression model, there were positive correlations of WHR, HDL-c, TG, and HbA1c with outcomes.Te results of the WQS model showed the association of WHR, TC, TG, and BMI with the occurrence of NAFLD in the nonobese population.In the BKMR model, we found that mixed metabolic markers were signifcantly associated with the development of NAFLD in the nonobese population.Te univariate exposure-response function showed a positive relationship between WHR, HbA1c, and TG.Te bivariate exposure-response function suggested a potential interaction between WHR and TG.From the results of the three models, we found that the WHR and TG were strong risk factors for predicting the development of NAFLD in a nonobese population.
Previous studies have reported BMI and waist circumference as traditional indicators of NAFLD.Xu et al. reported that BMI was associated with the presence and development of NAFLD in nonobese subjects [40,41].Ding et al. claimed that weight loss reduces the risk of NAFLD in a nonobese population, and a lower weight reduction target may sufce in this population [42,43].However, in our study, BMI played a small role in driving the development of NAFLD in nonobese people.Te phenomenon is consistent with Janssen et al.'s fndings that waist circumference is associated with obesity-related health risks compared to BMI [43].Johanna et al. reported that NAFLD in nonobese people occurs in association with an increase in visceral adiposity, independent of BMI [44].Approximately 20% of adults are classifed in the incorrect BMI category on the basis of self-reported height and weight [45].Tis may also account for the diference.BMI is not an accurate indicator of the degree of lipid accumulation in the body, and some of the subjects with a BMI of <25 also had elevated TG levels in our study.
In our study, WHR was identifed as a signifcant risk factor infuencing the occurrence of NAFLD in the nonobese population.A higher WHR was demonstrated in participants with NAFLD due to increased waist circumference than healthy subjects in the nonobese population.According to Shao et al., waist circumference is strongly associated with the development of moderate to severe hepatic steatosis and   6 International Journal of Clinical Practice fbrosis in people with a normal BMI [46].WHR is an important predictor for NAFLD cutof points as WHR varies in diferent populations [47,48].WHR is one of the indicators to assess visceral fat accumulation [49].Compared with healthy nonobese participants, nonobese participants with NAFLD had more visceral fat [50].Accumulation of visceral fat exposes the liver to high levels of free fatty acids, thus exacerbating the accumulation of TG in the liver [51].Terefore, maintaining a normal WHR is essential to prevent the development of NAFLD in nonobese participants.Dyslipidemia is a known risk factor for NAFLD [52].Leung et al. found that high serum triglyceride levels were a risk factor for developing NAFLD in nonobese people and a risk factor for advanced NAFLD-related liver disease [23].Tis is consistent with our study that TG levels are associated with the development of NAFLD.Elevated serum TG levels increase free fatty acids, producing excessive hepatic triglyceride deposition.Studies have reported that NAFLD is associated with dyslipidemia and dysglycemia, and fat deposition, including visceral fat, is an independent risk factor [52]. Insulin resistance (IR) may be a potential mediator of the relationship between TG levels and the development and progression of NAFLD.On the one hand, IR promotes elevated TG levels, which might further aggravate tissue IR [53].On the other hand, IR can boost TG lipolysis in adipose tissue and hepatic TG production from scratch [54].Although the current evidence makes it difcult to speculate on the role of high TG in developing NAFLD, it is one of the markers for NAFLD progression, especially in nonobese people.
Te WQS and BKMR models are recently developed statistical algorithms designed to elucidate metabolic-related indicators' combined efects on NAFLD occurrence in nonobese populations.Te WQS regression model was based on weighting determined empirically through bootstrap sampling to examine the whole-body burden of exposure to metabolically relevant indicators.Te BKMR model can provide a new perspective on nonlinear exposurereexposure response and potential interactions between metabolically related indicators.Te results were the same for the WQS and BKMR models.WHR and TG are signifcant contributors to the overall mixture efect.Te bivariate exposure-response function revealed a potential interaction between WHR and TG.
Our study has several limitations.First, the included participants may be biased because CAP values rather than biopsy-proven parameters were used to defne NAFLD.Second, the data used in this study were extracted from  International Journal of Clinical Practice a cross-sectional survey (NHANES) with a weak ability to investigate causal relationships.Terefore, more prospective studies are needed to validate our fndings.Logistic, WQS, and BKMR regression models were applied to evaluate the association between metabolic-related indicators and NAFLD in the nonobese population.Considering the results of the three models, we found that WHR and TG were most signifcantly associated with NAFLD in the nonobese population.In the future, our study could help clinicians fnd the "hidden" problems associated with NAFLD in the nonobese population, formulate efective prevention strategies, and carry out earlystage medical interventions to reduce the social and medical burden of NAFLD.

Figure 1 :
Figure 1: Flowchart of the study participant.CAP: controlled attenuation parameter.

Figure 2 :Figure 3 :
Figure 2: WQS regression model weights of each metabolism-related indicator.Te fgure shows the weights of each individual metabolism-related indicator contributing to the overall efect.Te models were adjusted for sex, age, smoking, race, hypertension, and T2DM.

Figure 3 :
Figure 3: BKMR model to assess the association between metabolism-related indicators and NAFLD in nonobese patients.(a) Cumulative efect of metabolism-related indicators.(b) Univariate exposure-response functions for each metabolism-related indicator, with other metabolites fxed at their median concentrations.(c) Posterior inclusion probabilities (PIPs) for NAFLD in nonobese population, using BKMR model.(d) Te single-exposure efect of metabolism-related indicators.

Figure 4 :
Figure 4: Bivariate exposure-response relationship between metabolism-related indicators and NAFLD in the nonobese population.

Table 1 :
Characteristics of the study population by nonobese NAFLD status.Weighted mean ± SE for continuous variables or n and weighted proportion for categorical variables.BMI, body mass index; WHR, waist-to-hip ratio; SBP, systolic blood pressure; DBP, diastolic blood pressure; TG, triglyceride; TC, total cholesterol; BMI, body mass index; T2DM, type 2 diabetes mellitus; HDL-c, high-density lipoprotein cholesterol; NAFLD, nonalcoholic fatty liver disease.

Table 2 :
Risk of NAFLD in nonobese population associated with metabolism-related indicators.