Content uploaded by Noriko Ishiguro- Tanaka
Author content
All content in this area was uploaded by Noriko Ishiguro- Tanaka on Jun 11, 2015
Content may be subject to copyright.
ORIGINAL ARTICLE
Noriko Ishiguro Æ Hiroaki Kanehisa Æ Masae Miyatani
Yoshihisa Masuo Æ Tetsuo Fukunaga
A comparison of three bioelectrical impedance analyses for predicting
lean body mass in a population with a large difference in muscularity
Accepted: 8 October 2004 / Published online: 17 December 2004
Springer-Verlag 2004
Abstract This study tested the hypothesis that, as com-
pared to whole-body bioelectrical impedance (BI) anal-
ysis, segmental BI analysis can estimate lean body mass
(LBM) more accurately in a population with a large
difference in muscularity. In addition to whole-body BI,
which determines impedance (Z) between the wrist and
ankle, two segmental BI analyses which determine the Z
value of every body segment in each of (1) the arms, legs
and trunk (distal BI) and (2) the upper arms, upper legs
and trunk (proximal BI) were applied to a group of 125
male athletes and 75 non-athletes. The subjects were
divided into validation and cross-validation groups.
Simple and multiple regression analyses were applied to
(length)
2
/Z (BI index) values for the whole-body and
each body segment, to develop the prediction equations
of LBM measured using air-displacement plethysmog-
raphy. In the validation group, the SE of estimation was
similar in the whole-body (3.4 kg, 5.4%), distal (3.4 kg,
5.5%) and proximal BI (3.3 kg, 5.2%) analyses. How-
ever, the whole-body and distal BI analyses produced
systematical errors in the estimates of LBM. Moreover,
the residuals in the two methods significantly (P<0.05)
correlated with the ratios of BI indices of the upper arms
and upper legs to those of the arms and legs, respec-
tively, calculated as variables approximating the relative
development of lean tissues at the proximal area of
limbs. On the other hand, the proximal BI analysis was
validated and cross-validated. Thus, the accuracy of
estimating LBM was similar in the whole-body and the
two segmental BI analyses. However, the prediction
equations derived from the use of the whole-body BI
index and a combination of the arms, legs and trunk BI
indices produced a systematical error relating to the
difference between the limb segments in lean tissue de-
velopment.
Keywords Body composition Æ Bioelectrical impedance
index Æ Whole body Æ Body segment Æ Athletes
Introduction
For athletes, the measurement of body composition is
helpful in determining suitable body weights for com-
petitions, and in evaluating the effects of training pro-
grams (Sinning 1996). There is increasing interest in the
use of bioelectrical impedance (BI) analysis to assess the
body composition of athletes (Lukaski et al. 1990;
Hortobagyi et al. 1992; Segal 1996 ; Pichard et al. 1997;
Fornetti et al. 1999; De Lorenzo et al. 2000; Eliakim
et al. 2000; Stewart and Hannan 2000; Huygens et al.
2002) because it is safe, non-invasive, convenient, easy,
and inexpensive (Baumgartner 1996). Previous studies
have estimated the percentage of fat mass or lean body
mass (LBM) from the ratio of (body height)
2
to
impedance (Z) measured between the wrist and ankle
with the single frequenc y of current. This approach as-
sumes that the human body is a cylindrical, isotrophic
conductor with a uniform cross-sectional area to which
Ohm’s law can be applied, indicating that the volume of
the whole body is proportional to the body height
squared divided by its Z value (Baumgartner 199 6). In
the aforementioned studies, however, prediction equa-
tions developed with the use of the whole-body BI
analysis have not been cross-validated. In addition, there
is evidence to suggest that the application of the
N. Ishiguro (&) Æ H. Kanehisa
Department of Life Sciences (Sports Sciences),
University of Tokyo, 3-8-1 Komaba,
Meguro-ku, 153-8902 Tokyo , Japan
E-mail: yuminori@lily.ocn.ne.jp
Tel.: +81-3-54546878
Fax: +81-3-54544317
M. Miyatani
Division of Health Promotion and Exercise,
National Institute of Health and Nutrition,
1-23-1 Toyama, Shinjuku-ku, 162-8636 Tokyo, Japan
Y. Masuo Æ T. Fukunaga
Department of Sport Sciences, School of Human Sciences,
Waseda University, 2-579-15 Mikajima,
Tokorozawa, 359-1192 Saitama, Japan
Eur J Appl Physiol (2005) 94: 25–35
DOI 10.1007/s00421-004-1259-2
whole-body BI analysis to athletic populations is invalid
(Stewart and Hannan 2000; Huygens et al. 2002). For
example, Huygens et al. (2002) have shown that for
estimating body composition in extreme power athletes,
BI analysis is not as accurate as anthropometric equa-
tions.
Conditioning programs result in considerable chan-
ges to the morphology of athletes compared with sed-
entary controls (Stewart and Hannan 2000). For
strength-trained athletes, muscle groups located in limbs
show a development specific to their own competitive
and training styles (Kanehisa and Fukunaga 1999).
Moreover, as a dynamic tissue aligned with the muscle
system, the skeleton also has been shown to exhibit
specific versus generalized changes in its mass with the
content of the trainin g, probably in the same way that
exercise increases the mass of the exercising muscle
(Hamdy et al. 199 4). These musculoskeletal changes
specific to performed exercise regimens may explain for
the invalidity of whole-body BI analysis for assessing
body composition in athletes, causing a deviation from
the standard cylindrical model to be large. If so, it is
hypothesized that segmental rather than who le-body BI
analysis, in which the human body is divided into seg-
ments such as the trunk, arms, and legs as cylindrical,
isotropic conductors, and the Z value of every body
segment is determined, produces a better prediction of
LBM in a population that includes not only untrained
individuals, but also athletes whose body shapes and/or
muscular development differ largely to each other.
