Proprioception and muscle performance unchanged by in-home step training in multiple sclerosis: secondary outcomes analysis

Trial design

This study was an assessor-blinded, parallel, randomised controlled trial. Participants were allocated to the intervention or control arm in a 1:1 ratio. The study was registered with the main iFIMS trial prior to recruitment of the first participant (ANZCTR trial registration number ACTRN12616001053415). Protocols and statistical analysis plans were publicly registered with Open Science Framework prior to data analysis (Djajadikarta, 2020; Menant et al., 2020) and the study was reported in line with CONSORT guidelines (CONSORT checklist: Supplemental 1) (Schulz et al., 2010).

Participants

This trial was approved by the Human Research Ethics Committees of South Eastern Sydney Local Health District (HREC 14/312) and the University of New South Wales (HREC number: HC14211) and was conducted according to the Declaration of Helsinki. All participants provided written, informed consent prior to their involvement.

Participants were included if they: were community-dwelling adults (aged ≥ 18 years), had been diagnosed with MS, had an Expanded Disability Status Scale (EDSS) (Kurtzke, 1983) score between 2.0–6.0, had no apparent cognitive impairment (i.e., were able to understand and follow study instructions), had no exacerbation of MS in the past 30 days (with or without disease-modifying drugs), were able to perform the choice stepping reaction time test (Tijsma et al., 2017), were able to walk 10 m without a bilateral mobility aid, had no existing conditions that prevented exercise, and had not been advised by a medical doctor to avoid exercise. Ability to perform the choice stepping reaction time test ensured that participants were capable of undertaking the intervention. All participants recruited to the main iFIMS trial were invited to participate in this sub-study until recruitment was complete. Prior to randomisation in the main iFIMS trial, participants were invited to participate in this sub-study. Sixty-six people agreed to participate and were enrolled, these people completed the iFIMS intervention but also attended an additional pre- and post-intervention assessment as part of this sub-study. Baseline data from 30 participants were previously published to compare proprioception and motor function in people with multiple sclerosis with age- and sex-matched healthy controls (Djajadikarta et al., 2020).

Participant characteristics

Prior to randomisation, information was obtained on participants’ demographics and type of multiple sclerosis. Average total weekly physical activity (planned and incidental) over the 3-months prior to baseline was measured using the Incidental and Planned Exercise Questionnaire (IPEQ; App Store; Merom et al., 2014). The authors have permission from the copyright holders to use this instrument.

Experimental protocol

Full details of the iFIMS main study’s intervention, outcome measures, setup, testing procedures, and data processing for the three balance and gait measures are reported in detail elsewhere (Hoang et al., 2024). We report them briefly here.

Interventions

The smart±step intervention involved playing home-based step training exergames. A research assistant installed the smart±step system in each participant’s home, which consisted of a custom-made step mat and a small computer connected to a television or a monitor screen. Participants stood and stepped on the mat to train balance, stepping ability, and cognitive function. Exercise intensity and type were selected by participants as they progressed through six game difficulty levels from ‘Easy’ to ‘Expert’. Participants were instructed to start on an easy level and when ready, progress to more challenging levels. They were encouraged to try to beat their highest score. Goal setting and barriers to training were discussed with participants to encourage compliance and exercise progression. There were eight exergame options available. To ensure that participants played games targeting speed and visuospatial processing, they were required to play two core games each day: ‘Stepmania’, where players stepped on (or avoided) floating arrows with precise timing, and ‘Brick Stacker’, a tile puzzle game where players stepped on arrows to move and rotate geometric shapes as they moved down the screen, to stack them into complete rows (Sturnieks et al., 2019). After this, the other games were made available (Hoang et al., 2024).

The intervention duration was 6 months and participants were asked to exercise for at least 120 min per week, broken into shorter sessions of 20–30 min. Compliance to intervention was monitored via daily transfer of data from the smart±step system to a computer server. If participants did less than 80 min of exercise per week for two weeks, they were contacted by telephone to encourage compliance to exercise.

The control group received usual care. Participants assigned to this group were instructed to continue their usual activities. Participants in both groups were given an educational booklet on exercise and fall prevention for people with MS as education alone does not typically improve fall rates (Gillespie et al., 2012).

Outcome measures

This sub-study explored effects of the intervention on secondary outcomes of ankle proprioception (movement detection threshold and reaction time) and plantarflexor muscle performance (maximal isometric voluntary torque, voluntary activation using twitch interpolation, twitch torque during electrical stimulation, decrements in voluntary torque, twitch torque and voluntary activation after a sustained isometric contraction, and recovery from these decrements. All testing procedures were performed before and after 6 months of step training or usual care. For both sessions, the same leg was tested, and custom-built chairs were set to the same measurements.

