Uniaxial accelerometers quantify physical activity (PA) as changes in movement or velocity in a single axis, primarily in the vertical (V) plane. However, human motion is not limited to movements in the V plane, especially during activities of daily living (ADL), a mode of PA that contributes considerably to daily activity energy expenditure (AEE). To increase the accuracy of AEE measures, the RT3 triaxial accelerometer detects motion in three orthogonal planes (V, anterior-posterior (AP), and medial-lateral (ML)). AEE is estimated by a proprietary algorithm using the vector magnitude (VM), which is a composite of counts from these three planes of motion:
;)
Comparison of the accuracy of predicting AEE between commonly used uniaxial accelerometers and the RT3 accelerometer has produced mixed results. High correlations have been reported between RT3 predicted AEE (RT3AEE) and measured AEE during sedentary (r = 0.82) and treadmill (TM) activities (r = 0.88) (3). The RT3 either overestimates (10) or underestimates (5,10,11) AEE for ADL. Therefore, although ADL involve more complex movement patterns in all three planes of motion compared with locomotion, it is unclear whether adding the accelerometer counts in the AP and ML planes improves the accuracy of predicting AEE for ADL. However, poor accuracy in predicting AEE in previous studies may be related to the differences between the monitors used in the comparison (e.g., differences in monitor technology and related prediction equations and differences in monitor placement) (4,18). Using a single device that is capable of measuring and reporting movement in all three axes separately and simultaneously eliminates these confounding variables for a true within-unit uniaxial versus triaxial comparison. Thus, the purpose of this study was to determine whether using three planes of acceleration signals is superior to using only the V plane of the same unit for predicting AEE during locomotion and ADL. It is hypothesized that there will be no difference in AEE prediction using the VM compared with the V plane regardless of activity type.
METHODS
Subjects.
Healthy, 20- to 60-yr-old participants were recruited (N = 276) from Amherst, MA, and the surrounding communities. The study was approved by the University of Massachusetts institutional review board, and each subject gave written and verbal consent. Subjects were free from cardiovascular or metabolic diseases or physical impairments that would interfere with participation in PA and were not taking any medications that would affect metabolism. Male subjects 40 yr or older and females 50 yr or older completed a physician-supervised stress test to screen for cardiovascular disease before participation in the study. Twenty subjects had a possible positive stress test and did not complete the study. An additional 47 subjects declined participation or had missing data resulting in a final sample size of 212 subjects.
Study protocol.
Participants reported to the Exercise Physiology Laboratory after a 4-h fast. Stature to the nearest 0.25 cm and weight to the nearest 0.1 kg were measured using a floor scale/stadiometer (Detecto, Webb City, MO). Subjects rested in a supine position for 15 min in a quiet temperature-controlled room before measuring resting metabolic rate (RMR) using the MedGem metabolic analyzer (MicroLife, Dunedin, FL) (12).
After the RMR measurement, subjects completed the TM and ADL protocols in random order. The TM protocol consisted of six bouts of up to 7 min each at speeds of 1.34, 1.56, and 2.23 m·s−1 at 0% and 3% grades with 4 min of rest between bouts. The ADL protocol consisted of up to 7 min ascending and descending up to 16 flights of stairs, moving a 4.5-kg box, and two randomly selected activities from a menu of 14 household and sport-related ADL. The activities were performed according to a randomized complete block design within age by gender groups. Each subject was asked to complete the ADL "as they would in their own home" with minimal instruction to allow for individual variability in accomplishing each task.
During each activity, total energy expenditure (EE) was measured using the Oxycon Mobile (OM) portable metabolic analyzer (Cardinal Health, Yorba Linda, CA). The OM, a lightweight device worn as a backpack, is a valid and reliable metabolic oxygen consumption system for measuring respiratory gas exchange in the field (1,15). The OM was calibrated before the TM and ADL protocols. The first 2 min and last 10 s of OM data were not used to compute total EE. The remaining data were averaged to determine the total EE during the activity expressed as kilocalories per minute (kcal·min−1). In cases when the subject did not complete 7 min of an activity (e.g., stair ascent/descent), a minimum of 60 s of valid data was necessary to determine the total EE for the activity. AEE was derived from subtracting the individually measured RMR from total EE.
