The new W@W-App was developed from a previous version [20], adding a commercially available sensor (MetaWearC; MbientLab Inc) to gather postural and movement information. The MetaWearC is a small sensor (24 mm × 6 mm; 5.6 g) covered with a waterproof round case. The sensor is a triaxial accelerometer with an amplitude range of ±16 g and a sampling rate of 6.25 Hz. Key features of the MetaWearC sensor are shown on the MbientLab web page [19]. Raw sensor data are synchronized with the W@W-App software via a low-energy Bluetooth system with a long battery life (>30 days) and a range of up to 10-15 meters. The data are directly processed and displayed in real time by the app on the phone and securely stored on the backend server. Figure 1 depicts the W@W-App (login page) and the MetaWearC sensor.

The algorithm for the W@W-App+MetaWearC (Figure 2) was designed to analyze accelerometer output from the MetaWearC sensor. The MetaWearC sensor is worn within a small bag inside an elastic and adjustable band (Figure 1) that is attached to the participant’s right thigh. The algorithm is based on two primary requirements: (i) data can only be recorded during the defined recording period (ie, working hours) and (ii) data can only be collected when both the device and the software are connected via Bluetooth. When these criteria have been met and the sensor detects an acceleration, the step counter begins and the sitting and standing counters are reset to 0. Stepping time is initiated when the sensor identifies a balance between false positives (ie, counting a step when the step has not happened) and false negatives (ie, not counting a step when the step has occurred). There are three sensitivity modes for the step detector: normal, sensitive, and robust. These modes balance sensitivity (false negatives) and robustness (false positives). Normal mode is used in most applications as it provides a balance between false positives and false negatives. An example of a false positive would be the detection of a step while an individual is in a sitting position, possibly as a result of stretching one’s leg.

The recognition of postures (sitting and standing) is based on the angle of the z-axis, where 0 indicates a completely vertical posture (standing) and 1 indicates a completely horizontal posture (sitting). When the sensor detects a value higher than 0.8 in the z-axis, the sitting counter is initiated while the standing counter remains to 0. When the sensor detects a value equal to or lower than 0.8 in the z-axis, the standing counter begins while the sitting counter returns to 0. If either the sitting or standing counters reach 75 readings (approximately every 2 seconds), this indicates that the sensor has not detected stepping during those 75 readings, and therefore the step counter stops and assumes that the user is either sitting or standing depending on which of these counters reaches 75.

Finally, if there is a difference greater than 15 minutes between the time counters for the W@W-App-MetaWearC (stepping, sitting, and standing) and the elapsed time, a weighted adjustment is completed. Normally, this difference is due to temporal disconnections of the sensor if it is kept more than 20 meters away from the mobile phone. For example, if the W@W-App-MetaWearC has counted for 100 minutes (75 minutes stepping, 10 minutes standing, and 15 minutes sitting), but the real time elapsed is 115 minutes, the weighted adjustment will correct the W@W-App-MetaWearC to 86 minutes stepping, 12 minutes standing, and 17 minutes sitting.

The activPAL3M (PAL Technologies Ltd) is referred to as the gold-standard, device-based measure for postural recognition in free-living conditions [21]. The activPAL3M was employed as the criterion measure for sitting, standing, and stepping times. The activPAL3M (25 mm × 45 mm × 5mm; 9g) was placed in a waterproof nitrile sleeve and attached on the midline of the anterior aspect of the participant’s thigh using a transparent film (10 cm × 10cm of hypoallergenic Tegaderm Foam Adhesive Dressing).

Office workers from the University of Vic-Central University of Catalonia (UVic-UCC) who owned a mobile phone (Android version 6.0.0/iOS version 10.0.0 or higher) were invited to participate in the study. A convenience sample was recruited (N=23). All volunteers provided written informed consent prior to participation. This study was conducted within a Spanish national project (W@WApp-Diab; PI17/01788) led by the UVic-UCC. Ethical approval was obtained by the research ethics committee of the Research Institute of Primary Care Jordi Gol (IDIAP).

