Feasibility and Preliminary Effectiveness of a 2-week in-place reactive balance training program in people with multiple sclerosis

Abstract

Background & purpose:

Current treatments for falls in people with multiple sclerosis (PwMS) are incompletely effective. Reactive balance can prevent falls after a loss of balance and is negatively impacted by MS. Perturbation training may improve reactive balance. However, the effects of extended training and its impacts on falls are poorly understood in PwMS.

Methods:

We conducted a task-specific, multi-baseline, in-place reactive balance training program. Over 2 weeks and 6 sessions, participants were exposed to approximately 192 support-surface perturbations in four directions (forward, backward, leftward, and rightward). Reactive stepping was assessed twice before training (Baseline-1 & Baseline-2; [B1, B2], 2-weeks apart), and twice after training (Post-1 & Post-2 [P1, P2], immediately and 2 months after training). Linear mixed models assessed exposure effects (B1-B2), immediate improvement (B2-P1), and retained improvement (B2-P2). Falls were prospectively assessed for 2 months before and after training. Primary outcomes were margin of stability (MOS), latency, and length of the first reactive step after a backwards loss of balance.

Results:

Twenty-seven PwMS at risk of falls completed the study through P1 (immediate improvement), with 96.9% session attendance, and minimal reported adverse events. Of these, 20 participants completed P2 (2-month retention) testing. For backward losses of balance, no significant changes were observed from exposure (B1-B2; p’s≥0.603). There were statistically significant immediate improvements in MOS and step latency (B2-P1; p=0.036 & p=0.012, respectively) and retained improvement in step latency (B2-P2; p=0.033). For forward losses of balance, participants exhibited immediate improvement in MOS (p=0.042) and step latency (p=0.042). Fall counts were not lower after training compared to pre-training (p=0.266). No adverse events were observed.

Discussion:

PwMS exhibited immediate and retained improvements in reactive stepping after 2-weeks of perturbation training, particularly in reactive step latency. This single-group, non-randomized clinical trial shows feasibility of reactive balance in PwMS and provides preliminary evidence of effectiveness of this intervention.

INTRODUCTION

People with multiple sclerosis (PwMS) exhibit frequent and debilitating falls and current fall-prevention therapies are inadequate [1]. Reactive balance, including quick protective steps after a slip or trip, are critical mechanisms of fall avoidance [2]. Protective steps are impacted by MS [3–5] and deficiencies in protective steps relate to increased fall frequency [6]. Therefore, it is important to understand whether protective steps can be improved in PwMS.

Despite research characterizing protective step improvement in fall-risk populations such as older adults and people with Parkinson’s disease (PD) [7–9], less work has focused on PwMS. Two recent investigations have assessed the effects of perturbation training on PwMS, overall showing that PwMS can improve reactive balance through perturbation training. Monjezi et al. 2022 [10] randomly assigned 34 PwMS to 12 sessions of walking and standing training either with or without balance-perturbing waist-pulls. Latency of responses after in-place (i.e., non-stepping) support-surface translations just before and after training were improved in the balance-perturbing group compared to the control group (mean improvement=16ms). Rates of falls 3 months after training were also assessed, however difference in fall rates between groups was not statistically significant. In 2023, Okubo and colleagues conducted a randomized controlled trial, in which 30 PwMS were randomized into either a walking trips & slip group, or a sham group (stepping over foam obstacles, n=16) [11]. All participants were exposed to unpredictable slips and trips at post-assessment. The intervention group exhibited improved dynamic stability and had fewer losses of balance at the post-intervention assessment compared to the control group. However, as with Monjezi et al., retention of improvement was not assessed. Finally, our group showed that 1-day of in-place perturbation training could improve reactive steps, and those improvements were retained 1 day later [12]. These studies provide key information demonstrating the ability of PwMS to improve several aspects of reactive balance through perturbation training. However, gaps in knowledge remain, including 1) whether extended, from-stance perturbation training also improves reactive balance, 2) the long-term retention of these improvements, and 3) the impact of standing slip training on falls in the home.

