An accelerometer is an electromechanical device used to measure acceleration forces and thereby detect motions [1
]. Since accelerometry functions are applicable to wearable activity monitors, accelerometer-based activity monitors have been widely accepted as a useful and practical device for monitoring and tracking physical activity as well as predicting energy expenditure [2
]. Further, the use of accelerometer-based activity monitors significantly contributes to the field of physical activity and health, such as the development of physical activity classification [3
], estimation of the mortality [5
], and application for different research settings [6
]. As such, physical activity assessment must be accurate; thus, researchers have validated accelerometer properties, placements, and/or data processing in regular physical activity settings [2
] but seldom in exergaming settings.
Exergaming combines body movements and video gaming and requires bilateral coordination skills of both upper and lower limb movement for different movement patterns (e.g., punching, kicking, jumping) in response to visual cues [8
]. Since exergaming increases energy expenditure and achieves moderate-to-vigorous levels of physical activity [9
], it has been widely implemented in clinical settings [11
] as well as in laboratory, home, schools, and the community [12
] as an innovative and alternative strategy to promote physical activity and health. To our knowledge, no exergaming studies have processed accelerometry data into quantifiable and interpretable information involving different monitors, placement sites, epoch lengths, or activity cut-points [2
]. There is thus an urgent need to validate the use of accelerometry for the assessment of physical activity in exergaming research.
In comparing subjective methods (e.g., diaries, questionnaires) for physical activity assessment, accelerometer-based activity monitors are regarded as the gold standard in detecting steps and quantifying the volume and intensity of physical activity [1
]. Such activity monitors have been used in a wide range of applications and in a variety of clinical and research settings [2
]. Despite their frequency of usage, validation studies have reported discrepancies in steps or physical activity levels when comparing activity monitors of different brands (e.g., activPAL, Hookie AM20, Polar Active vs. ActiGraph) at different placement sites [7
]; these validation studies have mostly assessed regular physical activities (e.g., walking, running) [15
] or free-living activities [15
], but one recent study compared the output of different monitors (pedometer vs. accelerometer) in an exergaming setting [18
One of the most commonly used activity monitor brands in physical activity research, ActiGraph has developed multiple generations of activity monitors [19
]. Researchers have validated different ActiGraph activity monitors—including GT3X vs. GT1M [4
], GT1M, GT3X, vs. GT3X+ [20
], and recently, GT3X+ vs. GT9X [21
]—placed at different sites such as hip vs. wrist [15
] during various physical activities. Although a hip placement site has been validated as an ideal location for accurately measuring steps and physical activity level in regular physical activities [15
], the evaluation of multiple placement sites (hip vs. wrist) in exergaming research is needed as more upper limb movements (unlike most regular physical activity) are required for exergaming [8
]. In addition, validation studies mainly focusing on young people (from preschoolers to adolescents) have evaluated epoch lengths using different sets of activity cut-points [24
], which impact the assessment of sedentary behavior and the different levels of physical activity intensity [25
]. Since the exergaming play we chose to evaluate here features acute bouts of intermittent and spontaneous physical activity, shorter epochs might be a better choice for capturing short bouts of frequently occurring activity [2
]. To date, there have been no studies comparing the effect of placement sites and epoch lengths on output especially from exergaming play or in young adults [2
]; thus, the most appropriate accelerometer data collection and scoring protocol remains unclear.
Of particular relevance here, studies comparing physical activity levels from different epoch lengths have not validated theses assessments with absolute measures of exercise intensity via indirect calorimetry (e.g., oxygen uptake, metabolic equivalent) or relative measures of exercise intensity via heart rate (HR) monitoring (e.g., %HRmax, %HR reserve (HRR)) [25
], which can be used as comparators to determine an appropriate epoch length for the accuracy of physical activity assessment. Whereas either relative or absolute measures can be used for classifying different levels of physical activity intensity [30
], the use of relative measures in comparing epoch lengths should be more feasible and effective for such an assessment [4
]. We believe that studies comparing epochs between activity counts and HR have never been reported, especially in an exergaming setting.
We aimed to examine the agreement of two recent generations of ActiGraph monitors (wGT3X-BT and GT9X Link, referred to below as wGT3X and GT9X, respectively) placed at different sites (hip and wrist). We sought to determine the most appropriate epoch length for physical activity assessment when validated using measurements of relative exercise intensity such as HR in healthy young adults in an exergaming setting. Our findings provide insight into effective data collection strategies for exergaming research, thereby improving the accuracy of physical activity assessment.
