Share this article on:

Energy Expenditure Estimated by Accelerometry and Doubly Labeled Water: Do They Agree?


Medicine & Science in Sports & Exercise: December 2006 - Volume 38 - Issue 12 - p 2165-2172
doi: 10.1249/01.mss.0000235883.94357.95
SPECIAL COMMUNICATIONS: Methodological Advances

Purpose: The purpose of this study was to compare energy expenditure derived from regression equations determined from accelerometry with energy expenditure obtained from the doubly labeled water method (DLW).

Methods: Thirteen subjects participated in a 7-d protocol during which total daily energy expenditure (TDEE) was measured with DLW. Simultaneously, during the 7 d, subjects wore a Tritrac-R3D and an Actigraph (ACT). Pearson and concordance correlations and one-sample t-tests were used to determine the agreement of six Tritrac and eight ACT regression equations that convert body acceleration to energy expenditure with the DLW measurements.

Results: Tritrac TDEE determined from the different Tritrac regression equations under- and overestimated TDEE determined with DLW that ranged from −10 to +101%. For ACT, the percent difference between DLW and ACT-TDEE determined with the regression equation developed by Hendelman and Swartz were not statistically significantly different from zero. The mean of the difference was −2 and −4%, but the range of the difference was large for both equations, −29 to +24%. TDEE determined with the six other ACT equations were significantly different compared with DLW.

Conclusion: Of the 14 different regression equations from the literature, only two developed for ACT compared favorably with DLW; however, the difference in TDEE between these two methods was variable and rather large. These results reemphasize the difficulty in converting body movement into energy expenditure on an individual basis from accelerometry. These results imply that researchers may want to avoid using accelerometers to predict energy expenditure in free-living conditions, instead using these instruments only to measure patterns of physical activity.

1Sport and Exercise Sciences Section, School of Physical Activity and Educational Services, and 2Department of Statistics, Ohio State University, Columbus, OH

Address for correspondence: Nicole Y. Leenders, Ph.D., The Ohio State University, General Clinical Research Center, S2000 Davis Medical Center, 480 Medical Center Drive, Columbus, OH 43210-1228; E-mail:

Submitted for publication December 2005.

Accepted for publication June 2006.

Assessment of physical activity used to depend on techniques such as retrospective questionnaires, activity diaries, physiological measurements via mechanical and electronic instrumentation, and the doubly labeled water method (DLW) (16). Although these methods measure some component of physical activity movement, the majority of these methods produce data that can be converted to total daily energy expenditure (TDEE) or physical activity-related energy expenditure (PAEE). In recent years, two activity monitors, the Tritrac and the Actigraph (ACT; formerly the Computer Science and Applications Inc. (CSA)) and the Manufacturing Technology Incorporated (MTI) activity monitor have gained popularity and have been widely used to assess physical activity in a variety of research settings. The Tritrac and ACT, when attached to the body, register body accelerations in three planes and one plane, respectively. Both instruments record this information as counts per minute in 1-min intervals. This information can then be downloaded and analyzed to determine patterns of physical activity during a day or week. For the Tritrac, these acceleration counts can be converted to energy expenditure using the manufacturer's proprietary equation. The ACT is not manufactured to convert counts into energy expenditure. Using a regression equation published by Freedson and coworkers (8), ACT counts can be converted into PAEE. The ability of the Tritrac and ACT to accurately convert body movement into PAEE using this equation has been reported earlier (11). Both instruments significantly underestimated DLW-PAEE when using regression equations developed from walking and jogging. Because of these inaccurate prediction models, several researchers have attempted to improve the prediction of the metabolic cost of physical activities from Tritrac and ACT activity counts. Several regression equations were developed in a laboratory or field setting using indirect calorimetry for comparison so that activity counts could be converted into energy expenditure (8,10,12,17,23). If researchers want to use these regression equations in a free-living environment to evaluate whether physical activities produce desired energy expenditure that may have health benefits or produce desired weight losses, it is very important that these regression equations are accurate; if they are not accurate, it is very important to know these equations' limitations to decide how to use such instruments in research studies. Recently, Matthews (15) reviewed the heterogeneity in the different regression equations for the ACT. However, a systematic validation of the ACT and Tritrac regression equations with DLW, the criterion method for determining TDEE in free-living conditions, has not been reported.

