Amplifying walking activity in Parkinson’s disease through autonomous music-based rhythmic auditory stimulation: randomized controlled trial

Abstract

Habitual moderate intensity walking has disease-modifying benefits in Parkinson’s disease (PD). However, the lack of sustainable gait interventions that collectively promote sufficient intensity, daily amount, and quality of walking marks a critical gap in PD rehabilitation. In this randomized controlled trial (clinicaltrials.gov#: NCT05421624, registered on June 6, 2022), we demonstrate the effectiveness of a real-world walking intervention delivered using an autonomous music-based digital rhythmic auditory stimulation (RAS) system. In comparison to an active-control arm (N = 20) of moderately intense brisk walking, the autonomous RAS system used in the experimental arm (N = 21) amplified moderate-to-vigorous walking intensities, increased daily steps, and improved (reduced) gait variability. While regular engagement in real-world walking with or without RAS each cultivated habits for walking, only the RAS intervention yielded a combination of strengthened habits and improved gait outcomes. Findings from this study supported the use of a personalized, autonomous RAS gait intervention that is effective, habit-forming and translatable to real-world walking in individuals with PD.

Introduction

Improving walking is a top priority for enhancing quality of life in people with Parkinson disease (PD)¹. The chronic and degenerative nature of PD requires easily accessible and sustainable interventions that effectively delay or reverse walking disability^2–8. Habitual walking at moderate intensities in sufficient quantities can alleviate motor symptoms in PD and promote improvements in gait and balance^5,6. However, regular engagement in moderate intensity walking can be challenging for people with PD, who typically experience motor (e.g., loss of walking automaticity⁹, bradykinesia) and non-motor (e.g., depression, lack of motivation) impairments. Commonly impaired gait quality mechanics in PD, such as increased variability in stepping patterns, limit walking stability and further contribute to gait disability in PD¹⁰. The development of rehabilitation interventions to address this complex array of neurophysiologic impairments, limited real-world walking performance, and reduced engagement in society is needed to counter gait disability¹¹.

PD research over the last several decades has included a focus on rhythmic auditory stimulation (RAS), a rehabilitation intervention leveraging the potent effects of auditory-motor entrainment to facilitate rhythmicity and stability in motor control during walking^12–14. Auditory-motor entrainment bypasses the internal rhythm deficits related to PD basal ganglia dysfunction¹⁵ by accessing cerebellar pathways involved in movement regulation^16,17. Further, RAS facilitates gait-related sensory-motor processing in the parietal areas¹⁸. RAS improves gait quality by increasing stride length^13,19, improving the timing of leg muscle activation¹³, and reducing step variability^19,20, which together enable faster walking speeds^13,19 and moderate intensity walking²¹. While clinical studies have supported RAS-based intervention^13,22–24, its translation to real-world environments has been limited²⁵ by barriers associated with its traditionally open-loop approach. In this approach, a trained individual is required to select and manually adjust isochronous (i.e. fixed tempo) rhythmic cues based on their evaluation of user gait characteristics. Further, the open-loop approach requires high levels of user vigilance^17,26,27, is prone to error accumulation²⁸, and fails to adapt to changing contexts and environments. Recent advances in user-worn digital technology have spawned the development of closed-loop (personalized) RAS systems, which autonomous deliver individualized cues based on user’s real-time gait performance^21,29–31. Closed-loop RAS fosters natural and stable entrainment^28,32,33, thereby making them potentially suitable for sustaining moderate-intensity walking and improving gait quality³³. However, the potential of closed-loop RAS interventions for reducing walking disability in PD has received only preliminary scientific scrutiny to date^21,31. Moreover, the effectiveness of closed-loop RAS used in real-world applications directed at increasing daily walking activity (e.g. daily walking intensity and amount) has not been examined.

Our research group recently investigated the feasibility and proof-of-concept of delivering a real-world walking intervention via an autonomous, closed-loop, music-based digital RAS system (MR-005, MedRhythms, Inc, Portland, ME, USA) designed for people with PD²¹. The results from the single-arm pilot study provided preliminary evidence supporting the potential benefits of the intervention for improving both real-world walking and gait quality during RAS-cued conditions²¹. Approaches in adopting behaviors associated with new interventions such as community-based walking with autonomous RAS include identifying and targeting modifiable determinants of action³⁴. Habits, as defined in psychology, are behavior patterns that are learned through context-dependent repetition. That is, through repeated performance, the link between context and behavior is reinforced leading to more automatic habitual responses³⁵. The cue-enriched delivery of music-based RAS has potential for harnessing habits for walking, which could be advantageous in promoting regular and sustainable walking regimens. Habit formation during adoption of new walking regimens has not been examined in PD.

