Original
Reassessment KarlJ
Smalley,
Research
Communications-general
of body mass indices13
Anita
N Knerr,
Zebulon
V Kendrick,
Jerry
A Colliver,
and Oliver
E Owen
ABSTRACT The accuracy of body mass indices (BMls), such as Quetelet’s index, for the definition ofobesity was investigated in a large sample ofhealthy humans. Two hundred thirteen women and 1 50 men with a wide spectrum of weights, heights, and ages underwent densitometric analysis for the determination of percent body fat (%BF). %BF was then contrasted with various well-established BMIs. Although %BF was correlated with all the BMIs(r 0.60-0.82), applying objective definitions of obesity based on BMIs or %BF by densitometry
the weight-height indices provided and, thus, obesity. These findings
often
study were to reassess the agreement among the different weightheight indices in a general as opposed to an obese population and to assess the accuracy ofthe weight-height indices as predictors of body fat and fat mass determined by densitometry.
=
produced
conflicting
results.
It was
also
95% confidence intervals for predicting Quetelet’s index were very wide. Because for individuals between densitometrically and body fat as estimated by BMIs, we should be used with caution as indicators C/in
Nuir
KEY
found
that
the
%BF by using of the wide variation determined body fat conclude that BMIs of obesity. Am J
1990;52:405-8.
WORDS
Body
densitometry,
body
mass
index,
obesity,
weight,
height,
dwarves
raised
ateness
of
more
general
question
indices
for
more typical population of humans The present study was conducted commonly
used
weight-height
fat mass as determined to obese
humans
about
normal
appropri-
obesity
in
with normal stature. to assess the value
indices
as estimates
by densitometry
with
the
estimating
The
a
of six
of%BF
on a spectrum
stature.
fat
and
of lean
purposes
of the
Methods
Subjects women and 1 50 men served as volunwere studied at Temple University. This population included subjects with a wide range ofages, weights, heights, and body compositions. Caucasians, blacks, and orientals were studied. The subjects’ physical characteristics are summarized in Table 1. The nature and purpose ofthe investigations were fully explained to each subject before the study. The study was approved by the Human Institutional Review Board at Temple University. After an overnight fast, weight (kg) and height (m) were measured with the subjects barefooted and lightly clothed. UnderTwo
composition
the
weight-height
poor estimates of body with the achondroplastic
hundred
thirteen
teers. All participants
Introduction The
increasing
increased
incidence
the need
cal method
for
to identify
measuring
ofobesity
in the United
a clinically obesity
(1).
States
applicable Clinically,
and the
has
practipresence
is frequently assessed through one ofthe commonly used body mass indices (BMIs). These indices are defined as different combinations ofweight and height, such as weight divided by height, weight divided by height squared, and weight expressed as a percentage of mean weight for a given height and sex. The indices represent different attempts to adjust body weight for height to derive a height-free measure ofobesity. Although a number of weight-height indices are available, research has shown that different indices measure the same thing of obesity
in an obese patients indices
strated.
population;
almost for
the
that
is, the different
identically
(2).
estimation
of obesity
indices
However,
the has
order
obese
been
demon-
of the
National
Library
of
Medicine
(Bethesda,
that the terms BMI or Quetelet (a commonly used index) appeared in 252 journal article titles or abstracts between 1986 and 1988. These indices have been used in our research as well (3-7). In one of our recent studies of the obesity of achondroplastic dwarves (8), several commonly used weight-height indices as well as a newer index developed by Abdel-Malek et al (9) were related to percent body fat (%BF) by densitometry, a more direct method of assessing body fat (10). The results showed that A,n
J C/in Nuir
1990:52:405-8.
Printed
in USA.
repeated ues within lung
were
fat-free six or more
25
volume
(12). Body
determined
for
densitometric
mass (1 1). The consecutive
g were
obtained
was
determined
density
was
weighing
times for
until
each
used
three
to estimate
oxygen
%BF
of was
similar
individual.