The purpose of this study was to determine the accuracy of different published regression equations that convert body movement measured with the Tritrac and ACT into energy expenditure in adult women in free-living conditions, using DLW as the criterion method to measure energy expenditure.

Back to Top | Article Outline



Thirteen healthy females participated in a 7-d experimental protocol. Characteristics of subjects and protocol for the 7-d experimental period have been described in detail elsewhere (11). Subjects were required to meet the following criteria to be eligible for the study: a) normal health status and no evidence of heart disease, diabetes, hypertension, or other chronic diseases as determined by a medical history and physical examination; b) no evidence of current pregnancy; and c) normal menstrual history and no use of oral contraceptives. The subject's PA pattern had to be primarily ambulatory; that is, they could not participate in any type of weight-training, cycling, or water activities. This was purposefully chosen because the activity monitors are designed to detect ambulatory activities. Greater error would have been expected if the patients were involved, for great amounts of time, in activities such as weight lifting, cycling, and swimming. Further, they exhibited a wide range of PA, from low (no or little reported PA) to high (meeting or exceeding the current PA recommendations). After the experimental protocol was explained, each subject signed an informed consent statement approved by The Ohio State University biomedical sciences human subjects review committee.

Back to Top | Article Outline

Experimental protocol.

The experimental protocol lasted 7 d and started with an overnight stay at the general clinical research center (GCRC). Subjects consumed the determined DLW dose in the evening, and resting metabolic rate (RMR) was measured the following morning. Energy expenditure determined by DLW was assessed for the 7-d experimental period. Total 7-d energy expenditure was divided by seven to estimate daily energy expenditure. Because RMR may be influenced by the phase of the menstrual cycle, the start of the experimental phase was always within 5-10 d after the onset of menstruation.

Before the start of the 7-d period, subjects reported to the GCRC at 1600 h and were given a medical examination by a physician. At approximately 2100 h, subjects were familiarized with the metabolic measurements (see below) for a period of 15 min. Thereafter, between 2130 and 2200 h, a baseline urine sample (10 mL) was obtained from all subjects. Subjects then ingested a well-mixed solution containing a measured dose of doubly labeled water (approximately 1.5 g per kilogram of body weight). The solution contained 10% enriched H2 18O and 99.9% enriched 2H2O (Cambridge Isotope Laboratories, Cambridge, MA) mixed in a 20:1 ratio. A 1:400 dilution of the stock solution was mixed together with tap water and was stored in a 10-mL Venoject vacutainer (Benton Dickinson and Company, Rutherford, NJ). The tap water-diluted tracer was analyzed to verify enrichment. The following morning, subjects were awakened at 0530 h to void their bladder. Subjects returned to bed for the measurement of RMR. At approximately 0615 h, while the subject rested in a supine posture and was awake but motionless, a canopy was placed over the subject's head. Metabolic measurements were obtained with a Deltatrac Metabolic Monitor (MBM-100, SensorMedics Corporation, Yorba Linda, CA) for a period of 45-50 min. The metabolic system was calibrated before each test with known verified standard calibration gases. An average of 8- to 10-min intervals was used for the calculation of RMR (CV < 4.5%) using equation 12 from de Weir (26). After the measurement of RMR, at approximately 0800 h, subjects provided a urine sample that was used for isotopic analysis. Urine samples were stored in sterile vacutainers and kept on ice until transferred into 10-mL evacuated tubes. After 7 d, subjects returned to the laboratory at approximately 0800 h to provide their second voided urine sample. This sample was used for isotopic analysis. All urine samples were stored at −80°C until analysis.