Building on our prior work, the current study was an 8-week prospective, assessor blinded randomized controlled trial with experimental and active control arms. It was designed to more rigorously examine the effects of autonomous RAS system for increasing daily moderate intensity walking, daily amount of walking, and improving stride time variability (STV, an indicator of walking automaticity and disability^10,36,37) in persons with PD. We also were interested in examining the habit-strengthening potential of walking with autonomous RAS system for promoting regular moderate intensity walking. Participants in both arms were first instructed to engage in six weeks of routine real-world walking in the community. For participants in the experimental “Amped-PD” group, the walking intervention was delivered via MR-005. For participants in the Active-Control group, no specific delivery modality was used. For the final 2-week “follow-on” period of the trial, both groups engaged in real-world walking as desired, without specific instructions from or monitoring by the research team. We hypothesize that engaging in the Amped-PD intervention compared to the Active-Control intervention will generate larger gains in moderate intensity walking and consequently higher daily steps during the 6-week intervention and 2-week follow-on periods. We further hypothesize that moderate walking intensity and daily steps will return to baseline following the withdrawal of the RAS intervention at study completion. These gains will be accompanied by larger training-related improvements in gait rhythmicity (reduced stride time variability) and stronger habits for walking in favor of the Amped-PD intervention over the Active-Control intervention.

Results

Participants

Fifty-six people with PD were screened for eligibility. Please see CONSORT Diagram (Fig. 1) and CONSORT Checklist (Supplementary Table 1) for details. Forty-four participants were enrolled and randomly assigned to one of the two intervention arms, with equal sample sizes in each. In the Amped-PD group, 21 out of 22 participants completed the full 8-week study (Fig. 2), while one participant dropped out early. In the Active-Control group, 20 out of 22 participants completed the 6-week walking intervention period, while two participants dropped out. Additionally, one participant in the Active-Control group did not complete the final 2-week follow-on period. In summary, the number of participants eligible for analyses was 21 for Amped-PD group and 20 for Active-Control group, with some exclusions per analysis.

Fig. 1. Study CONSORT diagram.

A flow chart summarizing participant enrollment, allocation, follow-up, and analysis, with detail on reasons for exclusion at each stage in the clinical trial.

Fig. 2. Flow diagram of assessment timepoints.

Real-world step assessment (moderate intensity minutes, daily steps) was performed at Baseline, during the intervention, during the follow-on period and at post-program completion. In-lab clinical assessments of gait stride time variability (STV) were performed at Baseline, Post 6-Week Assessment, and Post-8-Week Assessment. Self-reports of habit for “walking for exercise” were obtained at baseline and Post-8-week Assessment.

Both groups were comprised of older adults with idiopathic PD (Hoehn & Yahr Stage 2.2.5 (range 1–3), mild disease severity and relatively low disease burden³⁸ based on the Unified Parkinson Disease Rating Scale (UPDRS) motor score³⁹ and the Parkinson’s Disease Questionnaire-39⁴⁰. Demographic characteristics were similar across groups (Table 1). Both groups were physically and cognitively high functioning, displaying only mild impairments of balance, walking capacity, and walking endurance. Participants in both groups reported moderate confidence in their physical capability to walk at a moderately fast pace on the Self-Efficacy for Walking Duration Scale⁴¹. No changes in medications were noted during the trial.

Table 1.

Baseline demographic and clinical characteristics of participants

	Amped-PD (N = 21)	Active-control (N = 20)
DEMOGRAPHIC:
Age (y)	66.95 ± 8.84	60.65 ± 9.13
Sex (F, M)	11, 10	10, 10
Race/Ethnicity	21: White	19: White, 1: >1 race
	21: Not Hispanic or Latino	20: Not Hispanic or Latino
DBS	0	1
Education
High School Diploma/GED	2	2
Some College	1	2
College Degree (Associates)	3	0
College Degree (Bachelors)	8	5
Post-College	7	11
PD disease burden/Clinical:
UPDRS Part III (0–141 points)	20.71 ± 7.68	22.75 ± 10.60
Modified Hoehn & Yahr
1	0	1
2	15	15
2.5	5	4
3	1	0
PDQ39 – Total (0–100 points)	11.95 ± 8.24	14.10 ± 8.32
Cognition:
Mini-Mental State Exam (Out of 30)	28.71 ± 1.23	28.90 ± 1.12
Walking capacity & balance:
10MWT (m/s)
Comfortable speed	1.17 ± 0.17	1.25 ± 0.17
Fast speed	1.74 ± 0.36	1.80 ± 0.27
6MWT (m)	520 ± 87	546 ± 76
Mini-BESTest (Out Of 28 points)	24.23 ± 2.43	24.70 ± 2.74
Walking confidence:
SEWD (0–100%)	73.76 ± 18.85	70.05 ± 18.34

F Female, M Male, DBS History of undergoing Deep Brain Stimulation, GED General Education Diploma, UPDRS Unified Parkinson Disease Rating Scale, PDQ39 Parkinson’s Disease Questionnaire-39, 10MWT 10-Meter Walk Test, 6MWT 6-Minute Walk Test, SEWD Self-Efficacy of Walking-Duration Scale.

