Study at a glance
Summary
Disclaimer: The following summary details independent academic research. Theia3D is intended for research, sports performance and educational use only. It is not cleared as a medical device and is not intended to be used for the diagnosis, treatment, mitigation or prevention of any disease, injury or medical condition.
Editor's note: Theia3D version v2023.1.0.3161, as stated in the paper. This study uses Theia3D as the in-lab reference standard against which a wearable sensor system is validated, rather than evaluating Theia3D's own accuracy against marker-based motion capture.
Study Overview
Design
A prospective study, conducted as part of a larger cohort following individuals with knee osteoarthritis through a joint replacement procedure, comparing gait metrics derived concurrently from in-lab markerless motion capture and wearable shank-mounted sensors, and comparing those in-lab wearable sensor metrics against up to seven days of continuous free-living wearable sensor data from the same participants.
Participants
Forty-five older adults with end-stage knee osteoarthritis (mean age 65 years; mean height 167 cm; mean weight 94 kg; 14 men, 31 women), tested approximately two weeks prior to a scheduled joint replacement procedure as part of a larger cohort study. Mean self-reported pain was 7 out of 10 and mean Oxford Knee Score was 23 out of 48, indicating substantial functional limitation. Exclusion criteria were any significant disability preventing walking or an inability to provide informed consent.
Data Collection
Participants completed surveys assessing pain, function (Oxford Knee Score), and depressive symptoms (an 8-item screening questionnaire), then had two small wearable inertial sensors affixed to each shank, just below the knee. They then completed a series of movement tasks for an in-lab markerless motion capture assessment, including quiet standing, a 60-second self-selected-pace walk (the primary data source for this study), a faster-paced walk, a sit-to-stand test, and ascending and descending a short set of portable stairs. Ten synchronized Sony RX0-II cameras recorded the in-lab session and were processed through Theia3D to derive kinematic variables. Participants then left the lab wearing the same sensors, which continued recording continuously for up to seven days to capture free-living gait data during their normal daily routine.
Analysis
In-lab Theia3D data were processed through Visual3D to calculate stride, stance, and swing time, along with peak mediolateral shank angular velocity as a measure of sagittal-plane limb swing intensity. Wearable sensor data were processed using a custom pipeline including a machine-learning-based activity classifier to identify walking bouts and an event-detection method to segment individual strides, with an outlier-removal procedure applied to exclude strides with atypical waveform patterns. Agreement between in-lab motion capture and in-lab wearable sensor data, and separately between in-lab and free-living wearable sensor data, was assessed using Pearson correlation coefficients, Bland-Altman analysis, two-way random-effects intraclass correlation coefficients, and Cohen's d effect sizes. Correlations between gait variables and self-reported pain, function, and depressive symptoms were also calculated for both in-lab and free-living data.
Key Findings
- In-lab wearable sensor data showed moderate-to-excellent agreement with Theia3D-derived motion capture measurements, with the strongest agreement for stride time (mean difference 0.008 seconds, r = 0.96) and good agreement for stance time (mean difference 0.05 seconds, r = 0.89), supporting the validity of the wearable sensors as an in-lab measurement tool.
- Swing time and peak shank angular velocity showed comparatively weaker, though still moderate, correlation between motion capture and the in-lab wearable sensors (r = 0.71 and r = 0.68, respectively), with peak angular velocity in particular showing a consistent bias between the two measurement systems despite the overall moderate correlation.
- Agreement between in-lab and free-living wearable sensor data was only moderate across every gait variable tested (ICC > 0.67), a clear step down from the strong in-lab agreement between motion capture and the sensors. Free-living walking was consistently slower and more variable than in-lab walking across the week of continuous monitoring.
- Peak shank angular velocity showed the largest discrepancy between in-lab and free-living conditions (in-lab mean 288.3 deg/s vs free-living mean 254.2 deg/s; ICC = 0.75, Cohen's d = 0.70, P<.001), while swing time showed the smallest discrepancy between settings, with nearly identical in-lab and free-living means despite the reduced movement intensity reflected in the angular velocity measure.