The RT3 is a small device that sits in a belt-clip holster measuring 7.1 × 5.6 × 2.8 cm and weighing 65.2 g. The 1.5-V battery life is capable of storing 180 min of VM data plus data in each of three planes (V, AP, and ML). The monitor was secured to the nondominant hip during the activities and was initialized to sample data in 1-s epochs. Total EE was predicted from the RT3 VM counts using a proprietary algorithm. The RT3-estimated RMR, derived from a "published prediction equation," was subtracted from the RT3-predicted total EE to compute RT3AEE.
Analysis.
Linear regression analyses were used to examine the relationships among V counts and VM counts and AEE and RT3AEE. A paired t-test using Bonferroni adjustment (P < 0.002) was used to determine whether RT3AEE − AEE was significantly different from zero for all activities combined, by activity type (TM vs ADL) and by individual activity. Similar analyses were performed after RT3AEE was corrected for measured rather than estimated RMR. The accuracy of predicting AEE was reported as the sample bias and precision calculated as the mean percent difference between RT3AEE and AEE and the SD of this difference, respectively.
RESULTS
We recruited a heterogeneous sample of men (n = 91) and women (n = 121) ranging in age from 20 to 60 yr with 36% classified as overweight or obese (body mass index (BMI) ≥ 25 kg·m−2) and 15% minorities. Subject characteristics are presented in Table 1. AEE measured from the portable metabolic analyzer and RT3AEE predicted from VM algorithm are presented in Table 2.
TABLE 1:
General subject characteristics.
TABLE 2:
Comparison of measured and predicted AEE.
For all activities combined, the difference between AEE and RT3AEE was significantly different from zero with RT3AEE underestimating AEE by 8.4% (mean ± SEE: −0.47 ± 0.06 kcal·min−1; P < 0.001). When activities were grouped by type, RT3AEE overestimated the AEE for TM by 9.0% (0.54 ± 0.05 kcal·min−1) and underestimated the AEE for ADL by 34.3% (−1.75 ± 0.11 kcal·min−1; P < 0.001). When analyzed individually, RT3AEE underestimated AEE for activities with greater upper body movement (e.g., basketball, tennis, folding laundry, and painting) by 24.4%-64.5% and overestimated locomotion-type activities with less upper body movement (e.g., level TM walking or jogging and descending the stairs) by 20.6%-55.0% (Fig. 1). There was no significant difference in RT3AEE and AEE for high-intensity TM (2.23 m·s−1 3% grade), trimming, gardening, mowing, raking, and cleaning (Bonferonni adjustment, P < 0.002), whereas the differences between RT3AEE and AEE for all remaining activities were significantly different from zero.
FIGURE 1:
Mean measured and predicted activity energy expenditure (AEE and RT3AEE, respectively) across TM and ADL of increasing intensity. The number of tests (n) for each activity is in parentheses. *Bonferroni adjustment, P < 0.002.
Analysis of the sample bias between RT3AEE and AEE revealed that the RT3AEE overestimation or underestimation of AEE was independent of activity intensity (Fig. 2). For TM, the sample bias ranged from −0.43 ± 0.06 kcal·min−1 for inclined walking (1.34 m·s−1 3% grade) to 1.91 ± 0.16 kcal·min−1 for level jogging (2.23 m·s−1; P < 0.001) with the smallest, nonsignificant difference (0.30 ± 0.15 kcal·min−1; P = 0.04) for the highest-intensity TM (2.23 m·s−1 3% grade). During the most physically demanding task, a self-paced stair ascent to a maximum of 16 flights, AEE was significantly underestimated by 68.9 ± 1.4% (−7.43 ± 0.16 kcal·min−1; P < 0.001) using the RT3 VM counts.
FIGURE 2:
Sample bias and precision for predicting AEE from VM counts across all activities. The number of tests (n) for each activity is in parentheses. (Bias was calculated as the mean difference between predicted activity energy expenditure (RT3AEE) and measured AEE, and precision was calculated as the SD of this mean difference.)
When data from all activities were included in a linear regression model, there was a strong relationship between RT3AEE and VM counts (R2 = 0.89). However, a similar relationship existed between V counts and RT3AEE (R2 = 0.83). A weak but identical relationship existed between measured AEE and either VM counts or V counts (R2 = 0.35) for all activities combined. These relationships were consistent when comparing counts for any single axis from the RT3 (V, ML, or AP) and AEE (Fig. 3). However, when grouped by activity type, the relationship between AEE and VM counts was stronger for TM activities (R2 = 0.64) compared with ADL (R2 = 0.13; Fig. 4).
