Reliability of wearable sensors for assessing gait and chair stand function at home in people with knee osteoarthritis

Abstract

Objective.

To assess the reliability of wearable sensors for at-home assessment of walking and chair stand activities in people with knee osteoarthritis (OA).

Methods.

Baseline data from participants with knee OA (n=20) enrolled in a clinical trial of an exercise intervention were used. Participants completed an in-person lab visit and a video conference enabled at-home visit. In both visits, participants performed walking and chair stand tasks while instrumented with three inertial sensors. During the at-home visit, participants self-donned the sensors and completed two sets of acquisitions interspersed by a 15-minute break when they removed and re-donned the sensors. Participants completed a survey on their experience with the at-home visit. During the in-lab visit, researchers placed the sensors on the participants. Spatiotemporal metrics of walking gait and chair stand duration were extracted from the sensor data. We used intra-class correlation coefficients (ICC) and the Bland-Altman plot for statistical analyses.

Results.

For test-retest reliability during the at-home visit, all ICCs were good to excellent (between 0.85 and 0.95). For agreement between at-home and in-lab visits, ICCs were moderate to good (between 0.59 and 0.87). Systematic differences were noted between at-home and lab data due to faster task speed during the lab visits. Participants reported favorable experience during at-home visit.

Conclusion.

Our method of estimating spatiotemporal gait measures and chair stand duration function remotely was reliable, feasible, and acceptable in people with knee OA. Wearable sensors could be used to remotely assess walking and chair stand in participant’s natural environments in future studies.

INTRODUCTION

Osteoarthritis (OA) is a leading cause of chronic pain and disability among adults (1). For people with knee OA, assessment of movement patterns during daily activities like walking and chair stand are considered clinically important functional outcomes (2). Alterations in movement patterns during these activities in people with knee OA are related to worse functional outcomes and disease progression (3–5). For example, individuals with knee OA walk with greater stride duration and lower cadence compared to controls (6), and while getting up from a chair, people with knee OA take longer compared to controls (7). Hence, standardized tests of gait and chair stand function are recommended as core outcomes for clinical trials of interventions for people with knee OA (8, 9). However, assessments of gait and chair stand patterns are usually performed in tightly-controlled lab environments that require expensive and time-consuming motion capture technologies. These visits can not only be burdensome for participants (e.g., due to travel to site) and researchers alike, but also may not yield data that reflect movement patterns utilized in a person’s natural environments (10).

The COVID-19 pandemic has accelerated the adoption and implementation of telehealth (the practice of remote, virtual healthcare) (11) as well as the use of digital health technologies for remote assessment of participants in clinical trials (12). Wearable inertial sensors offer the possibility of remotely assessing gait and chair stand movements for people with knee OA using standardized tests in a person’s natural environment (13–18).However, it is important to determine whether such measures are reliable in individuals’ home environments (i.e., if they report consistent measurements from repeated tests) and how they agree with measures collected in well-controlled lab environments before large-scale use in clinical trials. While prior studies report excellent reliability in controlled lab-environments (19), only a few studies have been conducted in individual’s homes and none in people with knee OA (20–22).

Hence, our objective was to examine the reliability of wearable sensor metrics of walking gait and chair stand using standardized tests in participants’ homes and agreement between these metrics collected in laboratory and at-home. We hypothesized wearable-sensor derived walking gait and chair stand measures collected in a person’s home environment would show good to excellent reliability in repeated measures, and at-home data will agree with measures collected in a lab environment.

