European Journal of Clinical Nutrition (2001) 55, 145±152
ß 2001 Nature Publishing Group All rights reserved 0954±3007/01 $15.00
www.nature.com/ejcn
Predicting the resting metabolic rate of young Australian males
GE van der Ploeg1, SM Gunn1, RT Withers1*, AC Modra1, JP Keeves1 and BE Chatterton2
1
Exercise Physiology Laboratory, School of Education, Flinders University, Adelaide, South Australia, Australia; and 2Department of
Nuclear Medicine, Royal Adelaide Hospital, Adelaide, South Australia, Australia
Objectives: The aims of this study were: (a) to generate regression equations for predicting the resting metabolic
rate (RMR) of 18 to 30-y-old Australian males from age, height, mass and fat-free mass (FFM); and (b) crossvalidate RMR prediction equations, which are frequently used in Australia, against our measured and predicted
values.
Design: A power analysis demonstrated that 38 subjects would enable us to detect (a ˆ 0.05, power ˆ 0.80)
statistically and physiologically signi®cant differences of 8% between our predicted=measured RMRs and those
predicted from the equations of other investigators.
Subjects: Thirty-eight males (X s.d.: 24.3 3.3 y; 85.04 13.82 kg; 180.6 8.3 cm) were recruited from
advertisements placed in a university newsletter and on community centre noticeboards.
Interventions: The following measurements were conducted: skinfold thicknesses, RMR using open circuit
indirect calorimetry and FFM via a four-compartment (fat mass, total body water, bone mineral mass and residual)
body composition model.
Results: A multiple regression equation using the easily measured predictors of mass, height and age correlated
0.841 with RMR and the SEE was 521 kJ=day. Inclusion of FFM as a predictor increased both the R and the
precision of prediction, but there was virtually no difference between FFM via the four-compartment model
(R ˆ 0.893, SEE ˆ 433 kJ=day) and that predicted from skinfold thicknesses (R ˆ 0.886, SEE ˆ 440 kJ=day). The
regression equations of Harris & Benedict (1919) and Scho®eld (1985) all overestimated the mean RMR of our
subjects by 518 ± 600 kJ=day (P < 0.001) and these errors were relatively constant across the range of measured
RMR. The equations of Hayter & Henry (1994) and Piers et al (1997) only produced physiologically signi®cant
errors at the lower end of our range of measurement.
Conclusions: Equations need to be generated from a large database for the prediction of the RMR of 18 to 30-yold Australian males and FFM estimated from the regression of the sum of skinfold thicknesses on FFM via the
four compartment body composition model needs to be further explored as an expedient RMR predictor.
Sponsorship: Australian Research Council (small grants scheme).
Descriptors: four-compartment body composition model; hydrodensitometry; isotopic dilution; DXA
European Journal of Clinical Nutrition (2001) 55, 145±152
Introduction
The resting metabolic rate (RMR) represents 6300 kJ=day
or 60 ± 75% of the total daily energy expenditure for a 70 kg
person (Poehlman, 1989). It is therefore by far the largest
component of the 24 h energy expenditure compared with
those for the thermic effects of activity and feeding at 15 ±
30% and 10%, respectively (Poehlman, 1989). The RMR
*Correspondence: Dr RT Withers, Exercise Physiology Laboratory, School
of Education, Flinders University, GPO Box 2100, Adelaide, South
Australia 5001, Australia.
E-mail: bob.withers@¯inders.edu.au
Guarantor: GE van der Ploeg.
Contributors: GEP, SMG and ACM recruited the subjects, collected and
analysed the data and helped to write the paper; RTW conceived the study,
secured the research funding, supervised the project and wrote the paper;
JPK helped to write the paper; BEC supervised the DXA scans.
Received 16 June 2000; revised 9 October 2000; accepted 16 October 2000
consequently has a large impact on the regulation of body
mass and energy balance.
The FAO=WHO=UNU (1985) have advocated a method
for estimating the approximate energy intake requirements
of healthy adults. This method is based on the premise that
all estimates of energy requirements should be derived from
measurements of energy expenditure rather than energy
intake. The measured RMR is then multiplied by an activity
level, which is determined using the factorial method, to
yield a total energy requirement. However, it is not always
possible to measure RMR, which must therefore be predicted from such variables as age and mass. Furthermore, if
currently used equations overpredict RMR, then this error
will be compounded when it is multiplied by the activity
level. Hence, valid equations for predicting the RMR of
Australians would assist in the regulation of body mass
via the estimation of energy intake requirements. Such
Resting metabolic rate of Australian males
GE van der Ploeg et al
146
equations would also have implications for the treatment of
obesity, which is a major health hazard. Furthermore, the
accurate prediction of RMR is particularly important in the
clinical setting because it re¯ects almost all the energy
requirements of bedridden hospitalized patients.
The two most frequently used RMR prediction equations in
Australia are those of Harris & Benedict (1919) and Scho®eld
(1985). The former were generated on 239 adult American
males and females and published in 1919. The Scho®eld
equations were produced via pooling the data from 114
studies, 76 (66.7%) of which were published more than 50 y
ago. Most of the data on which the current RMR prediction
equations are based were therefore collected in the early part of
the last century. Since then puberty has occurred earlier and the
height and mass of many populations have tended to increase.