Adherence and safety

Both groups demonstrated excellent adherence during the 6-week walking intervention period (5x/week, 30 min), with the Amped-PD group completing 32 ± 4.62 out of 30 sessions (106.7% adherence) and the Active-Control group completing 30 ± 6.34 out of 30 sessions (100% adherence). When instructed to continue with a 2-week follow-on period (i.e., no prescribed amount or frequency; no research team contact), the Amped-PD group continued to engage in walking with the MR-005 system, completing 10 ± 3.9 sessions. In contrast, the Active-Control group completed modestly fewer (8.4 ± 5.1) walking sessions during the same period.

No adverse events or unanticipated problems occurred in the Amped-PD group. In the Active-Control group, two participants experienced a mildly injurious fall (not requiring medical attention) during a walking session.

Intensity and amount of real-world walking

During both the walking intervention and follow-on periods, the groups differed significantly in terms of daily minutes of moderate intensity walking gained relative to baseline. During the 6-week walking intervention period (Fig. 3A), the Amped-PD group gained 23.38 ± 8.84 min compared to 6.91 ± 18 min gained by the Active-Control group (F(1,34) = 12.89, p = 0.001), which are considered large effects (Cohen’s d effect size: 1.2). Similarly, during the 2-week follow-on period, (Fig. 3A), the Amped-PD group gained 20.85 ± 11.1 6 min compared to 8.0 ± 20.18 min gained by the Active Control group (F(1,34) = 5.91, p = 0.021; Cohen’s d: 0.82). The between-group difference in daily moderate intensity minutes gained disappeared after eight weeks at post-program completion, when both groups returned to baseline levels (F(1,34) = 0.01, p = 0.932).

Fig. 3. Effects of Amped-PD and Active-Control interventions on daily minutes of moderate intensity walking.

a Between-group comparisons on changes in moderate intensity walking. b Within-subject comparisons on moderate intensity walking by group. Red bars refer to Amped-PD. Gray bars refer to Active-Control. Blue dashed line denotes 30 min of moderate intensity walking, consistent with public health recommendations for adults with chronic conditions (e.g. daily 30-min moderate intensity aerobic activity)⁴². Data are presented as mean ± standard deviation.

When within-group changes were examined (Fig. 3B), the Amped PD group demonstrated significant increases in daily minutes of moderate intensity walking (χ²(3) = 46.50, p < 0.001) during both the walking intervention period (32.83 ± 10.14 min, Ζ = 2.05, p < 0.001) and the follow-on period (30.30 ± 11.45 min, Ζ = 1.70, p < 0.001) relative to baseline (9.45 ± 9.39 min). Notably, these increases in moderate intensity walking align with public health physical activity recommendations for adults with chronic conditions (e.g. daily 30-min moderate intensity walking⁴²). The increases, however, were not maintained following study completion when MR-005 was no longer available (Ζ = 0.15, p = 1.00). In contrast, the Active-Control group had no significant within-group changes in daily moderate intensity minutes during the walking intervention period, follow-on period, and after eight weeks (χ²(3) = 4.43, p = 0.218).

In addition to daily minutes of moderate intensity walking, we also examined between-group differences in the percentage of time spent walking at moderate intensity within individual intervention sessions. During the walking intervention period, the Amped-PD group spent 83.11 ± 16.40% of sessions walking at moderate intensity. In contrast, the Active-Control group spent 46.07 ± 35.49% of the time at that level (F(1,30) = 14.96, p < 0.001). Similarly, during the follow-on period, the Amped-PD group spent a greater percentage of sessions walking at moderate intensity (76.63 ± 27.77%) compared to the Active-Control group (37.91 ± 59.69%, F(1,29) = 11.86, p = 0.002)

A similar pattern of between-group and within-group results was observed for daily step counts. During the walking intervention period (Fig. 4A), the Amped-PD group gained significantly more daily steps relative to baseline (3343 ± 1642) compared to an almost negligible 172 ± 4009 steps gained by the Active Control group (F(1,34) = 10.39, p = 0.003), revealing large effect size following the Amped-PD intervention (d: 1.08). During the follow-on period, the between-group difference in daily steps gained approached significance ((F(1,34) = 3.57, p = 0.067, d = 0.63), with the Amped-PD group gaining 2543 ± 2555 steps relative to baseline compared to the Active-Control group gaining 614 ± 2759 steps. The magnitudes of daily step count gains in the Amped-PD group during the intervention and follow-on periods were 1.86-2.44 times a published minimum detectable change (MDC) value of 1366 steps⁴³, signifying that the changes were clinically meaningful. In contrast, the Active-Control group had gains below the MDC threshold during both periods. The between-group differences in daily step count gains during training dissipated at post-program completion (F(1,34) = 1.65, p = 0.207, d = 0.43). Moreover, the Active-Control group had a notable reduction in daily steps at study completion, which exceeded the MDC value.

