The prevalence of obesity in children has increased dramatically, putting children at incremental risk of developing chronic non-communicable diseases later in life.1 Non-alcoholic fatty liver disease (NAFLD) is the most common cause of chronic liver disease, affecting 10% of school-aged children and 44–70% of obese children,2 and 25% of the world’s population.3 In China, a study in the Yangtze River delta region reported a prevalence of NAFLD among children and adolescents to be 5.0% (7.5% in boys, 2.5% in girls, 5.6% in subjects with peripheral obesity, 12.9% in those with abdominal obesity, and 44.8% in those with mixed obesity from a cross-sectional survey completed in 2011).4 The principal risk factor for NAFLD is obesity. With excess adipose accumulation at the liver, in the intra-abdominal cavity, and under the dermal layers,5,6 abdominal adipose plays a critical role in the progress of NAFLD due to metabolic inflammation.7,8 Determining optimal prediction parameters for NAFLD stratified interventions could help reverse the progress to severe liver disease in later adulthood.9,10

Choosing appropriate assessments to diagnose NAFLD and estimation of abdominal adipose are key to early targeted health management interventions. Typically, NAFLD is diagnosed by blood tests, or imaging evaluation such as ultrasonography, computed tomography, magnetic resonance imaging, and elastography, among others.11,12,13 Although several blood biochemical markers are used to evaluate a patient with suspected NAFLD, none can distinguish the different stages of steatosis.14 The gold standard for diagnosing and assessing NAFLD severity is liver biopsy; however, liver biopsy is not practical in many healthcare settings due to its invasiveness, sampling error, high cost, and risk of complications.15 Ultrasonography measurement is recommended as the first-line screening method in assessing steatosis for NAFLD because of its practicality, minimum risk, repeatability, low cost, and well-acceptance.12,14

Abdominal obesity is usually estimated by anthropometric method such as waist circumference (WC), waist-to-hip ratio (WHR), and waist-to-height ratio (WHtR). These indicators are also used to assess the risk of NAFLD or metabolic syndrome.4,16,17,18 The principal limitation of anthropometric method is a lack of accuracy in assessing adipose mass and the location of its accumulation. Body composition measures can be used to quantify abdominal adipose, especially using trunk fat index (TFI) and visceral fat area (VFA).19,20 A recent longitudinal study among adults indicated that an endocrinological indicator, namely visceral adiposity index (VAI), can be a predictor for incident NAFLD;21 VAI is a combination of indicators from both endocrinological and anthropometric perspectives. Too many studies focus on indicators based on a single measurement method; there is a dearth of studies comparing the ability of traditional and new indicators to predict NAFLD in children.

In the present study, we sought to compare the performance of multiple abdominal obesity indicators in NAFLD identification in terms of accuracy and fastidious cut-off values through a population-based child cohort. Identification of sensitive, specific, and easy-to-use indicators for weight management can guide the prevention and stratified intervention of NAFLD among school-aged children.


Study design and participants

We enrolled children aged 6–8 years from the child cohort designed to study sensitization, puberty, obesity and cardiovascular risk (PROC) in the urban area of Shunyi District, Beijing. In brief, the PROC used a community-based census-like design with two-staged sampling. After consulting with the Shunyi District Education Commission, we decided to recruit one-half of all grade-one students in elementary schools in urban area for longitudinal follow-up, selecting six schools that had 66.4% coverage of local public non-boarding students, meeting the target recruitment rate. All 2394 first-grade children and their parents were approached in 2018, and 1914 children (80%) and their parents agreed to participate (parents via written informed consent; children via verbal assent). Sequential baseline surveys were conducted from October 2018 to June 2019. Enrollment criteria included being healthy/ no known serious illness, and both parents agreeing to three rounds of specimen collections from their children (blood, urine, feces, saliva, and hair) for biochemical and metal element assays, as well as sequential anthropometric measurement. Exclusion criteria were children with mental illness or disability, poor general health, and/or congenital cardiopulmonary insufficiency.