These changes pose questions concerning the composition of
this mass increase and whether it in¯uences the prediction of
RMR (Elia, 1992a). A further problem is that there are large
physiologically signi®cant differences of 13 ± 47% between
the RMRs predicted by frequently used equations (Elia,
1992a). While some of these disparities will be due to
biological differences, it is dif®cult to evaluate retrospectively
the accuracy and precision of the RMR methodology used by
many workers since their papers contain no such details.
The RMR of Australians was studied many years ago
(Hick et al, 1931; Wardlaw et al, 1934; Wardlaw & Horsley,
1928; Wardlaw & Lawrence, 1932) and this information is
included in the Scho®eld (1985) database. More recently,
Piers et al, (1997) demonstrated that the Scho®eld (1985)
equations signi®cantly overpredict (P < 0.001) the RMR of
18 to 30-y-old Australian males and females. The present
investigation extends the work of Piers et al (1997) by using
fat-free mass (FFM) as a predictor of RMR. The FFM is a
more valid predictor of RMR than body mass because it has
a much higher rate of resting energy expenditure (Elia,
1992b: 124.3 kJ kg71 day71) than adipose tissue (Elia,
1992b: 18.8 kJ kg71 day71). Hence, while adipose tissue
comprises signi®cant percentages of the body mass for
reference man (21.4%) and woman (32.8%), its contributions to their RMRs are only 4 and 6%, respectively (Elia,
1992b). Previous studies which used FFM as an RMR
predictor estimated FFM via either anthropometric measurements or two-compartment body composition models with
hydrodensitometry or underwater weighing being the most
popular (Elia, 1992a). However, the two-compartment
hydrodensitometric model (FFM and fat mass or FM) can
overestimate the FFM by up to 6% (Withers et al, 1998)
because it does not control for biological variability in both
the total body and bone mineral. We will therefore calculate
the FFM via a four-compartment criterion model (total body
water; bone mineral; FM; residual) of body composition
analysis which controls for interindividual variation in total
body water and bone mineral. However, this methodology is
time consuming and requires both expensive equipment and
considerable tester expertise. We accordingly propose to use
the sum of skinfold thicknesses plus other anthropometric
measurements as expedient and inexpensive measures of the
subcutaneous adipose tissue. The percentage body fat
European Journal of Clinical Nutrition
(%BF) will then be predicted from the regression of the
criterion of %BF via the four-compartment body composition model on the best weighted combination of anthropometric measurements, which have been largely ignored as
RMR predictors, thereby enabling the FFM to be calculated
by subtraction (FFM ˆ body mass 7 FM).
It is therefore proposed to test Australian males and
generate regression equations to predict RMR from age,
height, mass and FFM. The RMRs predicted by the equations of Harris & Benedit (1919), Scho®eld (1985), Hayter
& Henry (1994) and Piers et al (1997) will then be crossvalidated against our measured and predicted values.
Methods
Subjects
Thirty-eight males aged between 18 ± 30 y were recruited for
this pilot project (X s.d.: 24.3 3.3 y; 85.04 13.82 kg;
180.6 8.3 cm). This sample size enabled us to detect
(a ˆ 0.05; power ˆ 0.80) statistically and physiologically
signi®cant differences of 8% between our predicted=
measured RMRs and those predicted by the equations of
other investigators. Equal numbers of subjects were selected
across 3 y age bands with an attempt to recruit short,
medium and tall and light, medium and heavy persons
within each category. This allowed us to calculate more
accurately the regression surfaces than with a random
sample where the regression coef®cients for these two
predictor variables might be unduly in¯uenced by the few
cases at the extremes.
The sample was screened to exclude subjects who were:
smokers, not mass stable ( 2.0 kg) over the last year and
suffering from diseases or taking any medication which are
known to affect energy metabolism. Those with a history of
any clinical eating disorder were also excluded. This project
was approved by the Flinders Medical Centre's Committee
on Clinical Investigation. The aims, test protocols, possible
bene®ts and risks were explained to the subjects before they
gave their written consent to participate in accordance with
the established protocol for human subjects.
RMR
_ 2) was measured via open circuit
Oxygen consumption (VO
indirect calorimetry using the classical Douglas bag method
and Geppert & Zuntz (1888) transformation. All measurements were based on two 10 min collection periods. They
were conducted after 50 min of bedrest while the subject
was in the supine position with the head and shoulders
slightly elevated, breathing through a Hans Rudolph R2600
respiratory valve and wearing a noseclip. The subject was
always covered by a blanket and the temperature in his
vicinity was maintained at 24.0 0.5 C. All subjects were
habituated to this experimental situation on a previous day
to ensure that a true baseline was attained. Precautions (eg
phone off the hook and only the subject together with the
two experimenters allowed in the laboratory) were taken to
eliminate disturbing in¯uences that can affect the RMR. The
Resting metabolic rate of Australian males
GE van der Ploeg et al
subjects were also requested to adhere to the following
routine prior to the experimental trials in order to control for
the factors known to affect the RMR:
(1) no vigorous exercise during the preceding 36 h;
(2) no caffeine, alcohol and drugs during the preceding
12 h;
(3) consume a standardized evening meal between 19 30
and 20 00 h on the day before the test with only water to
be consumed afterwards; and
(4) be transported to the laboratory by car by the experimenters to eliminate uncontrolled activity.
Compliance with some of the preceding criteria was
checked by noting resting heart rates, which were monitored
continuously, and respiratory exchange ratios (RER).
