Given the limitations of self-report methods and the high cost and participant burden associated with other objective methods (e.g., HR monitoring, doubly labeled water), accelerometry has become the method of choice for measuring physical activity in free-living children and adolescents (21-23). Presently, several accelerometer makes and models are commercially available. However, one product in particular, the ActiGraph, has received a substantial amount of research attention and is one of the most widely used accelerometer-based motion sensors in studies involving children and adolescents (25).
Despite the widespread use of the ActiGraph monitor, considerable controversy exists about how to convert its output (counts per unit of time) into units of energy expenditure (EE) or estimates of physical activity intensity (1,24). To date, the most common approach has been to establish intensity-related cut points from a single regression equation that describes the linear or nonlinear relationship between counts and EE, but other methods such as receiver operating characteristic (ROC) curves have been used (1,6).
To date, at least five sets of youth-specific ActiGraph cut points have been independently developed and published in the peer-reviewed scientific literature (2,3,9,15,20). These cut points and their respective prediction equations are described in Table 1. To varying degrees, each equation and its associated cut points take into account the influence of growth and development on EE and accelerometer output. For example, the Freedson and Mattocks prediction equations include age as an independent variable in the regression model. The other equations/algorithms at least control for age-related declines in weight relative resting metabolic rate by expressing the energy cost of physical activity as multiples of resting EE (METs) or net EE (activity EE (AEE)) (4,15). Importantly, the method used to derive these cut points varied considerably from study to study. Some cut points were derived from samples with a large age range, whereas others were derived from a narrow age range or single age group. Some were derived from a single-sex group, whereas others were derived from both sexes. One set of cut points was based purely on treadmill walking and running, whereas the others were derived from a variety of free-play or lifestyle activities. One equation provided age-specific count cut points, whereas the others derived a single cut point for children and adolescents of different ages. Some were based on direct measures of EE, whereas others were based on direct observation.
;)
TABLE 1: Youth-specific predictions equations and/or cut points for the ActiGraph accelerometer.
The existence of multiple sets of intensity-related cut points for children and adolescents has significantly hindered research efforts to quantify, understand, and intervene on youth physical activity behavior. Indeed, the lack of consensus on cut point selection and the widespread practice of deriving new calibration equations/cut points for a single population group or single study has created what we refer to as the cut point conundrum (24). When reducing accelerometer data to behavioral end points (e.g., minutes of moderate- to vigorous-intensity physical activity (MVPA) per day), researchers must choose among multiple sets of cut points that differ significantly in magnitude and, as a result, report dramatically contrasting estimates of physical activity participation. Moreover, this choice is made in the absence of empirical studies simultaneously comparing the predictive validity of the cut points using a standardized protocol and a recognized criterion measure of physical activity intensity. The absence of comparative validity data is a serious gap in the research literature that must be urgently addressed if the field is to move toward a standardized approach to accelerometer data reduction. Therefore, the purpose of this study was to compare the classification accuracy of five previously published sets of ActiGraph cut points for children and adolescents using EE, measured via portable indirect calorimetry, as a criterion measure.
METHODS
Participants
A total of 206 children and adolescents between the ages of 5 and 15 yr participated in the study. The sample was evenly distributed across the age range and contained approximately equal numbers of boys and girls. Descriptive characteristics are presented in Table 2. Before participation, parental written consent and child assent were obtained. The study was approved by the institutional review boards of Oregon State University and Michigan State University.
TABLE 2: Participants' characteristics (n = 206).
Protocol
Participants performed 12 standardized activity trials. The trials were completed over two laboratory visits scheduled within a 2-wk period. Participants were asked to refrain from eating 2-3 h before each laboratory visit. On visit 1, participants completed the following six trials: lying down, hand writing, laundry task, throw and catch, comfortable overground walk, and aerobic dance. Visit 1 concluded with a 5-min treadmill familiarization trial. On visit 2, participants completed the remaining six trials: computer game, floor sweeping, brisk overground walk, basketball, overground run/jog, and brisk treadmill walk. Consistent with the recommendations of Welk (28), the selected activities ranged in intensity from sedentary to vigorous, included "free-living" or "lifestyle" physical activities typically performed by children and adolescents, and included both ambulatory and intermittent free-play activities. In addition, all 12 of the selected activities had been used in previously published accelerometer validation and calibration studies. A brief description of each activity trial is provided in Table 3. The script and other detailed instructions related to the administration of the activity trials are available from the authors on request.
