| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Vol. 83, Issue 6, 2112-2122, December 1997
Department of Biomedical Engineering and Medicine, Vanderbilt University, Nashville, Tennessee 37232-2279
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
ACKNOWLEDGEMENTS
FOOTNOTES
REFERENCES
ABSTRACT 
Chen, Kong Y., and Ming Sun. Improving energy
expenditure estimation by using a triaxial accelerometer.
J. Appl. Physiol. 83(6):
2112-2122, 1997.
In our study of 125 subjects (53 men and 72 women) for two 24-h periods, we validated energy expenditure (EE),
estimated by a triaxial accelerometer (Tritrac-R3D), by using a
whole-room indirect calorimeter under close-to-normal living
conditions. The estimated EE was correlated with the measured total EE
for the 2 days (r = 0.925 and
r = 0.855;
P < 0.001) and in minute-by-minute
EE (P < 0.01). Resting EE formulated
by the Tritrac was found to be similar to the measured values
[standard errors of estimation (SEE) = 0.112 W/kg;
P = 0.822]. The Tritrac significantly underestimated total EE, EE for physical activities, EE
of sedentary and light-intensity activities, and EE for exercise such
as stepping (all P < 0.001). We
developed a linear and a nonlinear model to predict EE by using the
acceleration components from the Tritrac. Predicted EE was
significantly improved with both models in estimating total EE, total
EE for physical activities, EE in low-intensity activities,
minute-by-minute averaged relative difference, and minute-by-minute SEE
(all P < 0.05). Furthermore, with
our generalized models and by using subjects' physical characteristics and body acceleration, EE can be estimated with higher accuracy (averaged SEE = 0.418 W/kg) than with the Tritrac model.
physical activity; metabolism; modeling; validation; whole-room
indirect calorimeter
INTRODUCTION WITH THE ESTABLISHMENT of a negative correlation
between energy costs of physical activities and incidence of
morbidity/mortality for some chronic diseases, such as coronary heart
disease, hypertension, and diabetes (2, 20), the quantification of
energy expenditure (EE) and daily physical activities has gained
considerable interest. Several methods and automated devices have been
adopted, including doubly labeled water (9, 27), indirect calorimeter
(27, 28), pencil and paper methods (30), heart-rate monitors (26, 27),
pedometers (6), and motion sensors (12, 16, 21). Researchers as well as
the general public use these methods to facilitate assessment of EE,
activity levels, and general fitness profiles (18). Each method has
strengths and limitations. The indirect calorimeter can measure EE
accurately under laboratory conditions and is regarded as the accepted
standard in validating energy costs of various physical activities
(15), but accurate minute-by-minute activity monitoring and EE
predictions in free-living conditions are needed as a more practical
alternative for the measurement of EE.
In recent years, a growing number of portable devices has become
available for daily physical activity monitoring, such as the Polar
heart-rate recorder (Polar Electro, Kempele, Finland) and the Caltrac
and the Tritrac-R3D accelerometers (both by Hemokinetics, Madison, WI).
Heart-rate monitors have been studied extensively in terms of technical
reliability and methodological usefulness in measuring physical
activities and predicting EE (27). Accelerometers have become more
attractive to the researchers because they have a longer recording time
period, a rugged nature, and are capable of providing an indication of
reliability and objectiveness (1, 14, 19). These devices use
piezoelectric accelerometers and advanced microcomputers to register
body motion (specifically, acceleration and deceleration during
activities that involve energy cost). A linear relationship has been
reported between EE and the body acceleration in walking (4, 5, 32).
However, the uniaxial accelerometer, such as the Caltrac, was found to be inaccurate in measuring certain activities, including some sedentary
activities (13, 19, 23) and running (7).
A triaxial accelerometer, the Tritrac-R3D, is a monitor developed to
correct some of the limitations of the Caltrac. This device, which
combines three independent sensors in orthogonal axes to detect
acceleration in the three-dimensional space, may improve accuracy,
especially during sedentary activities. The Tritrac monitor provides
minute-by-minute data in acceleration counts by which EE of physical
activity (EEact) is estimated. Moreover, with one-time programming, it
allows a recording period of up to 14 days, and this information can be
downloaded to a computer. These advances simplify the data-retrieval
process, can reduce some possible risks of subject tampering, and may
increase the reliability of this triaxial accelerometer in measuring
daily physical activity and the associated EE. A linear relationship (r > 0.77, P < 0.05) has been reported between
triaxial accelerometer output and EE for some physical activities (4,
14). These investigators have also suggested that a triaxial
accelerometer, such as the Tritrac, could be a good method for
correlating acceleration of physical activities with EE.
