Monitoring mobility in older adults using a global positioning system (GPS) smartwatch and accelerometer: A validation study

Beauchamp, Marla1 ; Kirkwood, Renata1 ; Cooper, Cody1 ; Brown, Matthew1 ; Newbold, Bruce1 ; Scott, Darren1

Published Jan 11, 2024; Updated Jan 15, 2024 on Dryad. https://doi.org/10.5061/dryad.bvq83bkg2

Abstract

Background

There is interest in identifying the most reliable method for detecting early mobility limitations. Accelerometry and Global Positioning System (GPS) could provide insight into declines in mobility, but few studies have used this multi-sensor approach to monitor mobility in older adults.

Methods

Thirty-two volunteers (66.2±6.3 years) agreed to participate in our validation study. We conducted two experiments to determine the validity of the TicWatch S2 and Pro 3 Ultra GPS models against the Qstarz receiver in measuring life-space mobility, trip frequency, duration, and mode. We also assessed the accuracy of the TicWatch in measuring step count and agreement with the ActiGraph wGT3X-BT for activity counts and sedentary behavior. Participants wore devices simultaneously for three consecutive days and recorded activity and trip information.

Results

The TicWatch Pro 3 Ultra GPS performed better than the S2 model and was similar to the Qstarz in all tested trip-related measures, and it was able to estimate both passive and active trip modes. Both models showed similar results to the Qstarz in life-space-related measures. The TicWatch S2 demonstrated good to excellent overall agreement with the ActiGraph algorithms for the time spent in sedentary and non-sedentary activities, with 84% and 87% agreement rates, respectively. Under supervised conditions, the TicWatch Pro 3 Ultra GPS measured step count consistently with the gold standard observer, with a bias of 0.4 steps. The thigh-worn ActiGraph algorithm accurately classified sitting and lying postures (97%) and standing postures (90%).

Conclusion

Our multi-sensor approach to monitoring mobility has the potential to capture both accelerometer-derived movement data and trip/life-space data only available through GPS. In this study, we found that the TicWatch models are valid devices for capturing GPS and raw accelerometer data, making them useful tools for assessing real-world mobility in older adults and advancing our knowledge of early mobility decline.

README: MONITORING MOBILITY IN OLDER ADULTS USING A GLOBAL POSITIONING SYSTEM (GPS) SMARTWATCH AND ACCELEROMETER: A VALIDATION STUDY

Brief description of dataset contents, experimental procedures and results.

Twenty-five volunteers participated in our validation study. We conducted two experiments to validate the TicWatch S2 and Pro 3 Ultra GPS for collecting accelerometer data against the ActiGraph wBT3X and life space measure and trip frequency, duration and mode against the Qstarz BT-Q1000X GPS Data Logger. We also assessed the accuracy of the TicWatch in measuring step count and agreement with the ActiGraph wGT3X-BT for activity counts and sedentary behavior. Participants wore devices simultaneously for three consecutive days and recorded activity and trip information.
In Experiment 1, participants were provided with three devices, the TicWatch S2, Qstarz GPS Data Logger and the ActiGraph. They were instructed to wear the TicWatch S2 and the ActiGraph on the non-dominant wrist simultaneously and to carry the GPS data logger with them whenever they travelled outside their homes. We asked participants to record meaningful bouts of activities in terms of their body posture (e.g., sitting, standing or lying) or whether they were exercising as they were able to throughout the day. Participants were also asked to write the address or the intersection of the places they visited. In Experiment 2, participants were provided with three devices, the TicWatch Pro 3 Ultra GPS, Qstarz GPS Data Logger and the ActiGraph. They were instructed to wear the TicWatch Pro 3 Ultra GPS on the non-dominant wrist, to carry the GPS whenever they travelled outside the home and to record the times they took the watch on and off. They were also instructed to attach the ActiGraph to the anterior aspect of the left or right thigh just above the kneecap using the adhesive patches provided to perform the body posture tasks described below. In both experiments, participants were asked to wear the devices for three consecutive days during waking hours and remove the devices for showering and water activity purposes.

