A progressive-individualized midstance gait perturbation protocol for reactive balance assessment in stroke survivors

Abstract

Restoration of balance control is a primary focus of rehabilitation after a stroke. The study developed a gait perturbation, treadmill-based, balance assessment protocol and demonstrated that it can be used to quantify improvements in reactive balance responses among individuals post-stroke. The protocol consists of a sequence of fifteen 90-second treadmill walking trials, with a single perturbation applied during the middle third of each trial. Gait was perturbed by rapid acceleration-deceleration of the treadmill belt at midstance of the unaffected leg during a randomly selected gait cycle. The initial perturbation magnitude was based on the participant’s maximum walking speed and increased or decreased in each trial, based on success or failure of recovery, as determined from an instrumented harness. The protocol was used before and after a 10-week period of therapy in twenty-four stroke survivors. Outcomes included maximum recoverable perturbation (MRP), self-selected gait speed, levels progressed through the algorithm, and falls versus recoveries. Participants were able to take recovery steps in response to the perturbation. Twelve participants completed the full assessment protocol before and after the therapeutic intervention. After the intervention, they had fewer falls and more recoveries (p < 0.001), progressed through more algorithm levels (p = 0.043), had a higher MRP (p = 0.005), and had higher gait speeds. The protocol was found to be feasible in stroke survivors with moderate gait deficits. The data supports the conclusion that this protocol can be used in clinical research to quantify improvements in balance during walking.

1. Introduction

Up to 73% of individuals post-stroke fall within 6 months of their release from the hospital (Mansfield et al., 2015). Falls often occur during walking for these individuals and it has been suggested that impaired reactive stepping correlates strongly with fall risk (Shumway-Cook and Woollacott, 2014). Individuals post-stroke experience difficulty executing accurate movements during both swing and stance phases of the gait cycle (Daly et al., 2011).

Adequate foot placement is imperative for maintaining balance and stability during walking (Hof, Gazendam and Sinke, 2005). Limitations in strength and movement speed often significantly impair stability in response to external perturbations (Roos and Dingwell, 2013). Balance can be especially challenged when a forward fall inducing perturbation is applied at mid-stance of one limb; the speed and precision of the response, and the resulting foot placement, are critical for recovery.

Treadmill-based gait perturbations are effective in training fall prevention strategies (Hamacher et al., 2011), and produce improved reactive responses (Yang, Bhatt and Pai, 2013; Mansfield et al., 2018). These reactive training studies advocate the use of randomly timed perturbations to avoid predictability (Wass, Taylor and Matsas, 2005). (Wang et al., 2019) investigated balance stability by inducing backward falls after older adults underwent repeated treadmill slips, however they did not control the timing of the perturbations. These perturbations were induced at random points in the gait cycle, leading to considerable variability in responses required for a successful recovery, which affects statistical power and clinical generalizability.

Instrumented treadmills can be an effective training tool for gait perturbation recovery as they can deliver precise, repeatable, and challenging perturbations; however, utilizing treadmill perturbations as an assessment tool to quantify improvements in balance before and after interventions has not been examined. The aim of this study was to develop and implement a progressive, treadmill-based, midstance gait perturbation protocol for reactive balance assessment, and to demonstrate that it can quantify improvements in balance responses after therapeutic interventions among people post-stroke.

2. Methods

Twenty-four post-stroke participants (17 males,13 left hemi-paretic; 61.5 ± 10.58 years; 86.6 ± 15.1 Kg) were recruited. Participants were ≥18 years, had a stroke for > 6 months, were able to walk independently, and self-identified balance difficulties. Participants were excluded if they had any musculoskeletal, neuromuscular, or cardiopulmonary conditions. The study was approved by the Institutional Review Board at Cleveland State University and written informed consent was obtained from all participants.

An instrumented R-Mill treadmill (Motekforce Link, Amsterdam, Netherlands) with six degree-of-freedom force plates and two side-by-side independent belts was used (Fig. 1A). Participants wore a custom instrumented harness (MASS Rehab, Dayton, Ohio, USA) during testing. Cortex 5.0.1 (Motion Analysis Corp., Rohnert Park, CA, USA) recorded vertical force, load cell, and motion data simultaneously. D-Flow software (Motekforce Link, Amsterdam, Netherlands) controlled the treadmill speed, while displaying and recording streaming data from Cortex.