Some studies have already tried to examine the
validity of segmental BI analysis in non-athletic popu-
lations (Chumlea et al. 1988; Baumgartner et al. 1989;
Bracco et al. 1996; Cha et al. 1997). Moreover, Organ
et al. (1994) developed vari ous electrode combinations
for determining the Z value of every body segment in
both sides of the body. In addition to the fact that no
study has tried to investigate the validity of the appli-
cation of segmental BI analysis to athletes, little infor-
mation on the influence of the electrode’s position on the
accuracy of predicting LBM is available from previous
studies. The measur es of whole-body Z value s are
dominated by arms and legs, and depend strongly on
variation in the cross-sectional areas of the distal
extremities (Baumgartner et al.1989; Fuller and Elia
1989). Scheltinga et al. (1991) and van Marken Lich-
tenbelt et al. (1994) have shown that positioning
impedance electrodes for measuring the Z value of the
proximal segments of limbs is superior to whole-body Z
measurement for predicting the total body water and/or
its change. The relative difference between strength-
trained athletes and non-athletes in the lean tissue cross-
sectional areas of limbs is more apparent in the proximal
(the upp er arms and upper legs) than in the distal seg-
ments (the forearms and lower legs) of limbs (Kanehisa
et al. 1998). Taking these findings into account, it may
be assumed that, if segmental BI analysis is applied, the
electrode placement on the proximal segments of limbs
may produce a higher accuracy of the estimation of
LBM in at hletes as compa red to that on the distal seg-
ments.
In the present stud y, the length and Z values of the
arms and legs and their proximal segments (the upper
arms and upper legs), and trunk other than the whole-
body, were determined in a sample that included athletes
and non-athletes. In addition, LBM was determ ined
using air-displacement plethysmography, and used as a
reference variable to develop prediction equations by the
application of BI analysis. The purpose of this study was
to test the hypothesis that the use of segmental rather
than whole-body Z values, especially those in the
proximal segments of limbs, can estimate LBM more
accurately in a population with a large difference in
muscularity.
Methods
Subjects
A group of 125 male athletes (6 Sumo wrestlers, 7 judo
athletes, 10 body builders, 17 wrestlers, 31 soccer play-
ers, and 54 American football players) and 75 non-ath-
letes participated voluntarily in this study. Physical
characteristics of subjects are shown in Table 1. All
athletes had participated in competitive meetings, either
at regional or national level in their own events, within
the preceding year. Data for the athletes were collected
during the period of pre-season training. Therefore,
none of the athletes were dehydrated to achie ve the re-
quired body mass for competition. The non-athletes had
not participated in any organized program of regular
physical exercise exceeding 30 min per day and 2 days
per week for at least 1 year prior to being tested. The
subjects were separated into a validation group (n=100;
3 Sumo wrestlers, 3 judo athletes, 5 body builders, 9
Table 1 Physical characteristics of subjects. Values are means (SD). BMI Body mass index, LBM lean body mass
n Age (years) Height (cm) Body mass (kg) BMI (kg m
2
) LBM (kg) Fat (%)
Sumo wrestlers 6 19.8 (1.0) 172.3 (6.9) 96.6 (25.3) 32.3 (7.3) 75.7 (13.2) 19.9 (8.9)
Judo athletes 7 23.7 (4.0) 173.1 (6.0) 77.3 (14.5) 25.6 (3.3) 69.5 (11.9) 9.7 (5.8)
Body builders 10 21.8 (2.5) 170.6 (5.1) 83.8 (16.9) 28.6 (4.4) 67.7 (7.1) 17.8 (9.3)
Wrestlers 17 23.8(2.5) 170.8(7.7) 79.1(12.8) 26.9(2.4) 69.8(9.3) 11.1(5.4)
Soccer players 31 20.4 (1.3) 174.1 (4.9) 66.5 (5.6) 21.9 (1.4) 58.4 (5.3) 12.2 (3.4)
American football players 54 20.8 (1.3) 176.2 (5.1) 81.3 (12.5) 26.2 (3.6) 67.7 (7.3) 16.1 (5.6)
Non-athlete 75 21.5 (4.9) 173.3 (5.1) 65.8 (8.5) 21.9 (2.5) 57.4 (5.8) 12.4 (5.8)
26
wrestlers, 15 soccer players, 27 American football play-
ers, and 38 non-athletes) and a cross-validation group
(n=100; 3 Sumo wrestlers, 4 judo athletes, 5 body
builders, 8 wrestlers, 16 soccer players, 27 American
football players, and 37 non-athletes). All measurements
for the athlete s were performed at intervals of more than
40 h after completion of a training session. This study
was approved by the Office of the Department of Sports
Sciences, University of Tokyo, and was consistent with
their requirements for human experimentation. The
subjects were fully informed about the procedures and
the purpose of this study. Written informed consent was
obtained from all of the subjects.
Anthropometric measurements
Protocols used for most measurements followed the
procedures described by Organ et al. (1994). Body height
and limb lengths were measured in a standing position.
Body height was measured to the nearest 0.1 cm on a
standard physician’s scale. The limb lengths of the right
side of the body were measured to the nearest 0.5 cm
with a flexible metal tape (Flat rule; KDS, Japan). In
this study, the length of each segment of the limbs was
defined as the distance between the electrodes placed to
determine the segmental Z value: forearm, distance be-
tween the head of the radius and the processus styloids;
upper arm, distance between the acromion process of the
sucapular and the lateral epicondyle of the humerus;
upper leg, distance between the greater trochanter of the
femur and articular cleft between the femur- and tibia-
condyles; lower leg, distance between the malleolus lat-
eralis and the articular cleft between the femur- and
tibiacondyles. The sum of the forearm and upper arm
lengths, and that of the upper leg and lower leg lengths,
was defined as the length of the arm and leg, respec-
tively. The length of the trunk was determined as the
difference between body height and leg length. In the
previous studies, trunk length was defined as either
the difference between acromiale height and leg length
(Chumlea et al. 1988; Baumgartner et al. 1989, 1998;
Bracco et al.1996) or that between height and leg length
(Gunnell et al. 1998, 2003; Wadsworth et al. 2002). In a
pilot study examining 38 adult males, we confirmed that
the trunk length taken in this study (i.e., height minus
leg length) was highly correlated (r=0.903, P<0.05) to
the difference between acromiale height and leg length.