The trial has been registered in Chinese Clinical Trial Registry (ChiCTR) (, No. ChiCTR2100044027). Results will be reported in accordance with standards described in the Consolidated Standard of Reporting Trials (CONSORT) statement and CONSORT extensions for nonpharmacologic trial (

There were 1350 student participants in the present study (71% of the 1914 children in the cohort) who provided complete data of anthropometric and body composition measurement, blood and urine sample collections, and ultrasound measurements. We excluded 564 children with missing data on body composition measurements or ultrasound imaging (Fig. 1).

Fig. 1: Flowchart for the study.
figure 1

The chart presents the recruitment and enrollment of the participants of the present study.

Ethical consideration

The study protocol was reviewed and approved by the Ethics Committee of Capital Medical University (No. 2018SY82), and met the ethical principles of the Declaration of Helsinki, later amendments, and comparable international ethical standards.

Anthropometric measurements

Anthropometric measurements were performed by trained medical staff from the Capital Medical University and the Shunyi District Center for Disease Control and Prevention. (1) Standing height was measured using a mechanical height meter (Zhenghe Medical Supply Manufacturer, Hebei, China) and rounded to 0.1 cm, averaging two measurements; (2) body weight was measured using a body composition analyzer with light clothes only and rounded to 0.1 kg; (3) body mass index (BMI) was calculated as weight in kilogram divided by height in meter squared (kg/m2); (4) WC was measured twice using inelastic tape measure (ABS 35202, Yiwu Qingyi Corp., Zhejiang, China, validated by steel rulers) at the midpoint horizontal level that links the iliac crest and the lower margin of the 12th rib to 0.1 cm, averaging two measurements; (5) hip circumference (HC) was measured twice using inelastic tape measured at the horizontal level of the widest portion of buttocks/trochanters to 0.1 cm, averaging two measurements; (6) WHR was calculated as WC divided by hip circumference; (7) WHtR was calculated as WC divided by height; (8) Z-score for age and sex of height, weight, and BMI were calculated per 2007 WHO standards,22 and obesity was defined as Z-score of BMI > 2.

Body composition measurement

Body composition was measured using a bioelectrical impedance body composition analyzer (multi-frequency and multi-section contact eight-electrode analyzer H-Key 350, Beijing Seehigher Technology Co., Ltd). Participants were instructed to stand barefoot on foot electrodes after fasting overnight and post-micturition and fecal discharge, as per the users’ manual. Body fat mass, trunk fat mass and other sections’ fat mass, visceral fat area, fat-free weight, total water, muscle mass, and other features were measured and recorded. The body composition measurements taken by bioelectrical impedance analysis has high consistency with the measurement taken by dual-energy X-ray absorptiometry,23,24 and with magnetic resonance imaging (Pearson correlation coefficient was 0.81).25 (1) TFI was calculated as trunk fat mass in kg divided by height in meter squared (kg/m2). (2) VFA was measured, rounding to 0.1 cm2.

Serum lipids and VAI

Serum lipids were measured using an automatic clinical chemistry analyzer Beckman Coulter AU5800, including total cholesterol, triglyceride, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and AST/ALT ratio. VAI was calculated by using the following formula:26

