G Model

YDLD-2688; No. of Pages 6 ARTICLE IN PRESS

Digestive and Liver Disease xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Digestive and Liver Disease

journal homepage: www.elsevier.com/locate/dld

Liver, Pancreas and Biliary Tract

A non-invasive prediction model for non-alcoholic steatohepatitis in

paediatric patients with non-alcoholic fatty liver disease
Katharine Eng a , Rocio Lopez c , Daniela Liccardo d , Valerio Nobili d , Naim Alkhouri a,b,∗
a
Department of Pediatric Gastroenterology, Cleveland Clinic, Cleveland, OH, United States
b
Digestive Disease Institute, Cleveland Clinic, Cleveland, OH, United States
c
Quantitative Health Sciences at the Cleveland Clinic Foundation, Cleveland, OH, United States
d
Liver Unit, Bambino Gesù Children’s Hospital and Research Institute, Rome, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Background: Non-alcoholic fatty liver disease encompasses a spectrum of diseases that range from sim-
Received 3 April 2014 ple steatosis to the aggressive form of non-alcoholic steatohepatitis. Non-alcoholic steatohepatitis is
Accepted 8 July 2014 currently diagnosed through liver biopsy.
Available online xxx
Aim: To develop a non-invasive predictive model of non-alcoholic steatohepatitis in children with non-
alcoholic fatty liver disease.
Keywords:
Methods: Anthropometric, laboratory, and histologic data were obtained in a cohort of children with
Children
biopsy-proven non-alcoholic fatty liver disease. Multivariable logistic regression analysis was employed
Nomogram
Non-alcoholic fatty liver disease
to create a nomogram predicting the risk of non-alcoholic steatohepatitis. Internal validation was per-
Non-alcoholic steatohepatitis formed by bootstrapping.
Results: Three hundred and two children were included in this analysis with a mean age of 12.3 ± 3.1
years, a mean body mass index percentile of 94.3 ± 6.9, and non-alcoholic steatohepatitis was present
in 67%. Following stepwise variable selection, total cholesterol, waist circumference percentile, and total
bilirubin were included as variables in the model, with good discrimination with an area under the
receiver operating characteristic curve of 0.737.
Conclusions: A nomogram was constructed with reasonable accuracy that can predict the risk of non-
alcoholic steatohepatitis in children with non-alcoholic fatty liver disease. If validated externally, this
tool could be utilized as a non-invasive method to diagnose non-alcoholic steatohepatitis in children
with non-alcoholic fatty liver disease.
© 2014 Editrice Gastroenterologica Italiana S.r.l. Published by Elsevier Ltd. All rights reserved.

1. Introduction
The incidence of non-alcoholic fatty liver disease (NAFLD) has
been increasing, and it has now become the most common cause
of chronic liver disease in the paediatric population [1,2]. NAFLD
encompasses a spectrum of diseases that can range from sim-
ple steatosis to non-alcoholic steatohepatitis (NASH), of which
NASH can ultimately progress to cirrhosis and even end-stage liver
disease [3]. Furthermore, children with NASH may benefit from
NASH-specific therapy that may include vitamin E, metformin, and
∗ Corresponding author at: Department of Pediatric Gastroenterology, Cleveland
Clinic Foundation, 9500 Euclid Avenue-A111, Cleveland, Ohio 44195, United States.
Tel.: +1 216 445 7126; fax: +1 216 444 2974.
E-mail address: alkhoun@ccf.org (N. Alkhouri).
newer agents that are being tested [4,5]. It is therefore important
to be able to identify those patients with NASH.
At the present time, there are no clinically available biomarkers
to reliably differentiate between steatosis and NASH, and instead
this differentiation is made through histologic evaluation by liver
biopsy. There are drawbacks to liver biopsy, however, which
include sampling error, variability in histopathology interpretation
by a pathologist, inadequate biopsy size, cost, and associated mor-
bidity. Liver biopsy is an invasive diagnostic procedure with its own
inherent risks that include pain, bleeding, and even other organ
perforation. Serious complications with percutaneous liver biopsy
have been found in 0.3% of cases, with a reported mortality rate
of 0.01% [6]. Given these inherent drawbacks to liver biopsies, an
ideal test would be a non-invasive test that is reproducible, sim-
ple, readily available, and less costly. In our current study, we have
developed a nomogram that utilizes clinical variables and routine

http://dx.doi.org/10.1016/j.dld.2014.07.016
1590-8658/© 2014 Editrice Gastroenterologica Italiana S.r.l. Published by Elsevier Ltd. All rights reserved.

Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016
G Model
YDLD-2688; No. of Pages 6 ARTICLE IN PRESS
2 K. Eng et al. / Digestive and Liver Disease xxx (2014) xxx–xxx
laboratory tests to predict the risk of NASH in paediatric patients
with NAFLD.
2. Materials and methods
2.1. Patient population
Between January 2003 and December 2009 at the Bambino Gesù
Children’s Hospital (Rome, Italy), paediatric patients with biopsy
proven NAFLD were seen and enrolled in this study. The study was
carefully explained to the patients’ guardians and written consent
was obtained for all patients. This study was approved by the ethics
committee at the Bambino Gesù Children’s Hospital and Research
Institute.
To be included in the study, subjects had characteristics
suspicious for NAFLD based on persistently elevated serum amino-
transferase levels and imaging studies that were suspicious for
fatty liver with a diffusely hyperechogenic liver, with a final
diagnosis made on liver biopsy of NAFLD. For patients with
a diagnosis of NAFLD, they were excluded if they had any
of the following: (1) hepatic viral infections (such as hepati-
tis A, B, C, D, and E; cytomegalovirus; and Epstein–Barr virus);
(2) alcohol consumption; (3) use of drugs that are known to
induce steatosis (e.g.: valproate, prednisone, or amiodarone),
affect carbohydrate metabolism or body weight; (4) history
of parental nutrition; and (5) known liver disease, such as
autoimmune hepatitis, metabolic liver disease, Wilson’s disease,
and alpha-1-antitrypsin associated liver disease. These were
ruled out using standard laboratory, clinical, and/or histologic
criteria.
2.2. Patient characteristics
Patient clinical variables were recorded, which included
standard height and weight, as well as the body mass index
(BMI) and Z-score, which were calculated [7]. Patients with a
BMI greater than or equal to 95th percentile adjusted for age
and sex were defined as obese. The waist circumference (WC)
of patients was measured at the highest point of the iliac crest
with a standing subject [8]. A WC to height ratio was also calcu-
lated.
Data collection in patients included laboratory evaluation
of aspartate aminotransferase (AST), alanine aminotransferase
(ALT), serum gamma-glutamyltransferase (GGT), total bilirubin,
albumin, INR, total cholesterol, HDL cholesterol, and triglyc-
erides. This was obtained utilizing standard laboratory meth-
ods.
Metabolic syndrome in this patient population was present
if there were three of more of the following: (1) impaired fas-
ting glucose (≥110 mg/dL), impaired glucose tolerance, or known
type 2 diabetes mellitus [9]; (2) hypertriglyceridemia, defined as
triglyceride level >95th percentile for age and sex [10]; (3) low
HDL-cholesterol, defined as concentrations <5th percentile for age
and sex [10]; (4) hypertension, defined as systolic or diastolic blood
pressures >95th percentile for age and sex [11]; and (5) abdominal
obesity, defined as WC ≥90th percentile for age [12].
2.3. Liver histology
Liver biopsy was performed in all patients who met criteria
because of an indication to assess for the presence of NASH, degree
of fibrosis, and/or to identify if other liver diseases were present.
After fasting overnight and under general anaesthesia, a liver biopsy
was performed utilizing ultrasound guidance and the automatic
core biopsy 18 gauge needle (Biopince, Amedic, Sweden). The Sono-
line Omnia ultrasound machine (Siemens, Munich, Germany) was
Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016
used, which is equipped with a 5-MHz probe (5.0C 50, Siemens)
and biopsy adaptor. For each patient, two biopsy passes in different
liver segments were made, and the length of the biopsy specimen
was recorded in millimetres. Only samples that included at least