In this study, using an acute bout of exergaming play with two recent generations of ActiGraph monitors, we found that (1) intermonitor differences in steps and activity counts between wGT3X and GT9X were not significant on the hip placement site but were significant in terms of step counts on the wrist placement site; and (2) a 1-s epoch of activity counts obtained from hip-worn activity monitors was the best choice for estimating sedentary and physical activity intensity levels in an exergaming setting when compared with measures of relative exercise intensity using HR. We believe that our work is the first to compare indirect activity intensity measures using HR with activity counts using different epoch lengths, which could be a practical and applicable method for the accuracy of physical activity assessment.
Since newer activity monitor models are continuously being produced by the manufacturers (replacing previous models), researchers have validated outputs (e.g., steps, activity counts) of activity monitors for the accuracy of physical activity assessment. Our results indicated that the differences in steps between wGT3X and GT9X depended on the placement site, although there were strong associations between both monitors on the hip and wrist. More specifically, intermonitor differences for steps between the hip worn-monitors were not significantly different and were generally in good agreement. For these monitors, bias was close to zero, indicating that they were producing similar results, and the 95% limits of agreement were small, suggesting that the hip-worn monitors could be used as an alternative to measure steps. Additionally, there were similar patterns in tri-axial counts, especially on the vertical axis where steps are calculated in the ActiLife step-counting algorithm [1
Our findings are consistent with those of previous studies using other ActiGraph models. These studies showed considerable intermonitor agreement for the vertical axis counts between GT1M and GT3X in young adults during treadmill exercise [4
], among GT1M, GT3X, and GT3X+ in children and adolescents with lab-based activities [20
], and between the GT3X+ and GT9X in young adults with lab-based activities [21
]. However, we observed a relatively poor intermonitor agreement for step counts between the wrist-worn wGT3X and GT9X, as indicated by large and significant intermonitor differences, but reasonably good intermonitor agreement in the vertical and other two axis counts. Some studies have examined possible factors for an intermonitor difference in steps or activity counts. For example, ActiGraph’s low-frequency extension filter (the detection of lower amplitude movements) affects the difference in step or activity counts within different generation models (GT3X+ vs. 7164) [37
] or in the same models (GT3X+) [38
]. In addition, ActiGraph’s sampling frequency (the processing of raw acceleration data to activity counts) influences the discrepancy in activity counts within the same models (GT3X+) [39
]. Since we used the same sampling frequency (30 Hz) and a normal filter instead of low-frequency extension filter when we compared the wGT3X and the GT9X, the source of the discrepancy in steps between the wrist-worn monitors remains unclear.
A recent study [22
] compared step outputs between hip and wrist-worn ActiGraph monitors and between wrist-worn GT3X+ and GT9X monitors during treadmill walking and showed that the discrepancy in tri-axial orientations between GT9X and GT3X+ or other previous ActiGraph monitors might significantly impact step-counting accuracy on the wrist. Further, the ActiGraph step-counting algorithm developed for the hip location might not work for the wrist location [7
]. Tudor-Locke et al. [15
] examined the accuracy of steps on the hip and wrist placement sites using the same ActiGraph GT3X+ monitors and found the hip site outperformed the wrist site at most treadmill speeds, regardless of the bandpass filter. Moreover, we cannot rule out the possibility that the discrepancy in step counts between the wGT3X and GT9X may be due to differences in individual movement patterns. Since the exergaming we studied requires irregular upper body movements, differences in an individual’s arm motion or speed may affect threshold crossing of the acceleration signal, perhaps inducing less step-counting accuracy on the wrist. Additionally, John et al. [22
] report that accelerations detected on the wrist were smaller in magnitude than those at the hip during treadmill walking at the same speed, indicating that a wrist-worn monitor would count fewer steps than a hip-worn monitor. However, we found that the wrist-worn GT3X+ monitors resulted in higher steps than hip-worn monitors, which can be explained by the fact that exergaming play involves more arm movements. Thus, when researchers seek to determine the accuracy of step-counting, it is important to take the placement site into consideration. Our result thus suggests that researchers can select either of the two monitors we used here to conduct exergaming research if the devices are placed on the hip.