For the athletes, we were obliged to shorten the time
requirement for the measurements in relation to their
training schedules. Hence, we used the difference be-
tween height and leg length as the trunk length. The
length values of the arm and leg, upper arm, upper leg,
and trunk were used for calculating the BI index [BI
index, (length)
2
/Z] of the related segment, and referred
to as L
arm
, L
leg
, L
upper arm
, L
upper leg
, and L
trunk
,
respectively.
BI (Z) measuremen ts
A BI acquisition system (Muscle a; Art Haven 9, Japan)
and disposable electrodes (Red Dot 2330; 3 M, USA)
were used to determine the Z value of each of the arms
and legs, upper arms, upper legs, and trunk, as well as
the whole body. The system applies a constant current of
Table 2 Positions of the electrodes and repeatability for impedance (Z) measurements.BIBioelectrical impedance. Z
whole-body
, Z
arm
, Z
leg
,
Z
upper arm
, and Z
upper leg
Z values of the whole-body, arms and legs, upper arms, and upper legs, respectively. Z
d-trunk
and Z
p-trunk
Z values
determined by the distal and proximal BI analyses, respectively
Method Current-injected
areas
Voltage measured
areas
Source electrodes Detector electrodes Intraclass
electrodes
%CV
Whole-body BI analysis
Z
whole-body
Right hand
to right foot
Right wrist
to right ankle
Dorsal surfaces
of the third metacarpal bone
of the right hand and third
metatarsal bone
of the right foot
Dorsal surface
of the right wrist at the level
of the head of radial and ulnar
bones and anterior surface
of the right ankle between
the protruding portions
of the tibial and fibular bones
0.970 1.9
Distal BI analysis
Z
arm
Right hand
to right foot
Right wrist
to left ankle
Dorsal surfaces
of the third metacarpal bone
of both hands and third
metatarsal bone of both feet
Dorsal surfaces of both wrists
at the level of the head of radial
and ulnar bones, anterior surface
of both ankles between the protruding
portions of the tibial and fibular bones
0.903 2.7
Z
leg
Right hand
o right foot
Right wrist
to left ankle
0.978 1.6
Z
d-trunk
Right hand
to left foot
Left wrist
to right ankle
0.763 3.8
Proximal BI analysis
Z
upper arm
Right hand
to right foot
Right elbow
to left elbow
Dorsal surfaces
of the third metacarpal bone
of both hands and third
metatarsal bone of both feet
Dorsal surfaces of both elbows
between the lateral epicondyles
of humerus and head of radius,
and articular cleft between
the femur- and tibiacondyles
0.853 2.9
Z
upper leg
Right hand
to right foot
Right knee
to left knee
0.896 2.8
Z
p-trunk
Right hand
to left foot
Left elbow
to right knee
0.760 4.0
27
500 lA at 50 kHz through the body. Subjects restrained
from vigorous exercise and alcohol intake during 24 h,
and from taking a meal during 4 h, preceding the
experiments. All Z measurements were taken with the
subjects supine, the arms relaxed at the side but not
touching the body, and the legs separated at least
25.0 cm at the ankles so that there was no contact be-
tween the upper legs. The room temperature was usually
kept at 23C (Caton et al. 1988; Liang et al. 2000).
In addition to whole-body BI analysis, two segmental
BI analyses were used to determine the Z values of every
body segment involved in each of: (1) the arms and legs
and trunk (distal BI), and (2) the upper arms, upper legs,
and trunk (proximal BI), by taking the findings of
Scheltinga et al. (1991) and van Marken Lichtenbelt
et al. (1994) into account. The positions of the electrodes
for measuring the Z value in each of the three BI anal-
yses are shown in Table 2 and Fig. 1.TheZ values of
the whole-body, arms and legs, upper arms, and upper
legs are referred to as Z
whole-body
, Z
arm
, Z
leg
, Z
upper arm
,
and Z
upper leg
, respectively. The Z values determined by
the distal and proximal BI analyses are referred to as Z
d-
trunk
and Z
p-trunk
, respectively.
The repeatability of the Z measurements for the
whole-body and each segment were assessed on 2 sepa-
rate days in a pilot study with 10 young ad ult males. For
the whole-body and segment Z values, the intraclass
correlation coefficients for the test–retest ranged from
0.760 to 0.978, and the coefficient of variation (%CV)
from 1.6 to 4.0% (Table 2). There were no significant
differences between the mean values of the two tests for
the whole-body and segment Z measurements.
Calculations of BI index
The ratios of [(length)
2
,cm
2
]toZ (ohms) for the whole-
body and every segment were calculated as BI indices,
and used as independent variables to predict LBM. The
BI indices were calculated as follows: (height)
2
/Z
whole-
body
for the whole body, (L
arm
)
2
/Z
arm
for the arms,
(L
leg
)
2
/Z
leg
for the legs, (L
upper arm
)
2
/Z
upper arm
for the
upper arms, (L
upper leg
)
2
/Z
upper leg
for the upper legs, and
(L
trunk
)
2
/Z
d-trunk
and (L
trunk
)
2
/Z
p-trunk
for the trunk in
the distal and proximal BI analyses, respectively.
