The recent GRANADA consensus on accelerometry [10], determined to establish future lines of research that include different analytical approaches to measure SB and PA by accelerometry. Therefore, the present study compares the application of six validated accelerometry protocols based on specific cut-off points to evaluate PA, showing very different estimations for SB and PA intensity levels in children; even considering the epoch lengths used in their validation studies; and independently of the NWT algorithm selected, NWT-20 or NWT-60, or the puberty stage, age and BMI.
One of the first aspects to be considered are the selection of the two possible criteria for NWT, the NWT-20 proposed by Cain et al. [7] or the NWT-60 proposed by Troiano et al. [15]. Both proposals showed several differences when compared among authors, with the exception that NWT-60 criteria led to a higher SB, and NWT-20 criteria that accumulated more time on PA intensities. For adults, NWT-20 has shown the lowest misclassification error, although it presents the inconvenience that it may result in slightly greater data loss (6% of the sample size) [9,23]. As the precision between NWT-20 and NWT-60 seems similar, the literature has suggested using NWT-60 without allowing interruptions in the collect criterion of counts as a general recommendation for adults [9]. However, in children, it could be very different. Thus, in our study, only a loss of 1.5% of the participants (n=8) with NWT-20 was detected. Therefore, it seems more adequeate NWT-20 for children. However, more studies are needed to examine the accuracy of different NWT detection algorithms in all age groups of children and adolescents.
Although it is difficult to establish a recommendation, in the present study, there are differences between both NWT. These results might be due to the time interval that must elapse without counts for NWT-60 is greater than in NWT-20; so, NWT-60 criteria may be interpreting this time as SB instead of NWT. Despite children or adolescents with OW or OB might present a longer time of consecutive 0 counts per minute (CPM) associated with a higher SB [24], especially in prepubertal children [25], this time could be misclassified as NWT.
Traditionally, SB and PA intensity have been estimated based on the number of CPM accumulated in a given period (length of time). The cut-off points are the thresholds of the activity counts used to categorize the activity as light, moderate and vigorous PA. This study selected 6 validated protocols based on different cut-off points and standards for PA interpretation. Although other standard measures can be found in the literature with similar mean cut-off points [7,9], the accelerometry protocol criteria were selected to represent the group of protocols more frequently used to estimate PA in school-age children, mainly with Actigraph accelerometers. These are also the protocols provided by ActiLife for estimating PA in school-aged children [7,9].
To calibrate the different range of accelerometer counts corresponding to predefined SB, the intensity levels or to estimate energy expenditure, authors usually involved movements as walking, running or stationary bicycle (only in the case of Evenson et al. [18]) alone or in combination with free-living activities (TV watching, arts and crafts) [21] in their study protocols. However, the methods used to analyze and quantify the physiological response of participants were different in each accelerometry protocol, such as: oxygen consumption (VO2) and the heart rate [18]; refitting the energy expenditure model with VO2 as the outcome [20]; calibrated against energy expenditure measures (kcal.kg−1.hr−1) obtained over a range of exercise intensities using a COSMED K4b2 portable metabolic unit [19]; 6‐hour energy expenditure measurements by room respiration calorimetry, activity by microwave detector, and heart rate by telemetry [22]; reviewed the calibration of different accelerometers used most frequently to assess PA and SB in children [21]; or based on the results of the National Health and Nutritional Examination Survey (NHANES)’s [15]. Although most of the accelerometry protocols used objective, validated, and standardized methods to associate the movement with their physiological response, Puyau et al. [22] seems to present a more controlled environment, specially to measure the SB, furthermore, Evenson et al. [18] used a robust statistical analysis compared to other accelerometry protocols.
On the other hand, the accelerometry protocols included in the present study only used the vertical axis to measure the movement. Nevertheless, the current Actigraph models (as GT3X) also include two more axes. Even though it has been verified that the Actigraphs with a single vertical axis are comparable with those with a triaxial axis [26,27], new protocols are trying to get recognition by the scientific community and the Actigraph Corporation for the GT3X model [28–31]. These accelerometry protocols were not considered for the present study, as they did not provide cut-off points for the different PA intensities.
