How Much Data Is Enough? A Reliable Methodology to Examine Long-Term Wearable Data Acquisition in Gait and Postural Sway

Abstract

Wearable sensors facilitate the evaluation of gait and balance impairment in the free-living environment, often with observation periods spanning weeks, months, and even years. Data supporting the minimal duration of sensor wear, which is necessary to capture representative variability in impairment measures, are needed to balance patient burden, data quality, and study cost. Prior investigations have examined the duration required for resolving a variety of movement variables (e.g., gait speed, sit-to-stand tests), but these studies use differing methodologies and have only examined a small subset of potential measures of gait and balance impairment. Notably, postural sway measures have not yet been considered in these analyses. Here, we propose a three-level framework for examining this problem. Difference testing and intra-class correlations (ICC) are used to examine the agreement in features computed from potential wear durations (levels one and two). The association between features and established patient reported outcomes at each wear duration is also considered (level three) for determining the necessary wear duration. Utilizing wearable accelerometer data continuously collected from 22 persons with multiple sclerosis (PwMS) for 6 weeks, this framework suggests that 2 to 3 days of monitoring may be sufficient to capture most of the variability in gait and sway; however, longer periods (e.g., 3 to 6 days) may be needed to establish strong correlations to patient-reported clinical measures. Regression analysis indicates that the required wear duration depends on both the observation frequency and variability of the measure being considered. This approach provides a framework for evaluating wear duration as one aspect of the comprehensive assessment, which is necessary to ensure that wearable sensor-based methods for capturing gait and balance impairment in the free-living environment are fit for purpose.

Chest-Based Wearables and Individualized Distributions for Assessing Postural Sway in Persons with Multiple Sclerosis

“…Consistently, we observed the features Jerk and Range to provide the strongest and most significant relationships to PRMs in addition to being able to distinguish fallers from nonfallers. However, based on our previous findings [38], the strength of these relationships is poor compared to those observed on postural sway data collected during daily life (e.g., Range and ABC had a correlation of 0.71 from free living data compared to the 0.36 observed using the SSD method). This difference may be related to the well documented issue that laboratory based tests are not able to adequately capture the variability with which people, and particularly PwMS whose symptoms are known to change dramatically from day to day, move in free living environments [33], [42].…”

The ChAMP App: A Scalable mHealth Technology for Detecting Digital Phenotypes of Early Childhood Mental Health

“…However, based on our previous work, it is likely that that more granular biomechanical features may provide improved model performance 26 . Thus, we aim to explore a greater variety and complexity of features (e.g., turning behavior, postural sway 27 ) in future work. Second, including all three mood induction tasks as one behavior battery is important as features from each appear to be independent, thereby complimenting, one another ( Figure 3c ).…”

Digital Phenotypes of Instability and Fatigue Derived from Daily Standing Transitions in Persons with Multiple Sclerosis

“…To identify daily activity transitions, we employed our activity classification pipeline, which leverages a deep learning model trained on over 100,000 four-second observations of acceleration from a variety of patient populations (as previously described [23], [24]), to identify periods of sitting and standing. Each sist or stsi transition was identified using an established technique; the cranial-caudal acceleration from thigh recording was filtered and the signal was inspected for a transition from 1g towards 0 g (stsi) or from 0 g to 1g (sist) within an 18-second window of data centered on transitions between the classified activities [23].…”

“…After identifying transitions, the following features were calculated from the thigh and chest accelerations: 5 th , 50 th , and 95 th percentile of cranial-caudal (CC) and horizontal plane (F5, F50, F95), jerk of CC and horizontal plane [23], range of CC and horizontal plane, 5 th , 50 th , and 95 th percentile frequency of CC and horizontal plane, total power in CC and horizontal plane, approximate entropy (ApEn) of CC and horizontal plane, and spectral edge frequency (SEF) [29]. We also computed transition time and postural sway features of Jerk, Range, and 50 th percentile frequency of chest acceleration from the standing bout immediately preceding or following the transition [24], [30]. Data were computed from custom MATLAB scripts using Medidata's Sensor Cloud Network Analytics service.…”

“…These transitions have been used extensively to assess patient stability, dating back to the introduction of the "get-up and go" test in 1986 [26]. More modernly, wearable monitoring of such tasks has been applied to augmentation of PBMs and home-monitoring of M symptoms [22], [23], but our understanding of these metrics remains limited [24], [27].…”