INTRODUCTION

Top Abstract Introduction Methods Results Discussion References

TO OVERCOME THE PROBLEM of unreliable self-report in children (3, 31), heart rate is commonly employed as an objective method of assessing children's physical activity. Heart rate does not measure physical activity directly but is based on the linear relationship between oxygen uptake and heart rate. The widespread use of heart rate monitoring is due to its ease of measurement, its ability to record values over time, and its reflection of the relative stress placed on the cardiopulmonary system due to physical activity (34). However, heart rate can also be elevated by emotional stress, which is independent of any change in oxygen uptake. The return of heart rate to baseline may also lag behind the return of oxygen uptake to baseline (24). Additionally, the heart rate-oxygen uptake relationship is moderated by the proportion of active muscle mass and whether the activity is continuous or intermittent (17).

Children's physical activity is highly transitory (2, 24). The relative delay in heart rate response to changes in movement suggests that heart rate monitors may mask potential information. The physical fitness levels of children are also a limiting factor when heart rate monitoring is used to assess physical activity. A fitter child has a higher stroke volume and, hence, a lower heart rate for any given activity (27). Mean daily heart rates may therefore be more representative of children's fitness than their activity level (27).

Despite these weaknesses, heart rate monitoring is commonly used to validate commercial accelerometers (14, 33). This is due to the lack of an adequate alternative criterion measure. Doubly labeled water is considered the gold standard for the assessment of energy expenditure in the field (21), but it is often inappropriate because of its high cost. In addition, information is limited to total energy expenditure, with no frequency, intensity, or duration information.

Conceptually, the use of activity monitors offers the ideal solution, particularly the new generation of uniaxial and triaxial accelerometers that facilitate temporal tracking of the frequency, intensity, and duration of activity. Studies of uniaxial and triaxial accelerometry have elicited encouraging results. However, the majority of studies have used indirect criterion measures, such as heart rate (14, 18, 33), or have restricted the activities to regulated walking and running on a treadmill (20). Few studies have validated accelerometry during a range of activities against the criterion of energy expenditure, and those studies have used adults as subjects (7, 19).

Laboratory studies elicit higher validity coefficients than field studies. This observation is also true for studies using adults compared with children (14). Children engage in a greater variety of movement than adults. Hence, whereas typical adult activities and laboratory activities (walking/running) may be adequately assessed by a uniaxial accelerometer (15), a triaxial accelerometer may be more sensitive to the increased range of movement in children. Some support for this concept is provided by Welk and Corbin (33), who obtained correlations of r = 0.46-0.74 (mean r = 0.60) between average vector magnitude from a triaxial accelerometer and average heart rate (corrected for resting values) compared with correlations of r = 0.51-0.69 (mean r = 0.57) obtained by Janz (14) with a uniaxial accelerometer. Both studies utilized 3 days of monitoring, but Welk and Corbin spread the days over an 8-mo period, whereas Janz assessed activity over 3 consecutive days. Although sample size was similar in the two studies, there was a much greater age range in the study by Janz (7-15 yr) than in the study by Welk and Corbin (9-11 yr), which limits the comparability between the studies.

Triaxial accelerometry appears to be more accurate for predicting energy expenditure (oxygen uptake) in adults undertaking a variety of activities than any dimension on its own (7). Bouten et al. (7) observed that the dominant dimension for predicting energy expenditure changed according to the activity being performed. For example, walking was predicted within 4% accuracy using only the anteroposterior component, but this dimension underestimated sedentary activities, on average, by >60% (7). When the vector magnitude of the three dimensions was used as the predictor, the energy expenditure of sedentary activities and walking was predicted with an accuracy of ~15%. Triaxial accelerometry has been reported to predict energy expenditure during low-level activities with greater accuracy than heart rate (19) and is highly associated with energy intake (r = 0.99, P < 0.025) during 1 wk under free-living conditions in adults (19). However, in the latter study, heart rate monitoring and triaxial accelerometry predicted free-living energy expenditure as 30% higher than energy intake, although higher individual differences were present when heart rate measurement was used.