Our findings indicate that the TicWatch Pro 3 Ultra GPS outperformed the S2 model and was similar to the Qstarz in all tested trip-related measures, and it was able to estimate both passive and active trip modes. Both models showed similar results to the Qstarz in life-space-related measures. The TicWatch S2 demonstrated good to excellent overall agreement with the ActiGraph algorithms for the time spent in sedentary and non-sedentary activities. Under supervised conditions, the TicWatch Pro 3 Ultra GPS measured step count consistently with the gold standard observer and the thigh-worn ActiGraph algorithm accurately classified sitting and lying postures and standing postures.

Description of the data and file structure

Demographics_Experiment 1 and Demographics_Experiment 2.xlsx: These files provide demographic data corresponding to Table 1 in the manuscript. They encompass demographic details for all 25 participants from Experiment 1 and 10 participants from Experiment 2. The data includes information such as sex, age in years, history of falls in the previous year, number of falls in the previous year, total number of comorbidities, level of education, and mobility level for each participant. Each row in the files corresponds to an individual participant.
Data on time spent sedentary and non-sedentary and CPM_Experiment 1.xlsx: This file supports the findings presented in Table 3 and Figure 2 of the manuscript. In Experiment 1, participants were directed to record meaningful activity bouts, specifying body posture (e.g., sitting, standing, or lying) and indicating exercise engagement whenever feasible throughout the day. Subsequently, these activities were categorized into sedentary and non-sedentary groups using cutoff points from Montoye et al., 2020 (reference 29 in the manuscript). The file incorporates data extracted from the ActiGraph and TicWatch S2, featuring the total duration of activities in minutes, the duration of sedentary and non-sedentary activities, as well as the total Counts per minute. (Mins = minutes and avg=average). The rows depict the activities recorded by each participant during the study period, with the participant numbers listed in the first column.
The files below contain selected activity counts per minute acquired from both the ActiGraph and TicWatch S2 during Experiment 1, serving as the data foundation for constructing Figures 3A to 3F in the manuscript. Figures 3A to 3F feature Bland-Altman plots, comparing the average counts per minute per epoch length of 60 seconds for select activities performed by six participants. These activities, chosen randomly, aim to demonstrate the agreement between the TicWatch S2 and the ActiGraph in providing counts per minute. Participants recorded these activities using the diary provided in Experiment 1. Each file is a recording of one participant's activity. Column A is the date the activity occurred, column B the time, and columns C and D the counts per minute determined by the Ticwatch S2 and Actigraph, respectively.
Activity_Walking the dog_Experiment 1.xlsx
Activity_Exercising dancing class_Experiment 1. xlsx
Activity_Lying watching TV_Experiment 1. xlsx
Activity_Sitting using a computer_Experiment 1. xlsx
Activity_Playing Pickleball_Experiment 1. xlsx
Activity_Standing Cleaning the house_Experiment 1. xlsx
Step count data_ Experiment 1 and 2.xlsx: This file supports Figure 4 and step count data results. The file has two tabs. Tab 1 displays the data from Experiment 1, showcasing both the number of steps recorded by the participant and the number of steps extracted from the TicWatch S2. In Experiment 1, we utilized the step counter that monitors the total number of steps taken over time. Tab 2 displays the data from Experiment 2, and shows the number of steps recorded by the participant and extracted from the TicWatch Pro 3 Ultra GPS. In Experiment 2, we employed the step detector, which identifies when a step is taken and generates an event each time it detects a step. Each row in the file corresponds to an individual participant.
Body Posture Analysis_Experiment 2. xlsx: This file supports the outcomes of body posture classification in Experiment 2, derived from the thigh-worn algorithm implemented on the ActiGraph. This algorithm relies on both movement patterns and thigh angle data to effectively distinguish between lying and sitting versus standing positions. Participants were instructed to undertake two tasks: sitting on the edge of a bed (or sofa) and standing still, with or without support, each for a duration of 5 minutes. The first row in the table indicates the specific activity performed by the participant, namely sitting or standing. The shaded columns display the duration of time during which the ActiGraph accurately identified the corresponding body posture. Starting on row 4, each row corresponds to an individual participant.
ComparingCHulls_Experiment 1 and ComparingCHulls_Experiment 2. xlsx: These files support the data presented in Table 2 and contain data related to the convex hull, defined as the smallest polygon where no internal angle exceeds 180 degrees and encompasses all collected points. The area and perimeter, reported in square meters and meters respectively, were extracted from the gold standard Qstarz, as well as from the TicWatch S2 and Pro 3 Ultra GPS. Each row in the file corresponds to an individual participant.
ComparingMaximumDistanceFromHome_Experiment 1 and ComparingMaximumDistanceFromHome_Experiment 2. xlsx: These files support the data presented in Table 2 and contain the data related to the maximum distance from home, defined as the farthest planar/straight-line distance (in meters) from the participants’ home location that was travelled to, extracted from the gold standard Qstarz, as well as from the TicWatch S2 and Pro 3 Ultra GPS. Each row in the file corresponds to an individual participant.
Trip Frequency and duration Experiment 1 and Trip Frequency and duration_Experiment 2. xlsx: These files support the data showcased in Table 2, encompassing life space mobility data, which includes trip frequency, duration in minutes, and mode of transportation (active or passive). In Experiment 1, the table presents data on trip frequency and duration in minutes directly sourced from participant diaries and the TicWatch S2. In Experiment 2, Tab 1 provides data on the trip frequency and duration in minutes directly sourced from the TicWatch Pro 3 Ultra GPS and the gold standard Qstarz. Tab 2 reports the mode of transportation (passive and active), obtained from both the gold standard Qstarz and the TicWatch Pro 3 Ultra GPS. Each row in the file corresponds to an individual participant.
OnBodyAnalysis_Experiment 2.xlsx: This file contains the wear time in hours reported by the participant and captured by the TicWatch Pro 3 Ultra GPS during Experiment 2. We present data from all 3 days of data collection as well as the average of all days. Each row in the file corresponds to an individual participant
PowerDischarge_Experiment 2.xlsx: This file contains data supporting the battery life modes, specifically "stay connected fix collection every 5 seconds" and "periodic fix collection every 10 seconds" for the TicWatch Pro 3 Ultra GPS. The table presents the duration in hours during which the device actively collected data in various modes, including normal mode, power saver mode, and extreme power saver mode. Each row in the file corresponds to an individual participant.

