Nonalcoholic fatty liver disease (NAFLD) is commonly linked to childhood obesity, with a prevalence of approximately 38% in obese children (1). The progression of liver damage is possible and may lead to the necessity of a liver transplant (1–3). Docosahexaenoic acid supplementation has been shown to be effective in reducing NAFLD in children (4). Therefore, early diagnosis of fatty liver in children may be useful for potential specific intervention.
Two methods may be used in clinical settings to measure fat liver content: magnetic resonance imaging (MRI) and histology; however, large-scale use of liver biopsy and MRI is not possible in general practice because of its invasiveness and/or lack of availability of these techniques. Other techniques have a lower sensitivity and only offer a qualitative estimate of fat accumulation (5).
To narrow the number of patients to refer to instrumental investigation, it would be helpful to have an accurate screening tool to identify the obese children who are most likely to be affected by NAFLD. Several single biochemical and anthropometrical parameters as well as the presence of metabolic syndrome are associated with NAFLD, either biopsy proven or assessed by MRI, and may be used as predictors (6–8). The aim of the present study was to build an accurate and handy equation able to estimate the risk of NAFLD in prepubertal obese children from the combination of anthropometric and biochemical parameters.
PATIENTS AND METHODS
A cohort of 56 white obese children was recruited from the Obesity and Nutrition Outpatient Clinic of the Pediatric Section of the University Hospital of Verona. Puberty development was clinically assessed and none of the patients had advanced puberty.
Inclusion criteria were ethnicity (white) and obesity according to the International Obesity Task Force body mass index (BMI) cutoffs (9). Exclusion criteria were any overt disease other than obesity, acute inflammation, blood or blood by-product transfusion, history of hepatic infectious disease, autoimmunity, impaired glucose tolerance, or diabetes. The protocol was in accordance with the 1975 Declaration of Helsinki, as revised in 1983, and was approved by the ethical committee of the University Hospital of Verona.
Patients arrived at the outpatient clinic at 8.00 AM after 12-hour fast. They were given a general checkup, and blood pressure and anthropometrical measures were taken according to standard methods as previously reported (10). BMI z scores (zBMI) were calculated using the least mean squares method and national reference values of BMI and least mean squares coefficients (11).
Baseline fasting blood samples were taken to measure plasma glucose, serum insulin, lipid profile, total adiponectin, and alanine aminotransferase (ALT). Plasma glucose concentration, lipid, and ALT were measured by standard in-house methods. Plasma insulin levels were measured by chemiluminescent immunometric assays (Euro/DPC Ltd, Llanberis, UK). Plasma adiponectin was measured by enzyme-linked immunosorbent assay (B-Bridge International, Mountain View, CA) according to the manufacturer's instructions.
After baseline blood drawing, a standard 3-hour oral glucose tolerance test was performed. Impaired glucose tolerance was defined by a 2-hour blood glucose of 140 to 199 mg/dL. Homeostasis model assessment of insulin resistance (HOMA-IR) was calculated as (fasting serum insulin [μU/mL] × fasting plasma glucose [mmol/dL]) × 0.00244 (12).
Following the oral glucose tolerance test, the children underwent body composition measurements by MRI. Hepatic fat accumulation was measured using MRI along with the Dixon method (13). A hepatic fat fraction cutoff of 5.5% was chosen as the threshold to define steatosis (8).
Data are presented as mean ± SD. Univariate binary logistic regression was used to identify variables associated with NAFLD. Pearson bivariate correlations between all of the NAFLD-associated variables were used to identify pairs of variables, with >0.5 correlation to keep only the variable with the strongest association with NAFLD in each of these “collinear” pairs for the subsequent multivariate analysis. This was meant to reduce possible redundant variables.
Selected variables were then used to build an NAFLD-predicting model by conditional backward binary logistic regression. Calibration of the model was evaluated by the Hosmer-Lemeshow goodness-of-fit test on 10 progressive predicted risk classes. Discrimination effectiveness of the model was evaluated by receiver operating characteristic (ROC) curve analysis of the model-issued NAFLD probability score. All of the statistics were performed with SPSS version 17 (SPSS Inc, Chicago, IL) for Windows.
