Where to Place Which Sensor to Measure Sedentary Behavior? A Method Development and Comparison Among Various Sensor Placements and Signal Types

in Journal for the Measurement of Physical Behaviour
View More View Less
  • 1 Karolinska Institutet
  • 2 ZHAW Zurich University of Applied Sciences
  • 3 Karolinska University Hospital
  • 4 Sophiahemmet University

Background: Sedentary behavior (SB) is associated with several chronic diseases and office workers especially are at increased risk. SB is defined by a sitting or reclined body posture with an energy expenditure of ≤1.5 metabolic equivalents. However, current objective methods to measure SB are not consistent with its definition. There is no consensus on which sensor placement and type should be used. Aim: To compare the accuracy of newly developed artificial intelligence models for 15 sensor placements in combination with four signal types (accelerometer only/plus gyroscope and/or magnetometer) to detect posture and physical in-/activity during desk-based activities. Method: Signal features for the model development were extracted from sensor raw data of 30 office workers performing 10 desk-based tasks, each lasting 5 min. Direct observation (posture) and indirect calorimetry (in-/activity) served as the reference criteria. The best classification model for each sensor was identified and compared among the sensor placements, both using Friedman and post hoc Wilcoxon tests (p ≤ .05). Results: Posture was most accurately measured with a lower body sensor, while in-/activity was most accurately measured with an upper body or waist sensor. The inclusion of additional signal types improved the posture classification for some placements, while the acceleration signal already contained the relevant signal information for the in-/activity classification. Overall, the thigh accelerometer most accurately classified desk-based SB. Conclusion: This study favors, in line with previous work, the measurement of SB with a thigh-worn accelerometer and adds the information that this sensor is also accurate in measuring physical in-/activity while sitting and standing.

Due to the associated detrimental health effects, sedentary behavior (SB) has received increasing attention from the research community. SB is an omnipresent behavior and an integral part of the modern lifestyle (Loyen et al., 2017; Matthews et al., 2018), especially in the office sector, where a large part of the population works (Keown, Skeaff, Perry, Haszard, & Peddie, 2018; Nooijen et al., 2018). Several chronic lifestyle diseases, like diabetes and metabolic syndrome, are associated with SB (Amirfaiz & Shahril, 2019; van der Velde et al., 2018). Accordingly, the use of active workplaces is currently recommended (Buckley et al., 2015), although, there is a lack of evidence whether they have an impact on health (Neuhaus et al., 2014; Shrestha et al., 2018; Tew, Posso, Arundel, & McDaid, 2015). One important reason for this lack is the use of inconsistent and nonobjective SB measurements (Hutcheson, Piazza, & Knowlden, 2018; Stephenson, McDonough, Murphy, Nugent, & Mair, 2017). While medicine is very skilled in diagnosing the diseases, measuring the dose of SB is still in its infancy, and it is too early to establish evidence-based health recommendations on SB (Stamatakis et al., 2019; van Uffelen et al., 2010).

By definition, SB involves a certain body posture (sitting or reclining) and a certain energy expenditure (≤1.5 metabolic equivalents, MET; Sedentary Behaviour Research Network, 2012). However, SB is currently not measured in line with this definition (Fanchamps, van den Berg-Emons, Stam, & Bussmann, 2017; Holtermann et al., 2017; Kang & Rowe, 2015). Existing devices can be separated into two groups, those measuring posture and those measuring activity (Kang & Rowe, 2015). This makes the SB measurement an unresolved challenge, since both the posture and activity must be measured at the same time (Holtermann et al., 2017). Posture-based accelerometers are typically attached to the thigh (Grant, Ryan, Tigbe, & Granat, 2006; Skotte, Korshoj, Kristiansen, Hanisch, & Holtermann, 2014), and the recorded acceleration is low-pass filtered to determine the thigh orientation versus gravity (often referred to as an inclinometer; Edwardson et al., 2017). These devices are known to have a very high sensitivity and specificity to detect sitting and standing (Kim, Barry, & Kang, 2015; Stemland et al., 2015). However, they are not able to separate active from inactive sitting and standing (Godfrey, Culhane, & Lyons, 2007; Grant et al., 2006). Their SB estimate is based on the posture information only (Edwardson et al., 2016; Kim et al., 2015). In contrast, activity-based devices are typically worn on a belt around the waist. Recently, these sensors have also been worn like a watch on the wrist with the aim to improve compliance (Kerr et al., 2017). For both placements, the recorded acceleration signal is converted into counts per minute and is used as a measure for the activity level (Migueles et al., 2017; Rosenberger et al., 2013). Activity-based devices are known to have a high accuracy to detect physical activity, that is, light-, moderate-, and vigorous-intensity activities (Rothney, Schaefer, Neumann, Choi, & Chen, 2008). However, their SB estimate is only based on a lack of activity (Migueles et al., 2017).

A simple solution to measure SB in line with its definition would be to combine a posture- and activity-based device (Ellingson, Schwabacher, Kim, Welk, & Cook, 2016). However, each study participant would then have to be equipped with two sensors, limiting the field of application and increasing the complexity of data processing. A more feasible solution would be to calibrate one single body-worn sensor against posture and activity. Unfortunately, there is a lack of information regarding where such a sensor should be worn, nor is it known which signal types are needed. So far, acceleration signals are used due to their ease of use. Accelerometers record in a respectable frequency over a long period, while having only a small sensor housing. Technical advances in recent years include gyroscopes and magnetometers to build so-called inertial measurement units (IMUs). The ActiGraph Link (ActiGraph LCC, Pensacola, FL), for example, includes a 3D accelerometer, a 3D gyroscope, and a 3D magnetometer, and the activPAL4 (PAL Technologies Ltd., Glasgow, Scotland) includes a 3D accelerometer and a 3D magnetometer. However, it is not known whether the additional signal types are of any value for the SB classification.

The primary aim of the present study was to develop and compare new models for 15 sensor placements in combination with four signal types. As an exhaustive Phase 1 study in the framework of Keadle et al. 2019, this study shall inform future Phase 1 and Phase 2 method developments in the choice of sensor placement and signal type for the measurement of SB in desk-based activities (Keadle, Lyden, Strath, Staudenmayer, & Freedson, 2019). The secondary aim was to provide the same information for the isolated detection of posture (i.e., sitting, standing, and walking) and in-/activity (≤1.5 MET/>1.5 MET).

Materials and Methods

This study calibrated 15 body-worn IMUs against valid reference criteria for posture (direct observation) and physical activity (indirect calorimeter). The data recording took place between September 12 and November 12, 2018, at a workplace at ZHAW Zurich University of Applied Sciences in Winterthur (Switzerland). For each sensor placement and signal combination (accelerometer only, accelerometer plus gyroscope, accelerometer plus magnetometer, accelerometer plus gyroscope, and magnetometer), a separate machine learning model was developed for the posture, as well as the in-/activity classification while sitting and standing. The models were then combined to classify the behavior on a minute-by-minute level into inactive sitting (equal to SB), active sitting, inactive standing, active standing, and walking.

Participants and Ethics

Thirty healthy office workers between 18 and 65 years, with ≥70% employment level and spending ≥50% of their working day at an office desk, were recruited through flyers, mail, and word of mouth. Persons with chronic or acute respiratory, neurological, or systemic diseases, as well as a silicon allergy, were excluded. Since the subjects booked their appointment online, inclusion and exclusion criteria were checked upon arrival. The subjects confirmed that they had refrained from eating and drinking sugary, caffeinated, and alcoholic beverages for 2 hr and had refrained from sport for 12 hr. Every subject signed an informed consent prior to the study inclusion. This binational project was approved by the ethics committee in Stockholm (DNR: 2018/554-31/1) and obtained a declaration of nonobjection from the ethics committee in Zurich. The participants averaged 38.8 ± 9.0 years, 174 ± 8 cm tall, and 71.2 ± 11.0 kg weight, and they worked 40.5 ± 6.6 hr a week, of which, 86 ± 11% was at an office desk (self-reported). No adverse events occurred.

Procedure

The subjects were equipped with the IMUs and the indirect calorimeter. While familiarizing themselves with the equipment, they filled out a questionnaire regarding their personal demographics. Subsequently, the aim of each task was orally explained before the measurement started. The task and condition order were randomized, but two identical tasks and conditions never occurred in succession. After completing all tasks, the resting metabolic rate was measured for 10 min in a supine position on a padded yoga mat with a head pillow.

Tasks and Conditions

The participants performed four tasks (Table 1) at a height-adjustable office desk in three different conditions: sitting on a conventional office chair (Vitra, Birsfelden, Switzerland), sitting on a saddle chair with an elevated seat height (HAG Capisco; Flokk, Oslo, Norway), and standing (Figure 1). The saddle chair was only used for the keyboard task, to challenge the posture classification. Together with the walking task, 10 tasks were recorded, each lasting 5 min. To account for different workplace designs, 2/3 of the study participants used a desktop PC with two 24 in. screens (diameter of 61.0 cm), and 1/3 used a laptop with a 15 in. screen (diameter of 38.1 cm). The execution of the tasks was neither demonstrated nor standardized in any form. The subjects had to place the working material their own way, and every subject completed the tasks at their own speed. The subjects were allowed to change the table and seat height at any time.

