Diagnostic performance of anthropometric measurements in detecting elevated serum alanine aminotransferase: a school-based screening of 7271 Shenzhen adolescents

Background Screening for elevated serum alanine aminotransferase (ALT) can help identifying individuals at the risks of chronic and metabolic diseases, but blood collection is invasive and cannot be widely used for investigations. Considered as simple and inexpensive screening indices, anthropometric measurements can be measured in a large crowd and may be important surrogate markers for ALT levels. Among adolescents, few studies focused on the diagnostic performance of anthropometric parameters such as body mass index (BMI) and BMI z-score as predictive factors for discerning an elevated ALT activity. We sought to fill this knowledge gap among Shenzhen adolescents. Methods A school-based study was performed from 9 high schools in Shenzhen during February 2017 and June 2018. Receiver operating characteristic curve was used to examine the diagnostic performance of each measurement. Results Altogether 7271 adolescents aged 9−17 years were involved. The proportion of elevated ALT greatly increased with increasing classification of BMI-z. By the sex-specific cut-offs for elevated ALT (30 U/L boys; 19 U/L girls), BMI showed the highest area under the curve of 0.789 (95% CI 0.765-0.812) and followed by weight (0.779 [0.755-0.802]), BMI-z (0.747 [0.722-0.772]), height (0.622 [0.597-0.647]), and age (0.608 [0.584-0.632]), while height-z was not capable. With the cut-off of 67.8 kg for weight and 22.6 kg/m2 for BMI, the accuracy was 87.1% for weight and 82.9% for BMI. Conclusions ALT in overweight or obese adolescents. BMI weight detecting ALT, by BMI-z,


Background
The liver enzyme of serum alanine aminotransferase (ALT) is commonly used to assess potential liver dysfunction and clinically as a screening tool for the detection of probable nonalcoholic fatty liver disease (NAFLD) 1− 4 . An elevation of serum ALT concentration is a key feature of NAFLD, and liver biopsy is the gold standard to confirm the diagnosis of NAFLD, characterized by pathological changes to liver structure and function such as fat accumulation, hepatocyte dysfunction, and fibrosis 4 . The elevation of ALT has also been recorded by published evidence to be strongly linked to, or to predict, the developing of type 2 diabetes mellitus, insulin resistance, and cardiovascular diseases (e.g. atherogenesis and coronary heart disease) 2, 3, 5− 11 .
Screening for the concentration of serum ALT and verifying an elevated ALT activity could assist in determining individuals at risk for the above chronic conditions and reducing the possibility of future liver diseases 12,13 . However, the measurement of ALT levels requires blood collection, which is invasive and could be impractical in large-scale investigations.
In contrast, measurements of anthropometric indices such as height, weight, and BMI are relatively simple, cheap, quick, and non-invasive, and therefore can be applied to a great number of people and easily conducted during common health examinations, especially for non-adults 12,14,15 . If these anthropometric indices showed a close correlation with serum ALT levels or had sufficient ability to detect an ALT elevation, they could be a useful surrogate for ALT levels.
Two multi-center, population-based epidemiological researches conducted in the mainland of China have consistently demonstrated that the higher values of body mass index (BMI), hip circumference, waist circumference, waist-to-hip ratio (WHpR), and waist-to-height ratio (WHtR) were able to predict an elevated ALT activity among adolescents and adults 14,16 . Recently, the nontraditional index of BMI z-score has also been put forward as another important anthropometric indicator of ALT elevation 4,17 . However, whether the potential of the z-score of height, a novel body index, for predicting elevated ALT was comparable or superior to the abovementioned parameters was still unclear up to now. In addition, although most previous studies on the relations of anthropometric measurements with ALT elevation were conducted among adults 14, 18− 20 , few investigations focused on the performance predication of those predictors for diagnosing elevated ALT among adolescents 12,17,21 .
Considering such a negotiable knowledge gap, we aimed to examine the diagnostic performance of individual characteristics (i.e. age and gender) and several anthropometric parameters (i.e. height, weight, BMI, BMI-z, and height-z) as predictive factors of an elevated ALT activity in Shenzhen adolescents, and to determine the optimal cut-off points for these parameters that would identify an adolescent with an elevation of serum ALT level. Assessing anthropometric measurements is crucial for early detection and prevention of an activity of elevated ALT or the subsequent development of NAFLD among children and adolescents.

