Upload
tatrc
View
0
Download
0
Embed Size (px)
Citation preview
764 Am J (un Nuir l992;55:764-70. Printed in USA. © 1992 American Society for Clinical Nutrition
Original Research Communications-general
Reliability of body-fat estimations from a four-compartmentmodel by using density, body water, andbone mineral measurements14
Karl E Fried!, Jane P DeLuca, Louis J Marchiteii, and James A Vogel
ABSTRACF Reliability of body-fat estimation by a four-
compartment model was tested in 10 subjects. Body densities
were measured by underwater weighing (UWW), total body wa-
ter (TBW) by deuterium dilution, and total body bone mass(TBBM) by dual-energy x-ray absorptiometry in three sessionsin 1 wk. Percent body fat was determined by [2.559/density
- 0.734 (TBW/weight) + 0.983 (TBBM/weight) - 1.841] X 100.
Reliability coefficients were 0.991 and 0.994, and within-subjects
standard deviations were ± 1.0 and ± 1 . 1 for percent body-fat
estimations from Sin’s two-compartment and the four-corn-
partment models, respectively; fat mass was ±0.8 kg with both
models. These data suggest that additive errors in the multicorn-
partrnent model do not offset the improved accuracy of fat es-
tirnations over those obtained from UWW alone. The greatest
source of error came from the UWW procedure itself (±0.002
g/crn3, or 1.0% of body weight), followed by error in TBW
(±0.5 L). More reproducible passive methods that are not de-
pendent on hydration or TBBM may be especially useful after
validation against the four-compartment model. Am J Clin
Nuir l992;55:764-70.
KEY WORDS Body composition, body fat, densitometry,
body-water analysis, x-ray absorptiometry, bone mass, reliability,
multicompartrnent models, electrical impedance
Introduction
Over 30 y ago, Allen et al (1) at the Army Medical Research
and Nutrition Laboratory proposed the use of a four-cornpart-
ment model to estimate body fat as an improvement over esti-
mations made by two-compartment models, which assume a
single density for the combined nonfat components (2, 3). At
that time, tracer methods used to measure total body water(TBW) were still relatively unavailable to most laboratories andtotal body bone mineral (TBBM) could only be roughly esti-
mated from anthropometry. Simplification of methods of anal-
ysis for TBW measurement using deuterium (4), and the de-
velopment of dual-energy x-ray absorptiometry (DEXA) for
TBBM (5, 6), have made it possible to practically and safely
measure these two components. This improved technology now
makes it practical to use four-compartment models as a criterion
method against which other indirect methods can be compared
and developed. The accuracy of this approach was demonstratedby Heymsfield et al (7), who compared it with the more-involved
criterion method of neutron-activation analysis, but the repro-
ducibility of these methods has not been examined.
This study was conducted to determine whether the additive
errors from the individual measurements for the four-compart-ment model might introduce more error in precision than gainsin accuracy obtained by assessing the additional major body
components of TBW and TBBM.
Methods
Ten soldiers volunteered to participate in this study and gavetheir written, informed consent; and the study was conductedin accordance with US Army Regulation 70-25, concerning the
use of volunteers in research. There were no selection criteria,other than age (< 40 y) and good health. Descriptive data col-lected at the start of the experiment are shown in Table 1. All
subjects were comfortable with water submersion but all werenaive to the procedure of underwater weighing (UWW) at theoutset of the study. The percentage of forced vital capacity cx-pired in the first second was > 75% for each subject.
General testing procedure
Each subject was studied in three separate sessions within 1wk. Subjects fasted for I 2 h before each experiment and were
encouraged to drink water during this period. Each experiment
began at 0800, with no additional water consumption until the
end ofthe study at 1 100. Subjects emptied their bladders at thebeginning ofthe study and the specific gravity ofurine was mea-
sured in this sample. Subjects were then weighed on an electronic
I From the Occupational Physiology Division, US Army Research
Institute of Environmental Medicine, Natick, MA.2 Presented in part at the Fifth Conference for Federally Supported
Human Nutrition Research Units and Centers, sponsored by the Inter-agency Committee on Human Nutrition Research, NIH, Bethesda, MD,February2O-2l, 1991.
3 The opinions or assertions contained herein are the private views of
the authors and are not to be construed as official or,as reflecting the
views of the Department ofthe Army or the Department of Defense.4 Address reprint requests to KE Fried!, Occupational Physiology Di-
vision, US Army Research Institute of Environmental Medicine, Natick,
MA 01760-5007.
Received July 19, 1991.Accepted for publication September 25, 1991.
