PEDIATRICOBESITY

ORIGINALRESEARCH doi:10.1111/ijpo.242

Diagnostic performance of body mass index to

ORIGINALRESEARCH
identify obesity as defined by body adiposity in
children and adolescents: a systematic review
and meta-analysis
A. Javed1, M. Jumean2, M. H. Murad3, D. Okorodudu4, S. Kumar1, V. K. Somers5,
O. Sochor5,6 and F. Lopez-Jimenez5
1
Department of Pediatric and Adolescent Medicine, Division of Pediatric Endocrinology Mayo Clinic, Rochester, MN, USA;
2
Department of Cardiovascular Diseases Tufts Medical Center, Boston, MA, USA; 3Division of Preventive, Occupational, and
Aerospace Medicine, Department of Medicine, Mayo Clinic, Rochester, MN, USA; 4Department of Internal Medicine, Duke
University Medical Center, Durham, NC, USA; 5Division of Cardiovascular Diseases, Department of Medicine, Mayo Clinic,
Rochester, MN, USA; 6International Clinical Research Center – Department of Cardiovascular Diseases, St. Anne’s University
Hospital Brno, Brno, Czech Republic

Received 24 July 2013; revised 7 October 2013; accepted 19 November 2013

Summary
Background: The ideal means of identifying obesity in children and adolescents has not been deter-
mined although body mass index (BMI) is the most widely used screening tool.
Objective: We performed a systematic review and meta-analysis of studies assessing the diagnostic
performance of BMI to detect adiposity in children up to 18 years.
Methods: Data sources were EMBASE, MEDLINE, Cochrane, Database of Systematic Reviews
Cochrane CENTRAL, Web of Science and SCOPUS up to March 2013. Studies providing measures of
diagnostic performance of BMI and using body composition technique for body fat percentage measurement
were included.
Results: Thirty-seven eligible studies that evaluated 53 521 patients, with mean age ranging from 4 to 18
years were included in the meta-analysis. Commonly used BMI cut-offs for obesity showed pooled sensitivity
to detect high adiposity of 0.73 (confidence interval [CI] 0.67–0.79), specificity of 0.93 (CI 0.88–0.96) and
diagnostic odds ratio of 36.93 (CI 20.75–65.71). Males had lower sensitivity. Moderate heterogeneity was
observed (I2 = 48%) explained in meta-regression by differences across studies in race, BMI cut-off, BMI
reference criteria (Center for Disease Control vs. International Obesity Task Force) and reference standard
method assessing adiposity.
Conclusion: BMI has high specificity but low sensitivity to detect excess adiposity and fails to identify over
a quarter of children with excess body fat percentage.

Keywords: Body mass index, meta-analysis, paediatric obesity, systematic review.

Introduction Childhood obesity tracks strongly into adult-

Obesity has become the most prevalent risk factor
for adverse cardiovascular outcome in children (1)
with 17% of American adolescents obese and 32%
either overweight or obese as defined by being equal
to or greater than the 85th percentile (but less than
the 95th) and equal to or greater than the 95th per-
centiles respectively on body mass index (BMI)
growth charts (1–4).
hood with increased risk for type 2 diabetes
mellitus and cardiovascular disease (5–8). Beyond
implications for physical well-being, obesity is often
associated with stigmatization and poor self-
esteem (9).
Timely recognition is of paramount significance in
mitigating these seriously adverse consequences.
The primary care provider, therefore, needs to be
equipped with an efficient screening tool suitable for

Address for correspondence: Dr F Lopez-Jimenez, Division of Cardiovascular Diseases, Department of Medicine, Mayo Clinic, 200 First
Street SW, Rochester, MN 55905, USA. E-mail: Lopez@mayo.edu
© 2014 The Authors
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
 2 | A. Javed et al.
widespread office-based use, yet sufficiently reliable
ORIGINALRESEARCH
to detect obesity at an early stage.
Obesity is defined as a medical condition in which
excess body fat accumulation can impact health
negatively (10). However, current assessment of
obesity is based on BMI which relies on body weight
regardless of body composition, and is calculated by
dividing the individual’s weight in kilograms by the
height in meter squared. Due to its simplicity and
epidemiologic data showing an association between
increased BMI and cardiovascular events (11), BMI
has been used in the clinical setting despite carrying
the inherent flaw of failing to distinguish between lean
and fat mass (12–14), both of which contribute to
BMI. Despite ethnic and racial differences in adipos-
ity, recommendations from several organizations,
including the American Academy of Pediatrics,
advise the use of BMI for age with national reference
data to diagnose paediatric obesity in the clinical
setting (15–19). However, previous studies have
highlighted the limitations of BMI, both in adults and
in the growing child (20–22). Further limitations of
BMI as a universal screening tool include the type of
BMI reference criteria used (e.g. National reference
for age data vs. Center for Disease Control [CDC]
(23) vs. International Obesity Task Force [IOTF]) (24),
pubertal stage and racial/ethnic differences between
the individual patient and cohort (12,25).
The cut-off value used to diagnose obesity in chil-
dren involves working out the distribution for a par-
ticular population and arbitrarily choosing particular
values, often the 85th or 95th centiles, to categorize
those with the highest BMIs (26). Several studies
have illustrated discrepancies between prevalence of
obesity within a population using different reference
growth curves (27,28).
Previous studies in children have analysed the per-
formance of BMI to detect obesity when compared
with techniques considered reference standard
methods for measurement of body adiposity, e.g.
dual energy X-ray absorptiometry (DXA), hydrostatic
weighing (HW), air-displacement plethysmography
(ADP), isotope dilution, bioelectrical impedance
analysis (BIA) and skin-fold thickness measurement
(29); however, these techniques are rarely used in
clinical practice because of their cumbersome nature
and lack of reference data on ‘normal’ cut-offs for
body fat percentage (BF%) values in children.
While a systematic review assessing the perfor-
mance of BMI in prediction of adverse cardiovascular
risk has been performed (11), no meta-analysis gen-
erating estimates of pooled sensitivity and accuracy
of BMI vs. reference standard methods of measuring
excess body fat or adiposity, such as BIA, DXA etc.,
© 2014 The Authors
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
has been conducted. The present meta-analysis
aims to assess an overall estimate of validity of BMI
for detection of obesity as defined by excess adipos-
ity in children and adolescents up to the age of 18
years, while recognizing that BMI is not used to
diagnose obesity in children less than 2 years of age.
Materials and methods
Search strategy and selection criteria
The predefined inclusion criteria for study selection
were: (i) the study must have assessed the diagnos-
tic performance of BMI to identify excess adiposity in
children aged 0–18 years; (ii) provided a 2 × 2 diag-
nostic table to allow for meta-analysis or information
to calculate these values; and (iii) used a reference
standard for body composition (e.g. DXA, ADP, HW).
Studies were excluded if they only assessed
cardiometabolic risk factors without use of any ref-
erence standard methods for measuring adiposity.
We searched the databases EMBASE (1988 to
March 2013), MEDLINE (1950 to March 2013),
Web of Science (1993 to March 2013), Cochrane,
Database of Systematic Reviews (from inception),
Cochrane CENTRAL (from inception) and SCOPUS
(1996 to March 2013). The search was conducted by
a librarian with expertise in meta-analyses with input
from investigators. Three domains were considered
absolute criteria during the search: (i) BMI or equi-
valent (e.g. Quetelet Index, BMI); (ii) diagnostic per-
formance or equivalent (e.g. sensitivity, specificity,
predictive values); and (iii) measurement of body fat
or equivalent (for example DXA, ADP, HW, BIA, body
composition). The detailed search strategy is pro-
vided (Supporting Information Appendix S1).
We eliminated irrelevant articles from our primary
search on the basis of information in title and
abstract (Fig. 1). The remaining studies were read in
their entirety by a single investigator (AJ) and those
not meeting inclusion criteria were excluded. The
search was supplemented with cross-references
from selected articles. A particular effort was made
to contact authors of articles with equivocal informa-
tion and investigators known to conduct research on
BMI diagnostic performance asking for studies that
may fulfil our inclusion criteria.
Quality assessment/data abstraction
Thirty of the studies were selected from our pri-
mary search based solely on title and abstract,
and independently reviewed by another investigator
(FL-J) and agreement coefficient determined. The
Quality Assessment of Diagnostic Accuracy Studies
(QUADAS) tool, a 0- to 14-point scale considering
 Diagnostic performance of BMI to identify obesity | 3