The CO2 and O2 concentrations of dried mixed expirate
were monitored by Beckman LB-2 and Electrochemistry S3A analysers, respectively, which were calibrated throughout the physiological range of measurement for mixed
expirate using gases that had been veri®ed by LloydHaldane analyses. The volume of the expirate, which was
collected in 150 l Douglas bags, was then determined using
a calibrated 350 l Tissot spirometer (Hart & Withers, 1996).
This instrument had been checked for constant crosssectional area and counterbalancing throughout its elevation and allowance was made for the volume pumped
through the gas analysers. Energy expenditure (kJ) was
_ 2 data in accordance with
calculated from the RER and VO
the recommendations of Elia & Livesey (1992). The most
recent data on the reliability and precision of our calorimetry system yielded an intraclass correlation coef®cient
(ICC), technical error of measurement (TEM) and %TEM
of 0.989, 119 kJ=day and 1.6%, respectively, for repeated
trials on 10 subjects.
Body composition
All tests were conducted on the same morning as the RMR
trials to minimize within-subject biological variability. Our
procedures for the measurement of body density (BD), total
body water (TBW) and bone mineral mass (BMM) via
hydrodensitometery, isotopic dilution and dual-energy
X-ray absorptiometry (DXA), respectively, have been
described previously (van der Ploeg et al, 2000; Withers
et al, 1998). The FFM was then calculated using a fourcompartment criterion model (Withers et al, 1998). Brie¯y,
the masses and volumes for TBW and BMM were subtracted from those determined for the whole body using
hydrodensitometry (BD ˆ mass=volume). This enabled the
remainder to be partitioned into fat and residual (protein,
nonbone mineral and glycogen) masses whose respective
densities were assumed to be 0.9007 (Fidanza et al, 1953)
and 1.404 g=cm3 (Allen et al, 1959; 37 C only samples).
The %BF was also estimated using the two-compartment
hydrodensitometric body composition model (BrozÏek et al,
1963; %BF ˆ 497.1=BD 7 451.9). Prior to this experiment,
repeated trials for BD (n ˆ 12), TBW (n ˆ 10) and BMM
(n ˆ 12) yielded respective ICCs and %TEMs of 0.998 and
0.1% (0.001 g=cm3), 0.998 and 0.6% (0.28 l), and 0.998
and 0.9% (27 g).
147
Anthropometry
Measurements were taken in accordance with the procedures of Norton et al (1996) by an ISAK (International
Society for the Advancement of Kinanthropometry) level
one anthropometrist. Height was determined with a wall
statiometer and body mass was measured to the nearest 20 g.
Two trials were conducted at each skinfold site with
Harpenden callipers and the mean was used if they differed
by < 10%. Otherwise, a third measurement was taken and
the median was used in further calculations. Measurements
of girths (arm relaxed, arm ¯exed and tensed, waist, gluteal
and calf) and breadths (biepicondylar humerus and femur)
were conducted with a ¯exible steel tape and Mitutoyo
vernier callipers as modi®ed by Carter (1980). The linear
measurements for all anthropometric equipment were
checked against standard rods and the dowscale jaw pressure of the Harpenden callipers was 8.09 ± 7.74 g=mm2 for
jaw openings from 5 to 40 mm (Carlyon et al, 1998). The
weighing scale was calibrated throughout the physiological
range of measurement using masses which were authenticated by an electrobalance at the South Australian Of®ce of
Fair Trading.
Statistical analysis
Forward stepwise regression was used to predict %BF,
which was derived via the four-compartment body composition model, from the best weighted combination of the
sum of skinfold thicknesses, girths and breadths. Only those
variables that resulted in a statistically signi®cant (P 0.05)
increase in prediction were included in the ®nal equation.
Logarithmic and quadratic transformations were used for
those variables whose relationships with the criterion
departed signi®cantly from linearity. Calculation of the
FM from the predicted %BF (FM ˆ %BF=100 body
mass) enabled the FFM to be estimated by subtraction
(FFM ˆ body mass 7 predicted FM).
The previously outlined forward stepwise regression was
also used to predict RMR (kJ=day) from: (a) age, height,
mass and FFM predicted via anthropometric measurements;
and (b) age, height, mass and FFM derived from the fourcompartment body composition model.
The dependent t-test was used to determine whether our
measured RMR mean was signi®cantly different (P 0.05)
from those predicted by the equations of Harris & Benedict
(1919), Scho®eld (1985), Hayter & Henry (1994) and Piers
et al (1997). Total errors were also calculated as follows:
Total error ˆ
s
S …our measured RMR ÿ their predicted RMR†2
n
European Journal of Clinical Nutrition
Resting metabolic rate of Australian males
GE van der Ploeg et al
148
This statistic includes two sources of variation, one due to
the lack of association between the two sets of measurement
(SEE) and one attributed to the difference between the
means (Lohman, 1981). Finally, prediction errors (our
measured RMR 7 their predicted RMR) were regressed
linearly on our predicted RMR. If neither the slope nor
the intercept differ signi®cantly (P 0.05) from zero then
the equation essentially does not differ from our own which
is optimal for our data. However, if the regression line's
slope is signi®cantly different from zero then it is implied
that the prediction errors vary across the RMR range;
alternatively, if the slope is not signi®cantly different from
zero but the intercept is then the equation has a consistent
bias across the RMR range.