TABLE 3: Description of the 12 activity trials.
Each activity trial lasted 5 min with the exception of the lying down trial, which lasted 10 min. To assist with motivation and to ensure even pacing during the overground walking and running trials, a research assistant walked/jogged alongside each participant. Verbal feedback was provided if the research assistant felt that the pace was inappropriate. The walking speed during the treadmill walk was set to equal the walking speed achieved during the brisk overground walking trial.
Instrumentation
Indirect calorimetry.
Oxygen uptake (O2) during each activity was measured on a breath-by-breath basis using the Oxycon Mobile (Yorba Linda, CA), a lightweight (950 g) portable indirect calorimetry system. A flexible facemask (Hans Rudolph, Kansas City, MO) held in place by a head harness covered the participant's nose and mouth. The mask was attached to a bidirectional rotary flow and measurement sensor (Triple V) to measure the volume of inspired and expired air. A sample tube running from the Triple V to the analyzer unit delivered expired air for the determination of O2 and CO2 content. Before each test, the Oxycon unit was calibrated according to manufacturer's guidelines. Flow control and gas calibration were performed using Oxycon's automated calibration system, with the CO2 and O2 analyzers calibrated against room air as well as to a reference gas of known composition (4% CO2 and 16% O2). The Oxycon Mobile has been shown to provide valid measures of oxygen uptake over a range of exercise intensities (5,18).
Accelerometry.
The ActiGraph GT1M (ActiGraph, LLC, Fort Walton Beach, FL) measures and records time-varying accelerations ranging in magnitude from approximately 0.05g to 2.5g. The accelerometer output is digitized by a 12-bit analog-to-digital converter at a rate of 30 Hz. Once digitized, the signal passes through a digital filter that band limits the accelerometer to the frequency range of 0.25-2.5 Hz. The filtered signal is then rectified and integrated over a user-specified interval known as an epoch. At the end of each epoch, the summed value or "activity count" is stored in memory and the integrator is reset. To facilitate synchronization with the indirect calorimetry data, the current study used 1-s epochs. Before each testing session, the ActiGraph was initialized according to the manufacturer's specifications and attached to a flexible elastic belt that was fastened snugly around the waist of the participant. The ActiGraph was positioned on the right midaxilla line at the level of the iliac crest.
Data Reduction
Before each test, the ActiGraph and Oxycon units were synchronized to an external timepiece. After download of the data from the respective instruments, customized software was used to calculate mean V˙O2 and mean counts per second recorded between minutes 2.5 and 4.5 of each activity trial. For the lying down trial, mean V˙O2 and mean counts per second were calculated from data collected between minutes 7.0 and 9.0. For each participant, the attainment of steady state was confirmed by inspection of recorded HR and V˙O2 values. Tolerance levels were ±5 bpm and ±10% for HR and V˙O2, respectively. Mean V˙O2 was converted into units of EE (kcal·min−1 and kcal·kg−1·min−1) using the Weir equation (27). METs were calculated by dividing mean weight relative V˙O2 by resting EE (REE). Activity EE values (AEE) were computed by subtracting REE from total EE. REE was predicted from the participant's sex, age, body mass, and height using Schofield's (19) equation for children aged 3-10 or 10-18 yr.
Prediction of EE.
After reintegration to 60- or 30-s epochs (depending on the equation), mean ActiGraph counts were converted into units of EE using the Freedson/Trost (FT), Puyau (PU), Treuth (TR), and Mattocks (MT) prediction equations. There was no prediction equation associated with the Evenson (EV) cut points. Because the equations were developed to identify cut points for MVPA and had y-intercepts significantly higher than typical REE values, all predictions were made using a count "flex point" of 100 counts per minute-a widely adopted count threshold for sedentary activity (8). Thus, for these calculations, predicted EE values were constrained to resting levels (i.e., 1 MET or AEE = 0) when count values for a given activity trial were <100 counts per minute.
Prediction of physical activity intensity.
