The field of physical activity assessment using activity monitors has expanded because of technical developments, providing the opportunity to collect and to analyze a vast amount of data at a high rate for long time periods. This has made it possible to process signals from multiple sensors and to take advantage of machine-learning techniques for activity recognition (18) and application of activity-specific algorithms for the assessment of energy cost (20). Studies in children and adults have shown that combining motion sensors with heart rate monitors results in less error in assessing energy cost compared with using them separately at both group and individual levels (3,5,21,22,24). Also, different relationships between energy cost and heart rate exist for arm work compared with leg work (21), and activities of the same metabolic cost show different acceleration scores (10). Hence, using one single algorithm for all activities results in differing prediction accuracy (7,23). However, even the most sophisticated and accurate activity monitor would fail in presenting high-quality data if it were too cumbersome to wear by the subject or to use by the investigator. Also, it has to be inexpensive to be feasible in large-scale studies. When measuring physical activity in children, the activity monitor must be easy to put on and take off and must remain secured in its optimal location. It should also be nonobtrusive, that is, it should not interfere with daily activities, and at the same time it should be resistant to shocks.
Two activity monitors that have taken advantage of combining multiple sensors with machine-learning techniques are the SenseWear® Pro2 Armband (SWA; BodyMedia, Inc., Pittsburgh, PA) and the Intelligent Device for Energy Expenditure and Activity® (IDEEA; Minisun LLC, Fresno, CA). Although in the SWA five different types of sensors have been incorporated into one single device worn as an armband around the upper arm, the IDEEA uses an array of mini-accelerometers attached to different parts of the body. According to the manufacturer, the SWA is accurate in subjects 7-65 yr old for resting, ambulatory movement, stationary bicycling, motoring, and weight-lifting activities. The SWA has been evaluated in children during resting and while performing different physical activities (2). It underestimated energy cost for most activities, an underestimation that increased with increasing intensity. This may reflect the fact that the number of activity-specific algorithms is limited in children and it takes time to develop them. The SWA has not been compared with other activity monitors in children. The IDEEA has shown promising results in adults for the assessment of energy cost (20,28), and the SWA and the IDEEA showed a high degree of agreement when compared with each other (27). The IDEEA has not yet been evaluated in children.
Another multisensor activity monitor is the ActiReg® (AR; PreMed AS, Oslo, Norway), which combines the signals from two sensors recording body position and motion for the assessment of energy cost. Compared with the SWA and the IDEEA, it uses simpler techniques and algorithm and is less expensive. Also, the algorithm is available to the user, and in the AR software, there are possibilities to elaborate with it. The AR has shown promising results in adults (1,11) but has never been tested in children. In contrast to the SWA, both the AR and the IDEEA have not yet gone through any systematic calibration procedure in children. Like the SWA, the IDEEA contains proprietary algorithms, and calibrations can only be made through the manufacturer. Hence, we were interested in evaluating these activity monitors as they are delivered by the manufacturers.
The aim of this study was to compare the ability of the SWA, the IDEEA, and the AR to assess energy cost in children during resting and during different physical activities using indirect calorimetry as the criterion method.
Study Design and Subjects
Energy cost during rest and during different physical activities was assessed by the SWA, the IDEEA, the AR, and the Oxycon Mobile portable metabolic system (OM; VIASYS Healthcare, Conshohocken, PA), which measured V˙O2 and V˙CO2. The OM was used as the criterion method for energy cost in this study. Fourteen healthy children, eight boys and six girls, aged 11-13 yr were recruited from an ongoing study of physical activity and aerobic fitness in children and adolescents in Gothenburg, Sweden. Written informed consent was received from the children and their parents. The study was approved by the Ethical Vetting Board of Western Sweden.
Assessment of Energy Cost
SenseWear Pro2 armband
The SWA (BodyMedia, Inc.) is worn on the back of the upper arm attached with an adjustable strap and is a multiple sensor device (85 × 53 × 19 mm, 79 g) collecting data from a skin temperature sensor, a near-body temperature sensor, a heat flux sensor, a galvanic skin response sensor, and a biaxial accelerometer. The skin temperature sensor and the near-body temperature sensor (a vent on the side of the armband) consist of sensitive thermistors in contact with the skin relying on changes in resistance with changing skin temperature. The heat flux sensor uses the difference between skin temperature and near-body temperature to assess heat loss. The galvanic skin response sensor measures the conductivity of the skin between two electrodes in contact with the skin. Conductivity of the skin varies due to physical and emotional stimuli. The biaxial accelerometer registers the movement of the upper arm and provides information about body position (lying or upright) by detecting gravity acceleration. Information from the sensors together with gender, age, height, and weight was incorporated into proprietary algorithms to estimate energy expenditure (EE). These algorithms are activity specific and are automatically applied based on analysis of the pattern of signals from the sensors. At its fastest rate, the SWA can store data up to 32 times per second with a storage capacity of 1 h, but when the rate is set to once a minute, it has a storage capacity of 10 d. One battery (LR03, 1.5 V) supplies energy for up to 14 d. The same SWA (version 6.03) was used in all subjects and was placed on the right arm over the triceps muscle 20 min before data collection. EE was calculated using InnerView® Professional software (version 5.1).
