Physical activity has long been recognized as a preventative tool for many chronic health conditions. In adults, both physical activity and physical fitness are inversely related to morbidity and mortality (2). However, quantification of the relationship may be hampered by the indirect measurement of physical activity, e.g., historical recall, diaries, or questionnaires (9).
Conceptually, a possible solution for the assessment of physical activity is the use of monitors that directly measure movement (8). However, it must be recognized that activity monitors are unable to detect the metabolic cost associated with standing, upper-body movements, static work, vertical lift, and changes in gradient (1). Accelerometers measure the accelerations of movement. A time sampling mechanism allows the capture of intensity, frequency, and duration information. The Computer Science and Applications, Inc., activity monitor (CSA model 7164, also known as the actigraph or the WAM) is a small, lightweight uniaxial accelerometer, which is capable of storing activity data for up to 22 d. It has been validated against oxygen consumption during typical children’s activities (4) and been shown to be a reliable tool, with an interinstrument coefficient of variation (CV) of less than 5% and an intra-instrument CV of less than 2% (7).
The RT3 accelerometer is a small, lightweight triaxial accelerometer that stores activity data for up to 21 d. The three-dimensional measure is potentially important when assessing activity. Eston et al. (4) showed that a triaxial accelerometer (TriTrac-R3D) was a more accurate predictor of scaled oxygen uptake in children across a variety of activities than a uniaxial accelerometer. The TriTrac has been successfully validated against energy expenditure measured by indirect calorimetry in the laboratory (mean r = 0.86 over a range of lab-based activities;11) and in the field (r = 0.62;6). A limitation associated with the TriTrac-R3D is its bulky nature (120 × 65 × 22 mm, 168 g). The RT3 is much smaller (71 × 56 × 28 mm, 65.2 g) than the TriTrac-R3D and was introduced as a more researcher- and user-friendly device. To our knowledge, no studies concerning the validity of the RT3 accelerometer have been published; however, the RT3 has been successfully validated against oxygen uptake in both children and adults, over a range of regulated and nonregulated activities in our own laboratory (r = 0.87; unpublished data).
To date, no study has considered the intermonitor technical variability of the RT3 accelerometer. Therefore, this study aims to determine the intermonitor technical variability of a sample of RT3 accelerometers, on each orthogonal axis at three frequencies of motion.
METHODS
Instrumentation and Test Procedures
The RT3 accelerometer.
The RT3 (Stayhealthy, Inc., Monrovia, CA) is a small (71 × 56 × 28 mm), lightweight (65.2 g), battery-powered instrument used as an experimental tool for measuring the physical activity of people. It is worn clipped to the waistband as an “accessory” during waking hours. Depending on its mode of operation, it can record data for up to 21 d, which is then downloaded to a PC for display and statistical processing. The sensor in the RT3 is an accelerometer sensitive along three orthogonal axes (X, Y, and Z), which represent vertical, anteroposterior, and mediolateral motion, respectively. The acceleration is measured periodically, converted to a digital representation, and processed to obtain an “activity count,” which is stored in memory. The exact relationship of the activity count to the acceleration (measured in meters per second squared or g, where 1 g = 9.81 m·s−2) is not clear.
The RT3 has four modes of operation: mode 1 samples and stores activity counts on individual axes at 1-s epochs; mode 2 samples and stores vector magnitude (a measure combining all three axes of motion) activity counts at 1-s epochs; mode 3 samples and stores accumulated activity counts on individual axes over 1-min epochs; and mode 4 samples and stores accumulated vector magnitude activity counts over 1-min epochs. The latter two modes store less detail about activity but are more economical in their use of memory, allowing longer duration experiments to be performed. Epoch duration of one min is generally used in the field and so was chosen for the trials reported. In this study, the RT3 activity monitor was tested along each orthogonal axis separately; therefore, while any one of the vectors was being tested, the other two vectors should have recorded zero. As vector magnitude is a culmination of the three vectors, it was not tested in this research. Twenty-three RT3 accelerometers were tested in total, all 6 months old and previously used in the field.