Figure 1 Regression of %BF via the four-compartment criterion model on
the sum of seven skinfold thicknesses.
Results
The descriptive statistics for the 38 subjects are contained in
Table 1. Figure 1 shows a curvilinear relationship between
%BF via the four compartment body composition model
and the sum of seven skinfold thicknesses. Hence, as
indicated in Figure 1, a quadratic transformation of the
skinfold thickness data resulted in a small increase in the
interclass correlation coef®cient and a small decrease in
the SEE compared with the linear model. Intermediate
values (R ˆ 0.960; SEE ˆ 2.0%BF) were registered for the
logarithmic transformation. None of the other anthropometric variables resulted in a statistically signi®cant
increase in the prediction of %BF.
Our regression equations for the prediction of RMR are
presented in Table 2. The multiple regression equation
generated from the easily determined predictors of mass,
height and age correlated 0.841 with RMR measured via
indirect calorimetry and the SEE was 521 kJ=day. The
incorporation of FFM as a predictor increased both the
Table 1
Descriptive statistics for the 38 male subjects
Age (y)
Mass (kg)
Height (cm)
Quetelet's index (kg=m2)
RMR (kJ=day)
RER
HR (rest)
%BF two-compartment model (UWW)
Four compartment model
%BF
FFM density (g=cm3)a
%TBW=FFMb
%BMM=FFMc
Sum of seven skinfold thicknesses (mm)d
a 1
b
BD
multiple correlation coef®cient and the precision of prediction but there was virtually no difference between FFM via
the four-compartment model (R ˆ 0.893; SEE ˆ 433 kJ=
day) and that predicted from the sum of seven skinfold
thicknesses (R ˆ 0.886; SEE ˆ 440 kJ=day).
Tables 3 and 4 contain the cross-validation results. The
regression equations of Harris & Benedict (1919) and
Scho®eld (1985) all overestimated (P 0.001) the RMR
of our subjects by mean differences of 518 ± 600 kJ=day.
The absolute mean differences and total errors were 618 ±
690 kJ=day and 745 ± 809 kJ=day, respectively. Figure 2 and
the prediction error analyses for slope (P 0.64) in Table 3
emphasise that these errors were fairly constant across the
range of measurement; however, the relative errors of 8.4 ±
11.7% in Table 4 were greatest at the lower end of the
distribution. The RMR means predicted from the regression
equations of Piers et al (1997) and Hayter & Henry (1994)
did not differ signi®cantly from our measured mean but
X
s.d.
Range
24.3
85.04
180.6
26.0
7668
0.829
54
19.0
3.3
13.82
8.3
3.5
923
0.048
7
7.4
18.8 ± 29.9
58.24 ± 115.21
166.0 ± 198.4
19.7 ± 33.4
5417 ± 9771
0.731 ± 0.980
39 ± 64
6.5 ± 32.2
21.3
1.1070
72.10
5.57
98.4
7.0
0.0034
0.77
0.37
42.0
10.0 ± 33.2
1.1002 ± 1.1138
70.85 ± 74.34
4.68 ± 6.25
36.3 ± 187.7
fFM
ˆ 0:9007
‡ FFMfFFM
density where fFM ‡ fFFM ˆ 1 and are from the four-compartment model.
%TBW=FFM ˆ fat-free mass hydration.
%BMM=FFM ˆ percentage of bone mineral mass in the fat-free mass.
Seven skinfolds ˆ triceps ‡ subscapular ‡ biceps ‡ supraspinale ‡ abdominal ‡ front thigh ‡
medial calf.
c
d
European Journal of Clinical Nutrition
Resting metabolic rate of Australian males
GE van der Ploeg et al
149
Table 2 Regression equations predicting RMR of 18 to 30-y-old males
Regression equations (kJ=day)
n
This study
RMR1 ˆ 53:5…M† ‡ 3116
RMR2 ˆ 45:5…M† ‡ 24:6…H† ÿ 635
RMR3 ˆ 48:2…M† ‡ 25:8…H† ÿ 49:6…A† ‡ 113
RMR4 ˆ 21:0…M† ÿ 56:2…A† ‡ 76:1…FFM 4C† ‡ 2202
RMR5 ˆ 109:2…FFM S7SF† ÿ 68:4…A† ‡ 2092
Harris & Benedict (1919)
RMR1 ˆ 54:96…M† ‡ 26:74…H† ÿ 1317
RMR2 ˆ 57:56…M† ‡ 20:94…H† ÿ 28:28…A† ‡ 278
Scho®eld (1985)
RMR1 ˆ 63:0…M† ‡ 2896
RMR2 ˆ 63:0…M† ÿ 0:42…H† ‡ 2953
Hayter & Henry (1994)
RMR ˆ 51:0…M† ‡ 3500
Piers et al (1997)
RMR ˆ 51:0…M† ‡ 3415
38
136
2879
478
39
r
r2
SEE (kJ=day)
0.802
0.823
0.841
0.893
0.886
0.643
0.677
0.707
0.797
0.784
559
539
521
433
440
0.819a
0.868a
0.671
0.753
497a
432a
0.65
0.65
0.423
0.423
641
641
0.67
0.449
NR
0.77
0.593
499
M ˆ mass (kg); H ˆ height (cm); A ˆ age (y); SEE ˆ standard error of estimate; NR ˆ not reported.