The purpose of the current study was to improve our understanding of the immediate and retained effects of task-specific perturbation training on reactive balance, and of feasibility (i.e., adverse events and training session completion rates) of reactive balance training. We also sought to provide preliminary information regarding possible impacts on daily-life falls. Based on previous work [10–13], we hypothesized that PwMS could improve and retain improvements in reactive balance through training.

METHODS

Participants

PwMS were recruited in the larger Phoenix and Flagstaff area via physician referral, electronic medical records in the Phoenix VA Medical Center, discussions with support groups, and the National MS Society. Inclusion criteria were 18–80 years of age, no known neurological conditions other than MS, and an ability to stand, unaided, for at least 5 minutes. Participants were “at risk for falls,” defined by: 1) one or more falls in the past year, 2) an Activities-Specific Balance Confidence (ABC) Scale score <80 [14], or 3) a Dynamic Gait Index (DGI) <19 [15]. Participants gave written informed consent to participate, and the protocol was approved by Arizona State University and Phoenix VA Medical Center IRBs. This study was registered as “Protective Step Training & MS” at ClinicalTrials.gov (NCT03551665).

Protocol

A within-subject, multiple-baseline design was used. This design allowed us to assess exposure effects (between which no training occurred), immediate training effects (just before and after training), and retention effects (the consistency of effects over time) [16].

At baseline, clinical, mobility, and cognitive tests were conducted, including the European Database for Multiple Sclerosis (EDMUS), Patient Determined Disease Steps (PDDS), Activities of Balance Confidence Scale (ABC), Mini Balance Evaluation Systems Test (MiniBEST) [17, 18], Montreal Cognitive Assessment (MoCA) [19], and Symbol Digit Modality Test. (SDMT) [20]. Falls were tracked via a falls calendar for 8 weeks before and after the intervention. Falls were defined as “any unexpected event that results in you ending up on the ground, floor, or any lower surface.” [21]

Participants’ reactive stepping was assessed at 4 timepoints: Baseline 1 and 2 (B1, B2), 2 weeks apart, post 1 (immediately after training) and post 2, 2 months after post 1, as described in [16]. Balance perturbations were elicited through quick movements of the support surface. Participants stood quietly on a Bertec split-belt instrumented treadmill (Bertec Corporation, Columbus, Ohio). The movement of the ground underfoot was modeled as a ramped protocol: with a 300ms acceleration, followed by a constant speed for 500ms, and a 300ms deceleration [12, 22–24]. The perturbation intensity was varied by changing the acceleration. During the B1 visit, 2 to 3 perturbations were first delivered at different accelerations to identify a speed that elicited a quick reactive step but did not induce a fall. The acceleration identified for each participant was held constant for all assessments. Average (SD) acceleration of assessment perturbations was 1.96m/s2 (0.70) for backward-loss-of-balance perturbations and 2.24 (0.73) for forward-loss-of-balance perturbations. Assessments (B1, B2, P1, P2) consisted first of 8 warmup trials (2 forward, backward, leftward, and rightward) to reduce first-trial startle responses [25]. Then, participants completed 6 (3 forward and 3 backward steps) reactive step perturbations to assess stepping performance. Direction of perturbations was pseudo-randomized to prohibit directional anticipation; however the sequence was the same across the assessment visits. The timing of perturbation delivery was pseudo-random to reduce any temporal anticipation. The following instructions were provided: “Stand still, do not anticipate the perturbation, and do your best to recover balance after the perturbation”. Leftward and rightward reactive stepping (6 trials total) were collected, but are not reported here. For all trials, participants wore a safety harness and a spotter to ensure safety. The length of the tethers from the harness to the upper truss were set to protect against a fall but not assist in stepping.