Studies on mechanical pedometers have generally concluded that they are inaccurate at counting steps or measuring distance walked (10, 16, 25, 32). However, the newer, commercially available electronic pedometers provide a reasonably accurate estimate of distance walked and number of steps taken (4). Bassett et al. (4) observed that the Yamax DW-500 pedometer was the most accurate, recording 100.7 and 100.6% of steps taken on the left and right foot, respectively.

Pedometers, therefore, show great potential for assessing daily activity. Sequeira et al. (29) demonstrated that the pedometer could differentiate between various levels of occupational activity in adults (sitting, standing, and moderate-effort occupational categories). However, heavy work did not differ from moderate work. The heavy work category contained a high proportion of static work, such as lifting heavy objects. Pedometers are unable to measure physical activity of this type and, hence, underestimated the energy cost of individuals in this occupational category. The contribution of static work to total daily energy expenditure was observed to be trivial in adults (19). In children, the contribution of static work to a day's energy expenditure is likely to be less than in adults, so the inability of the pedometer to measure this type of work is not a cause for concern. Pedometer readings from 4- to 6-yr-old children correlated highly with observation (26). The pedometer differentiated between the most and the least active children as predetermined by supervisor's questionnaire (P < 0.001) and confirmed by observation (P < 0.01). The validity of the pedometer as a measure of habitual activity needs to be tested using more stringent criteria. If validity is confirmed, the pedometer would be particularly suited to population studies, inasmuch as it is inexpensive, reusable, and objective.

The purpose of this study was to compare the accuracy of heart rate monitoring, triaxial accelerometry, uniaxial accelerometry, and pedometry to estimate oxygen uptake (energy expenditure) during a number of representative childhood activities (walking, running, hopping, catching, and sitting and crayoning), when used in isolation and in combination with one other.

	METHODS

Top Abstract Introduction Methods Results Discussion References

Subjects

Procedure

Each child was habituated to the Powerjog treadmill for 5 min. The child then walked at 4 and 6 km/h and ran at 8 and 10 km/h for 4 min at each speed. After a rest period to allow heart rate to return to resting levels, the subject played hopscotch for 4 min. Hopscotch involved alternately hopping and jumping on a hopscotch grid at the subject's preferred pace (knowing they would be required to continue for 4 min). The subject rested again to allow heart rate to return to baseline, then played catch with an assistant. A soccer ball was thrown between the assistant and the child (~3 m). The child selected the pace, and the rhythm was maintained for the 4 min. The order of the catch and hopscotch activities was interchanged between subjects. The subject rested again to allow heart rate to return to baseline. After resting heart rate was reached, the child sat and crayoned for 10 min. Before each activity started the three electronic pedometers were reset to zero, and at the end of each activity the total number of counts was recorded for each activity. Counts per minute were then calculated.

Instrumentation

Pedometry. A commercially available electronic pedometer (Digiwalker DW-200, Yamax, Tokyo, Japan) was used. This unit measures vertical oscillations, providing a total count of the accumulated movements. Units were firmly secured to the ankle and wrist using Velcro strips. A third unit was attached to a belt worn by the subject, with the pedometer positioned on the left side of the body.

Uniaxial accelerometry. The WAM accelerometer (model 7164, Computer Science Applications, Shalimar, FL) is a very small (5 × 4 × 1.5 cm), lightweight (43 g) unit with a time-sampling mechanism that allows it to provide a chronological measure of frequency, intensity, and duration of movement. It allows data to be analyzed over user-defined intervals (ranging from 1 s to several minutes). In this study, epoch duration was set at 1 min. This epoch was selected, as this was the epoch duration that would most likely be used in field-based studies, allowing data to be collected for up to 22 days with no download. The unit was stored in the pouch supplied, which allowed it to be threaded onto the belt worn by the subject. The WAM was positioned above the left hip.