by closed-circuit
then
analysis procedure
val-
Residual dilution
with
the
© 1990 American
Clinical Research Center, Temple University Philadelphia; the Biokinetics Research Laboratory, College of Health, Physical Education, Recreation and Dance, Temple University. Philadelphia: and the Division ofStatistics and Re. search Consulting and the Department oflnternal Medicine, Southern Illinois University School ofMedicine, Springfield, IL: 2 Supported in part by the US Public Health Service research grant MOl-RR 00349 (National Institutes of Health, General Clinical Research Centers Program). 3 Address reprint requests to OE Owen, Professor and Chairman, Department oflnternal Medicine, Southern Illinois University, School ofMedicine, P0 Box 19230, Springfield, IL 62794-9230. Receivediuly 10, 1989. Accepted for publication October 17, 1989. I
Nevertheless, the weight-height indices continue to be used. A literature search that used the Medlar database
widely MEDLINE MD) showed
weights
body fat and
of the
accuracy
not
water
Society
Downloaded from https://academic.oup.com/ajcn/article-abstract/52/3/405/4650856 by guest on 23 April 2018
From
the General
School of Medicine,
for Clinical
Nutrition
405
406
SMALLEY
TABLE Clinical
1 data5
ET
AL
indices.
The
advantage
generalized
Groups
Age
Weight
Percent body fat
Height
y
kg
m
%
34.6 ± 12.0 (15-82) 32.4 ± 12.0 (15-68) 37.7 ± 14.3 (18-82)
76.0 ± 18.9 (43.1-173.1) 69. 1 ± 16.5 (43.1-143.2) 85.8 ± 17.9 (59.8-171.3)
1.7 1 ± 0.09 (1.50-1.96) 1.66 ± 0.07 (1.50-1.88) 1.78 ± 0.07 (1.63-1.96)
24.0 ± 9.1 (5.0-55.2) 26.3 ± 9.4 (8.8-55.2) 20.9 ± 7.6 (5.0-52.9)
of using
indices
was
also
those used by Abdel-Malek
et al (9) were
our own estimates, the equation
women
for both
%BF All subjects (n = 363) Women (n = 213) Men (n = 150)
and statistical
Data
Pearson relationships sizes, such
i±
SD, range
sity was equation
by Brozek et al (10). Fat mass was calculated as
developed
(FATM)
etry
FATM
(%BF/lOO)
=
used
accuracy
X weight
weight-height
in estimating
indices
%BF
and
were
for indices,
evaluated
FATM.
The
based on different combinations ofweight (W, in kg)and height (H, in m), evaluated in this study were: W/H, W/H2 (Quetelet’s ( 13), W”3/H (ponderal index) ( 14), H/W”3 (Sheldon’s ( 1 5), cWt2/H33 (Abdel-Malek’s index, where c 4 x 106 for women and c 3 X 106 for men, and H is in cm) (9), and percent desirable body weight (%DBW, which is weight index) index)
more
similar
also
to develop
used
and men,
ofc,
m, and
to
k in
cWm/H(
%BF
BMIs.
W/H2 >
20%
or %DBW in men
and
to diagnose >
25%
obe-
in women
was used to determine obesity (7). Two commonly used weightheight measures ofobesity, W/H2 and %DBW, were compared with the densitometry standard. Obesity was indicated by a WI H2 27.3 in women and W/H2 27.8 in men (16) and by a
by densitom-
indices
Six commonly
their
the
(19) of using
determined.
120% in men
%DBW
Obesity
over
analyses
among
The validity
in parentheses.
=
indices
Methods
correlation coefficients ( 1 8) were used to assess the between all BMIs and measurements of body as %BF, FATM, weight, and height, as well as the
interrelationships S
sex-specific
evaluated.
and
women
(16).
Linear-regression analysis was also used to assess the utility of using W/H2 to predict %BF. The 95% confidence intervals for predicting %BF and for the mean %BF, given a W/H2, were calculated for the minimum, mean, and maximum W/H2 for both women and men. Regression analyses and confidence intervals were calculated by Minitab data-analysis software (version
6. 1, Minitab,
mc,
State
College,
PA).