The following variables were associated with NAFLD in univariate analysis: age, zBMI, waist-to-height ratio, ALT, HOMA-IR, and serum adiponectin (Table 1). zBMI and waist-to-height ratio showed a high correlation (R = 0.562, P < 0.0001), and only the waist-to-height ratio was kept for further multivariate analysis, given its stronger association with NAFLD in univariate analysis. Waist-to-height ratio, HOMA-IR, and adiponectin were independent predictors of NAFLD, with an overall Nagelkerke R 2 of 0.734 (Table 2).
The model-issued equation to predict NAFLD probability was
The model was well calibrated (Hosmer-Lemeshow χ 2 = 4.15, P = 0.76).
Discrimination between individuals with and without NAFLD was good, with an area under ROC of 0.94 (95% confidence interval [CI] 0.89–0.99, P < 10−5) (Fig. 1). The optimal cutoff point of 0.59 according to the Youden index showed 85% (71%–99%) sensitivity, 89.5% (78.5%–100%) specificity, a positive predictive value of 88.5% (72.2%–94.7%), and a negative predictive value of 86.5% (74.5%–99%).
Because adiponectin cannot be easily measured or may be considered cost-ineffective in the screening of obesity-associated NAFLD in some clinical settings, we also built a predictive model without adiponectin. For this purpose, we generated the model a second time after removing adiponectin from the variables. The resulting model had an overall Nagelkerke R 2 of 0.54 (Table 2). The model-issued equation to predict NAFLD probability was
This model was also well calibrated (Hosmer-Lemeshow χ 2 = 2.83, P = 0.90).
The discrimination accuracy of this model was worse than that shown by the model including adiponectin. In fact, by removing adiponectin, discrimination accuracy was associated with negative integrated discrimination improvement of −19% (P < 0.01), that is, a 19% decrease in the average sensitivity of all of the possible cutoffs adjusted for variation in specificity (14). Discrimination between individuals with and without NAFLD was still good, however, with an AUROC of 0.88 (95% CI 0.79–0.97, P < 10−5) (Fig. 1). The optimal cutoff point of 0.39 according to the Youden index showed 89% (77%–100%) sensitivity, 76% (60%–82%) specificity, a positive predictive value of 77.5% (62.5%–92.5%), and a negative predictive value of 88% (75%–100%).
The results of the present study show that it is possible to predict NAFLD in obese children by using a calculated score based on anthropometric and biochemical variables. Previous studies have been conducted on adults with the purpose of predicting fat accumulation in the liver by routinely available clinical and laboratory data (15–17). In particular, a recent study proposed a NAFLD liver score based on the following variables: metabolic syndrome, type 2 diabetes mellitus, fasting insulin, aspartate aminotransferase, and aspartate aminotransferase/ALT. This index showed good sensitivity (86%) and fair specificity (71%) in predicting increased liver fat content (17). To the best of our knowledge, this is the first study done on children, particularly obese prepubertal children, using MRI to measure hepatic fat fraction to develop a prediction equation for NAFLD. The clinical effect of early diagnosis of fatty liver is still unclear and little data are available on long-term prognosis of fatty liver in obese children (2). Recent evidence that fat accumulation in the liver declines after treatment with docosahexaenoic acid suggests that early diagnosis of NAFLD may be beneficial to obese children (4).
The equations proposed can be translated into smart automatic risk calculators for everyday clinical practice (online-only supplementary electronic appendix, http://links.lww.com/MPG/A55). Moreover, the model without adiponectin only requires routinely available measures and performed well despite a little loss of accuracy because adiponectin was not included. This suggests that the screening for fatty liver in obese children could be simple, safe, and cost-effective.