Triaxial accelerometry. The Tritrac-R3D accelerometer (model T303, version 6.0, Professional Products, Reining, Madison, WI) has the same time-sampling ability as the WAM but, in addition, assesses activity in three dimensions, giving output measures in mediolateral (x), anteroposterior (y), and vertical (z) dimensions, as well as the vector magnitude. The Tritrac is bulkier than the WAM (11.1 × 6.7 × 3.2 cm, 170 g). The sampling intervals for the Tritrac are 1-15 min, with a maximum of 14 days of data collection when the epoch interval is set at 1 min. The unit was programmed with the subject's age, height, mass, and gender and set to collect data every minute. This would ensure that the WAM and Tritrac were directly comparable. It was necessary to tape the unit securely to the belt worn by the subject to prevent any extraneous movement. The Tritrac was positioned above the right hip.

Heart rate telemetry. The heart rate monitor (BHL 6000 Medical, Fleurier, Belgium) was attached with small adhesive electrodes to minimize the discomfort to the child and improve accuracy of the recording. It was necessary to add adhesive tape to ensure that the electrodes did not slip. The heart rate monitor was programmed to record the average of every eight heartbeats.

Gas analysis. Oxygen uptake was measured by on-line gas analysis every 30 s during each activity (Biokinetics, Bangor, UK). This system involves a very lightweight, low-resistance mouthpiece from which samples of the expired air are drawn by 4-mm tubing and breath-by-breath volumes are monitored continuously.

Data Analysis

The output from the Tritrac was analyzed for each of the three dimensions (x, y, and z) and a three-dimensional vector [Tritrac_xyz = (x² + y² + z²)^0.5] representation of the count to determine whether the assessment of three dimensions offered any advantage over the assessment of any one dimension.

Descriptive statistics were calculated for all the output measures. Pearson product-moment correlations (r) were used to assess the concurrent validity of the pedometry, WAM, Tritrac, and heart rate monitoring procedures (N being the number of subjects times the number of activities). Simple linear regression equations were computed to predict oxygen uptake from each measure. Multiple regression equations were used to predict oxygen uptake from pairs of measures. An  level of 0.05 was used for all statistical tests. Where stated,  was adjusted to account for multiple tests of significance.

	RESULTS

Top Abstract Introduction Methods Results Discussion References

Values for Tritrac vector magnitude (Tritrac_xyz), Tritrac_x, Tritrac_y, Tritrac_z, WAM, heart rate, hip pedometer, ankle pedometer, wrist pedometer, and sO₂ during each of the activities are given in Table 1.

Each activity measure correlated significantly (P < 0.001) with sO₂ and heart rate (Table 2), with the exception of the wrist pedometer when sO₂ was predicted for treadmill activities alone. Correlations with sO₂ were consistently higher than corresponding correlations with heart rate. In addition, when heart rate was used as the criterion, a different trend in scores emerged, with Tritrac_x showing the best relationship. This was the weakest of the Tritrac variables when sO₂ was the criterion measure. There was no significant relationship between sO₂ and body mass (r = 0.024). This confirmed that the effects of mass had been factored out by the scaling.

View this table: [in this window] [in a new window]   Table 2.   Correlations of the various measures of energy expenditure with heart rate and sO2 for all activities combined, treadmill activities, and unregulated play activities

Planned comparisons of the correlations with sO₂ were carried out comparing Tritrac_xyz and heart rate, WAM and heart rate, hip pedometer and heart rate, Tritrac_xyz and WAM, and Tritrac_xyz and hip pedometer. In addition, the correlations between sO₂ and Tritrac_xyz and the best uniaxial predictor (Tritrac_z) were compared to assess whether the three-dimensional assessment was better than any one uniaxial assessment. These were carried out using an adapted t-test that takes into account the correlation between the two coefficients of correlation (12). To control for type I error, the Bonferroni correction was used (13);  (0.05) was divided by the number of tests to give P = 0.008. sO₂ correlated significantly better with Tritrac_xyz than with heart rate, WAM, hip pedometer, or Tritrac_z (P < 0.001). The correlation between heart rate and sO₂ was not significantly different from that between WAM or hip pedometer and sO₂.