Vibration
Each RT3 in turn was mounted securely in a test jig, which was screwed directly to the vibration table (Ling Dynamics 403). The vibration table was driven by a moving coil armature via a power amplifier. The amplifier input was provided by a signal generator, which was programmed accurately and reliably to vibrate the RT3 at frequencies of 2.1, 5.1, and 10.2 Hz.
There are two standard forms of industrial vibration tests, sinusoidal and random. Swept-frequency sinusoidal testing is easy to implement and usually used when the equipment being tested is subject to pulsating or oscillating forces of a periodic nature. Random testing is useful when the equipment may be subject to vibration of a stochastic nature, possibly exciting several resonant modes simultaneously. In this case, there was an advantage to the sinusoidal option. It is known the accelerometers used in the RT3 have a dynamic range of 0.05–2.00 g, are sensitive in the range 2–10 Hz, and are calibrated at 5.3 Hz. The output of the RT3 accelerometer is not available directly to the user, nor is it possible or desirable to seek these signals by breaching the encapsulation. The only output available in practice is the RT3 activity count, the fastest sampling rate being 1 Hz. According to Shannon’s sampling theorem (5), the highest frequency component of the measured variable should therefore be 0.5 Hz, if it is to be recoverable from the samples. Using the well-known relationship between the acceleration and amplitude of sinusoidal motion at a known frequency, it becomes clear that achieving 0.05 g (at the lower end of the accelerometer’s dynamic range) requires an amplitude of 49.7 mm at 0.5 Hz. This is outside the range of the Ling Dynamics 403, which has a maximum amplitude of 8.8 mm.
Therefore, the approach adopted was to vibrate the RT3 at frequencies of 2.1, 5.1, and 10.2 Hz, which are slightly offset from a multiple of the 1-Hz sampling frequency. This creates low-frequency aliases of the true vibration frequencies, which appear to have high peak accelerations at low displacements. The exact values of the accelerations are not important as long as they fall within the dynamic range of the RT3 accelerometer and are repeatable. A Kyowa AS-10B strain-gauge accelerometer (calibrated at 88.3 mV·g−1 ± 5%) was mounted on the same plate as the RT3 test jig. The sinusoidal waveform traces produced at the three frequencies and amplitudes (3.19, 2.09, and 0.99 mm, respectively) selected show that the peak test accelerations were 0.057, 0.219, and 0.414 g, respectively. When the RT3 counts are plotted against acceleration, there is a linear relationship (Fig. 1).
;)
FIGURE 1:
Variation of the RT3 count with peak acceleration (mean ± SD).
Sampling Issues
In practice, the RT3 must be switched on at some random time relative to the vibration input. It is of interest to know whether this affects the measurements. Considering sample values are given by:MATHwhere k = 0, 1, 2, 3, and 9; ωa = frequency of the alias; T = sample period; and φ = phase difference between vibration waveform and sample times, −π < φ < π.
Plotting ωa = 0.1 Hz, T = 1 s, and k = 0, 1, 2, and 3 (the remaining curves are similar) shows how the values measured at individual sample times vary with the phase, φ (Fig. 2). It is only when MATHthat the maximum value of 1 is recorded. However, even for the worst-case φ, a maximum value of 0.951 is recorded—an error of less than 5%. So the method records a sample very near to the peak acceleration applied irrespective of the time that the RT3 is actually switched on.
FIGURE 2:
Variation of phase with differing sample times. Procedures adopted mean that irrespective of the time the RT3 is switched on an error of less than 5% in the output value is apparent.
Procedures
Before each testing session, the apparatus was run for 8 min at each of the three frequencies, to ensure that the mechanics were sufficiently warmed up. Each RT3, in turn, was placed in the test jig aligned so that vibration would occur along the X axis (Fig. 3). Vibration of the test jig commenced at a frequency of 2.1 Hz. The frequency automatically increased to 5.1 Hz and subsequently to 10.2 Hz. Each frequency was maintained for 8 min. The initial and final minutes of data were discarded, leaving six complete epochs for analysis at each frequency. After each trial, the test jig was removed, rotated, and reaffixed to provide an alternative axis for vibration (Y and Z, respectively), and the procedures were repeated (Figs. 4 and 5).