METHODS

Participants:

We used data from a subset of participants (n=20) enrolled in our single-arm clinical trial of an exercise intervention in people with knee OA (clinicaltrials.gov NCT04243096). Participants were recruited from the community using print and online advertising and targeted social media strategies. Key inclusion criteria were age ≥ 50 years, BMI ≤ 40 kg/m², physician diagnosed knee OA, score ≥ 3/12 on weight-bearing questions (walking on flat surface, going up and down stairs, standing upright) from the Knee Injury and Osteoarthritis Outcome Score (KOOS) Pain Subscale in the index knee (23), and ability to walk for 20 minutes without assistance. Key exclusion criteria included contraindications to exercise, other pain in lower back or legs that is greater than knee pain, any knee surgery in the previous 6 months, joint replacement in either hip or ankle, previous knee osteotomy, partial or total knee replacement in either knee, corticosteroid or hyaluronic acid injections in either knee in the previous 3 months, other health conditions that may affect motor function, and received physical therapy for knee OA within past 6 months. All study procedures were approved by a Boston University Institutional Review Board. Participants signed an informed consent prior to any study procedures.

For each participant, a “study leg” was identified as the leg with the diagnosis of knee OA provided by the physician, or the more painful leg in case both knees were diagnosed with OA (9). In cases where individuals had knee OA diagnosed in both knees and equal pain scores, the study leg was chosen at random.

Data Collection:

All assessments took place prior to initiation of the study intervention. All participants completed the KOOS and provided information on their education, employment status, and family income. Participants also self-identified their sex assigned at birth and race from a fixed set of categories that included options to not provide this information. For race, participants could also select “unknown”, or select multiple options. Participants completed two study visits, an in-person lab visit and a remote at-home visit. The order of these visits was randomized across participants, half completing the in-lab visit first and half completing the at-home visit first, with participants completing both visits between 1 and 20 days of each other. Participants were asked to wear their same daily walking shoes during both visits, and we used the same equipment and instructions for each task across both visits. During the visits, timestamps for start and end of each trial of each task was recorded by a researcher using a custom system designed in REDCap electronic data capture tool hosted at Boston University (24). Participants who were randomized to complete their at-home visit first were required to attend an in-person session to complete the informed consent procedures and then took home the equipment for the remote at-home visit. Participants who were randomized to complete the in-person lab visit first took home the equipment following the first in-person visit.

Regardless of the sequence, all participants received the same equipment for the at-home visit. Specifically, we provided participants with a wearable system consisting of three inertial sensors and docking station with charging cable, two cones connected by a 7-meter rope, and an armless chair. We used Opal inertial sensors (APDM, Portland, USA). Each sensor contains an accelerometer (±16 g), gyroscope (±2,000 deg/s), and magnetometer (± 8 Gauss) and measures 43.7 x 39.7 x 13.7 mm (LxWxH) and weighs approximately 25 grams (Figure 1). Sensors were initialized to collect data in logging mode (data stored onboard the sensor) at 128 Hz. The initialization process also synchronized the internal sensor clock with the computer clock in our lab. We placed the sensors in “standby” mode to not deplete their battery. Sensors were provided to the participants in a briefcase provided by the manufacturer with sufficient padding for protection. Participants were also provided verbal and written instructions on the use of sensors. Additionally, participants were also provided with a tablet computer (Galaxy Tab S5e, Samsung, Seoul, South Korea). A video of the walking and chair stand tasks along with instructions was saved on the tablet computer for participants to review. We have included this video in the supplemental information.

Figure 1:

Opal inertial sensor (left) and example of three sensor system as worn by participants. Two sensors were placed with straps around the shoes (center) and one sensor was placed on the lumbar and buckled around the waist (right).