Samples were analyzed in triplicate for oxygen-18 and deuterium by isotope ratio mass spectrometry at the energy metabolism research unit in the department of nutrition sciences at the University of Alabama at Birmingham. The facility uses a VG Optima isotope ratio mass spectrometer (VG Isochrom-uG. Fisons Instruments Inc. Beverly, MA) (9). Carbon dioxide production was determined using equation R2 from Speakman et al. (21). Based on each subject's macronutrient intake calculated from a 3-d dietary record, the energy equivalent of CO2 was calculated, from which the food quotient (FQ) was calculated (7). The mean FQ was then used to convert CO2 production to EE using equation 12 from de Weir (26). The thermic effect of food was estimated at 10% of TDEE (19).

During all waking hours in this 7-d period, subjects wore a Tritrac-R3D accelerometer (Professional Products Reining Int., Madison, WI) and an ACT (model 7164, Fort Walton Beach, FL). These instruments were secured at the waist and attached to a belt.

Back to Top | Article Outline

Triaxial accelerometer.

The Tritrac (11.1 × 6.7 × 3.2 cm, 170 g) is a triaxial accelerometer that has three piezoceramic elements orthogonally mounted. When the Tritrac is attached to the body, it measures acceleration in the anteroposterior (x), mediolateral (y), and vertical (z) directions and summarizes that information as a vector magnitude (Vmag). The Vmag is the square root of the sum of the squared accelerations of each direction. For the purposes of this study, only raw accelerometer data were analyzed. The Tritrac was initialized to collect data at 1-min intervals. The daily accelerometer data for the 7-d period were then summed to produce the total accelerometer data for the 7-d experimental period. The average of the 7-d period was used in the statistical analysis. TDEE was calculated by converting the Vmag data into energy expenditure using several different published equations (Table 1). The following regression equations were used for the Tritrac: Chen et al. (C1, C2), Nichols et al. (N), Hendelman et al. (H1, H2), and Leenders et al. (L). The outcome of these calculations was used to compare the Tritrac with DLW.



Back to Top | Article Outline

Uniaxial accelerometer.

The ACT activity monitor (5.1 × 4.1 × 1.5 cm, 43 g) is a uniaxial accelerometer that measures acceleration only in the vertical direction (z). The monitor is designed to detect acceleration ranging in magnitude from 0.05 to 2.0 g with a frequency response between 0.25 and 2.5 Hz. The ACT contains a microprocessor that digitizes the filtered acceleration signals, converts the signal to a numeric value, and accumulates this value as activity counts over a selected time interval. At the end of the time interval, the activity counts are stored and the accumulator resets itself to zero. The activity monitor was initialized with a time stamp and a 1-min data collection time interval. TDEE was calculated from activity counts by converting the counts-per-minute data into energy expenditure using the method suggested by Crouter et al. (6). These authors recommend using an inactivity threshold of 1 MET when counts per minute are equal to 50; 1 MET is defined as the ratio of work metabolic rate to a standard RMR of 1.0 kcal·kg−1·h−1. Although this is a widely used value, it has been demonstrated that it can overestimate resting energy expenditure by 20% (2). Therefore, in the current study, each individual's MET level was used when counts per minute were equal to 50. If counts per minute were greater than 50, the following regression equations were used for the ACT: Freedson et al. (F), N, H1, H2, Swartz et al. (SW), L, and Yngve et al. (Y1, Y2) (Table 1). The outcome of these calculations was used to compare the ACT with DLW.

Back to Top | Article Outline

Collection of accelerometry data.

The Tritrac and ACT were placed either in a pouch (15.5 × 12.5 × 4 cm, Eagle Creek, San Marcos, CA) or nylon pocket (supplied by the company). These were attached to a belt that was worn by the subjects to minimize the recording of extraneous movements not caused by PA. The Tritrac and ACT were positioned over the right and left hips at the midaxial location. The hip location was selected for several reasons: 1) all 14 published regression equations were developed with the accelerometers placed on the hip; 2) it allows for measurement of whole-body movement; 3) it is a convenient position because it does not interfere with daily activities; 4) placement of the activity monitor does not have a significant impact on the prediction of time spent in moderate- and vigorous-intensity physical activity in the field (27); 5) it is the most commonly used location in physical activity and health-related research; and 6) the differences in accelerometer output between placement on the lower back and the hip are small, and the practical significance is considered questionable (25). Subjects were instructed on the precise positioning of the Tritrac and ACT and were instructed to ensure this location was used during the experimental period. Subjects returned to the laboratory after 7 d to return the Tritrac and ACT.