ORIGINALRESEARCH
Figure 1 Study flow chart illustration. Of the 1488 potentially relevant abstracts, 33 articles were included in the
meta-analysis along with four articles added via cross-reference.

Table 1 Diagnostic performance of body mass index

Pooled sensitivity Pooled specificity Positive likelihood Negative likelihood Diagnostic odds ratio
ratio ratio
Overall 0.73 (CI0.67–0.79) 0.93 (CI0.88–0.96) 10.58 (CI6.08–18.41) 0.29 (CI0.23–0.35) 36.93 (CI20.75–65.71)
Males 0.67 (CI0.56–0.76) 0.94 (CI0.84–0.98) 11.17 (4.22–29.54) 0.34 (CI0.27–0.47) 31.37 (11.53–85.31)
Females 0.71 (CI0.62–0.79) 0.95 (CI0.88–0.98) 14.56 (CI6.33–33.45) 0.30 (CI0.23–0.40) 48.02 (21.09–109.4)
CI, confidence interval.

factors that determine the validity of studies specifi- Statistical analysis

cally assessing diagnostic tests performance, was
used to review the quality of eligible studies (30). To
evaluate the quality of the study, the following criteria
were used: (i) standardization of the index method
i.e. BMI measurement, (ii) the reference standard
method used to assess body fat, (iii) blinding of BMI
measurement from BF% measurement, (iv) time
between BMI and BF% measurement and (v) details
of instructions given to subjects regarding diet and
exercise before body composition measurement.
Studies were classified as high quality (score of 10 or
greater) or low quality (score less than 10) based on
this assessment. Data regarding the population
studied, cut-off values used to define overweight or
obesity, reference standard methods used to
measure BF% and values of diagnostic performance
of BMI to detect high BF% were by a single investi-
gator (AJ) (Table 1).
Diagnostic data from each study were fitted in a
two-level mixed logistic regression model, with inde-
pendent binomial distributions for the true positives
and true negatives conditional on the sensitivity
and specificity in each study, and a bivariate normal
model for the log transforms of sensitivity and
specificity between studies. We estimated pooled
sensitivity, specificity, likelihood ratios (LR) and diag-
nostic odds ratios (DOR). The DOR, computed as
the positive LR (LR+) over negative LR (LR−), is
defined as the odds of having a positive test result in
patients with disease compared with the odds of a
positive test result in patients without disease (31).
The heterogeneity of diagnostic test parameters
was evaluated using the I2 statistic. The I2 statistic is
defined as the percentage of variation across studies
as a result of heterogeneity beyond that from chance