Table 1

Investigated Tasks

TaskInstruction and aim
MousePlay a computer game with the mouse (Microsoft Mahjong) to investigate intensive mouse use.
KeyboardWrite a text in Microsoft Word® (Microsoft Corp., Redmond, WA) to investigate intensive keyboard use (mouse use allowed).
DeskworkDo various short tasks with a physical folder and a Microsoft Excel® file (get the folder, search in it, do mental arithmetic, create tables, write notes, and switch screen views) to investigate successive short tasks with and without computer.
SortingOpen envelopes and stow the documents according to the instruction on the documents (in storage compartments or folders) to investigate successive manual tasks without computer.
WalkingWalk around to investigate nonstationary activities like walking to the printer or to a meeting.

Note. Selection is based on previous studies investigating typical desk-based activities (Burns, Forde, & Dockrell, 2017; Ellegast et al., 2012; Grooten, Conradsson, Ang, & Franzen, 2013). An example of each task is shown in Figure 1.

Figure 1
Figure 1

Investigated office tasks: (a) mouse, (b) keyboard, (c) deskwork, and (d) sorting, and conditions: (a and d) conventional office chair, (b) saddle chair, and (c) standing. Details for each task are given in Table 1.

Citation: Journal for the Measurement of Physical Behaviour 2020; 10.1123/jmpb.2019-0060

Measurement Equipment

Reference Criterion

Direct observation was used as the reference criterion for posture, and an indirect calorimeter (K5; COSMED, Rome, Italy) was used as the reference criterion for activity. The K5 has been shown to measure the energy expenditure of a given task reliably (Crouter et al., 2019). It was calibrated before each recording according to the manufacturer’s recommendation (flowmeter, scrubber, and room air). The data were recorded in the mixing chamber mode with 0.1 Hz. VO2 and VCO2 were exported to a CSV file for subsequent processing.

Inertial Measurement Units

The MVN BIOMECH Awinda (Xsens, Enschede, The Netherlands), in full-body configuration, without hands, was used. The system consists of 15 small IMUs, each featuring a 3D accelerometer (range ± 16 g), a 3D gyroscope (range ± 2,000°/s), and a 3D magnetometer (range ± 1.9 Gauss). The IMUs were placed according to the manufacturer’s recommendation on the following segments: head, sternum, and waist (unilateral), and wrist, upper arm, shoulder, thigh, shank, and foot (bilateral). All units were attached with elastic stripes, except for the shoulder and sternum (in a special shirt), as well as the waist (belt). The raw signals of the units (60Hz) were exported as MVNX files for subsequent processing.

Data Processing

All data were loaded into MATLAB 2019a (version 9.6.0; MathWorks Inc., Natick, MA) for processing and evaluation.

Reference Criterion

The energy expenditure was calculated using the Weir equation. Only steady-state data were used to express the energy expenditure of each task. The onset of steady state was defined by the first minute with <10% deviation from the median of all subsequent minutes, but earliest after 1 min and latest after 4 min. The median energy expenditure during the steady state of each task was then put in relation to the resting metabolic rate to calculate the MET. The resting metabolic rate was defined as the median energy expenditure during the second 5 min in the supine position (Borges et al., 2016; Popp, Tisch, Sakarcan, Bridges, & Jesch, 2016). All recorded minutes were subsequently assigned into body posture (sitting, standing, and walking) and in-/activity level (inactive: ≤1.5 MET, active: >1.5 MET).

Inertial Measurement Units

The IMU signals were also split into minute-by-minute data. For each minute, the same 562 signal features as in Kuster et al. (2020; except daytime) were calculated for each sensor type (feature calculation is shown in Supplementary Material 1 [available online]). The machine learning was split in three parts: (a) feature filtering, (b) feature inclusion, and (c) model optimization and training. The processing was conducted separately for the posture classification, the in-/activity classification while sitting, and the in-/activity classification while standing. To generate the overall behavior classification for SB, active sitting, inactive standing, active standing, and walking, the classification models of each sensor and signal combination were combined.

  1. (a)The feature filtering used a customized random forest classifier, programmed in Python (program available at https://github.com/RomanKuster/featureranking). Out of 100 classifier runs, the best 100 features were selected. The selection was done separately for each raw signal (accelerometer, gyroscope, and magnetometer). This step is referred to as feature filtering, as nonrelevant features are filtered out so that the subsequent computational demanding steps did not need to examine the full feature list.
  2. (b)In order to include only the most relevant features, the remaining features were stepwise included into a random forest classifier with five trees, separately for each investigated signal combination (accelerometer only, accelerometer plus gyroscope, accelerometer plus magnetometer, accelerometer plus gyroscope, and magnetometer). The first round selected the single best feature to solve the classification, and each subsequent round added the next best feature (Kuster et al., 2018). The stepwise inclusion was stopped when the maximum accuracy was reached (no increase for the next 10 features).
  3. (c)The training architecture for the classification models were then optimized for each feature number using MATLAB’s built-in hyperparameter optimization function for classification learners (fitcensemble with “OptimizeHyperparameters” set to “all,” see MATLAB code in Supplementary Material 2 [available online]). The optimization searched for the best learning algorithm, split criterion, number of learners, learning rate, minimum leaf size, and maximum number of splits. Further details about the parameter optimization can be accessed online (https://mathworks.com/help/stats/fitcensemble.html). The optimized parameters were finally used to train the models. For each sensor and signal combination, the feature number with the highest accuracy was selected and used in the statistical comparison.

Statistics

The feature inclusion and model training (including the parameter optimization) required a cross-validation technique to identify the most accurate feature in each step. The leave-one-subject-out cross-validation technique was used for this. The technique trains a model on all but one subject (the leave out) and analyzes the model accuracy on the leave-out subject. This procedure is repeated until every subject served once as leave out, and the accuracy is averaged over all leave-out subjects. To equally account for both the true positive and true negative detection of SB, the balanced sensitivity and specificity, which is the mean of sensitivity and specificity, was used as the measure of accuracy (Ellis, Kerr, Godbole, Staudenmayer, & Lanckriet, 2016).

To identify the best signal combination for each sensor, a Friedman test for dependent data was used. Unless there was a significant effect of signal type, only the accelerometer results are presented. Otherwise, the best signal combination was identified with a post hoc Wilcoxon test, taking multiple testing according to Bonferroni into account.

The accuracy among all sensor placements was compared with another Friedman test, again followed by a post hoc Wilcoxon test adjusted for multiple testing. Sensor placement comparison was done separately for the accelerometers only and the best signal combinations. The accuracy is presented with the median and nonparametric 95% confidence interval. For the isolated in-/activity classification while sitting and standing, the accuracy was merged since there were only two categories, and the sensitivity for inactivity equals the specificity for activity, and vice versa. Descriptive statistics are presented, after rejecting the normal distribution with Lilliefors test, with the median and interquartile range. The level of significance was set to 0.05.

Results

Out of 1,500 recorded minutes, 7 min from three subjects were lost due to system malfunction. Overall, the subjects spent 76.6% and 70.8% of all sitting and standing minutes inactive, respectively. The MET of each task and condition is shown in Table 2. The median (interquartile range) resting metabolic rate was 1,696 (607) kcal/day, or 3.5 (0.9) ml·kg−1·min−1 VO2.

Table 2

Average MET and Proportion of Time Spent Inactive for Each Condition and Task

Conditions
TasksConventional chairSaddle chairStandingWalking
Mouse
 MET1.19 (0.20)1.18 (0.25)
 % ≤1.5 MET96.7100.0
Typing
 MET1.34 (0.29)1.2 (0.27)1.28 (0.16)
 % ≤1.5 MET93.396.790.0
Deskwork
 MET1.26 (0.29)1.33 (0.25)
 % ≤1.5 MET76.776.7
Sorting
 MET1.72 (0.42)1.75 (0.33)
 % ≤1.5 MET16.716.7
Walking
 MET3.30 (0.91)
 % ≤1.5 MET0.0

Note. MET is presented with median and interquartile range in brackets, proportion of time spent inactive (≤1.5 MET) in percentage. MET = metabolic equivalent.

The best accelerometer placement to classify SB in desk-based activities was the right thigh. The results for the left thigh and the two shank accelerometers were similar (Table 3), while all other accelerometers performed significantly worse to detect SB, as well as inactive standing. However, when adding the gyroscope and magnetometer data to the waist, the accuracy significantly increased and was no longer different from the right thigh accelerometer. In contrast to the inactive behaviors, the accuracy to classify active sitting and standing was the same for most placements, and all accelerometer placements were able to detect walking with 100.0% accuracy (data not shown in Table 3). All models in Table 3 are shared on MATLAB Central (https://mathworks.com/matlabcentral/fileexchange/79469).