Three outcome measures of balance and gait that were obtained as part of the main trial were included: postural sway during quiet double stance (displacement of the body while standing on a foam mat for 30 s, with the eyes open; Lord, Menz & Tiedemann, 2003), comfortable walking speed (during the 10 Metre Walk Test; 10MWT; Peters, Fritz & Krotish, 2013) and walking endurance (distance walked in the Six Minute Walk Test; 6MWT; ATS Committee on Proficiency Standards for Clinical Pulmonary Function Laboratories, 2002).

Setup

Participants were instructed to wear the same flat, enclosed, low-top shoes for both sessions. All complied except for nine in the intervention group and five in the control group. Of these, seven intervention participants and two control participants wore similar shoe styles between sessions. Participants who wore different shoes were still included in the analysis. The tested limb was the limb identified by the participant as the one more affected by MS (intervention: 15 right leg; control: 15 right leg). A hand-held goniometer (Baseline 360 degrees) was used to measure voluntary ankle range of motion (Table 1) from maximal plantarflexion to maximal dorsiflexion (neutral position = 0°). The goniometer axis was positioned at the medial malleolus, the stationary arm spanned from the medial malleolus to the medial femoral condyle, and the moving arm was parallel to the base of the footwear.

Paired, recording electrodes (20-mm diameter, Ag/AgCl, Conmed, Anaheim, CA, USA) were placed over the tibialis anterior muscle (30-mm, centre-to-centre interelectrode distance) and over the muscle bellies and calcaneal tendon of the plantarflexors (soleus and medial gastrocnemius, using a belly-tendon configuration) (Fig. 1A). Electromyograms from each muscle were amplified (x100–300) and filtered at 16–1,000 Hz (CED 1902 amplifier; Cambridge Electronic Design (CED), Cambridge, UK). During proprioceptive testing, footplate position was recorded using a potentiometer (linearity of  ± 0.25%; Vishay Intertechnology, Malvern, PA, USA) and plantarflexor torque was recorded using a load cell underneath the footplate (Xtran S1W, Applied Measurement Australia, Baywater, Australia). During motor performance testing in a different custom-built device, plantarflexor torque was again recorded using a load cell (Xtran S1W). Electromyograms, torque, and footplate position were sampled at 2,000 Hz (CED Micro 1401-3 converter) and recorded on computer for analysis using Spike2 software (CED). Position changes of 0.01 degree could be identified from the recorded signal.

Proprioception testing procedures

Participants sat with the foot resting passively on a motorised footplate and the knee straight (Fig. 1A). The footplate position was initially set at midway between voluntary plantarflexion and dorsiflexion. The footplate moved the foot in dorsiflexion-plantarflexion at 13°/s and was controlled by a purpose-built LabVIEW program (LabVIEW v8.2, National Instruments, Austin, TX, USA). A table was positioned over the knees to conceal the legs and feet.

Movement detection threshold was measured using a double-staircase method (Cornsweet, 1962) to assess the smallest passive movement about the ankle that could be detected. Two staircase series were run concurrently, with steps of footplate rotation amplitude in each series alternating in a 1:1 ratio (see Fig. 1B for example). Before each step, the footplate was repeatedly moved up and down several times in a set pattern (a ‘waggle’), so that the movement history and muscle spindle sensitivity before each step was consistent (Proske, Tsay & Allen, 2014). For each step, the footplate randomly moved the foot in dorsiflexion or plantarflexion. Participants were instructed to verbally identify the direction of movement as ‘up’, ‘down’ or ‘I don’t know’. Responses were judged as correct when the reported movement direction matched the direction imposed by the footplate (dorsiflexion = ‘up’, plantarflexion = ‘down’). Participants received no feedback on whether or not responses were correct. For each staircase, a correct response resulted in a 50% decrease in the amplitude of movement of the following step in that staircase, and an incorrect/unidentified response resulted in a 50% increase. Each staircase ended after five reversal points (when the participant’s response changed from correct to incorrect or vice versa).

Reaction time was measured as the shortest time a participant took to respond to a movement at the ankle. The footplate passively moved the foot in dorsiflexion or plantarflexion at random, and participants actively plantarflexed into the footplate as soon as movement was detected. The amplitude of footplate rotation was twice the participant’s detection threshold, to ensure participants had no uncertainty that movement had occurred. The sound of the footplate motor was masked by noise played to participants through headphones. To alert participants that the footplate was about to rotate, a loud “beep” sounded, and the footplate rotated randomly after 520–1,500 ms. Five practice trials were provided, then 25 trials were performed with 5-s intervals between trials.