=
=
expressed
as a percentage
andsex)(16, 17). Two of the indices,
TABLE 2 Intercorrelations
Allsubjects(n H W/H W/H2 W”3/H H/W”3 cW’2/H33 %DBW Women(n=
ofweight,
mean
weight
for a given
Results
height
The %DBW
and
height,
cW’2/H33,
are sex-specific
BMIs,
intercorrelated
both
generalized
(Table
and
sex specific,
2). The intercorrelations
were
strongly from
ranged
and BMIs5 W
H
W/H
W/H2
0.46 0.98 0.89 0.76 -0.75 0.52 0.91
0.26 0.02 -0.21 0.20 -0.43 0.07
0.97 0.88 -0.87 0.69 0.98
0.97 -0.96 0.82 0.99
0.02 -0.15 -0.31 0.32 -0.27 -0.06
0.98 0.94 -0.93 0.95 1.00
0.98 -0.97 0.99 1.00
0.14 -0.06 -0.26 0.26 -0.20 0.06
0.98 0.92 -0.90 0.94 1.00
0.97 -0.96 0.99 0.99
W”3/H
H/W”3
cW’2/H33
363)
=
-1.00 0.90 0.95
-0.88 -0.94
0.82
0.99 0.96
-0.98 -0.95
0.98
-0.99 0.99 0.94
-0.96 -0.93
0.96
213)
H
0.20
W/H W/H2 W”3/H H/W”3 cW’2/H33 %DBW Men(n= 150) H W/H W/H2 W”3/H
0.98 0.94 0.86 -0.85 0.88 0.96 0.33 0.98 0.92 0.82
-0.8 1
13
W’2/H33 %DBW S ,
ofthe
weight;
0.85 0.96 H, height:
BMI, body
mass
index;
% DBW,
percent
desirable
Downloaded from https://academic.oup.com/ajcn/article-abstract/52/3/405/4650856 by guest on 23 April 2018
body
weight.
-1.00
BODY TABLE 3 Correlations
ofdensitometric
data with weight,
height,
407
INDICES A
Men
0
Women
and BMIs5
Wo men
All su bjects
MASS
Men
o
0
A
E 0
%BF
FATM
%BF
FATM
%BF
FATM
w
0.47
0.79
0.76
0.92
0.61
0.85
H
-0.32
-0.04
4.)
m U
-0.21
-0.07
-0.13
0.13
W/H
0.60
0.88
0.82
0.95
0.67
0.89
W/H2
0.70
0.92
0.84
0.95
0.70
0.90
W”3/H
H/W”3 W’2/H33
0.77
0.89
0.86
0.92
0.70
0.85
-0.77 0.82
-0.87 0.84
-0.86 0.84
-0.91 0.93
-0.69 0.70
-0.82 0.88
0.70
0.92
0.83
0.96
0.69
0.90
%DBW
4.)
0 %BF, percent
S
body
fat by densitometry:
FATM,
fat mass
to
1 .00
in absolute
value
in the
group
20
composed
30
Quetelet’s
FIG 0.82
10
by densitometry.
of all
Within the sexes the intercorrelation among all the BMIs ranged from 0.92 to 1.00 in absolute value. All of the indices were strongly correlated with weight and weakly correlated with height. The correlations of all the BMIs with measures of body size
40
Index
50
60
(/2)
I. Scatter plot of percent body fat (%BF) vs Quetelet’s The solid line is the line ofidentity (y = x).
(W/H2).
index
subjects.
the
sensitivities
50.6% cated
obesity
are presented
in Table 3. The sex-specific BMIs (cW’2/H33 and %DBW) were not more strongly correlated with %BF than were
the generalized BMIs. Generally, all the BMIs were strongly correlated with %BF (r = 0.60-0.82, in absolute value) and FATM (r = 0.87-0.92, in absolute value) in all subjects. In the group composed ofwomen, the correlations between the BMIs and %BF were slightly stronger than in the group composed of men (r = 0.82-0.86 vs r = 0.67-0.70, in absolute value). The estimates ofthe power parameters in the function relating %BF to weight and height (BF = cWm/Ho) based on our sample were m = 1.38, k = 3.64, and c = 8.76 X 106 for women and 6.58
yses
106 for
men.
The
index
based
on
these
estimates
S displays
Table
of %BF
though the 95% confidence value for W/H2 are narrow, predicting
%BF
example,
ifa man
his %BF body
will
given
a value
be within
obesity.
in Figure
1.
Discussion
quite
for obesity
height
rather
insensi-
Malek
et al (9) created
related
to %BF
in the population
tive
to obesity
TABLE
and
studied,
height
were
as determined
the definitions
found
to be somewhat
by densitometry.
For
all subjects,
4
Most
ofthree
Group
quantitative
Sensitivity
definitions
of obesity
Specificityt
False
False
negatives
positives
All subjects
57.2
93.4
42.8
6.6
Women
60.4 53.2
98.2 85.9
39.6 46.8
1.8 14.1
50.6 55.4
95. 1 98.2
49.4 44.6
4.9 1.8
44.3
90.1
55.7
9.9
Proportion
t Proportion
diagnosed
obese
diagnosed
lean
j:
100%
-
sensitivity.
