In this study, we developed a regression equation to predict MET score from the accelerometer counts. The model fit was good, as evidenced by a mixed model equivalent to R2
for linear regression of 0.84, even so, the large standard error of ±1.36 METs might suggest that the equation be used only as a gross estimate of energy expenditure. The use of the MET score for the prediction is a unique approach as previous papers typically use caloric expenditure in kilocalories (5
). The use of MET values also allows for the use of the Compendium of Physical Activity to estimate energy expenditure over a large range of activities (1
). Our equation will be useful to others interested in predicting a MET score with their own accelerometer data. It has already been useful in defining moderate-to-vigorous physical activity for the main trial data from TAAG. As well, this equation can be used in reducing the accelerometer data collected in TAAG.
There is relatively little other research using accelerometry to predict energy expenditure in children. In one study (17
), activity energy expenditure was predicted from Actigraph accelerometer counts using controlled laboratory conditions in a room calorimeter as well as free-living field activities. Both boys and girls of varying ages participated and a high correlation was found (R2
adj = 0.75). In adults, other laboratory-based studies have compared accelerometers to energy expenditure during physical activity (5
). Freedson et al. (5
) predicted energy expenditure (kcal·min-1
) measured in adults while walking at different speeds on the treadmill from Actigraph counts and BMI, reporting an R2
of 0.82. In two studies of adults by Welk et al. (26
), validation coefficients between several accelerometers (Actigraph, Biotrainer, and Actitrac) with energy expenditure measured under treadmill and field conditions were also moderate to high (range r = 0.62-0.91). Thus, our equation to estimate MET levels from accelerometer counts in adolescent girls appears comparable to previous reports in children and adults.
In order to define our activity thresholds appropriately, we utilized the
values above each girl’s own resting metabolic rate. One MET is defined as 3.5 mL·kg-1
for adults (1
), and in our study, this value was slightly higher (3.8 ± 0.9 mL·kg-1
). Utilizing the adult value for 1 MET might have biased the estimated energy cost of the physical activities measured in these adolescent girls. Thus, we chose to use the participants’ actual resting values to estimate MET levels.
Our three different approaches for determining our thresholds identified activities that were falsely classified. In approach 1 where bicycling was excluded, counts for watching TV, playing a computer game, sweeping the floor, and slow walking that were above the threshold were classifiedas false positive for moderate intensity activity, and counts for brisk walking, stair walking, shooting baskets, and running below the same threshold were classified as false negative. There were high numbers of false negatives, mostly related to step aerobics. This may have been expected since the accelerometer senses a walk (or step), but the participant had to lift her weight up a 6-inch step each step; thus, metabolic rate would be increased above walking. In approach 2, step aerobics and bicycling were both removed from the analysis. The numbers of false positives and false negatives were lower than excluding bicycling alone. This suggests that the accelerometer does not respond appropriately during activities in which body motion is somewhat independent of exercise intensity. Approach 3 had the lowest number of false positive and false negative classifications. This approach used differences between a slow and brisk walk to identify thresholds. Brisk walking has previously been categorized as a moderate activity (24
). In particular, the recent guidelines for meeting the minimum level of physical activity levels use brisk walking as an example of a moderate activity for individuals (2
). Thus, although we tested many activities with a wide range of intensities, a potentially limiting factor in the study is that we relied on differentiating between two walking speeds to determine our moderate threshold.
We experienced some difficulties using the accelerometer to establish the thresholds for light and vigorous activity. In our attempt to define a vigorous threshold, we designed the study to include activities that would be at an intensity > 6 METs. Previous literature has defined 6 METs as vigorous (24
). As shown in , shooting baskets and the stair walking yielded MET values greater than 6 METs, but the activity counts were very low. As adolescents perform activities in variable ways, it is evident that the accelerometer does not work well in evaluating the intensity of such activities. Perhaps accelerometers attached to the waist lack in their ability to account for upper extremity work. It is also possible that the accelerometers might detect movement when in fact the adolescent was sedentary (e.g., car/bus travel). This could have influenced the number of false negatives for the sedentary threshold. Thus, accelerometers are limited in their ability to estimate activity levels or metabolic rates during cycling, step aerobics, stair stepping, or other similar activities where the accelerometer counts are less accurate.
One study (17
) evaluated similar cut-points in children using the Actigraph accelerometer during treadmill walking/running and jogging outside at their own speed. A slightly higher value for the lower threshold for moderate activity was reported (<3200 counts·min-1
). This corresponded to <0.05 kcal·kg-1
). Their reported threshold for vigorous activity was >8200 counts, with sedentary defined as <800 counts·min-1
. Our values were lower and may be due to the types and intensities of activities chosen to determine these thresholds. For example, vigorous activity included treadmill walking (that specifically controlled speed and step rate) and jogging on a track (17
), whereas we did not use a treadmill. Greater variability would be expected with our run at 5 mph (8 kph). Our participants had to follow a set pace and were verbally encouraged to either speed up or slow down if the pace deviated.
Because we evaluated the fitness levels of the participants using a cycle ergometer test (PWC170
), we were able to verify how our chosen thresholds relate to O2max
. Moderate physical activity should correspond to 40-60% of O2max
. In these participants with a predicted average O2max
of ~39 mL·kg-1
, 40-60% would be equivalent to 16 -23 mL·kg-1
. The measured
values for the slow walk in our study are below this range whereas the value for brisk walking (4.3 METs and
= 15.6 mL·kg-1
) was at the low end for the range of 40 - 60% O2max
. Thus using slow and brisk walking as activities to separate light from moderate activity to define our activity thresholds appears to be appropriate. We can also compare our chosen activities and measured
to a percentage of this estimated O2max
. The 11 chosen activitierange from 11% to 79% of predicted O2max
. Thus, we can be confident that our prediction equation for MET score was developed from the entire range of exercise intensities. One limitation of our prediction is that these values were predicted from cycle ergometry, which is lower than what would be predicted from treadmill walking. Nevertheless, we did include a wide range of exercise intensities into the design of the study.
One unique component of this study was the setting of the accelerometers to collect activity counts in 30-s intervals. Children’s activities tend to be spontaneous, sporadic, and done in shorter intervals. Our rationale for this shorter data collection interval was that sensitivity might be increased and the activities might be better detected with the 30 s rather than 1 min collection time. We presented the data also in counts per minute for our thresholds, as this has been the standard in the literature (5
Based on our study, a prediction equation for estimating MET score from accelerometer counts is now available for use in adolescent girls. The accelerometer thresholds to define girls’ physical activity levels can also be used to determine the amount of sedentary, light, moderate, and vigorous physical activity. This type of determination is useful for surveillance research or to determine the effects of physical activity interventions designed to promote activity.