Triaxial accelerometry revealed that the major acceleration component varied in some subjects according to activity. Generally, the vertical plane, Tritrac_z, had the largest acceleration component followed by the anteroposterior plane, Tritrac_y (Fig. 1). However, mean values (Table 1) show that, during crayoning and catching, the anteroposterior plane had the greatest influence followed by the mediolateral (Tritrac_x) plane for crayoning and the vertical plane for catching.

View larger version (23K): [in this window] [in a new window]   Fig. 1.   A typical plot of Tritrac output measured in mediolateral (tri x), anteroposterior (tri y), and vertical (tri z) planes.

Because of the strong relationships, linear regression equations were computed to predict sO₂ from each of the measures (Table 3, Fig. 2). The best single measure was Tritrac_xyz, which accounted for 82.5% of the variance. Irrespective of the Tritrac measure used, the variance accounted for by the Tritrac was greater than any of the other measures (Table 3). The weakest of the Tritrac predictors (Tritrac_x) was still superior to the best of the rest, i.e., the pedometer worn at the hip (71.8% compared with 64.8%). It is surprising that the simple pedometer had less error associated with its predictions than did either the WAM or the heart rate monitor. The WAM data from one subject had to be rejected, because the counts seemed "jammed" at a high level. A small number of unusually high WAM counts were recorded during hopscotch and running at 10 km/h. However, these scores could not be eliminated, because counts before and after appeared normal. This contributed to the relatively poor relationship between WAM and sO₂. Additional pedometers worn at the wrist and ankle did not improve the estimate of sO₂ from the hip pedometer alone.

View this table: [in this window] [in a new window]   Table 3.   Regression analyses predicting sO2

View larger version (28K): [in this window] [in a new window]   Fig. 2.   Relationship between O2 uptake scaled for body mass (sO2) and Tritrac total magnitude counts (A), WAM counts (B), heart rate (C), and hip pedometer counts (D) during regulated treadmill and unregulated play activities in 30 boys and girls aged 8-10 yr. bpm, Beats/min.

Multiple regression analysis was used to establish the efficacy of using two measures simultaneously (Table 3). Each possible pair of measures was forced into the regression equation. The order of entry was alternated to illustrate the amount of variance accounted for by the second measure, over and above that already accounted for by the first measure. To reduce the possibility of type I error,  was adjusted to 0.01. The best model contained Tritrac_xyz and heart rate (R² = 0.849). Heart rate added 2% (P < 0.01) of variance to that already explained by Tritrac_xyz, whereas Tritrac_xyz was responsible for an extra 21.1% (P < 0.01) in addition to the variance accounted for by heart rate. The pedometer caused a similar increase in R² when added to heart rate, as heart rate did when used in addition to the pedometer (16.4 and 15.3%, respectively).

The relationships between sO₂ and Tritrac_xyz, WAM, heart rate, and hip pedometer are presented in Fig. 2. The reduced scatter can be seen in the Tritrac_xyz scattergram compared with the WAM and heart rate scattergrams. However, the pedometer data appear to deviate from the assumption of homogeneity of variance. This was confirmed by a graph of the residuals, which also showed the WAM data to deviate slightly from normal distribution.

It can be seen from Fig. 3 that the accuracy with which each individual activity was measured depended on which activity monitor was being used. This is highlighted when hopscotch is studied. WAM and heart rate assessed hopscotch more accurately than any other activity. Conversely, this activity was the least accurately assessed by Tritrac_xyz and the hip pedometer. Overall, the tendency to underestimate sO₂ as the exercise intensity increases is clearly shown, particularly as hopscotch had a higher oxygen cost than running at 8 km/h (Table 1).

View larger version (20K): [in this window] [in a new window]   Fig. 3.   Difference between actual and predicted sO2 for each activity measured by Tritracxyz, WAM, heart rate, and hip pedometer.