5–6 complete portal tracts and had a minimal length of ≥15 mm
met the requirements for this study [13]. Biopsies were processed
and routine staining of liver tissue included haematoxylin-eosin,
Periodic acid-Schiff diastase, Van Gieson, and Prussian blue stain. A
single hepatopathologist was blinded to the clinical and laboratory
data and reviewed the biopsies.
A single expert paediatric hepatopathologist reviewed the biop-
sies and established the histopathological diagnosis of NASH. Based
on this diagnosis, patients were divided into two groups: (1)
“NASH” or (2) diagnosis not compatible with NASH or “not NASH”.
Liver histology was scored using the NAFLD activity scoring (NAS)
system developed by the NASH Clinical Research Network [14].
The grade of steatosis (0–3), hepatocyte ballooning (0–2), and lob-
ular inflammation (0–3) were added together to determine the
NAS score (0–8). Portal inflammation (PI) was graded from 0 to
2 (0 = no PI, 1 = mild PI, and 2 = more than mild PI). Fibrosis was
staged as the following: 0 = no fibrosis, 1 = periportal or perisinu-
soidal, 2 = perisinusoidal and portal/periportal fibrosis, 3 = bridging
fibrosis, and 4 = cirrhosis.
2.4. Statistical analysis
Descriptive statistics were computed for all variables. These
include means, standard deviations and percentiles for continuous
variables and frequencies and percentages for categorical factors. A
univariate analysis was done to assess differences between subjects
with and without NASH; Student’s t-tests or the non-parametric
Wilcoxon rank sum tests were used to compare continuous and
ordinal factors and Pearson’s chi-square tests were used to compare
categorical variables. Multivariable logistic regression analysis was
performed to build a model for prediction of NASH. An automated
stepwise variable selection method performed on 1000 bootstrap
samples was used to choose the final model. All non-invasive fac-
tors were assessed and variables with inclusion rates of 30% or
more were included in the final model. Bilirubin was modelled
with restricted cubic spline to relax linearity assumptions and an
inverse transformation for waist circumference percentile ((1/WC
percentile) * 1000) was used.
Discrimination and calibration for model performance were
used for internal model validation. Discrimination is the ability
to rank patients by risk of NASH such that patients with a higher
predicted risk are more likely to have NASH. Discrimination was
measured by the area under the receiver operating characteristics
curve (AUROC). Calibration refers to the accuracy of prediction of
the model compared to observed outcome in our dataset; this was
assessed by constructing a calibration curve (a 45◦ curve would rep-
resent perfect prediction). The method described by Harrell et al.
was used to compute the validation metrics with over-fitting bias
corrected through bootstrap resampling [15]. A thousand boot-
strap samples (B = 1000) were drawn from the original data set
and a new model with the same model settings was built on
each bootstrap resample. Prediction on patients that were not
chosen in the resample was calculated. An optimism factor was
calculated over the 1000 new models and the bias-corrected vali-
dation metric was obtained by subtracting this optimism value
from the validation metric directly measured from the original
model. A p value <0.05 was considered statistically significant.
All analyses were performed using SAS (version 9.2, The SAS
Institute, Cary, NC) and R (version 2.13.1, The R Foundation for
Statistical Computing, Vienna, Austria; packages used: Design and
ROCR).
G Model
YDLD-2688; No. of Pages 6 ARTICLE IN PRESS
K. Eng et al. / Digestive and Liver Disease xxx (2014) xxx–xxx 3