FFM 4C ˆ fat-free mass (kg) via the four-compartment body composition model.
FFM S7SF ˆ fat-free mass (kg) predicted from sum of seven skinfold thicknesses (triceps ‡subscapular ‡ biceps ‡ supraspinale ‡ abdominal ‡ front
thigh ‡ medial calf).
a
Calculated from raw data.
Figure 2 shows that this was because overpredictions below
the intersections of the regression lines were balanced out
by underpredictions above the intersections. Consequently,
these were the only two equations where both the slopes
(P ˆ 0.03) and the intercepts (P 0.03) for the prediction
errors differed signi®cantly from zero and their largest
relative errors of 8.8% and 10.2% overpredictions in
Table 4 were at the lower end of the distribution. All the
total errors reported in Table 3 are larger than the SEE for
the original equations (Table 2).
Discussion
A major ®nding of this study is that predicting RMR from
FFM measured via the four-compartment body composition
model achieved little extra accuracy over FFM estimated
from the sum of seven skinfold thicknesses. This is
Figure 2 Regression of RMR predicted from cross-validated equations on
our predicted RMR. (RMR (kJ=day) ˆ 21.0 (mass) 7 56.2 (age) ‡
76.1 (FFM by four-compartment model) ‡ 2202).
Table 3 Cross-validation of RMR prediction equations against the measured RMR of 38 South Australian males aged 18 ± 30 y
Measured RMR and their predicted RMR
Equation
Harris & Benedict (1919)
Harris & Benedict (1919)
Scho®eld (1985)
Scho®eld (1985)
Hayter & Henry (1994)
Piers et al (1997)
d a
jdj
Predictors (kJ=day) (kJ=day)
H, M
A, H, M
M
H, M
M
M
ÿ518
ÿ600
ÿ585
ÿ566
ÿ168
ÿ83
618
673
690
676
458
436
t
P
r
7 5.87
ÿ7.11
ÿ6.37
ÿ6.15
ÿ1.88
ÿ0.93
< 0.001
< 0.001
< 0.001
< 0.001
0.07
0.36
0.823
0.834
0.802
0.801
0.802
0.802
Linear regression of prediction
errorsd on our predicted RMR
SEE
Total errorb
tslope ( P)
tintercept ( P)
(kJ=day) (kJ=day) (%c) (H0 : slope ˆ 0) (H0 : intercept ˆ 0)
532
515
559
560
559
559
745
790
809
796
571
551
9.7
10.3
10.6
10.4
7.5
7.2
ÿ0.14
ÿ0.07
0.45
0.47
2.21
2.21
(0.89)
(0.95)
(0.66)
(0.64)
(0.03)
(0.03)
ÿ0.47 …0:64†
ÿ0.67 …0:51†
ÿ1.11 (0.28)
ÿ1.11 (0.27)
ÿ2.41 …0:02†
ÿ2.30 (0.03)
A ˆ age; H ˆ height; M ˆ mass; SEE ˆ standard error of estimate.
Our measured
X Ð their predicted X .
p
Total error= …S…our measured RMRÿÿtheir predicted RMR†2 =n.
c
% ˆ total error=our mean (7668).
d
Prediction errors ˆ our measured RMR ÿ their predicted RMR.
a
b
European Journal of Clinical Nutrition
Resting metabolic rate of Australian males
GE van der Ploeg et al
150
Table 4
Prediction errors (our predicted RMR 7 their predicted RMR) with percentages in parentheses
Equations
Predictors
X ÿ 2 s:d:
X ÿ 1 s:d:
X
Our predicteda
Harris & Benedict (1919)
Prediction error
Harris & Benedict (1919)
Prediction error
Scho®eld (1985)
Prediction error
Scho®eld (1985)
Prediction error
Hayter & Henry (1994)
Prediction error
Piers et al (1997)
Prediction error
A, M, FFM
H, M
5823
6311
ÿ488 (8.4)
6410
ÿ587…10:1†
6502
ÿ679 (11.7)
6488
ÿ665 (11.4)
6419
ÿ596 (10.2)
6334
ÿ511 (8.8)
6746
7249
ÿ503 (7.5)
7339
ÿ593 (8.8)
7378
ÿ632 (9.4)
7362
ÿ616…9:1†
7128
ÿ382 (5.7)
7043
ÿ297 (4.4)
7668
8186
ÿ518 (6.7)
8268
ÿ600 (7.8)
8253
ÿ585 (7.6)
8234
ÿ566…7:4†
7836
ÿ168 (2.2)
7751
ÿ83 (1.1)
a
A, H, M
M
H, M
M
M
X ‡ 1 s:d:
X ‡ 2 s:d:
8591
9123
ÿ532 (6.2)
9198
ÿ607 (7.1)
9129
ÿ538 (6.3)
9107
ÿ516…6:0†
8546
45 (ÿ0:5)
8461
130 (ÿ1:5)
9514
10061
ÿ547 (5.7)
10127
ÿ613 (6.4)
10005
ÿ491 (5.2)
9980
ÿ466 (4.9)
9255
259 (ÿ2:7)
9170
344 (ÿ3:6)
RMR (kJ=day) ˆ 21:0 …mass† ÿ 56:2 …age† ‡ 76:1 …FFM by four-compartment model) ‡ 2202.
signi®cant for two reasons. First, the four-compartment
body composition model is a relatively time-consuming
and expensive method compared with the measurement of
skinfold thicknesses. Second, our measured RMR correlated
0.857 and 0.802 with FFM and body mass, respectively, yet
the latter easily measured variable has been used more
frequently as a predictor. Table 2 therefore indicates that
incorporating predicted FFM explained 7.7% more of the
RMR variance than a regression equation based on just the
traditional independent variables of age, height and mass;
furthermore, the SEE decreased from 521 to 440 kJ=day.