However, the validity of the Tritrac is under considerable
investigation. It still cannot adequately detect slow or subtle changes
in EE, such as during sleeping, the thermic effect of food, or EE
caused by some physical activities or exercises. Furthermore, theoretical questions have been raised about the linearity of the
relationship between body acceleration and EE (4). It has been reported
that the Tritrac overestimated time accumulated in sedentary activities
but underestimated active time components; thus it significantly
underestimated total EE under free-living conditions (14). Limited
studies using the Tritrac have been reported that correlated with heart
rate, the Caltrac, and self-reports in children (31) and adults (14)
under free-living conditions. But to our knowledge, no one has studied
the accuracy of EE estimation by using the Tritrac, compared with a
whole-room indirect calorimeter, in a minute-by-minute basis over 24-h
periods. Furthermore, no modified models have demonstrated more
accurate estimations of EE from the body accelerations. With an
accurate and fast-responding whole-room indirect calorimetry chamber,
we can accomplish these goals by using simultaneously measured EE and
body acceleration.
The primary purpose of this investigation was to use a whole-room
indirect calorimeter to evaluate the validity and accuracy of the
Tritrac for the normal adult population under conditions close to free
living and with different physical activities and exercises of various
intensities. We also compared the resting EE (REE) formulated by the
Tritrac to the values measured by the indirect calorimeter in two
separate 24-h periods. Mathematical models were established to predict
EE on a minute-by-minute and on a daily basis for each subject by using
the acceleration output from the monitor. Furthermore, we evaluated the
significance of these model parameters with respect to the subject's
characteristics, and we simplified the models from individual to
general forms that can be useful for the estimation of EE in large
populations in free-living conditions.
METHODS Subjects
Table 1.
Subject physical characteristics
Characteristics
Women
(n = 72)
Men (n = 53)
Mean ± SD
Range
Mean ± SD
Range
Body mass, kg
76.9 ± 24.0
46.0-154.4
89.1 ± 21.4
55.5-143.5
Height, cm
163.5 ± 6.1
150.5-184.0
177.3 ± 6.7
162.0-189.0
Age, yr
37.1 ± 9.8
20.0-62.0
35.7 ± 10.0
19.0-56.0
BMI,
kg/m2
28.6 ± 8.1
17.8-53.4
26.5 ± 6.6
18.8-45.5
Body fat, %
34.6 ± 10.5
5.4-58.9
24.1 ± 9.9
8.3-44.9
Values are means ± SD and ranges (minimum to maximum); n = no. of subjects. BMI, body mass index.
Experimental Procedures
Male and female adult volunteers were recruited from the Nashville, TN, area by means of posters, the Vanderbilt University periodical, and personal contact. Before participation, all subjects signed an informed consent form approved by the Vanderbilt University Committee for the Protection of Human Subjects. For each subject, two 24-h stays in the room calorimeter were scheduled within 8 consecutive days; these stays were separated by at least 1 day. Tritrac monitoring was recorded on a minute-by-minute basis with the same device during the 2 days. During one 24-h stay, the subjects were asked to structure their activity patterns as closely as possible to their normal daily activity routine (normal day). For the other day, the subject was asked to engage in a defined physical activity-exercise protocol (exercise day). The protocol consisted of three 10-min walking bouts (with average speeds of 0.6, 0.9, and 1.2 m/s) and four 10-min simple step up-and-step down movements (with an average speed of 12, 18, 24, and 30 steps/min), with each activity separated by a 10-min resting period. The same pattern of exercises was performed at two defined time periods in the morning and the afternoon. The order of participation in the normal and the exercise modes was randomized among subjects. During both the normal day and the exercise day, the subjects were also provided with free access to a step-exercise bench and a stationary bike for additional exercise needs. Meals were provided at exact times during the days. The REE measurement was obtained in a 30-min awake and quiet resting period preceded by overnight sleeping and fasting for all subjects on both days in the chamber. The subjects were also instructed to maintain a diary of physical activity during the entire study period to be used for verification purposes.Instrumentation