Participants then completed a 2-week, 6-session perturbation training protocol. The protocol procedure utilized repeated support surface perturbations from stance, delivered by a Bertec treadmill [16]. This training paradigm and assessment schedule was designed as a “task-specific” training program, where participants trained on a task like the one on which they were assessed. The training session consisted of 8 forward and 8 backward reactive step trials, followed by 8 leftward and 8 rightward perturbations (participants stood perpendicular to the belt direction). Participants were exposed to 32 perturbations per session and approximately 192 perturbations across sessions. The forward/backward and leftward/rightward perturbations within each session were separated into 2 blocks of 4 trials. The direction of perturbations was randomized within forward/backward and leftward/rightward blocks. Perturbation dose was chosen to elicit learning while managing fatigue. Perturbation acceleration was adjusted throughout the training. Across each 4-trial block, the number of “falls” into the harness (>10% of body weight applied to the harness, measured by force-transducers) was monitored. If a fall was observed in 1 of 4 trials acceleration was held constant. If no falls occurred (0 of 4 trials), acceleration was increased by 0.2m/s2 in the next block, and if falls occurred in 2 or more if the 4 trials, acceleration was reduced by 0.2m/s2.

Outcomes

Feasibility was assessed as adverse events and percentage of training session completion. Stepping kinematics were assessed with a 14-camera motion capture system (100hz; Motion Analysis Corporation, Santa Rosa, California). Marker clusters were placed bilaterally on the feet, shank, and thigh, and on the sacrum and T-12 vertebrae. Twenty Hz low-pass filters were applied to marker data. Data from force plates were collected at 2000 Hz and were low-pass filtered at 10 Hz.

Kinematic outcomes were calculated via The MotionMonitor xGen (MotionMonitor; Innovative Sports Training, Chicago, Illinois). Primary outcomes were step kinematics after a backward loss of balance. Specifically, we captured the margin of stability (MOS) in the anterior-posterior (AP) direction, step latency, and step length. To calculate MOS, we first calculated the continuous position of the extrapolated center of mass (xCOM [26, 27]) as:

$$\mathit{\text{xCoM}}=\frac{x+v_x}{Wo}$$

Here, x and Vx describe the AP position and velocity, respectively, of the COM. W_o is the eigenfrequency of the inverted pendulum, calculated as:

$$Wo=\sqrt{\frac{g}{L}}$$

Such that g is gravity (9.81 m/s²), and L is leg length. Then, we calculated the distance between the xCOM and base of support (the center of mass of the stepping foot) at the moment of foot contact of the first step [26, 27]. Larger positive MOS values reflect better stability, as the COM is within the base of support. Negative MOS values would require at least 1 additional reactive step.

We also calculated first step length as the distance (AP) between the feet at the moment of first foot contact). First step latency was defined as time between perturbation onset (initiation of treadmill movement) and step onset (movement of either foot relative to the treadmill belt).

Statistics and power

Statistical analysis was performed in R (version 4.4.0). Separate linear mixed effects (LME) models were run for each outcome (MOS, step latency, step length). Pre-planned contrasts were run within these 3 models to determine the effect of exposure (B1-B2), immediate training (B2-P1), and retention (B2-P2). Age and disease severity (measured via the EDMUS) were included as covariates in the models. LME models were used to model subject level random effects, and are robust to missing data [28].

A Wilcoxon signed-rank test compared number of falls in the 2 months before and after training.

Our a-priori primary variable of interest was margin of stability (MOS) during backward stepping (with step length and latency as secondary outcomes). When designing this study, MOS data were only available from 1-day, from-stance perturbation training data in people with PD. These data showed an effect size of short-term perturbation training on backwards-stepping MOS in people with PD to be 0.81, or a change in MOS after training of approximately 4 cm [29], suggesting a sample of 18 participants could detect changes in MOS with an alpha = 0.05 and 90% power.