Back to Top | Article Outline

Data analysis.

Descriptive statistics were used to summarize the data. The output from DLW, the Tritrac, and ACT are expressed as TDEE (kcal·d−1). The following regression equations were used for the Tritrac: C1, C2, N, H1, H2, and L. The following method was used to calculate TDEE from the ACT. When counts per minute were 50 or below, the individual's MET level was used. For the ACT, the following regression equations were used: F, N, H1, H2, SW, L, Y1, and Y2) (Table 1).Data were analyzed in the following ways:

1) Pearson product-moment correlation (r) was used to determine the relationships among DLW and the Tritrac and ACT regression equations.

2) A concordance correlation coefficient (ρ) was calculated to determine the agreement between DLW and the Tritrac equations and between DLW and the ACT equations that convert activity counts to EE (13). This correlation evaluates the extent to which DLW-TDEE and Tritrac-TDEE and DLW-TDEE and ACT-TDEE agree. The concordance correlation evaluates the degree to which pairs of estimates of TDEE (e.g., DLW-TDEE and Tritrac-TDEE) fall on the 45° line (or perfect line of agreement). The concordance correlation coefficient (ρ) incorporates the measurements of accuracy and precision. A departure from this line produces a concordance correlation coefficient that is less than one (ρ < 1). The formula used to calculate this coefficient was

where r = Pearson correlation coefficient, S1 = DLW, S2 = St.Dev activity monitor, M1 = mean of DLW, and M2 = mean of activity monitor.

3) Whether the percent difference between TDEE and DLW was different from zero was determined with a one-sample t-test.

4) The differences between the Tritrac and ACT equations and DLW were plotted against DLW values. Statistical significance was set a priori at P ≤ 0.05. The SAS-JMP (SAS Institute Inc., Cary, NC) statistical package was used to analyze the data. Data are presented as mean ± SEM.

Back to Top | Article Outline


Descriptive results.

Descriptive characteristics for the subjects are presented in Table 2. Subjects wore the monitors on average 13.5 ± 0.2 h·d−1, which was approximately 75-85% of their waking hours. For two subjects, activity counts were not available for the whole 7-d period because of technical difficulties with the data file for one subject; the other of these subjects wore the monitor only for 3 d. Therefore, data from the ACT are presented for 11 subjects.



The 7-d average for TDEE from DLW, Tritrac, and ACT using the different regression equations is presented in Table 3.



Back to Top | Article Outline

Tritrac regression equations.

The Pearson correlation coefficients between TDEE-DLW and TDEE calculated from C1, C2, H2, L, and N ranged from 0.51 to 0.67, which were all statistically significant (Table 4). Despite these significant correlations, none approached a perfect correlation of r = 1. This is further illustrated with the concordance coefficients between TDEE-DLW and C1, C2, H1, H2, N, and L, which were even lower at ρ = 0.34, ρ = 0.46, ρ = 0.42, ρ = 0.04, ρ = 0.45, and ρ = 0.38, respectively. These are much below 0.75, which is a generally accepted threshold for good concordance. TDEE calculated from the regression equations from C1, H1, and N underestimated DLW, ranging from −8 to −23%; the regression equations from C2, H2, and L overestimated DLW, ranging from +12 to + 101% (Table 3). All values were significantly different from zero. The standard deviations of the differences were also large, ranging from 10 to 31%. Using the less powerful but robust nonparametric test, the Wilcoxon signed rank test, led to the same conclusion, with similar P values.



Back to Top | Article Outline

ACT regression equations.