Study sitting and population
Situated approximately 1˚ south of the Tropic of Cancer, within the Pearl River Delta region of Guangdong province, and bordered Hong Kong, Shenzhen is one of the 4 firsttier cities in mainland China, the first Special Economic Zone of China (established in 1980), and a major financial centre and the worldly largest manufacturing base 22,23 . The population and economy have grown rapidly, with a total gross domestic product (GDP) of 255 billion USD and approximate 25,000 USD per capita in 2015 22,24 . The municipality covers an area of 1997.3 km 2 and the population is around 11.37 million; Bao'an District is the first largest area among 10 district-level jurisdictions of Shenzhen, with a total population of 2.86 million accounting for over a quarter of Shenzhen's total population at the end of 2016 22 .

Data collections and measures
Involved freshmen from 9 junior and senior high schools of Bao'an District, a crosssectional study organized with the assistance of the Bao'an Government of Shenzhen was officially carried out 25 . Data collection activities were completed during February 2017 and June 2018. This school-based study was approved by the Institutional Review Board of Baoan Central Hospital of Shenzhen, and written informed consents for each student and their parents were obtained.
Individual characteristic data such as gender and date of birth were taken from each participant, and anthropometric parameters of weight and standing-height were respectively measured to the nearest 0.1 kg and 0.5 cm by trained physicians using a standard weighing machine. At each high school, blood samples were drawn in the morning after participants fasting for at least 10 hours by trained nurses. ALT was measured with the fully automatic biochemical analyzer (Model AU5821, Tokyo, Japan). We restricted our analysis to adolescents but not adults, and a student aged between 9.0 and 17.9 years was regarded as an adolescent. Finally, a total of 7271 adolescents who had complete anthropometric and clinical data participated in the current investigation.
For each participant, age with 1 decimal was transformed based on the date of birth and BMI (kg/m 2 ) was calculated in weight divided by the square of height to set up a database.
Based on an international norm from the World Health Organization (WHO) growth reference (i.e. the 2007 WHO reference) for school-aged children and adolescents aged 5 − 19 years, anthropometric indicators of z-score values of height-for-age (height-z) and BMI-for-age (BMI-z) for both boys and girls were calculated by the R macro, provided on the WHO website (https://www.who.int/growthref/en/) 26,27 . Moreover, according to the zscore of BMI-for-age, we defined BMI-z > 2SD as obesity, 1SD < BMI-z ≤ 2SD as overweight, -2SD ≤ BMI-z ≤ 1SD as normal-weight, and BMI-z < -2SD as underweight; where SD was standard deviation of the BMI z-scores.

Definition of elevated serum ALT levels
The thresholds of elevated ALT needed to be determined consistent with previously reported levels, as there was no consensus on what level constituted an elevated serum ALT concentration for adolescents. For the most commonly use of definitions among adolescents, an activity of elevated serum ALT was defined as > 30 U/L for boys and > 19 U/L for girls (diagnostic criterion I) 2,3,8,28 . Main analysis was undertaken using these sex-specific cut-offs, and a secondary analysis for the sake of comparison was also reported, defining with higher thresholds (> 40 U/L for both boys and girls; diagnostic criterion II) that have been proposed in previous observational literatures 7, 13, 29 to explore whether higher thresholds led to the same or different results.