Fig. 4. Effects of Amped-PD and Active-Control interventions on daily walking based on walking amount (step counts).

a Between-group comparisons on changes in daily step counts. b Within-subject comparisons on daily step counts by group. Red bars refer to Amped-PD. Gray bars refer to Active-Control. Blue dashed line denotes minimum detectable change (MDC) in daily steps relative to baseline⁴³. Values equal or greater than the MDC line signify change in daily steps that is greater than measurement error. Data are presented as mean ± standard deviation.

With regard to within-group changes, the Amped-PD group (Fig. 4B) also demonstrated significant increases in daily steps (χ²(3) = 30.78, p < 0.001) during both the walking intervention period (12155 ± 3127 steps, Ζ = 1.60, p = 0.001) and follow-on period (11346 ± 2655 steps, Ζ = 1.15, p = 0.029) relative to their baseline (8812 ± 3542 steps). The increases were not maintained at the completion of study when MR-005 was no longer available (Ζ = 0.35, p = 1.00). The Active-Control group had no significant within-group changes in daily steps during the walking intervention or follow-on periods, albeit the results approach significance (χ²(3) = 7.73, p = 0.052), which was mainly due to a reduction in daily steps at study completion.

Stride time variability

Between-group and within-group differences on STV were examined in the lab during un-cued walking performance of the 6-Minute Walk Test⁴⁴ at baseline, 6-week, and 8-week clinical assessments. STV was quantified using the coefficient of variation (CoV), expressed as a percentage³⁷. Smaller STV values were presumed to reflect improved gait automaticity^10,36,37 and hence better gait quality.

At the 6-week clinical assessment (Fig. 5A), the Amped-PD group demonstrated a decrease (improvement) in STV relative to baseline (−0.73 ± 1.56 CoV), whereas STV in the Active-Control group increased (+0.30 ± 89 CoV). The between-group difference in STV change was statistically significant (F(1,32) = 5.20, p = 0.029) with moderate effect size (d: 0.79). Further, STV reductions for the Amped-PD group exceeded the minimum clinically important difference (MCID) of −0.67%CoV⁴⁵, thus these improvements are considered clinically meaningful. Similarly, at the 8-week assessment, STV in the Amped PD group decreased (−0.49 ± 1.25 CoV), while STV in the Active-Control group increased (+0.16 ± 1.41 CoV). For this comparison, however, the between-group difference in STV change was not significant (F(1,32) = 2.09, p = 0.158; d: 0.50). When comparing the magnitude of group differences, the Amped-PD group demonstrated considerable reduction in STV compared to the Active-Control group, with a large effect size (Cohen’s d: −0.88) at the 6-week assessment, and a medium effect size (Cohen’s d: −0.59) at the 8-week assessment.

Fig. 5. Effects of Amped-PD and Active-Control interventions on stride time variability (STV).

a Between-group comparisons on changes in STV. b Within-subject comparisons of STV by group. c Individual response analyses of STV. Lower STV values are indicative of improvement in gait variability, while higher STV values are indicative of worse gait variability. Blue dashed lines mark the threshold for minimum clinically important difference (MCID) of −0.67%CoV⁴⁵. Reductions in STV equal to or lower than the MCID are considered clinically meaningful. Data for (a, b) are presented as mean ± standard deviation, and data for (c) are individual participant responses.

There were no significant within-group differences in STV for either group across baseline, 6-week, and 8-week clinical assessments (Amped: χ²(2) = 4.11, p = 0.128; Active-Control: χ²(2) = 1.73, p = 0.420) (Fig. 5B). However, at an individual level (Fig. 5C), 70% of Amped-PD participants demonstrated reduced STV at both time points, whereas fewer than 47% of participants in the Active-Control group had reduced STV at either assessment.

Habit strength of walking

We measured the habit strength of “walking for exercise” as a health behavior using the Self-Report Habit Index (SRHI), which quantified habit strength as a percentage (0 = maximally weak; 100 = maximally strong). Baseline SRHI scores reflected similar degrees of moderate habit strength in both groups (Amped-PD group: 59.40 ± 26.57%; Active Control group: 44.08 ± 29.79%,44.08 ± 29.79%, (F(1,38) = 2.96, p = 0.096). After eight weeks (Fig. 6A), both groups demonstrated comparable increases in habit strength relative to baseline, with Amped-PD gaining 14.84 ± 19.09 SRHI percentage points and Active-Control gaining 13.77 ± 21.22 percentage points (F(1,38) = 0.03, p = 0.868). Significant habit strengthening (relative to individual baselines) occurred within each group (Fig. 6B), with the Amped-PD group increasing SRHI score to 74.25 ± 20.72% (t₂₀ = 3.56, p = 0.002) and the Active-Control group increasing SRHI score to 57.85 ± 24.68% (t₁₈ = 2.83, p = 0.006).