Table 3

Accuracy of Sensor Placement and Signal Type (Accelerometer Only and Best Signal Combination) to Classify Desk-Based Activities Into Posture and Physical in-/activity Level While Sitting and Standing (Inactive: ≤1.5 MET, Active: >1.5 MET)

Sedentary behaviorActive sittingInactive standingActive standing
AccelerometerBest signal combinationAccelerometerBest signal combinationAccelerometerBest signal combinationAccelerometerBest signal combination
SensorAccuracyAccuracyTypeAccuracyAccuracyTypeAccuracyAccuracyTypeAccuracyAccuracyType
Waist84.9

[80.0, 88.9]*
91.9

[88.0, 93.2]
+ G

+ M
87.8

[83.0, 90.7]
85.7

[82.3, 88.3]*
91.9

[89.2, 94.6]
+ G

+ M
90.0

[86.3, 98.9]
Thigh (right)93.4

[91.1, 96.3]
92.9

[89.6, 97.9]
95.0

[93.3, 96.8]
96.0

[80.0, 100.0]
Thigh (left)91.7

[87.6, 94.0]
95.0

[91.8, 97.5]
+ G90.0

[80.8, 97.9]
93.1

[90.5, 96.1]
96.7

[93.2, 100.0]
+ G93.0

[89.6, 98.9]
Wrist (right)75.0

[68.5, 79.8]*
72.2

[61.8, 94.5]*
60.0

[56.1, 69.1]*
80.0

[76.1, 91.2]*
Wrist (left)75.1

[70.0, 78.3]*
89.4

[71.6, 96.2]
68.5

[62.1, 76.8]*
80.0

[75.0, 94.6]
Sternum83.7

[78.7, 86.8]*
88.9

[78.4, 96.2]
81.0

[75.2, 85.6]*
87.8

[78.5, 95.7]
90.0

[87.2, 97.9]
+ G

+ M
Head71.7

[65.3, 73.8]*
81.5

[79.2, 85.0]*
+ G

+ M
69.4

[57.8, 90.0]*
88.9

[74.8, 95.7]
+ G65.3

[59.2, 68.3]*
85.2

[80.1, 91.8]*
+ G

+ M
85.9

[72.2, 96.2]
Shoulder (right)82.0

[76.7, 86.2]*
86.1

[70.7, 94.7]
82.5

[75.1, 87.0]*
85.2

[78.8, 91.5]
92.8

[85.6, 98.3]
+ G

+ M
Shoulder (left)83.3

[80.5, 88.3]*
87.8

[83.1, 92.8]*
+ G

+ M
89.4

[78.4, 91.7]
92.0

[87.8, 98.9]
+ G + M82.9

[76.7, 87.0]*
89.4

[78.2, 96.2]
Upper arm (right)78.0

[75.2, 82.3]*
86.1

[76.1, 92.8]
73.8

[68.8, 76.8]*
83.3

[72.4, 90.4]
Upper arm (left)77.0

[70.7, 81.2]*
81.6

[70.0, 96.2]
71.0

[65.0, 74.9]*
86.5

[76.6, 90.0]
Shank (right)91.8

[89.1, 95.0]
96.2

[94.2, 98.1]
+ G85.6

[77.8, 96.2]
90.0

[80.8, 98.3]
+ G91.0

[85.6, 94.5]
93.0

[87.9, 99.3]
Shank (left)90.0

[87.5, 94.3]
86.2

[70.0, 92.7]
89.4

[80.8, 97.8]
+ M90.9

[87.4, 92.8]
89.4

[76.8, 95.4]
Foot (right)81.7

[78.0, 87.0]*
70.0

[65.5, 87.1]*
76.3

[71.5, 82.8]*
87.8

[74.8, 90.7]*
Foot (left)82.3

[76.4, 86.7]*
88.7

[86.4, 91.8]*
+ G

+ M
77.8

[66.7, 85.0]*
79.8

[73.4, 83.0]*
83.0

[74.1, 95.7]*

Note. Sedentary behavior is equal to inactive sitting. Indicated is the median balanced sensitivity and specificity with 95% CI in brackets. The column “Best signal combination” indicates whether the addition of the gyroscope (+ G) and/or magnetometer (+ M) significantly improved the accuracy (empty if not). The accelerometer accuracy with lowest rank sum (Friedman test) of each behavior is marked in bold. MET = metabolic equivalent; CI = confidence interval.

*Significantly different.

If looking only at the posture classification, it is evident that the thigh placement classified posture most accurately, followed by the shank placement (Table 4). All other accelerometer placements performed significantly worse. Even when adding the gyroscope and magnetometer data to the waist, which improved the posture classification, the waist placement was not as accurate as the thigh. To detect the in-/activity level while sitting and standing, the sternum (sitting) and head (standing) placement performed best (Table 4). For sitting, those sensors placed on the upper body (except the right wrist) showed a significantly higher accuracy than those placed on the lower body and waist. For standing, the differences between the placements were only marginal and nonsignificant. For both the in-/activity classification while sitting and standing, adding the gyroscope and magnetometer data improved the accuracy only for one single placement while sitting (right shank).

Table 4

Accuracy of Sensor Placement and Signal Type (Accelerometer Only and Best Signal Combination) to Classify Desk-Based Activities Separately Into Posture (Sitting and Standing) and Physical in-/activity Level (Inactive: ≤1.5 MET, and Active: >1.5 MET)

Posture classificationin-/activity classification
SittingStandingSittingStanding
AccelerometerBest signal combinationAccelerometerBest signal combinationAccelerometerBest signal combinationAccelerometerBest signal combination
SensorAccuracyAccuracyTypeAccuracyAccuracyTypeAccuracyAccuracyTypeAccuracyAccuracyType
Waist89.6

[86.0, 94.0]*
94.0

[91.3, 98.0]*
+ G

+ M
90.0

[87.2, 92.5]*
95.0

[91.1, 98.3]*
+ G

+ M
90.0

[83.2, 95.0]*
92.9

[80.0, 100.0]
Thigh (right)100.0

[99.3, 100.0]
100.0

[99.1, 100.0]
90.0

[85.0, 95.7]*
92.5

[80.0, 100.0]
Thigh (left)98.0

[98.0, 100.0]
100.0

[100.0, 100.0]
+ G97.9

[97.5, 100.0]
100.0

[100.0, 100.0]
+ G86.3

[78.2, 95.3]*
90.0

[85.0, 96.7]
Wrist (right)68.0

[62.0, 74.7]*
66.7

[58.9, 72.5]*
90.8

[75.0, 97.5]*
92.9

[83.2, 96.7]
Wrist (left)74.0

[69.6, 78.0]*
72.9

[66.6, 78.3]*
97.2

[86.6, 100.0]
96.7

[81.3, 100.0]
Sternum82.0

[79.3, 84.5]*
80.0

[77.5, 85.3]*
96.2

[90.0
, 100.0]
91.3

[85.0, 96.4]
Head67.0

[62.0, 70.7]*
88.0

[80.0, 90.0]*
+ G

+ M
65.0

[59.7, 69.6]*
90.0

[81.5, 91.7]*
+ G

+ M
90.0

[82.4, 95.0]
95.6

[86.1
, 100.0]
Shoulder (right)83.0

[76.0, 88.0]*
86.0

[82.0, 90.0]*
+ M84.2

[76.4, 86.1]*
84.6

[79.7, 90.0]*
+ M94.4

[85.0, 100.0]
91.3

[84.4, 100.0]
Shoulder (left)84.0

[80.0, 88.0]*
92.0

[87.3, 94.7]*
+ G

+ M
84.6

[78.9, 87.5]*
92.1

[87.1, 93.9]*
+ G

+ M
90.5

[86.6, 97.5]
96.5

[80.0, 100.0]
Upper arm (right)75.0

[71.3, 79.9]*
73.3

[70.0, 78.9]*
92.4

[83.5, 100.0]
91.3

[83.6, 97.9]
Upper arm (left)76.0

[70.0, 78.0]*
72.5

[70.0, 75.6]*
79.6

[74.2, 82.8]*
+ G

+ M
91.0

[83.2, 100.0]
91.7

[86.1, 96.7]
Shank (right)97.9

[92.0, 98.0]*
100.0

[98.0, 100.0]
+ G

+ M
97.5

[93.0, 98.3]*
98.3

[97.5, 100.0]
+ G88.0

[78.2, 90.9]*
90.0

[83.6, 97.2]
+ G92.9

[81.3, 100.0]
Shank (left)96.8

[94.0, 98.7]
97.1

[94.2, 98.9]
83.5

[70.0, 91.8]*
85.0

[76.7, 92.8]
Foot (right)82.0

[80.0, 86.0]*
81.7

[78.0, 83.6]*
87.5

[79.1, 93.1]*
84.2

[75.0, 90.0]
Foot (left)84.0

[82.0, 86.7]*
90.0

[83.9, 92.7]*
+ G

+ M
81.3

[80.0, 86.1]*
82.9

[71.6, 90.0]*
80.6

[75.0, 96.7]

Note. Indicated is the median balanced sensitivity and specificity with 95% Cl in brackets. The column “Best signal combination” indicates whether the addition of the gyroscope (+ G) and/or magnetometer (+ M) significantly improved the accuracy (empty if not). The accelerometer accuracy with lowest rank sum (Friedman test) of each behavior is marked in bold. MET = metabolic equivalent; CI = confidence interval.