	DISCUSSION

Top Abstract Introduction Methods Results Discussion References

Prediction of sO₂

Although the heart rate monitor and WAM did significantly add to the variance accounted for by the Tritrac_xyz (P < 0.01), the increase was very small: 2.0 and 1.7% for heart rate and WAM, respectively. The justification of the cost of equipping a subject with two measures to effect this increase would be very questionnable. It was expected that the heart rate would account for rather more variance, over and above that explained by the Tritrac_xyz, as a different construct was being measured (heart rate as opposed to movement). Conversely, the ability of heart rate to explain sO₂ was greatly enhanced when Tritrac_xyz was added as a second predictor (21.1% increase in explained variance). It is interesting to note the significant increase in variance accounted for by the WAM, as this indicates that some of the uniaxial WAM output is independent of the Tritrac_xyz output.

The variance in sO₂ accounted for by the measures used to assess separate unregulated play activities was low, ranging from 3.5 to 26.1%, 2.2 to 3.7%, 0.1 to 8.4%, and 3.8 to 12.7% for Tritrac_xyz, heart rate, WAM, and hip pedometer, respectively. However, when all unregulated play activities were considered together, the accounted variance increased to 85.8% (Tritrac_xyz), 73.6% (heart rate), 72.6% (WAM), and 84.8% (hip pedometer). The Tritrac vector magnitude results compared favorably with the study of Bouten et al. (7), in which only 3-32% of the variance in energy expenditure during individual sedentary activities and 67% for all sedentary activities together were accounted for by a triaxial accelerometer. This highlights the low sensitivity of accelerometry when energy expenditure is predicted for short-term individual activities. The extremely low variance accounted for by heart rate was not surprising here, as all but one of the unregulated activities was of low intensity. The inability of heart rate to predict oxygen uptake at low activity levels is well documented (11).

In the present study, only raw heart rate data were used to predict sO₂. It is possible that an improved relationship would have been found had resting heart rate been accounted for. However, this was not found to be the case in a similar study with adults (19), in which a simple linear regression technique was found most appropriate. In addition, development of an individual heart rate-sO₂ regression line for each subject would not be feasible for a large population study.

Pedometer

The hip pedometer results were very encouraging. An inexpensive, "low-tech" measure appears to be as valid as the frequently used measure of heart rate and "high-tech" uniaxial accelerometry (WAM). In addition, the pedometer accounted for more variance when added to either of these two measures (16.4 and 16.6% for heart rate and WAM, respectively) than they did when added to the pedometer (15.3 and 12.2% for heart rate and WAM, respectively; Table 3). This offers potential for objective population studies on activity levels. When only unregulated play activities were assessed, supposedly where the pedometer would be least suited, the correlation with sO₂ was 0.921, which was significantly higher than the corresponding correlations for the WAM and heart rate (P < 0.005). The pedometer worn on the wrist was least suited to the assessment of physical activity. It was thought that three pedometers, one worn on the ankle, one on the wrist, and one on the hip, would provide a more accurate prediction of sO₂ than the hip pedometer alone. However, the extra pedometers did not significantly improve the prediction.

The high correlation for the hip pedometer may be misleading. The data points were not distributed uniformly, consisting of one clump of low values and another clump of high values. This may be due to the nature of the activities performed. Catching and crayoning both led to very few counts on the relatively insensitive pedometer, followed by a large increase when the subject began walking. It is possible that an activity that fitted between those that involved minimal movement and regular movement categories would have improved the fit of the data. When treadmill activities alone were studied, which would effectively eliminate the first clump of data points, the correlation remained high [r = 0.782: level with heart rate (0.784) and higher than WAM (0.692)]. The ensuing scatter plot and residuals graph revealed normality and equality of variance. However, treating the WAM data in the same way did not resolve the problem.

Error of Prediction for the Tritrac_xyz and the WAM

0.75

1

Dominant Axis

Measurement of Subgroups of Activities

1

1

During the unregulated play activities (hopping, catching, and crayoning), correlations between all measures were higher than when treadmill activities or when all activities were considered. The explanation is most likely that when walking or running on the treadmill the activity levels of the children were relatively homogenous, which would elicit lower correlations (9). During the unregulated play activities the range of movement and heart rate scores were more heterogenous and, hence, resulted in higher correlations. Children's habitual activity levels vary considerably between and within subjects. It is therefore important that each measure be validated across a whole range of activities and, hence, across a whole range of scores.