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
RELIABILITY OF BODY-FAT ESTIMATIONS 765
TABLE 1Subject characteristics; subjects are ordered by increasing density of
fat-free mass (FFM)*
Smoking
Subject Sex Age status Ethnicity Height BMI
y cm
1 M 19 NS NHW 174 23.42 M 19 FS NHW 174 21.73 M 20 SMK NHW 182 23.84 M 24 FS A/PI 164 27.4
5 M 22 NS NHW 171 23.4
6 F 23 FS NHW 174 23.4
7 M 21 SMK NHW 190 26.88 M 19 NS HISP 168 23.9
9 M 24 NS NHW 182 22.810 M 19 SMK NHW 187 24.4
i±SD 21±2 - - 176.6 ±8.4 24.1 ± 1.7
S NS, nonsmoker; ES, former smoker; SMK, current smoker, NHW,
non-Hispanic white; A/PI, Asian/Pacific Islander; H1SP, Hispanic; BMI,
body mass index (in kg/m2).
balance (model 708, Seca AG, Hamburg, Germany) while wear-
ing light-weight shorts or a swimsuit and their heights were mea-sured with a stadiometer. Volumes micturated during the cx-periment (all occurred after the first 90 mm but before UWW)were measured and deducted from body-water and body-weight
measurements. A baseline venous blood sample was obtained
and subjects were given a fixed dose of 30 g D20 (99.9% purity,
Cambridge Isotope Laboratory, Woburn, MA), which was con-sumed through a straw and was followed by two 30-mL rinsesof tap water. Each subject was then measured by DEXA, bio-electrical impedance, and UWW, in that sequence. Three hours
after the initial dose, at the end of all other testing, a second(deuterium-equilibrated) blood sample was obtained. Serum wasstored in airtight containers at - 135 #{176}Cfor � 3 mo, until deu-terium analysis.
Dual-energy x-ray absorptiometry
TBBM measurements were made by scanning subjects with
DEXA (DPX, Lunar Co, Madison, WI) in the 20-mm (medium
speed) scan mode (6). Bone-mineral-content measurements bythis device were calibrated daily against secondary standardsthat were compared with ashed-bone sections. TBBM was cal-
culated from the measured values obtained by DEXA by usingthe value I .000 g of bone mineral (2.982 g/cm3) yields 0.9582g ash (3). The software in this device also yields an estimate ofpercent body fat based on an extrapolation of fatness from theratio of soft tissue attenuation of two x-ray energies in pixelsnot containing bone. The basis of this body-fat algorithm was
previously described (6, 8).
Bioelectrical impedance
Total body resistance was measured by electrical impedance
plethysmography. Two electrodes were placed on the dorsal sur-
face ofthe right hand and two on the right foot, with 800 jzA at
50 kHz passed between the outer two electrodes (model BIA-
10 1 , RJL Systems, Detroit). Resistance measurements were ob-
tamed within 2 mm of placing subjects in a supine position.
TBW was estimated by using the equation of Kushner and
Schoeller (9).
Underwater weighing
UWW was performed after all other measurements. Subjects
were weighed in 1.0 m water in a 0.9 X 1.2 X 1.2 m aluminum
tank after residual volume determinations were made. Residualvolumes were measured by oxygen dilution by using the method
of Wilmore et al (10), with subjects in a bent-forward sitting
position; measurements were duplicated to within 50 mL. Sub-
jects were then submerged while in the same position while sittingon an aluminum chair suspended from an electronic load cell(model 6000, Ametek, Largo, FL), sensitive to 10 g; tare weightof the chair was established with the subject in the tank. The
load cell voltage was analyzed by a weight-calibrated analog-to-
digital converter and the weights were sampled at a rate of 4
samplings/s by computer (HP-85, Hewlett-Packard, Corvallis,OR), which computed an underwater weight of the subject av-
eraged over 4-7 s after complete exhalation. Each subject was
strongly encouraged to produce a complete exhalation during
each residual-volume measurement and during each of 10 UWWtrials. The density of a subject was calculated by using the twohighest underwater weights within 100 g of each other, of ten
trials; water temperature was corrected for (usually 37 #{176}C).Gas-trointestinal gas volume was assumed to be negligible becausethe subjects were fasted and no fixed estimate was deducted inthe density calculation.
