Background The purpose of this study was to determine the responsiveness of two motion sensors to detect change in sedentary behaviour (SB) and physical activity (PA) during an occupational intervention to reduce sitting time.
Methods SB and PA were assessed using a hip-worn Actigraph GTX3 (AG) and a thigh-worn activPAL (AP) during three consecutive workdays throughout baseline and intervention periods. Mean scores at baseline and intervention were estimated by hierarchical linear models (HLM) with robust SEs, adjusting for random variance of average scores between participants. Change scores (mean baseline minus mean intervention) were calculated for each device. Response to change was assessed for each device using the standardised response mean.
Results 67 adults (45±11 years; 29.3±7.7 kg/m2) wore the acceleration-based motion sensors for 8.3 (SD=1.2) and 8.3 (SD=1.1) h during the baseline and intervention periods, respectively. HLM showed that AP sitting/lying time (−16.5 min, −5%), AP stepping (+7.5 min, 19%), AP steps/day (+838 steps/day, +22%), AP sit-to-stand transitions (+3, +10%), AG SB (−14.6 min, −4%), AG lifestyle moderate-intensity PA (LMPA, +4 min, +15%) and AG MPA (+3 min, 23%) changed significantly between the baseline and the intervention period. Standardised response means for AP sitting/lying time, stepping, steps/day, sit-to-stand transitions and AG SB, LMPA and MPA were above 0.3, indicating a small but similar responsiveness to change.
Conclusions Responsiveness to change in SB and PA was similar and comparable for the AP and AG, indicating agreement across both measurement devices.
Statistics from Altmetric.com
The positive relationship between physical activity (PA) and health has been well established, and interventions to increase PA are prevalent. Recently, the ill effects of sedentary behaviour (SB) have come to light. Evidence is accumulating on the deleterious effects of sedentary time, showing that high levels of SB, defined as ‘any waking activity characterised by an energy expenditure ≤1.5 metabolic equivalents and a sitting or reclining posture’,1 are associated with numerous chronic acquired conditions including obesity,2–5 type 2 diabetes,6 ,7 metabolic syndrome,8 cardiovascular disease,9–11 certain types of cancer12 and cardiovascular and overall mortality7 ,13–15 in adults and older adults. Since these discoveries, researchers and practitioners have begun to intervene and attempted to break up prolonged sitting bouts and reduce total SB.
There are a number of self-report and objective tools used to assess SB and PA, each using a different approach or technology to measure these behaviours. Two of the most commonly employed objective tools are the activPAL accelerometer/inclinometer (AP; PAL Technologies Ltd, Glasgow, Scotland, UK) and the Actigraph accelerometer-based motion sensor (AG; ActiGraph LLC, Pensacola, Florida, USA). The AP assesses time spent sitting, standing and stepping through accelerometry and measurement of posture with an inclinometer. Time spent sitting or lying is commonly used as a marker of SB. Information on stepping and step rate are used as indicators of PA. The AG assesses times spent being sedentary and in PA of varying intensities by defined cut points.16–21 When comparing results across studies using different sedentary and PA measurement tools, it is essential that these tools provide comparable outcomes. Establishing this equivalency allows researchers to draw conclusions surrounding SB, PA and health from a multitude of studies using different assessment tools.
The AP and AG have been evaluated for validity and reliability in assessing sedentary time and PA.17 ,18 ,22–26 Additionally, the ability of the two devices to assess sedentary time has been compared to each other over a 1-day period in young, active, normal weight individuals,27 and to direct observation over a 6 h period in overweight office workers.17 Kozey-Keadle et al17 showed that both devices underestimated sitting time assessed by direct observation, but the AP was more precise and more sensitive to reductions in sitting time than the AG. Recently, the convergent validity of the AP and AG showed that the AG recorded over two additional hours of sedentary time over a 15 h period compared with the AP.27 While the validity and reliability of a device are paramount, it is also important that the device is able to detect a change in PA or sedentary time over the course of an intervention. However, no studies have compared the ability of the AP and AG devices to assess change in sedentary time and PA in adults over multiple days. Therefore, the purpose of this study was to determine congruency in the responsiveness of two accelerometer-based motion sensors (AP and AG) to detect change in free-living, occupational sedentary time and PA during an intervention to disrupt sitting time.