Fig. 1.

(A) Experimental setup for post stroke gait perturbation. An instrumented treadmill with two independent force plates (FP1 & FP2), load cell (solid red square), Harness system (solid blue rectangular), assistive 3-level steps. (B) Trial design consisted of a 90-second of walking of three periods: pre-perturbation, within perturbation, and after perturbation. Each participant in each trial was perturbed unexpectedly at exactly the same time in the gait cycle. Participants walked on the treadmill using their normal gait speed about 23 min (90 sec for maximum of 15 trials) in a single testing session. (C) Screen image from D-Flow system showing the console application parameters for individuals’ post-stroke; including the peak load as % of BW, affected side, NWS, MWS, and mPer. The data flow is shown on the left, in which the Lua scripting language was used to execute a perturbation under the unaffected foot exactly at midstance in one random gait cycle within the perturbation period.

Reactive balance was tested by a progressive perturbation protocol on the R-Mill treadmill before and after 10 weekly intervention sessions. Participants were randomly assigned to one of the three intervention groups: clinical physical therapy, reactive standing slip, and active video gaming. The analyses for this study use the combined data of participants from all three groups to demonstrate the utility of the assessment protocol.

The protocol consisted of a maximum of fifteen 90-second walking trials, each with three 30-second periods: pre-perturbation, perturbation, and recovery (Fig. 1B). A perturbation was induced once during the perturbation period by increasing the treadmill belt speed by a specified value for 0.25 seconds, then returning it to its previous speed (Fig. 1B). All speed changes were done at a constant acceleration of 15 m/s². These perturbations, which induce a forward fall, were applied at mid-stance of the unaffected leg, during a randomly preselected gait cycle within the perturbation period.

Average stance time was determined from vertical force data from gait cycles prior to the perturbation and used to calculate the mid-stance time after the randomly selected heel-strike (Osman et al., 2019). A custom application using D-flow software and the scripting language Lua controlled the treadmill when executing the gait-perturbation protocol. Participant parameters were used to individualize the session to each individual (Fig. 1C).

To determine their self-selected normal and fast walking speeds, participants walked on the treadmill, without handrail support, while the investigator adjusted the belt speed: participants self-identified their normal walking speed (NWS) and maximum walking speed (MWS) when each was reached. The treadmill ran at each participant’s NWS throughout the perturbation testing sessions.

The perturbation magnitude (mPer) in the first trial was based on the individual’s MWS and balance ability as measured by the Mini-BEST. The Mini-BEST test is a performance-based measure of balance deficits addressing anticipatory and reactive control, sensory orientation, and dynamic gait; higher scores denote better balance ability, the maximum score is 28. Three ranges of scores were used to create three levels of percent MWS to determine mPer (Fig. 2A).

Fig. 2.

(A) Initial perturbation magnitude (mPer) was calculated based on the participant’s’ Mini-BEST score and their maximum walking speed (MWS in m\s). There are 3 levels of intensities: high, medium, and low. Each level matched with Mini-BEST total score and % of MSW. The same initial mPer was used for pre-and post- intervention testing. (B) A stepwise progression based on trial outcome then modulates the magnitude for subsequent trials. Trial outcomes were classified based on the peak load (PKL) as percentage of body weight (BW). R represents the recovery trial (PKL< 5 % of BW), M represents the intermediate trials (PKL 5% −30 % of BW), and F represents the fall trials (PKL> 30 % of BW), following the stepwise progression algorithm.

Trial outcomes were classified based on the peak load (PKL) as percentage of body weight (BW): R: recovery (PKL< 5 % BW); F: fall (PKL> 30 % BW), and M: intermediate trials (Bhatt et al., 2011). Subsequent to the initial perturbation, mPer was determined through a progression algorithm (Fig. 2B): After the first two trials, at each subsequent trial, mPer was increased by 20% if the two previous trials had been recoveries, mPer was decreased by 20% if the two previous trials had been falls; intermediate trials were repeated until two consecutive falls or recoveries occurred.