Accelerometer. The Tritrac activity monitors were used to measure minute-by-minute acceleration in three dimensions (x or anteroposterior axis, y or medial-lateral axis, and z or vertical axis). The monitor (weighing 170 g and measuring 11.1 × 6.7 × 3.2 cm) was worn on the right hip, in a nylon pouch secured to a belt at the waistline, during all activities throughout the study days except during sleeping. Part of the output of the Tritrac is expressed as integrated acceleration over each minute in the three axes. The subject's physical characteristics are entered (gender, age, height, and weight) on initializing the monitor. This information is used to calculate an individual's REE, in kilocalories per minute (4.19 kJ/min), based on established predictive equations (shown in Eqs. 1 and 2) used by the Tritrac (8)
![Men: <FR><NU><AR><R><C>[473 × weight (lb)] + [971 × height (in.)] </C></R><R><C>− [513 × age (yr)] + 4,687</C></R></AR></NU><DE>100,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img001.gif)
|
![Women: <FR><NU><AR><R><C>[331 × weight (lb)] + (352 × height (in.)] </C></R><R><C>− [353 × age (yr)] + 49,854</C></R></AR></NU><DE>100,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img002.gif)
|
Other physical measurements. Body mass was measured to the nearest 0.05 kg with the use of a digital scale. Height was measured to the nearest 0.5 cm. Body fat and fat-free mass were determined by hydrodensitometry (underwater weighing). The subjects were weighed underwater while their residual lung volume was measured by the nitrogen-dilution technique (10). Body fat percent was calculated from body density by using Siri's equation (24), whereas fat mass and fat-free mass were calculated from body mass.
Establishment of Models
The output of the integrated minute-by-minute body acceleration from the Tritrac and the subject's physical characteristics were used to estimate EEact (EEact = EE
REE). For each subject, a linear and
a nonlinear model were used. We first established individual models for
each subject with independent parameters for optimal estimation, as
shown in Fig. 2. These parameters were then
generalized so that only body mass, height, gender, and age were needed
from a subject to predict his or her EE by using the acceleration
output. Data from the exercise day
were used to generate the models because of the rich signal information
in acceleration and EE recordings. Each model (individual or
generalized) was then used to predict EE of the normal
day and compared with the measured EE from the room
calorimeter. Before modeling, vertical (V) and horizontal (H)
acceleration components were synchronized to EE measured
by the indirect calorimeter and were low-pass filtered by a three-point
moving-average filter to reduce artifacts, such as the monitor's
sudden movements caused by the subject's adjustment of the pouch.
to estimate EE for physical activity (EEact). Process
arrows 1b, 2b, and
2c: linear and nonlinear models, using
horizontal
(
) and vertical (z) to estimate EEact.
Process arrow 3a: generalization of model parameters
by using stepwise linear-regression analysis. Process
arrows 3 and
4: EEact estimated by each model were
compared with actual EEact measured.
p1 and
p2, Power parameters;
aL,
bL: regression
parameters in linear model (see Eq. 3);
aN, bN:
regression parameters in nonlinear model (see Eq. 4).
Linear model. The Tritrac calculates the EEact on the basis of the combined acceleration of all three axes (the vector magnitude
). In our models, the acceleration in the
z axis was isolated from the
x- and
y-axes as the V component, as shown in
Fig. 2. The H component was defined as the square root of the sum of
squared signals of the x- and
y-axes
(). The rationale for separating z from
x and
y signals is that acceleration in the
V component differs from the rest of dimensions because of gravity.