RESULTS

A Consort diagram illustrating participant completion is shown in Figure 1. Of the 38 individuals that enrolled, 27 completed training and P1 assessments, and 20 completed the P2 (retention) visit. On average, participants completed 96.9% (SD: 9.29) of their training dose. No adverse events were recorded, although some participants reported feeling anxious about the slip tests despite the harness. Participant anxiety did not contribute to study drop out.

Figure 1:

Consort diagram

Supplemental Table shows participant characteristics, including B1 stepping outcomes, between those completing P1 and P2 (follow-up visits). Individuals who completed only P1 exhibited lower (worse) MiniBEST scores and later step latency at B1 (p=0.003 and p=0.001, respectively).

Figure 2 shows reactive balance outcomes at each timepoint (B1, B2, P1, and P2) for both backward stepping (primary outcome) and forward stepping. Model results of exposure (B1-B2), immediate training (B2-P1), and retained improvements (B2-P2) are detailed in Table 2 (backward stepping) and Table 3 (forward stepping). Mean stepping outcome values across the 4 timepoints are shown in Table 4.

Figure 2:

Stepping outcomes across all timepoints for forward (f) and backward (b) stepping outcomes. Margin of stability in the AP direction (mosap), first step latency, and first step length are shown.

Table 2:

Model outputs for each backward reactive stepping outcome (Margin of stability, Step Latency, Step Length)

model	Contrast	Estimate	SE	df	T value	P value	Lower CI	Upper CI
Margin of stability (m)	Base1 - Basel2	−0.007	0.014	71	−0.523	0.603	−0.034	0.020
	Base2 - Post1	−0.029	0.014	71	−2.134	0.036	−0.056	−0.002
	Base2 - Post2	−0.021	0.015	71	−1.404	0.165	−0.051	0.009
Step Latency (s)	Base1 - Basel2	−0.003	0.016	71	−0.207	0.836	−0.035	0.029
	Base2 - Post1	0.041	0.016	71	2.575	0.012	0.009	0.073
	Base2 - Post2	0.039	0.018	71	2.176	0.033	0.003	0.074
Step Length (m)	Base1 - Basel2	−0.006	0.017	71	−0.350	0.727	−0.039	0.027
	Base2 - Post1	−0.026	0.017	71	−1.579	0.119	−0.059	0.007
	Base2 - Post2	−0.020	0.018	71	−1.106	0.273	−0.057	0.016

Table 3:

Model outputs for each forward reactive stepping outcome (Margin of stability, Step Latency, Step Length)

model	Contrast	Estimate	SE	df	T value	P value	Lower CI	Upper CI
Margin of stability (m)	Base1 - Base2	0.007	0.009	70	0.766	0.446	−0.011	0.026
	Base2 - Post1	−0.019	0.009	70	−2.068	0.042	−0.038	−0.001
	Base2 - Post2	0.000	0.011	70	−0.031	0.975	−0.021	0.021
Step Latency (s)	Base1 - Base2	0.010	0.016	70	0.640	0.524	−0.021	0.041
	Base2 - Post1	0.009	0.016	70	0.573	0.569	−0.022	0.040
	Base2 - Post2	−0.036	0.018	70	−2.077	0.042	−0.071	−0.001
Step Length (m)	Base1 - Base2	0.021	0.018	70	1.182	0.241	−0.014	0.056
	Base2 - Post1	−0.024	0.018	70	−1.360	0.178	−0.059	0.011
	Base2 - Post2	−0.015	0.020	70	−0.743	0.460	−0.054	0.025

Table 4:

Stepping outcomes

Timepoint	B1 (n=27)	B2 (n=27)	Training	P1 (n=27)	P2 (n=20)*
Backward Stepping
Margin of Stability (cm)	−7.46 (10.13)	−6.75 (9.73)		−3.85 (6.90)	−3.79 (7.53)
Step Length (cm)	21.15 (11.74)	21.73 (10.69)		24.36 (10.44)	25.54 (11.22)
Step Latency (ms)	298.58 (128.67)	301.91 (148.17)		260.49 (88.24)	225.83 (54.52)
Forward Stepping
MOS (cm)	12.71 (5.95)	12.00 (5.84)		13.93 (5.02)	12.40 (6.48)
Step Length (cm)	37.43 (14.33)	35.36 (12.83)		37.75 (13.50)	38.28 (14.73)
Step Latency (ms)	313.15 (69.09)	303.15 (66.80)		294.20 (55.61)	330.53 (116.25)