Activity trials were classified as sedentary, light-, moderate-, or vigorous-intensity physical activity using the count cut points derived from each prediction equation or ROC curve analysis (Table 1). To evaluate the classification accuracy of these cut points, activity trials were also classified as sedentary, light-, moderate-, or vigorous-intensity physical activity on the basis of measured EE. For METs, the following classification scheme was adopted. Sedentary activity (SED) was defined as <1.5 METs. Light activity (LPA) was defined as ≥1.5 and <4 METs. Moderate activity (MPA) was defined as ≥4 and <6 METs. Vigorous activity (VPA) was defined as ≥6 METs. There has been debate within the field regarding the selection of MET intensity thresholds for children and adolescents (4,17). The MET thresholds used in the present study were selected because they closely approximated the intensity thresholds used in most of the original calibration studies (2,9,20). Although some calibration studies (3,15) used the adult value of 3 METs or its equivalent to define MPA, there is consistent evidence that brisk walking, a key behavioral indicator of moderate-intensity physical activity, is associated with an energy cost of approximately 4 METs in children and adolescents (9,13,16,20). For AEE, the following classification scheme was used. Sedentary activity was defined as <0.015 kcal·kg−1·min−1. Light activity was defined as ≥0.015 and <0.05 kcal·kg−1·min−1. Moderate activity was defined as ≥0.05 and <0.10 kcal·kg−1·min−1. Vigorous activity was defined as >0.10 kcal·kg−1·min−1. These thresholds were selected because they replicated the AEE intensity thresholds identified by Puyau et al. (15).
Statistical Analyses
Differences between measured and predicted EE values were evaluated using dependent t-tests with a Bonferroni adjustment for multiple comparisons (P < 0.004). Classification accuracy for each set of cut points was evaluated by calculating weighted κ statistics, sensitivity (Se), specificity (Sp), and area under the receiver operating characteristic curve (ROC-AUC). An area of 1 represents perfect classification, whereas an area of 0.5 represents an absence of classification accuracy. ROC-AUC values of ≥0.90 are considered excellent, 0.80-0.89 good, 0.70-0.79 fair, and <0.70 poor (11). To confirm that classification accuracy was comparable across the entire sample, we also evaluated sex and age group differences in classification accuracy. All statistical analyses were performed using SAS Version 9.1 (Cary, NC).
RESULTS
Of the 2472 possible activity trials (206 × 12 = 2472), complete V˙O2 and ActiGraph data were available for 2313 trials (93.6%). Trials were excluded if 1) the accelerometer failed to initialize or download, 2) the Oxycon Mobile malfunctioned, 3) V˙O2 failed to meet the criteria for steady state, or 4) participants were absent, failed to complete the entire trial, or did not follow the instructions. Table 4 displays descriptive statistics for V˙O2, METs, and ActiGraph counts for the 12 activity trials. On the basis of average MET level, the lying down and computer game trials fell into the SED category; the hand writing, throw and catch, laundry, sweeping, aerobic dance, and comfortable walking trials were considered LPA; the brisk overground and treadmill walking trials were considered MPA; whereas the basketball and running trials were considered VPA. There was, however, substantial individual variability in the energy cost of each trial, and many of the activity trials were completed at absolute intensities ranging from light to vigorous. Median ActiGraph counts increased in accordance with physical activity intensity and also demonstrated substantial individual variability within each trial.
TABLE 4: Descriptive statistics for V˙O2, METs, and ActiGraph counts for the 12 activity trials.
Differences between measured and predicted EE values for the FT, PU, TR, and MT equations are displayed in Figures 1A, B, C, and D, respectively. The FT equation significantly underestimated mean METs during the lying down, computer game, hand writing, throw and catch, laundry, sweeping, and basketball trials and significantly overestimated mean METs during the slow and brisk overground walking trials. Predicted METs were not significantly different from observed METs during the aerobics, brisk treadmill walking, and overground running trials. The TR equation significantly underestimated mean METs during the lying down, computer game, hand writing, throw and catch, laundry, sweeping, brisk treadmill walking, basketball, and overground running trials and significantly overestimated mean METs during the slow and brisk overground walking trials. Predicted METs were not significantly different from observed METs during the aerobics trial.