Intelligent device for energy expenditure and activity
The IDEEA (Minisun LLC) consists of five sets of sensors (16 × 14 × 4 mm, 2 g) attached with medical tape over the sternum, to the front side of each thigh and under each foot, connected with thin soft cables to a microprocessor/storage unit (70 × 44 × 18 mm, 59 g) that can be worn on the belt. The sensors send information about angles of body segments and motion (acceleration) in two orthogonal directions to the microprocessor for the identification of type of activity, gait analysis during walking and running, and calculation of duration, frequency, and intensity of activity. The IDEEA can identify most postures and motions, except for bicycling and jumping on one foot. It stores data 32 times per second and has a storage capacity of 6 d. Two batteries (LR1, 1.5 V) supply the unit with energy for more than 48 h. Stored activity data together with information about age, gender, body weight, height, and estimated fitness level (1-10) are used in activity-specific proprietary algorithms for the calculation of energy cost. The results can be displayed and analyzed in the software ActView™. The same IDEEA version 3.01 was used in all subjects. No instructions are provided by the manufacturer regarding how to apply the 10-level fitness scale. Hence, we assigned an arbitrary value of 6 for all subjects in this study, considering them to have at least an average fitness level or more but not belonging to the highest fitness levels. Attachment of sensors and calibration was performed according to the manufacturer's instructions.
The AR system (PreMed AS) consists of two components, an activity monitor (ActiReg®) and a computer program (ActiCalc32®) for processing and presenting the AR data and for calculating energy cost. The AR has two pairs of sensors attached by medical tape, one over the sternum and one to the front of the right thigh. Each pair consists of one body position sensor and one motion sensor. The chest and the thigh sensors are secured to 20 × 42- and 27 × 38-mm plastic brackets, respectively. Together, the body position sensors can discriminate between standing (both sensors in vertical position), sitting (chest sensors in vertical position and thigh sensor in horizontal position), bending forward (chest sensor in horizontal position and thigh sensor in vertical position), and lying (both sensors in horizontal positions). The break point when the sensors go from vertical to horizontal position is when they deviate more than 45°. Each motion sensor detects if there is motion or no motion. The sensors are connected to a storage unit (60 g, 82 × 45 × 15 mm) through thin soft cables, and the storage unit is placed in a tight material pocket in an elastic belt around the waist and has a storage capacity of more than 30 d. Each second, 1 of 16 different combinations of body position and body motion is stored as a unique code, 60 codes per minute. Two batteries (LR1, 1.5 V) supply the energy needed for up to 14 d, and a lithium battery (CR2430, 3.0 V) supplies the internal clock if the main batteries are missing or if the voltage gets below 1 V.
The stored codes are transferred to a computer where ActiCalc32® translates them into information about body position and intensity level each minute. One main body position is assigned for each minute. Calculation of the intensity level, called the activity factor (AF), is based on the 60 codes per minute. If a code indicates no motion in either sensor, it receives the weight 0.00. If there is motion in one of the sensors, the code receives the weight 0.50, and motion in both sensors receives the weight 1.00. The AF is the average of the 60 weights per minute and will attain a value between 0.00 and 1.00. The AF is categorized into three physical activity (PA) levels: very low PA, AF < 0.10; low PA, 0.10 ≥ AF < 0.90; and moderate-to-vigorous PA, AF ≥ 0.90. Each body position at each PA level is assigned a multiple of basal metabolic rate (BMRmultiple) taken from published reference values (8, p. 92): very low PA, lying 1.0, sitting 1.2, bending forward and standing 1.4; low PA, lying and sitting 2.0, bending forward and standing 2.5; and moderate-to-vigorous PA, all positions 5.0. AF 1.00 is reached at a speed of 5 km·h−1. Hence, to be able to capture higher intensities, a second variable for intensity level is included in the calculation of energy cost, position changes (PC). The PC are combinations of real PC and acceleration forces on the position sensors, increasing proportionally with the physical activity intensity and attains a value between 0 and 60 per minute. The final algorithm for energy cost each minute is E = BMR × BMRmultiple (1 + 0.025 PC), where BMR can be either measured or predicted. In this algorithm, the PC part (BMRmultiple × 0.025PC) adds variation to the intensity assigned for each body position (1 × BMRmultiple) within the low and moderate-to-vigorous PA level, where PC occur. Energy cost is then achieved by multiply with BMR (kJ·min−1). The same AR version 2.9 and ActiCalc32® were used in all subjects together with predicted BMR using FAO/WHO/UNU equations (8, p. 37): boys, BMR (MJ·d−1) = 0.074 × weight (kg) + 2.754; girls, BMR (MJ·d−1) = 0.056 × weight (kg) + 2.898. The sensors were attached over the sternum and to the front side of the right thigh midway between the hip and the knee using medical tape.
