Background
It is well-recognized that osteoarthritis (OA) of the knee or hip impairs gait [
1‐
4]. Indeed, individuals with knee or hip OA walk less during daily life and their quality of gait is compromised [
5]. Yet, objective gait assessments are not part of routine clinical evaluation, and gait difficulties in OA are insufficiently captured by patient-reported outcome measures [
6‐
8]. In part, this may be due to limited time available during clinical visits, considering that gait analysis is traditionally conducted in a gait laboratory, making it time consuming and not easily accessible. Recent advances in inertial sensor technology have opened up new avenues to quickly and objectively assess gait quality in a clinical setting.
Small inertial measurement units (IMUs) can be used to quickly and accurately obtain gait parameters without being restricted to a fixed (laboratory) environment [
9,
10]. Moreover, compared to gait analysis in a lab, substantially more strides can be collected in a shorter period of time. On the downside, an important issue of gait assessment with IMUs is that it typically results in a large number of outcome parameters, with numerous correlated parameters. For example, many gait parameters share covariance with gait speed [
11‐
15]. Hence, for clinical implementation, it is important to identify gait parameters from independent gait domains that best describe the gait adaptations in individuals with knee and hip OA compared to healthy controls.
So far, ambulatory gait assessments in individuals with knee and hip OA have mostly been limited to simple, straight-ahead walking paradigms [
16]. Parameters reflecting more complex and relevant aspects of gait, including dual-task gait, turning, and compensatory trunk motion are less frequently reported in studies using IMUs. Turning and dual-task performance have been shown to be important aspects of daily life ambulation in elderly populations and can easily be assessed using wearable sensors [
17‐
20]. Turning is a common cause of falling in community dwelling elderly, and may be more sensitive to sensorimotor impairments than straight-ahead gait [
19,
21]. Dual-task performance, on the other hand, reflects the amount of attentional resources allocated to gait [
22]. In order to compensate for gait difficulties caused by OA, a strategy could be to allocate more attention to gait. The extent to which a secondary cognitive task affects gait performance (i.e. dual-task cost (DTC)) may therefore be larger in individuals with OA. A recent scoping review indicated that DTC was not different between individuals with knee OA and healthy controls during quiet standing and forward induced falls [
23]. However, DTC during gait has not yet been compared between those groups. A third gap in literature regarding wearable sensors and OA is the lack of attention for upper body movement. Upper body motion is important for maintaining stability, but may also be indicative of compensatory gait changes that reflect OA-related pain or disability [
24‐
26].
The aim of this study was therefore to investigate turning, dual-task performance, and upper body motion in addition to spatiotemporal gait parameters in individuals with knee or hip OA, taking shared covariance between gait parameters into account. More specifically, we aim to test if 1) turning, dual-task gait, and upper body motion constitute independent domains of gait in our sample, and 2) gait parameters in these gait domains can discriminate individuals with knee or hip OA from healthy controls. Together, these findings may contribute to a better understanding of the multidimensional aspects of gait, and how this is affected in knee and hip OA.
Discussion
The aim of the present study was to investigate turning, dual-task performance, and upper body motion in addition to spatiotemporal gait parameters in individuals with knee or hip OA. To avoid redundancy of gait parameters, we conducted a factor analysis. Four independent gait domains were identified: speed-spatial, speed-temporal, dual-task cost, and upper body motion. Turning did not constitute its own domain but was related to speed-temporal. Three domains held parameters sensitive to knee or hip OA: speed-spatial (stride length), speed-temporal (cadence), and upper body motion (lumbar sagittal RoM). Dual-task cost was not sensitive to knee or hip OA.
Factor analysis effectively reduced the dimensionality of our dataset from twenty-five gait parameters to four independent domains of gait, including domains related to dual-task gait and compensatory trunk motion. Turning, however, was part of a factor together with cadence. The factors explaining most of the variance in our sample, i.e. speed-spatial and speed-temporal, were both dependent on gait speed (Table
2). In the literature, these factors reflecting the spatial and temporal aspects of gait speed are consistently reported [
38‐
42]. Other factors related to gait are variability [
38,
39,
41,
42], asymmetry [
39,
41,
42], postural control [
39], and trunk motion [
40]. Dual-task cost has not previously been evaluated in a factor analysis approach, but may contain unique information about gait that is informative of disease-specific compensations related to the re-allocation of attentional resources. Importantly, dual-task cost and upper body motion are interesting domains as they were independent of gait speed, evidenced by the absence of a cross-loading of gait speed on these domains in our study. Dual-task cost and upper body motion may therefore provide promising gait parameters for clinical evaluation of gait, in addition to the more commonly used speed-related measures.
In our analysis steps, turning parameters were excluded in favor of the stronger factor loading that was obtained for cadence. However, effect sizes for turning were large when comparing both knee and hip OA with healthy participants (SMD > 0.9, Fig.