$${{{{{{{\mathrm{Male:}}}}}}}}\,{{{{{{{\mathrm{VAI}}}}}}}} = \frac{{{{{{{{{\mathrm{WC}}}}}}}}}}{{{{{{{{{\mathrm{39}}}}}}}}{{{{{{{\mathrm{.68 + 1}}}}}}}}{{{{{{{\mathrm{.88 \times BMI}}}}}}}}}} \times \frac{{{{{{{{{\mathrm{TG}}}}}}}}}}{{{{{{{{{\mathrm{1}}}}}}}}{{{{{{{\mathrm{.03}}}}}}}}}} \times \frac{{{{{{{{{\mathrm{1}}}}}}}}{{{{{{{\mathrm{.31}}}}}}}}}}{{{{{{{{{\mathrm{HDL}}}}}} \mbox{-} {{{{{\mathrm{C}}}}}}}}}},\\ {{{{{{{\mathrm{Female}}}}}}}}:\,{{{{{{{\mathrm{VAI}}}}}}}} = \frac{{{{{{{{{\mathrm{WC}}}}}}}}}}{{{{{{{{{\mathrm{36}}}}}}}}{{{{{{{\mathrm{.58 + 1}}}}}}}}{{{{{{{\mathrm{.89 \times BMI}}}}}}}}}} \times \frac{{{{{{{{{\mathrm{TG}}}}}}}}}}{{{{{{{{{\mathrm{0}}}}}}}}{{{{{{{\mathrm{.81}}}}}}}}}} \times \frac{{{{{{{{{\mathrm{1}}}}}}}}{{{{{{{\mathrm{.52}}}}}}}}}}{{{{{{{{{\mathrm{HDL}}}}}} \mbox{-} {{{{{\mathrm{C}}}}}}}}}}$$

Ultrasonography measurement

Liver ultrasound imaging was performed by a single certified physician (who is well- trained and experienced in heart, liver, and renal ultrasonic imaging) from Beijing Anzhen Hospital, an advanced third-tiered hospital, using Canon Aplio500 with a multi-frequency convex transducer probe (4–11 MHz) to investigate the presence of NAFLD with hyperechoic texture or a bright liver standard including alteration of the fine echoes, visualization of diaphragm, and blurring intra hepatic vessels. Scores ranged from 1 to 3, with 1 representing either hepatorenal echo contrast or that bright liver was positive, 2 representing that both were positive, and a score of 3 indicating that aside from hepatorenal echo contrastness, bright liver was severe.27 All participants were required to fast overnight to minimize the measurement bias.

Statistical analysis

The main outcome indicator was whether or not the child had NAFLD. The abdominal obesity indicators including WC, WHR, WHtR, VAI, TFI, and VFA were used to predict NAFLD. Descriptive analysis like anthropometric indices and abdominal obesity indicators were presented as mean ± standard deviation, and using independent t-tests, we compared differences between normal and NAFLD groups. Using Spearman’s correlation coefficients, we estimated the association between NAFLD and each abdominal obesity indicator. Using Pearson’s correlation coefficients, we estimated potential relationships between abdominal obesity indicators and other anthropometric indices. We use receiver operating characteristic (ROC) curves to estimate key parameters including the empirical area under the curve (AUC), Youden index [values calculated as (sensitivity + specificity−1)], cut-off values (values with maximum Youden index), sensitivity (the ability to correctly identify patients with a disease) and specificity (the ability to correctly identify people without the disease), positive predictive value (PPV, the probability that participants with a positive screening test truly have the disease) and negative predictive value (NPV, the probability that participants with a negative screening test truly without the disease). Using binormal models to smooth the ROC curve, we estimated the fitted AUC and 95% confidence interval (95% CI), a frequently used tool for parametric ROC analysis that is more robust than empirical ROC curves.28 For convenience of comparison with other studies, we provided empirical ROC curves for each indicator.

Using logistic regression analysis, we identified the association between NAFLD and abdominal obesity indicators with crude and adjusted odds ratios (cOR and aOR) with 95% CI. We assessed the performance of abdominal obesity indicators predicting NAFLD and developed an assessment table with rankings of each indicator. The overall assessment is the ranking sum of predictive ability, risk assessment ability, and easy-to-use cut-off values. The smaller the ranking, the better overall performance. A two-tailed P value of 0.05 was used to determine statistical significance. All data were analyzed using Statistical Analysis System (V.9.4, SAS Institute Inc., Cary, North Carolina, USA).


