Rehabilitation improves recovery after stroke, with better outcomes when applied intensively and tailored to individual needs [1-4]. Wearable technology has the potential to support increased self-managed rehabilitation by providing performance feedback during everyday activities such as walking in the community but needs to be accurate [5].

As an important feature of independent living, the recovery of walking ability is a major rehabilitation goal [6]. Reduced walking ability is common after stroke, particularly in hemiplegic stroke, which causes specific gait impairments such as reduced stance time duration on the hemiplegic side [3,7]. Speed is the most widely reported and clinically accessible metric in stroke rehabilitation, used as a global measure of mobility recovery [8,9]. Speed alone, however, does not provide meaningful information on the underlying impairments needed to inform effective rehabilitation interventions [10]. One metric that could provide this insight and is clinically relevant and sensitive to change, is stance duration symmetry [11-13]. Measuring this metric during everyday rehabilitation is problematic but achievable with wearable sensors; however, such an approach should consider measurement burden and potential Hawthorne effects [14]. Integrating these metrics into existing systems, for example physical activity monitors, may be a more acceptable approach.

One well-established wearable device, designed for measuring free-living physical activity, is the ActivPAL4+ (AP4) activity monitor (PAL technologies), which is a uniaxial accelerometer attached to the anterior thigh, using proprietary algorithms to measure physical activity and posture of a healthy and impaired populations within free-living environments [15]. This includes standing, sitting, walking (durations and transitions) and measurements of stepping cadence, step counts and energy expenditure [16]. The AP4 has good validity for measuring walking bouts at normal walking speeds, intraclass correlation coefficient [ICC (2,1)=0.78] when compared to direct observation [15,17-19].

Many algorithms and analysis techniques have been developed to measure gait parameters from wearable sensors, from frequency domain metrics to more complex approaches that calculate joint angular displacements [20,21]. To detect gait events specifically, most approaches have used peak detection algorithms and zero crossing techniques, but these have not been found to be robust [22-24]. Gait event detection using the Teager Kaiser Energy Operator (TKEO) with AP4 data has been attempted before with Huntington’s disease, to determine initial (IC) and final contact (FC) events, with the resulting stance phase calculation consistently underestimated (0.08 s), compared to video analysis [25,26]. While this technique appears promising from tests in healthy and impaired participants populations data, there is a need to test with a hemiplegic population post stroke who stand to benefit from the enhanced gait rehabilitation, that could be enabled by the feedback from this approach.

The aim of this study, therefore, was to test the concurrent validity of a thigh-mounted physical activity sensor (AP4) for measuring the stance phase duration of hemiplegic gait, with a 3D motion capture system details acting as the gold standard measure. The hypothesis was that the AP4 would have acceptable levels of concurrent validity through an ICC (3,1) greater than 0.75, and a low (<0.1 s) absolute difference between the two systems experience.

The Montreal cognitive assessment (MoCA) is a screening instrument that evaluates general mental capabilities, such as visuospatial abilities, executive functions, and orientation to time and space. The MoCA is rated between 1 and 30, with increasing score dictating better cognitive ability [31]. The functional ambulation category (FAC) evaluates walking ability in 6 levels, with a score of 0 defining no ability to walk, or requiring the help of 2 physiotherapist, and a score of 5 defining full capability to walk independently, including stairs [32]. The Rivermead mobility index (RMI) is an outcome measure used to assess mobility after stroke, rated between 0 and 15, with increasing score indicating better mobility [33] (Table 1).

The AP4 (mean stance time=0.85) saw consistent underestimation in comparison to the ECMS (mean stance time=0.88). Despite this, both methods had a high agreement for stance time measurement (T=0.61, P=.54, 95% CI for difference=−0.07s, 0.13s, ICC [3,1]=0.79). The differences between the AP4 and ECMS were shown to be within a standard distribution (D=0.11, P=.10).

The Bland-Altman plot (Figure 2) displays a consistent spread between the techniques during the performed stroke gait which ranged between 0.49 and 1.90. This shows close agreement between the AP4 and ECMS, irrespective of the value. A bias of 0.03 seconds is reported, suggesting excellent accuracy of the AP4 in comparison to ECMS. The LOA were 0.28s and −0.22s, which contain the 95% of the recorded datapoints. This indicates good agreement between the AP4 and ECMS.

This study aimed to test the concurrent validity of detecting gait events (ie, initial and final foot contact) and the derived metric—stance phase duration—from thigh acceleration data analyzed with the TKEO, compared with a gold standard 3D motion capture (Vicon) in chronic survivors of stroke. The excellent accuracy (bias=0.03 s) were similar to Flood et al (IC error=0.01s, FC error=0.02 s) [26]and Lozano-García et al (bias=0.08 s) [25] . In the context of a typical stance duration (nonparetic =1.01 s), the reported difference of 0.03 s (2 sampling frames) represents a 2.97% difference, with performance comparable to other commonly used methods of wearable sensor gait parameter measurement, collected by Pacini Panebianco [34] et al. The reported bias of 0.03 seconds could be considered clinically unimportant. A study by Wang [35] et al denoted stance time averages and variability from survivors of stroke at different gait speeds. They noted 0.90 (SD 0.34) seconds and 1.01 (SD 0.41) seconds for the paretic and nonparetic sides, respectively. The magnitude of the bias, therefore, will have little effect on the readings that the AP4 will make by using the TKEO technique. These findings should be confirmed with a larger sample that preserves the diverse nature of the current sample (speed ranged between 0.06‐1.39_ms⁻¹ ).

This positive outcome opens opportunity to expand the output of wearable sensors currently used as physical activity monitors, to provide feedback on gait parameters (stance duration, symmetry and weight transference) during the rehabilitation of patients with hemiplegic stroke, without increasing measurement burden.

The results should be interpreted in light of the study’s limitations. The testing took place in a laboratory setting, reducing the generalizability of results. This controlled environment may not reflect everyday gait [36,37]. To obtain ecologically valid gait data, there is a need to capture in free-living environments. The sample of participants with a history of chronic stroke means that only a specific subset (chronic) of the population with stroke has been validated for this technique. Future studies are encouraged to include subacute populations who may have more variable gait parameters. The study had a small sample size, limiting the statistical power and potentially not representing the whole population, although the range of walking ability (walking speed 0.06‐1.39_ms⁻¹ , FAC 1‐5; Table 1) is reassuringly wide.

In conclusion, the AP4 sensor, in conjunction with the TKEO technique, has been validated against a gold standard 3D Motion Capture system, for stance duration measurement in participants with chronic stroke. However, this positive finding is limited by the study’s setting and small sample. Future work should consider a bigger sample and collect gait data in free-living environments. The outcomes of this study could be exploited to enhance the function of wearable sensors, in order to provide the gait parameters valuable for self-rehabilitation after stroke, such as symmetry.