The percent difference between DLW and ACT-TDEE determined with the regression equation developed by Hendelman and Swartz were not statistically significantly different from zero (P = 0.36 and P = 0.23). The mean of the difference between DLW-TDEE and TDEE-H2 was −2 ± 5%, and the range of the difference was −24 to 23%. The mean of the difference between DLW-TDEE and TDEE-SW was −4 ± 5%, and the range of the difference was −29 to 24%. The percent difference for TDEE determined with the six other ACT equations and DLW were all significantly different from zero. Use of the less powerful test, the Wilcoxon signed rank nonparametric test, led to the same conclusion with similar P values.

The Pearson correlation coefficients between TDEE-DLW and TDEE estimated from ACT using the regression equations F, H1, H2, SW, N, L, Y1, and Y2 ranged from 0.41 to 0.67 and were statistically significant (Table 4). Only the correlation between TDEE-DLW and TDEE estimated from SW was not statistically significant (P = 0.21). However, none of them approached a perfect correlation of r = 1, and the best R2 achieved was under 50%. Illustrating the less than perfect agreement between TDEE-DLW and the ACT regression equations, the concordance correlations between TDEE-DLW and F, H1, H2, SW, N, L, Y1, and Y2 ranged from 0.11 to 0.50 (Table 4, Fig. 1). These are much below 0.75, the generally accepted threshold for good concordance.

Figure 1

Figure 1

Back to Top | Article Outline


From the published literature, six different regression equations for the Tritrac and eight different regression equations for the ACT that convert body acceleration into energy expenditure were compared for their ability to predict the energy expenditure derived with DLW. It is important to understand the accuracy of these different regression equations when one is deciding how to use this type of instrumentation in studies designed to measure TDEE or to measure changes in TDEE using physical activity behavior-change strategies.

Four different statistical approaches based on the following questions were used to determine the accuracy of the 14 regression equations. 1) How well do the TDEE values correlate (Pearson correlation coefficient)? 2) How close were the TDEE values to a perfect agreement (concordance correlation coefficient)? 3) Was the mean percent difference between TDEE and DLW different from zero? 4) How can the individual differences between Tritrac and ACT equations and DLW measurements be visualized? (Fig. 2).

Figure 2-Diffe

Figure 2-Diffe

The concordance correlation (ρ), given in equation (1), is a multiple of Pearson correlation (r). The multiplier there does not exceed 1, and it equals 1 only when M1 = M2 and S1 = S2; that is, when the two methods show equality of sample measures of centrality (accuracy) and spread (precision) for the paired data. Any imbalance in these measures' results shrinks the concordance correlation towards 0 (5). This is illustrated in Table 4, where pairs with similar r values exhibit marked variation in ρ; in fact, the method that produces the highest Pearson correlation has the second lowest concordance correlation.

The results demonstrate that compared with measurement of energy expenditure using the criterion method, DLW and using an inactivity threshold combined with the regression equations from H2 and SW (ACT only) produced results that were within acceptable statistical limits (P = 0.05) for some but not all of the statistical analyses. The H2 and SW equations demonstrated reasonable agreement between DLW (2381 kcal·d−1) and ACT-TDEE (H2: 2294 kcal·d−1; SW: 2238) with a mean difference of −2 and −4%, respectively. Both percentages were not significantly different from zero. However, for both equations, the range in the differences between the DLW-TDEE and the ACT-TDEE were large: −24 to 23% (1121 kcal·d−1; Fig. 2)) and −29 to 24% (1293 kcal·d−1), and the standard deviation of the differences was 15 and 17%, respectively. The less than perfect agreements between TDEE-DLW and the H2 and SW equations are also demonstrated, with moderate concordance correlations of 0.495 and 0.35 and correlation coefficients of 0.556 and 0.407, respectively. These moderate values were found even though our subjects exhibited a wide range of PA, from low (no or little reported PA) to high (meeting or exceeding the current PA recommendations) and thus a wide range of EE (1665-2814 kcal·d−1). One may speculate that even lower correlation coefficients would have been found with a less wide range of EE. For each of these equations, it indicates that large errors can be expected when these ACT equations are being used to estimate TDEE on an individual basis.