During the at-home study visit, researchers guided the participants via video conference. Participants were asked to place the sensors into the docking station that was connected to a power supply. This process activated the sensors so that they were no longer in “standby” mode and started logging data, time-synchronized to each other. Then we guided the participants to place three sensors on their body. One sensor was placed on the dorsum of each foot, and one was placed on the lower back as per manufacturer’s guidelines (Figure 1). The sensors were secured to each segment using straps attached to the devices. After donning the sensors, we guided the participants through two trials each of (a) a standardized walking task, and (b) a chair stand task. For the standardized walking task, we asked the participants to walk at their self-selected, comfortable pace for two laps of a 7-meter path defined by the previously provided cones and rope for a total walking distance of 28 meters. We selected 7-meters for each direction because the manufacturer of the inertial sensors recommended a minimum of 7-meters for extraction of gait metrics. All but two participants had enough room for the 7-meter walking course; for those two individuals they walked as far as they could in a straight line before turning around. Before beginning the walking test, participants were instructed to stand still for 30 seconds for sensor initialization, as stated in the device manual provided by the manufacturer. For the standardized chair stand task, we asked the participants to stand up from the provided chair five times as quickly as possible with arms crossed across the chest, similar to a traditional five time sit-to-stand test (25). After the first set of tests, participants removed the sensors, waited 15-minutes, re-donned the sensors, and performed two more trials of each task. At the end of the at-home visit, participants completed a REDCap survey evaluating the ease of using the devices, their comfort with performing the tests, and the likelihood they would participate in a similar visit in the future. Participants returned the equipment to us either at their first intervention visit or their in-person lab baseline data collection, depending on the randomization of their at-home visit. During the in-person lab visit, participants performed two trials of the same tasks as in the at-home visit, this time with a researcher placing the sensors on the participants.

Data Processing:

On return of the sensors, we downloaded the data from the devices. We extracted spatiotemporal walking gait metrics for both legs using the manufacturer provided software (MoveoExplorer v1.0.0.201904110002) (26, 27). The algorithm excluded turns using a validated method that uses the angular velocity signal around the vertical axis from the lower back sensor (28, 29). For each gait trial, metrics were averaged across all strides. Step and stride duration are reported in seconds; stance duration, swing duration, double support duration, and terminal double support duration are reported in % of gait cycle time; gait speed is in meters per second; cadence is in steps per minute and stride length is in meters. For the chair stand test, duration of chair stand (in seconds) was extracted from the sensors as an average of all chair stands detected. All data underwent visual inspection for errors or inconsistencies. A significant time shift was noticed in some participants’ data (16 at-home and 9 in-lab recordings), due to a drift in the clocks of all sensors which were time-synchronized with each other (Supplemental Figure 1). The maximum drift was 24 seconds. To correct for this, a researcher visually inspected the raw sensor data for a walking trial from each participant and manually recorded the time difference between when a participant began walking during the gait test and when we expected them to begin walking (30 seconds after the initial timestamp was recorded due to 30-second still period for sensor initialization). The time offset correction was then applied to all timestamps for all sensors of that participant’s visit. All at-home visits were conducted 1-9 days from the initialization and provision of sensors to participants, with no association observed between amount of time between initialization and data collection with sensor clock drift. Three sets of walking gait and chair stand measures were generated, i.e., mean of the first two trials from the at-home visit, mean of the second two trials from the at-home visit, and mean of the two trials from the in-lab visit. While gait data were extracted for both legs, since our analyses are within-person, we are only reporting gait data from the study leg as defined above. Gait data for the contralateral leg as well as data reported as right and left legs can be found in the Supplemental Material (Supplemental Table 1–3).

Sample size estimation:

The sample size was estimated using a prior study in healthy adults (n=32, 65-85 years old) (30). In the cited study, the test-retest reliability based upon ICC for gait speed between two lab visits derived from an inertial sensor placed on the lower back was 0.85 [95% Confidence Interval: 0.71, 0.92](31). We computed an ICC H₀ of 0.42 by generating a null distribution based on 10,000 permutations of the gait speed across participants between visit 1 and visit 2 and taking the 99th percentile of this distribution as ICC H₀. Using this ICC H₀ and a one-tailed test with alpha=0.05 and power=0.8, we would need at least 11 participants to detect an ICC Of 0.85. We estimated that 12 participants would be needed assuming a 10% attrition. We overenrolled to ensure we had sufficient power for all measures of interest and anticipating that not all participants may be able to complete all tasks. We used an R package ICC.Sample.Size to perform the sample size calculations (32).