Our two best equations in Table 2 indicate that 79.7%
and 78.4% of the variance in the criterion variable of RMR
was attributable to the variance of the combined predictor or
independent variables. While these percentages compare
very favourably with those of other equations in Table 2,
the coef®cient of determination is only a valid measure of
the relative worth of multiple regression equations if all the
sample sizes and standard deviations for the criterion or
dependent variable are identical. This is certainly not the
case for the present comparison and in such circumstance
the SEE is a more valid indicator of an equation's predictive
accuracy. In this respect, the Scho®eld equations in Table 2
stand out as those with the highest SEE.
The SEE for our best equations in Table 2 is 433 kJ=day.
However, one of the independent variables is FFM via the
four-compartment model, which requires just as much time
but more expensive equipment compared with direct measurement of RMR. A more user-friendly equation is the one
which employs age and FFM estimated from the sum of
seven skinfold thicknesses as predictors and has a comparable SEE of 440 kJ=day. The probability at the mean is
therefore 0.95 that a person's measured RMR lies within the
range of the predicted RMR 862 kJ=day. This imprecision
raises the question as to whether it is more acceptable to
tolerate the error in the prediction or the dif®culty of direct
measurement.
Figure 2 illustrates that the equations of Harris &
Benedict (1919) and Scho®eld (1985) overpredicted the
RMR of our subjects by a relatively constant amount of
European Journal of Clinical Nutrition
500 ± 600 kJ=day throughout the range of measurement.
Hence, their regression lines in Figure 2 are above and
parallel to the one for our data. These consistent biases
across the RMR range were con®rmed by the statistically
signi®cant differences (P < 0.001) between our measured
RMR mean and their predicted RMR means and the linear
regressions of the prediction errors on our predicted RMR,
which all resulted in slopes that were not signi®cantly
different from zero (P 0.64). The most likely reason for
the constant overprediction of our subjects' RMR by these
investigators' equations is that their subjects had a lower
%BF and hence higher relative FFM than our sample. As
explained previously, the FFM has a much higher RMR than
adipose tissue. However, neither study reported body composition data so it is impossible to verify this hypothesis.
Nevertheless, while both groups were signi®cantly lighter
(Harris & Benedict: 64.1 10.3 kg, P < 0.001; Scho®eld:
63.0 8.7 kg, P < 0.001) and shorter (Harris & Benedict:
173.0 7.6 cm, P < 0.001; Scho®eld: 170.0 7.3 cm,
P < 0.001) than our subjects, it is possible to calculate
Quetelet's index (mass (kg)=height (m)2), which is used as
a body composition marker by epidemiologists. The
National Heart Foundation of Australia (1990) accordingly
regards persons with a Quetelet's index of 20 ± 25 kg=m2
inclusive to be of acceptable weight, whereas those with
scores greater than 25 and 30 are regarded as being overweight and obese, respectively. Table 1 shows that the mean
for our subjects was 26.0 kg=m2 with a range of 19.7 ±
33.4 kg=m2. The data of Harris & Benedict (1919) and
Scho®eld (1985) yielded much lower respective means of
21.3 and 21.8 kg=m2. Also the correlation between Quetelet's index and %BF via the four-compartment body composition model was 0.833 for our heterogeneous sample.
However, it must be noted that Quetelet's index can be
insensitive to body composition because it is possible to
have a high score yet to be very lean (Withers et al, 1997).
Additional support for our hypothesis that the subjects of
Harris & Benedict (1919) and Scho®eld (1985) had a lower
%BF than our subjects is that the overpredictions decreased
to 293 ± 435 kJ=day when the data for our leanest males
Resting metabolic rate of Australian males
GE van der Ploeg et al
(n ˆ 13; X s.d.: 13.5 2.2 %BF; range 10.0 ± 16.7 %BF)
were analysed separately. Only the difference between our
measured RMR and that predicted by the Harris & Benedict
(1919) equation that used mass, height and age as predictors
was statistically signi®cant (P ˆ 0.02). Finally, while the
data reported in 1919 by Harris & Benedict are on nonathletes, it is likely that they were relatively more active
than untrained persons today and this would have resulted in
a greater relative FFM.
The database of Scho®eld (1985) contains a disproportionate number of Italian subjects who comprise 56% of the
18 to 30-y-old male cohort. Many of these were military
cadets, military servicemen, labourers and miners who are
not representative of the Italian population. Piers et al
(1997) accordingly agree with our ®ndings that the Scho®eld (1985) equations overpredict the RMR of Australian
subjects and they hypothesize that this is due to the greater
relative FFM of the Italian subjects in the Scho®eld (1985)
database. Interestingly, Hayter & Henry (1994) generated
RMR prediction equations for North European and American subjects by excluding the Italian subjects from the
Scho®eld (1985) database. Table 3 demonstrates that their
resultant equation for males agrees far more closely with our
measured RMR than the Scho®eld (1985) equation and this
is in accordance with previous work on Australian subjects
(Piers et al, 1997). Figure 2 illustrates the similarity between
the values predicted by the Hayter & Henry (1994) and Piers
et al (1997) equations. The prediction errors are least around
those points where the regression lines for these two crossvalidated equations intersect that for the South Australian
subjects: they then progressively underpredict RMR above
the intersection to a maximum of 340 kJ=day and progressively overpredict below the intersection to a maximum
of 620 kJ=day. The errors therefore balanced out such that
neither equation resulted in a statistically signi®cant difference (P 0.07) between the means for measured and predicted RMR but their absolute mean differences in Table 3
were much greater than the differences between the means.