Statistical analyses
The normal distribution of each continuous anthropometric measurement was determined by the one-sample Kolmogorov-Smirnov test. The non-normally distributed measurements were described as median with inter-quartile range (25th percentile − 75th percentile; IQR); significance for differences between elevated and normal serum ALT was evaluated by the Mann-Whitney U test. Stratified by the classification of BMI z-score, the crude prevalence along with 95% confidence interval (CI) of elevated ALT was quantitatively estimated, and the trend of the proportion of elevated ALT based on BMI-z was carried out with a chi-square test. The Spearman's rank correlation analyses among the skewed parameters were also performed.
Extensively used in clinical epidemiology for evaluation of diagnostic ability of biomarkers (e.g. serum markers) or a diagnostic test with dichotomous outcome (i.e. positive or negative result) in classification of the diseased from healthy population, the receiver operating characteristic (ROC) curve, defined as a plot of the sensitivity of a test as y-axis versus 1-specificity as x-axis, is an effective method and graphical technique to describe the accuracy of a prediction model or diagnostic test 30, 31 . Sensitivity and specificity, the basic measures of accuracy of a diagnostic test, vary with different cut-off thresholds, and sensitivity is inversely related to specificity 15,31 . Positive predictive value (PPV) is defined as the probability of disease for positive test results, and negative predictive value (NPV) is defined as the probability of being healthy for negative test results.
However, influenced by the prior prevalence of disease in a population, PPV rises while NPV decreases with an increased prevalence of disease 30 .
In the current study, we utilized the ROC curve analysis to find out the optimal cut-off points and to examine the diagnostic performance of each anthropometric measurement as indicators of ALT elevation. An anthropometric parameter value with the highest Youden index was chosen as the optimal cut-off point 15 . The observed agreement and the area under the curve (AUC) were also determined. Statistical analyses were processed in R 3.5.1 (http://www.R-project.org) and SPSS for Windows 16.0, with two-tailed P-value < 0.05 considered statistically significant.

Characteristics of participants
Altogether 7,271 adolescents aged 9−17 years were recruited in the present study for the operative statistical analysis, involving 4,014 (55.2%) boys and 3,257 (44.8%) girls.
Characteristics of study subjects with and without elevated serum ALT activity are listed in for BMI-z. Each anthropometric index but not height-z was significantly greater among adolescents with an elevated ALT activity than those with a normal ALT level, regardless of the use of definitions (most P<0.001).

Crude prevalence of elevated ALT by the BMI-z classification
On the basis of ALT activity thresholds of diagnostic criterion I, an elevated ALT activity was present in 2.48% (6/242) of underweight, 3.79% (208/5493) of normal-weight, 12.58%  (Table S1). Regardless of the use of diagnostic criteria, the abnormality prevalence rate of ALT greatly increased with increasing classification of BMI z-score and peaked in the obesity group for all participants, boys, and girls (all P trend <0.001). Overweight or obese adolescents were much more likely to obtain increased ALT concentrations than normal-weight or underweight adolescents. Figure S1 illustrates the diagnostic performance of anthropometric indices for identifying elevated serum ALT by ROC curves, and Table 2  . As compared to the above 3 obesity indices, age and height had a poorer diagnostic efficiency in detecting abnormal ALT. The discriminatory power of height z-score in the prediction of elevated ALT was null (both P > 0.1), and gender also had no ability to evaluate elevated ALT by the same upper limit of normal (ULN) between boys and girls for ALT of the diagnostic criterion I (P = 0.435). In the stratification analyses by gender, similar outcomes were also provided in Table S2. Table 3 elucidates the optimal cut-off points and the probability of meaningful and significant parameters in diagnosing abnormal ALT, based on the ROC analyses. The  Table S3 displays the Spearman's correlation coefficients between anthropometric indices and ALT levels. BMI had a very high positive correlation with BMI z-score (r = 0.891) and weight (r = 0.829), suggesting a multicollinearity among each obesity parameter. Weight was also quite strongly and positively correlated with height (r = 0.673) and BMI-z (r = 0.657). Table 4