Four outcomes were measured: trial outcomes (the number of falls and recoveries), algorithm level progressions (PrgL), maximum recoverable perturbation (MRP), and gait speeds. Algorithm progression was the number of levels a participant advanced (higher mPer) through the algorithm in one session. The MRP was the largest perturbation magnitude (m/s) achieved throughout the testing session. Gait speeds, NWS and MWS (m/s), were measured by the treadmill.

Statistical analyses were performed using Minitab software (Minitab Inc., State College, PA). Paired t-tests for mean differences were used to compare outcomes pre to postintervention. Data were presented as mean ± standard deviations and a significance level of α = 0.05 was used.

3. Results

Two participants withdrew from the study before the post-test, and 10 were not able to complete the protocol. Of the remaining 12, one was able to complete only 11 trials at pre- and all 15 trials at post-testing; another was limited to 11 trails in both pre- and post-testing. Their data were deemed sufficiently complete to be included in the analysis. Table 1 shows the demographic information for all study participants.

Table 1.

Demographics and anthropometric information of the study participants.

Protocol	Feasible (n =12)	Not feasible (n =10)
Gender (M / F)	9 / 3	6 / 4
Age (yrs.)	60.4 ± 13.0	61.3 ± 4.7
Affected - Side (R / L)	8 / 4	3 / 7
Weight (KG)	88.6 ±14.96	89.0 ± 13.3
Mini-BEST* Score	19 ± 3	15 ± 3

M= Male, F= Female, R = Right, L= Left. Data presented as mean ± standard deviations.

^*P< 0.05 for paired compared means.

Figure 3 shows an example of the treadmill velocity, vertical forces, and the knee angles for a small and a large perturbation in a left hemi-paretic participant. The perturbation was generated at the specified time (mid-stance) in a random gait cycle.

Fig. 3.

Illustration of velocity and the ground forces reaction results (top). A left side hemi- paretic stroke participant response to different perturbation magnitudes; small perturbation and large perturbation on the right foot perturbed. The perturbation consisted of the treadmill belt speeding up to the calculated magnitude for 0.25 seconds then returning to its previous speed and was delivered at exactly at midstance in one random gait cycle within 30 seconds in the perturbation period. An example of a participant’s data (bottom). This to show how the levels of progression was calculated before and after balance testing; R represents the recovery trial, M represents the intermediate trials, and F represents the fall trials following the stepwise progression algorithm. Participants showed improvement in reactive balance response in pre-to post tests.

NWS was faster at post-testing than pre (p = 0.003) as was MWS (p = 0.010) as shown in Fig. 4. Levels progressed (PrgL) increased significantly pre- to post intervention within participants (p = 0.043; Fig 5).

Fig. 4.

Difference of means for the normal walking speed (NWS) and maximum walking speed (MWS) were significantly different before and after the tests (p < 0.05). Error bars are the standard deviations. Note: The treadmill ran at each participant’s NWS throughout the perturbation testing sessions. Participant’s maximum gait speed was used to calculate the perturbation magnitude (mPer) in the first trial.

Fig. 5.

Comparing pre to post tests, change of means for FLP group (n =12) were found statistically different (P < 0.05) between the outcome measures; Falls and recoveries (top), progression levels (PrgL) in (middle), and maximum recoverable perturbation magnitudes (MRP) in the (bottom). Error bars are the standard deviations.

There were significantly more recoveries than falls (P < 0.05) in both sessions (pre:1.7 ± 1.7 falls vs. 11.7 ± 2.6 recoveries; post: 0.6 ± 0.9 falls vs. 13.5 ±1.93 recoveries). Compared to the pre-session, there were fewer falls at post-testing (p = 0.015) and more recoveries (p = 0.001; Fig. 5).

The first trial perturbation magnitude for all participants in both sessions started at the lowest level (Fig. 2A) as the Mini-BEST scores ranged from 14–23. MRP was larger at post than at pre-test (0.331 ± 0.328 m/s, p = 0.005: Fig. 5). There was a positive correlation between Mini-BEST score and MPR in both sessions (pre: r = 0.60; post: r = 0.45).

4. Discussion

We implemented an individualized, treadmill-based, midstance gait-perturbation protocol for reactive balance assessment and demonstrated that it quantifies a number of improvements in balance responses pre to post intervention. In this progressive protocol, gait perturbations of participant tailored magnitude were induced under the unaffected leg during its mid-stance.