Statistical Analysis:

We calculated test-retest reliability ICCs (95% confidence intervals) for each measure of function between repeated measures obtained in the at-home visit. We also computed agreement between the in-lab visit and the first set of measures from at-home visit using ICCs. For all of these, we used ICC (2,1) based on absolute agreement in a two-way random effects model (33). We interpreted ICCs as poor (<0.5), moderate (between 0.5-0.75), good (0.75-0.90), and excellent (>0.90) (33). We also calculated Pearson’s correlation between the first and second set of measures from the remote at-home visit. We report the standard error of measurement (SEM) and minimum detectable change (MDC) (Eq 1, 2) for assessment of walking gait and chair stand tasks at home as calculated similarly in previous studies (26, 34–36). We also report SEM%, and MDC% (Eq 3, 4) calculated as shown below:

$$SEM=SD\times \sqrt{1-ICC}$$ Eq (1)

$$MDC=SEM\times 1.96\times \surd 2$$ Eq (2)

$$SEM\%=100\times \left(SEM/\overline{x}\right)$$ Eq (3)

$$MDC\%=100\times \left(MDC/\overline{x}\right)$$ Eq (4)

where $\stackrel{-}{x}$ is the mean for all observations (37). These values are independent of the units of measurement and allow comparison of inherent error between measures.

For agreement between the lab and at-home visits, we additionally performed Bland-Altman analyses to observe any possible biases, and paired t-tests to determine if differences were significant. ICCs and Bland-Altman plots were created using R statistical software (version 4.1.1) using the packages psych and BlandAltmanLeh respectively, and Pearson’s correlation were calculated in Matlab (MathWorks, Inc., Natick, MA).

RESULTS

We enrolled 20 participants from the parent study for this substudy. Data from 2 of the 20 participants were unusable due to challenges with participants following our instructions on how to use the sensor, resulting in battery depletion and lack of data collection in some sensors. Participant characteristics are shown in Table 1.

Table 1:

Participant characteristics. Data are presented as mean (standard deviation) where applicable.

Characteristic	Participants (n = 20)
Age, years	70.5 (4.7)
BMI, kg/m2	30.6 (4.7)
KOOS Pain, 0-100	60.2 (10.6)
KOOS ADL, 0-100	68.8 (14.2)
KOOS weight bearing pain, more painful knee, 0-12	5.15 (1.35)
KOOS weight bearing pain, less painful knee, 0-12	3.15 (2.06)
Study Leg = Left leg, n (%)	9 (45%)
Unilateral knee OA, n (%)	10 (50%)
Had previous knee injury, n (%)	13 (65%)
Sex assigned at birth, n (%)
Female,	17 (85%)
Male	3 (15%)
Race, n (%) *
White	19 (95%)
Education, n (%)
Without a college degree	2 (10%)
Undergraduate	2 (10%)
Graduate	13 (65%)
Doctorate	3 (15%)
Annual Income, n (%)
<$50,000	4 (20%)
$50,000-$150,000	6 (30%)
>$150,000	4 (20%)
Did not report	6 (30%)
Currently Employed, n (%)	11 (55%)

KOOS = Knee injury and Osteoarthritis Outcome Score

^*For privacy reasons, the race of one participant who did not self-identify as White is not reported.

Test re-test reliability during at-home visit:

Figure 2 shows scatter plots for selected gait measures and the chair stand duration from the first and second set of trials in the at-home visit. During the at-home visit, the test-retest reliability of walking speed (ICC = 0.85) and chair stand duration (ICC = 0.89) were good, and reliabilities of all other gait measures were excellent (ICC > 0.9) when reported for the study leg (Table 2). Similar reliability was noted for the contralateral leg, left leg, and right leg (Supplemental Tables 1–3). Pearson’s correlation ranged from 0.81 to 0.97 for these tests (Table 2). SEM, SEM%, MDC, and MDC% for the wearable sensor derived gait and chair stand metrics from the at-home visit are shown in Table 3. Participant feedback indicated that, in general, they were highly accepting of the at-home visit (Supplemental Table 4).

Figure 2:

Scatter plots for selected gait metrics from the left leg (A-C) and chair stand duration (D) derived from the first and second set of trials during the remote visit. Blue circles represent collected data, the solid orange line shows a linear fit between the data, the dotted pink lines represent the 95% prediction interval, and the dotted black line represents the line of unity between the two sets of at-home tests.