Traditionally, the relationship between bodily movement measured with the Tritrac or ACT and EE derived from IC were determined during laboratory protocols using level treadmill walking and running as the mode of physical activity. In general, the R2 ranged between 0.81 and 0.93 between the Tritrac and EE when walking and running between 2 and 10 METs (12). For the ACT, when using level treadmill walking and running, the R2 for EE ranged between 0.74 and 0.89, and a R2 of 0.59 was found during outdoor walking between 2 and 5 METs was used (Table 1). In recent years, however, field studies have been conducted to determine the relationship between EE resulting from bodily movement and activity counts. In a field validation study, Hendelman et al. (10) reported an R2 of 0.78 between Tritrac and EE and an R2 of 0.59 between ACT and EE when walking between 2 and 5 METs. When the same subjects performed lifestyle activities at intensities between 2 and 5 METs, only 39% of the variance in EE could be explained by Tritrac, and only 32% of the variance in EE could be explained by ACT (10). Similarly, Swartz et al. (23) had subjects perform lifestyle activities between 1.5 and 8.0 METS and reported that only 32% of the variance in EE could be explained by using ACT. Despite the lower percentage of variance explained in EE in a laboratory protocol using a variety of lifestyle activities, the results of the present study indicate that these regression equations, in combination with using an inactivity threshold, estimate TDEE in a free-living environment better than the equations that are developed using walking and/or running as the only activity mode.

Surprisingly, the equations developed by Chen et al. (4) did not seem to produce an accurate agreement with DLW. These investigators used a whole-body indirect calorimeter and had 125 subjects perform a variety of activities while wearing a Tritrac. The subjects stayed in the calorimeter two times for 24 h. From the accelerometer data collected by the Tritrac, a linear and nonlinear model produced regression equations that estimated EE. With this approach, real free-living conditions were closely duplicated. Therefore, it may be speculated that the small sample size in our study may have prevented a closer agreement between the regression equations developed with this study and DLW or the fact that the subjects did not wear their accelerometers for a 24-h period (see below). Further, the use of an inactivity threshold as suggested by Crouter et al., (6) could not be used for the Tritrac because one has not been established for the Tritrac.

Part of the discrepancy between the EE estimated with the regression equations and DLW might also be related to the fact that for activities that involve static exercise (carrying a load, bicycling, or walking into a head wind), accelerometer output is not proportional to the increase in energy expenditure. None of the subjects participated in any type of weight lifting or bicycling, but in a free-living environment, carrying loads such as book bags or groceries might contribute to the underestimation of PAEE (3). When climbing stairs, the body is not accelerated in proportion to energy expenditure, and measurements of body accelerations will underestimate energy expenditure (20). Earlier research demonstrated that accelerometers do not detect the increase in energy expenditure associated with an increase in the grade of walking. The amount of stair climbing by subjects in this study was not assessed, and this may contribute to any amount of underestimation. Compliance of wearing the activity monitors in this study was good. From the subjects' log, it was observed that the monitors were worn during the majority of the day (≥ 13 h; 75-80% of waking hours), and recording for 10 h·d−1 can be considered a representative activity-recording day (14). However, there was always some discrepancy between sleeping and waking hours and time that the monitors were worn. Thus, during some of the time the subjects were awake and moved, there was no registration of body movement, and therefore their own MET was calculated for that minute of data collection. It is likely that this produced some amount of underestimation of EE. For example, from their physical activity recall data (11), it was determined that the subjects slept, on average, 8 h·d−1. Thus, if a subject performed some type of activity during the daily 2-h period that the monitors were not worn, EE could have been underestimated by approximately 100-150 kcal·d−1. The only way this could be controlled for however, is if subjects were known to wear the accelerometers all day or perhaps if the monitors were somehow permanently attached to the subjects. Because fidgeting can also contribute to spontaneous physical activity and, therefore, to TDEE, it is likely that this movement and the associated energy expenditure are not detected by an accelerometer worn at the waist (18). Thus, these accelerometers do not accurately detect all movement-associated energy expenditure and also do not accurately estimate energy expenditure using published regression equations on an individual basis.

