Table 1. Methods used to triangulate the validity of Fitbit-derived PA metrics.
| Method | Description |
|---|---|
| Step count | |
| Manual | Scripted tasks were video-recorded and two assessors manually counted steps according to a validated standard operating procedure. The two assessors’ counts were averaged. Manual counts were used as criterion measures during scripted tasks only. |
| Actigraph (Standard) | During post-processing, a band pass filter is applied to Actigraph’s raw accelerometer signal to remove movement artifact outside the range of human motion. Actigraph’s step count algorithm detects footfalls by identifying peaks in the accelerometer signal, and can therefore be affected by choice of filter. The standard filter was developed by the manufacturer for use in healthy populations, but has also been used in populations with MS. It is known to underestimate step count, especially in populations with walking impairments [33,34]. |
| Actigraph (LFE) | The Low Frequency Extension (LFE) is a revised band pass filter which enhances the Actigraph’s sensitivity to slow movements. It is recommended by the manufacturer in populations with impaired mobility, and is also frequently used in MS populations. The LFE has been shown to increase sensitivity to slow stepping in laboratory settings. However, it is also known to overestimate step count during free-living activity [33]. |
| Fitbit | Fitbit’s proprietary step detection algorithm derived step count from the device’s raw accelerometer signal. Fitbit provides step counts at up to minute-level granularity through its application programming interface (API). |
| Time in PA | |
| Actigraph (Vertical) | PA intensity is derived from the Actigraph by applying cutpoints to the number of activity counts identified per minute. The Actigraph (Vertical) method differentiates between sedentary behavior and PA depending on whether a minimum cutpoint of 100 vertical axis counts per minute is met [35]. This method was validated on an older model of Actigraph which is roughly equivalent to data derived from the LFE on the Actigraph GT3X. It is widely used in MS populations [36,37]. |
| Actigraph (VM) | The Actigraph Vector Magnitude (VM) method accounts for three-dimensional motion, rather than motion in the vertical axis. This method categorizes minutes as sedentary or PA with a cutpoint of 150 VM counts per minute. It is typically used with the standard filter applied. It was derived from healthy individuals [35,38,39], but is also used in persons with MS [40,41]. |
| Fitbit | Fitbit’s proprietary PA classification algorithm classifies minutes into four intensity categories: sedentary, lightly active, fairly/moderately active, and very active [42]. These categories loosely align with those used by research-grade devices: sedentary, light PA, moderate PA, vigorous PA. We defined all non-sedentary minutes as time in PA. The factors which influence PA classification are not publicly available. However, movement intensity, heart rate, and breathing rate are mentioned by moderators of Fitbit’s support forum [43]. Fitbit provides PA classifications at the minute level through its API, which may then be aggregated into hourly or daily metrics. |
| Time in MVPA | |
| Actigraph (Uniform) | The Actigraph (Uniform) method was developed to differentiate between light and MVPA in populations with MS and was validated during over-ground walking at multiple speeds. It defines minutes which exceed 1745 vertical axis counts to be MVPA, and uses the LFE filter [44]. |
| Actigraph (Severity) | The Actigraph (Severity) method was developed alongside the Actigraph (Uniform) method, but proposes different cutpoints according to MS severity: • Mild/moderate MS (EDSS < 6.0): 1980 vertical counts per minute • Severe MS (EDSS > = 6.0): 1185 vertical counts per minute Both cutpoints are designed to be used with the LFE filter [44]. |
| Actigraph (Sasaki) | The Actigraph (Sasaki) method uses a cutpoint of 2690 VM counts per minute to differentiate between light and MVPA. This method was developed in healthy controls during treadmill walking at varying speeds [38]. It has since been used in populations with MS [40,45]. |
| Fitbit | We defined all minutes characterized as ‘fairly/moderately active’ and ‘very active’ by Fitbit’s proprietary algorithm as time in MVPA [42]. This grouping aligns with the output of the Actigraph Uniform, Severity, and Sasaki methods, which do not differentiate between moderate PA and vigorous PA [38,44]. |
MS: multiple sclerosis; PA: physical activity; LFE: Low frequency extension; MVPA: moderate to vigorous physical activity; EDSS: expanded disability status score