*Significantly different.

Discussion

This study compared the accuracy of 60 placement-signal type combinations to classify SB, as well as active sitting, inactive standing, active standing, and walking. Furthermore, the study also presents the isolated posture, as well as the in-/activity classification. The results of this study support future method developments and algorithm refinements for SB in the choice of sensor placement and signal type.

Overall, the SB classification was most accurately solved by the thigh and shank accelerometers, as well as the waist IMU. In particular, the classification of SB and inactive standing strongly depended on sensor placement, while the placement dependence was less pronounced for the active behaviors, regardless of body posture. The isolated analysis of posture and in-/activity revealed that the difference results mainly from the posture classification. The thigh and shank accelerometers solved this classification significantly best, even when adding the gyroscope and/or magnetometer data to the other placements. This result demonstrates that posture is best measured with a sensor worn on the lower extremity, while the accelerometers attached to the upper extremity and trunk performed better to detect the in-/activity level, in particular, for sitting. Presumably, this is due to the fact that the activity while sitting is mainly caused by upper body motions that are less easy to detect with sensors worn at the lower extremities. But still, even the lower body sensors were able to detect the activity level to a certain amount, presumably because upper body motions raising the MET level above 1.5 cause a certain motion pattern of the lower body, which the models learned to detect. Interestingly, the inclusion of additional sensor signals has not improved the in-/activity classification. We therefore conclude that the acceleration already contains the relevant signal information to solve the in-/activity classification. In contrast, the inclusion of additional sensor signals improved the posture classification for some sensor placements (e.g., waist, Table 4). Overall, SB was most often confused with inactive standing, and inactive standing with SB, and the active behaviors were confused with the inactive behaviors in the same posture (see misclassification table in Supplementary Material 3 [available online]).

So far, three accelerometer placements are commonly used to measure SB: the thigh, the waist, and the wrist (Matthews, Hagstromer, Pober, & Bowles, 2012). In the present study on detecting SB in desk-based office work, the thigh accelerometer significantly outperformed the other two accelerometer placements. However, previous studies used the thigh sensor as a posture-based method only (Kim et al., 2015; Skotte et al., 2014). This study shows that an accelerometer worn on the thigh is also accurate in measuring the in-/activity level while sitting as well as standing. The presented data suggest that the combination of an accelerometer and a gyroscope might perform even better than an accelerometer only (results for left thigh in Table 3), but future studies are needed to draw a sound conclusion. However, the inclusion of a magnetometer as can be found in the activPAL4, did not improve the SB classification. In contrast, the waist accelerometer alone performed significantly worse, and the combination with the data of a gyroscope and magnetometer significantly improved the posture and, thus, the SB classification. Accordingly, if a future study wants to calibrate a waist-worn sensor to measure SB, we recommend the use of a nine-dimensional IMU, for example, like the ActiGraph Link. However, it must be noted that the waist-worn IMU was the only sensor with lower accuracy to detect the standing-like sitting posture on the saddle chair as compared with the normal sitting posture on the conventional chair, and that the waist-worn IMU model uses many more features than the thigh-worn accelerometer model (Supplementary Material 4 [available online]). The presented data show that the wrist placement is accurate to classify the activity level, but we noticed a limited accuracy to measure posture and, thus, SB. The inclusion of additional sensor signals has not improved the accuracy of the wrist placement. This makes wrist placement not the preferred choice measure SB from an accuracy perspective.

In this regard, it is important to note that accuracy is only one aspect on which the choice of sensor placement and signal type should be based. Other important and more pragmatic aspects include sensor pricing, wear comfort, protocol compliance, and the primary study aim. Since each sensor placement has its own pros and cons, there is no placement that outperforms all others. A large-scaled epidemiological study focusing on the total time spent sedentary per day might have to use another method than a workplace intervention study aiming to break up prolonged SB. The former might consider comfort and compliance to be most important, while the latter has to ensure a high accuracy to detect behavioral changes. However, in any case, both studies should base their sensor placement and signal type decision on the available evidence and discuss their findings in terms of the decisions made, be it a limited accuracy or a pragmatic limitation in data recording.

Methodological Considerations

This study examined the fundamental accuracy of different sensor placements, whether accelerometers or IMUs, to detect body posture and in-/activity level during typical desk-based activities. For this reason, the study used a workplace at the ZHAW Zurich Univeristy of Applied Sciences and prescribed activities to ensure a safe handling of the indirect calorimeter and the motion capture system. Other studies used direct observation as the only reference criterion for the SB classification in combination with the Compendium of Physical Activity (Ainsworth et al., 2011; Lyden, Keadle, Staudenmayer, & Freedson, 2014). Direct observation allows for a much more field-like data recording; however, it classifies seated activities as SB, and nonseated activities as non-SB. This, in fact, turns the SB classification into a posture classification and neglects the MET component of SB. The protocol used in this study classified the posture and in-/activity level independently and individually. The same task could be classified for one subject as SB and for another as active sitting, and each task was performed while sitting, as well as standing (Table 2). This is also in contrast to the vast majority of calibration studies that used a predefined SB classification on task level, with only a small fraction of sedentary tasks and an artificial large gap between the prescribed sedentary and nonsedentary tasks in terms of posture and in-/activity (e.g., Montoye, Pivarnik, Mudd, Biswas, & Pfeiffer, 2016; Staudenmayer, He, Hickey, Sasaki, & Freedson, 2015; Zhang et al., 2012). As a consequence, the trained models in those studies detect SB through a correctly classified posture or in-/activity level, while the trained models in the present study detect SB through a correctly classified posture and in-/activity level. Accordingly, the accuracies presented in this study for the thigh, waist, and wrist are lower than those presented in other studies, but are expected to be much closer to reality, where SB is very common, not every sitting task is necessarily SB, and the separation between SB and non-SB is not always that obvious. Furthermore, to account for the variability in real life, the prescribed tasks were not standardized and were only orally explained without demonstration, and 1/3 of the study population used a laptop while 2/3 used a desktop computer. No differences between the two workplaces were found.

The indirect calorimeter used to measure the MET level requires a steady state to determine the true energy expenditure. With three pilot subjects doing 17 tasks, we observed that steady state was typically reached after 2 min, but no later than 4 min. Consequently, the subjects of this study performed each activity for 5 min, and each minute was categorized into inactive/active behavior using the steady-state energy expenditure and a cut point of 1.5 MET. Contrary to the Terminology Consensus Project of the Sedentary Behavior Research Network (Tremblay et al., 2017), the 1.5 MET cut point was also applied to separate inactive/active standing, as we see no evidence that this cut point should be higher for standing than sitting. In our data, there was no MET difference with respect to body posture, and the time spent >1.5 MET should be considered light physically active, regardless of posture. To calculate the MET, the steady-state energy expenditure of each task was referenced to the resting metabolic rate (Kozey, Lyden, Staudenmayer, & Freedson, 2010). Most calibration studies used a standardized MET of 3.5 ml · kg−1 · min−1 VO2 or an approximated MET based on personal characteristics (Kim et al., 2017; Kozey et al., 2010). However, both neglect individual variation and ambient factors, like temperature (Borges et al., 2016; Popp et al., 2016). Although the resting metabolic rate was on average 3.5 ml · kg−1 · min−1 VO2, the results of the present study should not be interpreted as if the standardized value was taken.

To record the body-worn sensor signals, the Xsens Biomech Awinda was used. The system is a user-friendly whole-body measurement system allowing one to record with 15 IMUs synchronously. However, the recording frequency cannot be adjusted (set to 60 Hz). The subjects in this study additionally wore four standalone MSR145 accelerometers (MSR Electronics GmbH, Seuzach, Switzerland) recording with 20 Hz at both thighs, the waist, and the sternum (range ±2 g). With exactly the same data processing, the developed models for these sensors showed the same significances as the presented models. From this observation, we conclude that the presented differences between the sensor placements do not depend on the sensor manufacturer, nor the recording frequency and the recording range, as long as a model is used with the same sensor as it was developed with.