Fig. 6. Effects of intervention on Self-Report Habit Index (SRHI).

a Between-group comparisons of change in SRHI scores. b Within-subject comparisons of SRHI scores by group. c Relationship of SRHI and total amount of walking sessions with groups collapsed. Red bars refer to Amped-PD. Gray bars refer to Active-Control. Data for (a, b) are presented as mean ± standard deviation, and data in (c) (green scatter dots) represent each data point across both groups.

Given their comparable adherence rates and improvements in habit strength, we pooled data from both groups and conducted an exploratory analysis of the link between habit strength and engagement in a walking intervention. To do so, we ran a linear regression to examine the relationship between total session count and 8-week SRHI score. The analysis (Fig. 6C) revealed that adherence explained 16% of the variation in habit strength at the completion of intervention (F(1,31) = 5.86, p = 0.022), suggesting that higher engagement in a walking intervention, either with or without using MR-005, was beneficial for habit-building.

Secondary outcomes

After completing the 8-week walking trial, 84% of participants in the Amped-PD group reported favorable changes in overall physical activity, as reflected on the Global Rating of Change scale, in comparison to 58% of participants in the Active-Control group. Similarly, 74% of the Amped-PD group reported improvements in their walking, in contrast to 58% of the Active-Control group. No significant differences were present for other secondary clinical outcomes based on between-group comparisons on changes at 6-week and 8-week assessments relative to baseline (p > 0.05) (see Supplementary Table 2).

To better understand the overall experience of Amped PD participants in the trial, we conducted a thematic analysis of interview user experience data collected during the 8-week assessment. Five themes emerged from the data, spanning behavioral and physical benefits (Fig. 7). Overall, participants were positive about the walking intervention given its perceived physical and functional benefits, as well as its motivational, engaging, and enjoyable nature. Participants also expressed considerations for improvement, which included integration of feedback and encouraging prompts during walking sessions, and the ability to select specific songs instead of music genre.

Fig. 7. User experience with Amped-PD intervention.

Thematic data from interview of participants in Amped-PD group.

Discussion

The lack of accessible and sustainable gait interventions that collectively target intensity, amount, and quality of walking in PD marks a critical gap in the rehabilitation of individuals of PD. The current clinical trial presents evidence of the effectiveness of a walking intervention delivered by an autonomous, closed-loop music-based RAS system. In comparison to the Active-Control intervention, the Amped-PD intervention using the MR-005 system amplified moderate-to-vigorous walking intensities, increased daily steps, and improved (reduced) gait variability. While both interventions cultivated “walking for exercise” habits, only the intervention with the MR-005 system yielded a combination of strengthened habits, amplified walking activity, and improved gait quality. Findings from this study advance our understanding on the potential of technology-driven gait rehabilitation that is safe, personalized, habit-forming, and translatable for community use by people with PD.

Left alone, people with PD experience natural declines in daily walking intensity and amount of walking year after year⁴⁶ at rates that outpace the decline of other walking-related impairments and participation restrictions⁴⁷. Conversely, people with PD who purposefully engage in regular moderate intensity exercise have a better clinical course of PD, including slower declines in postural control and gait functions⁵. In the current study, we capitalized on the benefits of auditory-motor entrainment delivered via a closed-loop music-based RAS intervention to promote increased walking activity in the community. Our findings demonstrated that MR-005 effectively led to significant and substantial improvements in both daily walking intensity and amount during use, indicative of additive benefits on top of customary walking activity. Notably, the Amped-PD group achieved improvements in daily moderate intensity walking that approximated or exceeded 30 min during both the 6-week walking intervention period and the 2-week follow-on period. By extension, this number of daily moderate intensity minutes of walking equated to ≥150 min per week, thus harmonizing with public health guidelines for moderate-to-vigorous physical activity⁴². These increases in real-world walking activity with the autonomous RAS intervention significantly surpassed the effects induced by a closely matched control intervention of brisk walking without RAS. Furthermore, there were no significant changes in daily walking intensity and volume with brisk walking without RAS.