FIGURE 1: Measured versus predicted EE values for the Freedson/Trost (FT) (A), Puyau (PU) (B), Treuth (TR) (C), and Mattocks (MT) (D) prediction equations. For FT, PU, and TR, sample sizes are as follows: RE (n = 180), CG (n = 189), HW (n = 188), TC (n = 194), LY (n = 195), SW(n = 184), AE (n = 198), CW (n = 187), BW (n = 201), TM (n = 198), BB (n = 199), and RU (n = 200). For MT, sample sizes are as follows: RE (n=180), CG (n = 189), HW (n = 188), TC (n = 159), LY (n = 137), SW (n = 129), AE (n = 155), CW (n = 161), BW (n = 184), TM (n = 177), BB (n = 182), and RU (n = 194). #Statistically significant (P < 0.004).
For girls between the ages of 5 and 10 yr, the MT equation produced negative PAEE estimates when counts per minute fell below 4879, 3951, 3022, 2093, 1165, and 236, respectively. For boys between the ages of 5 and 10 yr, negative PAEE estimates resulted from counts per minute below 5310, 4381, 3453, 2524, 1594, and 667, respectively. Thus, for the comparison of measured and predicted values, negative predictions were set to missing and excluded from the analysis. On the basis of these data, the MT equation significantly underestimated mean PAEE during the lying down, computer game, hand writing, throw and catch, laundry, sweeping, basketball, and overground running trials. Predicted PAEE was not significantly different from observed PAEE during the aerobics, slow overground walk, brisk overground walk, and brisk treadmill walk trials. However, these nonsignificant differences seemed to be more a function of greater variability in predicted PAEE values than a lack of bias in the estimates. The PU equation significantly underestimated mean AEE in all 12 activity trials.
For the EV, FT, and TR cut points, there were no significant sex or age group differences in classification accuracy for all levels of physical activity intensity. Classification accuracy for the MT and PU cut points did not differ significantly by sex; however, three age-related differences were noted. For the MT, classification accuracy for MVPA among children aged 5-8 yr (ROC-AUC = 0.69, 95% confidence interval (CI) = 0.66-0.73) was significantly lower than children aged 9-10 yr (ROC-AUC = 0.79, 95% CI = 0.76-0.83), 11-12 yr (ROC-AUC = 0.81, 95% CI = 0.77-0.84), and ≥13 yr (ROC-AUC = 0.80, 95% CI = 0.77-0.82). For the PU, classification accuracy for LPA among children aged 5-8 yr (ROC-AUC = 0.34, 95% CI = 0.30-0.37) was significantly lower than children aged 9-10 yr (ROC-AUC = 0.42, 95% CI = 0.39-0.46), 11-12 yr (ROC-AUC = 0.44, 95% CI = 0.41-0.48), and ≥13 yr (ROC-AUC = 0.47, 95% CI = 0.44-0.49). PU classification accuracy for MVPA among children aged 5-8 yr (ROC-AUC = 0.68, 95% CI = 0.65-0.71) was also significantly lower than children aged 9-10 yr (ROC-AUC = 0.76, 95% CI = 0.73-0.79), 11-12 yr (ROC-AUC = 0.79, 95% CI = 0.76-0.82), and ≥13 yr (ROC-AUC = 0.82, 95% CI = 0.79-0.84). No other significant age-related differences were detected (see Table, Supplemental Digital Content 1, https://links.lww.com/MSS/A70, which reports the complete results of the sex and age group comparisons). Given the generally consistent pattern of findings observed across sex and age groups and the widespread practice of applying cut points developed for a single age and/or sex group to more diverse samples of youth, classification accuracy for SED, LPA, MPA, VPA, and MVPA were evaluated using data for the entire sample. The results are reported in Table 5.
TABLE 5: Sensitivity (Se %), specificity (Sp %), and area under the ROC curve (ROC-AUC) values for the classification of sedentary, light, moderate, vigorous activity, and MVPA.
The sedentary activity cut point of 100 counts per minute (used for the FT, TR, MT, and EV cut points) exhibited excellent classification accuracy (ROC-AUC = 0.90). The sensitivity or true-positive rate was 100%, whereas the false-positive rate was approximately 20% (Sp = 79.0%-79.4%). The PU sedentary cut point of 800 counts per minute also provided 100% sensitivity, but with substantially lower specificity (60.7%), resulting in a classification accuracy rating of only fair to good (ROC-AUC = 0.80).