The OM (VIASYS Healthcare) is a battery operated, portable, wireless metabolic system measuring gas exchange breath by breath while attached to the body in a vest system. It has been tested against the stationary system Oxycon Pro (accurate compared with the Douglas bag method) (19) and showed lower V˙O2 values for higher workloads but were still valid for V˙O2 at lower intensities and for V˙CO2 across all intensities (16). A flow sensor unit is connected to a face mask (Hans Rudolf, Inc., Kansas City, MO) and detects the air flow by the rotation of a low-resistance, bidirectional turbine for determination of ventilation. Via a sampling line connected to the flow sensor unit, the expired air is analyzed for O2 and CO2 concentrations in a sensor box using a microfuel cell and thermal conductivity, respectively. A data exchange unit collects the data and sends it telemetrically to a base station connected to a computer. After 30 min of warm-up time and immediately before data collection, a two-point (0.2 and 2.0 L·s−1) air flow calibration was performed using the automatic flow calibrator. At the same time, the gas analyzers were calibrated against room air and a certified gas mixture of 16% O2, 5% CO2, and 79% N2 (Reissner-Gase Gmbh & Co, Lichtenfels, Germany) together with determination of measurement delay time. EE was calculated from the gas exchange data using the equation of Weir (26): EE (kJ) = 4.184 (3.9 V˙O2 + 1.1 V˙CO2). Before the start of measurement and data collection, the face mask was checked to prevent air leakage, and gas exchange values were confirmed to be within normal limits.
Each subject arrived at the laboratory having been in the postprandial state for at least 3 h. Body weight was determined to the nearest 0.1 kg using a digital scale, and body height was determined to the nearest centimeter using a horizontal headboard with an attached wall-mounted metric rule. The SWA, the IDEEA, the AR, and the OM were time synchronized and attached to the body. After a short instruction and adaptation to the equipment, resting energy expenditure (REE) was assessed for 30 min with the subject lying down listening to soft music (Table 1). During a 90-min break when the OM was detached from the body (the SWA, the IDEEA, and the AR remained attached), the subject was offered a small meal, followed by instructions and time for getting acquainted to the activities included in the study. The OM was again attached to the body. After 10 min in sitting position, the subject performed five different activities of 5-min duration each and separated by 5 min in sitting position: 1) sitting quietly; 2) stationary bicycling (Cardio Care 827E; Monark Exercise AB, Vansbro, Sweden); 3) jumping on a trampoline; 4) playing basketball; and 5) stair walking.
After a 10-min break sitting quietly, the subject performed walking at three different paces and running at two different paces on a 50-m marked track (Table 1). During the previous 90-min break, the subject had been instructed how to perform the walking and the running activities and was given time to establish and to practice the different paces. Total duration of the laboratory visit was 4 h.