- At the individual level, free-living stride times were often substantially more variable and right-skewed than in-lab stride times, with participants' in-lab values frequently falling near the faster end of their own free-living stride-time distribution, even though average stride time looked similar between settings at the group level, illustrating that group-level summary statistics can mask meaningful within-person differences between structured and everyday walking.
- Free-living peak shank angular velocity, but not in-lab peak angular velocity, was significantly associated with self-reported pain (free-living: r = -0.31, P=.04; in-lab: not significant), while longer stride and stance time in the lab were significantly associated with worse self-reported function and greater depressive symptoms, associations that were weaker and not statistically significant in the free-living data.
What This Means for Movement Science Researchers and Mobility Monitoring Teams
For researchers and teams studying knee osteoarthritis and other conditions where everyday mobility matters, this study offers a clear, practically important finding: gait assessed during a brief, structured lab visit using markerless motion capture may systematically overstate how someone moves in daily life, with free-living walking in this cohort consistently slower and showing less vigorous limb swing than in-lab walking. The specific promise shown by peak shank angular velocity, a measure that can be derived directly from a single wearable sensor without the spatial calibration that gait speed or stride length require, and that correlated with self-reported pain and function in a pattern comparable to motion-capture-derived gait speed, positions it as a practically accessible metric worth tracking in extended, real-world monitoring alongside, or as a complement to, lab-based motion capture assessment.
The Role of Theia3D in Movement Analysis
Theia3D served as the in-lab reference measurement system throughout this study, providing the kinematic gait data against which the wearable sensor system's in-lab accuracy was established before extending the sensors into continuous free-living data collection. Because this study's central research question concerned how well in-lab assessment reflects free-living gait, rather than evaluating Theia3D's own measurement accuracy, Theia3D's role here was as a trusted, already-established laboratory-grade measurement standard, providing the anchor point needed to validate the wearable sensors before they were deployed on their own in participants' everyday environments.
Study Limitations and Future Directions
The authors note that this study examined only a limited set of gait parameters; the wearable-sensor-based analysis did not include measures such as stride length or gait speed, common markers of gait health and disease progression in knee osteoarthritis, because these variables are difficult to estimate accurately from a single shank-mounted sensor, particularly in free-living data collected over an extended period, so the study instead focused on temporal and native sensor-derived variables that could be estimated more robustly under these conditions. They also note that wearable-sensor-derived estimates may be prone to error over extended data collection periods, highlighting the potential future value of further validating native sensor-derived variables as alternatives to conventional motion-capture-based gait metrics. This study focused exclusively on individuals with late-stage knee osteoarthritis preparing for joint replacement, so these relationships and agreement levels may not generalize to other populations, though the authors note their findings provide a useful framework for evaluating gait measures in other contexts. Finally, although the sample size was comparable to or larger than many studies of this type, the use of a convenience sample without a formal prior power calculation is a limitation that may increase the risk of over- or under-detecting true effects. The authors identify examining sensitivity to practically meaningful change, responsiveness following joint replacement, and validity across a broader range of osteoarthritis severity and participant populations as important next steps.
Citation: Ruder, M., Di Bacco, V. E., Madden, K., Adili, A., & Kobsar, D. (2026). Comparing gait metrics from in-lab gait analyses to free-living assessment from wearable sensors in end-stage osteoarthritis patients. Gait & Posture, 129, 110232. https://doi.org/10.1016/j.gaitpost.2026.110232
Link: Read the full study
Does In-Lab Gait Assessment Reflect How People With Knee Osteoarthritis Actually Walk at Home?
This study uses Theia3D as the in-lab reference standard to show that wearable sensors agree strongly with lab-based motion capture during structured assessment, but that free-living gait in people with knee osteoarthritis is consistently slower and more variable than what a brief lab visit captures. If your work involves tracking real-world mobility or validating wearable sensors against lab-based gait assessment, talk to our team about applying Theia3D as a reference standard in your own research.