For light-intensity physical activity, classification accuracy for the EV, FT, and TR cut points was fair (ROC-AUC = 0.68-0.70), whereas the MT and PU cut points exhibited poor classification accuracy (ROC-AUC = 0.43-0.64). The generally poor classification accuracy of the LPA cut points was a function of poor sensitivity with the true-positive rate ranging from just 11.8% (PU) to 58.8% (MT).
For moderate-intensity physical activity, the FT, TR, and EV cut points exhibited fair classification accuracy, with the FT (ROC-AUC = 0.79) exhibiting significantly better classification accuracy than the EV (ROC-AUC = 0.74) and TR (ROC-AUC = 0.71). Classification accuracy for the PU and MT cut points was poor (ROC-AUC = 0.56-0.63), primarily the result of poor sensitivity or a high false-negative rate.
For vigorous physical activity, the EV cut point exhibited good classification accuracy (ROC-AUC = 0.84), whereas the FT (ROC-AUC = 0.77) and TR (ROC-AUC = 0.73) exhibited fair classification accuracy. The higher vigorous cut points associated with PU and MT (>6000 counts per minute) exhibited very low sensitivity and poor classification accuracy (ROC-AUC = 0.54-0.66). With the exception of the EV cut point, which provided reasonable sensitivity (73.7%) and excellent specificity (93.8%), the cut points for VPA were highly specific (Sp = 97.5%-100%) but not sensitive (Se = 7.5%-57.1%). For moderate and vigorous activity combined (MVPA), the EV and FT cut points exhibited excellent classification accuracy (ROC-AUC = 0.90), the TR cut point exhibited good classification accuracy (ROC-AUC = 0.85), whereas the classification accuracy for the PU and MT cut points was fair (ROC-AUC = 0.77-0.78).
Applying the rubric of Landis and Koch (7), the EV (κ = 0.68, 95% CI = 0.66-0.70), FT (κ = 0.66, 95% CI = 0.64-0.68) and TR (κ = 0.62, 95% CI = 0.60-0.64) cut points exhibited substantial agreement across all four levels of intensity; however, agreement for the EV and FT was significantly better than the TR. For the MT (κ = 0.54, 95% CI = 0.52-0.56) and PU (κ = 0.36, 95% CI = 0.34-0.38), agreement across all four level of intensity was moderate to fair.
DISCUSSION
The existence of multiple sets of intensity-related cut points for children and adolescents has significantly hindered research efforts to quantify, understand, and intervene on youth physical activity behavior. To help move the field toward consensus on the application of intensity-related cut points for children and adolescents, the present study simultaneously evaluated the classification accuracy of five independently developed and widely applied sets of cut points for the ActiGraph accelerometer. For the classification of MVPA, the EV and FT cut points exhibited significantly better classification accuracy than the TR, MT, and PU cut points. The EV and FT cut points exhibited excellent classification accuracy, the TR good classification accuracy, whereas the MT and PU cut points exhibited only fair classification accuracy. The widely applied cut point of 100 counts per minute for sedentary activity exhibited good to excellent classification accuracy, whereas the higher PU sedentary cut point of 800 counts per minute exhibited fair classification accuracy. Notably, all cut points exhibited lower classification accuracy for light-intensity physical activity, and only the EV cut points exhibited acceptable levels of classification accuracy for all four levels of physical activity intensity. Collectively, these findings support the application of the EV or FT ActiGraph cut points in field-based studies, with the EV cut points being the best overall performer across all intensity levels. Conversely, our findings do not support the continued use of the TR, PU, and MT cut points.
The results are consistent with our previous investigation evaluating the predictive validity of different ActiGraph energy prediction equations during overground walking and running. In that study, the FT cut points exhibited significantly better classification accuracy for MVPA than the PU cut points. On average, the FT equation overestimated the energy cost of walking and running by 13%, whereas the PU equation underestimated the mean energy cost of walking and running by 33% (26). Our findings are also consistent with those of McClain et al. (10), who compared directly observed MVPA to ActiGraph-based estimates of MVPA using the FT, TR, and MT cut points. Compared with directly observed MVPA, the MT and TR cut points underestimated average time spent in MVPA by 39%-74%. In contrast, the MVPA estimates provided by the FT cut points were not significantly different from directly observed MVPA levels.