For the assessment of REE from the SWA, the IDEEA, and the OM, the average of 16-25 min of the 30-min lying down was used; 1-15 min was used for the subject to become relaxed. The AR was not included in the assessment of REE due to the known use of 1 × BMR for this activity, and BMR was predicted using FAO/WHO/UNU equations, which are not evaluated in this study. However, the SWA and the IDEEA use proprietary algorithms for resting, which are evaluated in this study. For the assessment of energy cost of all other activities (except for normal running), the average of 3-5 min was used, considering 1-2 min as an equilibrium period. For normal running, the average of 2-3 min was used for the assessment of energy cost, considering the first minute as an equilibrium period. The selection of equilibrium period was based on the observed change in gas exchange as well as change in EE assessed by SWA across the measurement period for each activity. The SWA measures skin temperature to assess heat flux that also has an equilibrium period. The largest changes occurred during the first 1-2 min, and thereafter a plateau was reached with minor further changes. The following mean (SD) changes in V˙O2 were observed across the final 3 min of each 5-min measurement period for the assessment of energy cost (2 min for normal running): sitting quietly, −1.0% (10.3); stationary bicycling, −1.2% (6.7), jumping on a trampoline, −7.6% (8.6); playing basketball, −1.8% (9.0); stair walking, −1.0% (5.3); slow walk, 1.2% (4.5); normal walk, 3.5% (9.3); brisk walk, −3.2% (6.8); slow running, 2.7% (7.0); and normal running, 0.9% (5.6). The mean (SD) changes in EE assessed by the SWA were as follows: sitting quietly, 6.0% (15.1); stationary bicycling, 5.0% (5.6); jumping on a trampoline. 2.0% (19.9); playing basketball, −0.5% (4.6); stair walking, 0.1% (3.8); slow walk, 4.2% (7.4); normal walk, 2.6% (4.6); brisk walk, 0.2% (5.2); slow running, 0.9% (3.0); and normal running 0.8% (2.4). Hence, we consider the equilibrium period as adequate for V˙O2 and SWA and interindividual variation because of variation of the performance of the activity.
Energy cost assessed by the SWA, the IDEEA, and the AR was compared with the OM, and the differences between the methods were tested statistically using t-test. Most of the differences between the methods showed approximal normal distribution according to the Shapiro-Wilk test. The MET values in this study were established from the OM data by the quotient of the energy cost of the activity and the REE. Speed of walking and running was assessed from the distance covered for the last 3 min (last 2 min for normal running) on the 50-m track. Results are presented as mean (SD). All statistical analysis was performed using SPSS version 14.0 (SPSS Inc., Chicago, IL).
Complete results for all activities were collected from 12 out of 14 children. One of the boys was not able to complete the two running activities, and a technical error in the OM resulted in lost data for the walking activities for a second boy. Subject characteristics for all 14 children are presented in Table 2. One of the girls showed larger body size, especially body weight, compared with the other girls, which had an effect on the statistical test of difference in body weight between boys and girls. The boys showed a nonstatistically significant higher weight compared with the girls, and because there was no sex difference in height, they also showed a higher BMI. The boys also showed a nonstatistically significant higher REE compared with the girls. Data are presented for all children together (Table 3). Due to variation in body size in our sample, energy cost is presented as kilojoules per kilogram per minute. The children were considered as having mostly normal weight for their height.
The SWA underestimated energy cost for resting by 0.01 (0.01) kJ·kg−1·min−1 (P = 0.003), and both the AR and the SWA underestimated energy cost for sitting quietly by 0.01 (0.02) kJ·kg−1·min−1 (P = 0.008) and 0.03 (0.02) kJ·kg−1·min−1 (P < 0.001), respectively (Table 3). The high energy cost for resting (0.13, 0.14, and 0.16 kJ·kg−1·min−1) and sitting quietly (0.15, 0.17, and 0.20 kJ·kg−1·min−1) assessed by the IDEEA for three of the children resulted in large SD of the difference in energy cost between the IDEEA and the OM and hence a nonstatistically significant overestimation of energy cost of 0.01 (0.03) kJ·kg−1·min−1 (P = 0.13) and 0.02 (0.03) kJ·kg−1·min−1 (P = 0.08), respectively, (Fig. 1A). The AR showed the closest estimate of energy cost for stationary bicycling, +0.06 (0.16) kJ·kg−1·min−1 (P = 0.17), and the SWA for jumping on a trampoline, +0.04 (0.24) kJ·kg−1·min−1 (P = 0.58). However, these activity monitors showed large individual variations in the ability to assess energy cost for these activities (Fig. 1A). The AR overestimated the energy cost for playing basketball by 0.13 (0.15) kJ·kg−1·min−1 (P = 0.007), whereas this activity was underestimated by the SWA and the IDEEA by 0.22 (0.17) kJ·kg−1·min−1 (P < 0.001) and 0.24 (0.17) kJ·kg−1·min−1 (P < 0.001), respectively. Only the IDEEA accurately assessed energy cost for stair walking, ±0.00 (0.05) kJ·kg−1·min−1 (P = 0.90).