Whether a two-regression approach based on the observed CV among 10-s accelerometer counts to estimate TDEE would have improved the agreement between TDEE-DLW and TDEE estimated from H2 and SW could not be determined (6). However, using an inactivity threshold in combination with the H2 and SW regression equations developed using lifestyle activities may be useful when interested in change of TDEE over time. A similar approach needs to be established for the Tritrac. Future studies using DLW and accelerometry should be performed to confirm how well the two-regression approach described by Crouter et al. (6) can predict TDEE in free-living individuals (24).

Back to Top | Article Outline


The difficulty of predicting EE from activity counts measured with the Tritrac and ACT is already demonstrated by the fact that 14 different prediction models have been published in the literature. The results of the present study demonstrated that on the basis of four statistical analyses, only the equations developed by Hendelman and Swartz using daily life activities predicted TDEE on a group basis in free-living conditions reasonably well. Individual differences, however, were large. Therefore, these results confirm that extrapolating EE estimated from regression equations developed in a laboratory or a field situation to free-living conditions remains difficult.

The Tritrac and ACT can store data for a couple of weeks and can provide information about PA patterns, a feature not available with DLW. Thus, in studies where PA patterns are important, the accelerometers may be useful to determine patterns of those movements; if a prediction of TDEE is desirable, based on the current study for the ACT, the equations by Hendelman or Swartz are recommended for use. However, the results of this study and others (22) indicate that energy expenditure should not be predicted on an individual basis using the Tritrac or ACT accelerometers unless one is interested in change over time.

The authors would like to thank Dr. Carlijn Bouten, Dr. Andrea Dunn, Dr. Diane Habash, and Dr. Farah Ramirez-Marrero for their suggestions. We thank Dr. M. Goran and the late H. Vaugh for interactions and technical assistance related to the doubly labeled water measurements when they were at the University of Alabama at Birmingham. We also like to thank Dr. Stewart Trost for his thoughtful suggestion for writing this paper and Jason Winnick for the useful feedback and discussions while writing this manuscript.

Thanks to Honey Marsh, Tim Nelson, Stephanie Pigman, and Jon Kepner for help with data collection and analysis and to all the subjects who were willing to wear the activity monitors. Thanks are also due to The Ohio State University Hospital General Clinical Research Center for the use of their facilities and to the nursing staff.

This study was partially funded by the School of Physical Activity and Educational Services, Graduate Student Alumni Research Award at The Ohio State University, a student research grant from the Gatorade Sports Science Institute, and the General Clinical Research Center grant M01-RR-00034 from the National Center of Research Resources of the National Institutes of Health.

The results of the present study do not constitute endorsement of the product by the authors or ACSM.