B1- baseline assessment 1; B2- baseline assessment 2; P1- post assessment 1, P2- post assessment 2;

^*Forward Stepping P2 n=19

Backward reactive stepping

Across all 3 outcomes (MOS, step latency, and step length), no significant effect of exposure was observed (B1-B2 contrast; p’s≥0.603). Statistically significant improvements in immediate learning (B2-P1) were observed for both MOS (increased after training; Estimate [CI]= 0.029 [−0.056, −0.002], p=0.036) and step latency (reduced after training; Estimate [CI]= 0.041 [0.009, −0.073], p=0.012; see Tables 2 & 4, Figure 2). No significant improvement (increase) in step length was observed immediately after training (Estimate [CI]= −0.026 [−0.059, 0.007], p=0.119). Improvements in step latency remained statistically different 2 months after training (B2-P2 contrast; Estimate [CI]= 0.039 [0.003, −0.074], p=0.033). There was no statistically significant retention of MOS or reactive step length.

Forward reactive stepping

Across all 3 outcomes (MOS, step latency, and step length), no effect of exposure (B1-B2 contrast; p’s≥0.241) was observed. A significant improvement in MOS immediately after training (B2-P1) was observed (Estimate [CI]= −0.036 [−0.071, −0.001], p=0.042). Step latency showed a statistically significant increase from B2-P2 (Estimate [CI]= −0.019 [−0.038, −0.001], p=0.042; See Table 3, Figure 2).

Falls

Of the 27 people with pre- & post-falls data, 15 reported no change in falls, 7 reported a reduction in falls, and 5 reported an increase in falls (Figure 3). The average number of falls was reduced from 0.70 (1.51) to 0.41 (0.75) pre- to post-training (Table 5). The change in falls pre-to post-training was not statistically significant (p=0.266).

Figure 3:

Change in prospectively captured falls from the 2 months prior to training to the 2 months after training.

Table 5:

Mean (SD) of falls before and after training

N	27
Pre-Train Falls	0.70 (1.51)
Post-Train Falls	0.41 (0.75)
Change in Falls	−0.30 (1.17)

DISCUSSION

PwMS who completed a 2-week, from-stance perturbation training protocol reported no adverse events (although some expressed anxiety with repeated perturbations) and completed 96.9% of scheduled trainings. Further, they showed immediate improvements in MOS and step latency, with retention of step latency improvements (a domain particularly impacted in MS) 2 months post-training. Although the average number of falls decreased in the two months following training compared to the two months prior, this change was not statistically significant. This work, along with previous reports shows 1) the feasibility of reactive balance training in PwMS, 2) provides preliminary data supporting the ability of PwMS to improve reactive balance through training, and 3) extends previous work to note possible retention of these effects. However, given the lack of randomized control group, this evidence should be considered with caution. Further, additional studies will be necessary to determine the degree to which these interventions can improve falls.