FIGURE 3:
Linear regression analysis of EE versus activity counts. A, VM counts versus RT3AEE. B, Vertical activity counts (V) versus RT3AEE. C, VM versus measured AEE. D, V versus AEE. E, ML activity counts versus AEE. F, AP activity counts versus AEE. Activity counts are expressed as counts per minute, and EE is expressed as kilocalories per minute (kcal·min−1; N = 2079).
FIGURE 4:
Linear regression analysis of AEE (kcal·min−1) vs VM and V counts (counts·min−1) by activity type: TM using VM counts (A) and V counts (B) (n = 1159); ADL using VM counts (C) and V counts (D) (n = 920).
DISCUSSION
The primary goal of this study was to determine whether there was an added benefit of using the composite acceleration-based counts of three orthogonal axes (VM) to predict the energy cost of TM and ADL compared with using only the V axis acceleration-based counts from the same RT3 accelerometer. Our primary finding indicated no improvements in the accuracy of predicting AEE from VM counts compared with the single-axis V counts alone across a wide range of activity types and intensities. These results are not consistent with previous research, which suggested that the RT3 accelerometer provides a more accurate estimate of AEE than uniaxial devices (3,10,14,16,21). Rothney et al. (16) compared the RT3 to two uniaxial accelerometers (ActiGraph and Actical) for the prediction of AEE from three different regression models during ADL and TM. Using whole-room calorimetry as the criterion measure, no single regression equation consistently and accurately predicted AEE across activity type or intensity. However, the triaxial RT3 accelerometer was reportedly superior to the ActiGraph and Actical in predicting AEE. Similarly, Hendelman et al. (10) found a stronger correlation between activity counts from the Tritrac (predecessor to the RT3) accelerometer than the uniaxial CSA (predecessor to the ActiGraph) monitor. However, Welk et al. (21) found that the Tritrac accelerometer overestimated EE by as much as 101%-136% during TM and lifestyle activities.
In these and other studies, the comparison of uniaxial and triaxial devices has been complicated by differences in monitor technology, calibration, and placement on the body (4,16,18,19). Chen and Bassett (4) detailed the sensor properties of commonly used uniaxial and triaxial accelerometers. They concluded that sensor technology (e.g., cantilever beam vs integrated chip) and data processing (different sampling frequencies and band pass filtering) directly influence the sensitivity to human movement resulting in varying outcome measurements (counts per minute) between monitors. This difference in outcome measures has led to monitor specific equations for predicting energy cost of PA. The equations developed from various calibration methods are a known source of error in estimating energy cost and limit the capacity to compare the accelerometers' accuracy in assessing human movement (16). Using more than one device requires placing the monitors either bilaterally or side by side on the body. Because movement patterns of the body are not symmetrical about a given joint or bilaterally, this results in the measurement of slightly different movement patterns (18). Therefore, the use of two different devices compromises a true one-axis versus three-axes comparison.
The current study used data from a single device, the RT3, which simultaneously measured and reported acceleration-based counts in all three orthogonal planes (V, AP, and ML) of motion eliminating different monitor placement, prediction equations, and device technologies as factors contributing to differences in predicting EE. The composite of the signal from all three planes, the VM, was used to estimate AEE in a large, diverse sample during a wide variety of TM and ADL ranging in intensity from light (washing dishes and folding laundry [2.1 ± 0.3 and 2.4 ± 0.3 METs]) to very vigorous (ascending up to 16 flights of stairs [10.1 ± 1.8 METs] and TM jogging at 2.23 m·s−1 3% grade [10.3 ± 1.5 METs]) intensity. The VM-predicted AEE was accurate for 6 of the 23 different activities in this study including trimming, mowing, cleaning the room, raking, gardening, and inclined TM jogging. For all other activities, the estimates of AEE were significantly different from the criterion measure. The range in activity intensity from light to very vigorous indicates that RT3AEE accuracy was independent of activity intensity. The weak relationship among activity counts and AEE was similar for single-axis counts and the VM counts indicating no improvement in this estimation error across activity intensity with the composite three-axes measure. This suggests that other factors are contributing to this estimation error.