Participants installed and configured the W@W-App, following guidance provided by the researchers: (i) registration on the Walk at Work web platform [22], (ii) user verification through email, (iii) W@W-App installation and initialization, (iv) recording day and time period configuration (ie, between 3 and 8 working hours), and (v) recognition of the MetaWearC sensor via Bluetooth. In adherence to the European Union General Data Protection Regulation, participants could read the private policy of the W@W-App, which is written using clear and straightforward language, on the Walk at Work website [22]. In addition, participants provided affirmative consent prior to using the W@W-App when they voluntarily registered on the web platform.

Researchers initialized the activPAL3M and the W@W-App+MetaWearC and placed both devices on the midpoint (ie, one over the other) of the anterior aspect of the thigh of the same leg to avoid measurement bias due to asymmetric leg positions and movements. To ensure that the timestamp of the W@W-App+MetaWearC and the activPAL3M aligned for data analysis, they were initialized from the same PC.

Participants wore the W@W-App+MetaWearC and the activPAL3M sensor in occupational free-living conditions for 3-8 hours. They were required to keep their mobile phone within a 5-meter radius throughout the measurement period (ie, participants were asked to keep their mobile phones with them at all times).

The variables recorded and quantified by the W@W-App+MetaWearC were time spent in sitting and standing postures and time spent stepping. These variables were extracted from the W@W-App software. For the activPAL3M, files were processed via the activPAL Professional Software (version 7.2.32) upon completion of data collection. Data were then exported to a Microsoft Excel (Microsoft Excel 2016; Microsoft Corporation) file format, providing data on sitting, standing, and stepping in 15-second epochs. This enabled the quantification of the number of minutes spent sitting, standing, and stepping. In addition, the time spent sitting and standing were added together to quantify stationary time, while the amount of time spent standing and stepping were added together to compute upright time. Total recording time (ie, minutes) from both devices was calculated by summing the amount of time spent sitting, standing, and stepping.

Descriptive characteristics (mean [SD] and median [IQR]) were used to describe the data. Differences between the W@W-App+MetaWearC and the activPAL3M were examined using paired-samples t tests and Wilcoxon signed-rank tests. Pearson correlation coefficients were used to determine the strength and direction of association between variables quantified by the two measures when the data were normally distributed. Spearman rank-order correlation coefficients were employed when data were not normally distributed. The concordance correlation coefficient (CCC), using Lin’s approach [23], was used to examine the level of agreement between the W@W-App+MetaWearC variables and the activPAL3M determined variables. The CCC values were interpreted using the categorization recommended by McBride [24]. Bland-Altman plots with mean bias and limits of agreement (LoA) were constructed to examine the agreement between the W@W-App+MetaWearC variables and the activPAL3M determined variables using similar approaches reported previously [25]. Equivalence was determined using two one-sided paired t tests (90% CI) for the mean difference between the W@W-App+MetaWearC variables and the activPAL3M determined variables [26]. Equivalence was supported if the CI for the mean difference was within 15% of the activPAL3M-determined time spent sitting, standing, and stepping. The equivalence region was arbitrarily defined, as limited biologically and analytically relevant criteria can be defined for the equivalence regions for sitting, standing, and stepping. Less conservative equivalence regions were also tested in case the equivalence was not supported for the 10% level. Additional tests to determine the region of equivalence were completed using increments of 5%. This approach was selected to provide a clear estimation of the accuracy of the W@W-App+MetaWearC [27]. Measures were expected to differ by no more than 30 minutes for sitting, 11 minutes for standing time, and 4 minutes for stepping. All statistical analysis was conducted using IBM SPSS Statistics 25 (IBM Corporation) and Microsoft Excel (Microsoft Corporation).

This study examined the validity of the W@W-App+MetaWearC to measure occupational sitting, standing, and stepping times in a free-living workplace environment. Our findings indicated that the W@W-App+MetaWearC is a valid tool for self-monitoring occupational sitting, standing, stationary, and upright times, demonstrating moderate to very strong validity when compared to the criterion measure (activPAL3M). However, the analysis demonstrated that the findings for stepping from the W@W-App+MetaWearC are not equivalent to those from the activPAL3M.