Equations or algorithms to predict the energy cost of a single epoch (15-60 s in length) were developed for the purpose of constructing cut points and were never intended to provide point estimates of daily expenditure. Nevertheless, comparisons between predicted and measured EE values for the different activity trials provide insight into each cut point's performance relative to the task of classifying the intensity of physical activity. In the present study, all prediction equations tended to underestimate the energy cost of moderate-to-vigorous nonambulatory activities such as aerobic dance and playing basketball, which contributed to the generally modest levels of agreement between predicted and measured intensity level (κ = 0.36-0.68). In addition, the lower classification accuracy exhibited by the light-intensity cut points reflected the tendency of waist-mounted accelerometers to underestimate the energy cost of nonambulatory light-intensity lifestyle activities such as folding laundry, sweeping the floor, and playing catch. Nevertheless, it is important to note that, relative to the FT and TR prediction equations, the PU and MT equations provided considerably larger underestimations of energy cost, contributing to substantial misclassification error across the moderate to vigorous end of the intensity spectrum. This was particularly true for the PU equation that significantly underestimated the energy cost of all 10 nonsedentary activity trials.
In light of the growing body of evidence identifying sedentary behavior as an independent risk factor for several adverse health conditions (12), the assessment of sedentary behavior via accelerometry has become a topic of considerable research interest. The present study evaluated the predictive validity of the widely used 100 counts per minute threshold for sedentary activity. Within our age-diverse sample of just >200 children, this threshold exhibited excellent classification accuracy, with a sensitivity or true-positive rate of 100% and generally acceptable level of specificity. With the development of effective algorithms to differentiate sedentary activity from non-wear time, the use of accelerometers to measure time in sedentary behavior holds great promise (8,12). Nevertheless, the propensity for waist-mounted accelerometers to misclassify static light to moderate-intensity activities such as folding laundry and sweeping as sedentary remains a legitimate concern. Combining accelerometry with other monitoring devices such as thigh-mounted inclinometers may be a viable solution to this problem. However, the feasibility and validity of this approach among younger children will need to be evaluated in future studies.
The use of age-specific cut points versus a single cut point for youth of different ages is a key methodological issue that has not been adequately addressed in the research literature. Although there is a strong theoretical basis for the existence of age-related thresholds (24), only one equation, the FT equation, provides age-specific cut points for MVPA. In the present study, the age-specific FT cut points exhibited excellent sensitivity and specificity for MVPA; however, the single value EV cut point exhibited nearly identical sensitivity and marginally better specificity. This result was somewhat perplexing given that, among children ≤10 yr, the FT cut points are substantially lower (1400-1910 counts per minute) than the EV cut point of 2296 counts per minute. Because excessively low cut points would tend to misclassify light-intensity activity as MVPA (a false positive) and overestimate time spent in MVPA, we decided to compare the number of false positives provided by the two thresholds during the sedentary and light-intensity activity trials.
Consistent with the high level of specificity exhibited by the FT cut points, only 148 of the 1364 sedentary and light-intensity activity trials were misclassified as MVPA. Of these 148 false positives, 73 occurred during comfortable walking, 30 during brisk walking, 33 during aerobic dance, and 12 during brisk treadmill walking. Notably, no false positives were recorded during the sedentary and light-intensity free-living trials. Importantly, just >60% of the false positives occurred in trials completed by children ≤10 yr for whom the MVPA cut point was significantly lower than 2000 counts per minute. In comparison, the EV cut point misclassified just 113 of 1364 non-MVPA trials. Of these 113 false positives, 49 occurred during comfortable walking, 27 during brisk walking, 29 during aerobic dance, and 8 occurred during the brisk treadmill walking. Similar to the FT cut points, none of the sedentary or light-intensity lifestyle activities was misclassified as MVPA. However, in contrast to the FT cut points, less than half (48.7%) of the false positives occurred in trials completed by children aged ≤10 yr. Moreover, the EV cut point provided substantially fewer false positives during the comfortable walking trial (49 vs 73).