Table 1 Table 2
Demographic and clinical characteristics of subjects. Histological features of subjects.

Factor NASH (N = 203) Not NASH (N = 99) p value Factor NASH (N = 203) Not NASH (N = 99) p value

Male 68 (33.5) 42 (42.4) 0.13 Steatosis <0.001

Age at 1st visit (y) 12.3 ± 3.1 12.3 ± 3.1 0.98 <5% 0 (0.0) 2 (2.0)
Obese 188 (92.6) 80 (80.8) 0.002 5–33% 34 (16.7) 68 (68.7)
BMI (kg/m2 ) 26.3 ± 4.2 25.5 ± 3.7 0.11 34–65% 85 (41.9) 26 (26.3)
BMI percentile 95.2 ± 4.8 92.4 ± 9.6 <0.001 ≥66% 84 (41.4) 3 (3.0)
Waist circumference (cm) 91.7 ± 11.4 89.1 ± 9.5 0.044
Lobular inflammation <0.001
WC percentile 94.6 ± 3.5 92.3 ± 3.6 <0.001
None 5 (2.5) 19 (19.2)
WC/height ratio 0.60 ± 0.04 0.58 ± 0.04 <0.001
<2 under 20× 147 (72.4) 80 (80.8)
Total cholesterol (mg/dL) 163.5 ± 32.6 149.2 ± 25.3 <0.001
2–4 under 20× 48 (23.6) 0 (0.0)
HDL (mg/dL) 53.0 [39.0,69.0] 56.0 [38.0,71.0] 0.96
>4 under 20× 3 (1.5) 0 (0.0)
Non-HDL cholesterol (mg/dL) 105.8 ± 40.2 92.3 ± 32.8 0.004
Triglycerides (mg/dL) 114.0 ± 62.9 84.6 ± 41.8 <0.001 Portal inflammation <0.001
Hypercholesterolemia 121 (59.6) 41 (41.4) 0.003 None 48 (23.6) 55 (55.6)
Hypertriglyceridemia 136 (67.0) 52 (52.5) 0.015 Mild 134 (66.0) 43 (43.4)
Hypertension 55 (27.1) 29 (29.3) 0.69 More than mild 21 (10.3) 1 (1.0)
HOMA-IR 2.7 ± 1.7 2.6 ± 2.0 0.83
IGT/diabetes 88 (43.3) 45 (45.5) 0.73 Ballooninga <0.001
Metabolic syndrome 131 (64.5) 37 (37.4) <0.001 None 66 (32.8) 94 (94.9)
ALT (U/L) 73.0 [50.0,99.0] 70.0 [50.0,89.0] 0.41 Few 55 (27.4) 5 (5.1)
AST (U/L) 57.5 ± 29.3 49.9 ± 16.3 0.017 Many 80 (39.8) 0 (0.0)
GGT (U/L) 27.0 [20.0,40.0] 25.0 [17.0,33.0] 0.031 Fibrosis <0.001
Total bilirubin (mg/dL) 0.65 ± 0.24 0.73 ± 0.22 0.007 0 33 (16.3) 75 (75.8)
Albumin (g/dL) 4.6 ± 0.53 4.6 ± 0.73 0.64 1 137 (67.5) 18 (18.2)
INR 1.1 ± 0.21 1.2 ± 0.23 0.009 4.6 ± 0.73 0.64 2 15 (7.4) 4 (4.0)
NASH, non-alcoholic steatohepatitis. 3 18 (8.9) 2 (2.0)
Values presented as mean ± SD with t-test; Median [P25, P75] with Wilcoxon rank NASa 4.5 ± 1.4 2.2 ± 0.65 <0.001
sum test, or N (%) with Pearson’s chi-square test. NASH, non-alcoholic steatohepatitis; NAS, non-alcoholic fatty liver disease activity
score.
Values presented as mean ± SD with t-test or N (%) with Wilcoxan rank sum test.
3. Results a
Ballooning and NAS not available for 2 subjects.

3.1. Demographic and clinical characteristics

A total of 302 paediatric patients with biopsy-proven NAFLD

were included in the analysis. Of the total subjects, 63.6% were
female and 36.4% were male. The mean age at time of initial visit
was 12.3 ± 3.1 years. 203 subjects (67.2% of all subjects) had biopsy
proven NASH, and 99 subjects (32.8% of all subjects) were without
NASH. 88.7% of all subjects were found to be obese.
Table 1 presents a summary of patient anthropometric, clini-
cal, and laboratory characteristics. Obesity, higher BMI percentile,
metabolic syndrome, WC percentile and WC/height ratio were
significantly associated with NASH. Total cholesterol and total
bilirubin were significantly higher in NASH subjects versus not-
NASH subjects. In addition, triglycerides, AST, GGT, non-HDL
cholesterol and INR were found to be associated with the presence Fig. 1. Paediatric non-alcoholic steatohepatitis predictive model for predicting the
of NASH. presence of non-alcoholic steatohepatitis in paediatric non-alcoholic fatty liver dis-
ease. WC, waist circumference; NASH, non-alcoholic steatohepatitis.

3.2. Liver biopsy findings

Table 2 presents a summary of histological features of the
patients. As expected, subjects with NASH were more likely to have
higher grades of steatosis, lobular inflammation, and ballooning,
thus higher NAS scores than those without. PI was present in 76.3%
of patients with NASH as compared to 44.4% of patients without
NASH. The presence of fibrosis was significantly higher in patients
with NASH as compared to those without NASH; 83.8% vs 24.2%
respectively, p < 0.001. The mean NAS score in patients with NASH
was 4.5 ± 1.4, as compared to those patients without NASH who
had an NAS was 2.2 ± 0.65, p < 0.001.
variable scale according to each patient’s values. Each scale posi-
tion has corresponding prognostic points (top axis). Point values for
each predictor are determined and summed to calculate the total
point value. This value is then located in the bottom axis, where a
straight line is drawn down to determine the probability of having
NASH.
The model was internally validated by means of bootstrap-
ping and demonstrated good discrimination with AUROC of 0.737
Table 3
Wald statistics for non-alcoholic steatohepatitis: multivariable logistic regression.