© 2014 The Authors

Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
 4 | A. Javed et al.
(32) with a value of 0% indicating no observed het-
ORIGINALRESEARCH
erogeneity and >50% denoting substantial heteroge-
neity. Data analysis was conducted using the
statistical program Stata (Version 11, StataCorp,
College Station, TX, USA).
To explain heterogeneity, we conducted multiva-
riable meta-regression in which the outcome variable
was the log of DOR and the explanatory variables
were covariates known to affect BMI (21). These
variables were chosen a priori as potential causes of
between-study heterogeneity and included: (i) BMI
cut-off values to define obesity, (ii) BF% cut-off values
to define obesity, (iii) gold standard used to assess
BF%, (iv) race and (v) quality assessment score.
Studies were grouped based on BMI definition of
excess adiposity using (i) BMI cut-off values into
≥95th percentile, ≥85th percentile or other, or (ii) CDC
definition or national reference criteria vs. the IOTF
definition of obesity. When more than one BMI cut-off
values were reported, the diagnostic accuracy of
BMI using the CDC or national reference criteria was
used for the overall pooled analysis. Studies were
also grouped based on BF% groups using cut-off
values to define obesity by adiposity and not by BMI,
as follows: (i) BF% of 20–25% in males and <30% in
females, (ii) other definitions of BF% cut-off, (iii) ≥95th
percentile BF% cut-off. These BF% cut-offs were
based on definitions of obesity by fat content previ-
ously used in study populations assessing adiposity
at the individual level, and mostly derived from adult
data shown to correlate with metabolic risk. Studies
included in the meta-analysis used different methods
as reference standard to assess body composition.
Based on comparable accuracy, studies that used
DXA, HW, ADP and isotope dilution measurement of
total body water were grouped together and the
pooled estimates were reported and compared with
those studies that used lower accuracy measures as
the reference standard method (BIA and skin-fold)
(33–35). For studies reporting diagnostic accuracy of
BMI when compared with more than one BF com-
position measure, data from the higher accuracy BF
composition method were used in overall pooled
analysis. Given geographic differences in body com-
position, studies were grouped based on race of the
study population into Caucasian, Asian, Hispanic or
Black. Finally, studies were categorized into two
subgroups based on QUADAS score into optimal or
suboptimal groups (cut-off score = 10) (Table 2).
Results
The search strategy yielded 1488 potential studies.
Subsequently, 37 articles met all inclusion criteria
© 2014 The Authors
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
(Fig. 1). The concordance assessment using a
sample of abstracts showed perfect agreement (cor-
relation coefficient of 1.0) between co-authors (AJ
and FL-J) when selecting studies based on title and
abstract. The studies included in the final analysis
evaluated a total of 53 521 patients, with a mean age
ranging from 4 to 18 years. Gender-specific data
were present in 34 of the 37 studies, including
18 429 males and 22 781 females.
Meta-analysis showed pooled sensitivity to detect
high adiposity of 0.73 (95% confidence interval [CI]
0.67–0.79) and pooled specificity of 0.93 (CI 0.88–
0.96). LR+ was 10.58 (CI 6.08–18.41), LR− was 0.29
(CI 0.23–0.35) and DOR was 36.93 (CI 20.75–
65.71). Receiver operating curve of BMI to detect
adiposity is shown in Fig. 2. In males, BMI showed
pooled sensitivity of 0.67 (CI 0.56–0.76) and pooled
specificity of 0.94 (CI 0.84–0.98). LR+ was 11.17
(4.22–29.54), LR− was 0.34 (CI 0.27–0.47) and DOR
was 31.37 (11.53–85.31). As for females, BMI
showed pooled sensitivity of 0.71(CI 0.62–0.79),
pooled specificity of 0.95 (CI 0.88–0.98), LR+ of
14.56 (CI 6.33–33.45), LR− of 0.30 (CI 0.23–0.40)
and DOR of 48.02 (21.09–109.4) (Table 1).
Moderate heterogeneity was observed across
studies (I2 = 48%), explored through multi variable
meta-regression. While race, BMI cut-off value,
national or CDC vs. IOTF reference criteria, definition
of BMI and reference standard method used to
measure adiposity explained heterogeneity, body fat
cut-off used or study quality did not (Appendix S2).
The characteristics of individual studies selected for
meta-analysis are presented in Table 2.
Discussion
This is the first systematic review and meta-analysis
assessing diagnostic performance of BMI in identify-
ing excess body fat as compared with reference
standard techniques of measuring adiposity, e.g.
DXA, ADP etc., in children. The results show that BMI
has high specificity in identifying paediatric obesity,
but moderate sensitivity. Pooled results from the 37
studies showed sensitivity of 73%, suggesting over a
quarter of children not labelled as obese by BMI
might indeed have excess adiposity.
These results have implications for clinicians,
public health and policymakers as well as the indi-
vidual child. Suboptimal recognition of obesity at this
crucial time translates into missed opportunities to
institute appropriate lifestyle interventions to mitigate
future health risks (65).
Previous studies report low sensitivity and high
specificity of BMI in children (14,17,38,40,42,49,52)
 Table 2 Characteristics of individual studies

Author Quality Population/ n Mean age Age SD/ % % Reference Dx criteria Dx criteria Prevalence Sensitivity Specificity
score location (years) SEM± Male Female standard according according according % %
method to reference to BMI to reference
standard standard

Eto et al. (36) 9 Japan 486 4.35 0.70 47.94 52.06 BIA 20% BF (M) 90th percentile 18.07 32.98 95.95
25% BF (F)
Wickramasinghe 11 Australia 96 9.43 2.25 45.83 54.17 DD 20% BF (M) 95th percentile 50.00 11.75 100.00
et al. (37) 30% BF (F)
Sri Lankan 42 9.28 3.18 64.29 35.71 DD 20% BF (M) 95th percentile 54.76 13.40 100.00
migrants living 30% BF (F)
in Australia
da Veiga et al. (25) 11 Brazil 1540 (10–17.9) 46.5 53.5 BIA 25% BF (M) 95th percentile 16.60 28.90 99.22
30% BF (F)
Sardinha et al. (38) 10 Portugal 55 (10–11) * 100 0 DXA 25% BF 19 kg m−2 27.3% 96 86
52 (12–13) * 100 0 DXA 25% BF 19.4 kg m−2 86 76
58 (14–15) * 100 0 DXA 25% BF 24 kg m−2 50 92
54 (10–11) * 0 100 DXA 30% BF 19.6 kg m−2 44.8% 83 87
60 (12–13) * 0 100 DXA 30% BF 21.2 kg m−2 N/A 67 97
49 (14–15) * 0 100 DXA 30% BF 21.9 kg m−2 N/A 77 89
Neovius et al. (39) 10 Sweden 474 16.84 0.26 42.19 57.81 ADP 25% BF (M) 85th percentile 29.93 48.48 94.37
30% BF (F)
Reilly et al. (40) 9 United Kingdom 238 8.50 0.40 51.68 48.32 SF 25% BF in males z score = 2 ∼ 97.7th 6.72 43.75 98.65
32% BF in females percentile
Taylor et al. (41) 10 New Zealand 368 11.89 1.97 48.64 51.36 DXA 25% BF in males Male z score = 24.99 86.18 93.11
35% BF in females 0.41 ∼ 65.8th percentile
Female z score =
0.89 ∼ 79.1th percentile
Marshall et al. (42) 10 Canada 540 10.88 0.40 49.3 50.7 HW BF >20% in males, >120% value of BMI 14.10 71.10 91.60
>25% in females
Nichols & Cadogan 10 Island of Tobago 3749 14.50 1.66 42.9 57.1 BIA BF >25% in males, >95th percentile 30.11 57.12 93.58
2009 (43) >30% in females
Fu et al. (44) 10 Singapore 623 8.35 0.24 51.5 49.5 BIA BF >95th percentile IOTF criteria extrapolated 75.00 95.80
adult > 30 kg m−2
Li et al. (13) 7 China 587 9.64 0.50 47.5 52.5 BIA BF >25% males, BMI cut-off of 20 kg m−2 34.80 26.40 99.20
>30% females used
Candido et al. (45) 10 Brazil 159 pre-pubertal 6–15 47.7 52.3 BIA >85th percentile Differing cut-offs based on 87.50 99.30
girl least misclassification
121 pubertal girl 88.90 88.50
123 post-pubertal 78.00 84.90
girl
183 pre-pubertal 95.00 92.00
boy
Diagnostic performance of BMI to identify obesity