To further assess the effect of the lower age-specific cut points on the false-positive rate for MVPA, we calculated the relative odds of misclassifying light-intensity activities as MVPA among children ≤10 yr compared with those >10 yr. Applying the FT age-specific cut points, children ≤10 yr were 2.3 times more likely than older children to misclassify light-intensity walking and aerobic dance as MVPA (OR = 2.3, 95% CI = 1.6-3.2). In contrast, age was not associated with the likelihood of misclassifying light intensity walking and aerobic dance as MVPA when the Evenson MVPA cut point was applied (OR = 1.2, 95% CI = 0.8-1.8).
Altogether, the above analyses demonstrate that, despite exhibiting comparable levels of sensitivity and specificity for MVPA when examined over all 12 activity trials, the lower cut points associated with the FT equation for children aged ≤10 yr may overestimate time spent in MVPA because of their tendency to misclassify light-intensity activities such as slow walking as MVPA. Among children >10 yr, however, the FT and EV cut points for MVPA provided comparable classification accuracy because they are comparable in magnitude. Moreover, our findings suggest that a single MVPA cut point of around 575 counts per 15 s (2300 counts per minute) may function equally well for youth between the ages of 6 and 15 yr because age-specific cut points seem to increase rather than decrease the likelihood of age-related misclassification errors. However, longitudinal validity studies comparing the classification accuracy of different cut points over time are needed to confirm this finding.
The present study had several limitations that warrant consideration. First, to preserve internal validity and obtain steady-state measures of EE, participants completed a series of controlled activity trials, which do not accurately reflect the intermittent activity patterns of free-living children and adolescents. Accordingly, additional research is needed to evaluate the predictive validity of the different ActiGraph cut points for youth under free-living conditions. Second, because the selected lifestyle activities (folding laundry, sweeping, throw and catch) were well below the moderate-intensity threshold, the study would have benefited from inclusion of additional moderate-intensity lifestyle trials that were nonambulatory in nature. Third, because our resting trial did not adhere to established guidelines for assessment of resting EE, we used predicted values to calculate MET and AEE scores. However, measured MET and AEE values were in close agreement with previously published values (2,3,9,15,16,20). Offsetting these limitations were several notable strengths. First, our sample size of just >200 children and adolescents between the ages of 6 and 15 yr is significantly larger and more age-diverse than any previously published accelerometer calibration/validation study. Second, by including a variety of commonly performed intermittent lifestyle and ambulatory activities ranging from sedentary to vigorous in intensity, our study design adhered closely to the best practice recommendations described by Welk (28). Finally, to account for the considerable between-subject variability in the energy cost of performing a given physical activity (14), we used EE, measured by portable indirect calorimetry, as a criterion measure. This was an important design feature because some calibration/validation studies have relied on direct observation to derive intensity-related cut points.
In summary, the EV and FT MVPA cut points exhibited significantly better classification accuracy than the TR, MT, and PU cut points. MVPA cut points of ≥3000 counts per minute were associated with an increased false-negative rate, significantly lower sensitivity, and marginal classification accuracy. Of the five sets of cut points examined, only the EV cut points provided acceptable classification accuracy for all four levels of physical activity intensity. In addition, the EV MVPA cut point performed well among children of all ages. In contrast, the lower FT MVPA cut points for children aged 10 yr and younger were associated with an increased likelihood of misclassifying light-intensity activities as moderate, leading to potential overestimations of time spent in MVPA. The widely applied sedentary cut point of 100 counts per minute exhibited excellent classification accuracy. However, the inability of waist-mounted accelerometers to accurately record the energy cost of static light-intensity activities will continue to be a limitation. On the basis of these findings, we recommend that researchers use the EV ActiGraph cut points to estimate time spent in sedentary, light-, moderate-, and vigorous-intensity activity in children and adolescents.
This study was supported by funding by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD R01 55400).
The authors thank Matthew Pfeiffer for writing customized data reduction software.
The authors have no conflict of interest to declare.
The results of the present study do not constitute an endorsement by the American College of Sports Medicine.
REFERENCES
1. Bassett DR, Rowlands AV, Trost SG. Calibration and validation of wearable monitors.
Med Sci Sports Exerc. (in press).
2. Evenson KR, Cattellier D, Gill K, Ondrak K, McMurray RG. Calibration of two objective measures of physical activity for children.
J Sports Sci. 2008;26:1557-65.
3. Freedson PS, Pober D, Janz KF. Calibration of accelerometer output for children.
Med Sci Sports Exerc. 2005;37(11 suppl):S523-30.