FIGURE 1-(A-B) Indiv...Image Tools
For slow and normal walking, the SWA accurately assessed energy cost, which differed by ±0.00 (0.04) kJ·kg−1·min−1 (P = 0.83) and -0.01 (0.03) kJ·kg−1·min−1 (P = 0.21) from the OM, whereas the IDEEA overestimated energy cost for these activities by 0.02 (0.03) kJ·kg−1·min−1 (P = 0.02) and 0.03 (0.03) kJ·kg−1·min−1 (P = 0.01) (Table 3). However, the SWA showed increased underestimation of energy cost with increasing intensity, whereas the IDEEA was better in assessing energy cost for higher intensities, ±0.00 (0.04) kJ·kg−1·min−1 (P = 0.63) for brisk walking, −0.05 (0.07) kJ·kg−1·min−1 (P = 0.04) for slow running, and +0.08 (0.13) kJ·kg−1·min−1 (P = 0.06) for normal running compared with the OM. The AR overestimated energy cost for all intensities, except for normal run where the difference was -0.02 (0.13) kJ·kg−1·min−1 (P = 0.51) compared with the OM. There was less variation in the assessment accuracy for the walking-running activities compared with the other more complex activities (Fig. 1).
Although underestimation increased with increasing intensity, the SWA showed the same difference in slopes as the OM for energy cost going from walking to running (Fig. 2). The IDEEA had the strongest agreement with the OM across the different walking and running intensities but without the apparent change in slope for the running activities. The energy cost curve of the AR did not show any similarity to that of the OM, starting with a high overestimation of energy cost at slow walk, with a modest increase in energy cost with increasing intensity to a plateau at slow running, not responding to any further increase in intensity. When calculating the energy cost for the whole measurement period for each activity (kJ·kg−1 per 5 min and kJ·kg−1 per 3 min for normal running) from the minute energy cost (kJ·kg−1·min−1) and then adding together the energy cost for all walking and running intensities, there was only a difference of +0.29 (0.95) kJ·kg−1 (P = 0.30) between the IDEEA and the OM (Table 3). For the same activities, the SWA underestimated energy cost by 1.39 (1.20) kJ·kg−1 (P = 0.002), and the AR overestimated energy cost by 6.00 (1.90) kJ·kg−1 (P < 0.001). Also, of the three activity monitors, the IDEEA showed the closest estimate of energy cost for all activities together (not including energy cost for resting in the calculation) with an underestimation of 2.90 (2.31) kJ·kg−1 (P = 0.001) compared with an underestimation of 4.49 (2.65) kJ·kg−1 (P < 0.001) by the SWA and an overestimation of 9.59 (3.22) kJ·kg−1 (P < 0.001) by the AR.
FIGURE 2-Mean energy...Image Tools
The ability of the SWA, the IDEEA, and the AR to assess energy cost in children during resting and during different physical activities was compared in this study using a portable indirect calorimeter as the criterion method. For resting and sitting, the three activity monitors showed comparable ability to assess energy cost. None of the methods could accurately assess energy cost of stationary bicycling, jumping on a trampoline, or playing basketball. Only the IDEEA was accurate for stair walking. For walking and running, both the SWA and the IDEEA showed higher accuracy with less individual variation compared with stationary bicycling, jumping on a trampoline, and playing basketball. However, there was an increased underestimation of energy cost by the SWA with increasing intensity not shown by the IDEEA. The AR systematically overestimated energy cost of walking and running and was not able to respond to any increase in running speed. Overall, the IDEEA showed the smallest difference from the criterion method to assess energy cost of physical activity in children, largely because of its ability to assess energy cost during walking and running.
The SWA has been calibrated to assess resting energy cost in children. However, REE was underestimated by 0.01 (0.01) kJ·kg−1·min−1 (P = 0.003) in this study and by 0.02 (0.02) kJ·kg−1·min−1 (P < 0.001) in our previous study (2). REE assessed by SWA showed better agreement with REE from prediction equations (2). Papazoglou et al. (15) found that REE assessed by the SWA in obese adults was highly correlated with REE predicted by the Harris-Benedict equations (r = 0.96). The SWA has the ability to include metabolic signals in the assessment of energy cost. However, it seems to calculate resting energy cost based on physical characteristics only and does not take into account that the children in our study were in a resting condition rather than BMR condition. Also, the SWA did not respond to the change from resting to sitting, resulting in an increase in the underestimation of energy cost with the change of position. The IDEEA has not yet been calibrated in children but showed the same ability to assess resting energy cost as SWA. However, three of the children showed extreme values of resting energy cost from IDEEA (Fig. 1A). If they were excluded from the statistical calculations, there would be no difference between the IDEEA and the OM, ±0.00 (0.01) kJ·kg−1·min−1 (P = 0.78). It seems that the extreme overestimation for resting affected the results for sitting because the same three children received extreme values for this activity too. If they were again excluded from statistical calculation, the difference between the IDEEA and the OM during sitting would be +0.06 (0.03) kJ·kg−1·min−1 (P = 0.49). The raw data and the personal characteristics for calculating energy cost for these three children did not differ from the other 11, and hence there is no suitable explanation for the extreme values other than that the IDEEA was out of calibration in these children. However, for the other activities, the IDEEA did not produce any extreme values for these children. Both the AR and the IDEEA use personal characteristics to calculate energy cost and seem to multiply the resting value by a constant factor to assess the sitting energy cost. Hence, the assessment accuracy for resting and sitting depends on the accuracy of the algorithm developed for resting. For the AR, the FAO/WHO/UNU algorithms for BMR (8) were used in this study, which would by necessity underestimate resting energy cost because the BMR criteria were not fulfilled in this study. However, the factor 1.2 used by the AR for sitting energy cost seems to be correct because the MET value assessed by the OM for sitting was 1.2. A MET value of 1.3 was determined in our previous study (2). The choice of using FAO/WHO/UNU algorithms in the AR probably explains the underestimation in sitting energy cost made by the AR. Hence, none of the activity monitors seem to be superior to the other for resting and sitting, and all of them need further calibration to better capture these activities in children.