Total body water
TBW volume was calculated from serum concentrations of
deuterium. Deuterium was equilibrated between equal volumes(1.5 mL) of serum and deionized water at 37 #{176}Cfor 48 h in
8.3-cm diameter, Conway diffusion dishes (Bel-Air Products,
Pequannock, NJ) tightly sealed with Parafilm (American Can
Company, Greenwich, CT) ( I 1). These volumes were found tobe within the optimum range of recoveries in a test of all com-
binations ofO.5-mL increments between 1.0 and 2.5 mL. Each
baseline and 3-h postdosing serum sample was purified in trip-
licate samples. Recoveries by this method averaged 98% of initial
concentrations (ranging from 95% to 102%) and the methodyielded the same results obtained after vacuum sublimation pu-
rification in this laboratory (JP DeLuca, unpublished observa-
tions, 1991).
Absorption was measured by using a single-beam, infra-red spectrophotometer with a 4.0-tim fixed filter (Miran 1FF,Foxboro, Foxboro, MA) and a digital display device (Applied
Measurement, Acton, MA) in a cold ( 1 5 #{176}C)room. Samples
were measured in a calcium fluoride cell with a path length of
0.2 mm and deuterium oxide concentrations were calculated
from a linear standard curve. Measured concentrations rangedfrom 0.5 to 0.8 g/L after a single deuterium dose, with a 1.0%intraassay CV (for three replicates), equivalent to �0.5 L of
TBW. Measurements were calculated as liters ofTBW and this
was converted to mass by using 0.9937 g/mL. An additional
correction of 2% was used to account for the overestimate of
TBW due to proton exchange (12, 13), which is known to occur
between the tracer and proteins and carbohydrates in the bodywithin the first few hours of equilibration (14).
Percent-body-fat calculations
Percent-body-fat calculations were made as follows. Two-
compartment models, from density (2):
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
%BFuww [4.95/density - 4.5] 100
766 FRIEDL El AL
TABLE 2Principal measurements and characteristics of FFM calculated from the four-compartment model
Principal measurements Four-compart ment-model measurements
SubjectBody
weight Body density TBBM TBW�� TBW�
Residual
volume FFM density
TBBM/FFM
TBW�/FFM
kg g/cm3 kg L L L g/cm3 % %
I #{149}
2
3
45
67
89
10
70.2 ± 0.9
65.5 ± 0.2
79.4 ± 0.4
73.7 ± 0.268.3 ± 0.2
70.6 ± 0.496.3 ± 0.3
68.1 ±0.675.2 ± 0.4
84.9±0.4
1.077 ± 0.006
1.077 ± 0.0011.058 ± 0.002
1.039 ± 0.0011.063 ± 0.002
1.037 ± 0.0031.051 ± 0.002
1.062±0.0021.063 ± 0.0011.069±0.001
2.87 ± 0.01
3.04 ± 0.013.18 ± 0.01
2.71 ± 0.05
2.98 ± 0.02
3.09 ± 0.024.45 ± 0.09
3.07±0.013.23 ± 0.033.96±0.02
48.7 ± 0.5
44.2 ± 0.648.4 ± 0.9
39.4 ± 1.2
41.7 ± 0.3
36.6 ± 0.353.7 ± 1.539.9± 1.443.9 ± 1.850.7±0.1
47.0 ± 0.9
43.4 ± 0.650.6 ± 0.2
41.1 ± 0.3
42.7 ± 0.3
36.8 ± 0.153.7 ± 1.5
42.8±0.448.3 ± 1.052.5±0.5
1.44 ± 0.16
0.81 ± 0.020.95 ± 0.050.96 ± 0.06
1.31 ± 0.040.95 ± 0.041.65 ± 0.040.91 ±0.031.68 ± 0.061.06±0.05
1.096 ± 0.0031.098 ± 0.002
1.098 ± 0.0011.101 ± 0.0031.102 ± 0.002
1.106 ± 0.0021.107 ± 0.004
1.108±0.0051.108 ± 0.005
1.110±0.001
5.62 ± 0.026.55 ± 0.086.20 ± 0.126.45 ± 0.086.67 ± 0.017.92 ± 0.057.71 ± 0.25
7.09±0.086.76 ± 0.21
7.14±0.03
74.2 ± 0.973.8 ± 0.473.3 ± 0.172.8 ± 1.072.7 ± 0.572.9 ± 0.772.4 ± 1.171.6± 1.471.1 ± 1.3
71.1±0.1
i±SDWithin-
subject
SD
75.2±9.4
0.4
1.060±0.014
0.002
3.26±0.53
0.04
44.7±5.6
1.0
45.9±5.4
0.7
1.17±0.32
0.06
1.103±0.006
0.003
6.81 ±0.67
0.12
72.6± 1.3
0.9
and from TBW (15):
%BFTBw [(body weight - TBW/0.732)/body weight]lOO
Three-compartment model, from density and TBW (2):
%BFUWWTBW [2. 1 1 8/density
- 0.780(IBW/weight) - 1.354] 100Four-compartment model, from density, TBW mass, and
IBBM:
%BFUWWTBwTBBM = [2.559/density - 0.734(IBW/weight)
+ 0.983(IBBM/weight) - l.841J100
This four-compartment model was derived by using essen-tially the same density assumptions used by Selinger (16) andHeymsfield et al (7), with component densities of 0.9007 (fat),0.9937 (water), and 2.982 (bone mineral) from the original es-
timations of Brozek et al (3) but with the remainder (largelyprotein, nonosseous mineral, and glycogen) based on the actualdensity measurements for the residual mass made by Allen etal (17): 1 .39 g/cm3. The results obtained by our equation and
the calculations suggested by Heymsfieid et al (7) were very
similar.