The linear model estimates EEact (in kJ/min) with the acceleration
signals from the Tritrac
|
(3) |
|
(4) |
Statistics
The accuracy of a particular model was evaluated by the difference between predicted and actual EE under various conditions over the 24-h period. The periods during the stays inside the room calorimeter were also segmented according to the intensity of EE. Sleeping periods were identified and isolated as the lowest intensity category. The other four categories (sedentary, light, moderate, and high intensities) were designated to represent 1-2.5, 2.6-4.0, 4.1-6, and >6.0 times the REE (MET = EE/REE). Correlation coefficient (Pearson's r), averaged relative differences (ARD), and standard errors of estimation (SEE) were used as the evaluation criteria, as shown in Eqs. 5 and 6
|
(5) |
|
(6) |
RESULTS
From the study of 125 adults, we found a significant intra-individual
correlation between the two separate 24-h stays of total EE measured by
the indirect calorimeter (r = 0.914, P < 0.001) as well as of total EE
estimated by the Tritrac (r = 0.959, P < 0.001). For EEact, the
correlation between the days was lower in measured values
(r = 0.534, P < 0.01) and in estimated values (r = 0.429, P < 0.01). REE measured in the two
stays in the calorimeter were not significantly different
(t value = 0.802, P = 0.424) by paired
t-test. The Tritrac also predicted the
REE accurately (SEE = 0.112 W/kg; P = 0.822) compared with the measurement from the calorimeter, as shown in
Fig. 3. Thus the REE formulated by the
Tritrac was also used in our estimation models. Within each of the 24-h
periods, total EE estimated by the Tritrac was significantly correlated
(P < 0.001) with that
measured by the whole-room indirect calorimeter (Fig.
4). Correlations between the measured and
the estimated EEact were lower than the total EE for the exercise and
the normal days (r = 0.737 and 0.539 respectively; P < 0.01). On a
minute-by-minute basis, the Tritrac-estimated EE values were strongly
correlated with measured values for the exercise
day (P < 0.001) and
the normal day
(P < 0.01), as illustrated in Tables 2 and
3.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Despite these significant correlations, the Tritrac significantly underestimated total EE for both the exercise day and the normal day (SEE = 0.523 and 0.468 W/kg, respectively; P < 0.001) compared with the calorimeter. There also existed a consistent underestimation of EEact by the Tritrac for both days combined (P < 0.001). This underestimation occurred in all the intensity categories except sleeping, in which a significant overestimation (P < 0.001) by the Tritrac was observed, as shown in Tables 4 and 5.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Individual Models
Linear model. By separating the V component (z-axis) from the other components, the linear model significantly reduced the prediction error of the total EE in the exercise day (P < 0.001), the normal day (P = 0.014), and the total EEact for both days (P < 0.001 and P = 0.003) compared with the Tritrac model. Such improvement existed in all of the four nonsleeping activity categories for the 2 days (P < 0.01) except in the sedentary activities (1.0-2.5 MET) during the normal day, as shown in Tables 4 and 5. The ARD also showed significant improvements (P
0.001) for the two stays, and
the SEE was significantly improved for the exercise
day (P = 0.002), as
seen in Fig. 5. However, the total EE and
EEact estimated by the linear model were still significantly lower than
that measured by the calorimeter for both days
(P < 0.01). Significant
(P < 0.01) difference in EE during
activities of lower intensities in the exercise
day and during all activities on the
normal day contributed to the
underestimation of the daily totals.
Significantly
different from linear model (individual or general);
P < 0.05.
Nonlinear model. After introducing power parameters to the H and V components of the body acceleration, total EE estimated by the nonlinear power model was comparable to the measured values by the calorimeter for the two 24-h stays (P > 0.05). The differences between the measured and estimated total EE and EEact were also significantly decreased over the Tritrac model and the linear model (P < 0.001 and < 0.01, respectively), as seen in Fig. 4. This improved accuracy was a result of the significant (P < 0.001) improvement in estimation of the sedentary and light-intensity activities for both days (Tables 4 and 5). As shown in Fig. 5, the ARD and SEE were also significantly lower for both days (P < 0.001), whereas the minute-by-minute correlation between the measured and estimated EE in the exercise day was higher (P < 0.01), as shown in Tables 2 and 3. However, because the Tritrac cannot distinguish sleeping from other activities for which the acceleration registration is zero, and we utilized the same REE formulated by the Tritrac for our models, an overestimation (8-10%) exists for sleeping. Such overestimation slightly affects the overall performance of the estimation models. Furthermore, when focusing on the walking and stepping protocol on the exercise day, the Tritrac significantly underestimated the total EE during stepping exercises (EEstep) (P < 0.001) but not during walking exercises (EEwalk) (P > 0.05). The linear model significantly reduced (P < 0.01) the estimation error of EEstep but not for EEwalk (P > 0.05). However, total EEstep estimated by using this model was still significantly lower (P < 0.05) than measured values. The nonlinear model did not significantly (P = 0.232) improve the prediction over the linear model in EEstep or EEwalk (Table 6).