Sociodemographic characteristics

The prevalence of NAFLD was 7.0% (95 of 1350, 95% CI: 5.7–8.4%) among our urban Chinese children, significantly higher in 674 boys (10.4%) than 676 girls (3.7%). The obesity prevalence was 19.7% (133 of 674) for boys and 10.2% (69 of 676) for girls. Children with NAFLD had higher age-specific height and weight compared to normal children, and had higher obesity indices like Z-score of BMI, WC, HC, WHR, WHtR, VAI, TFI, and VFA (all P < 0.001). The ALT level of children with NAFLD was significantly higher than that of normal children, while the AST/ALT ratio was the opposite (Table 1).

Table 1 Sociodemographic characteristics of the children aged 6–8 from PROC in Beijing, China (N = 1350).

The relationship between abdominal obesity and NAFLD

All six abdominal obesity indicators were significantly correlated with NAFLD after controlling for sex; the three highest Spearman’s correlation coefficients were TFI (0.46), WC (0.45), and VFA (0.45) for boys, and VFA (0.30), TFI (0.30), and WC (0.29) for girls (Table 2).

Table 2 The correlations between abdominal obesity indicators and non-alcoholic fatty liver disease among children aged 6–8 from PROC in Beijing, China (N = 1350).

ROC curve analysis showed all abdominal obesity indicators to have the diagnostic capacity to predict NAFLD and we juxtaposed ROC curves for comparison (Fig. 2). The fitted and empirical curves showed that the predictive ability of WC, VAI, TFI, and VFA was better among girls than boys (Fig. 2a, d–f), while WHR was a superior predictor for boys (Fig. 2b). The prediction capacity of WHtR was similar for girls and boys (Fig. 2c). The ranking for the six best abdominal obesity indicators to predict NAFLD was as follows: TFI, WC, VFA, WHtR, WHR, and VAI among boys (Fig. 2g), and VFA, TFI, WC, WHtR, WHR, and VAI among girls (Fig. 2h). The size of the bubble represents the value of Youden index; the closer the bubble to the upper left corner, the better the predictive ability of the indicator was (Fig. 2i).

Fig. 2: The ROC curves of abdominal obesity indicators predicting non-alcoholic fatty liver disease among 1350 children aged 6–8 years in Beijing, China.
figure 2

The solid line presents empirical curve of the ROC line, while the dash lines presents the fitting curve; blue line for boy and red line for girl. af Show ROC curve for waist circumference, waist-to-hip ratio, waist-to-height ratio, visceral adiposity index, trunk fat index, visceral fat area, respectively; g and h show comparison of indicators among boys and girls respectively using empirical curves; i shows comparison of Youden index.

The top three fitted AUC were TFI [0.94 (0.92–0.97)], WC [0.93 (0.89–0.96)], and WHtR [0.92 (0.88–0.95)] for boys, and TFI [0.96 (0.92–0.99)], WC [0.95 (0.91–0.99)], and VFA [0.93 (0.86–0.99)] for girls. Among all six indicators, TFI seemed to be the best single indicator with cut-off values of 1.83 kg/m2 for boys and 2.31 kg/m2 for girls. WC is the most widely used indicator of abdominal obesity, but it was slightly less effective to predict NAFLD than TFI, with cut-off values of 63.0 cm for boys and 59.0 cm for girls. The sensitivity and specificity of the cut-off values and other parameters are shown in Table 2.

Logistic regression and risk assessment

After grouping by each cut-off value, boys with higher TFI (aOR = 13.8, 95% CI = 3.8–49.8), higher VFA (aOR = 11.1, 95% CI = 2.9–42.9), higher WHtR (aOR = 3.1, 95% CI = 1.1–9.2), and higher VAI (aOR = 2.8, 95% CI = 1.3–6.1) were more likely to have NAFLD. Girls with higher TFI (aOR = 21.0, 95% CI = 1.9–233.0) and higher VFA (aOR = 17.5, 95% CI = 1.7–181.5) were more likely to have NAFLD. The variables in the multivariable models were adjusted by Z-score of height and Z-score of weight (Table 3).