Sharing/Access information

Data was derived from the following sources:

ActiGraph accelerometer data: raw data was extracted using the ActiLife software from ActiGraph.
Counts per minute: we determined the activity counts per minute (CPM) using a Python script (BrØnd et al., 2017) that generates the ActiGraph physical activity counts (BrØnd JC, Andersen LB, Arvidsson D. Generating ActiGraph Counts from Raw Acceleration Recorded by an Alternative Monitor. Med Sci Sports Exerc. 2017 Nov 1;49(11):2351–60).
TicWatch accelerometer and GPS data: a custom data-collection application was developed by our software engineer to optimize data quality and battery life. Ivy is an app designed to run on any wearable device running Google Wear OS and to allow for continuous, non-intrusive, long-term measurements of key mobility-related parameters. It collects and stores data from the following onboard sensors: accelerometer, gyroscope, location (GPS), and heart rate. The app additionally collects information on step counts, current activity type predictions (using the Google Activity Recognition API), on-body detection, and device power/battery states. Ivy also provides a modular list of settings which control behavior of the TicWatch’s sensors, namely the frequency of accelerometry and GPS receiver to address battery constraints. Our second app, Clover, has been designed to support Ivy as a Windows companion app with a user-friendly graphical interface that automates the process of setting up Ivy on a smartwatch, downloading collected data, and preparing devices for distribution. Clover has the capability to conduct initial checks on data quality, including an assessment of the number of days the watch was worn.
Measures related to the life-space area, such as maximum distance from home, minimum convex hull (MCH), and standard deviational ellipse (SDE), were calculated using ArcGIS® Pro, a desktop Geographic Information System application developed by Esri®.