Factor Chi-Square d.f. p value

3.3. Nomogram model Cholesterol 11.86 1 <0.001
(1/WC percentile) * 1000 18.61 1 <0.001
A nomogram (Fig. 1) for the risk of NASH was built. The final Total bilirubin 15.33 2 <0.001
proposed model consists of cholesterol, total bilirubin, and WC Nonlinear 11.76 1 <0.001
Total linear 42.64 4 <0.001
percentile. Table 3 presents the Wald statistic test for the model.
The nomogram is used by locating the position on each predictor d.f., degrees of freedom; WC, waist circumference.

Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016
G Model
YDLD-2688; No. of Pages 6 ARTICLE IN PRESS
4 K. Eng et al. / Digestive and Liver Disease xxx (2014) xxx–xxx
Fig. 2. Receiver operating characteristic curve estimating the accuracy of the pae-
diatric non-alcoholic steatohepatitis predictive model for predicting the presence
of non-alcoholic steatohepatitis on liver biopsy. AUROC, area under the receiver
operating characteristics curve.
(Fig. 2). This means that the model is correct 74% of the time
in identifying which patient has a higher risk among all possi-
ble comparable patient pairs. In addition, the calibration curve
(Fig. 3) showed good agreement between the observed and pre-
dicted probabilities. Further analysis revealed that a nomogram
cutoff value of 40% probability would give a 29.3% specificity and
94.1% sensitivity of having NASH, with a 73.2% positive predictive
value and 70.7% negative predictive value. Therefore, a nomogram
score of less than 40% would be reasonable to rule out NASH (in
Fig. 3. Calibration plot showing good agreement between the observed and
predicted probabilities for non-alcoholic steatohepatitis using the paediatric non-
alcoholic steatohepatitis predictive model. NASH, non-alcoholic steatohepatitis.
Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016
this study, 29 out of 41 subjects in this score category did not have
NASH). In contrast, a nomogram cut-off value of 75% would give
an 80.8% specificity and 50.7% sensitivity of having NASH, with a
84.4% positive predictive value and 44.4% negative predictive value.
A nomogram score of more than 75% therefore would be reasonable
to rule in NASH (in this study, 103 out of 122 subjects in this score
category did have NASH). Extrapolating this data to the subject pop-
ulation, by utilizing this nomogram with the above cut-off values,
theoretically a total of 163 out of 302 subjects (54% of subjects)
would have avoided a liver biopsy.
To enable widespread use, the nomogram was used to cre-
ate an online prediction tool accessible to the general public.
This prediction tool is called the paediatric NASH predictive
model or PNPM, and it can be found at: http://www.rcalc.com/
calculator.aspx?calculator id=WPFZWDPO. In Supplementary
Table S1, two study patient examples (one with biopsy proven
NASH versus one without NASH) are provided that reflect imple-
mentation of PNPM to assess for the probability of having NASH in
comparison to biopsy findings. Histologic slides of the two study
patient examples can be found in Supplementary Fig. S1.
4. Discussion
The principal finding of this current study relates to the devel-
opment of a new nomogram, the PNPM, to predict the presence
of NASH in children with NAFLD. By utilizing this easily accessible
tool that is derived from simple clinical variables, it is estimated
that around 54% of children may be able to avoid the need for liver
biopsy to diagnose NASH.
Currently, an invasive liver biopsy is the only reliable way to
diagnose NASH and identify the degree of liver damage. In such
a prevalent condition, however, an invasive test with significant
potential morbidity is not an ideal screening test. As a result, there
has been a recent focus on identifying more non-invasive measures
of screening for NASH. Several recent adult studies have examined
the combination of biochemical and clinical markers to diagnose
NASH. One of these studies has looked at clinically readily avail-
able factors, such as hypertension, presence of type II diabetes,
sleep apnoea, ALT levels, and ethnicity to develop a NASH clinical
scoring system in morbidly obese patients [16]. Other adult studies
have looked at a combination of clinical and biochemical mark-
ers, of which some studies have included alpha2macroglobulin,
transforming growth factor-B, adiponectin, and/or type IV colla-
gen, to develop a non-invasive screening model for distinguishing
between simple steatosis and NASH [17–20]. However, the most
promising biochemical serum marker has been cytokeratin-10 (CK-
18), which is a marker of hepatocyte apoptosis, and it has been
used alone as well as with other markers to identify NASH patients
[21,22]. Furthermore, CK-18 has been studied in the paediatric
population and has been found to be highly accurate in identify-
ing NASH with an excellent AUC of 0.933 [23]. Unfortunately this
promising marker is not clinically readily available.