The Amped-PD group clearly engaged in intervention sessions that were substantially more vigorous than in the Active-Control group, where close to 80% of the entire session’s duration was performed at moderate-to-vigorous intensities. In contrast, the Active-Control group walked at moderate intensities less than half the session’s duration. We have shown previously that the MR-005 system was capable of eliciting immediate gait-enhancing benefits during use, improving walking speed, stride length, cadence and variability²¹. Lab-based studies have shown energy cost savings during faster walking, and penalties for walking slower in people with PD⁴⁸. Taken together, the gait-improving benefits of MR-005, along with presumed improvements on gait efficiencies associated with faster walking may enabled access to latent capacities to engage in more vigorous walking. At study completion, the intensity of walking for both groups returned to baseline. In support of our hypothesis, the withdrawal of the autonomous RAS system from the Amped-PD group at study completion resulted in walking intensities returning to baseline levels due to removal of the technology that enabled access to latent capacities for more vigorous walking. This suggests the need for continuous access to the autonomous RAS system for daily use. Further, the ongoing improvements on walking activity while using the autonomous RAS system offer potential application for use as assistive aid especially for users with more gait impairments. It will be important for future studies to examine prolonged (e.g., 6-month) impact of MR-005 use on natural walking activity and how it might compare with standard-of-care walking interventions. The extent to which it might result in a more long-lasting change in walking activity following use has not yet been determined. This issue is especially important, given the degenerative nature of PD, expected natural declines in walking activity without intervention, and the known health benefits of regular moderate intensity walking^5,6.

Stride-to-stride variations in stepping are linked with basal ganglia function, and heightened (worse) stride time variability is believed to be due to loss of walking automaticity^10,36,37. Thus, gait variability serves as an important marker of stability of walking performance⁴⁹ and a measure of the neurodynamics of walking⁵⁰. Importantly, STV is associated with disease severity and disability in PD¹⁰. Following the 6-week walking intervention period, the Amped-PD group demonstrated significant improvement (reduction) in STV during uncued walking, whereas the Active-Control group demonstrated worsened (increased) STV. Similar patterns and trends were observed at the 8-week assessment. The absence of between-group differences in other gait-related secondary outcomes suggested that benefits on gait variability were independent and likely a product of task-specific training with the autonomous closed-loop RAS system. The modest within-group improvements in fast gait speed and stride length following each intervention were likely attenuated by minimal deficits on walking speeds at baseline. Nonetheless, these changes in walking demonstrated a response in the desired direction. Longer interventions are needed to examine each intervention’s comprehensive effects on gait variability and other walking quality metrics.

The resultant improvements in STV during uncued walking at the completion of training provided encouraging and intriguing evidence of motor learning, even in the presence of PD, consistent with other studies that have explored this possibility^51–55. Other studies have demonstrated immediate, short-term carryover effects of RAS leading to improvements in variability only when RAS tempos were faster than baseline walking speeds²⁰. Therefore, our intervention’s delivery of progressively fast tempos in addition to the high volume of practice (i.e. 30 min, 5x/week x 6 weeks of walking intervention) were likely contributors to training-related effects on gait variability. Given that our assessments of variability were performed at the completion of the intervention, studies that examine longer retention and no-intervention periods are needed to examine durability of these effects.

The observed difference in intervention training effects on STV may have been influenced by group differences in baseline STV, which were higher (worse) in the Amped-PD group [4.2 ± 2.4%COV] than in the Active-Control Group [2.9 ± 0.80%COV] (p = 0.043). Indeed, a recent systematic review supported the idea that RAS effects on gait variability can be relatively pronounced in individuals with higher gait variability at baseline^24,49. Tosserams and colleagues²⁴ extended this idea to a variety of gait compensation strategies (including external cueing): individuals with higher baseline variability tend to show larger improvements. While a plausible interpretation of our results, we propose two additional points for consideration. First, the Active-Control group had baseline STV that approximated CoV in other studies involving people with PD (e.g. 2.6 ± 1.0% in the study by Hausdorff et al.²⁰; 2.4 ± 0.6% in the study by Lo et al.⁵⁶). Notably, the study by Hausdorff et al. showed significant RAS-related reductions in CoV from 2.6% to 2.2%²⁰, demonstrating a propensity for improvement. Thus, baseline STV in the Active-Control group, although lower than in the Amped-PD group, was still abnormally high, did not suffer from ceiling effects, and could have been improved if the intervention had involved RAS. Second, although the Amped-PD group demonstrated a significant inverse relationship between baseline STV and training-related effects on STV at 6-weeks (r_s(19) = −0.80, p < 0.001) and 8 weeks (r_s(19) = −0.56, p = 0.01), no such relationships were found in the Active Control group. Thus, baseline STV may have mattered only for the Amped-PD group. Traditional methods for examining confounding baseline effects (e.g., ANCOVA, linear mixed model) were not appropriate for this dataset due to violations of model assumptions. Future studies with larger samples with more balanced gait variability at baseline between groups are needed to more directly examine training-related responses on gait variability adjusted to baseline gait variability.

It is also important to acknowledge the worsening in variability that occurred with the Active-Control group. The results suggest that the technology-free walking program performed by the Active-Control group failed to promote task-specific training on walking rhythmicity, thereby carrying a cost on gait variability. This finding reinforces the need for walking interventions with multi-objective targets that include walking quality (gait variability) together with intensity and amount of walking.