112 pubertal boy 100.00 86.60

|

74 post-pubertal 90.00 92.20

boy

© 2014 The Authors

5

Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••

ORIGINALRESEARCH
 ORIGINALRESEARCH 6
|

Table 2 Continued

Author Quality Population/ n Mean age Age SD/ % % Reference Dx criteria Dx criteria Prevalence Sensitivity Specificity
score location (years) SEM± Male Female standard according according according % %

© 2014 The Authors

method to reference to BMI to reference
standard standard

Bedogni et al. (46) 8 Italy 934 10.00 1.00 49.89 50.11 BIA 85th percentile 95th percentile N/A 39.00 99.00
A. Javed et al.

Gaskin & Walker (47) 9 Jamaica 303 7.75 55.45 44.55 Sum SF 85th percentile Predicted 25 kg m−2 at 3.63 50.10 99.12
age 18
306 11.81 0.35 55.45 44.55 Sum SF 85th percentile Predicted 25 kg m−2 9.48 72.69 97.36
at age 18
Ghosh (48) 9 India 450 7.46 1.16 0 100 SF 85th percentile 95th percentile N/A 49.00 94.00
Lazarus et al. (49) 9 Australia 230 12.17 51.74 48.26 DXA 85th percentile 95th percentile N/A 29.00 99.00
Mei et al. (50) 8 NHANES III 4285 2–5 SF 85th percentile 85th percentile N/A 78.30 88.30
NHANES III 3279 6–11 SF 85th percentile 85th percentile 92.70 91.50
NHANES III 3189 12–15 SF 85th percentile 85th percentile 84.70 90.50
America, Italy, 162 4.76 0.58 33.33 66.66 DXA 85th percentile 85th percentile 88.50 79.40
New Zealand
America, Italy, 466 8.52 1.43 41.42 58.58 DXA 85th percentile 85th percentile 98.60 67.70
New Zealand

Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••

America, Italy, 292 14.28 2.27 46.57 53.43 DXA 85th percentile 85th percentile 100.00 72.20
New Zealand
Moreno et al. (51) 10 Spain 286 14.75 * 40.56 59.44 DXA 85th percentile 85th percentile 21.04 67.33 90.42
Sarria et al. (17) 10 Spain 175 12.80 100 0 HW 85th percentile 85th percentile N/A 50.00 91.00
Warner et al. (52) 11 United Kingdom 40 11.80 3.10 57.5 42.5 DXA 85th percentile z score = N/A 67.00 96.00
1 ∼ 84.1th percentile
United Kingdom 40 11.80 3.10 57.5 42.5 SF 85th percentile z score = 44.00 100.00
1 ∼ 84.1th percentile
Duncan et al. (53) 11 Five ethnicities 1676 11.64 0.34 0% 100% BIA >85% percentile Differing BMI cut-offs
New Zealand (age
and ethnic
specific)
Maori 216 11.50 2.60 69th 90.00 80.30
East Asian 242 12.10 2.60 68th 90.30 78.90
Pacific Island 355 11.90 2.90 69th 90.40 79.20
South Asian 183 11.30 2.50 58th 91.40 77.90
European 680 11.40 2.90 70th 92.30 66.70
Nwizu et al. (54) 10 Nigerian 753 13.59 0.29 50.1 49.9 BIA >85th percentile >85th percentile 5.40 97.50 97.19
 Freedman et al. (12) 11 North America 162 11.7 3.00 100 0 DXA >84th percentile >95th percentile N/A 65.00 N/A
White males
Black 130 12.3 4.00 83.00
Hispanic 94 12.8 3.00 82.00
Asian 192 12.0 4.00 65.00
White 141 11.5 3.00 >85th percentile 72.00
females
Black 140 11.8 3.00 89.00
Hispanic 79 12.6 3.00 88.00
Asian 166 11.8 4.00 50.00
Field et al. (55) 7 North America 596 10.00 2.20 40 60 DXA 90th percentile BMI z score > 2 60.00 51.00 100.00
Himes & Bouchard 8 Canada 316 8.4–18.99 * 50.32 49.68 HW 90th percentile 85th percentile 16.13 25.47 99.48
(56) (range)
Yoo et al. (57) 9 Korea 938 9.70 1.20 51.6 48.4 BIA >35% body fat 95th percentile N/A 71.40 79.00
Ellis et al. (22) 9 North America 979 11.37 41.47 58.53 DXA 95th percentile 95th percentile 5.11 82.00 91.71
Oliveira et al. (58) 10 Brazil 418 9–18 * 100 0 DXA 95th percentile 95th percentile N/A 81.20 90.90
years
Reilly et al. (14) 11 United Kingdom 3948 7.00 * 50.91 49.09 BIA 95th percentile 95th percentile 5.00 88.00 94.00
Zimmermann et al. 10 Sweden 2431 9.80 1.77 100 0 SF 95th percentile 95th percentile N/A 91.40 96.90
(59) 1235 9.84 1.8 0 100 SF 95th percentile 95th percentile N/A 67.90 97.30
Mast et al. (60) 10 Germany 2286 6.24 0.33 50.1 49.9 BIA 90th percentile 90th percentile 5.82 82.70 94.52
2286 6.24 0.33 50.1 49.9 SF 90th percentile 90th percentile 9.80 65.62 93.00
Mei et al. (61) 11 North America 1196 12.00 3.40 52.3 47.7 DXA >95% (range of >95th percentile N/A 92.40 89.40
cut-offs)
Bartok et al. (62) 10 North America 173 9.00 * 0 100 DXA >95th percentile 19.1 kg m−2 29.00 90.00 94.00
longitudinal
study
172 11.00 * 0 100 DXA >95th percentile 20.4 kg m−2 24.00 95.00 83.00
151 13.00 * 0 100 DXA >95th percentile 22 kg m−2 20.00 93.00 88.00
160 15.00 * 0 100 DXA >95th percentile 17 kg m−2 24.00 96.00 95.00
Schaefer et al. (16) 8 Germany 2554 11.82 49.9 50.1 SF >95th percentile >95th percentile national N/A 55.00 97.00
Malina & Katzmarzyk 11 Canada (White) 275 13.13 56.36 43.64 SF >25% M >95th percentile 5.00 54.28 99.63
(63)
Neovius & 8 Sweden 3334 15.20 0.60 51.92 48.08 BIA >95th percentile IOTF 5.00 44.30 98.74
Rasmussen (64)
Reilly et al. (11) 11 North America 7722 9.90 0.50 49.24 50.76 DXA >90th percentile >95th percentile 9.40 95.32 95.47