Statistics
Data were analyzed by using reliability procedures, t test, and
analysis of variance (ANOVA) with SPSS-X statistical software
(SPSS, Inc, Chicago). Repeatability of the measurements and
fat estimations was expressed by within-subjects standard de-viations and by the standardized Cronbach’s alpha reliability
statistic.
Results
Means and group standard deviations for each ofthe principal
measurements in this experiment are shown in Table 2. Thereliability coefficients for these measurements were 0.999 forweight and TBBM, 0.992 for density and residual volume, and0.989 for TBW�. Within-subjects standard deviations were0.4 kg for body weight, 0.002 g/cm3 for body density, 0.06 L forresidual volumes, 0.04 kg for TBBM, 1.0 L for TBW�, and0.7 L for TBWBIA.
A learning effect for the underwater exhalation procedure was
evident for halfofthe subjects and this also produced an overalltrend to a reduction in estimated %BF��, between the threeexperiments, with average estimates increased by 0.002 g/cm3
or reduced by � 1 %BF from the first to the third experiment;there was no comparable change in the residual-volume mea-surements or body weights. The within-subjects standard devia-tion for UWW encompassed this between-experiments learningdecrement: ±0.002 g/cm3, or 1.0 %BF.
Body-water measurements by deuterium dilution varied
within subjects between trials by ± 1.0 L (range ofactual within-
subjects differences: 0.2-3.4 L) (Table 2). As a fraction of thefat-free mass (FFM) (determined from the four-compartment
model), TBW averaged 72.6% but individual means varied sig-nificantly (ANOVA, P < 0.003), ranging from 7 1 . 1 ± 0.01% for
a subject with consistently poor hydration (average urine speci-fic gravity = 1.031 ± 0.002), to 74.1 ± 0.9% (urine specificgravity = 1.021 ± 0.006). This is a significant difference betweennormal individuals because the difference between these cx-tremes was larger than the within-individual variations (±0.9%TBW/FFM) and the precision of the deuterium purificationand analysis (±0.5 L).
Body-water estimations from the Kushner and Schoeller
equation (10), in which height, weight, and resistance are used,were very close approximations ofthe deuterium-measured vol-
umes (r = 0.92; TBW�o TBWsIA average = - 1.2 ± 2.1 L)
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
RELIABILITY OF BODY-FAT ESTIMATIONS 767
* Within-subjects standard deviation.
TABLE 3Percent body fat estimated by various methods in this study
Three-compartment Four-compartmentSubject DEXA TBW UWW method method
%
1 8.7 ± 0.3 5.8 ± 0.2 9.6 ± 2.4 7.4 ± 1.1 7.1 ± 1.3
2 9.3 ± 0.7 8.4 ± 1.1 9.8 ± 0.1 9.0 ± 0.7 9.2 ± 0.6
3 17.6 ± 0.4 17.2 ± 1.1 17.9 ± 0.9 17.6 ± 1.0 17.4 ± 1.1
4 25.7 ± 0.6 27.4 ± 2.2 26.6 ± 0.4 27.1 ± 1.4 27.1 ± 1.3
5 16.0 ± 0.2 17.0 ± 0.5 15.6 ± 0.8 16.4 ± 0.3 16.5 ± 0.36 31.5 ± 0.5 29.6 ± 0.1 27.3 ± 1.2 28.6 ± 0.5 29.3 ± 0.67 24.4 ± 0.9 24.3 ± 2.1 21.0 ± 0.8 22.9 ± 1.2 23.4 ± 1.2
8 18.3 ± 0.2 20.5 ± 2.9 16.1 ± 0.8 18.7 ± 1.6 18.8 ± 1.59 15.4 ± 0.4 20.8 ± 2.9 15.8 ± 0.4 18.7 ± 1.5 18.6 ± 1.4
10 13.1 ± 0.4 18.9 ± 0.6 13.0 ± 0.6 16.4 ± 0.6 16.5 ± 0.6
i±SD* 18.0±0.5 19.0± 1.7 17.3± 1.0 18.3± 1.1 18.4± 1.1Cronbach’s ‘� 0.999 0.986 0.99 1 0.994 0.994
S Within-subjects standard deviation.