Daily walking activity, a vital component of human behavior, is reduced in PD^46,47,57. Rehabilitation experts recognize the urgent need to design effective PD interventions to promote regular, habitual walking activity^34,35,58 Our study uniquely documented how habits pertaining to ‘walking for exercise’ form in people with PD. We quantified habit strength using the Self-Report Habit Index (SRHI), which is a standardized and reliable measure of habit strength for health behaviors, including physical activity⁵⁸. Habits are automatically recurring behaviors that arise from contextual cues associated with the behavior^35,59. In this study, we posited that the MR-005 system would deliver strong contextual (music-based) cues during walking that would promote stronger habits for walking in favor of the Amped-PD group over the Active-Control group. However, our findings revealed that habits were effectively increased by engaging in real-world walking interventions, whether participants used RAS or not. This outcome may be explained by two considerations. First, in addition to RAS, other elements of the intervention may have emerged as contextual cues for participants’ habit formation. Particularly, both groups used a study diary (tracking log) and received weekly phone check-ins with the physical therapist. The collective effects of these procedures may have served as cue mechanisms that promoted regular engagement in the walking activity, high adherence, high repetition of the behavior, and ultimately a perceived improvement in habit strength in both groups. Second, the relatively brief intervention length may not have been sufficient to reveal between-group differences with habit formation. While the current study shows that walking with autonomous RAS is habit-forming, we expect that longer intervention periods will show divergence in group responses, where stronger associations of walking with autonomous RAS will outweigh effects of other contextual cues (e.g. diary). Nonetheless, our finding on strengthening of habits was very encouraging, given that habit control is principally governed by basal ganglia circuitry, which is compromised in PD⁶⁰. Our study sample included people with mild-to-moderate PD with low disease burden, which suggested that habit formation may be more amenable in earlier stages of PD than in later stages. This evidence extended support for initiating habit-forming gait interventions early in the course PD, to take advantage of the responsiveness to behavioral change approaches that may have a favorable impact on walking activity in subsequent stages of the disease.

In regards to habit and walking outcomes, while the Amped-PD group had an improvement in habit strength at 8 weeks, we saw a drop in walking outcomes at study completion after 8 weeks when the device was withdrawn. This incongruity between walking activity and habit is largely due to withdrawal of RAS system itself which served as an important contextual cue for the walking behavior. Future studies are needed to examine habit and walking activity with long-term use and more continuous access to the RAS device to more robustly examine their association.

To our knowledge, our study was the first to use the SRHI in connection with PD walking interventions. We therefore explored its construct validity by examining the relationship between behavior repetition and habit strength. Our findings demonstrated that 16% of the variance in amount of walking intervention explained by habit formation, consistent with findings of a systematic review on the use of SRHI on health behaviors (i.e. R² ≈ 0.20)⁵⁸. This finding suggested that regular engagement in a walking intervention is an important driver of habit formation. However, it is noteworthy that while strengthening habits were achieved through either intervention, only the Amped-PD intervention resulted in combined improvements in habit strength, amplified real-world walking activity, and improved gait variability. Thus, the critical ingredients of the walking intervention through MR-005 matters. Building walking habits is critical; however, simply engaging in walking (not enhanced by MR-005) is not sufficient to achieve robust changes in clinically-relevant walking activity outcomes.

Both groups demonstrated high adherence during the 6-week walking intervention period. During the 2-week follow-on period, the Amped-PD group maintained high engagement with MR-005, whereas the Active-Control group had a slight reduction in engagement. Our study implemented a relatively short intervention in comparison to usual clinical trials in PD. It is possible that longer intervention durations will reveal more pronounced divergence in adherence rates between groups. Future studies are needed to test this hypothesis.

It is noteworthy that none of the Amped-PD participants experienced any serious adverse events or any fall-related events during 8 weeks of community-based walking at moderate-to-vigorous intensities. In contrast, there were two serious adverse events in the Active-Control intervention involving non-injurious falls during the walking intervention. Specifically, one participant tripped due to a toe catch, and another participant fell due to festinating gait. These occurrences, unfortunately represent common challenges and risks during daily walking in PD. Based on user-reported accounts, Amped-PD participants noted improvements in walking quality such as reduced toe scuffing, longer strides, and prevention of shuffling gait during cued walking. These participant comments align with the observed action of MR-005, which is to improve gait mechanics (quality) during walking²¹. Additionally, MR-005 has an integrated multi-tiered decision-making approach that considers real-time user performance and gait quality to determine appropriate tempo progression (See Methods). Together, these intervention components serve as important safety features that are critical in mitigating some of the inherent risks associated with community walking and gait disability in PD.