Participants were recruited from a large Midwestern university via flyers posted on university bulletin boards and emailed to university employees. Included in this study were individuals over 20 years of age whose occupation was sedentary, such as working at a desk or on a computer, and self-reported sitting for at least 60% of their workday. This quantification of sedentary occupation was based on previous research showing that office workers sat an average of 66% of their workday.28 Self-reported workday sitting time of at least 60% was confirmed during the baseline monitoring period using the AP. Individuals were excluded from participation if they walked with a limp, had a lower extremity fracture within the past 3 months, had any amputations of the lower limbs other than toes or used an assistive device to aid in ambulation.
Throughout the 2-week study period, participants met with the research team three times. The first visit took place in the laboratory. Participants read and signed an informed consent that had been approved by the University Institutional Review Board. A health history questionnaire was completed and height and weight were measured following standardised procedures.29 Height was measured using a stadiometer (Detecto 3PHTROD-WM, Webb City, Missouri, USA) and weight was measured using a balance beam scale (Detecto 339, Webb City, Missouri, USA). Finally, participants were given instructions on how to wear the AP and AG activity monitors.
There were two data collection periods, a baseline period and an intervention monitoring period. For each collection period, researchers met with participants at their workstation to issue the monitors. The intervention prompts (a wrist-worn device and a desktop computer application) were provided after completion of the baseline period and before the intervention period. These prompts were programmed to disrupt sedentary time every 60 min.30 The baseline and intervention monitoring periods assessed SB and PA during all working hours on the same three consecutive weekdays, over 2 weeks. At the end of each monitoring period, monitors were collected by the research team.
The variables of interest for the current study included average time spent in sitting/lying, standing, stepping; steps/day; number of sit-to-stand transitions from the AP monitoring device and time spent in SB, light-intensity PA (LPA), lifestyle moderate-intensity PA (LMPA), MPA and vigorous-intensity PA (VPA) from the AG.
AP accelerometer-based motion sensor
The AP is a small (35×53×7 mm), lightweight (20 g), piezoresistive uniaxial accelerometer, with a dynamic range >2 g. The AP was worn on the midline of the right thigh and was attached using the manufacturer provided PALstickies hydrogel adhesive pads between the device and leg and athletic tape over the top of the device. Data from the AP were recorded in 15 s epochs at a sampling rate of 10 Hz. In addition to the accelerometer, the AP contains an inclinometer that provides information based on the posture of an individual. Using manufacturer-determined angle and acceleration thresholds, the inclinometer can detect sit-to-stand (angle greater than 32° and a 0.53 g acceleration for a minimum of 10 s) and stand-to-sit transitions (angle less than 22° and a 0.38 g acceleration for a minimum of 10 s). The AP has previously been shown to be valid and reliable in detecting body positioning and activity in different populations.22 ,23
The AG GT3X (Pensacola, FL; 3.8×3.7×1.8 cm; 27 g) is a triaxial accelerometer-based PA monitor with a dynamic range of 2 g. Participants wore the AG accelerometers on the right hip. Data were collected in 15 s epochs at 30 Hz. The AG signals are passed through a bandpass filter (0.25–2.5 Hz) that excludes signals outside the range of human movement. In addition, this monitor offers a low-frequency extension option in order to increase sensitivity to low-intensity activities. For the current study, this low-frequency extension was not employed since it has not been shown to improve detection of steps accumulated through slow walking and light-intensity activity.31 The AG accelerometer has been shown to be a valid and reliable activity monitor across many different populations.17 ,18 ,24–26
Data for the AP and AG were downloaded onto a laboratory computer and processed using manufacturer provided software. All additional data processing was performed in Excel. Data were averaged over the 3-day baseline period (T1) and the intervention period (T2).