Both the SWA and the IDEEA showed higher accuracy in assessing energy cost during walking and running compared with bicycling, jumping on a trampoline, and playing basketball with less individual variation. Although the SWA showed increased underestimation of energy cost with increasing speed, there were similarities in the energy cost curves between the SWA and the OM (Fig. 2). It has been shown that the relationship between speed and energy cost differs for walking compared with running (17,25). The speed-energy cost curve for walking shows an increasing slope with increasing speed, whereas the relationship is linear for running with a significantly lower slope. The different relationships for walking and running were observed for the SWA. Similar results were found in our previous study for treadmill walking and running (2). However, there was a leveling-off of the energy cost assessed by the SWA for increasing running speed. Actually, the SWA detected an increased energy cost for increased running speed for most subjects in both our studies. However, a marked decrease in energy cost was assessed by the SWA for two of the subjects in our previous study, whereas a marked increase in energy cost was assessed for two of the subjects in the present study. These changes were not accompanied by similar changes in the OM. If they were excluded from the statistical calculations, the SWA assessed an increase in energy cost between the two running speeds from 0.66 (0.08) to 0.68 (0.06) kJ·kg−1·min−1 in our previous study compared with a change from 0.63 (0.06) to 0.67 (0.07) kJ·kg−1·min−1 in the present study. These increases of 3.0% and 6.3 %, respectively, can be compared with increases of 17.3% and 21.8%, respectively, assessed by the OM. Whether this is an indication that the SWA may detect a difference in running economy between walking and running or a limitation by the monitor to respond to higher intensities or a combination of both may be further investigated. A different shape of the energy cost curve was shown for the IDEEA with overestimation of energy cost at the highest running speed. We observed a larger overestimation of speed by the IDEEA for normal running compared with slow running (data not shown), which may explain the overestimation of energy cost and the shape of its energy cost curve. The deviation of the energy cost curves by the SWA and the IDEEA from that of the OM may also be explained by the SWA having only been calibrated in children for walking using InnerView® 5.1 analyzing software, whereas the IDEEA is not calibrated in children for any activity. The large overestimation of energy cost for walking and slow running by the AR is explained by an error in its algorithm. At 3 km·h−1 when the AF 0.9 is reached, a BMRmultiple of 5.0 is applied, which generates a large increase in energy cost assessed by the AR. A new algorithm using a more continuous transition between the low and the moderate-to-vigorous intensity level would be necessary to increase the accuracy of the AR. Also, it needs to be adjusted to increase its responsiveness to higher intensities. At running speeds higher than 8 km·h−1, the energy cost assessed by the AR ceases to increase (Fig. 1B), explained by a leveling-off of the number of PC (data not shown). A simple adjustment of the AR can be performed to increase its sampling rate.
All three activity monitors had problems with assessing energy cost of stationary bicycling, jumping on a trampoline, and playing basketball. The SWA has been developed to recognize and to assess energy cost of stationary bicycling in children. It did not respond correctly to the intensity of stationary bicycling, with the same degree of underestimation as in our previous study (2). The IDEEA has not yet been developed to identify bicycling (28), hence the systematic underestimation of this activity. Even if the AR showed the smallest mean difference compared with the OM, the leg sensor responded very differently among the children. In some cases, there was only a weak response and consequently a large underestimation of energy cost. Instead, in those children where there was a normal response, there was an overestimation of energy cost caused by the error in the algorithm for moderate-to-vigorous physical activity. By using pattern recognition, the IDEEA has the ability, through proper calibration procedure, to assess energy cost for bicycling, assuming the sensors are sensitive to the constant changing direction of the acceleration when pedaling. The AR is not using pattern recognition, and together with the large variation in response by the leg sensor, it will be difficult to calibrate it to assess energy cost of bicycling. The SWA must rely on the assumption that increases in heat and sweat dissipation from the arm truly reflect the increased intensity of bicycling. The large underestimation of energy cost in the present study indicates the need of a recalibration for this activity. Bicycling is a common activity among children, at least in Sweden, which requires that the activity monitor be able to detect this activity if it is going to be a useful tool to assess total physical activity in children.