The results of our study must be considered preliminary. In particular, the prediction of accuracy was evaluated only in the sample in which the models were built because the limited size of the cohort did not allow for internal validation in a subsample that was different from that used for model building. Consequently, accuracy estimates could be slightly overoptimistic; however, the predictive models were built from a parsimonious number of variables to avoid model overfitting. Moreover, abdominal fat, ALT, insulin resistance, and adiponectin have been consistently and strongly associated with obesity-related NAFLD, making it likely that the combination of 3 or all of these risk factors actually are dependable and rather accurate in predicting NAFLD in obese children (6–8).
The models proposed need to be validated in other cohorts of obese children before being proposed for clinical use. Of course, model adaptation or redevelopment in local clinical settings would be possible, beyond validation, to achieve the best accuracy. This would be necessary, for example, in case of differences in dosage methods and/or normal ranges of serum insulin, which would require the HOMA-IR coefficient in the predictive equation to be reassessed.
In conclusion, our pilot study suggests that a NAFLD predictive equation that includes a few routinely available parameters could be a helpful screening tool for selecting obese children likely to be affected by NAFLD.
1. Schwimmer JB, Deutsch R, Kahen T, et al. Prevalence of fatty liver in children and adolescents. Pediatrics
2. Feldstein AE, Charatcharoenwitthaya P, Treeprasertsuk S, et al. The natural history of non-alcoholic fatty liver disease in children: a follow-up study for up to 20 years. Gut
3. Schreuder TC, Verwer BJ, van Nieuwkerk CM, et al. Nonalcoholic fatty liver disease: an overview of current insights in pathogenesis, diagnosis and treatment. World J Gastroenterol
4. Nobili V, Bedogni G, Alisi A, et al. Docosahexaenoic acid supplementation decreases liver fat content in children with non-alcoholic fatty liver disease: double-blind randomised controlled clinical trial. Arch Dis Child
5. Graif M, Yanuka M, Baraz M, et al. Quantitative estimation of attenuation in ultrasound video images: correlation with histology in diffuse liver disease. Invest Radiol
6. Manco M, Marcellini M, Devito R, et al. Metabolic syndrome and liver histology in paediatric non-alcoholic steatohepatitis. Int J Obes
7. Schwimmer JB, Pardee PE, Lavine JE, et al. Cardiovascular risk factors and metabolic syndrome in pediatric nonalcoholic fatty liver disease. Circulation
8. Burgert TS, Taksali SE, Dziura J, et al. Alanine aminotransferase levels and fatty liver in childhood obesity: associations with insulin resistance, adiponectin, and visceral fat. J Clin Endocrinol Metab
9. Cole TJ, Bellizzi MC, Flegal KM, et al. Establishing a standard definition for child overweight and obesity worldwide: International survey. BMJ
10. Maffeis C, Pietrobelli A, Grezzani A, et al. Waist circumference and cardiovascular risk factors in prepubertal children. Obes Res
11. Luciano A, Bressan F, Zoppi G. Body mass index reference curves for children aged 3-19 years from Verona, Italy. Eur J Clin Nutr
12. Matthews DR, Hosker JP, Rudenski AS, et al. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia
13. Fishbein MH, Gardner KG, Potter CJ, et al. Introduction of fast MR imaging in the assessment of hepatic steatosis. Magn Reson Imaging
14. Pencina MJ, D’Agostino RB, D’Agostino RB, et al. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med
15. Poynard T, Ratziu V, Naveau S, et al. The diagnostic value of biomarkers (SteatoTest) for the prediction of liver steatosis. Comp Hepatol
16. Bedogni G, Bellentani S, Miglioli L, et al. The fatty liver Index: a simple and accurate predictor of hepatic steatosis in the general population. BMC Gastroenterol
17. Kotronen A, Peltonen M, Hakkarainen A, et al. Prediction of non-alcoholic fatty liver disease and liver fat using metabolic and genetic factors. Gastroenterology
Keywords:Copyright 2011 by ESPGHAN and NASPGHAN
accuracy of biochemical parameters and anthropometry in predicting NAFLD; NAFLD; obese children