ADP, air-displacement plethysmography; BF, body fat; BIA, bioelectrical impedance analysis; BMI, body mass index; DD, deuterium dilution; Dx, Diagnostic; DXA, dual energy X-ray absorptiometry; F, female; HW, hydrostatic weighing;
IOTF, International Obesity Task Force; M, male; NHANES, National Health and Nutrition Examination Survey; N/A, not available; SD, standard deviation; SEM, standard error of the mean; SF, skin-fold. Sensitivity results ranged from
11.75% in an Australian population [Wickramasinghe et al. (37)] to 100% in Caucasian populations [Mei et al. (50), Candido et al. (45)]. Specificity ranged from 66.7% in a New Zealand population [Duncan et al. (53)] to a value of 100%
Diagnostic performance of BMI to identify obesity

[Warner et al. (52), Field et al. (55), Wickramasinghe et al. (37)].

|

© 2014 The Authors

7

Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••

ORIGINALRESEARCH
 8 | A. Javed et al.
ORIGINALRESEARCH
Figure 2 Receiver operating characteristics of body
mass index to detect adiposity in children.
but available studies employed different definitions of
excess adiposity and were conducted on varying eth-
nicities and age groups (46). Obesity denotes excess
body fat and BMI relies on body weight regardless of
composition. The inherent flaws of BMI stem first from
its reliance on the assumption that every individual
with normal BMI has normal body fat and secondly
that increase in BMI above normal is proportional to
increase in BF%. Increases in BMI have been accom-
panied by larger increases in BF% and concomitant
decreases in fat-free mass according to recent inves-
tigations suggesting increases in adiposity may be
higher than indicated by BMI (66–68). In adults, coex-
istence of increased BF% with normal BMI, termed
‘normal weight obesity’, has been reported to convey
an increased cardiovascular risk (69).
The moderate heterogeneity observed between
studies (I2 = 48%) was explained by race, obesity
definition using BMI, reference criteria of BMI and the
reference standard method to measure adiposity
used, but not by body fat cut-off or study quality.
We demonstrated modest benefit of using the IOTF
criteria over the CDC criteria. The decreased sensitiv-
ity of BMI to detect adiposity in males speaks to the
differential ability of BMI to diagnose obesity based on
gender. The developmental pattern of adipose tissue
accumulation and distribution is gender-specific and
changes rapidly during puberty (70). Previous studies
regarding validity of BMI report conflicting results,
some reporting higher sensitivity of BMI for boys than
girls (13,25,39,43–45,56,58–60) while others report
the opposite (14,22,37,47,49,51,54).
© 2014 The Authors
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
The practical implications of this analysis are that in
children with a BMI denoting obesity, the child is
almost definitely obese. However, with BMI being
normal, secondary tools, such as ADP or skin-fold
measurements, may aid in reliably ruling out obesity.
Despite logistic limitations, techniques, such as ADP
(BOD-POD), could be incorporated into evaluation
of children suspected of having excess adiposity
despite normal BMI (71,72). Incorporating skin-fold
measurements in clinical practice has been sug-
gested (25,73) while others refute its benefit (11,
12,41). The incorporation of a secondary technique
may be particularly important for male children with
normal BMI, although this study cannot endorse
these for widespread clinical use without evidence
that these techniques improve identification, preven-
tion and management of childhood obesity.
Visceral fat accumulation is a risk factor for obesity-
related complications (74,75). Although it is conceiv-
able, imaging modalities may offer benefit in assess-
ing visceral adiposity; studies assessing the diagnos-
tic performance of BMI were based on the assump-
tion that BMI has been intended to identify total,
not regional, body adiposity. Furthermore, studies in
obese children do not demonstrate the benefit of
imaging modalities over anthropometry (75,76).
Diagnostic tests are reported as continuous variables
therefore, there is a trade-off between sensitivity and
specificity depending on cut-off utilized with high cut-
off values yielding higher specificity whereas lower
cut-off values impact sensitivity favourably. Thus,
changing the cut-off BMI value to diagnose obesity is
likely not the solution, as lowering the cut-off will be
at the expense of decreased specificity. The current
systematic review suggests that although BMI is an
invaluable epidemiologic tool to assess population-
level risk and track national trends in obesity, it has
limitations in diagnosing obesity at the individual
level.
Limitations
There is a risk for publication bias inherent to meta-
analyses in which positive results or results with
‘expected’ findings are more likely to be published
which we attempted to minimize by contacting
investigators in the field. Also, there was moderate
heterogeneity between studies explained by race,
obesity definition using BMI, reference criteria for BMI
and reference standard method to measure adipos-
ity. In these studies, fat mass (FM) was often reported
as rankings instead of absolute values (46,49). Large
epidemiologic and prognostic studies assessing
value of different reference standard methods to for-
mulate prediction equations in children can help
 Diagnostic performance of BMI to identify obesity
determine if techniques, such as BIA or ADP, may