Muscle performance testing procedures

On a different testing device, participants were seated with the knee straight, and the foot strapped onto a footplate set at 45° from horizontal. Stimulating electrodes (10-mm diameter, Ag/AgCl, Ambu A/S, Ballerup, Denmark) were placed over the tibial nerve in the popliteal fossa (Fig. 1A) and over the medial femoral condyle. Optimal electrode position over the tibial nerve was first determined using low intensity stimulation (10–20 mA; 1-ms duration; DS7A, Digitimer, Garden City, UK) delivered through a custom-built hand probe. The probe was repositioned over the approximate location of the tibial nerve until the largest plantarflexor response was found. A self-adhesive electrode was attached over the final position of the probe. Stimulation intensity was then progressively increased until maximal plantarflexion twitch torque and maximal peak-to-peak amplitude of the muscle compound action potentials from soleus and medial gastrocnemius muscles were identified. The stimulation intensity used for testing was 120% of that used to obtain maximal responses.

Participants first practised isometric maximal voluntary contractions (MVCs) of the plantarflexors on the footplate until torque was consistent (within 10%) over 3–4 MVCs. Participants were given verbal encouragement and visual feedback to attain maximal effort. Five MVCs (∼3-s duration) were performed, with 2-min rests between contractions to prevent exercise-related decrements in performance (Fig. 1C). During each MVC, the tibial nerve was stimulated at peak plantarflexion torque to generate a superimposed twitch, and immediately after the MVC to generate a resting twitch.

Participants were then instructed to perform a maximal isometric plantarflexion contraction against the footplate for 2 min to fatigue the plantarflexors. Similar fatiguing tasks have been carried out with other muscles in people with MS and in other patient groups (e.g., Neyroud et al., 2017; Schwid et al., 1999; Steens et al., 2012; Wolkorte, Heersema & Zijdewind, 2016). Ten superimposed twitches were elicited during the MVC (∼13 s apart), and two resting twitches (2–3 s apart) were elicited immediately after the fatiguing contraction.

Finally, five MVCs (∼3-s duration) were performed to investigate recovery of motor outcomes following the 2-min sustained contraction. Participants performed MVCs at 15, 30, 60, 90 and 120 s after the fatiguing contraction. Superimposed and resting twitches were elicited during each of the recovery MVCs.

Protocol deviation

One of our planned proprioception secondary outcomes (ankle joint position sense using a motorised foot plate) was not included in analysis, as the measurement of joint position sense using this device is not a validated technique (Vetter & Mascha, 2017).

Randomisation

Following the baseline assessment, participants were randomly allocated to the intervention or control group using a central web-based randomisation program. Permuted block randomisation (block sizes: 2, 4, 6, 10) using computer-generated random numbers were applied to form two groups of similar size in the main study. All participants were invited to do the sub-study, and recruitment finished when 66 participants agreed to participate.

Blinding

To ensure allocation concealment, the software and schedule for randomisation were developed by individuals external to the study team. All baseline assessments, follow-up assessments, and statistical analyses were conducted by investigators blind to participant group allocation.

Data analysis

Detection threshold values were extracted from Spike2 files using a custom-written CED script. First, a median filter (200-ms window) was applied to smooth the footplate angle signal. Next, the five reversal points in each staircase were identified. The footplate angular movement at each reversal was calculated as the difference between the start position (midway point between voluntary plantarflexion and dorsiflexion) and the final position of the plate. An average of the movement amplitudes at the five reversals was calculated for each staircase, and then a single detection threshold was calculated as the average of both.

Reaction time was measured as the mean time (s) between footplate movement onset and voluntary force production (measured as time at 10% of maximal rising slope of torque increase). Trials that were performed incorrectly were excluded (e.g., if the participant applied force before the footplate was moved). Trials that were >3 standard deviations (SD) from the participant’s mean were also excluded.

Muscle performance outcomes of MVC torque, resting twitch torque and voluntary activation were obtained before (non-fatigued), at the end of a 2-min fatiguing MVC, and during recovery after the fatiguing MVC (Fig. 1C). Non-fatigued muscle performance outcomes were obtained from the largest value across the five non-fatigued MVCs. MVC torque was measured as the mean torque over 200 ms before the superimposed twitch. The amplitude of superimposed and resting twitch torques were measured from twitch onset to peak. Voluntary activation (%) (Herbert & Gandevia, 1999; McKenzie et al., 1992) was calculated as: ${\displaystyle \left(1-\text{superimposed twitch/resting twitch}\right)\times 100.}$

Voluntary activation values that were calculated as negative were set at 0% (n = 2 fatigued MVCs included in the analysis). Negative values occur if the resting twitch is smaller than the superimposed twitch because of non-linearity in the measured relationship between the superimposed twitch and voluntary torque at low torques. Such values reflect very poor voluntary activation. Here, use of 0% rather than the non-physiological negative values altered the fall in voluntary activation over the fatiguing contraction from 85.3 to 76.9% (baseline session, intervention participant with concurrent 85% fall in maximal torque) and from 48.6 to 27.7% (baseline session, control participant with concurrent 35% fall in maximal torque). Voluntary activation calculated with superimposed twitches with negative amplitudes (fall in torque after stimulation) were set at 100% (n = 4 over all MVCs).