Table 2:

Test-retest reliability data. Gait data reported for study leg. Mean (standard deviation) are reported for all metrics along with ICC (lower bound, upper bound) values.

Sensor variables	Home 1*	Home 2*	Home 1 vs Home 2 ICC (2,1)*	Home 1 vs Home 2 r	Lab**	Home 1 vs Lab ICC (2,1)**	Home 1 vs Lab p-value
Gait Speed (m/s)	1.01 (0.11)	1.05 (0.15)	0.85 (0.62, 0.93)	0.90	1.06 (0.15)	0.63 (0.31, 0.82)	0.01#
Cadence (steps/min)	111 (11)	114 (11)	0.92 (0.77, 0.97)	0.97	111 (8)	0.74 (0.51, 0.87)	0.61
Stride Length (m)	1.09 (0.10)	1.11 (0.13)	0.91 (0.80, 0.96)	0.94	1.14 (0.12)	0.59 (0.28, 0.79)	0.02#
Stride Duration (s)	1.09 (0.11)	1.07 (0.10)	0.92 (0.75, 0.97)	0.95	1.08 (0.08)	0.75 (0.53, 0.88)	0.37
Step Duration (s)	0.55 (0.06)	0.53 (0.06)	0.92 (0.73, 0.97)	0.95	0.54 (0.04)	0.78 (0.59, 0.89)	0.25
Stance (%GCT)	61.57 (1.71)	61.43 (1.97)	0.93 (0.86, 0.97)	0.94	61.21 (2.03)	0.81 (0.67, 0.91)	0.22
Swing (%GCT)	38.43 (1.71)	38.57 (1.97)	0.96 (0.91, 0.98)	0.94	38.79 (2.03)	0.81 (0.67, 0.91)	0.22
Double Support (%GCT)	23.5 (3.2)	23.0 (3.6)	0.95 (0.89, 0.98)	0.97	22.6 (4.0)	0.87 (0.67, 0.94)	0.01#
Terminal Double Support (%GCT)	11.66 (1.91)	11.34 (2.14)	0.94 (0.86, 0.97)	0.96	11.02 (2.08)	0.84 (0.64, 0.92)	0.04#
Chair Stand Duration (s)	1.05 (0.31)	1.11 (0.34)	0.89 (0.76, 0.95)	0.90	0.98 (0.21)	0.66 (0.40, 0.83)	0.12

^*Gait data during remote home visit were available from 18 participants (one participant did not wear the sensors correctly and sensors for one participant lost power). Chair stand data during home visit were available from 18 participants (two participants were not able to perform the task)

^**Chair stand data during lab visit were available from 16 participants (one person could not perform the task, sensors did not detect any chair stands for two participants, and data for one participant were of poor quality)

^#Statistically significant difference between lab and home values (p < 0.05)

Table 3:

Standard error of measurement (SEM) and minimum detectable change (MDC) values for at-home data.

Sensor variables	MDC	SEM	MDC%	SEM%
Gait Speed (m/s)	0.15	0.06	14.8	5.34
Cadence (steps/min)	8.51	3.07	7.58	2.73
Stride Length (m)	0.10	0.04	8.84	3.18
Step Duration (s)	0.04	0.02	8.25	2.98
Stride Duration (s)	0.08	0.03	7.42	2.68
Stance (%GCT)	1.34	0.48	2.17	0.78
Swing (%GCT)	1.34	0.48	3.47	1.25
Double Support (%GCT)	2.08	0.75	8.93	3.22
Terminal Double Support (%GCT)	1.36	0.49	11.83	4.27
Chair Stand Duration (s)	0.25	0.09	23.15	8.35

Agreement between at-home and lab visits:

Figure 3 shows the Bland-Altman plots for selected gait measures and the chair stand duration from the first set of trials in the at-home visit and the trials from the lab visit. For the at-home versus in-lab visits, agreement was moderate (0.5 < ICC < 0.75) for gait speed, stride length, cadence, and chair stand duration, and good (0.75 < ICC < 0.9) for stride duration, step duration, stance, swing, double support, and terminal double support (Table 2). Mean differences between values derived in the lab vs at home are also shown in Table 2. There were small systematic differences noted in the Bland-Altman plots in all metrics likely due to participants walking faster during the lab visit compared to the remote at-home visit (Figure 3).