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Generalized Models
Linear model. When the linear model parameters of all subjects were generalized, body mass was a significant (P < 0.05) contributing factor for both aL and bL (Eq. 3) by stepwise multiple-linear regression. Parameter aL also showed a significant correlation with height and age of the 125 subjects studied (P < 0.05). Thus a simple linear model estimated EEact by using the H and V acceleration components from the Tritrac; the parameters aL and bL were generalized
![<IT>a</IT><SUB>L</SUB> = <FR><NU><AR><R><C>[5.78 × mass (kg) + 11.95 × height (cm)</C></R><R><C>+ 6.89 × age (yr) − 2,001]</C></R></AR></NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img009.gif)
|
(7) |
![<IT>b</IT><SUB>L</SUB> = <FR><NU>[5.96 × mass (kg) + 349.5]</NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img010.gif)
|
(8) |
![<IT>p</IT>1 = <FR><NU>[2.66 × mass (kg) + 146.72]</NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img011.gif)
|
(9) |
![<IT>p</IT>2 = <FR><NU>[−3.85 × mass (kg) + 968.28]</NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img012.gif)
|
(10) |
![<IT>a</IT><SUB>N</SUB> = <FR><NU>[12.81 × mass (kg) + 843.22]</NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img013.gif)
|
(11) |
![<IT>b</IT><SUB>N</SUB> = <FR><NU><AR><R><C>[38.90 × mass (kg) − 682.44</C></R><R><C>× gender + 692.50]</C></R></AR></NU><DE>1,000</DE></FR>](/content/vol83/issue6/fulltext/2112/img014.gif)
|
(12) |
DISCUSSION
In this study, accuracy of EE estimation by the Tritrac monitor on a minute-by-minute basis and on a daily basis was evaluated by using a whole-room calorimeter. We found the intraindividual variation of the Tritrac in estimating total EE was small between 2 separate days, and the estimated total EE was correlated to measured EE. The REE, the major component of the total EE, was estimated accurately with subjects' characteristics by the Tritrac compared with the REE measured in the calorimeter in our study. We found a significant underestimation of total EEact by the Tritrac in the whole-room indirect calorimeter under conditions close to free living. The Tritrac explained ~50% of the EEact during the exercise day and 30% for the normal day. Underestimation by the Tritrac was significant in all waking activities, especially for sedentary and light intensities. Total EE was partially compensated for during sleeping because the formulated REE (Eqs. 1 and 2) was higher than actual sleeping EE. A similar result was also found with the Caltrac (3). The total EE, expressed as the superimposed EEact and REE, was underestimated by the Tritrac compared with the measured values as a net result.
Estimating EE by accelerometers has been studied previously under laboratory and free-living settings. The most commonly used device, the Caltrac, is a single-direction accelerometer that records acceleration counts from which EE is estimated. Some studies in adults and children reported significant correlations between EEact estimated by the Caltrac and by other proven accurate methods (7, 9, 32), but they also found significant underestimation of total EE (3, 11, 17, 19) by the monitor. Pambianco et al. (21) also found that the accuracy of the Caltrac changed with gender, body mass, and walking speed. With a three-dimensional accelerometer, the predicted EE was slightly improved. Ayen and Montoye (1) found that a combination of three uniaxial accelerometers mounted orthogonally could better estimate EE in walking, running, and step exercise than could any of the uniaxial accelerometers. Meijer et al. (16) reported the SEE of 1.32 W/kg for 16 subjects measured under laboratory conditions by using a triaxial portable accelerometer compared with respirometry. Bouten et al. (4) found a correlation value of 0.95 (P < 0.001, SEE = 0.70 W/kg) between EEact measured by an indirect calorimeter and estimated by a triaxial accelerometer. Matthews and Freedson (14) reported high correlation values (0.82 and 0.77, both P < 0.001) of total EE in free-living conditions as measured by the Tritrac, by 3-day physical activity log and by a 7-day recall. However, they concluded that the Tritrac still significantly underestimated EE in free-living conditions. Our calculations of the SEE (over the 24-h periods) were slightly lower than those found by other investigators (4, 16). We found the errors of estimation in EE during the awake period (Fig. 5) were close to the values previously reported.