Their total errors were also much lower than those for the
Harris & Benedict (1919) and Scho®eld (1985) equations.
Nevertheless, both the Hayter & Henry (1994) and Piers
et al (1997) equations exhibited prediction errors
(measured RMR 7 predicted RMR) which, when linearly
regressed on our predicted RMR, yielded slopes and
intercepts which differed signi®cantly (P 0.03) from
zero. However, the prediction errors were only of physiological signi®cance at the lower end of our range of
measurement.
It must be emphasized that the present study was a pilot
project. Hence, while the sample was speci®cally chosen to
represent equal numbers of short, medium and tall and light,
medium and heavy persons within 3 y age bands, the
preceding cross-validation analyses suggested that equations
need to be generated from a large database for the prediction
of the RMR of Australian males. The resultant increase in
the ratio of subjects to independent variables would enhance
the generalizability of the equations (Tabachnick & Fidell,
1996).
The SEE of 1.9 %BF for the prediction of body fat via the
four-compartment criterion body composition model from
the sum of seven skinfold thicknesses (see Figure 1) is lower
than that of 2.9 %BF reported by Williams et al (1992) for
the sum of nine skinfold thicknesses of 91 American males
aged 34 ± 84 y. Their coef®cient of determination indicated
80% shared variance between the dependent and independent variables compared with 93% for this study. Our
correlation and SEE were also much higher and lower,
respectively, than for equations in the literature (Norton,
1996) where anthropometric variables (skinfolds, girths and
breadths) have been used to predict body density from
which the %BF has been estimated via the two-compartment body composition model using either the Siri (1961) or
BrozÏek et al (1963) equations. Nevertheless, as far as our
predicted %BF is concerned, one is still left with a 95%
con®dence interval of 3.7 %BF at the mean.
The two-compartment hydrodensitometric body composition model yielded a lower %BF than the four-compartment criterion model (P < 0.001; Table 1), albeit the
interclass correlation coef®cient between these two methods
was 0.989. The former model assumes that the body can be
partitioned into the FM and FFM whose respective densities
are 0.9007 (Fidanza et al, 1953) and 1.1000 g=cm3 (BrozÏek
et al, 1963). The FFM compartment comprises four components whose percentages and densities (in parentheses) at
36 C are assumed (BrozÏek et al, 1963) to be as follows:
19.41%
protein
73.72%
water
(0.99371 g=cm3),
3
(1.34 g=cm ), 5.63% bone mineral (2.982 g=cm3) and
1.24% non-bone mineral (3.317 g=cm3). Persons with a
FFM density greater than 1 1000 g=cm3 will therefore
have their %BF underestimated via hydrodensitometry,
whereas the converse applies to those whose FFM density
is below 1 1000 g=cm3. Table 1 indicates that the mean
FFM hydration for our subjects of 72.1% was below the
two-compartment hydrodensitometric assumption of
73.72% that is based on analyses of just three male cadavers
aged 25, 43 and 46 y. Hence, their FFM density was greater
than 1.1000 g=cm3 because water has by far the lowest
density of any of the four FFM components and their
%BF was therefore underestimated via hydrodensitometry.
In summary, our data on a heterogeneous sample of 38,
18 to 30-y-old Australian males suggested the following
conclusions:
151
(1) The coef®cient of determination and SEE for our best
RMR prediction equation compared extremely favourably with those of other equations in the literature.
However, the 95% con®dence interval at the mean of
862 kJ=day raised the question as to whether it was
more acceptable to tolerate the error in the prediction or
the dif®culty of direct measurement.
(2) FFM predicted from the sum of skinfold thicknesses,
which has been validated against the criterion of FFM
via the four-compartment body composition model,
needs to be further explored as an RMR predictor.
(3) Cross-validation analyses produced prediction errors
that were unequivocally of long-term physiological
European Journal of Clinical Nutrition
Resting metabolic rate of Australian males
GE van der Ploeg et al
152
signi®cance. Equations therefore need to be generated
from a large database for the prediction of the RMR of
Australian males.
References
Allen TH, Krzywicki HJ & Roberts JF (1959): Density, fat, water, and
solids in freshly isolated tissues. J. Appl. Physiol. 14, 1005 ± 1008.
BrozÏek J, Grande F, Anderson JT & Keys A (1963): Densitometric analysis
of body composition: revision of some quantitative assumptions. Ann.
NY Acad. Sci. 110, 113 ± 140.
Carlyon RG, Bryant RW, Gore CJ & Walker RE (1998): Apparatus for
precision calibration of skinfold calipers. Am. J. Hum. Biol. 10, 689 ±
697.
Carter JEL (1980): The Health-Carter Somatotype Method, pp 2 ± 4. San
Diego, CA: San Diego State University.