Table 3 Association of non-alcoholic fatty liver disease and abdominal obesity indicators using multivariable logistic regression among children aged 6–8 from PROC in Beijing, China (N = 1350).

Assessment of abdominal obesity indicators

Among Chinese children aged 6–8 years, (1) TFI has the best ability to identify NAFLD, has a robust and stable risk assessment ability, and is moderately easy to use, though sex-specific cut-off values may be hard to remember; (2) WC has a similar predictive ability to TFI, but risk assessment may be affected by height and weight, and is also moderately easy to use with sex-specific cut-off values; (3) VFA has an excellent ability to identify NAFLD, has a stable and robust risk assessment ability, and is more suitable for girls, but has more complex sex-specific cut-off values; (4) WHtR has a similar predictive capacity to VFA, but is affected by both height and weight; its cut-off values are easy to remember and use; (5) WHR has a moderate ability to identify NAFLD, and risk assessment is affected by height and weight, but cut-off values are easy to remember and use; (6) VAI has a moderate ability to identify NAFLD, risk assessment may be affected by height and weight, and is difficult to use because of complicated calculations (Table 4). All abdominal obesity indicators show significant correlations with each other as well as with other anthropometric indexes (Table 5).

Table 4 The performance ranking of abdominal obesity indicators predicting non-alcoholic fatty liver disease among children aged 6–8 from PROC in Beijing, China.
Table 5 The correlations between abdominal obesity indicators and anthropometric indices among Chinese children aged 6–8 years (N = 1350).


This study used anthropometric and ultrasonographic data from 1350 children aged 6–8 years in Beijing to compare six abdominal obesity indicators for predicting pediatric NAFLD. We found optimal cut-off values of body composition indicators such as TFI and VFA were more robust to identify children with NAFLD than the traditional anthropometric measurement indicators like WC, WHtR and WHR, and an endocrine indicator VAI. The cut-off values for TFI, WC, and VFA are 1.83 kg/m2, 63.0 cm, and 24.6 cm2, and 2.31 kg/m2, 59.0 cm, and 32.7 cm2 for boys and girls, respectively. The reliability of cut-off values of TFI or VFA for predicting NAFLD is not affected by height and weight, while previously widely used indicators such as WC and WHR can be modified by height and weight, as validated by multivariable analysis.

In a 2019 publication, the prevalence of pediatric NAFLD was estimated to range from 2.6 to 17.3%, increasing with age.3 Our results report the prevalence of NAFLD in Beijing was 7.0% (10.4% in boys, 3.7% in girls). This was a bit higher than the prevalence in a Yangtze River delta region study in China that reported 5.0% NAFLD prevalence in 7229 school children aged 7–18 from the southern China had NAFLD (7.5% in boys, 2.5% in girls); rates of NAFLD were 12.9% in those with abdominal obesity and 44.8% in those with mixed obesity.4 The discrepancy is alarmingly high since the Yangtze study was conducted about 10 years ago, covering a wide age range of 7–18. Children approaching puberty are more likely having adipose accumulation. Excessive body fat, mainly abdominal adipose, is associated with NAFLD.8,9,29 We observed significant differences in all indicators, reflecting markedly altered abdominal circumference or fat accumulation in NAFLD children compared to others. The height, weight, and BMI of children with NAFLD were significantly higher than those without NAFLD, which indicates that the faster development may potentially modify later life health trajectories.30,31,32