Body composition measurements
The air-displacement plethysmograph (Bodpod, model
2000A; Life Measurement Instruments, USA) was used
to determine LBM. The measurement protocol was
similar to that described in prior studies (McC rory
et al. 1995; Dempster and Aitkens 1995). Briefly, the
subjects wore only a tight-fitting swimsuit and swim
cap during this experiment. Following body mass
measurement to within accuracy of 0.01 kg on a cali-
brated electric scale, the subjects sat quietly to deter-
mine the body volume in the fiberglass chamber with
normal respiration. This measurement was performed
two times, and the average volume was adopted for the
LBM calculation. To determine the thoracic gas
volume, the subjects were connected to a breathing
circuit within the system via the breathing tube, fitted
with a nose clip, and instructed to continue normal
breathing. The tidal breathing of the subjects was
recorded and displayed on the computer monitor, and
after two to three cycles of a pattern, the airway was
occluded. The subjects were then signaled by the
investigator, and they puffed against the closed airway
for about 3 s. The body density (Db) was calculated
with the equation Db (g·cm
-3
) = (body mass, g)/(body
volume, cm
3
). Once the Db value was known, the
percentage of fat mass (%Fat) was calculated using the
equation developed by Brozˇ ek et al. (1963):
%Fat = (4.570/Db 4.124)·100. Then LBM was
derived by subtra cting fat mass from the total body
mass. The measurements of body composition were
Fig. 1 Schematic representation
of the position of the electrodes
position for each bioelectrical
impedance (BI) analysis
28
completed within 3–5 min. In a pilot study, we evalu-
ated the validity of the LBM determination using the
air-displacement plethysmograph technique through
comparison with data obtained by a hydrostatic
weighing method (n=10). There was a high correlation
(r=0.966) between LBM values obtained by the two
methods, without a significant diff erence between the
average values. In addition, the slope and intercept of
the regression line for the relationship between LBM
values obtained by the two methods did not signifi-
cantly differ from one and zero, respectively. With no
correction, therefore, we used the LBM determined
using the air-displacement plethysmograph technique
as a dependent variable for developing prediction
equations with BI indices as independent variables. The
repeatability of the LBM measurements was also tested
on 2 separate days in a pilot study with nine young
males. The intraclass correlation coefficient for the test–
retest of the LBM measurements was 0.967 (P<0.05),
and %CV was 1.6%. There was no significant differ-
ence between the mean values of the two tests.
Data analysis
Firstly, in the validation group, equa tions were devel-
oped for predicting the measured LBM with the use of
BI indices as independent variables, determined in each
of the whole-body, distal, and proximal BI analyses. For
the who le-body BI, a simple regression analysis was
applied to develop a prediction equation for LBM with
(height)
2
/Z
whole-body
as an independent variable. For the
distal and proximal BI analyses, multiple regression
analyses were used to develop prediction equations for
LBM using two sets of BI indices as independent vari-
ables: (L
arm
)
2
/Z
arm
,(L
leg
)
2
/Z
leg
and (L
trunk
)
2
/Z
d-trunk
for
the distal BI analysis, and (L
upper arm
)
2
/Z
upper arm
,(L
upper
leg
)
2
/Z
upper leg
and (L
trunk
)
2
/Z
p-trunk
for the proximal BI
analysis. Secondly, it was confirmed that the regression
slope and intercept for the relationship between the
measured and estimated LBM values did not signifi-
cantly differ from one and zero, respectively. In addition
to the test of the significance of the difference between
the measured and estimated LBM values, the difference
between the two variables was plotted against the mean
LBM of the two methods to examine for systematic er-
ror, as described by Bland and Altman (1986). Thirdly,
predicted values of LBM were calculated for individuals
in the cross-validation group using the equations derived
for the validation group, which satisfied the three con-
ditions mentioned above. In the cross-validation group,
too, the same analyses as used in the validation group
were applied to examine the validity of the prediction
equation.
The standard error of the estimate (SEE) was calcu-
lated to evaluate the accuracy of the LBM estimated
with the equation in each condition. The SEE was ex-
pressed as an absolute value and relative to the mean of
the measured LBM. In addition, the standard regression
coefficient of each independent variable was calculated
in the multiple regression equations derived from the
validation group. For every independent variable se-
lected, the product of the standard regression coefficient
in the multiple regression equation and the simple cor-
relation coefficient in the relationship with the measured
LBM, expressed as a percentage, was calculated as an
index presenting its relative contribution for estimating
the measured LBM. Descriptive values are presented as
means (SD). Student’s t-test was used to test the signif-
icance of differences between the mean values of the
validation and cross-validation groups in the measured
variables, and between those of the measured and esti-
mated LBM. A simple linear regression analysis was
used to calculate the correlation coefficient (r). The
probability level for statistical signi ficance was set at
P<0.05.
Results
Baseline characteristics of the validation
and cross-validation groups
There were no significant differences between the vali-
dation and cross-validation groups in the anthropo-
metric and body composition variables (Table 3).
Table 3 Physical characteristics of the validation and cross-validation groups. MaxMaximum, Min minimum
Variable Validation group (n=100) Cross-validation group (n=100)
Mean SD Max Min Mean SD Max Min
Age (years) 21.2 3.2 45.0 18.0 21.6 3.7 47.0 18.0
Height (cm) 173.6 5.1 185.0 159.4 174.0 6.1 191.0 155.5
Body mass (kg) 73.2 13.6 120.1 48.6 73.7 14.2 126.9 50.6
BMI (kg m
2
) 24.3 4.3 42.3 16.9 24.2 3.8 38.3 17.9
LBM (kg) 62.5 9.0 88.8 42.1 63.2 9.0 87.9 46.3
Fat (%) 13.9 6.1 34.1 2.4 13.4 6.2 34.3 4.1
Segment length (cm)
Arm 56.2 2.3 62.0 50.7 56.4 2.7 63.5 50.5
Leg 79.9 3.5 86.0 71.5 80.2 3.7 88.5 72.0
Upper arm 32.4 1.9 36.0 28.4 32.5 1.7 36.5 28.7
Upper leg 39.5 1.9 42.9 34.5 40.5 2.0 44.0 36.0
Trunk 93.5 3.0 101.3 86.5 93.8 3.6 103.3 82.2
29
Moreover, no significant differences were found between
the two groups in Z values and BI indices (Table 4).
Again, there was no significant difference between
(L
trunk
)
2
/Z
d-trunk
and (L
trunk
)
2
/Z
p-trunk
, reflecting that the
current injection and voltage measurement compartment
were equivalent in principle.