One of the possible sources of the error in estimating AEE may be due to upper body movements that were not detected by the activity monitor. Studies have shown that both uniaxial and triaxial accelerometers commonly underestimate AEE during ADL due to upper body movements that are not detected by monitors placed on the hip (2,4,10,20). As a result of this limitation, no single linear regression model that only uses a hip-positioned monitor will accurately estimate the AEE across a range of different lifestyle activities. This was evident in the current study when the regression model was applied separately to locomotion activities that involve little upper body movements (TM) and daily living tasks that involve greater upper body movements (ADL). A stronger relationship existed between RT3AEE and the criterion measure for TM compared with ADL. Therefore, the RT3AEE accuracy was dependent on the activity type (TM vs ADL). This is in agreement with other studies reporting that the predictive accuracy of accelerometers (uniaxial or triaxial) is much stronger for walking and jogging than for lifestyle activities (10,21). The reduced accuracy of predicting EE during activities with greater upper body movement is an inherent limitation of hip-only accelerometers and is not improved with a greater number of axes incorporated into a device. To account for this limitation of accelerometry, studies have included a second device on the wrist to measure upper body movements. However, only minor improvements in the prediction of AEE during ADL were found, accounting for an additional 2.6%-9% of the variability in EE (13,17).
Another possible source of error in estimating AEE from the RT3 may be the use of an estimated RMR rather than a measured RMR. The RT3 software calculates total EE from VM counts and subtracts an estimated RMR, calculated from a "published prediction equation," from this total to determine RT3AEE (4,5). RMR prediction equations have been shown to differ as much as 43% from measured RMR (6-9). However, adjusting RT3AEE for the subject's measured RMR in the current study resulted in a modest improvement (4%) in the AEE prediction accuracy.
The VM equation is the square root of the sum of the squares of three planes of motion. Therefore, the activity counts from the V, AP, and ML planes are equally weighted in the calculation of the VM. However, the magnitude of the V counts represents 52.6% of the magnitude of the VM counts across all the activities (range = 31.2%-68.0%). Interestingly, this difference between VM and V counts was systematic in nature, increasing with increased activity intensity (R2 = 0.65; Fig. 5). A weak relationship existed between AEE and activity counts from the V plane and the VM. The activity counts from the VM accounted for 89% of the variability within the predicted AEE. Similarly, the activity counts from the V accelerometer signal accounted for 83% of the variability in the predicted AEE. The strong relationships among VM and V counts and RT3AEE also suggested that the use of either VM counts or V counts resulted in similar errors of estimation of AEE. The similar prediction of AEE using the single-axis counts and the VM indicated that the additional accelerometer signals from the AP and ML planes did not improve the accuracy of predicting AEE in ADL.
FIGURE 5:
The activity counts (counts per minute) from the VM and V axis for all activities separated by activity type (TM vs ADL). The number of tests (n) for each activity is in parentheses (means ± SD).
Nonlinear regression models or other more sophisticated statistical modeling techniques may take into account the different movement patterns common in ADL and may increase the accuracy of predicting AEE from the VM or from each of the axes separately in nonconstrained activities. Understanding the degree of movement within each of the three planes of motion for different activities may reveal other ways to more accurately estimate AEE with the triaxial accelerometer. Developing corresponding regression equations to account for variations of movement within the three planes may reduce the error in estimating AEE from the triaxial RT3 accelerometer. Using a multiple regression model with each axis' counts as separate independent variables will weigh the VM equation according to the contribution of movement within each plane of motion and may increase the accuracy of predicting AEE. A multiple regression model using the differences in percent contribution of the three axes during ADL may also improve the accuracy of predicting AEE from the VM counts.
CONCLUSIONS
The relationship between estimated AEE was similar for VM and V counts. This suggests that a triaxial accelerometer is not superior to the uniaxial signal from the same monitor when simple linear regression modeling of the counts is used to estimate AEE. Accuracy in estimating AEE with the RT3 is moderate at best and tends to underestimate ADL more than TM activities. This error was dependent more on activity type rather than intensity suggesting a need to improve the VM equation or regression model used to accommodate complex movement patterns of ADL when converting accelerometer output to AEE. More sophisticated statistical techniques using different models are necessary to determine whether adding the additional two axes, or perhaps a combination of two axes, improves the accuracy of predicting AEE from activity monitors. Until then, this study does not support the evidence that the triaxial accelerometer is superior to the uniaxial accelerometer in the assessment of energy cost of human movement.