In the paediatric population it has been a struggle to identify sin-
gle clinical variables to identify paediatric patients with progressive
disease from NAFLD. In a large multi-centre trial of 176 children, an
analysis was performed to identify clinical factors that may corre-
late with pathology of paediatric NAFLD. While certain components
of laboratory tests, such as AST and GGT were predictive of both
NAFLD pattern and presence of NASH, ultimately further analysis
did not reveal sufficient discriminate power to accurately identify
NASH and replace liver biopsy in paediatric NAFLD [24]. Our cur-
rent study brings forth a new potential screening tool that utilizes
multiple readily available variables to identify NASH in paediatric
patients with NAFLD.
G Model
YDLD-2688; No. of Pages 6 ARTICLE IN PRESS
K. Eng et al. / Digestive and Liver Disease xxx (2014) xxx–xxx
Our nomogram currently utilizes the variables of WC per-
centile, total cholesterol and total bilirubin. The rationale for the
predictive power of these variables has been previously demon-
strated. A patient’s WC percentile is a clinical measure of central
obesity. Studies have found that increased visceral obesity can
result in increased production of proinflammatory adipokines and
cytokines [25–27]. In addition, it is believed that visceral obesity
can conversely decrease the production of protective adipokines
[26,28]. It has been hypothesized that this abnormal balance in pro-
tective adipokines and proinflammatory adipokines and cytokines
may contribute to the pathophysiology of NASH. The variable of
total cholesterol in our nomogram may be reflective of an abnor-
mal lipid metabolism that may contribute to NASH [29]. Lastly, the
contribution of total bilirubin to the nomogram may be reflective
of some of the anti-oxidative protective effects that bilirubin may
have. In a study by Puri et al. [30] serum bilirubin was found to
be inversely related to the presence of NASH in children. In our
nomogram, the presence of low levels (0.3–0.5 mg/dL) or high lev-
els (1.2–1.3 mg/dL) of total bilirubin contributed to the most points
in the nomogram, compared to intermediate levels (0.6–1.1 mg/dL).
There may be a delicate balance in which total bilirubin plays an
anti-oxidative protective effects versus a reflection of disease pro-
cess.
This study has several strengths, of which one is the inclusion
of a large group of over 300 paediatric patients with biopsy-proven
NAFLD. These patients had extensive description of their anthropo-
metric, laboratory and histologic data, and they exhibited a varying
spectrum of the disease and varying degrees of fibrosis. To our
knowledge, this is the first paediatric study to develop a nomo-
gram to predict the risk of NASH in paediatric patients with NAFLD.
With our PNPM model, utilizing the nomogram cut-off value of 40%
probability provided a 29.3% specificity and a 94.1% sensitivity for
having NASH, of which was a good rule-out value. By utilizing this
lower cut-off value we were able to increase sensitivity to stratify
the risk of NASH, although this does increase false positive rates.
Currently this model does need to be validated by future studies
involving a more diverse population in order to be used clinically
while understanding the false positive risks. If validated this has
implications for access towards earlier treatment.
Currently, the mainstay of management of paediatric NAFLD,
including NASH, is that of lifestyle modification composed of
dietary modifications and increased physical activity [31]. How-
ever, we are now entering into the era of NAFLD and NASH specific
medical therapies, and current research has included evaluating
the use of metformin and vitamin E therapy in NASH [4,32], as well
as docosahexaenoic acid [33] among other medications. If future
studies were able to validate the effective use of these medications
in paediatric NAFLD, and in particular NASH, early recognition of
those children at risk for NASH would be important to ensure earlier
access to these medications.
This study does have some limitations, however. The major-
ity of our NAFLD patients had biopsy proven NASH, as well as a
higher prevalent rate of fibrosis. This patient cohort was from a
large tertiary care medical centre, thus these findings may not be
reflective of the general paediatric population in the community.
Lastly, a majority of the children were of Caucasian descent, thus it
is unclear if this nomogram would be applicable in other ethnici-
ties. Future studies are warranted to assess the fit of this model in
NAFLD patients from a general paediatric community patient pop-
ulation, as well as a more diverse patient population that includes
other ethnicities.
In conclusion, we have successfully constructed a nomogram
that utilizes simple clinical variables and routine tests that are
readily available to most practitioners. An online prediction tool