Of particular importance, we confirmed that epoch lengths differentially influenced assessment of sedentary and different physical activity levels, which is consistent with the previous studies. We found that time spent in SB and physical activity intensity levels varied when assessed using different epoch lengths (1, 5, 10, 15, 30, and 60 s). For instance, as epoch lengths decreased on the hip-worn monitors, estimates of SB and VPA increased while estimates of LPA and MPA decreased. We observed similar patterns in SB and MPA but a different pattern in LPA on the wrist-worn monitors. Our findings here are consistent with those of previous studies showing a varying effect of epoch length with earlier generations of ActiGraph monitors placed on the hip for seven days in a free-living condition. For example, Edwardson and Gorely [26
] used single-activity cut-points (i.e., Freedson) and observed that shorter epoch lengths among the four epochs (5, 15, 30, and 60 s) were associated with more time spent in SB and VPA and less time spent in MVPA, MPA, and LPA in children wearing an ActiGraph GT1M monitor on the hip. Banda et al. [25
] used multiple activity cut-points (e.g., Evenson, Treuth, Puyan) and showed that shorter epoch lengths among the six epochs (1, 5, 10, 15, 30, and 60 s) were related to more time spent in SB, MPA, and VPA and less time spent in LPA in children wearing the ActiGraph GT3X+ monitor on the hip. Finally, a study examining physical activity levels in middle-aged adults found that an epoch of 4 s among the three epochs (4, 20, and 60 s) was associated with longer time spent in VPA and shorter in LPA [28
]. The consistency in findings from the previous studies and our study might be associated with the form and intensity of an intermittent and spontaneous physical activity in which shorter epoch lengths such as a 1-s epoch [25
] or a 2-s epoch [29
] were the most appropriate epoch length to capture short bouts of vigorous or more intense physical activity. The physical activity form and/or intensity might be comparable to the exergaming play we used, which is characterized by rapid changes from sedentary to intense physical activity occurring frequently in a short period [40
]. Taken together, although the results from different epoch lengths vary, shorter epoch lengths may be appropriate for capturing short bouts, especially in more intense physical activity.
However, previous studies examining the effect of epochs on assessment of physical activity levels have not apparently compared relative or absolute measures of exercise intensity [25
], which might attenuate their findings. We compared physical activity intensity based on a cut-point set of HR with each of two cut-point sets of activity counts and found that the amount of time spent in SB, MPA, and VPA with the 1-s epoch length on the hip-worn monitors was similar to that in SB, MPA, and VPA of the HR but that this did not hold for any of the other longer epoch lengths from the hip- or wrist-worn monitors (Figure 4
). This is a novel result with respect to previous validations of cut-points of activity counts or raw data (accelerations) against indirect calorimetry (absolute measure) and HR monitoring (relative measure) for physical activity intensity. Previous studies have validated cut-points for physical activity intensity using indirect calorimetry as a gold standard measure of energy expenditure and metabolic equivalent for regular physical activities (e.g., treadmill walking/running) [4
]. Other studies have used HR monitoring as a relatively less expensive but feasible instrument to support the validation of cut-points with different analytical methods. For instance, Ozemek et al. [42
] suggested that activity counts were comparable to heart rate using %HRR at relative moderate (40% HRR) and vigorous (60% HRR) intensities but added that this depended on individual fitness levels. Two studies compared raw data and HR for sedentary activities and different intensity levels of physical activity and showed a strong correlation (r
= 0.97) [43
] and excellent agreement (receiver operating characteristic area under the curve = 0.99) [44
]. Thus, measures of indirect calorimetry or HR monitoring can be comparable to intensity assessed using cut-points of activity counts. We found that a 1-s epoch length in conjunction with a hip-worn monitor was the most similar to HR-derived measures and should be the most accurate method for measuring sedentary and various physical activity intensity levels in an exergaming setting.
We should note some important limitations to our results. Based on our statistical methods, we cannot determine the better of the two monitors we tested, but either of the two models of activity monitors can be used interchangeably on the hip placement site. Although hip-worn monitors seemed to be more appropriate for assessing step counts as well as physical activity intensity levels, we cannot generalize our results to other physical activity conditions or other activity monitor brands. Further, even though we demonstrated that a 1-s epoch would be the most appropriate epoch length for detecting short bursts of intense physical activity, it is unclear how an estimate of sedentary and physical activity intensity levels can be comparable to objective measures of energy expenditure [45
]. Thus, the fact that we did not use an indirect calorimetry technique as a criterion measure for physical activity intensity [46
] may be considered a limitation and therefore may require further investigation.