Different activities with similar sO₂ and heart rate responses did not always elicit similar counts from the activity monitors. When catching and walking at 4 km/h or hopscotch and running at 8 km/h were compared, the counts per minute were very different, whereas sO₂ and heart rate remained fairly constant (Table 1). We believe that the nature of the activities causes these discrepancies. Catching predominantly involves the arms with little torso movement; hopscotch is a jumping activity and, hence, uses more energy per movement than walking or running. It was our intention to deliberately select activities that would reflect normal "free-living" activities typical of children's behavior. This would ensure a stringent test for the accelerometers and pedometers and enable us to make ecologically valid observations. Despite these activities being included, when all activities were considered together, the activity monitors still provided a very good prediction of sO₂. This gives us confidence that, when daily activity is assessed as a whole, the presence of these activities at times during the day should not throw out the overall estimation of the day's activity. This again emphasizes the importance of validating the monitors over a range of activities and assessing how the monitors perform overall as opposed to during individual activities.

All monitors underestimated sO₂ at 10 km/h, particularly the WAM (Fig. 3). A likely explanation for this is the much greater increase in stride length in relation to stride frequency. It is well documented that stride length increases in greater proportion than stride frequency in adults (6) and children (30). Accelerometry or pedometry methods will not account adequately for stride length changes, which leads to underestimation at higher speeds. Accelerometers have been found to underestimate energy expenditure during high-intensity activities (19). However, the Tritrac vector magnitude score was superior to the heart rate monitor (an underestimation of 1.93 ± 9.4042 ml · kg^0.75 · min¹ compared with an underestimation of 7.26 ± 14.2071 ml · kg^0.75 · min¹, respectively, during 10 km/h). Conversely, the sedentary crayoning activity was overestimated by all monitors. The accuracy of the WAM was more closely related to the accuracy of the heart rate monitor than were the other movement counters. WAM and heart rate had less absolute error (Fig. 3) when predicting sO₂ for hopscotch and the most error when predicting sO₂ for crayoning. Conversely, the Tritrac vector magnitude and the hip pedometer provided their worst estimates when the activity was hopscotch. The examination of the accuracy of monitors to assess isolated activities is largely academic. As a measure of habitual physical activity, it is important that the monitor provide an accurate picture of varied activity over long periods of time.

Studies that have examined activity levels throughout a whole day have obtained lower correlations for the WAM and Tritrac with heart rate compared with those observed in the present study. In this study, correlations of the Tritrac vector magnitude and the WAM with heart rate were 0.791 and 0.684, respectively (P < 0.001). Using the Tritrac activity monitor on 30 schoolchildren, aged 9-11 yr, Welk and Corbin (33) obtained correlations with heart rate of 0.34-0.50 over 3 days (mean = 0.44). When resting heart rate was accounted for, the correlation increased slightly to 0.46-0.74 (mean = 0.60). Janz (14), who also accounted for resting heart rate, obtained correlations of 0.51-0.70 (mean 0.57, P < 0.05) over 3 days between the WAM and heart rate for 31 schoolchildren, aged 7-15 yr.

A stringent test concerning the validity of triaxial accelerometry in adults was carried out by Bouten et al. (8) on the Tracmor unit. Average daily metabolic rate was assessed over 7 days using the field study gold standard of doubly labeled water. Sleeping metabolic rate was measured during 2 nights in a respiration chamber. Physical activity level (defined as the ratio of these 2 values) correlated highly (r = 0.73, P < 0.001) with Tracmor output (corrected for vibrations caused by transport).

Heart Rate vs. Triaxial Accelerometry

Summary

The regression equations developed in this study should be cross validated on a different sample of children, if possible undertaking different activities. If the cross validation of the Tritrac confirms the results of this study, the Tritrac will provide the ideal criterion measure for use in finding less expensive, field-based assessments of activity levels, such as the pedometer. The Tritrac appears to have greater accuracy than the heart rate monitor, is capable of collecting data for a longer period of time, and does not interfere with normal activity. Results indicate potential for the hip pedometer as a measure of habitual physical activity in large samples.