Methods

We conducted two experiments to validate the TicWatch for collecting movement and navigation data: 1) we compared the TicWatch S2 against the Qstarz BT-Q1000X GPS Data Logger to measure life-space mobility, trip frequency, and duration. We also assessed the agreement of the TicWatch S2 in measuring step count against an observer, activity counts per minute (CPM), and sedentary and non-sedentary activity using ActiGraph's proprietary algorithms; 2) we compared the TicWatch Pro 3 Ultra GPS against a stand-alone Qstarz BT-Q1000X GPS Data Logger to measure life-space mobility, trip frequency and duration, and mode of transportation. We evaluated the level of agreement between the TicWatch Pro 3 Ultra GPS in measuring steps compared to direct measures of step count reported by the participants. Additionally, we tested two different GPS configurations of the TicWatch Pro 3 Ultra GPS in a free-living study setting to observe battery life performance: a) periodic fix collection every 10 seconds (i.e., the GPS receiver turned off for 10 seconds before searching for a new location point), and b) stay connected fix collection every 5 seconds (i.e., the GPS receiver never turns off, allowing for more continuous data collection). Finally, for assessing body posture, we evaluated the agreement of the thigh-worn algorithms of the ActiGraph wGT3X-BT with an observer in identifying lying and sitting from standing. The study results not only inform the use of the TicWatch for assessing and monitoring early changes in mobility in the MacM3 cohort study but also offer valuable information on the validity of wearable devices for researchers considering collecting movement and navigation data in their research protocols.

Protocol

In Experiment 1, participants were provided with three devices, the TicWatch S2, Qstarz GPS Data Logger, and the ActiGraph. They were instructed to wear the TicWatch S2 and the ActiGraph on the non-dominant wrist simultaneously and to carry the GPS data logger with them whenever they travelled outside their homes. They were instructed to charge the TicWatch S2 and Qstarz every night using the chargers provided.

In Experiment 2, participants were provided with three devices, the TicWatch Pro 3 Ultra GPS, Qstarz GPS Data Logger, and the ActiGraph. They were instructed to wear the TicWatch Pro 3 Ultra GPS on the non-dominant wrist, to carry the GPS whenever they travelled outside the home, and to record the time they put the watch on and off. They were also instructed to attach the ActiGraph to the anterior aspect of the left or right thigh just above the kneecap using the adhesive patches provided to perform the body posture tasks.

Data Reduction

The GPS data collected by the Qstarz and TicWatch models were first cleaned by excluding any points with speeds above 160 km/h, as the fastest roadways in our study area have a maximum speed limit of 110 km/h. We then processed each participant’s data to ensure the time periods compared between devices were identical, such that discrepancies due to battery life or participant error were excluded from the analysis. Measures related to the life-space area, such as maximum distance from home, minimum convex hull (MCH), and standard deviational ellipse (SDE), were calculated using ArcGIS® Pro, a desktop Geographic Information System application developed by Esri®. Each participant’s trip frequency and trip duration were determined using GPS data collected by both Qstarz and TicWatch. To accomplish this, we adapted the stop and trip detection algorithm of Montoliu et al.¹ We chose to use algorithm settings proposed by Fillekes et al² for trip detection and used them to derive trip frequency and duration independently for each device. Using the Qstarz data, we manually verified the algorithm results and adjusted the values for accuracy. We then compared the algorithm-derived measures from the TicWatch against these results. Additionally, we used the method proposed for segmenting GPS segments into active (non-motorized) and passive (motorized) trips, which is adapted from the work of Carlson et al.³ and Vanwolleghem et al.⁴ Specifically, trips with 90th percentile speed ≥ 25 km/h were classified as passive, whereas trips below that threshold were classified as active.