Table 4 Descriptive data on Z values and BI indices in the validation and cross-validation groups. L Length
Variable Validation group (n=100) Cross-validation group (n=100)
Mean SD Max Min Mean SD Max Min
Z(ohm)
Whole-body BI analysis
Z
whole-body
414.8 66.3 622.8 280.9 407.6 55.0 536.9 283.6
Distal BI analysis
Z
arm
212.8 42.2 363.9 135.6 209.2 36.9 298.5 134.1
Z
leg
191.3 27.7 278.7 136.1 187.9 22.0 237.9 138.3
Z
d-trunk
23.0 3.0 30.7 15.7 22.5 2.9 30.7 15.1
Proximal BI analysis
Z
upper arm
96.4 19.9 154.6 58.7 94.5 18.4 137.2 51.0
Z
upper leg
55.8 7.8 77.4 40.4 54.4 7.1 70.9 37.3
Z
p-trunk
23.0 3.0 29.8 16.2 22.5 3.2 29.8 14.4
BI index (cm
2
/ohm)
Whole-body BI index
(Height)
2
/Z
whole-body
74.5 12.5 108.4 50.3 75.8 12.1 114.0 52.5
Distal BI index
(L
arm
)
2
/Z
arm
7.6 1.6 12.2 4.8 7.8 1.7 12.6 4.9
(L
leg
)
2
/Z
leg
16.9 2.8 23.8 12.4 17.3 2.7 24.1 10.9
(L
trunk
)
2
/Z
d-trunk
387.2 54.8 559.8 268.5 398.1 57.4 593.6 297.4
Proximal BI index
(L
upper arm
)
2
/Z
upper arm
5.6 1.2 8.6 3.3 5.8 1.3 9.5 3.3
(L
upper leg
)
2
/Z
upper leg
14.8 2.4 21.0 9.8 15.2 2.7 22.9 9.0
(L
trunk
)
2
/Z
p-trunk
386.5 54.9 542.5 279.7 398.5 63.2 596.9 286.1
Fig. 2 Relationship between
the whole-body BI index and
measured lean body mass
(LBM) in the validation group.
Z
whole-body
Whole-body BI
Table 5 Multiple regression model for predicting the LBM mea-
sured from the BI indices of each distal and proximal BI analysis in
the validation group. For the distal BI analysis, the measured LBM
(kg)=(3.39·X
1
)+(0.67·X
2
)+(0.03·X
3
)+15.47. For the proximal
BI analysis, the measured LBM (kg)=(3.97·X
4
)+(1.40·X
5
)+
(0.02·X
6
)+13.45
Method BI index r
2
Regression
coefficient
Regression
standard
Contribution, %
Distal
BI analysis
X
1
,(L
arm
)
2
/Z
arm
0.791 0.889 0.615 54.7
X
2
,(L
leg
)
2
/Z
leg
0.660 0.812 0.204 16.6
X
3
,(L
trunk
)
2
/Z
d-trunk
0.526 0.725 0.155 11.2
Proximal
BI analysis
X
4
,(L
upper arm
)
2
/Z
upper arm
0.789 0.888 0.523 50.0
X
5
,(L
upper leg
)
2
/Z
lower leg
0.722 0.849 0.369 31.3
X
6
,(L
trunk
)/Z
p-trunk
0.554 0.744 0.095 7.1
30
Prediction equations derived from the validation
group
The whole-body BI index was significantly correlated to
the measured LBM (r=0.912, P<0.05; Fig. 2) in the
validation group. In addition, all BI indices derived from
each of the distal and proximal BI analyses were selected
as significant contributors to predict the measured LBM
(Table 5). In the distal BI analysis, the relative contri-
bution of every BI index to the prediction of the mea-
sured LBM varied from 11.2% for (L
trunk
)
2
/Z
d-trunk
to
54.7% for (L
arm
)
2
/Z
arm
. The corresponding values for
the proximal BI analysis ranged from 7.4% for (L
trunk
)
2
/
Z
p-trunk
to 50.0% for (L
upper arm
)
2
/Z
upper arm
.
The relationships between the measured and esti-
mated LBM values are presented in Figure 3A–C. The
observed R
2
and SEE values for the three methods were
almost the same among the three equations: 0.825 to
0.849 for R
2
, and 3.3 (5.2%) to 3.4 kg (5.5%) for SEE.
The analysis indicated that the slope and intercept of the
regression equation for the relationship between the
measured and estimated LBM values in each of the three
methods were not significantly different from one and
zero, respectively. Moreover, there were no significant
differences between the measured and estimated LBM
values in all the three BI analyses. However, Bland–
Altman plots showed significant systematic errors in the
whole-body and distal BI analyses (Fig. 3D–F). Hence,
the two prediction equations derived from the whole-
body and dis tal BI analyses were excluded from the
following analysis in the cross-validation group.
Cross-validation of the prediction equation
The prediction equation derived from the validation
group by proximal BI analysis, in which (L
upper arm
)
2
/
Z
upper arm
,(L
upper leg
)
2
/Z
upper leg
and (L
trunk
)
2
/Z
p-trunk
were used as independent variables, was applied to
estimate the measured LBM in the cross-validation
group. Figure 4 indicates the relationship between the
measured and estimated LBM values (Fig. 4A), and
between the residual value of the estimate and the mean
value of the measured and estimated LBM values
Fig. 3 Relationships between
the measured and estimated
LBM (A–C), and between the
residual (difference between the
measured and estimated LBM
values) and the mean LBM
determined by two methods (D–
F) in the validation group. A
and D The corresponding
relationships for the whole-
body, B and E the
corresponding relationships for
the distal BI analysis, C and F
the corresponding relationships
for the proximal BI analysis.
SEE Standard error of estimate,
continuous lines regression lines,
dashed lines lines of identity,
dotted lines, lines of ±2SD
31
(Fig. 4B). There was no significant difference between
the measured and esti mated LBM values. The slope and
intercept of the regression equation for the relationship
between the measured and estimated LBM values were
not significantly different from one and zero, with R
2
and SEE being 0.862 and 3.5 kg (5.5%), respectively.
Moreover, no significant systematic error was found in
the Bland–Altman plot.
Relationship between residuals and BI index
ratios among segme nts
In this study, we hypothesized that the whole-body BI
analysis could not reflect accurately the difference be-
tween segments in muscular development. To confirm
whether the segment-related difference in muscular
development becomes a factor producing a systematic
error in the predicting equations in the whole-body and/
or distal BI analyses, the relationship between residuals
and BI index ratios among the segments were examined.