To develop simple classification models with only relevant features, the study used a stepwise feature inclusion. The stepwise feature inclusion is a wrapper method that identifies the most relevant feature set with consideration of the final model algorithm. However, the method has two significant limitations. First, it is computationally very expensive. This is why it was combined with a previous feature filtering, to limit the number of features to be inspected. Second, every step requires a performance analysis to identify the most relevant feature. This requires a cross-validation approach, like the leave-one-subject-out. The leave-one-subject-out approach provides a good estimate for the model performance in the recorded data, but other studies reported a 0%–15% overestimation for the model’s generalizability to new data (Gyllensten & Bonomi, 2011; Kerr et al., 2016; Montoye, Westgate, Fonley, & Pfeiffer, 2018). These studies also reported a substantially smaller overestimation for sitting and standing than for walking, running, and cycling. Accordingly, we expect that the overestimation is closer to 0% than 15%, but it is impossible to make a reliable conclusion unless an independent field validation is conducted. Since the overestimation applies equally to all models, we consider the observed differences to be accurate, and the values presented in Tables 3 and 4 could be interpreted in terms of model generalizability as maximum accuracies that can be reached. An alternative solution to estimate the model generalizability would be to split the sample into a training and testing set and report only the accuracy of the testing set. However, this would generate a similarly biased estimation of the generalizability, as the testing set was recorded in exactly the same setting as the training set. Furthermore, splitting the sample means decreasing the training sample and thus weakening the model development. To analyze the models’ generalizability to new data, it is essential to perform an independent field validation. However, there are three main reasons why this was not included in the present study. First, this study aimed to compare the fundamental accuracy of various sensor placements and signal types to inform future Phase 1 and Phase 2 algorithm developments and refinements and not to present a field-ready classification model (Keadle, Lyden, Strath, Staudenmayer, & Freedson, 2019). Second, the sensors of the motion capture system record only for a limited amount of time (up to 8 hr) and require a wireless connection to a computer, and the system does not allow one to record with a user-specified number of sensors. Accordingly, the system’s usability in field studies is severely limited, but its usability in the present study was very high. Third, there is still no valid reference criteria to measure physical in-/activity in desk-based office work on an individual subject level in field settings. With respect to future method developments and validations, we see an urgent need to have such a method. We therefore highly recommend analyzing whether steady-state detection algorithms for indirect calorimetry data as used in other research areas (e.g., the analysis of steady-state energy expenditure with variable walking speeds, Plasschaert, Jones, & Forward, 2009; Schwartz, 2007) could be adapted to our field of research. An independent field validation without such a method is pointless for the presented models. In this regard, it should generally be considered more critical that measuring SB is often equated with measuring sitting (when using a posture-based device) or a lack of physical activity (when using an activity-based device), but in both cases interpreted as it would be SB. For posture-based devices, we recommend referring to “sitting,” and for activity-based devices, we recommend referring to “minimal physical activity” (Holtermann et al., 2017). This study not only presents the SB accuracy, but also presents the accuracy for the isolated posture and in-/activity classification for each sensor placement and signal type. This information might be useful for future studies, to uncover the relevance of the two aspects of SB: posture and physical inactivity.

All developed models were at last statistically compared in order to identify the one with the highest accuracy for each placement. We thereby took into account that the inclusion of an additional signal type makes the sensor more expensive and the data processing more complex. Unless there was a significant improvement, only the accelerometer results are presented. Furthermore, this study started with a very large feature set informed by previous studies in this field of research, and calculated each feature for each signal type. Since previous studies used accelerometers almost exclusively, there might be other features more suitable for the gyroscope and magnetometer data. It remains subject to future studies to investigate whether the inclusion of other signal features for those two sensor types improves the classification accuracy. Last, the numbers presented in Tables 3 and 4 are based on equally fractioned office tasks; however, it is unlikely that office workers spend their time equally balanced in real life. The interested reader, therefore, finds in Supplementary Material 5 (available online) the accuracy of each sensor placement and signal type separated by behavior and office task. From this table, they can inform themselves about the most accurate sensor placement in relation to the expected behavior in a future data collection.

Conclusion

The presented method development and comparison shows that posture is best measured with lower body sensors, while upper body sensors performed better to detect the in-/activity level, in particular, for sitting. The study also shows that the acceleration signal contains the relevant signal information to solve the in-/activity classification, but the inclusion of additional sensor signals improved the posture classification for some placements. Future algorithm developments and refinements should consider the results of this study in combination with pragmatic aspects derived from field studies as a basis for decision making when choosing sensor placement and signal type. In line with previous work (Kozey-Keadle, Libertine, Lyden, Staudenmayer, & Freedson, 2011; Montoye et al., 2018), this study favors the measurement of SB with a thigh-worn accelerometer and adds the information that such a sensor is also accurate in measuring the in-/activity level while sitting and standing.

Acknowledgments

R.P. Kuster was partially funded through a personal grant from the Swiss National Science Foundation (grant ID P1SKP3_187637). The other authors did not receive any funding from external parties. They do not declare a conflict of interest. The authors acknowledge the support of Cahit Atilgan in programming the Python-based feature filtering program, and the support of Mary Huggler in recruiting the participants.

References

  • Ainsworth, B.E., Haskell, W.L., Herrmann, S.D., Meckes, N., Bassett, D.R., Jr., Tudor-Locke, C., … Leon, A.S. (2011). 2011 Compendium of Physical Activities: A second update of codes and MET values. Medicine & Science in Sports & Exercise, 43(8), 15751581. PubMed ID: 21681120 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Amirfaiz, S., & Shahril, M.R. (2019). Objectively measured physical activity, sedentary behavior, and metabolic syndrome in adults: Systematic review of observational evidence. Metabolic Syndrome and Related Disorders, 17(1), 121. PubMed ID: 30272527 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Borges, J.H., Langer, R.D., Cirolini, V.X., Pascoa, M.A., Guerra-Junior, G., & Goncalves, E.M. (2016). Minimum time to achieve the steady state and optimum abbreviated period to estimate the resting energy expenditure by indirect calorimetry in healthy young adults. Nutrition in Clinical Practice, 31(3), 349354. PubMed ID: 26888859 doi: doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buckley, J.P., Hedge, A., Yates, T., Copeland, R.J., Loosemore, M., Hamer, M., … Dunstan, D.W. (2015). The sedentary office: An expert statement on the growing case for change towards better health and productivity. British Journal of Sports Medicine, 49(21), 13571362. PubMed ID: 26034192 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Burns, J., Forde, C., & Dockrell, S. (2017). Energy Expenditure of Standing Compared to Sitting While Conducting Office Tasks. Human Factors, 59(7), 10781087.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crouter, S.E., LaMunion, S.R., Hibbingi, P.R., Kaplan, A.S., Quarantillo, M.E., & Bassett, D.R. (2019). Accuracy of the Cosmed K5 Portable Metabolic System. Medicine & Science in Sports & Exercise, 51(65), 147 doi: doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Edwardson, C.L., Rowlands, A.V., Bunnewell, S., Sanders, J., Esliger, D.W., Gorely, T., … Yates, T. (2016). Accuracy of posture allocation algorithms for thigh- and waist-worn accelerometers. Medicine & Science in Sports & Exercise, 48(6), 10851090. PubMed ID: 26741122 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Edwardson, C.L., Winkler, E.A.H., Bodicoat, D.H., Yates, T., Davies, M.J., Dunstan, D.W., & Healy, G.N. (2017). Considerations when using the activPAL monitor in field-based research with adult populations. Journal of Sport and Health Science, 6(2), 162178. PubMed ID: 30356601 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellegast, R.P., Kraft, K., Groenesteijn, L., Krause, F., Berger, H., & Vink, P. (2012). Comparison of four specific dynamic office chairs with a conventional office chair: Impact upon muscle activation. Applied Ergonomics, 43(2), 296307.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellingson, L.D., Schwabacher, I.J., Kim, Y., Welk, G.J., & Cook, D.B. (2016). Validity of an integrative method for processing physical activity data. Medicine & Science in Sports & Exercise, 48(8), 16291638. PubMed ID: 27015380 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellis, K., Kerr, J., Godbole, S., Staudenmayer, J., & Lanckriet, G. (2016). Hip and wrist accelerometer algorithms for free-living behavior classification. Medicine & Science in Sports & Exercise, 48(5), 933940. PubMed ID: 26673126 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fanchamps, M.H.J., van den Berg-Emons, H.J.G., Stam, H.J., & Bussmann, J.B.J. (2017). Sedentary behavior: Different types of operationalization influence outcome measures. Gait & Posture, 54, 188193. PubMed ID: 28324755 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Godfrey, A., Culhane, K.M., & Lyons, G.M. (2007). Comparison of the performance of the activPAL Professional physical activity logger to a discrete accelerometer-based activity monitor. Medical Engineering and Physics, 29(8), 930934. PubMed ID: 17134934 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grant, P.M., Ryan, C.G., Tigbe, W.W., & Granat, M.H. (2006). The validation of a novel activity monitor in the measurement of posture and motion during everyday activities. British Journal of Sports Medicine, 40(12), 992997. PubMed ID: 16980531 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grooten, W.J.A., Conradsson, D., Ang, B.O., & Franzen, E. (2013). Is active sitting as active as we think? Ergonomics, 56(8), 13041314.