There are some known instances in which EEact is underestimated by the Tritrac. Such activities include weight lifting, stationary bicycling, movements of upper extremities, activities involving extra carried weight, and walking uphill (19). Also, in the case of running, inaccuracy toward overestimation of EE (+68%) was found by Matthews and Freedson (14). One possible cause may be that the Tritrac predicts EE by a linear equation, which was most likely developed in level walking (14), with the use of the vector magnitude of all three axes and the subject's characteristics (mainly body mass).
We believe that the accuracy of equations that estimate EEact by the body acceleration can be improved by using the models developed in this investigation. The reasons are as follows. First, the acceleration recorded in the separate directions can contribute with various strength to enhance EEact prediction. Acceleration and deceleration of the V axis may cause more energy to be expended because of the work against gravity. This effect was shown in the underestimation of EE during the stepping protocol by using the vector magnitude of all three acceleration components (the Tritrac model) vs. the improvement by the linear models when the accelerations in V and H were separated. The V axis of body acceleration is the major component in daily activities such as walking and stepping (7, 32). Earlier studies that involved acceleration and EE had relied solely on this component for estimation, e.g., the Caltrac. But EEact could relate differently to acceleration components in other axes, depending on the type and nature of the activities. For example, Bouten et al. (4) found the best prediction of EE in walking was by using the acceleration in the anterior-posterior (x) direction. Therefore, separating the direction of acceleration components in estimation of EEact may improve the accuracy.
Second, the linear relationship between the acceleration and EEact may not be the best fitting for certain activities. Using a force-platform inside the whole-room calorimeter, we previously found a linear relationship between the EE of some exercises and the mechanical work performed (28). However, Bouten et al. (4) reported that a better estimation of EE was achieved by using the linear prediction than with the quadratic prediction using integrated acceleration for walking and other sedentary activities. We generated two nonlinear variables in the power parameters of an estimation model to optimize the EE estimation.
Third, the room calorimeter enables the subject to perform nearly free-living activities while measuring EE and accelerations accurately on minute-by-minute basis. In addition to walking at various speeds (0.6-1.2 m/s) and stepping (12-30 steps/min), >40 other activities of various intensities were performed by the subjects in the whole-room indirect calorimeter. These activities provided a wide range of daily activity types and patterns for the development of accurate models. With the accurate and quickly responding EE measurement system (28, 29), we were able to validate the Tritrac as well as to develop models on a minute-by-minute basis and in separate intensity categories to best describe the total EE on a daily basis.
The linear model was formulated by reestablishing the H and V components of body acceleration. The general model used the same parameters from the subject characteristics as did the Tritrac model and achieved slightly higher accuracy of estimating EEact over the existing model. As predicted by Eqs. 3, 7, and 8, EEact correlated positively with body mass, age, and height. Body mass was a strong determinant of converting acceleration to EE because of the mechanical work needed for motion. Body stature also contributed to the estimation of EEact, possibly because of the additional energy for the movements of upper and lower extremities; the Tritrac cannot measure these. Moreover, with increasing age, EEact for the same amount of physical activity (acceleration over time) in the H direction would also increase. This may suggest a decrease in work efficiency with age for certain activities. This hypothesis is supported by a negative correlation reported between running economy and age (22) and is consistent with the model. Nevertheless, the underestimation of EE still existed with the linear approach. One possible cause is the inherent limitation of recording acceleration and converting it linearly to EEact, especially during the sedentary and light-intensity activities.
The nonlinear model incorporated the variables in the power parameters
for the acceleration components. The noted changes in EE estimation for
the 2 days (Fig. 4) showed improvements of the nonlinear models over
the Tritrac and the linear models in total EE. Furthermore, the
nonlinear model estimated ~96 and 98% of the EEact on the
exercise and the
normal days for the group. The
generalized nonlinear model also proved to have similar accuracy in
EEact estimation. The overall reduction in the ARD and SEE suggested
that the nonlinear model was a better method. The most important
improvement of the nonlinear model was its ability to enhance the
acceleration components in sedentary to light-intensity activities (4
MET) in comparison with the linear models in which the estimation of EE
in moderate- to high-intensity activities (>4 MET) was similar
(Tables 4 and 5). In the generalized model (Eqs.