We firstly tried to analyze the relationships between the
measured LBM and the ratios of (L
upper arm
)
2
/Z
upper arm
to (L
arm
)
2
/Z
arm
and (L
upper leg
)
2
/Z
upper leg
to (L
leg
)
2
/Z
leg
.
The regression analyses revealed low but significant
correlations between the measured LBM and the ratios
of (L
upper arm
)
2
/Z
upper arm
to (L
arm
)
2
/Z
arm
(r=0.220,
P<0.05) and (L
upper leg
)
2
/Z
upper leg
to (L
leg
)
2
/Z
leg
(r=0.210, P<0.05). Secondly, we examined the re-
lationships between the residual values of LBM esti-
mated from the whole-body and distal BI analyses and
the two ratios of the BI index. The residual values in the
whole-body and distal BI analyses were significantly
(P<0.05) correlated to the ratios of (L
upper arm
)
2
/Z
upper
arm
to (L
arm
)
2
/Z
arm
(r=0.288 for the whole-body BI
analysis, and r=0.202 for the distal BI analysis) and
(L
upper leg
)
2
/Z
upper leg
to (L
leg
)
2
/Z
leg
(r=0.422 for the
whole-body BI analysis, and r=0.396 for the distal BI
analysis) (Fig. 5). On the other hand, the residual values
in the proximal BI analysis were not significantly cor-
related to either the ratio of (L
upper arm
)
2
/Z
upper arm
to
(L
arm
)
2
/Z
arm
(r=0.149, n.s.) or the ratio (L
upper leg
)
2
/
Z
upper leg
to (L
leg
)
2
/Z
leg
(r=0.003, n.s.).
Discussion
This study is the first to compare the accuracy of esti-
mating LBM between whole-body and segmental BI
analyses, and to cross-validate the prediction equation
validated. In the validation group, the SEE wa s 3.4 kg
(5.4%) for the whole-body BI analysis, 3.4 kg (5.5%) for
the distal BI analysis, and 3.3 kg (5.2%) for the proxi-
mal BI analysis. Previous studies using BI analysis for
the prediction of LBM have used untrained and trained
individuals separately as subject samples. We analyzed
data on both athletes and non-athletes, with the inten-
tion of elucidating the potency of BI analys is for accu-
rately predicting LBM in a population with large
variations in muscularity. Consequently, we cannot di-
rectly compare the SEE values observed in the validation
group to those reported in previous studies. However,
the absolute SEE values of the three BI analyses exam-
ined here are comparable with those reported in previ-
ous studies using the whole-body (Lukaski et al. 1985;
Jackson et al. 1988) and segmental BI analyses (Chum-
lea et al. 1988; Baumgartner et al. 1989), i.e., 2.6–4.1 kg.
On the other hand, the observed similarity in SEE values
among the three BI analyses rules out the hypothesis set
at the start of this study, i.e., that the use of segmental
rather than whole-body Z values could more accuratel y
estimate LBM by excluding the possible influences of the
difference in muscularity among the subjects on the
LBM estimates. In other words, the present result for the
SEE values in the prediction equations derived from the
three methods implies that the accuracy of esti mating
LBM in a sample with large variations in muscularity is
similar regardless of the type, whole-body or segmental,
of BI analysis.
Fig. 4 Relationships between the measured and estimated LBM
(A), and between the residual (difference between the measured and
estimated LBM values) and the mean LBM determined by two
methods (B) in the cross-validation group. Continuous line
Regression line, dashed line line of identity, dotted lines lines of
±2SD
32
In the validation group, however, the prediction
equations derived from the whole-body and distal BI
analyses showed a systematic error suggesting that the
difference in LBM influenced the accuracy of estimating
LBM with the two methods. For the validat ion groups,
the measured LBM was significantly correlated to the
ratios of (L
upper arm
)
2
/Z
upper arm
to (L
arm
)
2
/Z
arm
, and
(L
upper leg
)
2
/Z
upper leg
to (L
leg
)
2
/Z
leg
. In addition, the
residuals for whole-body and distal BI analyses in the
validation group were significantly correlated to the ra-
tios of (L
upper arm
)
2
/Z
upper arm
to (L
arm
)
2
/Z
arm
, and
(L
upper leg
)
2
/Z
upper leg
to (L
leg
)
2
/Z
leg
. It is known that the
BI index of a limb is highly correlated to lean tissue mass
(Bracco et al. 1996; Cha et al. 1997; Nun
˜
ez et al. 1999;
Tagliabue et al. 2001). Hence, the BI index ratios cal-
culated here correspond to the variables approximating
the relative development of the lean tissue masses in the
upper arms and upper legs within the arms and legs,
respectively. Considering that the proximal BI analysis,
which measured Z values in both sides of the upper arms
and upper legs and trunk, did not show systematical
error, it is likely that Z values determined by the whole-
body and distal BI analyses do not reflect the predomi-
nant development of lean tissue in the proximal areas of
the arms and legs with increasing LBM, and so might
result in systematic errors in the prediction equations
derived using the two methods. Alternatively, it may be
said that the proximal BI analysis makes it poss ible to
predict LBM without the influence of the difference in
the development of lean tissue between segments within
the same limb.
The argument mentioned above is based on the no-
tion that an increasing LBM makes a deviation from the
assumption that the human body and the arms and legs
are a cylindrical, isotropic conductor with a uniform
cross-sectional area (CSA). However, the quantitative
difference between segments in tissues making up limbs
might also explain why the proximal BI analysis was
validated, but the whole-body and distal BI analyses
were not. It is recognized that muscle is a highly con-
ductive tissue as compared to bone and fat, which are
highly resistive and/or dielectric tissues (Baumgartner
et al. 1998). In general, muscle CSA is less in the distal
than the proximal segment in each of the arms and legs.