Elia M (1992a): Energy expenditure in the whole body. In Energy
Metabolism: Tissue Determinants and Cellular Corollaries, ed. JM
Kinney, HN Tucker, pp 19 ± 59. New York: Raven Press.
Elia M (1992b): Organ and tissue contribution to metabolic rate. In Energy
Metabolism: Tissue Determinants and Cellular Corollaries, ed. JM
Kinney, HN Tucker, pp 61 ± 79. New York: Raven Press.
Elia M & G Livesey (1992): Energy expenditure and fuel selection in
biological systems. The theory and practice of calculations based on
indirect calorimetry and tracer methods. In Metabolic Control of Eating,
Energy Expenditure and the Bioenergetics of Obesity, ed. AP Simopoulos, p 89. Basel: Karger.
FAO=WHO=UNU (1985): Energy and Protein Requirements. WHO Technical Report Series 724, pp 1 ± 206. Geneva: WHO.
Fidanza F, Keys A & Anderson JT (1953): Density of body fat in man and
other mammals. J. Appl. Physiol. 6, 252 ± 256.
Geppert J & Zuntz N (1888): Ueber die Regulation der Athmung. P¯uÈegers
Arch. 42, 189 ± 245.
Harris JA & Benedict FG (1919): A Biometric Study of Basal Metabolism
in Man, Publication no 279, pp 1 ± 266. Washington DC: Carnegie
Institution.
Hart JD & Withers RT (1996): The calibration of gas volume measuring
devices at continuous and pulsatile ¯ows. Aust. J. Sci. Med. Sport 28,
61 ± 65.
Hayter JE & Henry CJK (1994): A re-examination of basal metabolic rate
predictive equations: the importance of geographic origin of subjects in
sample selection. Eur. J. Clin. Nutr. 48, 702 ± 707.
Hicks CS, Matters F & Mitchell ML (1931): The standard metabolism of
Australian Aborigines. Aust. J. Exp. Biol. Med. Sci. 8, 69 ± 82.
European Journal of Clinical Nutrition
Lohman TG (1981): Skinfolds and body density and their relation to body
fatness: a review. Hum. Biol. 53, 181 ± 225.
National Heart Foundation of Australia (1990): Risk Factor Prevalence
Study Survey No 3 1989, p 46. Canberra: National Heart Foundation of
Australia and Australian Institute of Health.
Norton K (1996): Anthropometric estimation of body fat. In Anthropometrica, ed. K. Norton, T Olds, pp 171 ± 198. Sydney: University of New
South Wales Press.
Norton K, Whittingham N, Carter L, Kerr D, Gore C & Marfell-Jones M
(1996): Measurement techniques in anthropometry. In Anthropometrica,
ed. K Norton, T Olds, pp 25 ± 75. Sydney: University of New South
Wales Press.
Piers LS, Diffey B, Soares MJ, Frandsen SL, McCormack LM, Lutschini MJ &
O'Dea K (1997): The validity of predicting the basal metabolic rate of
young Australian men and women. Eur. J. Clin. Nutr. 51, 333 ± 337.
Poehlman ET (1989): A review: exercise and its in¯uence on resting energy
metabolism in man. Med. Sci. Sports Exerc. 21, 515 ± 525.
Scho®eld WN (1985): Predicting basal metabolic rate, new standards and
review of previous work. Hum. Nutr. Clin. Nutr. 39C(Suppl 1), 5 ± 41.
Siri WE (1961): Body composition from ¯uid spaces and density: analysis
of methods. In Techniques for Measuring Body Composition, ed. J
BrozÏek, A Henschel, pp 223 ± 244. Washington, DC: National Academy
of Science, National Research Council.
Tabachnick BG & Fidell LS (1996): Using Multivariate Statistics, pp 132 ±
133. New York: Harper Collins College Publishers.
van der Ploeg GE, Gunn SM, Withers RT, Modra AC & Crockett AJ
(2000): Comparison of two hydrodensitometric methods for estimating
percent body fat. J. Appl. Physiol. 88, 1175 ± 1180.
Wardlaw HSH & Horsley CH (1928): The basal metabolism of some
Australian Aborigines. Aust. J. Exp. Biol. Med. Sci. 5, 263 ± 272.
Wardlaw HSH & Lawrence WJ (1932): Further observations on the basal
metabolism of Australian Aborigines. Aust. J. Exp. Biol. Med. Sci. 10,
157 ± 166.
Wardlaw HSH, Davis HW & Joseph MR (1934): Energy metabolism and
insensible perspiration of Australian Aborigines. Aust. J. Exp. Med. Sci.
12, 63 ± 74.
Williams DP, Going SB, Lohman TG, Hewitt MJ & Haber AE (1992):
Estimation of body fat from skinfold thicknesses in middle-aged and
older men and women: a multiple component approach. Am. J. Hum.
Biol. 4, 595 ± 605.
Withers RT, Noell CJ, Whittingham NO, Chatterton BE, Schulz CG &
Keeves JP (1997): Body composition changes in elite male bodybuilders
during preparation for competition. Aust. J. Sci. Med. Sport 29, 11 ± 16.
Withers RT, LaForgia J, Pillans RK, Shipp NJ, Chatterton BE, Schultz CG
& Leaney F (1998): Comparisons of two-, three-, and four-compartment
models of body composition analysis in men and women. J. Appl.
Physiol. 85, 238 ± 245.