Discussion
Screening for increased ALT levels is conducive to identifying individuals at the risks of some chronic and metabolic diseases, but blood collection of serum ALT is invasive and cannot be widely used for a massive epidemiological survey among children and adolescents because of subject burden and cost. Considered as a simple, easily obtained, inexpensive, and convenient screening measure, anthropometry can be monitored in a large crowd and might be an important surrogate marker for ALT concentrations, if the discriminative accuracy of the anthropometric parameters is very high for diagnosing abnormal ALT. Primary focused on the diagnostic performance, anthropometric data for Shenzhen adolescents aged 9-17 years were used to detect elevated serum ALT in our present survey, and we confirmed the belief that anthropometric measurements were important predictors of elevated ALT among Shenzhen adolescents. An elevated ALT activity was more common in overweight or obese adolescents than those with a normal or relatively low BMI z-score. BMI and weight showed an approximate diagnostic performance for predicting the presence of elevated ALT, followed by the z-score of BMI, height, and age.
Consistent findings that the proportion of elevated ALT greatly increased with increasing degree of obesity (e.g. the percentile or z-score of BMI), and higher proportion among overweight or obese adolescents than normal-weight ones were shown in previous studies and our research, regardless of the use of diagnostic criteria for an elevated ALT activity 7,13,18,32 . Based on the thresholds of ALT activity (40 U/L for boys and girls), nearly 6.6% of the Mexican youths (9.8% of boys and 3.8% of girls) had elevated ALT levels by using the baseline data from 1262 participants of 8-19 years in the Mexican Health Worker Cohort Study (MHWCS); elevated ALT concentration was also observed in 2.7% of normal-BMI, 14.2% of overweight, and 28.9% of the obese children and adolescents 13  Screening for school-adolescents, it is clearly more convenient to collect anthropometric measures than biochemical measures, but the abilities of several individual characteristics and anthropometric indices to correctly predict elevated serum ALT in Shenzhen children are questionable and need to be assessed. Determined by the univariate ROC curves, our results showed that the anthropometric parameter of BMI had the best superiority of discerning the presence of elevated ALT (AUC = 0.789 for criterion I and 0.850 for II), weight and BMI-z displayed the second and third highest detection accuracy (AUC range from 0.747 to 0.850), and followed by height, age, and gender (from 0.490 to 0.695). As such, we believed that the single variable of height, age, or gender was not reliable surrogate measure of an activity of elevated ALT among Shenzhen adolescents of 9 − 17 years. Further, age, gender, and height were considered as covariates together with each predictor of obesity indices (i.e. weight, BMI, or BMI-z) to predict elevated ALT in the subsequent multivariate model analyses, although the diagnostic accuracy of the combined models did not markedly improve. On the other hand, the current study firstly, as far as we known, investigated the association of height z-score with serum ALT in adolescents and estimated the usefulness of height-z as a predictive index of elevated ALT. Nevertheless, the present result showed a null discriminative power of height z-score for predicting an elevated ALT serum activity.
For BMI, several studies had pointed out that BMI was able to determine the presence of elevated ALT among adults, although the accuracy was not very high in absolute term -  17 . In addition, based on the data from the National Health and Nutrition Examination Survey (NHANES) during 1999 to 2014, an United States study consisted of 5019 adolescents of 12-19 years suggested a significant correlation between BMI z-score and serum ALT levels (r = 0.29, P < 0.0001) 4 .
Significant positive associations of ALT with obesity indices (e.g. BMI, body fat percentage, truncal fat mass, total fat mass, waist circumference, and WHtR) were further confirmed in Korean male adolescents 21 .
Several potential limitations of our research should be recognized. First, participants were only from a single city of China, and it might be therefore difficult to generalize these findings to other adolescent populations. Second, our cross-sectional study omitted to measure some important individual anthropometric parameters -hip circumference, waist circumference, neck circumference, and the subsequent indicators of WHpR, WHtR, and A Body Shape Index (ABSI), which might achieve a better diagnostic performance and to be more sensitive predictive indicates for diagnosing an elevated ALT activity among the indigenous adolescents. For predicting the ALT levels, WHtR and to some extent BMI were congruously shown to be the best body indices in some Asians, as compares to waist circumference, hip circumference, WHpR, and the useless index of ABSI 12,14,16 . Another limitation was our inability to consider the potential confounders of ethyl alcohol intake and hepatitis B virus (HBV) and C virus (HCV) infections, which were well-known risk factors for increases in ALT 4,12,17,19 . However, alcohol consumption and HBV and HCV infections were less prevalent in school-adolescents of China 33, 34 , and we hypothesized that the obesity indices could be even more important predictors of elevated ALT among the adolescents of Shenzhen, and the above potential confounders were not likely to have a substantial impact on current results.

Conclusions
An elevated ALT activity was more frequent in overweight or obese adolescents than normal-weight subjects. Anthropometric obesity indices of BMI, weight, and BMI-z achieved a very high diagnostic performance for predicting elevated ALT among Shenzhen adolescents and followed by height and age, while height-z was not capable.

Declarations
Ethics approval and consent to participate: This school-based study was approved by the Institutional Review Board of Baoan Central Hospital of Shenzhen, and written informed consents for each student and their parents were obtained.

Consent for publication:
All of the authors have read and approved the paper.   The optimal cut-off points were based on the maximum Youden index for each parameter; the probability of parameters for identifying elevated ALT is expressed as %.
# The best performance for each conventional diagnostic test index.