and yielded a greater precision (±0.7 L) than the TBW values parison with the measurement by the four-compartment modelfrom direct measurement by deuterium (± 1.0 L). is shown in Figure 2. Four-compartment corrected values showed
The measurement ofTBBM varied within subjects by ±40 g, consistent differences (ie, adjustments in the same direction inwith the largest individual difference shown as ±90 g. As a per- three experiments) for subjects where the mean difference wascentage ofthe FFM determined by the four-compartment model, > � 1 kg fat.
TBBM ranged from 5.6% to 7.9% of FFM (Table 2) and this The DEXA soft tissue analysis, which did not involve UWW,did not vary appreciably between trials within subjects. reduced the precision error by half. Estimates from TBW (two-
Percent body fat estimated by two-, three-, and four-com- compartment model) produced the highest variability in the
partment models is shown in Table 3. Reliabilities were slightly measurements.
higher with correction for day-to-day variation in TBW in thethree-compartment model (0.994) than for UWW alone (0.991);the four-compartment model did not further increase this reli- Discussionability with correction for TBBM. Fat mass from the variousmodels gave the same pattern of reliabilities reflected in the ret- The data from this study indicate that the multicompartmentative fat estimations (Table 4). The individual day-to-day vari- model, which includes measurements ofTBW and TBBM, can
ations in the difference of the fat weight measured by UWW be used to improve the accuracy ofbody-fat measurement fromand the two-compartment model is shown in Figure 1; its corn- UWW without being invalidated by the sum of errors from the
TABLE 4Fat mass estimated by various methods reported in this study
Three-compartment Four-compartment
Subject DEXA TBW UWW method method
kg
1 6.1 ± 0.2 4.1 ± 0.2 6.7 ± 1.8 5.2 ± 0.9 5.0 ± 1.0
2 6.1 ± 0.4 5.5 ± 0.7 6.4 ± 0.1 5.9 ± 0.4 6.0 ± 0.43 14.0 ± 0.4 13.7 ± 0.8 14.2 ± 0.7 13.9 ± 0.7 13.8 ± 0.8
4 18.9 ± 0.4 20.3 ± 1.6 19.6 ± 0.3 20.0 ± 1.0 20.0 ± 1.05 10.9 ± 0.1 1 1.6 ± 0.3 10.6 ± 0.6 1 1.2 ± 0.2 1 1.2 ± 0.2
6 22.3 ± 0.2 20.9 ± 0.1 19.3 ± 0.9 20.2 ± 0.4 20.7 ± 0.5
7 23.5 ± 0.8 23.4 ± 2.1 20.2 ± 0.7 22.1 ± 1.2 22.6 ± 1.18 12.4 ± 0.2 14.0 ± 2.0 1 1.0 ± 0.6 12.7 ± 1.2 12.8 ± 1.1
9 11.5±0.3 15.7±2.1 11.9±0.4 14.1±1.1 14.0±1.110 1 1.1 ± 0.4 16.0 ± 0.6 1 1.1 ± 0.6 13.9 ± 0.6 14.0 ± 0.6
.�±SD 13.7±0.4 14.5± 1.3 13.1 ±0.8 13.9±0.8 14.0±0.8Cronbach’s a 0.999 0.989 0.992 0.995 0.995
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
1 2 3 4 5 6
Subject
7 8 9 10
Difference in fat weight (kg)
.3-i -�.J -
FFM d.nafty (k�t):
1.056 1.OQS 1.008 1.101 1.102 1.106 1.107 1.106 1.108 1.110
1 2 3 4 5 6
768 FRIEDL ET AL
FIG 1. Fat weight estimated from density and the two-compartment
equation of Siri in 10 healthy young subjects whose weight was studiedin three separate experiments within I wk.
multiple measurements. The largest error in this model appears
to come from the procedure ofUWW itself. UWW was no morereliable than ± 1.0% body fat or ±0.002 g/cm3, as previouslyreported (18-2 1). The precision of this measurement is largely
dependent on how well gas spaces are accounted for, which is
affected by methodological differences in the method of residual-
volume measurement (22), biological variations in intestinal gas
(23, 24), and the ability of the subjects to consistently match
their exhalation performances from the residual-volume mea-
surement to their efforts underwater. The method of trial selec-
tion also varies from study to study and, in this study, averaging
trials in a fixed sequence (eg, first 6 or last 3 of the 10 trials)
instead ofselecting best trials produced an additional difference
in density of -0.003 ± 0.002 g/cm3 (P < 0.001) or > + 1% body
fat. We believe that, because ofthe greater difficulty encounteredby subjects in exhaling to residual volume underwater, the best-
trials approach is the more accurate reflection of the residual-volume measurements used in the calculation of density. This
difference between in-water and out-of-water exhalation perfor-
mance is illustrated by the learning effect between experiments
for half of the subjects in this study, whereas residual-volume
measurements remained unchanged.