The AP software (activPAL V.6.4.1) classified individuals as sitting, standing or walking. Steps/day, number of sit-to-stand transitions are also recorded by the device.
The AG data were analysed using ActiLife 6 software. Only accelerations in the vertical plane were analysed. A combination of Freedson et al18 and Matthews and coworkers16 ,26 cut points were used to determine time spent sedentary and in PA. Sedentary was defined as <100 cpm,16 LPA was defined as 100–759 cpm, LMPA was defined as 760–1951 cpm, MPA was defined as 1952–5724 cpm and VPA was defined as ≥5725 cpm from the AG.16 ,18 ,26 In conjunction with log data, each participant's data were individually evaluated to precisely determine monitor on/off time, and to ensure that data from the two monitors were time synced for comparability.
Descriptive statistics were run for demographic, PA and sedentary variables. Response to change was assessed for each device using the standardised response mean (SRM) and effect size. SRM (also known as the mean standardised response) was calculated as the absolute value of the mean change score divided by the SD of the individual's change score. Effect size was calculated as the absolute value of the mean change score divided by the SD of the individual's baseline score. SRM and effect size can be interpreted as follows: <0.20 trivial, ≥0.20 to <0.50 small, ≥0.50 to <0.80 moderate and ≥0.80 large.32–34
Change scores were calculated for each device based on mean values during baseline monitoring minus mean values during intervention monitoring. Owing to the fact that the scores, repeatedly measured at baseline and intervention, were nested within each participant, mean scores at baseline and intervention were estimated by hierarchical linear models (HLMs) with robust SEs, adjusting for random variance of average scores between participants. Also, the changes between the two time points (ie, slope for time) were treated as randomly varied between individuals. This estimation model also controls for participant characteristics including age, gender, race, education and body mass index (BMI). All these variables were mean-centred. A generic two-level model was specified as: 1
π0i is a score of the participant i at baseline. π1i is the coefficient for time 2 (intervention), which is the measure of change between the two time points. The individual score (π0i) is then predicted by the background variables in the level-2 model, equation (2) with a random error (r0i). The coefficient for time, π1i, is also specified to vary between individuals (r1i) in the equation (3). All statistical analyses were conducted using HLM 735 and SPSS V.21.0 for Windows (SPSS Inc, Chicago, Illinois, USA). Significance was set at an α level of p<0.05.
Ninety individuals were enrolled and screened for participation in this study. Twelve did not qualify because they were too active as determined by AP assessed sitting time during the baseline period; 11 were excluded because they did not meet the wear time criteria or had equipment malfunction. Sixty-seven individuals successfully completed the study. Participants were predominantly white (87%) and female (72%), with an average age of 45 years. Eight per cent of the participants completed high school, 50% completed college and 42% completed graduate school. Participants, in general, were ‘overweight’ based on their average BMI score (mean=29.3 kg/m2, SD=7.7 kg/m2; table 1). On average, participants wore the acceleration-based motion sensors for 8.3 (SD=1.2) h during the baseline period and 8.3 (SD=1.1) h during the intervention period. Because wear time was not significantly different between baseline and intervention days (p<0.05), and results of analyses using outcome measures as absolute values and relative to wear time were similar, all SB and PA data are reported in absolute values.
The HLM showed that SB and PA measured by the AG and AP were, in general, changed significantly between the baseline and the intervention period with the exception of standing time, LTPA and VPA (see tables 2 and 3 for time 2 coefficients, π1). Models also showed that the baseline averages of those measures were significantly varied between individual participants (r0). The changes of activity between the two time points, however, did not vary between participants (r1). Individual demographic variables did not influence SB and PA (β01–β04). BMI was negatively associated with sitting and lying time with p<0.05 (b=−16.68; SE=1.72).