From the findings of Kanehisa and Fukunaga (1999),
the muscle CSA of the upper leg was greater in strength-
trained athletes than in untrained subjects, but that of
the lower leg was similar between the two groups, when
the difference in (LBM)
2/3
was normalized. Baumgartner
(1996) pointed out that most of the Z value of a limb will
be determined by conductive tissue with the smallest
CSA. Considering these points, it seems that the Z val-
ues of the arms and legs would be mostly affected by
muscle mass in the distal parts of these limbs, i.e.,
forearms and lower legs, respectively, and might pro-
duce a systematic error in LBM estimated by the whole-
body and distal BI analyses. Furthermore, Heymsfield
et al. (1998) found that bone and fat masses were sig-
nificant contributors to the model for estimating the
skeletal mass of the arm in BI analysis. At the site where
whole limb CSA is maximum, the percentage of bone
CSA to the whole limb CSA is high in the forearm and
lower leg compared to the upper arm and upper leg,
respectively (Miyatani et al. 2001). In contrast, the cor-
responding percentage of fat CSA is lower in the fore-
arm and lower leg than in the upper arm and upper leg,
respectively (Miyatani et al. 200 1). Considering these
findings, therefore, it seems that the differences between
limb segments in the relative distributions of bone and
fat tissues might be a factor explaining why the whole-
Fig. 5A–D Relationships
between the residual (difference
between the measured and
estimated LBM value) in the
whole-body BI analysis and
each of the BI index ratios of
the upper arms to the arms and
the upper legs to the legs. A and
B The corresponding
relationships for the whole-
body analysis, C and D the
corresponding relationships for
the distal BI analysis.
Continuous lines Regression
lines
33
body and distal BI analyses were not validated. In any
case, we have no data on the possible differences be-
tween segments in the masses of muscle, bone and fat
tissues and their relative distributions along the limb
length. Further study is needed to test the assumptions
mentioned above.
Only the proximal BI analysis was validated and
cross-validated. In the prediction equation derived from
the proximal BI analysis, the contribution of the trunk
BI index to the estimation of LBM was very low (7.1%)
compared with that of the upper arm (50.0%) and the
upper leg BI index (31.3%). This has already been ob-
served in previous studies (Chumlea et al. 1988; Baum-
gartner et al. 1989; Organ et al. 1994; Bracco et al. 1996).
With regard to the lower contribution of the trunk BI
index for estimating LBM, Baumgartner et al. (1989)
pointed out the following: (1) the current may not be
conducted in the trunk in the same way as in the limbs,
or (2) the conduc tivity of the trunk would be 2.5–3.0
times lower than that of the leg or the arm because of the
large CSA of the trunk, anisotropy of the muscle fiber
(Baumgartner et al. 1989), and complexity of the inter-
nal structure (Chumlea et al. 1988). However, the trunk
contains about one-half of all the body mass and lean
tissue mass (Organ et al. 1994). From the findings of
Kondo et al. (1994), who investigated the relationship
between LBM and limb muscle CSA, as LBM increases,
muscle CSA also increases, but a further increase in CSA
is not apparent in LBM above 80 kg. This finding im-
plies that an increase in LBM above 80 kg largely de-
pends on the hypertrophy of lean tissues involved in the
trunk. For most of the subjects examined in this study,
the LBM values were less than 80 kg. Therefore, there is
a possibility that the sample of the subjects might be
related to the present result that the proximal BI met hod
was validated and cross-validated in spite of the lower
contribution of the trunk BI index. At the same time, the
previous finding cited above suggests a necessity for
improving the technique of Z measurement in the trunk
to accurately predict LBM in the sample of subjects with
an LBM of more than 80 kg.
In summary, the accuracy of estimating LBM was
similar in the whole-body, distal and proximal BI anal-
yses, in contrast to the hypothesis set at the start of this
study. However, the whole-body and distal BI analyses
showed a systematic error that the accuracy of the pre-
diction equation s derived using the two methods is af-
fected by the difference in LBM. On the other hand, the
proximal BI analysis, which measured Z values in both
sides of the upper arm and upper leg and trunk, was
validated and cross-validated.
Reference
Baumgartner RN (1996) Electrical impedance and total body
electrical conductivity. In: Roche AF, Heymsfield SB, Lohman
TG (eds) Human body composition. Human Kinetics, Cham-
paign, Ill., pp79–108
Baumgartner RN, Chumlea WC, Roche AF (1989) Estimation of
body composition from bioelectric impedance of body seg-
ments. Am J Clin Nutr 50:221–226
Baumgartner RN, Ross R, Heymsfield SB (1998) Does adipose
tissue influence bioelectric impedance in obese men and women?
J Appl Physiol 84:257–262
Bland JM, Altman DG (1986) Statistical methods for assessing
agreement between two methods of clinical measurement.