Siri (2) estimated that the standard deviation in fat estimated
by UWW was ±4.6% body-fat units and, with no errors in mea-
suring density, the error would still be ±3.8% body-fat units
“because of the normal variability in body constituents.” He
estimated the degree of hydration to be the next largest error in
fat estimation, with variations of ±6.5% of TBW in normally
hydrated individuals; thus, with the combination of UWW andTBW, Siri suggested that the estimate could be reduced to ± 1.5%
body-fat units ifthe error in body-water measurement could be
reduced to ± 1% of body weight. The error of the technique we
used is slightly better than ± 1% of body weight.
Adjustments for hydration produced sizeable changes in thefat estimates for some of our subjects and these adjustments
appeared to be more important than corrections for fractional
bone mass. This was indicated by the difference in fat-weight
means between those calculated from UWW and those fromthe three-compartment model of Siri, with little additional
change from the correction for TBBM (Table 4). Normal vari-
ations in the TBW of our 10 subjects ranged between 71% and
74% of FFM. These ranges are significant departures from the
assumed value of 73.2%, decreasing the fat estimate by UWW
by 2.2% body fat and increasing it by 3.4% body fat at the cx-
tremes of hydration (Table 3). Hydration, as a key determinant
of the departures from the assumed density of 1. 100 g/cm3 forthe FFM, progressively decreased throughout our subjects or-dered by increasing FFM density (Table 2).
The degree ofhydration also markedly affected the fat estimate
by the other two-compartment model in this study, based on
TBW. The differences between fat estimates from UWW andTBW in our 10 subjects ranged between +3.8% and -5.9% body-
fat units; however, the differences were smaller between TBWor UWW fat estimates and estimates made by three- and four-compartment models. Because the fractional bone mass also
tended to be higher in our subjects with lower degrees of meanhydration, this greater difference between fat estimates by thetwo two-compartment models suggests that TBW incorrectlyaccentuates differences when the fractional bone mass also de-parts from the 7% TBBM-FFM reference value. Thus, the two-
compartment model using TBW underestimates fatness whenthere is less bone mineral and overestimates fatness when thereis more than the assumed fraction of bone mineral, as a con-
sequence of the increased body mass, which is not consideredin the TBW-determined estimate of FFM. The error produced
by deviations in the assumed TBBM in fat estimation from
UWW is in the opposite direction. This may explain the puzzling
overestimate of the fractional bone mineral in black subjects
encountered by Schutte et al (25) when they tried to estimate
FFM density from a comparison of TBW and UWW fat esti-mates.
The 2% correction that we applied to the isotope-dilution
method is nothing more than a best guess at the magnitude ofthe exchange that is known to occur. This value was recom-mended as the upper limit by Pinson (12) but others estimateda higher theoretical maximum of 5.2% (26). Whatever the upperlimit is, equilibrium is reportedly not achieved until 24 h (27),
3
2
0
-1
-2
-4
Sublect
7 8 9 10
FIG 2. Difference in fat weight estimated from density by the two-compartment equation compared with the estimate from the four-com-partment equation, including assessments of TBW and TBBM. The re-
producibility of the four-compartment model is illustrated by the dif-ferences obtained in three separate experiments performed within 1 wk
for each of 10 subjects. Subjects are ordered by their mean FFM density(1. 100 is the assumed value in the two-compartment model).
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
RELIABILITY OF BODY-FAT ESTIMATIONS 769
and on this basis we chose not to apply the higher correction
factor to samples equilibrated in vivo for only 3 h.
Our estimates of TBW from electrical impedance were also
remarkably similar to our values for TBW determined by deu-
terium dilution and they actually yielded smaller day-to-day
variations than did our measurements by deuterium. This finding
suggests either greater precision by electrical impedance than by
the deuterium method or a decreased sensitivity to small differ-ences. These measurements would have been even closer (0.24
± 2.2 L; � ± SD) if we had made no adjustment for proton
exchange, just as no adjustment was made in the original deri-
vation of this equation (9). This suggests the possibility thatimpedance could become a useful substitute for estimations of
TBW, offering the convenience ofimmediate results. Estimationofbody fat by using electrical impedance currently suffers from
the same limitations encountered with fat estimation from TBW,
namely the variation in hydration. However, this problem may
be overcome with multifrequency impedance measurements (28)
iffat estimation is based on intracellular water instead of TBW,as suggested by Keys and Brozek (29).