Playing basketball represents a more complex activity consisting of several different movement patterns, including arm involvement. Both the SWA and the IDEEA have been trained to recognize simple activities and to apply activity-specific algorithms for energy cost, mostly in adults. Also, they have been tested for their ability to assess energy cost during simple and standardized activities in adults (6,9,12-15,20,28). Hence, the underestimation of energy cost of playing basketball by the SWA and the IDEEA may be explained by the limitation in assessing higher intensities and the limitation in detecting arm movement, respectively, and because none of the monitors are easily calibrated for the complex nature of this activity. The overestimation of energy cost by the AR for playing basketball is again explained by the error in the algorithm for moderate-to-high physical activity. The IDEEA has been trained to recognize and to assess energy cost of stair walking, and high accuracy was shown in this study. This was not the case for the SWA, which underestimated energy cost for all children. Again, this may be due limitations in the algorithms for children because the SWA has the ability to assess energy cost of stair stepping in adults (12).
Overall, the IDEEA showed the smallest difference from the criterion method in assessing energy cost in this study. All three activity monitors showed limitations in their ability to assess energy cost of most activities, largely explained by insufficient calibration in children. Multisensor devices were introduced to overcome the limitations of single-sensor waist-mounted activity monitors (4). The AR was an early attempt to combine body position with movement of the leg and trunk. The IDEEA developed this approach by adding more and smaller sensors but also by introducing pattern analysis to identify more activities and positions. The SWA took the step to analyze the pattern of different physiological signals to assess physical activity. The use of activity-specific algorithms by the SWA and the IDEEA needs to consider many different factors such as age, gender, and activity type. As indicated in the present study, to be able to assess physical activity and energy cost in children, the SWA and the IDEEA need to be calibrated to more activities in children. Both the SWA and the IDEEA are only in the beginning of their development potential for activity-specific assessment of energy cost in children, and more needs to be done before researchers can fully evaluate the abilities and the limitations of these activity monitors. Despite the higher accuracy of the IDEEA, the attachment of all sensors and cables decreases its wearability in children and its usefulness under free-living conditions. The approach of an armband seems much more feasible and may increase the compliance in children. Also, the SWA was the most functional activity monitor to use for performing physical activity measurements, including the procedure for starting up, downloading, viewing, and analyzing data.
Although a limitation of this study is the small sample size where some extreme output from the activity monitors will affect the group results, we propose that the results in this study give a clear indication of the performances of the activity monitors in children. Another limitation is that the activity monitors were tested in activities and/or in an age group for which the monitors have not been calibrated. The SWA is stated to be accurate in subjects 7-65 yr old for resting, ambulatory movement, stationary bicycling, motoring, and weight-lifting activities, but the IDEEA and the AR are not yet calibrated in children. However, there are few multisensor activity monitors available, and we were interested in assessing whether they could be used in children today. Interestingly, in the present study, the IDEEA showed a comparable accuracy to the SWA or in some activities even higher accuracy. The children in this study were offered a small meal after the resting activity. This may to some extent affect the V˙O2 and V˙CO2 and the energy cost assessed by the OM for the activities that started 60 min later. However, the meal was considered small and was offered to improve the compliance throughout the 4-h visit. This study was performed in a laboratory setting with many controlled physical activities. Hence, comparing the performances of these activity monitors in the field is encouraged using reliable criterion methods for EE.
To be able to capture children's physical activity, all three activity monitors need to be further developed. For resting and sitting, the three activity monitors showed comparable abilities to assess energy cost. None of the methods could accurately assess energy cost of stationary bicycling, jumping on a trampoline, or playing basketball. Only the IDEEA was accurate for stair walking. Both the SWA and the IDEEA showed higher accuracy for walking and running, whereas these activities were systematically overestimated by the AR. However, the SWA showed increased underestimation of energy cost with increasing walking-running speed, which was not seen in the IDEEA. Although the IDEEA showed the highest ability to assess energy cost in this study, the SWA may be more feasible for use in children under free-living conditions.
No financial support was provided for this study by any manufacturer of the instruments used.
The results of the present study do not constitute endorsement by ACSM.