Back to Top | Article Outline


1. Bassett, D. R., JR., B. E. Ainsworth, A. M. Swartz, S. J. Strath, W. L. O'Brien, and G. A. King. Validity of four motion sensors in measuring moderate intensity physical activity. Med. Sci. Sports Exerc. 32:S471-S480, 2000.
2. 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. 26:1516-1523, 1994.
3. Byrne, N. M., A. P. Hills, G. R. Hunter, R. L. Weinsier, and Y. Schutz. Metabolic equivalent: one size does not fit all. J. Appl. Physiol. 99:112-119, 2005.
4. Chen, K. Y., and M. Sun. Improving energy expenditure estimation by using a triaxial accelerometer. J. Appl. Physiol. 83:2112-2122, 1997.
5. Choudhary, P. K., and H. N. Nagaraja. Measuring agreement in method comparison studies - a review. In: Advances in Ranking and Selection, Multiple Comparisons, and Reliability, N. Balakrishnan, N. Kannan, and H. N. Nagaraja (Eds.). Boston, MA: Birkhauser, 2005, pp. 215-244.
6. Crouter, S. E., K. G. Clowers, and D. R. Bassett, JR. A novel method for using accelerometer data to predict energy expenditure. J. Appl. Physiol. 100:1324-1331, 2006.
7. Elia, M. Energy equivalents of CO2 and their importance assessing energy expenditure when using tracer techniques. Am. J. Physiol. 260:E75-E88, 1991.
8. Freedson, P., E. Melanson, and J. Sirard. Calibration of the Computer Science and Applications Inc., accelerometer. Med. Sci. Sports Exerc. 30:777-781, 1998.
9. Goran, M. I., W. I. Carpenter, A. McGloin, R. Johnson, M. Hardin, and R. L. Weinsier. Energy expenditure in children of lean and obese parents. Am. J. Physiol. 268:E917-E924, 1995.
10. Hendelman, D., K. Miller, C. Bagget, E. Debold, and P. S. Freedson. Validity of accelerometry for the assessment of moderate intensity physical activity in the field. Med. Sci. Sports Exerc. 32:S442-S449, 2000.
11. Leenders, N. Y. J. M., W. M. Sherman, H. N. Nagaraja, and C. L. Kien. Evaluation of methods to assess physical activity in free-living conditions. Med. Sci. Sports Exerc. 33:1233-1240, 2001.
12. Leenders, N. Y. J. M, T. E. Nelson, and W. M. Sherman. Ability of different physical activity monitors to detect movement during treadmill walking. Int. J. Sports Med. 24:43-50, 2003.
13. Lin, L. I. A concordance correlation coefficient to evaluate reproducibility. Biometrics 45:255-268, 1989.
14. Masse, L. C., B. F. Fuemmeler, C. B. Anderson, et al. Accelerometer data reduction: a comparison of four reduction algorithms on select outcome variables. Med. Sci. Sports Exerc. 37:S544-S554, 2005.
15. Matthews, C. E. Calibration of accelerometer output for adults. Med. Sci. Sports Exerc. 37:S512-S522, 2005.
16. Montoye, H. J., H. C. G. Kemper, W. H. M. Saris, and R. A. Washburn. Measuring Physical Activity and Energy Expenditure. Champaign, IL: Human Kinetics, 1996, pp. 17-25.
17. Nichols, J. F., C. J. Morgan, J. A. Sarkin, J. F. Sallis, and K. J. Calfas. Validity, reliability, and calibration of the Tritrac accelerometer as a measure for physical activity. Med. Sci. Sports Exerc. 31:908-912, 1999.
18. Ravussin, E., S. Lillioja, T. E. Anderson, L. Christin, and C. Bogardus. Determinants of 24-hour energy expenditure in man. Methods and results using a respiratory chamber. J. Clin. Inv. 78:1568-1578, 1986.
19. Schutz, Y., T. Bessard, and E. Jequier. Diet-induced thermogenesis measured over a whole day in obese and non-obese women. Am. J. Clin. Nutr. 40:542-552, 1984.
20. 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.
21. Speakman, J. R., K. Speekumaran Nair, and M. I. Goran. Revised equations for calculating CO2 production from doubly labeled water in humans. Am. J. Physiol. 264:E912-E917, 1993.
22. Strah, S. J, D. R. Bassett, JR., and A. M. Swartz. Comparison of MTI accelerometer cut-points for predicting time spent in physical activity. Int. J. Sports Med. 24:298-303, 2004.
23. Swartz, A. M., S. J. Strah, D. R. Bassett, JR., W. L. O'Brien, G. A. King, and B. E. Ainsworth. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites. Med. Sci. Sports Exerc. 32:S450-S456, 2000.
24. Troiano, R. P. Translating accelerometer counts into energy expenditure. Invited editorial. J. Appl. Phys. 100:1107-1108, 2006.
25. Trost, S. G., K. L. Mciver, and R. R. Pate. Conducting accelerometer-based activity assessments in field-based research. Med. Sci. Sports Exerc. 37:S531-S543, 2005.
26. Weir, J. B. New methods for calculating metabolic rate with special reference to protein metabolism. J. Physiol. 109:1-9, 1949.
27. Yngve, A., A. Nilsson, M. Sjostrom, and U. Ekelund. Effect of monitor placement and of activity setting on the MTI accelerometer output. Med. Sci. Sports Exerc. 35:320-326, 2003.


©2006The American College of Sports Medicine