4 and 9-12),
body mass contributed differently to the power parameters
(p1 and
p2) of the H and V components. It
suggests that larger body mass tends to push the H component towards a linear relationship (p1 = 1 at
320.8 kg) and pull the V component away
(p2 = 0 at 251.5 kg) from it.
Whereas the parameters
aN and bN were
positively correlated with body mass, the slope for
bN was steeper,
perhaps to partially compensate for the effect of p2. Finally, gender contributed in
bN, showing that
women have a smaller parameter in the V component than men in EEact
estimation.
In conclusion, this investigation evaluated the Tritrac under close to
free-living conditions in a whole-room indirect calorimeter with a
large group of healthy adult subjects. We found a significant correlation of the total EE between two 24-h stays as well as between
the minute-by-minute EE as estimated by the Tritrac and measured by the
calorimeter. We also found that the Tritrac fairly accurately estimated
the REE and EE of walking. However, the underestimation of EE by the
Tritrac was significant in daily totals and in all intensities except
during sleeping. The major underestimation was in the low-intensity
activities (EE 4 MET). We developed two simple models that
incorporated each subject's body mass, height, age, and gender to
estimate EE by using the body acceleration measured and recorded by the
Tritrac. Results showed significant improvements over the Tritrac model
(1-2% total EE difference, compared with a whole-room indirect
calorimeter, SEE = 0.349-0.628 W/kg).
ACKNOWLEDGEMENTS
The authors thank Dr. C. N. Sun for his work in processing the calorimeter data, and Drs. M. Buchowski, J. Ford, and M. Nies for their editorial reviews.
FOOTNOTES
Address for reprint requests: K. Y. Chen, C2104 MCN, 21st and Garland Ave., Vanderbilt Univ., Nashville, TN 37232-2279 (E-mail: kong.chen{at}mcmail.vanderbilt.edu).
Received 30 December 1996; accepted in final form 28 July 1997.
REFERENCES
| 1. | Ayen, T. G., and H. J. Montoye. Estimation of energy expenditure with a simulated three-dimensional accelerometer. J. Ambulatory Monit. 1: 293-301, 1988. |
| 2. | Blair, S. N., H. W. Kohl III, R. S. Paffenbarger, P. G. Clark, K. H. Cooper, and L. K. Gibbons. Physical fitness and all-cause mortality: a prospective study of healthy men and women. JAMA 262: 2395-2401, 1989[Abstract]. |
| 3. | Bray, M. S., W. W. Wong, J. R. Morrow, Jr., N. F. Butte, and J. M. Pivarnik. Caltrac versus calorimeter determination of 24-h energy expenditure in female children and adolescents. Med. Sci. Sports Exerc. 26: 1524-1530, 1994[Medline]. |
| 4. | Bouten, C. V., K. R. Westerterp, M. Verduin, and J. D. Janssen. Assessment of energy expenditure for physical activity using a triaxial accelerometer. Med. Sci. Sports Exerc. 12: 1516-1523, 1994. |
| 5. | Cotes, J. E., and F. Meade. The energy expenditure and mechanical energy demand in walking. Ergonomics 3: 97-119, 1960. |
| 6. | Gretebeck, R., and H. J. Montoye. Variability of some objective measures of physical activity. Med. Sci. Sports Exerc. 24: 1167-1172, 1992[Medline]. |
| 7. | Haymes, E. M., and W. C. Byrnes. Walking and running energy expenditure estimated by Caltrac and indirect calorimetry. Med. Sci. Sports Exerc. 25: 1365-1369, 1993[Medline]. |
| 8. | Hemokinetics, Inc. Tritrac-R3D Research Ergometer Operations (rev. 3). Madison, WI: Hemokinetics, 1993. |
| 9. | Heyman, M. B., P. Fuss, V. R. Young, W. J. Evans, and S. B. Roberts. Prediction of total energy expenditure using the Caltrac activity monitor. Int. J. Obesity 15, Suppl. 1: 23, 1991. |
| 10. | Hsieh, S., G. Kline, J. Porcari, and F. I. Katch. Measurement of residual volume sitting, and lying in air and water (and during underwater weighing) and its effects on computed body density (Abstract). Med. Sci. Sports Exerc. 17: 204, 1985. |
| 11. |
Klesges, R. C.,
L. M. Klesges,
A. M. Swenson,
and
A. M. Phely.
A validation of two motion sensors in the prediction of child and adult physical activity levels.
