The aim of the present study was the calibration of the Move4 accelerometer for children aged 8–13 years. In more detail, firstly mean values for the selected activities (in mg for MAD and MAI metric) were assessed and then these activities and the energy expenditure were validated by using the heart rate. Secondly, cut-off points were modelled and determined to distinguish different intensity levels by using two different metrics (MAD and MAI) as well as by differentiating between the four sensor positions.
Validity
First of all, the determination of activities and energy expenditure level was validated by using the heart rate measures. Overall, a strong correlation between the heart rate and the MAD values (in mg) of each activity was found. With increasing MET values, the heart rate of the activities increased. Based on this finding, it is possible to determine the cut-off points by using the MAD/MAI values of the activities. The present results are consistent with previous reports stating that heart rate is effective in detecting a variety of activity patterns [40]. Nevertheless, as we randomized the order of the activities, it was possible, that in some groups SB activities followed right after VPA activities. In that case, the three minutes rest could not have been enough time to normalize the heart rate. This could explain slightly higher heart rates of 105 bpm for SB compared to physiological studies indicating a resting heart rate of 90–95 bpm in this age range [62, 63]. In this context, it is worth mentioning that the validation of the activities was based on children with an overall good fitness level (mean activity minutes per day = 61.2) and a normal weight (overall BMI = 17.9 kg/m− 2, 60th percentile). Nevertheless, there were differences in the amount of activity and weight status between the children (see Table 2), resulting in differences in the HR while performing the activities [64]. This needs to be considered when interpreting the data. Still, as the mean value is quite good, we assume that the determined cut-off points can be used for children with a normal weight status and fitness level.
Selection of activities
In this context, the selection of the activities for the present calibration study should be discussed. First of all, a combination of locomotor activities (e.g., slow walking with 2 km/h) and other free-play activities (e.g., throwing and catching a ball) was used to better simulate the different types of activities that children engage in. These were common to children of this age group and provided both, varying intensity levels and ranges of accelerometer counts [22]. This is also a principle suggested by Welk [65] and was used in several studies so far [66,67,68,69]. Furthermore, nine activities were used which is also applied in other studies [40, 41, 65, 68, 70]. The selection and classification of the used activities is based on the Youth Compendium of Physical Activity [50]. Unfortunately, only one activity (slow walking) was selected for LPA. Thus, it is suggested that further studies should apply an equal number of activities for each of the four PA intensity levels.
Cut-off points
There is a need for using raw acceleration data instead of activity counts for measuring the intensity of PA [71,72,73]. Thus, the present study compared two different raw acceleration metrics (MAD and MAI) for the calibration of the Move4 sensor across four sensor positions in children aged 8–13 years. Within the discussion, further focus will set on the results of the MAD metrics, because, to the best of our knowledge, there is no validation and calibration study using the MAI metric so far. Nevertheless, this metric is also important as it uses a bandpass filter that ensures that accelerations that do not come from physical movements tend to be filtered out and is more and more used in studies [74, 75], so it was decided to determine the cut-off points for both MAD and MAI metrics. In comparison to Aittasalo et al. [70] who used the MAD metric for a hip worn accelerometer in children aged 13–15 years, our cut-off points differ especially in Cut 1 (SB-LPA: 52.9 vs. 26.9 mg;) and Cut 2 (LPA-MPA: 173.3 vs. 332 mg). One reason for the difference in the lower values in our study for LPA-MPA cut-off points might be the allocation of the activities to the intensities: In the present study, slow walking was the only activity for LPA and normal walking for MPA (oriented to the Youth Compendium of Physical Activity [50]), whereas Aittasalo et al. [70] allocated slow walking as well as normal walking to LPA. Another study using the MAD metric in 11 year old children indicated as the optimal cut-off points for LPA-MPA (= 3 MET) 91 mg and the MPA-VPA (= 6 MET) cut-off points was at 414 mg [47]. However, in this study participants performed a pace-conducted non-stop test on a 200 m long oval indoor track with initial speed of 0.6 m/s and it was increased by 0.4 m/s at every 2.5 min [47]. As free-living activities were also included, this could be the reason for the differences in the cut-off point from LPA to MPA (173.3 vs. 91 mg). In summary, our data show that the values respectively the cut-off points differ between the studies. This could be due to different samples, different activities and therefore it is important that for each sensor cut-off points are formed to make them usable for studies.
Sensor positions
To the best of our knowledge, this was the first study comparing four sensor positions of any accelerometer. Existing studies compared in particular hip and wrist worn accelerometer [41] or hip, wrist and thigh [68], but none compared the hip, thigh, wrist and chest position for sensor location. As the different body positions are involved differently in the nine activities, it is not surprisingly that the cut-off points differ slightly across the four sensor positions.