The number of epochs established at the set-up moment also seems to determine the protocol precision. In a recent systematic review, Migueles et al. [9] recommend for children the Hänggi et al. [31] cut-off points developed in 1-sec epoch for the hip due to the excellent classification accuracy (ROC-AUC > 0.90 for all cut-points) obtained and the cover for almost the whole spectrum of PA intensities. The ranges obtained by Hänggi et al. [31] were "<3 counts for SB, 3–56 counts for LPA and > 56 counts for MVPA". If values from 1-sec epoch to 60-sec epochs are transformed, the results are within the following ranges "<180 for SB, 180- 3360 for LPA and > 3360 for MVPA". These are very similar to those proposed by Mattocks' et al. [20] (SB: ≤ 100; LPA: 101-3580; MPA: 3581-6129; VPA: ≥ 6130), that was the protocol included in the present study. The latter has the advantage of being able to study separately MPA and VPA intensities.
Once analyzed the accelerometer protocols, it was found that the lower value in min obtained for SB was 471/504.7 min [18] vs 631.5/663.7 min in the upper value [22], respectively for NWT-20 and NWT-60; and highlighting the MVPA which was 21/20.7 min [20] vs 180.2/178.4 min [21] (Table 3). Indeed, the current literature reporting PA children data measured by accelerometry must be interpreted with caution, paying attention to the analysis protocol when comparing one study's results with others [11].
To evaluate PA in children is essential to consider age and puberty stage. All selected accelerometry protocols for this study included criteria for school-age children and some of them for adolescents. Although the puberty stages were not specified in the protocols' validation, the age range was between 5 to 19 years. Usually, the studies include an age range higher than a couple of years and usually comprises children from 5-6 to 14. During puberty, males gain greater amounts of fat free mass and skeletal mass, whereas females acquire significantly more fat mass [32]. Therefore, an age range very wide, e.g., 6-18 years, may lead to a less specific measure. Only Mattocks et al. [20], and Puyau et al. [22] did not show differences when compared by puberty stage or age quartiles in the MPA and MVPA intensities, both for NWT-20 and NWT-60. This may indicate that both accelerometry protocols seem to be less precise for ages outside those included in their protocols.
Other consideration is that all the selected protocols used children with NW to validate their cut-off, except for Troiano et al. [15], who included 2-3% of OW, but no OB. Despite establishing various cut-off by each accelerometry protocol, no differences were obtained by BMI category in SB and the different PA intensities comparing them. The fact that none of the accelerometry protocols has included OB children in their validation makes wonder whether the estimates of SB and PA in children with OB measured with accelerometry are reliable or not, considering that this methodology is commonly used in the evaluation or in interventions related with childhood obesity [33]. This question has not been exactly resolved so far, although few studies have provided approximations and interesting data [12,34–36]. Robertson et al. [34] conducted an investigation only in children with OB, concluding that accelerometers are acceptable to most of the children, although their use at school is problematic for some of them because they may underestimate children's PA, as some children with OB are unwilling to wear accelerometers at school and during sports because they feel they are at risk of stigma and bullying. The aim of Moura et al. study [35] was to analyze the impact of cut-off points in defining SB time and prevalence in adolescents from Northeastern Brazil. Also in this context, Migueles et al. [12], aimed to examine how cut‐points relative to different attachment sites affect the final estimations of SB and PA in children with overweight/OB. Similar to our study, the cut-off points examined by them produced significant differences in SB and PA estimates. Gaba et al. [36] reported a curvilinear analysis that indicated the optimal thresholds for CPM and MVPA derived from the Puyau et al. [22], which was very useful in classifying children according to their BMI and fat mass percentage to overweight and obesity prevention but only considering MVPA.
According to the Bland-Altman plots, the accelerometry protocol of Evenson et al. [18] method showed large mean differences with that of Puyau et al. [22] for SB and LPA, which had a more controlled environment to validate cut-off points with energy expenditure. However, despite there being also differences, the mean differences were lower for Eveson et al. [18] vs. Puyau et al. [22], and Evenson et al. [18] vs Mattocks et al. [20] for MPA, VPA and MVPA.