Although a small mean bias of 1.15 minutes for stepping was observed between the W@W-App+MetaWearC and the activPAL3M, poor agreement, wide CIs and nonequivalence would suggest that the W@W-App+MetaWearC should not be recommended for use in detecting stepping time. However, it is quite plausible that these observations can be attributed not only to variance from the W@W-App+MetaWearC tool but also variance in the activPAL3M device. The activPAL3M is primarily used to examine postural position, namely sitting and standing, and has demonstrated high levels of accuracy in the detection of these behaviors in lab-based and free-living conditions [28], justifying its use as a device-based comparison for the measurement of sitting and standing times. However, lower levels of validity for the activPAL3M have been highlighted for stepping time and step count, particularly during activities of daily living. Therefore, future research should aim to utilize more accurate methods of movement when validating the W@W-App+MetaWearC tool. It should be acknowledged that W@W-App+MetaWearC performs relatively well in the detection of steps in free-living conditions when compared with findings from other commercially available activity monitors [29,30].

For stationary time (ie, sitting and standing), the W@W-App+MetaWearC demonstrated high levels of accuracy when compared with previous validation studies employing a range of activity monitors [31]. This is likely due to the ability of the W@W-App+MetaWearC to detect sitting and standing postures based on thigh acceleration. Recent studies have developed and validated self-monitoring devices that also provide real-time feedback on an integrated display, including the SitFit [32] or through a mobile phone app via Bluetooth synchronization such as the VitaBit [33] and Chair&App [34]. Similar to the findings presented here, the SitFit and Chair&App devices reported that sitting time was highly accurate when compared to the activPAL3M in free-living conditions. However, the W@W-App+MetaWearC reported a lower mean bias (W@W-App+MetaWearC) in comparison to other studies (SitFit). In contrast, the VitaBit device did not accurately distinguish between sitting and standing in free-living conditions but was accurate in the detection of movement. These findings are unsurprising, since the device used as the comparison measure (ActiGraph) struggles to accurately distinguish sitting and standing behaviors [35]. Both the SitFit and the VitaBit were designed to be worn in the pocket of a user’s trousers, which may be a usability barrier when wearing clothes without pockets. Chair&App, as well as the W@W-App, focused on office-based jobs, but Chair&App used a regular office chair equipped with pressure sensors instead of a thigh-attached device. That may remove compliance issues related to recording time but standing, stepping, and sitting away from one’s personal desk cannot be captured. The W@W-App has demonstrated high levels of validity for sitting, standing, stationary, and upright times, while the wearer’s location and attachment site may increase compliance with wearing a self-monitoring tool in the workplace.

The W@W-App+MetaWearC is a novel tool that simulates the activPAL3M activity monitor in accurately recognizing postural position at the workplace. The output from the W@W-App+MetaWearC tool for sitting, standing, stationary, and upright times were identified as equivalent to the current gold-standard, device-based postural measure, the activPAL3M. This suggests that this self-monitoring tool, which provides real-time feedback to users, is worthwhile for use in interventions that aim to reduce sitting behavior in the workplace. Self-monitoring is vital for increasing individuals’ awareness and empowerment toward behavior change [7]. This may result in a more accurate, affordable, and accessible device than those currently available, enabling the more cost-effective inclusion of SB self-monitoring as a function of SB interventions in the future.

The strengths of this study include (i) the examination of the complete range of occupational sedentary and activity behavior types (sitting/lying, standing, and stepping), (ii) the examination of the validity of these measures in occupational free-living conditions, and (iii) the use of a gold-standard, objective measurement device to determine the validity of the W@W-App+MetaWearC.

The present study is not without limitations. Although the activPAL3M has been described as the gold standard for measuring sitting time [21] and is an acceptable field-based measure for activity behaviors in youth and adult populations [36,37], it is not the gold standard for comparison of stepping time. This should then be considered when interpreting the Bland-Altman plots, as these are designed to support the comparison of a new measure to a previous gold standard. The relatively small sample size with a large percentage of females (16/20) and the homogeneity of the workplace setting (ie, all sampled from a university context) might differ from the general office population. Furthermore, the data gathered included an average of 5 hours per subject, providing a limited timeframe. Additionally, the wide range of operating systems and hardware available added complexity to app development and subsequent validation.

The W@W-App+MetaWearC self-monitoring system demonstrates high levels of accuracy in determining postural position. This tool is a low-cost alternative tool for the examination of occupational sitting and standing times. It has demonstrated high levels of validity in detecting postural position and provides real-time feedback to users. Future research should examine the interventional effect of utilizing this system as a self-monitoring tool for modifying activity behaviors in office-based workers.