We observed significant positive correlations between all abdominal obesity indicators and NAFLD; all correlation coefficients were greater for boys than girls. Previous studies reported similar findings for WC,33 WHR,17 WHtR,18 VAI,21 and VFA20 among inconsistently sampled populations that are hard to compare. Studying all six indicators in one fairly representative sample of school children found them all significantly related to NAFLD. We found that body composition indicators were more strongly predictive of NAFLD than anthropometric measurements and endocrinological indicator (VAI) by Spearman correlation coefficient. Though there are concerns that VAI was developed and validated among adults rather than children and adolescents, possibly limiting its application among children. The inconsistencies in the literature may also be due to small samples sizes, both in pre- and post-pubertal children.34 From a physiological perspective, puberty is complex stage of one’s life. A recent Brazilian study echoed our findings that VAI was a good predictor of MS in both adolescents (OR = 12.2), and adults (OR = 9.7).35 A cross-sectional study of 978 Chinese adults reported liver controlled attenuation parameters (CAP) grading of fatty liver changes to have a good association with body fat mass, especially with TFI and VFA,36 further promoting the predictive potential of TFI and VFA. All abdominal obesity indicators correlate well with each other, which indicates a potential implication for a more complex body composition indicator to identify NAFLD in children. Furthermore, most previous studies chose indicators based primarily on a single measurement method; there is a dearth of studies comparing the ability of traditional vs. newer indicators to predict NAFLD in children.

Based on the ROC curves, TFI best discriminated NAFLD. Moreover, the cut-off values of WC in our study are similar with the recommended 90th percentile values of (61.5, 64.6) cm in boys aged 6 and 7, and (60.5, 63.3) cm in girls aged 6 and 7 from the latest 8-country study on WC percentile cutoffs for central obesity.16 VFA was more suitable for girls, and WHtR performed the same for both boys and girls. WHR performed less well, and VAI showed comparatively lower predictive ability. Previous studies show that among children, WHtR is a useful index for NAFLD, with cut-off values around 0.474,18 slightly lower than our results, possibly due to the narrow age range and being comparatively younger age in our population. TFI, VFA, WHtR, and VAI as predictors have better sensitivity than specificity for screening NAFLD both for boys and girls, but WHR has better specificity than sensitivity for NAFLD screening. The sensitivity of WC is similar with specificity in boys, but much greater in girls.

With an adjustment for Z-score of height and weight, the cut-off values of body composition indicators showed a more stable predictive and risk assessment ability than anthropometric measurement indicators, as with TFI (aOR: 13.8 vs. 21.0) and VFA (aOR: 11.1 vs. 17.5) for boys and girls, respectively. Along with the wide use of body composition measurements, TFI or VFA are recommended to NAFLD screening for their convenient use and discriminative potency. Given the age range of our participants (between infancy and adulthood), we seek to contribute to the limited literature on the association of VAI and pediatric NAFLD. According to the objective analysis of the six indicators in our study, VAI was not the best indicator for predicting NAFLD.

The major strengths of our study are the robust and complete data from a population-based cohort with a representative sample of school children aged 6–8 in China, using a strong ultrasonic imaging method with reliable quality control assurance. We used one single observer for imaging for a more robust and reliable standard, avoiding interobserver variation. Consistency is vital for a comparative prospective study. We measured abdominal obesity by anthropometric, body composition, and endocrine methods, providing an opportunity for intensive comparative analysis with validation and cross-assessment. Study limitations include the fact that diagnosis of NAFLD was based on ultrasonography rather than liver biopsy, which may lead to misclassification and underestimate the prevalence. Nonetheless, ultrasound screening for NAFLD is more acceptable and is recommended by the European Association for the Study of the Liver (EASL). Another limitation is our narrow age range of 6–8 years old, which is well in advance of puberty; however, this actually helps in data interpretation. While our cut-off values are quite robust, we have less generalizability to other age groups. In future, we will follow-up the participants to validate cut-off values for wider age range of children and adolescents, as well as verify their longer-term associations with NAFLD.


In conclusion, we present a comparison with cut-off values for six abdominal obesity indicators measured by different methods to predict NAFLD among children aged 6–8 years in China. Body composition indicators such as TFI and VFA show a greater application potential in helping physicians early screen for pediatric NAFLD and perform easy, reliable, and interpretable weight management to prevent the progress of liver damage.