The authors thank the subjects who volunteered their time and energy for this study and the many graduate and undergraduate students who were essential for collecting the data. This study was supported by a National Institutes of Health RO1 research grant CA-121005.
There is no conflict of interest. The results of the present study do not constitute endorsement by ACSM.
REFERENCES
1. Attinger A, Tuller C, Souren T, Tamm M, Schindler C, Brutsche M. Feasibility of mobile cardiopulmonary exercise testing.
Swiss Med Wkly. 2006;136:13-8.
2. Bassett DR Jr, Ainsworth BE, Swartz AM, Strath SJ, O'Brien WL, King GA. Validity of four motion sensors in measuring moderate intensity physical activity.
Med Sci Sports Exerc. 2000;32(9 Suppl):S471-80.
3. Bouten CV, Westerterp KR, Verduin M, Janssen JD. Assessment of energy expenditure for physical activity using a triaxial accelerometer.
Med Sci Sports Exerc. 1994;26(12):1516-23.
4. Chen KY, Bassett DR Jr. The technology of accelerometry-based activity monitors: current and future.
Med Sci Sports Exerc. 2005;37(11 Suppl):S490-500.
5. Chen KY, Sun M. Improving energy expenditure estimation by using a triaxial accelerometer.
J Appl Physiol. 1997;83:2112-22.
6. Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review.
J Am Diet Assoc. 2005;105:775-89.
7. Frankenfield DC, Muth ER, Rowe WA. The Harris-Benedict studies of human basal metabolism: history and limitations.
J Am Diet Assoc. 1998;98:439-45.
8. Frankenfield DC, Rowe WA, Smith JS, Cooney RN. Validation of several established equations for resting metabolic rate in obese and nonobese people.
J Am Diet Assoc. 2003;103:1152-9.
9. Garrel DR, Jobin N, de Jonge LH. Should we still use the Harris and Benedict equations?
Nutr Clin Pract. 1996;11:99-103.
10. Hendelman D, Miller K, Baggett C, Debold E, Freedson P. Validity of accelerometry for the assessment of moderate intensity physical activity in the field.
Med Sci Sports Exerc. 2000;32(9 Suppl):S442-9.
11. Matthews CE, Freedson PS. Field trial of a three-dimensional activity monitor: comparison with self report.
Med Sci Sports Exerc. 1995;27(7):1071-8.
12. McDoniel SO. A systematic review on use of a handheld indirect calorimeter to assess energy needs in adults and children.
Int J Sports Nutr Exerc Metab. 2007;17:491-500.
13. Melanson EL Jr, Freedson PS. Validity of the Computer Science and Applications, Inc. (CSA) activity monitor.
Med Sci Sports Exerc. 1995;27(6):934-40.
14. Ott AE, Pate RR, Trost SG, Ward DS, Saunders R. The use of uniaxial and triaxial accelerometer to measure children's "free-play" physical activity.
Pediatr Exerc Sci. 2000;12:360-70.
15. Rietjens GJ, Kuipers H, Kester AD, Keizer HA. Validation of a computerized metabolic measurement system (Oxycon-Pro) during low and high intensity exercise.
Int J Sports Med. 2001;22:291-4.
16. Rothney MP, Schaefer EV, Neumann MM, Choi L, Chen KY. Validity of physical activity intensity predictions by ActiGraph, Actical, and RT3 accelerometers.
Obesity (Silver Spring). 2008;16:1946-52.
17. Swartz AM, Strath SJ, Bassett DR Jr, O'Brien WL, King GA, Ainsworth BE. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites.
Med Sci Sports Exerc. 2000;32(9 Suppl):S450-6.
18. Trost SG, McIver KL, Pate RR. Conducting accelerometer-based activity assessments in field-based research.
Med Sci Sports Exerc. 2005;37(11 Suppl):S531-43.
19. Ward DS, Evenson KR, Vaughn A, Rodgers AB, Troiano RP. Accelerometer use in physical activity: best practices and research recommendations.
Med Sci Sports Exerc. 2005;37(11 Suppl):S582-8.
20. Welk GJ. Principles of design and analyses for the calibration of accelerometry-based activity monitors.
Med Sci Sports Exerc. 2005;37(11 Suppl):S501-11.
21. Welk GJ, Blair SN, Wood K, Jones S, Thompson RW. A comparative evaluation of three accelerometry-based physical activity monitors.
Med Sci Sports Exerc. 2000;32(9 Suppl):S489-97.