FIGURE 3:
Test jig vibrating along the X axis:arrow indicates the direction of movement of the vibration table.
FIGURE 4:
Test jig vibrating along the Y axis:arrow indicates the direction of movement of the vibration table.
FIGURE 5:
Test jig vibrating along the Z axis:arrow indicates the direction of movement of the vibration table.
Statistical Analysis
Data was downloaded into Excel and imported into SPSS 9.0 for analysis. Descriptive data were calculated for all variables. A priori inclusion criteria defined outliers as activity counts greater than two standard deviations above or below the mean.
Frequency and axis effects.
A two-way fully repeated measures ANOVA was used, with repeated measures on axis (X, Y, and Z) and frequency (2.1, 5.1, and 10.2 Hz). Where the assumptions of sphericity were violated, the Greenhouse Geisser correction factor was employed. Significant results were followed up using an adapted Tukey’s test for repeated measures (10).
Interinstrument variation.
The relationship between accelerometers for each axis and for all axes combined was investigated using intraclass correlation coefficients (ICC) with a two-way random model for absolute agreement. The interinstrument coefficient of variation (CV = SD/mean) was calculated for each axis at each frequency.
Intra-instrument variation.
The intra-instrument CV was calculated for each RT3, over the six epochs available for each axis at each frequency. Alpha was set at P < 0.05.
RESULTS
Four RT3 accelerometers were identifies as outliers (two SD above or below the mean). Data were analyzed including and excluding the outliers. Overall descriptive data for the total sample of 23 RT3, and the remaining 19 RT3s are presented in Table 1.
TABLE 1:
Descriptive statistics for each axis at each frequency, including all 23 monitors and excluding the four outliers (activity counts, mean ± SD).
Frequency and axis effects.
Main effects for axis (F1.1,19.4 = 10,577.7, P < 0.001) and frequency (F1.5,26.1 = 15.6, P < 0.001) were observed, accompanied by an interaction between the two (F2.1,36.8 = 19.9, P < 0.001). Post hoc tests revealed no significant difference between the axes at 2.1 Hz. However, the Y axis counts were significantly higher than the X and Z axes at 5.1 and 10.2 Hz (Fig. 6). Inclusion of the outliers in the analysis still resulted in a frequency by axis interaction (F1.4,31.1 = 5.2, P < 0.02), with no significant difference in activity counts at 2.1 Hz. At 5.1 Hz and 10.2 Hz, the Y axis was still significantly higher than the X and Z axes; however, the Z axis was also significantly lower than the Y and X axes.
FIGURE 6:
Activity counts by axis and frequency. *Y axis significantly higher than the X and Z axes, P < 0.05.
Interinstrument variability.
The ICC across frequency for activity counts for all axes was 0.99 (F53,1025 = 41.3, P < 0.001). The ICC across frequency for the X, Y, and Z axes were 0.99 (F17,341 = 35.2, P < 0.001), 0.99 (F17,341 = 38.8, P < 0.001) and 0.99 (F17,341 = 38.6, P < 0.001), respectively. ICC did not differ when all 23 monitors were included (Table 2). Interinstrument coefficients of variation (CV) showed greatest variability at 2.1 Hz; this reduced as frequency increased regardless of whether outliers were included or not (Table 3).
TABLE 2:
Intraclass correlation coefficients across frequency for each axis, including all 23 monitors and excluding the four outliers, all significant P < 0.001.
TABLE 3:
Interinstrument coefficient of variation (CV %) for the mean activity counts at each axis and frequency, including all 23 monitors and excluding the four outliers.
Intra-instrument variability.
Intra-instrument CV showed greatest variability at 2.1 Hz along the Y axis compared with the X and Z axes regardless of whether all 23 monitors were included or not. The variability reduced as the frequency increased for all axes (Table 4).
TABLE 4:
Intrainstrument coefficient of variation (CV %) for each axis at each axis at each frequency, including all 23 monitors and excluding the four outliers, range (mean ± SD).