improve adiposity assessment (46,49).
The diagnostic performance of BMI was not differ-
ent regardless of the reference standard used, sug-
gesting that different accuracy of various methods
to assess body fat does not explain limitations of
BMI. Inclusion of studies from different geographic
areas increases applicability of findings, despite
increased heterogeneity. The use of methods for
body composition assessment not well validated
in non-Caucasian populations, particularly children,
challenges the validity of the studies performed in
non-Caucasian populations. Another practical limita-
tion is the current definition of obesity on a normative
curve instead of health consequences as prospec-
tive, well-controlled studies on BMI cut-off levels
posing risk to future health are lacking in children.
Conclusion
BMI is a very specific but less sensitive method to
assess increased body adiposity in children and fails
to identify excessive adiposity in a significant per-
centage of children. There is a need for paediatric
studies to develop reference values for skin-fold
thickness, ADP, DXA and to assess their additive or
independent ability to predict development of
obesity-associated comorbidities.
Conflict of Interest
Statement
No conflict of interest was declared.
Authors statement
This manuscript is not being submitted for publica-
tion, nor has it been published or presented in whole
or in part, elsewhere. The text and images in the
manuscript are the original work of the authors.
We attest to the validity and legitimacy of the data
and its interpretation, and attest to the fact that all au-
thors listed on the title page have read the manuscript
and agreed to its submission to Pediatric Obesity.
References
1. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of
obesity and trends in body mass index among US children
and adolescents, 1999–2010. JAMA 2012; 307: 483–490.
2. National Center for Health Statistics. Health, United
States, 2011: with Special Feature on Socioeconomic
Status and Health. Hyattsville MD: Hyattsville, MD, 2012.
3. Lo JC, Maring B, Chandra M, et al. Prevalence of
obesity and extreme obesity in children aged 3–5 years.
Pediatr Obes 2013; [Epub ahead of print].
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
| 9
4. Rolland-Cachera MF. Towards a simplified definition of
ORIGINALRESEARCH
childhood obesity? A focus on the extended IOTF refer-
ences. Pediatr Obes 2012; 7: 259–260.
5. Whitaker RC, Wright JA, Pepe MS, Seidel KD, Dietz
WH. Predicting obesity in young adulthood from childhood
and parental obesity. N Engl J Med 1997; 337: 869–873.
6. Steinberger J, Moran A, Hong CP, Jacobs DR Jr,
Sinaiko AR. Adiposity in childhood predicts obesity and
insulin resistance in young adulthood. J Pediatr 2001; 138:
469–473.
7. Srinivasan SR, Bao W, Wattigney WA, Berenson GS.
Adolescent overweight is associated with adult overweight
and related multiple cardiovascular risk factors: the
Bogalusa Heart Study. Metabolism 1996; 45: 235–240.
8. Freedman DS, Mei Z, Srinivasan SR, Berenson GS,
Dietz WH. Cardiovascular risk factors and excess adiposity
among overweight children and adolescents: the Bogalusa
Heart Study. J Pediatr 2007; 150: 12–17, e2.
9. Griffiths LJ, Parsons TJ, Hill AJ. Self-esteem and quality
of life in obese children and adolescents: a systematic
review. Int J Pediatr Obes 2010; 5: 282–304.
10. WHO. Fact Sheet on Obesity and Overweight. World
Health Organization Media Center, Geneva, Switzerland,
2006.
11. Reilly JJ, Kelly J, Wilson DC. Accuracy of simple clini-
cal and epidemiological definitions of childhood obesity:
systematic review and evidence appraisal. Obes Rev 2010;
11: 645–655.
12. Freedman DS, Wang J, Thornton JC, et al. Racial/
ethnic differences in body fatness among children and
adolescents. Obesity (Silver Spring) 2008; 16: 1105–1111.
13. Li S, Zhang M, Yang S, Okada T, Iwata F, Harada K.
Age- and sex-specific body composition of Chinese chil-
dren. Acta Paediatr 2005; 94: 1139–1142.
14. Reilly JJ, Dorosty AR, Emmett PM. Identification of the
obese child: adequacy of the body mass index for clinical
practice and epidemiology. Int J Obes Relat Metab Disord
2000; 24: 1623–1627.
15. August GP, Caprio S, Fennoy I, et al. Prevention and
treatment of pediatric obesity: an endocrine society clinical
practice guideline based on expert opinion. J Clin
Endocrinol Metab 2008; 93: 4576–4599.
16. Schaefer F, Georgi M, Wuhl E, Scharer K. Body mass
index and percentage fat mass in healthy German school-
children and adolescents. Int J Obes Relat Metab Disord
1998; 22: 461–469.
17. Sarria A, Moreno LA, Garcia-Llop LA, Fleta J, Morellon
MP, Bueno M. Body mass index, triceps skinfold and
waist circumference in screening for adiposity in male chil-
dren and adolescents. Acta Paediatr 2001; 90: 387–392.
18. Barlow SE. Expert committee recommendations
regarding the prevention, assessment, and treatment of
child and adolescent overweight and obesity: summary
report. Pediatrics 2007; 120(Suppl. 4): S164–S192.
19. Centre for Public Health Excellence at NICE (UK),
National Collaborating Centre for Primary Care (UK).
Obesity: the Prevention, Identification, Assessment and
Management of Overweight and Obesity in Adults and
© 2014 The Authors
 10 | A. Javed et al.
Children. National Institute for Health and Clinical Excel-
ORIGINALRESEARCH
lence: London, 2006.
20. Romero-Corral A, Somers VK, Sierra-Johnson J, et al.
Accuracy of body mass index in diagnosing obesity in the
adult general population. Int J Obes (Lond) 2008; 32:
959–966.