(PNPM) was created which provides for widespread public utiliza-
tion of this nomogram. By employing this nomogram, a little over
Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016
5
half of children who would have previously undergone liver biopsy
for the diagnosis of NASH may be able to avoid the risks of this
invasive diagnostic test. Future studies are needed to validate this
model externally, and if validated, this tool could be utilized as a
non-invasive method to stratify the risk of NASH in children with
NAFLD.
Conflict of interest
None declared.
Appendix A. Supplementary data
Supplementary material related to this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.dld.2014.07.016.
References
[1] Schwimmer JB, Deutsch R, Kahen T, et al. Prevalence of fatty liver in children
and adolescents. Pediatrics 2006;118:1388–93.
[2] Wieckowska A, Feldstein AE. Nonalcoholic fatty liver disease in the paediatric
population: a review. Current Opinion in Pediatrics 2005;17:636–41.
[3] Vajro P, Lenta S, Socha P, et al. Diagnosis of nonalcoholic fatty liver dis-
ease in children and adolescents: position paper of the ESPGHAN Hepatology
Committee. Journal of Pediatric Gastroenterology and Nutrition 2012;54:
700–13.
[4] Lavine JE, Schwimmer JB, Van Natta ML, et al. Effect of vitamin E or metformin
for treatment of nonalcoholic fatty liver disease in children and adolescents:
the TONIC randomized controlled trial. Journal of the American Medical Asso-
ciation 2011;305:1659–68.
[5] Dohil R, Meyer L, Schmeltzer S, et al. The effect of cysteamine bitartrate on
adiponectin multimerization in non-alcoholic fatty liver disease and healthy
subjects. Journal of Pediatrics 2012;161, 639–45.e1.
[6] Piccinino F, Sagnelli E, Pasquale G, et al. Complications following percutaneous
liver biopsy. A multicentre retrospective study on 68,276 biopsies. Journal of
Hepatology 1986;2:165–73.
[7] Kuczmarski RJ, Ogden CL, Grummer-Strawn LM, et al. CDC growth charts:
United States. Advance Data 2000:1–27.
[8] Li C, Ford ES, Mokdad AH, et al. Recent trends in waist circumference
and waist–height ratio among US children and adolescents. Pediatrics
2006;118:e1390–8.
[9] Genuth S, Alberti KG, Bennett P, et al. Follow-up report on the diagnosis of
diabetes mellitus. Diabetes Care 2003;26:3160–7.
[10] American Academy of Pediatrics. National Cholesterol Education Program:
report of the expert panel on blood cholesterol levels in children and ado-
lescents. Pediatrics 1992;89:525–84.
[11] Report of the Second Task Force on Blood Pressure Control in Children – 1987.
Task force on blood pressure control in children. National Heart, Lung, and
Blood Institute, Bethesda, Maryland. Pediatrics 1987;79:1–25.
[12] Fernandez JR, Redden DT, Pietrobelli A, et al. Waist circumference percentiles
in nationally representative samples of African-American, European-American,
and Mexican-American children and adolescents. Journal of Pediatrics
2004;145:439–44.
[13] Poynard T, Halfon P, Castera L, et al. Variability of the area under the receiver
operating characteristic curves in the diagnostic evaluation of liver fibrosis
markers: impact of biopsy length and fragmentation. Alimentary Pharmacology
and Therapeutics 2007;25:733–9.
[14] Kleiner DE, Brunt EM, Van Natta M, et al. Design and validation of a his-
tological scoring system for nonalcoholic fatty liver disease. Hepatology
2005;41:1313–21.
[15] Harrell Jr FE, Califf RM, Pryor DB, et al. Evaluating the yield of medical tests.
Journal of the American Medical Association 1982;247:2543–6.
[16] Campos GM, Bambha K, Vittinghoff E, et al. A clinical scoring system for pre-
dicting nonalcoholic steatohepatitis in morbidly obese patients. Hepatology
2008;47:1916–23.
[17] Sumida Y, Yoneda M, Hyogo H, et al. A simple clinical scoring system using
ferritin, fasting insulin, and type IV collagen 7S for predicting steatohepatitis in
nonalcoholic fatty liver disease. Journal of Gastroenterology 2011;46:257–68.
[18] Shimada M, Kawahara H, Ozaki K, et al. Usefulness of a combined evaluation
of the serum adiponectin level, HOMA-IR, and serum type IV collagen 7S level
to predict the early stage of nonalcoholic steatohepatitis. American Journal of
Gastroenterology 2007;102:1931–8.
[19] Poynard T, Ratziu V, Charlotte F, et al. Diagnostic value of biochemical markers
(NashTest) for the prediction of non alcoholo steato hepatitis in patients with
non-alcoholic fatty liver disease. BMC Gastroenterology 2006;6:34.
[20] Palekar NA, Naus R, Larson SP, et al. Clinical model for distinguishing nonalco-
holic steatohepatitis from simple steatosis in patients with nonalcoholic fatty
liver disease. Liver International 2006;26:151–6.
[21] Younossi ZM, Jarrar M, Nugent C, et al. A novel diagnostic biomarker
panel for obesity-related nonalcoholic steatohepatitis (NASH). Obesity Surgery
2008;18:1430–7.
G Model
YDLD-2688; No. of Pages 6 ARTICLE IN PRESS
6 K. Eng et al. / Digestive and Liver Disease xxx (2014) xxx–xxx