Figure 3:

Bland-Altman plots for selected gait metrics (A-C) and chair stand duration (D) from the remote and lab visits. Lines show mean difference (dotted) and 95% limits of agreement (dashed).

DISCUSSION

We evaluated the reliability of wearable inertial sensor derived walking gait and chair stand metrics collected remotely in a person’s home environment and compared those measures to those obtained in a research laboratory environment. Our results show good to excellent test-retest reliability of these measures obtained at home. We also observed moderate to good agreement for the gait and chair stand measures across lab and home environments. Our findings suggest that, in future studies in people with knee OA, wearable sensors could be used for standardized assessment of gait and chair stand function in individual’s homes.

To our knowledge, there is limited prior information on reliability of gait or chair stand metrics from wearable sensors in a person’s home environment. In individuals with multiple sclerosis, ICC values similar to ours were reported (0.91 to 0.95) for mean spatiotemporal gait metrics derived from inertial sensors in a smartphone collected during multiple at-home daily self-administered gait assessments (20). A meta-analysis of studies on reliability of gait metrics from wearable sensors in a lab environment in healthy adults reported good to excellent reliability for stride time, stride length, stance time, and swing time metrics (0.85 to 0.92) (19). In people with knee OA, a prior study that used treadmill walking to study the reliability of gait and sensor metrics found good to excellent reliability for step length, single-limb support time, and ground reaction force first and second peaks walking on an instrumented treadmill at two different speeds and inclinations (38). Reliability of raw acceleration waveforms from wearable sensors on shank and thigh (39) and the foot and lower back (35) during treadmill walking has also been reported to be acceptable (ICC > 0.75) in people with knee OA. Thus, our approach for collecting these measures remotely can be used to reliably measure walking gait and chair stand movement patterns in people with knee OA in their home environments. Importantly, implementation of our approach in future studies should consider the use of similar rigorous methods including written and video instructions for participants and a guided at-home virtual visit.

For chair stand, a study in healthy young adults reported good to excellent reliability for measuring acceleration during chair stand using a single inertial sensor embedded in smart glasses (21). Another study reported ICC of 0.84 and 0.87 for chair stand duration measured using a single hip-worn inertial sensor during a five time sit to stand test performed as quickly as possible in young and old healthy adults respectively (22). Our findings are similar to these prior studies. However, in the prior study using a hip-worn inertial sensor, the reliability for chair stand duration was worse when the task was performed at self-selected pace (ICC of 0.25 and 0.66 in young and old respectively)(22). Hence, while our results suggest that wearable sensors may be used to remotely assess chair stand performance at home in future studies in people with knee OA, it is possible that our findings may only be generalizable to the standardized sit to stand test performed as quickly as possible.

As expected with inertial sensors, temporal gait metrics outperformed spatial metrics, with stride length and gait speed (which is derived from stride length) having the lowest ICC values compared to the temporal metrics in both testing scenarios. Step and stride length measures are typically extracted from inertial sensors using single or double inverted pendulum biomechanical models that require signal integration and can accumulate errors (27, 40). A meta-analysis of published studies supports this finding where temporal metrics (i.e., step time and stride time) were identified as having the strongest body of evidence for excellent validity and reliability (19). Importantly, our findings were consistent irrespective of knee pain severity in this population (see results for study leg and contralateral leg). We also report data for left and right leg which can serve as reference for future studies in healthy populations or where a distinction based on study and contralateral leg is not needed.