Am. J. Epidemiol.
122:
400-410,
1985 |
| 12. | Laporte, R., H. J. Montoye, and C. Casperson. Assessment of physical activity in epidemiologic research: problems and prospects. Public Health Rep. 100: 131-146, 1985[Medline]. |
| 13. | Maliszewski, A., P. Freedson, and C. Ebbeling. Validity of the Caltrac accelerometer in estimating energy expenditure, and activity in children and adults. Pediatr. Exerc. Sci. 3: 141-151, 1991. |
| 14. | Matthews, C. E., and P. S. Freedson. Field trial of a three-dimensional activity monitor: comparison with self report. Med. Sci. Sports Exerc. 7: 1071-1078, 1995. |
| 15. | McArdle, W. D., F. I. Katch, and V. L. Katch. Measurement of human energy expenditure. In: Exercise Physiology: Energy, Nutrition, and Human Performance (3rd ed.). Philadelphia, PA: Lea & Febiger, 1991, chapt. 8, p. 145-157. |
| 16. | Meijer, G. A., K. R. Westerterp, M. H. Verhoeven, B. M. Koper, and F. Hoor. Methods to assess physical activity with special reference to motion sensors and accelerometers. IEEE Trans. Biomed. Eng. 38: 221-228, 1991[Medline]. |
| 17. | Miller, D. J., P. S. Freedson, and G. M. Kline. Comparison of activity levels using the Caltrac accelerometer and five questionnaires. Med. Sci. Sports Exerc. 26: 376-382, 1994[Medline]. |
| 18. | Montoye, H. J., H. C. G. Kemper, W. H. M. Saris, and R. A. Washburn. Movement assessment devices. In: Measuring Physical Activity and Energy Expenditure. Champaign, IL: Human Kinetics, 1995, chapt. 2, p. 6-14. |
| 19. | Montoye, H. J., R. Washburn, S. Servais, A. Ertl, J. G. Webster, and F. J. Nagle. Estimation of energy expenditure by a portable accelerometer. Med. Sci. Sports Exerc. 15: 403-407, 1983[Medline]. |
| 20. | Paffenbarger, R. S., R. T. Hyde, A. L. Wing, and C. Hsieh. Physical activity, all-cause mortality, and longevity of college alumni. N. Engl. J. Med. 314: 605-613, 1986[Abstract]. |
| 21. | Pambianco, G., R. R. Wing, and R. Robertson. Accuracy and reliability of the Caltrac accelerometer for estimating energy expenditure. Med. Sci. Sports Exerc. 22: 858-862, 1990[Medline]. |
| 22. | Pate, R. R., C. A. Macera, S. P. Bailey, W. P. Bartoli, and K. E. Powell. Physiological, anthropometric, and training correlates of running economy. Med. Sci. Sports Exerc. 24: 1128-1133, 1992[Medline]. |
| 23. | Servais, S. B., J. G. Webster, and H. J. Montoye. Estimating human energy expenditure using an accelerometer device. J. Clin. Eng. 9: 159-170, 1984. |
| 24. | Siri, W. E. Gross composition of the body. In: Advances in Biological and Medical Physics, edited by J. H. Lawrence, and C. A. Tobias. New York: Academic, 1956, vol. 4, p. 239-280. |
| 25. |
Sharp, T. A.,
G. W. Reed,
M. Sun,
N. H. Abumrad,
and
J. O. Hill.
Relationship between aerobic fitness level and daily energy expenditure in weight-stable humans.
Am. J. Physiol.
263 ((Endocrinol. Metab. 26):
E121-E128,
1992 |
| 26. |
Shulz, S.,
K. R. Westerterp,
and
K. Bruck.
Comparison of energy expenditure by the doubly labeled water technique with energy intake, heart rate, and activity recording in man.
Am. J. Clin. Nutr.
49:
1146-1154,
1989 |