The authors thank the children who with great patience performed all the activities in this study and the personnel at the physiotherapy section at the Queen Silvia Children's Hospital where we performed the activities. We are also grateful for the financial support from the Swedish Heart and Lung Association and the Swedish Heart and Lung Foundation and to Mark Fitch at the Department of Nutritional Sciences and Toxicology, University of California at Berkeley, for proofreading the manuscript.
1. Arvidsson D, Slinde F, Nordenson A, Larsson S, Hulthen L. Validity of the ActiReg system in assessing energy requirement in chronic obstructive pulmonary disease patients. Clin Nutr
2. Arvidsson D, Slinde F, Larsson S, Hulthén L. Energy cost of physical activities in children. Validation of SenseWear Armband. Med Sci Sports Exerc
3. Brage S, Brage N, Franks PW, et al. Branched equation modeling of simultaneous accelerometry and heart rate monitoring improves estimate of directly measured physical activity energy expenditure. J Appl Physiol
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. Corder K, Brage S, Wareham NJ, Ekelund U. Comparison of PAEE from combined and separate heart rate and movement models in children. Med Sci Sports Exerc
6. Cole PJ, LeMura LM, Klinger TA, Strohecker K, McConnell TR. Measuring energy expenditure in cardiac patients using the Body Media Armband versus indirect calorimetry. A validation study. J Sports Med Phys Fitness
7. Crouter SE, Churilla JR, Bassett DR Jr. Estimating energy expenditure using accelerometers. Eur J Appl Physiol
8. FAO/WHO/UNU. Human Energy Requirement. Report of a Joint FAO/WHO/UNU Expert Consultation, 17-24 October 2001, Rome, Italy
. Rome: FAO/WHO/UNU; 2004. p. 37, 92.
9. Fruin ML, Rankin JW. Validity of a multi-sensor armband in estimating rest and exercise energy expenditure. Med Sci Sports Exerc
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. Hustvedt BE, Christophersen A, Johnsen LR, et al. Description and validation of the ActiReg: a novel instrument to measure physical activity and energy expenditure. Br J Nutr
12. Jakicic JM, Marcus M, Gallagher KI, et al. Evaluation of the SenseWear Pro Armband to assess energy expenditure during exercise. Med Sci Sports Exerc
13. King GA, Torres N, Potter C, Brooks TJ, Coleman KJ. Comparison of activity monitors to estimate energy cost of treadmill exercise. Med Sci Sports Exerc
14. Malavolti M, Pietrobelli A, Dugoni M, et al. A new device for measuring resting energy expenditure (REE) in healthy subjects. Nutr Metab Cardiovasc Dis
15. Papazoglou D, Augello G, Tagliaferri M, et al. Evaluation of a multisensor armband in estimating energy expenditure in obese individuals. Obesity
16. Perret C, Mueller G. Validation of a new portable ergospirometric device (Oxycon Mobile) during exercise. Int J Sports Med
17. Paróczai R, Kocsis L. Analysis of human walking and running parameters as a function of speed. Technol Health Care
18. Pärkkä J, Ermes M, Korpipää P, Mäntyjärvi J, Peltola J, Korhonen I. Activity classification using realistic data from wearable sensors. IEEE Trans Inf Technol Biomed
19. 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
20. Rothney MP, Neumann M, Beziat A, Chen KY. An artificial neural network model of energy expenditure using nonintegrated acceleration signals. J Appl Physiol
21. Strath SJ, Bassett DR Jr, Swartz AM, Thompson DL. Simultaneous heart rate-motion sensor technique to estimate energy expenditure. Med Sci Sports Exerc
22. Strath SJ, Bassett DR Jr, Thompson DL, Swartz AM. Validity of the simultaneous heart rate motion sensor technique for measuring energy expenditure. Med Sci Sports Exerc
23. Swartz AM, Scott J, Bassett DR Jr, O'Brein WL, King GA, Ainsworth BE. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites. Med Sci Sports Exerc
24. Treuth MS, Adolph AL, Butte NF. Energy expenditure in children predicted from heart rate and activity calibrated against respiration calorimetry. Am J Physiol
25. Walker JL, Murray TD, Jackson AS, Morrow JR Jr, Michaus TJ. The energy cost of horizontal walking and running in adolescents. Med Sci Sports Exerc
26. Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism. J Physiol
27. Welk GJ, McClain JJ, Eisenmann JC, Wickel EE. Field validation of the MTI Actigraph and BodyMedia armband monitor using the IDEEA monitor. Obesity
28. Zhang K, Pi-Sunyer FX, Boozer CN. Improving energy expenditure estimation for physical activity. Med Sci Sports Exerc