4. Harrell JS, McMurray RG, Baggett CD, Pennell ML, Pearce PF, Bangdiwala SI. Energy costs of physical activities in children and adolescents.
Med Sci Sports Exerc. 2005;37(2):329-36.
5. Hendrickson NR, Harmon J, Burckhard J, Bacharach DW. Validity of the Oxycon Mobile open-circuit spirometry system.
Med Sci Sports Exerc. 2007;39(5 suppl):S178.
6. Jago R, Zakeri I, Baranowski T, Watson K. Decision boundaries and receiver operating characteristic curves: new methods for determining accelerometer cut points.
J Sports Sci. 2007;25:937-44.
7. Landis J, Koch G. The measurement of observer agreement for categorical data.
Biometrics. 1977;33:159-74.
8. Matthews CE, Chen KY, Freedson PS, et al. Amount of time spent in sedentary behaviors in the United States, 2003-2004.
Am J Epidemiol. 2008;167:875-81.
9. Mattocks C, Leary S, Ness A, et al. Calibration of an accelerometer during free-living activities in children.
Int J Pediatr Obes. 2007;2:218-26.
10. McClain JJ, Abraham TL, Brusseau TA, Tudor-Locke C. Epoch length and accelerometer outputs in children: comparison to direct observation.
Med Sci Sports Exerc. 2008;40(12):2080-7.
11. Metz CE. Basic principles of ROC analysis.
Semin Nucl Med. 1978;8:283-98.
12. Owen N, Healy GN, Matthews CE, Dunstan DW. Too much sitting: the population health science of sedentary behavior.
Exerc Sports Sci Rev. 2010;38;105-13.
13. Pate RR, Stevens J, Pratt C, et al. Objectively measured physical activity in sixth-grade girls.
Arch Pediatr Adolesc Med. 2006;160;1262-8.
14. Pfeiffer KA, Schmitz KH, McMurray RG, Treuth MS, Murray DM, Pate RR. Variability in energy expenditure of selected activities in adolescent girls.
Am J Prev Med. 2006;31:328-31.
15. Puyau MR, Adolph AL, Vohra FA, Butte NF. Validation and calibration of physical activity monitors in children.
Obes Res. 2002;10:150-7.
16. Ridley K, Ainsworth BE, Olds TS. Development of a compendium of energy expenditures for youth.
Int J Behav Nutr Phys Act. 2008;5:45.
17. Ridley K, Olds TS. Assigning energy costs to activities in children: a review and synthesis.
Med Sci Sports Exerc. 2008;40(8):1439-46.
18. Rosdahl H, Gullstrand L, Salier-Eriksson J, Johansson P, Schantz P. Evaluation of the Oxycon Mobile metabolic system against the Douglas bag method.
Eur J Appl Physiol. 2010;109:159-71.
19. Schofield WN. Predicting basal metabolic rate, new standards and review of previous work.
Hum Nutr Clin Nutr. 1985;39(1 suppl):5-41.
20. Treuth MS, Schmitz K, Catellier DJ, et al. Defining accelerometer thresholds for activity intensities in adolescent girls.
Med Sci Sports Exerc. 2004;36(7):1259-66.
21. Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer.
Med Sci Sports Exerc. 2008;40(1):181-8.
22. Troiano RP. A timely meeting: objective measurement of physical activity.
Med Sci Sports Exerc. 2005;37(11 suppl):S487-9.
23. Trost SG. Objective measurement of physical activity in youth: current issues, future directions.
Exerc Sports Sci Rev. 2001;29(1):32-6.
24. Trost SG. Measurement of physical activity in children and adolescents.
Am J Lifestyle Med. 2007;1:299-314.
25. Trost SG, McIver KL, Pate RR. Conducting accelerometer-based activity assessments in field-based research.
Med Sci Sports Exerc. 2005;37(11 suppl):S531-43.
26. Trost SG, Way R, Okely AD. Predictive validity of three ActiGraph energy expenditure equations for children.
Med Sci Sports Exerc. 2006;38(2):380-7.
27. Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism.
J Physiol. 1949;109:1-9.
28. Welk GJ. Principles of design and analyses for the calibration of accelerometry-based activity monitors.
Med Sci Sports Exerc. 2005;37(11 suppl):S501-11.