Estimated means (adjusted for age, gender, race, education and BMI) of outcome variables at times 1 (at baseline) and 2 (during intervention), and the changes between these two time points for AP-measured and AG-measured SB and PA are presented in table 4. Both devices were able to determine changes in SB and aspects of PA from the baseline to the intervention period. According to the AP output, average sitting/lying time decreased significantly by approximately 5% from baseline, while stepping time, steps/day and sit-to-stand transitions increased by 19%, 22% and 10%, respectively. As also shown in table 2, standing time did not change from baseline to intervention for the group. AG data detected a significant decrease in SB (4%), while LMPA and MPA increased by 15% and 23%, respectively.
SRM and effect size values for the AP and AG are shown in table 4. SRM values for the AP sitting/lying time, stepping, steps/day and sit-to-stand transitions were above 0.3, indicating a small responsiveness to change. SRM values for the AG SB, LMPA and MPA were also above 0.3, indicating a small responsiveness to change. SRM values for the AP and AG across all measures of SB and PA were therefore comparable.
When comparing results across studies using different SB and PA measurement tools, it is essential that these tools are congruent. Establishing this consistency allows researchers to draw conclusions regarding SB, PA and health from a multitude of studies using different assessment tools. Although accelerometer-based motion sensors have become a widely used tool to assess these SB and PA, and research has published studies on the validity and reliability of these devices, little research has focused on comparing the ability of these devices to detect change in these behaviours. This study fills an important gap in the literature, demonstrating that the AP and AG were similarly responsive to change in SB and PA during a free-living intervention that aimed to disrupt sedentary time, while controlling for factors that may impact the outcome such as age, gender, education and BMI. AP SRMs were highest when assessing change in sitting/lying time, stepping time, steps/day and sit-to-stand transitions. AG SRMs were highest when detecting change in SB, LMPA and MPA.
The SRM values for the AP and AG are considered small.32 ,33 Despite various statistical methodologies, populations and intervention designs, the SRMs are comparable to other studies that evaluated objective sedentary and PA monitoring devices.17 ,36 ,37 The AP has previously been shown to be sensitive to changes in SB in overweight inactive office workers over a 6 h period; however, during 6-month intervention to reduce dietary intake, reduce SB and increase PA in breast cancer survivors, the reported responsiveness index38 (RI) was lower than in the current study (RI=0.13).17 ,37 The AG was able to detect change in SB (SB<100 cpm) in older adults between two 6-day monitoring periods (responsiveness statistic=0.39)36 ,39 but was not sensitive to change in SB (SB<50, 100, 150, 200, 250 cpm; low-frequency filter applied) in overweight inactive office workers between two 6 h periods.17
The AP and AG SRM values to detect change in SB were similar (AP: 0.300; AG: 0.327), indicating that both devices were able to detect a similar magnitude of change in SB. Interestingly, when examining the mean SB values during the baseline and intervention periods, the AP recorded 30.8 and 32.7 fewer minutes in sedentary time compared with the AG, but change scores were similar (AP: −16.5, AG: −14.6 min), demonstrating their congruency to detect change in SB. The difference in recorded SB between the devices could be due to the body position of the participants during sitting activities, or the AG SB cut point applied. At times, the research staff observed participants sitting at their desks with their legs extended in front of their bodies. Since the AP is worn on the thigh (and AG on the hip), this could cause a misclassification of the body position by the AP due to the separate stand and sit thresholds. For the AP to capture a movement from the sitting to the upright posture, the acceleration must exceed the threshold of 0.53 g (approximately 32°), and maintain that posture for a minimum of 10 s. If individuals extend their legs out in front of them while sitting, this threshold could be exceeded and therefore improper classification could occur. The difference in recorded SB between the devices could also be due to the AG SB cut point applied (SB<100 cpm).There are multiple cut point values available to determine time spent in SB (<50, <100, <150 cpm), and the application of cut point values has been shown to alter time spent in SB.26 ,40 The higher SB values by the AG compared with the AP in the current study are in contrast to the results of Kozey-Keadle et al17 whose results showed the AG (SB<100 cpm) recording of approximately 9 fewer minutes of sedentary time than the AP, over a 6 h period in a small sample of overweight or obese adults. However, the purpose of this study was to determine congruency in responsiveness of the AP and AG to detect change in sedentary time and PA, which is independent of any cut point applied.