The improvement in reactive stepping is consistent with recent clinical trials assessing the impact of walking perturbations in PwMS. These reports showed that slips and trips during walking [11], waist-pulls during walking and standing [10], and from-stance, support surface translations [12] resulted in immediate improvements in reactive balance in PwMS. We extend previous literature in 2 important ways. First, we measured the retained improvement in reactive balance in PwMS. We observed PwMS to retain improvements in step latency. The retained improvements in reactive step latency are encouraging as temporal aspects of reactive balance are particularly impacted in PwMS [3, 5], and may be related to falls [30, 31]. Second, we deployed discrete, from-stance, support surface perturbations. While reactive balance responses from-stance and during walking likely rely on similar neural networks, there are key differences. For example, slips and trips delivered during walking allow participants to have robust proactive biomechanical changes, such that when they enter the same environment in which they slipped previously, they walk with altered gait (e.g., shorter, wider steps) [32–34]. Alternatively, from-stance perturbations provide more limited (albeit non-zero) opportunities for pro-active, biomechanical adaptations to improve perturbations [35]. Previous work has indicated that perturbations released in different circumstances (e.g., treadmill vs. overground), may not elicit similar responses [36]. Given the numerous types of losses of balance one can experience, identification of training that is generalizable remains an unmet need.

Our findings of retained improvements in reactive stepping in PwMS are partially consistent with a recently published intervention in people with PD [16]. Over a similar 2 week, from-stance reactive step training, people with PD exhibited retained improvements in MOS, and improvements, albeit less robust, in step length. The improvements in step length in people with PD, as opposed to the step latency improvements in PwMS, may be due to the relative deficits observed across populations. Specifically, people with PD exhibit relatively more spatial deficits and MS more temporal deficits in reactive balance compared to age-matched controls[3, 5, 29]. Further, it is notable that although the effects on MOS in the current study were not statistically significant, the overall magnitude of effect sizes are consistent across studies, and the somewhat smaller effect sizes reported here may be related to the reduced sample.

It is notable that those individuals who did not complete P2 (follow-up) exhibited better balance (measured via the MiniBEST) and took faster reactive steps at B1 (Supplemental Table 1). As shown in Figure 1, the majority of dropouts from P1 to P2 was because we completed a wave of training just before the COVID-19 outbreak, and P2 visits were scheduled after travel restrictions were implemented. The somewhat less affected nature of participants recruited after COVID-19 restrictions may have been due to more severe participants being less willing or able to participate in research studies of this sort after travel restrictions were lifted. Aside from the lower power in the retention analyses, interpretation of our statistical contrasts of B2-P1 (immediate learning) and B2-P2 (retained changes) are unlikely to be impacted by these dropouts. However, it is important to consider these group differences for future work.

The moderate to low effect of perturbation training on daily life falls in PwMS in the current and previous [10] manuscripts should be considered preliminary. In the current report, we acknowledge that a 2-month assessment in 27 participants is underpowered to identify an effect of perturbation training in PwMS. Provision of these data establish preliminary effect sizes for future studies. Notably, the average number of falls after training was less than pre-training levels, showing potential promise for interventions of this sort to impact falls. Furthermore, to most effectively reduce falls, a comprehensive rehabilitative intervention, focused on many clinically targetable predictors of falls should be undertaken. For example, Huoang at al 2014 showed many outcomes that predict falls in PwMS including sway, leaning balance, choice stepping reaction time, walking speed, and executive function, among others, and should be considered in the development of rehabilitative interventions [37]. Integration of reactive balance into such rehabilitative approaches for PwMS is not fully known and larger, randomized clinical trials on perturbation training are necessary to better answer this question. However, results of the current and recent studies are promising, and show that, at minimum, reactive balance (which are related to falls [6, 30, 31, 38–42], are likely modifiable through training, making them a possible target for such rehabilitative interventions.

In general, changes in performance in forward reactive stepping were less robust than backward stepping. However, some changes were observed. First, we observed a significant forward-stepping improvement in MOS immediately after training, although this change was not retained. Interestingly, during forward reactive stepping, step latency remained stable after training but was significantly later 2 months after training compared to pre-training. The increased step latency after a forward loss of balance is intriguing. Previous work has indicated that forward and backward reactive stepping may be controlled by different neural mechanisms. Specifically, while first exposures to perturbations may result in a “first trial” startle effect [25], after repeated exposures, forward perturbations may elicit less neural circuitry associated with startle effects [43]. Startling stimuli can hasten large-scale movements [44]. As such, repeated forward perturbations may result in a diminished startle effect, and longer step latencies, which could simultaneously be more effective. However, further study is necessary to test this prediction. Importantly, and unlike changes in backward stepping, the forward reactive stepping changes were not robust across timepoints. For example, the immediate effect of training on forward reactive stepping was to hasten steps (albeit non-significantly; see Figure 2). The lack of consistency across the B2-P1 and B2-P2 timepoints in this outcome calls into question the robustness of the finding and therefore should be considered with caution.