The other significant error in the assumptions of the two-
compartment models is the effect of the bone. This error was
well recognized in the earliest presentation ofbody-density mea-surements, when Behnke et al (30) noted that “excess fat is
viewed as the prime factor governing the level ofspecific gravity,”
but “precise measurements ofthis excess fat will necessarily await
a knowledge of the relative percentage variation of the weight
of the skeleton in lean persons.” This variable component cannow be more reliably measured than either body density or TBWand the measurement of TBBM is important because it varieswith ethnicity (3 1), physical training (32), and even total weight
(33), although in our subjects TBBM was of lesser importancethan TBW. A 20% increase in the average TBBM of 6.8% ofFFM in this study would produce an approximate correction of
+0.7 kg offat, an error ofunderestimation that might be typically
observed when these assumptions are applied to the average
black subject by using UWW. Nevertheless, this technology still
remains to be validated when, for example, subject hydration
and subject size are varied; subject size may artifactually reducethe bone measure with increasing body thickness (34).
The four-compartment-model approach to percent-body-fat
estimation improves upon the two-compartment models of Siri
(2) and Brozek et al (3) in terms of accuracy, by accounting for
the bone mineral and water components, which are otherwise
assumed to be of fixed proportions when they can be quite vari-able. The largest remaining error, the precision error of ±0.002
g/cm3 measured in our three UWW sessions, coupled with theerror in the TBW-measurement technique (±0.5 L), makes the
four-compartment model estimates offat no better than ± 1 per-
cent, but probably the accuracy is within Sin’s goal of 1.5%.
Further improvements in precision will probably only occur
when a technique that is easier for subjects to perform is adopted.
Thus, passive methods that do not require subjects to forciblyexhale underwater (eg, DEXA soft tissue analyses) may be pref-erable because of their superior reproducibility, particularly if
they are found to be accurate when compared with the four-
compartment model. 0
We thank Sherryl Kubel, Maijorie Harp, Sonja Moore, and Robert
Mello for their care and expert technical assistance in data collection.We are grateful to Henry Lukaski (USDA, Grand Forks, ND) for his
timely suggestions that led to this line of investigation and to John F
Patton for his helpful suggestions concerning this manuscript.
References
1. Allen TH, Welch BE, Trujillo TT, Roberts JE. Fat, water and tissue
solids ofthe whole body less its bone mineral. J Appl Physiol 1959;14:1009- 12.
2. Siri WE. Body composition from fluid spaces and density: analysisofmethods. In: Brozek J, Henschel A, eds. Techniques for measuring
body composition. Washington, DC: National Academy of Sciences�1961:223-44.
3. Brozek J, Grande F, Anderson JT, Keys A. Densitometric analysis
of body composition: revision of some quantitative assumptions.
Ann NY Acad Sci l963;l 10:113-40.4. Lukaski HC, Johnson PE. A simple, inexpensive method of deter-
mining total body water using a tracer dose of D2O and infraredabsorption ofbiological fluids. Am J Gin Nutr l985;41:363-70.
5. Heymsfield SB, Wang J, Heshka 5, Kehayias JJ, Pierson RN. Dual-photon absorptiometry: comparison ofbone mineral and soft tissue
mass measurements in vivo with established methods. Am J Gin
Nutr l989;49: 1283-9.6. Mazess RB, Barden HS, Bisek JP, Hanson J. Dual-energy x-ray ab-
sorptiometry for total-body and regional bone-mineral and soft-tissue
composition. Am J Gin Nutr l990;5l:l 106-12.7. Heymsfield SB, Lichtman 5, Baumgartner RN, et a1. Body com-
position ofhumans: comparison oftwo improved four-compartmentmodels that differ in expense, technical complexity, and radiation
exposure. Am J Clin Nutr l990;52:52-8.8. Mazess RB, Peppler WW, Gibbons M. Total body composition by
dual-photon (‘35Gd) absorptiometry. Am J Gin Nutr l984;40:
834-9.9. Kushner RR, Schoeller DA. Estimation oftotal body water by bio-
electrical impedance analysis. Am J Clin Nutr l986;44:4l7-24.10. Wilmore JH, Vodak PA, Parr RB, Girandola RN, Billing JE. Further
simplification of a method for determination of residual volume.Med Sci Sports Exerc 1980; 12:216-8.