Regarding sensitivity and specificity of each sensor position, our results indicated overall for the hip as well as for the thigh the highest specificity (MAD hip: 88.8%; MAD chest: 82.1%) as well as sensitivity (MAD hip: 89.3%; MAD chest: 91.1%) whereas the sensitivity and specificity for the wrist indicated in MAD metrics 85.2% respectively 81.3%. In contrast, Johansson et al. [41] indicated for wrist and hip worn sensors same sensitivity (SB: 100%, MVPA: 70%) and specificity (SB: 60%, MVPA: 100%) values in preschool children. Sensitivity and specificity values for two hip worn accelerometers indicated in children aged 13–15 years almost perfect values for all cut-off values (98.6-100%) [70]. The different values of sensitivity and specificity in various studies could be explained by the selection of the activities. Depending on how far the MET values of the activities are from the MET-cutpoint (e.g., 3 MET), there are different metric cut-off points and different accuracies in the detection. In comparison, a study investigating adults found high accuracy of the thigh-worn accelerometer for predicting time spent in each PA intensity category, as seen by sensitivities and specificities > 99% for correctly classifying each PA intensity category [68]. One possible explanation for the differences between children and adults could be the inconsistent performance of activities in children whereas adults could more consequently perform activities over a certain time period [76].
Furthermore, regarding the accelerometer output (in mg) within one intensity level, there are differences according to the body position to which the sensor is attached. In particular, the MAD metrics for SB varied widely in our study: 52.9 mg for the hip placement, 62.4 mg for thigh, 86.4 mg for the wrist sensor and 45.9 mg for the chest position. The high values of wrist worn accelerometers in SB could be explained by the fact that this sensor position captures movements performed by the arms, unlike a hip worn monitor [41]. Especially younger children have problems to stand still without moving their arms [41]. Our findings are in accordance with results from other studies [36, 41, 77] which showed higher values for wrist-worn accelerometer compared with hip worn sensors, while measuring simultaneously.
Recommendations for sensor positions
We suggest to choose the sensor position depending on the research question. Overall, the sensor at the hip is really comfortable and shows good values and is already commonly used [32, 33, 78, 79]. Furthermore, the hip worn sensor indicated good sensitivity and specificity values in our study. Nevertheless other body positions should be considered while planning a study [68]. In particular, accelerometers worn on the thigh have shown high accuracy for measuring several different PA levels as well as SB and sleep [33, 38, 80,81,82,83,84]. Further, if there is an interest in the heart rate of the participants, the EcgMove4 accelerometer worn at the chest is suggested. Thus, it is easy to assess time spent in different intensity levels as well as the heart rate. The least favorable and efficient position seems to be the wrist due to low sensitivity and specificity.
Test-retest agreement
Lastly, to assess the accuracy of the Move4 sensor, a test-retest design for the agreement of the cut-off points between T1 and T2 for all sensor positions was used. Overall, agreement indicated good values with small differences between T1 and T2. Regarding the cuts, Cut 1 and 2 indicated higher deviations compared to Cut 3. This could be explained by the types of activities within one PA level allowing greater variations in the execution. Especially during the SB activities (standing and lying), some children problems to hold the position and not to move their bodies, especially their arms. In contrast, VPA activities required the whole body to move which allows less variations in execution.
Regarding the sensor positions, only the wrist worn accelerometer showed great differences especially for SB-LPA and LPA-MPA cut-off points. A problem is that the wrist worn accelerometer output is highly depending on the movements of the hands [41]. In particular, the task standing for four minutes was highly challenging for some children and variances were recognized between the children (inter-individual) and also between the two measurement points (intra-individual) in relation to the movement of the hands. This could explain the differences of 16.3% between T1 and T2 cut-off points.
Strengths and limitations
The main strength of our study is our sample size of 53 children aged 8–13 years which was numerous compared to other studies investigating between 20 and 47 participants [21, 41, 47, 70]. Furthermore, we calibrated and tested the Move4 sensor at four different body positions (hip, thigh, non-dominant wrist, and chest). Thus, the validation of the sensors was successful regarding a wide range of application possibilities. In addition, the different activities were not selected randomly, rather following the Youth Compendium of Physical Activity. Butte et al. [50] developed various activities and their resulting energy consumption in MET values. This is a meaningful list of activities and is a valuable resource. Furthermore, to ensure that the activities are performed with high accuracy, one research assistant was leading and participating in the exercise. This was highly important for the walking and running activities to lead the pace.
A limitation of this study relates to different weather conditions during the time period of the data collection. Therefore, some of the exercises were carried out indoors, which may have affected the children’s movement. Secondly, the sensors have partially fallen off during the movements. Although they were immediately reattached, a few seconds of activity had to be cut out. Further, some participants had difficulties to perform the activity the whole duration of four minutes. Thus, the data preparation contains cut outs to clean the raw data. Besides performing for four minutes, the accuracy of the execution lacked (e.g., standing). In this context, the validation of the activities and thus the determination of the cut-off points need to be slightly limited as the sample differed within the amount of activity (fitness level) and weight status, which might result in variability of the heart rate within one activity. Lastly, we only had one activity for the LPA intensity level that could be not representative for this level. Nevertheless, our data show good validity of the activities and the MET values. Further studies should consider that all intensity levels include more than one activity.