  • Gyllensten, I.C., & Bonomi, A.G. (2011). Identifying types of physical activity with a single accelerometer: Evaluating laboratory-trained algorithms in daily life. IEEE Transactions on Biomedical Engineering, 58(9), 26562663. PubMed ID: 21712150 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Holtermann, A., Schellewald, V., Mathiassen, S.E., Gupta, N., Pinder, A., Punakallio, A., … Ellegast, R. (2017). A practical guidance for assessments of sedentary behavior at work: A PEROSH initiative. Applied Ergonomics, 63, 4152. PubMed ID: 28502405 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutcheson, A.K., Piazza, A.J., & Knowlden, A.P. (2018). Work site-based environmental interventions to reduce sedentary behavior: A systematic review. American Journal of Health Promotion, 32(1), 3247. PubMed ID: 27780893 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kang, M., & Rowe, D.A. (2015). Issues and challenges in sedentary behavior measurement. Measurement in Physical Education and Exercise Science, 19(3), 105115. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Keadle, S.K., Lyden, K.A., Strath, S.J., Staudenmayer, J.W., & Freedson, P.S. (2019). A framework to evaluate devices that assess physical behavior. Exercise and Sport Sciences Reviews, 47(4), 206214. PubMed ID: 31524786 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Keown, M.K., Skeaff, C.M., Perry, T.L., Haszard, J.J., & Peddie, M.C. (2018). Device-measured sedentary behavior patterns in office-based university employees. Journal of Occupational and Environmental Medicine, 60(12), 11501157. PubMed ID: 30308622 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kerr, J., Marinac, C.R., Ellis, K., Godbole, S., Hipp, A., Glanz, K., … Berrigan, D. (2017). Comparison of accelerometry methods for estimating physical activity. Medicine & Science in Sports & Exercise, 49(3), 617624. PubMed ID: 27755355 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kerr, J., Patterson, R.E., Ellis, K., Godbole, S., Johnson, E., Lanckriet, G., & Staudenmayer, J. (2016). Objective assessment of physical activity: Classifiers for public health. Medicine & Science in Sports & Exercise, 48(5), 951957. PubMed ID: 27089222 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, D., Lee, J., Park, H.K., Jang, D.P., Song, S., Cho, B.H., … Kim, I.Y. (2017). Comparing the standards of one metabolic equivalent of task in accurately estimating physical activity energy expenditure based on acceleration. Journal of Sports Sciences, 35(13), 12791286. PubMed ID: 27556835 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, Y., Barry, V.W., & Kang, M. (2015). Validation of the ActiGraph GT3X and activPAL accelerometers for the assessment of sedentary behavior. Measurement in Physical Education and Exercise Science, 19(3), 125137. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozey, S., Lyden, K., Staudenmayer, J., & Freedson, P. (2010). Errors in MET estimates of physical activities using 3.5 ml × kg−1 × min−1 as the baseline oxygen consumption. Journal of Physical Activity and Health, 7(4), 508516. PubMed ID: 20683093 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozey-Keadle, S., Libertine, A., Lyden, K., Staudenmayer, J., & Freedson, P.S. (2011). Validation of wearable monitors for assessing sedentary behavior. Medicine & Science in Sports & Exercise, 43(8), 15611567. PubMed ID: 21233777 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kuster, R.P., Grooten, W.J.A., Baumgartner, D., Blom, V., Hagstromer, M., & Ekblom, O. (2020). Detecting prolonged sitting bouts with the ActiGraph GT3X. Scandinavian Journal of Medicine & Science in Sports, 30(3), 572582. PubMed ID: 31743494 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kuster, R.P., Huber, M., Hirschi, S., Siegl, W., Baumgartner, D., Hagströmer, M., & Grooten, W. (2018). Measuring sedentary behavior by means of muscular activity and accelerometry. Sensors, 18(11). doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Loyen, A., Clarke-Cornwell, A.M., Anderssen, S.A., Hagstromer, M., Sardinha, L.B., Sundquist, K., … van der Ploeg, H.P. (2017). Sedentary time and physical activity surveillance through accelerometer pooling in four European countries. Sports Medicine, 47(7), 14211435. PubMed ID: 27943147 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lyden, K., Keadle, S.K., Staudenmayer, J., & Freedson, P.S. (2014). A method to estimate free-living active and sedentary behavior from an accelerometer. Medicine & Science in Sports & Exercise, 46(2), 386397. PubMed ID: 23860415 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matthews, C.E., Hagstromer, M., Pober, D.M., & Bowles, H.R. (2012). Best practices for using physical activity monitors in population-based research. Medicine & Science in Sports & Exercise, 44(1, Suppl. 1), S68S76. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matthews, C.E., Kozey-Keadle, S., Moore, S.C., Schoeller, D.S., Carroll, R.J., Troiano, R.P., & Sampson, J.N. (2018). Measurement of active and sedentary behavior in context of large epidemiologic studies. Medicine & Science in Sports & Exercise, 50(2), 266276. PubMed ID: 28930863 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Migueles, J.H., Cadenas-Sanchez, C., Ekelund, U., Delisle Nystrom, C., Mora-Gonzalez, J., Lof, M., … Ortega, F.B. (2017). Accelerometer data collection and processing criteria to assess physical activity and other outcomes: A systematic review and practical considerations. Sports Medicine, 47(9), 18211845. PubMed ID: 28303543 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montoye, A.H.K., Pivarnik, J.M., Mudd, L.M., Biswas, S., & Pfeiffer, K.A. (2016). Validation and comparison of accelerometers worn on the hip, thigh, and wrists for measuring physical activity and sedentary behavior. AIMS Public Health, 3(2), 298312. PubMed ID: 29546164 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montoye, A.H.K., Westgate, B.S., Fonley, M.R., & Pfeiffer, K.A. (2018). Cross-validation and out-of-sample testing of physical activity intensity predictions with a wrist-worn accelerometer. J Appl Physiol, 124(5), 12841293. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Neuhaus, M., Eakin, E.G., Straker, L., Owen, N., Dunstan, D.W., Reid, N., & Healy, G.N. (2014). Reducing occupational sedentary time: A systematic review and meta-analysis of evidence on activity-permissive workstations. Obesity Reviews, 15(10), 822838. PubMed ID: 25040784 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nooijen, C., Kallings, L., Blom, V., Ekblom, O., Forsell, Y., & Ekblom, M. (2018). Common perceived barriers and facilitators for reducing sedentary behaviour among office workers. Journal of Physical Activity and Health, 15(10), S94S95. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Plasschaert, F., Jones, K., & Forward, M. (2009). Energy cost of walking: Solving the paradox of steady state in the presence of variable walking speed. Gait & Posture, 29(2), 311316. PubMed ID: 19027301 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Popp, C.J., Tisch, J.J., Sakarcan, K.E., Bridges, W.C., & Jesch, E.D. (2016). Approximate time to steady-state resting energy expenditure using indirect calorimetry in young, healthy adults. Frontiers in Nutrition, 3, 49. PubMed ID: 27857943 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rosenberger, M.E., Haskell, W.L., Albinali, F., Mota, S., Nawyn, J., & Intille, S. (2013). Estimating activity and sedentary behavior from an accelerometer on the hip or wrist. Medicine & Science in Sports & Exercise, 45(5), 964975. PubMed ID: 23247702 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rothney, M.P., Schaefer, E.V., Neumann, M.M., Choi, L., & Chen, K.Y. (2008). Validity of physical activity intensity predictions by ActiGraph, Actical, and RT3 accelerometers. Obesity, 16(8), 19461952. PubMed ID: 18535553 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, M.H. (2007). Protocol changes can improve the reliability of net oxygen cost data. Gait & Posture, 26(4), 494500. PubMed ID: 17855095 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sedentary Behaviour Research Network. (2012). Letter to the editor: Standardized use of the terms “sedentary” and “sedentary behaviours”. Applied Physiology, Nutrition, and Metabolism, 37(3), 540542. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Shrestha, N., Kukkonen-Harjula, K.T., Verbeek, J.H., Ijaz, S., Hermans, V., & Pedisic, Z. (2018). Workplace interventions for reducing sitting at work. Cochrane Database of Systematic Reviews, (6), CD010912. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Skotte, J., Korshoj, M., Kristiansen, J., Hanisch, C., & Holtermann, A. (2014). Detection of physical activity types using triaxial accelerometers. Journal of Physical Activity and Health, 11(1), 7684. PubMed ID: 23249722 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stamatakis, E., Ekelund, U., Ding, D., Hamer, M., Bauman, A.E., & Lee, I.M. (2019). Is the time right for quantitative public health guidelines on sitting? A narrative review of sedentary behaviour research paradigms and findings. British Journal of Sports Medicine, 53(6), 377382. PubMed ID: 29891615 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Staudenmayer, J., He, S., Hickey, A., Sasaki, J., & Freedson, P. (2015). Methods to estimate aspects of physical activity and sedentary behavior from high-frequency wrist accelerometer measurements. Journal of Applied Physiology (Bethesda, Md.: 1985), 119(4), 396403. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stemland, I., Ingebrigtsen, J., Christiansen, C.S., Jensen, B.R., Hanisch, C., Skotte, J., & Holtermann, A. (2015). Validity of the Acti4 method for detection of physical activity types in free-living settings: Comparison with video analysis. Ergonomics, 58(6), 953965. PubMed ID: 25588819 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stephenson, A., McDonough, S.M., Murphy, M.H., Nugent, C.D., & Mair, J.L. (2017). Using computer, mobile and wearable technology enhanced interventions to reduce sedentary behaviour: A systematic review and meta-analysis. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 17. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tew, G.A., Posso, M.C., Arundel, C.E., & McDaid, C.M. (2015). Systematic review: Height-adjustable workstations to reduce sedentary behaviour in office-based workers. Occupational Medicine (Oxford, England), 65(5), 357366. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tremblay, M.S., Aubert, S., Barnes, J.D., Saunders, T.J., Carson, V., Latimer-Cheung, A.E., … Participants, S.T.C.P. (2017). Sedentary Behavior Research Network (SBRN)—Terminology Consensus Project process and outcome. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 75. PubMed ID: 28599680 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van der Velde, J., Schaper, N.C., Stehouwer, C.D.A., van der Kallen, C.J.H., Sep, S.J.S., Schram, M.T., … Koster, A. (2018). Which is more important for cardiometabolic health: Sedentary time, higher intensity physical activity or cardiorespiratory fitness? The Maastricht Study. Diabetologia, 61(12), 25612569. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van Uffelen, J.G.Z., Wong, J., Chau, J.Y., van der Ploeg, H.P., Riphagen, I., Gilson, N.D., … Brown, W.J. (2010). Occupational sitting and health risks: A systematic review. American Journal of Preventive Medicine, 39(4), 379388. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhang, S., Murray, P., Zillmer, R., Eston, R.G., Catt, M., & Rowlands, A.V. (2012). Activity classification using the GENEA: Optimum sampling frequency and number of axes. Medicine & Science in Sports & Exercise, 44(11), 22282234. PubMed ID: 22617400 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation

If the inline PDF is not rendering correctly, you can download the PDF file here.

Kuster, Hagströmer, and Grooten are with the Division of Physiotherapy, Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden. Kuster and Baumgartner are with the Institute of Mechanical Systems, School of Engineering, ZHAW Zurich University of Applied Sciences, Winterthur, Switzerland. Hagströmer and Grooten are also with Function Area Occupational Therapy and Physiotherapy, Allied Health Professionals, Karolinska University Hospital, Stockholm, Sweden. Hagströmer is also with the Department of Health Promoting Science, Sophiahemmet University, Stockholm, Sweden.

Kuster (roman.kuster@alumni.ethz.ch) is corresponding author.
  • View in gallery

    Investigated office tasks: (a) mouse, (b) keyboard, (c) deskwork, and (d) sorting, and conditions: (a and d) conventional office chair, (b) saddle chair, and (c) standing. Details for each task are given in Table 1.

  • Ainsworth, B.E., Haskell, W.L., Herrmann, S.D., Meckes, N., Bassett, D.R., Jr., Tudor-Locke, C., … Leon, A.S. (2011). 2011 Compendium of Physical Activities: A second update of codes and MET values. Medicine & Science in Sports & Exercise, 43(8), 15751581. PubMed ID: 21681120 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Amirfaiz, S., & Shahril, M.R. (2019). Objectively measured physical activity, sedentary behavior, and metabolic syndrome in adults: Systematic review of observational evidence. Metabolic Syndrome and Related Disorders, 17(1), 121. PubMed ID: 30272527 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Borges, J.H., Langer, R.D., Cirolini, V.X., Pascoa, M.A., Guerra-Junior, G., & Goncalves, E.M. (2016). Minimum time to achieve the steady state and optimum abbreviated period to estimate the resting energy expenditure by indirect calorimetry in healthy young adults. Nutrition in Clinical Practice, 31(3), 349354. PubMed ID: 26888859 doi: doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buckley, J.P., Hedge, A., Yates, T., Copeland, R.J., Loosemore, M., Hamer, M., … Dunstan, D.W. (2015). The sedentary office: An expert statement on the growing case for change towards better health and productivity. British Journal of Sports Medicine, 49(21), 13571362. PubMed ID: 26034192 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Burns, J., Forde, C., & Dockrell, S. (2017). Energy Expenditure of Standing Compared to Sitting While Conducting Office Tasks. Human Factors, 59(7), 10781087.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crouter, S.E., LaMunion, S.R., Hibbingi, P.R., Kaplan, A.S., Quarantillo, M.E., & Bassett, D.R. (2019). Accuracy of the Cosmed K5 Portable Metabolic System. Medicine & Science in Sports & Exercise, 51(65), 147 doi: doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Edwardson, C.L., Rowlands, A.V., Bunnewell, S., Sanders, J., Esliger, D.W., Gorely, T., … Yates, T. (2016). Accuracy of posture allocation algorithms for thigh- and waist-worn accelerometers. Medicine & Science in Sports & Exercise, 48(6), 10851090. PubMed ID: 26741122 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Edwardson, C.L., Winkler, E.A.H., Bodicoat, D.H., Yates, T., Davies, M.J., Dunstan, D.W., & Healy, G.N. (2017). Considerations when using the activPAL monitor in field-based research with adult populations. Journal of Sport and Health Science, 6(2), 162178. PubMed ID: 30356601 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellegast, R.P., Kraft, K., Groenesteijn, L., Krause, F., Berger, H., & Vink, P. (2012). Comparison of four specific dynamic office chairs with a conventional office chair: Impact upon muscle activation. Applied Ergonomics, 43(2), 296307.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellingson, L.D., Schwabacher, I.J., Kim, Y., Welk, G.J., & Cook, D.B. (2016). Validity of an integrative method for processing physical activity data. Medicine & Science in Sports & Exercise, 48(8), 16291638. PubMed ID: 27015380 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellis, K., Kerr, J., Godbole, S., Staudenmayer, J., & Lanckriet, G. (2016). Hip and wrist accelerometer algorithms for free-living behavior classification. Medicine & Science in Sports & Exercise, 48(5), 933940. PubMed ID: 26673126 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fanchamps, M.H.J., van den Berg-Emons, H.J.G., Stam, H.J., & Bussmann, J.B.J. (2017). Sedentary behavior: Different types of operationalization influence outcome measures. Gait & Posture, 54, 188193. PubMed ID: 28324755 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Godfrey, A., Culhane, K.M., & Lyons, G.M. (2007). Comparison of the performance of the activPAL Professional physical activity logger to a discrete accelerometer-based activity monitor. Medical Engineering and Physics, 29(8), 930934. PubMed ID: 17134934 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grant, P.M., Ryan, C.G., Tigbe, W.W., & Granat, M.H. (2006). The validation of a novel activity monitor in the measurement of posture and motion during everyday activities. British Journal of Sports Medicine, 40(12), 992997. PubMed ID: 16980531 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grooten, W.J.A., Conradsson, D., Ang, B.O., & Franzen, E. (2013). Is active sitting as active as we think? Ergonomics, 56(8), 13041314.