1 1. Davis JM, Lamb DR, Burgess WA, Bartoli WP. Accumulation of
deuterium oxide in body fluids after ingestion of D20-labeled bev-
erages. J AppI Physiol l987;63:2060-6.12. Pinson EA. Water exchanges and barriers as studied by the use of
hydrogen isotopes. Physiol Rev l952;32: 123-34.13. Sheng HP, Huggins RA. A review ofbody composition studies with
emphasis on total body water and fat. Am J Clin Nutr l979;32:
630-47.14. Schloerb PR, Friis-Hansen BJ, Edelman IS, Sheldon DB, Moore
FD. The measurement of deuterium oxide in body fluids by the
falling drop method. J Lab Clin Med 195 l;37:652-62.15. Pace N, Rathbun EN. Studies on body composition. III. The body
water and chemically combined nitrogen content in relation to fat
content. J Biol Chem l945;158:685-9l.
16. Selinger A. The body as a three component system. Champaign, IL:University of Illinois at Urbana-Champaign, 1977.
17. Allen TH, Krzywicki HJ, Roberts JE. Density, fat, water and solidsin freshly isolated tissues. J Appl Physiol 1959;l4:l005-8.
18. Katch Fl. Apparent body density and variability during underwaterweighing. Res Q 1968;39:993-9.
19. Welch BE, Crisp CE. Effect ofthe level ofexpiration on body densitymeasurement. J Appl Physiol l958;l2:399-402.
20. Durnin JVGA, Taylor A. Replicability of measurements of densityof the human body as determined by underwater weighing. J AppiPhysiol 1960; I 5:142-4.
21. Mendez J, Lukaski HC. Variability of body density in ambulatorysubjects measured at different days. Am J Gin Nutr l98l;34:78-81.
22. Forsyth R, Plyley MJ, Shephard Ri. Residual volume as a tool inbody fat prediction. Ann Nutr Metab l988;32:62-7.
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from
770 FRIEDL ET AL
23. Bedell ON, Marshall R, DuBois AB, Harris JH. Measurement of
the volume of gas in the gastrointestinal tract. Values in normalsubjects and ambulatory patients. J Gin Invest 1956;35:336-45.
24. Durnin JVGA, Satwanti. Variations in the assessment of the fatcontent of the human body due to experimental technique in mea-
suring body density. Ann Hum Biol 1982;9:22l-5.25. Schutte JE, Townsend EJ, HuggJ, Shoup RE, Malina RM, Blomqvist
CG. Density oflean body mass is greater in blacks than in whites.
J Appl Physiol l984;56: 1647-9.26. Culebras JM, Moore FD. Total body water and the exchangeable
hydrogen. I. Theoretical calculation of nonaqueous exchangeable
hydrogen in man. Am J Physiol l977;232:R54-9.27. Hallaway BE, Benson ES. Hydrogen exchange in proteins. Equilib-
rium and kinetic isotope effects in bovine plasma albumin studiedwith tritium and deuterium. Biochim Biophys Acta 197 l;243:
380-7.28. Van Loan MD, Mayclin PL. Use of multi-frequency bioelectrical
impedance analysis for the estimation of extracellular fluid. Med
Sci Sports Exerc 1991 ;23:S7 1 (abstr).
29. Keys A, Brozek J. Body fat in adult man. Physiol Rev l953;33:245-
325.
30. Behnke AR Jr, Feen FG, Welham WC. The specific gravity of healthymen: body weight:volume as an index ofobesity. JAMA 1942;l 18:495-8.
31. Cohn SH, Abesamis C, Yasumura 5, Aloia JF, Zanzi I, Ellis KJ.
Comparative skeletal mass and radial bone mineral content in blackand white women. Metabolism l977;26:17l-8.
32. Dalsky GP, Stocke KS, Ehsani AA, Slatopolsky E, Lee WC, BirgeSJ. Weight-bearing exercise training and lumbar bone mineral con-tent in postmenopausal women. Ann Intern Med l988;l08:824-8.
33. Mazess RB, Barden HS, Drinka PJ, Bauwens SF, Orwoll ES, BellNH. Influence of age and body weight on spine and femur bone
mineral density in US white men. J Bone Miner Res l990;5:645-52.
34. Buhl K, Heymsfield 5, Russell-Aulet M, Wang J, Pierson RN Jr,Lichtman S. Effect of tissue thickness on bone density and bone
mineral by dual energy x-ray absorptiometry. FASEB J l963;5:A924(abstr).
by on March 27, 2006
ww
w.ajcn.org
Dow
nloaded from