Towards the end of the fatiguing contraction, the mean torque over 200 ms before the final superimposed twitch was measured and expressed as a percentage of non-fatigued MVC torque. The mean torque of the two resting twitches immediately after the fatiguing contraction was expressed as a percentage of non-fatigued twitch torque. The mean resting twitch and final superimposed twitch were used to calculate voluntary activation. The changes in MVC torque, resting twitch torque and voluntary activation were expressed as the differences from prior to the fatiguing contraction to the end of the fatiguing contraction.

Time-to-recovery outcomes of MVC torque, twitch torque and voluntary activation were measured from the five recovery MVCs. MVC torque and twitch torque were expressed as percentages of non-fatigued values, and voluntary activation was expressed as the difference from non-fatigued values. For each outcome of each participant, a third order polynomial was fitted over time and used to determine the time at which the outcome recovered relative to the non-fatigued value. MVC torque and twitch torque recovered if they returned to at least 95% of non-fatigued values. Voluntary activation recovered if it returned to within 5% of the non-fatigued value. Third order polynomials were chosen to model time-to-recovery data as they allowed for non-linear relationships. Plots of individual participant raw and fitted data were inspected. Time-to-recovery data were analysed using custom-written scripts in Python v3.8.

Statistical analyses

In total, there were two proprioception outcomes (detection threshold, reaction time), nine muscle performance outcomes (MVC torque, twitch torque and voluntary activation and the same parameters after a fatiguing contraction, and time-to-recovery of those outcomes), and three balance and gait outcomes (postural sway, walking speed, and gait endurance).

Distributions of the data were assessed by visual inspection of histograms and boxplots. Proprioception outcomes were not normally distributed, and potential outliers were present (∼8 for detection threshold, ∼6 for reaction time).

We used statistical tests different to those in the original statistical analysis plan (Djajadikarta, 2020), as they were more appropriate.

Students t tests (or Mann Whitney U test) for independent groups were used to compare baseline demographic data. Robust linear regression was used to determine between-group effects of intervention for all proprioception, muscle performance, and balance and gait outcomes using the -rreg- command in Stata (Verardi & Croux, 2009; Vickers & Altman, 2001). Robust linear regression is weighted such that extreme observations influence the overall effect less, compared to non-extreme observations. Data were analysed by intention-to-treat.

A complier average causal effect (CACE) analysis was conducted to estimate the average effect of treatment in participants who complied with the intervention. Compliance to the intervention was measured as the total amount of step time (h) performed for each participant in the intervention group, expressed as a percentage of total intervention time. Participants in the control group did not have access to the intervention, and so performed 0 h of intervention. The CACE estimates the mean effect of treatment in participants who undertake 100% of the intervention if allocated to the intervention group and 0% of the intervention if allocated to the control group (Angrist & Imbens, 1995; Fairhall et al., 2017). The CACE was estimated with instrumental variable regression using the -ivregress- command with the two-stage least squares estimator in Stata. The instrument was the randomly allocated intervention. Compliance to the intervention was entered as a continuous variable.

Time-to-recovery outcomes were analysed using mixed effects parametric survival models with the -mestreg- command in Stata (Cleves et al., 2008). Survival models account for non-normal and right-censored data (i.e., when participants had not recovered after the 120-s period of observation).

Statistical analysis was conducted with IBM SPSS Statistics software (v25, IBM, Armonk, NY, USA) and Stata (v16; StataCorp, College Station, TX, USA). All de-identified data and computer code used for analysis are available from the public repository (http://www.osf.io/gazfw).

Limitations

As a sub-study of a trial aimed at reducing falls through home-based step training, the current study was not specifically designed to improve ankle proprioception or muscle performance. Therefore, the lack of change in these parameters does not necessarily indicate that they cannot be improved by other exercise regimens. Adherence to the goal of 120 min of training per week was low and the main study was ineffective at reducing falls (Hoang et al., 2024). In addition, the number of participants in the sub-study was low. However, the lack of change in ankle proprioceptive and muscle performance is consistent with the secondary outcomes of the main trial which showed no improvement in physical functions like balance, gait, and knee and hip strength (Hoang et al., 2024). Finally, the baseline performance of proprioceptive detection threshold did not differ from that of healthy control participants (Djajadikarta et al., 2020) so a test that revealed more impairment may have offered more chance for improvement with training.