  • Gyllensten, I.C., & Bonomi, A.G. (2011). Identifying types of physical activity with a single accelerometer: Evaluating laboratory-trained algorithms in daily life. IEEE Transactions on Biomedical Engineering, 58(9), 26562663. PubMed ID: 21712150 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Holtermann, A., Schellewald, V., Mathiassen, S.E., Gupta, N., Pinder, A., Punakallio, A., … Ellegast, R. (2017). A practical guidance for assessments of sedentary behavior at work: A PEROSH initiative. Applied Ergonomics, 63, 4152. PubMed ID: 28502405 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutcheson, A.K., Piazza, A.J., & Knowlden, A.P. (2018). Work site-based environmental interventions to reduce sedentary behavior: A systematic review. American Journal of Health Promotion, 32(1), 3247. PubMed ID: 27780893 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kang, M., & Rowe, D.A. (2015). Issues and challenges in sedentary behavior measurement. Measurement in Physical Education and Exercise Science, 19(3), 105115. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Keadle, S.K., Lyden, K.A., Strath, S.J., Staudenmayer, J.W., & Freedson, P.S. (2019). A framework to evaluate devices that assess physical behavior. Exercise and Sport Sciences Reviews, 47(4), 206214. PubMed ID: 31524786 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Keown, M.K., Skeaff, C.M., Perry, T.L., Haszard, J.J., & Peddie, M.C. (2018). Device-measured sedentary behavior patterns in office-based university employees. Journal of Occupational and Environmental Medicine, 60(12), 11501157. PubMed ID: 30308622 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kerr, J., Marinac, C.R., Ellis, K., Godbole, S., Hipp, A., Glanz, K., … Berrigan, D. (2017). Comparison of accelerometry methods for estimating physical activity. Medicine & Science in Sports & Exercise, 49(3), 617624. PubMed ID: 27755355 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kerr, J., Patterson, R.E., Ellis, K., Godbole, S., Johnson, E., Lanckriet, G., & Staudenmayer, J. (2016). Objective assessment of physical activity: Classifiers for public health. Medicine & Science in Sports & Exercise, 48(5), 951957. PubMed ID: 27089222 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, D., Lee, J., Park, H.K., Jang, D.P., Song, S., Cho, B.H., … Kim, I.Y. (2017). Comparing the standards of one metabolic equivalent of task in accurately estimating physical activity energy expenditure based on acceleration. Journal of Sports Sciences, 35(13), 12791286. PubMed ID: 27556835 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, Y., Barry, V.W., & Kang, M. (2015). Validation of the ActiGraph GT3X and activPAL accelerometers for the assessment of sedentary behavior. Measurement in Physical Education and Exercise Science, 19(3), 125137. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozey, S., Lyden, K., Staudenmayer, J., & Freedson, P. (2010). Errors in MET estimates of physical activities using 3.5 ml × kg−1 × min−1 as the baseline oxygen consumption. Journal of Physical Activity and Health, 7(4), 508516. PubMed ID: 20683093 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozey-Keadle, S., Libertine, A., Lyden, K., Staudenmayer, J., & Freedson, P.S. (2011). Validation of wearable monitors for assessing sedentary behavior. Medicine & Science in Sports & Exercise, 43(8), 15611567. PubMed ID: 21233777 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kuster, R.P., Grooten, W.J.A., Baumgartner, D., Blom, V., Hagstromer, M., & Ekblom, O. (2020). Detecting prolonged sitting bouts with the ActiGraph GT3X. Scandinavian Journal of Medicine & Science in Sports, 30(3), 572582. PubMed ID: 31743494 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kuster, R.P., Huber, M., Hirschi, S., Siegl, W., Baumgartner, D., Hagströmer, M., & Grooten, W. (2018). Measuring sedentary behavior by means of muscular activity and accelerometry. Sensors, 18(11). doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Loyen, A., Clarke-Cornwell, A.M., Anderssen, S.A., Hagstromer, M., Sardinha, L.B., Sundquist, K., … van der Ploeg, H.P. (2017). Sedentary time and physical activity surveillance through accelerometer pooling in four European countries. Sports Medicine, 47(7), 14211435. PubMed ID: 27943147 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lyden, K., Keadle, S.K., Staudenmayer, J., & Freedson, P.S. (2014). A method to estimate free-living active and sedentary behavior from an accelerometer. Medicine & Science in Sports & Exercise, 46(2), 386397. PubMed ID: 23860415 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matthews, C.E., Hagstromer, M., Pober, D.M., & Bowles, H.R. (2012). Best practices for using physical activity monitors in population-based research. Medicine & Science in Sports & Exercise, 44(1, Suppl. 1), S68S76. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Matthews, C.E., Kozey-Keadle, S., Moore, S.C., Schoeller, D.S., Carroll, R.J., Troiano, R.P., & Sampson, J.N. (2018). Measurement of active and sedentary behavior in context of large epidemiologic studies. Medicine & Science in Sports & Exercise, 50(2), 266276. PubMed ID: 28930863 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Migueles, J.H., Cadenas-Sanchez, C., Ekelund, U., Delisle Nystrom, C., Mora-Gonzalez, J., Lof, M., … Ortega, F.B. (2017). Accelerometer data collection and processing criteria to assess physical activity and other outcomes: A systematic review and practical considerations. Sports Medicine, 47(9), 18211845. PubMed ID: 28303543 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montoye, A.H.K., Pivarnik, J.M., Mudd, L.M., Biswas, S., & Pfeiffer, K.A. (2016). Validation and comparison of accelerometers worn on the hip, thigh, and wrists for measuring physical activity and sedentary behavior. AIMS Public Health, 3(2), 298312. PubMed ID: 29546164 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montoye, A.H.K., Westgate, B.S., Fonley, M.R., & Pfeiffer, K.A. (2018). Cross-validation and out-of-sample testing of physical activity intensity predictions with a wrist-worn accelerometer. J Appl Physiol, 124(5), 12841293. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Neuhaus, M., Eakin, E.G., Straker, L., Owen, N., Dunstan, D.W., Reid, N., & Healy, G.N. (2014). Reducing occupational sedentary time: A systematic review and meta-analysis of evidence on activity-permissive workstations. Obesity Reviews, 15(10), 822838. PubMed ID: 25040784 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nooijen, C., Kallings, L., Blom, V., Ekblom, O., Forsell, Y., & Ekblom, M. (2018). Common perceived barriers and facilitators for reducing sedentary behaviour among office workers. Journal of Physical Activity and Health, 15(10), S94S95. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Plasschaert, F., Jones, K., & Forward, M. (2009). Energy cost of walking: Solving the paradox of steady state in the presence of variable walking speed. Gait & Posture, 29(2), 311316. PubMed ID: 19027301 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Popp, C.J., Tisch, J.J., Sakarcan, K.E., Bridges, W.C., & Jesch, E.D. (2016). Approximate time to steady-state resting energy expenditure using indirect calorimetry in young, healthy adults. Frontiers in Nutrition, 3, 49. PubMed ID: 27857943 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rosenberger, M.E., Haskell, W.L., Albinali, F., Mota, S., Nawyn, J., & Intille, S. (2013). Estimating activity and sedentary behavior from an accelerometer on the hip or wrist. Medicine & Science in Sports & Exercise, 45(5), 964975. PubMed ID: 23247702 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rothney, M.P., Schaefer, E.V., Neumann, M.M., Choi, L., & Chen, K.Y. (2008). Validity of physical activity intensity predictions by ActiGraph, Actical, and RT3 accelerometers. Obesity, 16(8), 19461952. PubMed ID: 18535553 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, M.H. (2007). Protocol changes can improve the reliability of net oxygen cost data. Gait & Posture, 26(4), 494500. PubMed ID: 17855095 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sedentary Behaviour Research Network. (2012). Letter to the editor: Standardized use of the terms “sedentary” and “sedentary behaviours”. Applied Physiology, Nutrition, and Metabolism, 37(3), 540542. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Shrestha, N., Kukkonen-Harjula, K.T., Verbeek, J.H., Ijaz, S., Hermans, V., & Pedisic, Z. (2018). Workplace interventions for reducing sitting at work. Cochrane Database of Systematic Reviews, (6), CD010912. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Skotte, J., Korshoj, M., Kristiansen, J., Hanisch, C., & Holtermann, A. (2014). Detection of physical activity types using triaxial accelerometers. Journal of Physical Activity and Health, 11(1), 7684. PubMed ID: 23249722 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stamatakis, E., Ekelund, U., Ding, D., Hamer, M., Bauman, A.E., & Lee, I.M. (2019). Is the time right for quantitative public health guidelines on sitting? A narrative review of sedentary behaviour research paradigms and findings. British Journal of Sports Medicine, 53(6), 377382. PubMed ID: 29891615 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Staudenmayer, J., He, S., Hickey, A., Sasaki, J., & Freedson, P. (2015). Methods to estimate aspects of physical activity and sedentary behavior from high-frequency wrist accelerometer measurements. Journal of Applied Physiology (Bethesda, Md.: 1985), 119(4), 396403. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stemland, I., Ingebrigtsen, J., Christiansen, C.S., Jensen, B.R., Hanisch, C., Skotte, J., & Holtermann, A. (2015). Validity of the Acti4 method for detection of physical activity types in free-living settings: Comparison with video analysis. Ergonomics, 58(6), 953965. PubMed ID: 25588819 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stephenson, A., McDonough, S.M., Murphy, M.H., Nugent, C.D., & Mair, J.L. (2017). Using computer, mobile and wearable technology enhanced interventions to reduce sedentary behaviour: A systematic review and meta-analysis. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 17. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tew, G.A., Posso, M.C., Arundel, C.E., & McDaid, C.M. (2015). Systematic review: Height-adjustable workstations to reduce sedentary behaviour in office-based workers. Occupational Medicine (Oxford, England), 65(5), 357366. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tremblay, M.S., Aubert, S., Barnes, J.D., Saunders, T.J., Carson, V., Latimer-Cheung, A.E., … Participants, S.T.C.P. (2017). Sedentary Behavior Research Network (SBRN)—Terminology Consensus Project process and outcome. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 75. PubMed ID: 28599680 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van der Velde, J., Schaper, N.C., Stehouwer, C.D.A., van der Kallen, C.J.H., Sep, S.J.S., Schram, M.T., … Koster, A. (2018). Which is more important for cardiometabolic health: Sedentary time, higher intensity physical activity or cardiorespiratory fitness? The Maastricht Study. Diabetologia, 61(12), 25612569. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van Uffelen, J.G.Z., Wong, J., Chau, J.Y., van der Ploeg, H.P., Riphagen, I., Gilson, N.D., … Brown, W.J. (2010). Occupational sitting and health risks: A systematic review. American Journal of Preventive Medicine, 39(4), 379388. doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhang, S., Murray, P., Zillmer, R., Eston, R.G., Catt, M., & Rowlands, A.V. (2012). Activity classification using the GENEA: Optimum sampling frequency and number of axes. Medicine & Science in Sports & Exercise, 44(11), 22282234. PubMed ID: 22617400 doi:

    • Crossref
    • Search Google Scholar
    • Export Citation
All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 115 115 34
PDF Downloads 39 39 12