DISCUSSION
Without an accurate measure of physical activity, it is difficult to quantify relationships with health (3). Additionally, when determining whether a population meets the national guidelines for physical activity, it is necessary that the tool chosen to quantify physical activity is valid and reliable. The inter- and intramonitor variability of the CSA uniaxial accelerometer has been determined and found to be acceptable (7). To our knowledge, this is the first study to assess the technical variability of the RT3 accelerometer.
When assessing the variability of the output from the RT3s, it is important that a range of scores obtained in human motion is used. Data from our laboratory (unpublished) have related RT3 counts to energy expenditure (METs) in children. In this study, a frequency of 2.1 Hz (acceleration = 0.057 g) resulted in a mean activity count of 100.01 ct·min−1, which reflected activities lower than 3 METs (e.g., crayoning); at 5.1 Hz (acceleration = 0.219 g), a mean of 871.06 ct·min−1 again reflected activities lower than 3 METs (e.g., playing catch), and at 10.2 Hz (acceleration = 0.414 g) a mean of 1741.06 ct·min−1 reflected activities of 3–6 METs (e.g., walking). It must be noted that the higher frequencies used for testing are possibly out of the range of human motion, although the output amplitude is still within the range; however, frequencies of 5.1–10.2 Hz are in the dynamic capacity of the monitor and avoid the sampling issues discussed earlier in the method. Therefore, the results from the higher frequencies are relevant, as it is the technical variability of the monitor being assessed. However, in terms of human motion, the results at the 2.1-Hz frequency are perhaps most interesting.
At 2.1 Hz, the axes did not differ. At higher frequencies the Y axis consistently read higher than both the X and Z axes. On closer examination of the data, this was the case for virtually all of the RT3s tested. Whether this is also the case for other RT3 monitors needs further investigation.
ICC are impressive for the total sample and for individual axes (r = 0.99), reflecting the strong relationship between axes across all frequencies and across RT3s. The high inter- and intramonitor CV reported reflects the greater variability within and between RT3s at 2.1 Hz, compared with 5.1 and 10.2 Hz.
Although no instrumentation or test procedures are stated, Stayhealthy report data on their website (http://www.Stayhealthy.com) from the central 30 min of 40 min of testing using six randomly chosen monitors. The activity counts achieved were on average 1794.26 (±SD, ± 48.6), 1380.42 (± 34.6), and 1444.56 (± 16.7) ct·min−1 at the X, Y, and Z axes, respectively. These activity counts are comparable to the activity counts achieved at the 10.2-Hz frequency in the present study (1683.4 ± 71.3, 1847.9 ± 132.2, and 1691.9 ± 119.7 ct·min−1 for X, Y, and Z axes, respectively). However, only interaxes comparisons can be made, and no other comparison is possible between the two studies as testing procedures were not made available.
Interinstrument CV for the company data are impressive at less than 2.7%, 2.5%, and 1.2% for X, Y, and Z, respectively, in comparison with less than 5.3% for both fast and medium speeds using the CSA uniaxial accelerometer (7) and less than 7.2% at 10.2 Hz using the RT3 triaxial accelerometer. The relatively high CV obtained in the current study may be inflated by the individual vibration of accelerometers in turn compared with the “Stayhealthy” procedure of simultaneous shaking of all six RT3s. Details of the instrumentation, test procedures and frequency “Stayhealthy” used when testing the RT3s are not available. It is acknowledged a larger sample of RT3 monitors would provide more conclusive results.
Inclusion of the four outlying monitors did not change any results. ICC did not differ maintaining the strong relationship between axes across all frequencies and across all RT3s. However, in all cases, the CV was reduced postexclusion (excluding interinstrument CV on the X axis at 5.1 Hz, where there was no change).