Data collected at home may have greater ecological validity than those collected in the lab. Bland-Altman plots of at-home minus in-lab gait data showed a small bias towards participants walking with a greater cadence, faster gait speed, and longer stride length in the lab compared to in their home environments. Further, gait speed and stride length were significantly faster on average for in-lab values compared to at-home. These results are consistent with previous literature showing that both healthy individuals and those with Parkinson’s disease tend to walk faster in a laboratory or hospital environment respectively than in free-living and home settings (10, 30, 41). Given the importance of gait speed for predictions of functional decline and other outcomes in people with knee OA, future studies may consider measuring gait speed in a person’s natural environment (42, 43). Overall. these results suggest that, while still comparable, caution should be used in instances where data are collected in different environments and aggregated.

In our study, feedback from the participants showed high acceptability of our approach. Previous work found it to be feasible to assess gait using wearable sensors in both the clinic and in the homes of individuals with mild Alzheimer’s disease (44). Participants reported the sensors as very easy to apply and completely comfortable to wear. The participants also noted that the level of commitment for the visit (45-60 min) was very manageable and that they would be very likely to participate in a similar visit again. Hence, our approach could be used to collect such data over repeated visits in clinical trials of interventions for people with knee OA.

We have provided important information that could be used to design and implement future studies. Consistent inertial sensor placement on the body is known to be important (45, 46). While half of our participants performed the at-home assessment after wearing the sensors on their own without any prior familiarity, all participants were provided instructions and were also guided by researchers in real-time via video. While ICC provides a relative measure of reliability, the SEM and MDC (Table 3) provide measures of absolute reliability. MDC represents the minimum amount of change that needs to take place to overcome error in the measurement (shown as SEM in Table 3). These values can be used in future studies where measured changes larger than the MDC can be considered a real change for a given participant (for example, a change > 0.15 m/s in gait speed). The SEM% and MDC% are similar to SEM and MDC but are independent of units of measurement. With SEM% being < 10 for all measures and MDC% being <10 for most measures, it suggests that these values are sensitive and could be used to detect small effects of interventions in future studies. It is important to note that MDC is not the same as minimally important change that is considered clinically meaningful (47).

There are some limitations to this study that should be considered. In our cohort, 95% (19/20) of participants self-identified their race as White, compared to the greater knee OA population. Individuals from minoritized races or ethnicities report greater challenges with technology in general (48) and specifically for health-related purposes (49, 50), which may limit the current generalizability of our results to a more diverse population. We also had a larger proportion of women in our cohort than what is reflective of knee OA. Our sample size was small but justified a priori. Additionally, we scheduled only a 15-minute gap between the two at-home collections as that was sufficient to demonstrate test-retest reliability because the participants removed and re-wore the sensors. However, implementation of our approach in future studies will likely include larger gaps between visits (for e.g., baseline and follow-up in a clinical trial) and it is possible that having a longer delay between the collections may yield different results. The gap between the lab and home visits was variable ranging from 1-20 days due to scheduling challenges across participants. However, given the chronic nature of OA pathology, this variability is unlikely to influence our results. Finally, we did not record the types of walking surfaces during the at-home visits which could partially explain the differences in gait parameters between home vs. lab visits.

CONCLUSION

In this cohort of people with knee OA who had moderate pain and disability, our method of estimating spatiotemporal gait measures and chair stand duration remotely was reliable, feasible, and participant accepted. Wearable sensors could be used to remotely monitor gait and chair stand function in participant’s natural environments at a lower cost, reduced participant and researcher burden, and greater ecological validity overcoming many limitations of lab visits. Hence, our approach could be used in future longitudinal studies or clinical trials of people with knee OA.

Significance and Innovation:

• Given the importance of daily activities as outcomes for interventions, standardized assessment of walking and chair stand in a person’s natural environments is important for knee OA research and clinical practice.

• In this study, we observed good to excellent reliability in remotely assessing walking gait and chair stand activities using wearable inertial sensors at-home in adults with knee OA; the agreement between at-home and in-person lab assessments showed a small bias, explained by participants walking faster in the lab environment. Participants were highly accepting of the at-home visit.

• Our approach could be used to monitor gait and chair stand activities reliably and remotely in individuals’ natural environments at a lower cost, reduced participant and researcher burden, and greater ecological validity.