[22] Feldstein AE, Wieckowska A, Lopez AR, et al. Cytokeratin-18 fragment lev-
els as noninvasive biomarkers for nonalcoholic steatohepatitis: a multicenter
validation study. Hepatology 2009;50:1072–8.
[23] Feldstein AE, Alkhouri N, De Vito R, et al. Serum cytokeratin-18 fragment levels
are useful biomarkers for nonalcoholic steatohepatitis in children. American
Journal of Gastroenterology 2013;108:1526–31.
[24] Patton HM, Lavine JE, Van Natta ML, et al. Clinical correlates of histopath-
ology in paediatric nonalcoholic steatohepatitis. Gastroenterology 2008;135,
1961–71.e2.
[25] Angulo P. Nonalcoholic fatty liver disease. New England Journal of Medicine
2002;346:1221–31.
[26] Lyon CJ, Law RE, Hsueh WA. Minireview: adiposity, inflammation, and athero-
genesis. Endocrinology 2003;144:2195–200.
[27] Ziccardi P, Nappo F, Giugliano G, et al. Reduction of inflammatory cytokine con-
centrations and improvement of endothelial functions in obese women after
weight loss over one year. Circulation 2002;105:804–9.
[28] Lihn AS, Bruun JM, He G, et al. Lower expression of adiponectin mRNA in visceral
adipose tissue in lean and obese subjects. Molecular and Cellular Endocrinology
2004;219:9–15.
[29] Koruk M, Savas MC, Yilmaz O, et al. Serum lipids, lipoproteins and apolipopro-
teins levels in patients with nonalcoholic steatohepatitis. Journal of Clinical
Gastroenterology 2003;37:177–82.
[30] Puri K, Nobili V, Melville K, et al. Serum bilirubin level is inversely associated
with nonalcoholic steatohepatitis in children. Journal of Pediatric Gastroen-
terology and Nutrition 2013;57:114–8.
[31] Chalasani N, Younossi Z, Lavine JE, et al. The diagnosis and management
of non-alcoholic fatty liver disease: practice guideline by the American
Gastroenterological Association, American Association for the Study of
Liver Diseases, and American College of Gastroenterology. Gastroenterology
2012;142:1592–609.
[32] Sanyal AJ, Chalasani N, Kowdley KV, et al. Pioglitazone, vitamin E, or
placebo for nonalcoholic steatohepatitis. New England Journal of Medicine
2010;362:1675–85.
[33] 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. Archives of Disease in Child-
hood 2011;96:350–3.

Please cite this article in press as: Eng K, et al. A non-invasive prediction model for non-alcoholic steatohepatitis in paediatric patients
with non-alcoholic fatty liver disease. Dig Liver Dis (2014), http://dx.doi.org/10.1016/j.dld.2014.07.016