We did not observe an exposure effect for any of the reactive balance outcomes. That is, the approximately 12 balance perturbations delivered at Baseline 1 to assess reactive balance did not result in a statistically significant change in performance at Baseline 2 (2 weeks post B1). Previous work in older adults noted that even a few walking slips may influence future gait performance [32–34]. These partially conflicting results may be due to the different populations studied in the two experiments, such that PwMS may take more exposure or training to see adaptation (although preliminary work has shown PwMS to exhibit improvements in reactive stepping through 1-day of training [12, 45]). Therefore, an alternative explanation for these partially conflicting findings may be the type of perturbation delivered. As noted above, perturbations delivered during walking often result in proactive biomechanical changes, while proactive responses during from-stance perturbations are less pronounced [35]. As such, from-stance perturbations may require more reactive responses, increasing the training dose necessary to exact a change in performance. Although both types of losses of balance are likely important for maintaining balance in the face of external perturbations, these differences are worth considering when interpreting current results.

Limitations

Several limitations should be noted. First, this was not a randomized clinical trial. Rather, our within-subject design measured the effect of perturbation exposure via multiple baselines and timepoints. As such, well-powered, randomized clinical trials, with longer falls and retention follow-up will be necessary to confirm and extend the current findings. Next, 7 participants that completed training and P1 testing were not able to complete P2 (retention) testing. This was due primarily to an interruption of the study due to COVID-19, where participants were unable to attend the P2 collection, reducing power to assess retention. Third, we deployed a task-specific, laboratory-based intervention to understand the degree to which PwMS could improve stepping in a controlled environment. Ecologically valid and clinically deployable perturbation interventions for PwMS should be conducted.

Conclusions

Results show the feasibility of delivery of a 2-week, from-stance reactive balance intervention to PwMS. Fruther, we show preliminary evidence of this intervention, as PwMS were able to improve the overall performance (MOS) and step latency, of reactive steps. Improvements in reactive step latency were retained 2 months later. These results are consistent with, and extend, previous work showing the ability of PwMS to improve reactive balance through perturbation training. Additional work is necessary to establish whether in-clinic interventions improve stepping and reduce falls in PwMS.

Table 1:

Subject Characteristics

N (female)	27 (16)
Age (mean (SD))	56.49 (13.97)
Diagnosis (%)
Relapsing remitting	23 (85.2)
Primary progressive	3 (11.1)
Secondary progressive	1 (3.7)
ABC (mean (SD))	70.72 (11.83)
MiniBEST (mean (SD))	19.62 (5.38)
MoCA (mean (SD))	26.37 (2.47)
SDMT(mean (SD))	42.44 (10.33)
Fall in prev. year = Yes (%)	16 (59.3)
GDS (mean (SD))	3.78 (3.12)
EDMUS (%)
2	3 (11.1)
3	3 (11.1)
4	7 (25.9)
5	6 (22.2)
6	5 (18.5)
6.5	2 (7.4)
7	1 (3.7)
PDDS (mean (SD))	2.69 (1.85)
FSEQ (mean (SD))	24.17 (25.04)

ABC- Activities of Balance Confidence Sale; MiniBEST- Mini Balance Evaluation Systems Test; MoCA- Montreal Cognitive Assessment; SDMT- Symbol Digit Modality Test; GDS- Geriatric Depression Scale; EDMUS- European Database of MS; PDDS- Patient Determined Disease Steps; FSEQ- Falls Self Efficacy Questionnaire