21. Okorodudu DO, Jumean MF, Montori VM, et al. Diag-
nostic performance of body mass index to identify obesity
as defined by body adiposity: a systematic review and
meta-analysis. Int J Obes (Lond) 2010; 34: 791–799.
22. Ellis KJ, Abrams SA, Wong WW. Monitoring childhood
obesity: assessment of the weight/height index. Am J
Epidemiol 1999; 150: 939–946.
23. Kuczmarski RJ, Ogden CL, Grummer-Strawn LM,
et al. CDC growth charts: United States. Adv Data 2000;
314: 1–27.
24. Bellizzi MC, Dietz WH. Workshop on childhood
obesity: summary of the discussion. Am J Clin Nutr 1999;
70(1 Pt 2): 173S–175S.
25. da Veiga GV, Dias PC, dos Anjos LA. A comparison of
distribution curves of body mass index from Brazil and the
United States for assessing overweight and obesity in Bra-
zilian adolescents. Rev Panam Salud Publica 2001; 10:
79–85.
26. Troiano RP, Flegal KM. Overweight prevalence among
youth in the United States: why so many different
numbers? Int J Obes Relat Metab Disord 1999; 23(Suppl.
2): S22–S27.
27. Flegal KM, Ogden CL, Wei R, Kuczmarski RL,
Johnson CL. Prevalence of overweight in US children:
comparison of US growth charts from the Centers for
Disease Control and Prevention with other reference values
for body mass index. Am J Clin Nutr 2001; 73: 1086–1093.
28. Chinn S, Rona RJ. International definitions of over-
weight and obesity for children: a lasting solution? Ann
Hum Biol 2002; 29: 306–313.
29. Sweeting HN. Measurement and definitions of obesity
in childhood and adolescence: a field guide for the unini-
tiated. Nutr J 2007; 6: 32.
30. Whiting P, Rutjes AW, Reitsma JB, Bossuyt PM,
Kleijnen J. The development of QUADAS: a tool for the
quality assessment of studies of diagnostic accuracy
included in systematic reviews. BMC Med Res Methodol
2003; 3: 25.
31. Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt
PM. The diagnostic odds ratio: a single indicator of
test performance. J Clin Epidemiol 2003; 56: 1129–
1135.
32. Higgins JP, Thompson SG, Deeks JJ, Altman DG.
Measuring inconsistency in meta-analyses. BMJ 2003;
327: 557–560.
33. Warolin J, Whitaker LE, Choi L, Acra S, Buchowski
MS. Comparative assessment of change in fat mass
using dual X-ray absorptiometry and air-displacement
plethysmography. Clin Obes 2012; 2: 66–72.
34. Minderico CS, Silva AM, Teixeira PJ, Sardinha LB,
Hull HR, Fields DA. Validity of air-displacement
plethysmography in the assessment of body composition
© 2014 The Authors
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
changes in a 16-month weight loss program. Nutr Metab
(Lond) 2006; 3: 32.
35. Frisard MI, Greenway FL, Delany JP. Comparison of
methods to assess body composition changes during a
period of weight loss. Obes Res 2005; 13: 845–854.
36. Eto C, Komiya S, Nakao T, Kikkawa K. Validity of the
body mass index and fat mass index as an indicator of
obesity in children aged 3–5 year. J Physiol Anthropol Appl
Human Sci 2004; 23: 25–30.
37. Wickramasinghe VP, Cleghorn GJ, Edmiston KA,
Murphy AJ, Abbott RA, Davies PS. Validity of BMI as a
measure of obesity in Australian white Caucasian and
Australian Sri Lankan children. Ann Hum Biol 2005; 32:
60–71.
38. Sardinha LB, Going SB, Teixeira PJ, Lohman TG.
Receiver operating characteristic analysis of body mass
index, triceps skinfold thickness, and arm girth for obesity
screening in children and adolescents. Am J Clin Nutr
1999; 70: 1090–1095.
39. Neovius MG, Linne YM, Barkeling BS, Rossner SO.
Sensitivity and specificity of classification systems for
fatness in adolescents. Am J Clin Nutr 2004; 80: 597–603.
40. Reilly JJ, Savage SA, Ruxton CH, Kirk TR. Assessment
of obesity in a community sample of prepubertal children. Int
J Obes Relat Metab Disord 1999; 23: 217–219.
41. Taylor RW, Falorni A, Jones IE, Goulding A. Identifying
adolescents with high percentage body fat: a comparison
of BMI cutoffs using age and stage of pubertal develop-
ment compared with BMI cutoffs using age alone. Eur J
Clin Nutr 2003; 57: 764–769.
42. Marshall JD, Hazlett CB, Spady DW, Conger PR,
Quinney HA. Validity of convenient indicators of obesity.
Hum Biol 1991; 63: 137–153.
43. Nichols SD, Cadogan F. BMI-based obesity cutoffs
and excess adiposity in a Caribbean adolescent popu-
lation of African origin. Eur J Clin Nutr 2009; 63: 253–
258.
44. Fu WP, Lee HC, Ng CJ, et al. Screening for childhood
obesity: international vs population-specific definitions.
Which is more appropriate? Int J Obes Relat Metab Disord
2003; 27: 1121–1126.
45. Candido AP, Freitas SN, Machado-Coelho GL.
Anthropometric measurements and obesity diagnosis in
schoolchildren. Acta Paediatr 2011; 100: e120–e124.
46. Bedogni G, Iughetti L, Ferrari M, et al. Sensitivity and
specificity of body mass index and skinfold thicknesses in
detecting excess adiposity in children aged 8–12 years.
Ann Hum Biol 2003; 30: 132–139.
47. Gaskin PS, Walker SP. Obesity in a cohort of black
Jamaican children as estimated by BMI and other indices
of adiposity. Eur J Clin Nutr 2003; 57: 420–426.
48. Ghosh A. Receiver operating characteristic (ROC)
curve analysis in 5–10-year-old Bengalee girls from Cal-
cutta, India. Ann Hum Biol 2004; 31: 364–369.
49. Lazarus R, Baur L, Webb K, Blyth F. Body mass index
in screening for adiposity in children and adolescents: sys-
tematic evaluation using receiver operating characteristic
curves. Am J Clin Nutr 1996; 63: 500–506.
 Diagnostic performance of BMI to identify obesity
50. Mei Z, Grummer-Strawn LM, Pietrobelli A, Goulding A,
Goran MI, Dietz WH. Validity of body mass index com-
pared with other body-composition screening indexes for