3). In addition, factor loadings were not substantially lower compared to cadence. Taken together, we are unsure whether this factor represents a combination of gait and turning, or better reflects turning itself. Future research should therefore indicate as to what extent turning parameters are driven by cadence or gait speed, and how meaningful the unexplained variance is for evaluation of physical functioning in individuals with knee and hip OA.
To facilitate assessment of the between-group differences, we opted to select single gait parameters from the independent factor, to represent the respective factor. From the factors that we obtained, only dual-tasking parameters did not discriminate between knee or hip OA and healthy controls (SMD < 0.5). This indicates that, compared to healthy controls, individuals with OA did not need more attentional resources for the motor task. Thus, although gait was affected in OA, this was not compensated by more attentional resources.
Many of the gait parameters that showed large between-group effect sizes (Fig.
3) were grouped either under the speed-spatial or under the speed-temporal domain. This suggests that the two main components determining gait speed, stride length and cadence, are inherently linked with various gait adaptations prominent in individuals with knee and hip OA. As such, gait speed may also be considered as the final common pathway for various gait adaptations, and could be used as a very general, but highly sensitive marker for functional status in individuals with OA. Next to this, our findings further stress the need to take gait speed differences into account when evaluating gait in individuals with OA. More specifically, for parameters that are correlated with gait speed, it may be more appropriate to assess them at a standardized, matched speed, as it may be difficult to separate effects of gait speed from the effects of OA itself [
43]. Finally, these findings underline the importance of data reduction techniques when investigating gait using IMUs or motion capture systems, as statistical testing of all gait parameters would increase the probability of finding false positives.
That speed-related gait parameters have good discriminatory capacity in OA has been reported before. Two systematic reviews reported lower gait speed and stride length in individuals with knee and hip OA compared to healthy participants [
1,
3]. In studies employing IMUs, similar changes in stride length and cadence were found [
25,
44]. In absolute numbers, slight differences with our values can be discerned. Reasons for this may include the relatively short walking distance (6 m) in this study that was necessary to reliably assess turning, versus the longer distances (~ 20 m) that are commonly used. Nevertheless, our findings corroborated previous findings about the discriminatory capacity of stride length and cadence.
In addition to spatiotemporal differences, individuals with hip OA walked with distinct upper body motion, which was most evident in the sagittal plane at the lumbar level. However, upper body motion is difficult to capture by just one parameter, as is illustrated by the relatively low factor loadings lying close together in this domain (Table
1). Altered trunk motion may point toward the use of compensatory strategies to unload the arthritic joint [
45]. More specifically, increased pelvic RoM in the sagittal plane may enable more effective anteflexion of the lower limbs and may thereby, to a certain extent, preserve stride length [
46]. In addition, anterior pelvic tilt combined with lateral trunk lean can reduce the lever arm between the hip joint center and center of mass [
25]. We observed more lumbar sagittal RoM and more RoM of the trunk in the coronal plane in individuals with hip OA compared to healthy controls, in line with previous reports [
25,
46]. Unfortunately, the exact reason for the use of these compensatory mechanisms remains speculative and may relate to pain, muscle weakness, or joint instability [
47]. Future research should therefore investigate the importance of upper body motion in individuals with OA, to inform us about potential mechanisms underlying these gait adaptations.
With regard to the use of wearable sensors in clinical practice, our study showed that quick and easy gait assessments with wearable sensors are useful for evaluating gait impairments in individuals with knee and hip OA. In comparison to optical motion capture systems, wearable sensors are more feasible for large-scale use and could be utilized to routinely assess physical functioning. From all gait parameters, gait speed was found to be a very general but highly sensitive marker for mobility limitations, combining both the effects on stride length and cadence. Besides the basic spatiotemporal measures, trunk motion and turning appeared to be relevant for individuals with knee and hip OA. We therefore recommend to use sensor configurations that allow to look beyond these basic spatiotemporal parameters. In the future, wearable sensors should also be utilized to their fullest potential to enable remote monitoring at home, which would allow to more accurately capture the habitual gait patterns.
This study had several limitations that merit attention. First, we did not obtain factors representing gait asymmetry or variability, which may have been related to the low number of gait parameters related to those domains that were initially entered into factor analysis. We were therefore limited in our conclusions regarding the potential value of those measures for individuals with knee or hip OA. Second, five potentially valuable gait parameters were removed from further analysis due to sampling inadequacy (KMO value < 0.5). Larger sample sizes are therefore required to identify the potential value of these parameters. Related to this, we did not include demographic or clinical variables in the factor analysis, as this could have affected the accuracy of factor analysis due to the relatively small sample size. Finally, including individuals with isolated, unilateral knee or hip OA was important for our study purposes, although the majority of the OA population have complaints in more than one joint [
48]. We expect that widening the inclusion criteria would have resulted in larger differences of OA groups compared to healthy controls, but in less specificity for each OA group. In addition, it is important to note that individuals in this study had end-stage OA and were scheduled for joint replacement. Our results may thus not be representative of gait in individuals with less severe OA.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.