Despite differing technologies, assessment methodologies and outputs of PA, SRMs for assessing PA by the AP (stepping=0.376, steps/day=0.436) and AG (LPA=0.245, LMPA=0.350, MPA=0.371, VPA=0.220) were comparable, demonstrating a small but similar responsiveness to measures of PA. Because these devices do not measure the same aspects of PA, studies on PA including both monitors are scarce; however, research has demonstrated that the responsiveness of numerous PA questionnaires is similar to the responsiveness of the AP and AG for PA seen in this study. Responsiveness values assessed by Tuley et al's formula38 for the Community Health Activities Model Program for Senior (CHAMPS) Questionnaire, Active Australia Questionnaire (AAQ) and the US National Health Interview Survey (USNHIS) ranged from 0.45 (USNHIS) to 0.43 (AAQ) for walking frequency and from 0.27 (USNHIS) to 0.24 for walking duration; the AAQ showed a responsiveness level of 0.43 for moderate-to-vigorous intensity 0.43 (AAQ), and the CHAMPS showed a responsiveness level of 0.32 for moderate-to-vigorous duration.41 Therefore, although responsiveness of the AP and AG for changes in PA has not been explored previously, the rates are similar to widely used subjective methods for assessing PA.
The use of SRM is a strength of this study. It is a fairly simple calculation with easy to understand output. Further, SRM allows the responsiveness to changes to be evaluated across a spectrum (small vs moderate vs large). The limitations of this study warrant consideration when evaluating the results. This study does not include a criterion measure of SB such as direct observation, and therefore the accuracy of the responsiveness was not determined. Second, this study included a short duration baseline and intervention monitoring periods. Future studies should evaluate the responsiveness of these devices over a longer period of time. Finally, the population in this study included office workers with sedentary occupations, and therefore generalisability is limited.
The results of this study demonstrate the agreement in responsiveness of two accelerometer-based motion sensors (AP and AG) to detect change in free-living, occupational sedentary time and PA during an intervention to disrupt sitting time. This important new information is vital when assessing change in these behaviours as the result of an intervention, and enabling synthesis of the literature in this area to draw conclusions pertinent to efficacy and health outcomes.
What are the new findings?
The activPAL and Actigraph GTX3 acceleration-based motion sensors were able to determine change in sedentary behaviour (SB) and aspects of physical activity (PA) from the baseline to the intervention period.
The responsiveness of both devices was comparable with the most standardised response mean values for activPAL and Actigraph GTX3 variables above 0.3, indicating a small responsiveness to change.
ActivPAL standardised response means were highest when assessing change in sitting/lying time, stepping time, steps/day and sit-to-stand transitions.
Actigraph standardised response means were highest when detecting change in SB, lifestyle moderate-intensity PA and moderate-intensity PA.
How might it impact on clinical practice in the near future?
The activPAL or the Actigraph GTX3 are viable options to detect change in daily SB over the course of an intervention.
The activPAL or the Actigraph GTX3 are viable options to detect change in daily PA behaviour over the course of an intervention.
Standardised response means for activPAL and Actigraph in this study were similar to previously published standardised response means for questionnaires.
The authors would like to acknowledge the Clinical and Translational Science Institute of Southeastern Wisconsin (1-UL1-RR031973) for support of this project. They also thank Nick Thielke for assistance with data collection. Finally, they thank the participants for their contributions to this study.
Contributors All authors contributed substantially to the manuscript. AMS, AER and SJS contributed to the conception and design of the study. AMS, AER, YIC, SJS and WAW contributed to the acquisition of data or analysis and interpretation of data. All authors were significantly involved in the writing and/or editing of the manuscript and provided final approval of the version published.
Funding This work was supported by funding source 1-UL1-RR031973.
Competing interests None.
Ethics approval University of Wisconsin-Milwaukee IRB.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.