Perturbations were generated as designed, at specifically mid-stance of a random gait cycle; the perturbations elicited responses as expected (Fig. 3). Participants produced effective responses to perturbations of increasing magnitudes over the course of the pre intervention trials and showed improved responses to higher magnitude perturbations in the post intervention sessions (Fig. 3).

Training protocols have used the idea of modulating intensity vs. constant intensity perturbations and have found that a progressively increasing protocol was better than a progressively decreasing protocol. (Liu, Bhatt and Pai, 2016). However, the starting point was not tailored to each individual, which is quite important for individuals post stroke since they have extremely wide-ranging motor abilities, particularly for gait and balance. Our protocol is both tailored in perturbation magnitude to each individual participant and progressive in the perturbation magnitudes. This allowed us to assess fall resistance to an individually scaled perturbation and to assess degree of adaptation and learning for increasingly larger perturbations within a session.

Ten participants were not able to complete the protocol while 12 were. We recognized three requirements under which this perturbation protocol was feasible: walking speed greater than 0.2 m/s to generate a noticeable perturbation; able to walk on the treadmill without holding the handrail for the 90 sec trials to elicit solely a stepping strategy for recovery; and able to walk consistently on both force plates (each foot landing on its side of the treadmill) to trigger perturbations under the unaffected leg. We conclude that the protocol should not be used for participants who do not meet these conditions.

Participant age, weight, and gender were similar between those who completed the protocol and those who could not, but hemiplegic side was dissimilar (Table 1). However, participants who could not complete the protocol had lower Mini-BEST scores and slower gait speeds. This suggests that there is a cut-off, albeit a low one, of balance ability to participate in this protocol; further research should clarify this.

Our 12 participants’ normal gait speeds ranged from 0.25 – 0.70 m/s (pre) to 0.25 – 0.85 m/s (post). These are slower than reported normal gait speed for high-functioning individuals with stroke (0.78 m/s) (Beyaert, Vasa and Frykberg, 2015). We were able to assess and demonstrate fall resistance improvements among people post-stroke who are not high functioning, i.e., those who have slower gait speeds and greater balance challenges and for whom balance training programs would be important.

Our protocol was able to demonstrate improvements from pre-intervention to post-intervention in gait speed, trial outcomes (fall vs. recover), algorithm level progressions (PrgL), and maximum recoverable perturbation (MRP). Normal and maximum gait speeds improved significantly pre to post intervention (Fig. 4). Compared to the pre-session, there were fewer falls at post-testing and more recoveries, showing improvement in participants’ ability to generate a successful response.

The results showed that all participants recovered from a range of perturbation magnitudes (Fig. 5). The MRP was higher at post intervention testing than at pre-testing (Fig. 5) demonstrating an improved ability to resist falling at higher level perturbations. We found a positive correlation between Mini-BEST scores and MRP in both sessions, suggesting that the magnitude of a perturbation to which one can resist falling may be related to overall balance ability.

Clinical tests are frequently used to assess balance control or ability (Godi et al., 2013; Tsang et al., 2013). Some of these clinical measures can be predictive of fall risk based on studies of their scores among people who prospectively experience falls or have a history of falling. This is imprecise and questions have been raised about the ability of clinical tests to predict fall risk in older adults (Pai et al., 2010). We suggest that our findings hold implications for people post-stroke for the potential predictive ability for fall risk where non-perturbation inducing tests might not predict fall risk. Further research is needed to confirm this.

A limitation of this work is that the within-session improvements in response to perturbations could have been influenced by task learning or adaptation (Pai et al., 2014). However, all participants experienced similar learning opportunities in pre as post-testing. We propose that the improved learning across the post session vs. the pre appears to have been positively impacted by the intervening balance training intervention.

These results demonstrate a method to deliver an individualized and appropriately challenging progression of perturbations to assess improvements in balance responses from pre- to post- balance training intervention. In future work, we plan to quantify these responses and the specific movement strategies used. Moreover, we have demonstrated a potential direct approach for assessing reactive balance improvements that may offer a valuable alternative to longitudinal falls diary data collection.