Lancet 1:307–310
Bracco D, Thiebaud D, Chiolero RL, Landry M, Burckhardt P,
Schutz Y (1996) Segmental body composition assessed by bio-
electrical impedance analysis and DEXA in humans. J Appl
Physiol 81:2580–2587
Brozˇ ek J, Grande F, Anderson JT, Keys A (1963) Densitometric
analysis of body composition: revision of some quantative
assumptions. Ann N Y Acad Sci 110:113–40
Caton JR, Mole PA, Adams WC, Heustis DS (1988) Body com-
position analysis by bioelectrical impedance: effect of skin
temperature. Med Sci Sports Exerc 20:489–491
Cha K, Shin S, Shon C, Choi S, Wilmore DW (1997) Evaluation of
segmental bioelectrical impedance analysis (SBIA) for measur-
ing muscle distribution. J ICHPER SD-ASIA 11–14
Chumlea WC, Baumgartner RN, Roche AF (1988) Specific
resistivity used to estimate fat-free mass from segmental
body measures of bioelectric impedance. Am J Clin Nutr
48:7–15
De Lorenzo A, Bertini I, Iacopino L, Pagliato E, Testolin C,
Testolin G (2000) Body composition measurement in highly
trained male athletes. A comparison of three methods. J Sports
Med Phys Fitness 40:178–183
Dempster P, Aitkens S (1995) A new air displacement method for
the determination of human body composition. Med Sci Sports
Exerc 27:1692–1697
Eliakim A, Ish-Shalom S, Giladi A, Falk B, Constantini N (2000)
Assessment of body composition in ballet dancers: correlation
among anthropometric measurements, bio-electrical impedance
analysis, and dual-energy X-ray absorptiometry. Int J Sports
Med 21:598–601
Fornetti WC, Pivarnik JM, Foley JM, Fiechtner JJ (1999) Reli-
ability and validity of body composition measures in female
athletes. J Appl Physiol 87:1114–1122
Fuller NJ, Elia M (1989) Potential use of bioelectrical impedance of
the ‘whole body’ and of body segments for the assessment of
body composition: comparison with densitometry and anthro-
pometry. Eur J Clin Nutr 43:779–791
Gunnell DJ, Smith GD, Frankel SJ, Kemp M, Peters TJ (1998)
Socio-economic and dietary influences on leg length and trunk
length in childhood: a reanalysis of the Carnegie (Boyd Orr)
survey of diet and health in prewar Britain (1937–1939). Pae-
diatr Perinat Epidemiol 12 [Suppl 1]:96–113
Gunnell DJ, May M, Ben-Shlomo Y, Yarnell J, Smith GD (2003)
Height, leg length, and cancer: the Caerphilly Study. Nutr
Cancer 47:34–39
Hamdy RC, Anderson JS, Whalen KE, Harvill LM (1994) Re-
gional differences in bone density of young men involved in
different exercises. Med Sci Sports Exerc 26:884–888
Heymsfield SB, Gallagher D, Grammes J, Nun
˜
ez C, Wang Z,
Pietrobelli A (1998) Upper extremity skeletal muscle mass:
potential of measurement with single frequency bioimpedance
analysis. Appl Radiat Isot 49:473–474
Hortobagyi T, Israel RG, Houmard JA, O’Brien KF, Johns RA,
Wells JM (1992) Comparison of four methods to assess body
composition in black and white athletes. Int J Sport Nutr 2:60–
74
Huygens W, Claessens AL, Thomis M, Loos R, Van Langendonck
L, Peeters M, Philippaerts R, Meynaerts E, Vlietinck R, Beunen
G (2002) Body composition estimations by BIA versus
anthropometric equations in body builders and other power
athletes. J Sports Med Phys Fitness 42:45–55
Jackson AS, Pollock ML, Graves JE, MaharmT (1988) Reliability
and validity of bioelectrical impedance in determining body
composition. J Appl Physiol 64:529–534
34
Kanehisa H, Fukunaga T (1999) Profiles of musculoskeletal
development in limbs of college Olympic weightlifters and
wrestlers. Eur J Appl Physiol 79:414–420
Kanehisa H, Kondo M, Ikegawa S, Fukunaga T (1998) Body
composition and isokinetic strength of professional Sumo
wrestlers. Eur J Appl Physiol 77:352–359
Kondo M, Asbe T, Ikegawa S, Kawakami Y, Fukunaga T (1994)
Upper limit of fat-free mass in humans: a study on Japanese
sumo wrestlers. Am J Hum Biol 6:613–618
Liang MT, Su HF, Lee NY (2000) Skin temperature and skin
blood flow affect bioelectric impedance study of female fat-free
mass. Med Sci Sports Exerc 32:221–227
Lukaski HC, Johnson PE, Bolonchuk WW, Lykken GI (1985)
Assessment of fat-free mass using bioelectrical impedance
measurements of the human body. Am J Clin Nutr 41:810–817
Lukaski HC, Bolonchuk WW, Siders WA, Hall CB (1990) Body
composition assessment of athletes using bioelectrical imped-
ance measurements. J Sports Med Phys Fitness 30:434–440
McCrory MA, Gomez TD, Bernauer EM, Mole PA (1995) Eval-
uation of a new air displacement plethysmograph for measuring
human body composition. Med Sci Sports Exerc 27:1686–1691
Miyatani M, Kanehisa H, Masuo Y, Ito M, Fukunaga T (2001)
Validity of estimating limb muscle volume by bioelectrical
impedance. J Appl Physiol 91:386–394
Nun
˜
ez C, Gallagher D, Grammes J, Baumgartner RN, Ross R,
Wang Z, Thornton J, Heymsfield SB (1999) Bioimpedance
analysis: potential for measuring leg skeletal muscle mass.
JPEN J Parenter Enteral Nutr 23:96–103
Organ LW, Bradham GB, Gore DT, Lozier SL (1994) Segmental
bioelectrical impedance analysis: theory and application of a
new technique. J Appl Physiol 77:98–112
Pichard C, Kyle UG, Gremion G, Gerbase M, Slosman DO (1997)
Body composition by x-ray absorptiometry and bioelectrical
impedance in female runners. Med Sci Sports Exerc 19:1527–
1534
Scheltinga MR, Jacobs DO, Kimbrough TD, Wilmore DW (1991)
Alterations in body fluid content can be detected by bioelec-
trical impedance analysis. J Surg Res 50:461–468
Segal KR (1996) Use of bioelectrical impedance analysis mea-
surements as an evaluation for participating in sports. Am J
Clin Nutr 64 [Suppl]:469S–471S
Sinning WE (1996) Body composition in athletes. In: Roche AF,
Hemsfield SB, Lohman TG (eds), Human body composition.
Human Kinetics, Champaign, Ill., pp 257–273
Stewart AD, Hannan WJ (2000) Prediction of fat and fat-free mass
in male athletes using dual X-ray absorptiometry as the refer-
ence method. J Sports Sci 18:263–274
Tagliabue A, Andreoli A, Comelli M, Bertoli S, Testolin G, Oriani
G, De Lorenzo A (2001) Prediction of lean body mass from
multifrequency segmental impedance: influence of adiposity.
Acta Diabetol 38:93–97
van Marken Lichtenbelt WD, Westerterp KR, Wouters L, Lu-
ijendijk SC (1994) Validation of bioelectrical-impedance mea-
surements as a method to estimate body-water compartments.
Am J Clin Nutr 60:159–166
Wadsworth ME, Hardy RJ, Paul AA, Marshall SF, Cole TJ (2002)
Leg and trunk length at 43 years in relation to childhood
health, diet and family circumstances; evidence from the 1946
national birth cohort. Int J Epidemiol 31:383–390
35