Accelerometer data were collected at different frequencies for the ActiGraph and TicWatch devices. TicWatch data was adjusted to match ActiGraph's frequency for comparison.

In Experiment 1, the accelerometer data from the ActiGraph and TicWatch S2 were screened for the period of wear times using the method described by Choi et al.⁵ We first determined the activity counts per minute (CPM) using a Python script that generates the ActiGraph physical activity counts.⁶ We applied the script on both the S2 and ActiGraph devices. Based on the activity counts and using an epoch length of 60 seconds, non-wear time was defined as 90 consecutive minutes of zero counts, with an allowance of 2 minutes of nonzero counts, provided there were 30-minute consecutive zero counts before and after that allowance⁵. Wear times in Experiment 2 were obtained using the on-body sensor of the TicWatch Pro 3 Ultra GPS.

To evaluate PA intensity, we computed the vector magnitude (VM) by taking the square root of the summed squared counts per minute for each axis on both devices. The VM counts were then calculated per 60-second epoch, and we applied the cut-off scores developed by Montoye et al.⁷ specifically for wrist-worn devices. We classified activities as "sedentary" if the VM counts were below 2,860 and collapsed light and moderate/vigorous activity categories into "non-sedentary" which included VM counts of 2,860 or higher.⁷ Following this, we determined the time, in minutes, spent on sedentary and non-sedentary behaviour. We also calculated the mean activity counts per epoch length of 60-second for various activities, including exercising, sitting, lying, and walking, as reported by the participants, for both ActiGraph and TicWatch S2. To ensure an accurate comparison of accelerometer data, we restricted our analysis to the periods when participants reported wearing both the TicWatch and ActiGraph devices simultaneously.

Step count was obtained directly using the step count and step detector sensors from the TicWatch models. In Experiment 1, we selected the step counter that keeps track of the total number of steps taken over time. In Experiment 2, we used the step detector that detects when a step is taken and generates an event each time it detects a step, but it does not keep track of the total number of steps taken. Body posture classification in Experiment 2 was obtained using the thigh-worn algorithm from the ActiGraph that relies on movement and the thigh angle to accurately classify lying and sitting vs. standing positions.⁸

REFERENCES

Montoliu, R., Blom, J. & Gatica-Perez, D. Discovering places of interest in everyday life from smartphone data. in Multimedia Tools and Applications vol. 62 179–207 (2013).
Fillekes, M. P., Giannouli, E., Kim, E. K., Zijlstra, W. & Weibel, R. Towards a comprehensive set of GPS-based indicators reflecting the multidimensional nature of daily mobility for applications in health and aging research. Int J Health Geogr 18, 17 (2019).
Carlson, J. A. et al. Association between neighbourhood walkability and GPS-measured walking, bicycling and vehicle time in adolescents. Health Place 32, 1 (2015).
Vanwolleghem, G. et al. Children’s GPS-determined versus self-reported transport in leisure time and associations with parental perceptions of the neighborhood environment. Int J Health Geogr 15, (2016).
Choi, L., Liu, Z., Matthews, C. E. & Buchowski, M. S. Validation of accelerometer wear and nonwear time classification algorithm. Med Sci Sports Exerc 43, 357–364 (2011).
BrØnd, J. C., Andersen, L. B. & Arvidsson, D. Generating ActiGraph Counts from Raw Acceleration Recorded by an Alternative Monitor. Med Sci Sports Exerc 49, 2351–2360 (2017).
Montoye, A. H. K. et al. Development of cut-points for determining activity intensity from a wrist-worn ActiGraph accelerometer in free-living adults. J Sports Sci 38, 1–10 (2020).
ActiGraph. How is inclination determined (for thigh wear location)? 1–1 https://actigraphcorp.my.site.com/support/s/article/How-is-Inclination-Determined-for-Thigh-Wear-Location (2019).