the assessment of body fatness in children and adoles-
cents. Am J Clin Nutr 2002; 75: 978–985.
51. Moreno LA, Blay MG, Rodriguez G, et al. Screening
performances of the International Obesity Task Force body
mass index cut-off values in adolescents. J Am Coll Nutr
2006; 25: 403–408.
52. Warner JT, Cowan FJ, Dunstan FD, Gregory JW. The
validity of body mass index for the assessment of adiposity
in children with disease states. Ann Hum Biol 1997; 24:
209–215.
53. Duncan JS, Duncan EK, Schofield G. Accuracy of
body mass index (BMI) thresholds for predicting excess
body fat in girls from five ethnicities. Asia Pac J Clin Nutr
2009; 18: 404–411.
54. Nwizu SE, Njokanma OF, Okoromah CA, David NA.
Relationship between bioelectrical impedance analysis and
body mass index in adolescent urban Nigerians. West Afr
J Med 2011; 30: 99–103.
55. Field AE, Laird N, Steinberg E, Fallon E,
Semega-Janneh M, Yanovski JA. Which metric of relative
weight best captures body fatness in children? Obes Res
2003; 11: 1345–1352.
56. Himes JH, Bouchard C. Validity of anthropometry in
classifying youths as obese. Int J Obes 1989; 13: 183–193.
57. Yoo S, Lee SY, Kim KN, Sung E. Obesity in Korean
pre-adolescent school children: comparison of various
anthropometric measurements based on bioelectrical im-
pedance analysis. Int J Obes (Lond) 2006; 30: 1086–1090.
58. Oliveira FL, Taddei JA, Escrivao MA, et al. Accuracy of
obesity diagnosis in Brazilian adolescents: comparison of
Cole, et al and Must, et al criteria with DXA percentage
of fat mass. Nutr Hosp 2006; 21: 484–490.
59. Zimmermann MB, Gubeli C, Puntener C, Molinari L.
Detection of overweight and obesity in a national sample of
6–12-y-old Swiss children: accuracy and validity of refer-
ence values for body mass index from the US Centers for
Disease Control and Prevention and the International
Obesity Task Force. Am J Clin Nutr 2004; 79: 838–843.
60. Mast M, Langnase K, Labitzke K, Bruse U, Preuss U,
Muller MJ. Use of BMI as a measure of overweight and
obesity in a field study on 5–7 year old children. Eur J Nutr
2002; 41: 61–67.
61. Mei Z, Grummer-Strawn LM, Wang J, et al. Do
skinfold measurements provide additional information to
body mass index in the assessment of body fatness
among children and adolescents? Pediatrics 2007; 119:
e1306–e1313.
62. Bartok CJ, Marini ME, Birch LL. High body mass index
percentile accurately reflects excess adiposity in white
girls. J Am Diet Assoc 2011; 111: 437–441.
63. Malina RM, Katzmarzyk PT. Validity of the body mass
index as an indicator of the risk and presence of overweight
in adolescents. Am J Clin Nutr 1999; 70: 131S–136S.
64. Neovius M, Rasmussen F. Evaluation of BMI-based
classification of adolescent overweight and obesity: choice
Pediatric Obesity © 2014 World Obesity. Pediatric Obesity ••, ••–••
| 11
of percentage body fat cutoffs exerts a large influence. The
ORIGINALRESEARCH
COMPASS study. Eur J Clin Nutr 2008; 62: 1201–1207.
65. Barton M. Screening for obesity in children and ado-
lescents: US Preventive Services Task Force recommen-
dation statement. Pediatrics 2010; 125: 361–367.
66. Wells JC, Coward WA, Cole TJ, Davies PS. The con-
tribution of fat and fat-free tissue to body mass index in
contemporary children and the reference child. Int J Obes
Relat Metab Disord 2002; 26: 1323–1328.
67. Ruxton CH, Reilly JJ, Kirk TR. Body composition of
healthy 7- and 8-year-old children and a comparison with
the ‘reference child’. Int J Obes Relat Metab Disord 1999;
23: 1276–1281.
68. McCarthy HD, Ellis SM, Cole TJ. Central overweight
and obesity in British youth aged 11–16 years: cross sec-
tional surveys of waist circumference. BMJ 2003; 326: 624.
69. Romero-Corral A, Somers VK, Sierra-Johnson J, et al.
Normal weight obesity: a risk factor for cardiometabolic
dysregulation and cardiovascular mortality. Eur Heart J
2010; 31: 737–746.
70. Taylor RW, Grant AM, Williams SM, Goulding A. Sex
differences in regional body fat distribution from pre- to
postpuberty. Obesity (Silver Spring) 2010; 18: 1410–1416.
71. Ma G, Yao M, Liu Y, et al. Validation of a new pediatric
air-displacement plethysmograph for assessing body
composition in infants. Am J Clin Nutr 2004; 79: 653–660.
72. Michels N, Huybrechts I, Bammann K, et al. Caucasian
children’s fat mass: routine anthropometry v. air-displace-
ment plethysmography. Br J Nutr 2013; 109: 1528–1537.
73. Nooyens AC, Koppes LL, Visscher TL, et al. Adoles-
cent skinfold thickness is a better predictor of high body
fatness in adults than is body mass index: the Amsterdam
Growth and Health Longitudinal Study. Am J Clin Nutr
2007; 85: 1533–1539.
74. Mostazir M, Jeffery A, Voss L, Wilkin T. Gender-
assortative waist circumference in mother-daughter and
father-son pairs, and its implications. An 11-year longitu-
dinal study in children (EarlyBird 59). Pediatr Obes 2013;
[Epub ahead of print].
75. Koot BG, Westerhout R, Bohte AE, et al. Ultrasonog-
raphy is not more reliable than anthropometry for assess-
ing visceral fat in obese children. Pediatr Obes 2013; [Epub
ahead of print].
76. Brambilla P, Bedogni G, Moreno LA, et al.
Crossvalidation of anthropometry against magnetic reso-
nance imaging for the assessment of visceral and subcu-
taneous adipose tissue in children. Int J Obes (Lond) 2006;
30: 23–30.
Supporting Information
Additional Supporting Information may be found in
the online version of this article at the publisher’s
web-site:
Appendix S1. Search strategy.
Appendix S2. Diagnostic performance of BMI
across predefined subgroups.
© 2014 The Authors