§
100%
-
specificity.
by the by the
BMI BMI
ifobese iflean
by the
anal-
of the
are
data.
Al-
%BF given a intervals for
quite
This
range
broad.
of%BF
borderline
variance
For reflects
malnutrition
is further
strongly
in obesity
as the ones
research
prior
area
with
than
by %BF. by %BF.
Downloaded from https://academic.oup.com/ajcn/article-abstract/52/3/405/4650856 by guest on 23 April 2018
here.
has
been
weight
and
directly
and
makes
presented
in this
correlated
to
demonstrated
thus
to obesity.
with
%BF than
method
is the most
quantitatively
will
it appears not
index.
accurately
However,
these
On
diagnose
indices
obesity,
at least
with
the correlation between W/H2 W/H2 was somewhat insensitive
as determined
true of %DBW. ered
a greater
would diseases
error
beneficial Although
is truly associated
density
by densitometry than
with
therapy. it has been
may obesity suggested
stud-
composition
(Table
4). This might
equal-magnitude
an individual obese,
alone
and that
this
was
also
be consid-
overprediction
as lean,
put
(19).
and %BF was strong to the detection of
of obesity
an
presented
in population
to body
Underprediction
be. Classifying
data
and height
validity of using weight-height be influenced by the 2.5% error
Although (r = 0.70),
body
this
for an individual.
may still have value
relating
more
for construct-
of the
on weight
not
indices,
method
associated
obesity
it was
the other
basis
is
with
body fat. Abdelthat is directly
Although
based
of the
that
correlated
with
were
the
ofmost
a BMI
weakly
direct
that an index
goal
to create
(cW’2/H33)
correlated
a weight-height
use of weight-height The
correlated
an index
strongly
individual S
obese
we can only be 95% certain
1 .7%.
ies. Our determinations of the indices to diagnose obesity may
%DBWvs%BF
Men
such
here,
%
Men W/H2 vs %BF All subjects Women
mdi-
linear-regression
W/H2
from
wide
of the research
indices
ing
Comparison
for
10.0-3
This
were
subgroup
of27,
ranging
moderate
that
and
W/H2
intervals for the man the 95% confidence
has a W/H2
compositions
men
in each
%DBW
for
of men,
ofsimple
W/H2
results (r = 0.8 1 between %BF and the index in all subjects) and, therefore, was not presented in tabular form. Weight was as strongly correlated with measures of body size as the BMls were with measures ofbody size within each sex. Table 4 demonstrates the validity of using W/H2 or %DBW to detect obesity. Although the specificity of each index was high
halfofthe
a summary
versus
57.2%
were
composed
produced
comparable
based on weight
indices
In the group
in less than
%BF criterion.
x
of the
for W/H2.
when
in fact the
individual
potentially sex-specific
at risk delay BMIs
for
possible would
SMALLEY
408 TABLES Linear-regression
analysis
of% BF vs Quetelet’s
ET AL
index5 95% confidence
Predicted
Group
95%
for mean
value(x)
Women(n = 213) [5.l]t Minimum Mean Maximum Men(n= l50)[5.5J
confidence
16.4 25.2 49.6
interval
%BFgiven
x
%
%BF
given
x
%
68 38
14.8 26.3 58.2
±
1.2
14.8 25.2
10.1
0.7
8 3
±
±
±
10. 1
± 2.9
5
58.2
± 10.4
18 77
Minimum
20.3
14.2
± 1.4
10
14.2
± 10.9
Mean Maximum
27.0 58.9
20.9 52.2
±
0.9 5.3
4 10
19.8 52.2
±
10.9
52
±
12.0
23
S
value
Predictions for the minimum, and an individual prediction.
t Root mean square f Interval expressed
mean,
and maximum
error given in brackets. as a percentage ofpredicted
±
of Quetelet’s
index
generalized
BMIs.
Garn
et al (21)
attributed
the
relationship
of
BMls to stature, lean body mass, and body frame or proportion as three limiting factors ofBMIs. These factors may be responsible for the inability of these indices to indicate obesity with greater accuracy. However, there may be other fundamental factors racy
that
have
not been
ofweight-height
different
identified
indices.
forms
Another
risks
in individuals,
W/H2