This study did not test the robustness of the RT3 accelerometers. However, from a sample of 24 accelerometers, four failed (one due to water damage) during 6 months of use with children in physical activity research. In comparison, Metcalf et al. (7) reported that only one of the original 24 CSA uniaxial accelerometers failed during 12 months of use. As the monitors used in the present study were 6 months old, the results presented cannot be generalized to newly bought RT3 activity monitors. It must also be acknowledged that the holster (provided by the company with each RT3 monitor for the attachment to a waist band) was attached firmly to the vibrator plate and, even at the highest test frequency, can be regarded as a rigid structure (i.e., the holster was moving with the plate and not contributing some “extra” vibration of its own). However, it cannot be ruled out that there was some movement of the RT3 within its holster during the high-frequency measurement, although it was never observed to have moved at the end of a test.
In conclusion, the inter- and intra-instrument variability of the RT3 accelerometer has been assessed. The use of the activity monitor has research potential, allowing intensity, frequency, and duration of physical activity to be measured. However, inter- and intramonitor variability exists. Additionally, the Y axis reads consistently higher than the X and Z axes at 5.1 and 10.2 Hz. Therefore, it is recommended that all studies using the RT3 accelerometer perform trials to identify any outlying monitors and to assess the intermonitor variability of RT3s. This level of quality control would ensure confidence in data obtained.
Future research should assess the technical reliability of the RT3s using apparatus with a larger field of movement, generating results for more direct comparison with human movement. Additionally, accelerometers should be tested repeatedly to allow comparisons on several occasions producing measures of validity and reliability. Ideally, this could lead to the publication of a standard procedure recommending how often the RT3 activity monitors need to be tested. After controlled reliability tests, the RT3s should be tested in a lab and field-based environment on human subjects.
REFERENCES
1. Bassett, D. R., Jr. Validity and reliability issues in objective monitoring of physical activity. Res. Q. Exerc. Sport 71: 30–36, 2000.
2. Blair, S. N., H. W. Kohl, R. S. J. Paffenbarger, D. G. Clarh, K. H. Cooper, and L. W. Gibbons. Physical fitness and all-cause mortality: a prospective study of healthy men and women. JAMA 262: 2395–2401, 1989.
3. Boreham, C., and C. Riddoch. The physical activity, fitness and health of children. J. Sport Sci. 19: 915–929, 2001.
4. Eston, R. G., A. V. Rowlands, and D. K. Ingledew. Validity of heart rate, pedometry, and accelerometry for predicting the energy cost of childrenís activities. J. Appl. Physiol. 84: 362–371, 1998.
5. Franklin, G. F., D. Powell, and A. Emami-Naeini. Feedback Control of Dynamic Systems, 4th Ed. Upper Saddle River NJ: Prentice Hall, 2002, pp. 910.
6. Hendleman, D., K. Miller, C. Baggett, E. Debold, and P. Freedson. Validity of accelerometry for the assessment of moderate intensity physical activity in the field. Med. Sci. Sports Exerc. 32: 442–449, 2000.
7. Metcalf, B. S., J. S. H. Curnow, C. Evans, L. D. Voss, and T. J. Wilkin. Technical reliability of the CSA activity monitor: the EarlyBird Study. Med. Sci. Sports Exerc. 34: 1533–1537, 2002.
8. Rowlands, A. V. Field methods of assessing physical activity and energy balance. In: Kinanthropometry and Exercise Physiology Laboratory Manual: Tests, Procedures and Data, 2nd Ed., R. G. Eston and T. Reilly (Eds.). London: Routledge, 2001, pp. 151–167.
9. Rowlands, A. V., D. K. Ingledew, and R. G. Eston. The effect of type of physical activity measure on the relationship between body fatness and habitual physical activity in children: a meta-analysis. Ann. Hum. Biol. 27: 479–497, 2000.
10. Stevens, J. Applied Multivariate Statistics for the Social Sciences, 3rd Ed. Hillsdale NJ: Lawrence Erlbaum, 1996, pp. 479–480.
11. Welk, G. J., S. N. Blair, K. Wood, S. Jones, and R. W. Thompson. A comparative evaluation of three accelerometry-based physical activity monitors. Med. Sci. Sports Exerc. 32: 489–497, 2000.
Keywords:
PHYSICAL ACTIVITY; ACTIVITY ASSESSMENT; TRIAXIAL ACCELEROMETRY; ACTIVITY MONITOR
©2003The American College of Sports Medicine