Finally, the Amped-PD group perceived the walking intervention to be more helpful in improving overall walking and physical activity than the Active-Control group, based on their higher scores on the Global Rating of Change scale. This rating was complemented by our user experience interview data, revealing a prominent theme related to physical benefits and functional improvements following the Amped-PD intervention. The interview data also revealed a favorable influence on affect, where participants reported enjoyment, motivation, and engagement related to the walking intervention. Positive impact on one’s affect is important in promoting long-term engagement, and enjoyment related to music cues has been shown to even have associated gait improvements in people with PD^19,61. Finally, we also obtained user feedback on technology considerations that would further enhance the user experience. Participants valued the integration of encouraging prompts commonly used by physical therapists in gait rehabilitation, and the ability to select specific songs instead of music genre. These suggestions highlight the value placed on personalized gait interventions. All in all, these data indicated an overall positive user experience on engaging in a community-based walking intervention with the autonomous music-based RAS system.

Altogether, our study findings suggest that music-based RAS facilitates more typical walking patterns in PD and enhancing rhythmicity (reduced variability) that are likely to help individuals sustain walking activities at greater amounts and intensities. The heterogeneity of motor problems across individuals, as well as variations in walking for each individual due to task, environment, and other personal factors critically underscores the need for a personalized gait interventions^24,62. Our study findings offer preliminary evidence in support of the potential of an autonomous, music-based RAS system in providing a personalized approach that is sustainable, effective, and engaging.

Our study has several limitations. The intervention duration utilized in this study was relatively short. Subsequent studies with longer treatment durations are needed to explore its impact on long-term adherence to walking intervention and to examine training-related effects on other gait and clinical outcomes. Next, our study participants included people with PD who were relatively high functioning based on walking capacity and performance. While our results show encouraging findings impacting walking activity, gait quality, and habit formation, future studies should examine the effects of this intervention on participants who have greater levels of gait disability and disease severity to enhance generalization of results.

Methods

Study design

The study was a prospective, single-blinded (assessor), 2-arm randomized controlled trial examining the effects of using MR-005, an investigational, autonomous, closed-loop digital RAS system that uses music as an auditory modality (MedRhythms, Inc, USA) to improve walking in persons with mild-to-moderate PD. We conducted the trial (clinicaltrials.gov registration #: NCT05421624, registered on June 6, 2022) at Boston University from August 2022 to November 2023. All aspects of the study protocol were performed in accordance with the Declaration of Helsinki, and under approval of the Boston University – Charles River Campus Institutional Review Board (#6518). Signed informed consents were obtained prior to study enrollment. The trial (Fig. 1 and 2) was initiated with a phone screen, followed by an in-person clinical screening assessment to determine full eligibility to participate in the study. Enrolled participants first underwent a baseline assessment that included PD-specific assessments, instrumented walking tests, and real-world (naturalistic) monitoring of customary daily walking activity. Participants subsequently engaged in a 6-week walking intervention period (i.e., defined schedule and remotely monitored), either delivered with the MR-005 system (experimental Amped-PD group) or without any specific delivery modality or RAS (Active-Control group). See Intervention section for details on MR-005. Regardless of group, participants wore an activity monitor on select days during this phase of the trial. Following a clinical reassessment, all participants were instructed to engage in a follow-on period (i.e., ad lib) for two additional weeks. The Amped PD group retained possession of MR-005 for use as desired during this period, and real-world walking activity was monitored in all participants on select days. Participants returned for a final post-intervention clinical reassessment, 8 weeks after the baseline assessment. Amped PD participants returned MR-005 to the research team at that time. Following the final clinical assessment, all participants underwent an additional period of real-world activity monitoring. No important changes to study design were implemented after trial commencement.

Participants

Participants were recruited from the Greater Boston area and surrounding New England locales through patient registries at the BU Center for Neurorehabilitation, patient support groups, Parkinson newsletters, flyers, and through word-of-mouth. Participants with idiopathic, typical Parkinson’s disease were included based on the following inclusion criteria: 40–85 years of age, mild to moderate disease severity (Modified Hoehn & Yahr stages 1–3), able to walk independently without physical assistance or an assistive device for at least 10 min, and stable PD medications for at least 2 weeks prior to enrollment. We excluded those who were less than 40 years of age, had atypical Parkinsonism, had moderately or significantly disturbing freezing of gait during daily walking based on the New Freezing of Gait Questionnaire⁶³ (i.e. self-reported scoring of 2: Moderately or 3: Significantly on item 7 which asks “How disturbing are the freezing episodes for your daily walking?), had a history of more than one fall for any reason or a singular fall that was PD-related over the previous 3 months, had a Mini-Mental State Exam score less than 24, had self-reported significant hearing impairment, or who reported the presence of other orthopedic, neurologic, and medical conditions that impacted walking and mobility. Individuals who were receiving ongoing physical therapy for gait rehabilitation at the time of the study and those who were already performing regular walking >3x/week for at least 30 minutes also were excluded.

Randomization and blinding

Upon completing the baseline assessment (i.e., first study visit), participants were randomly assigned to either the Amped-PD group or the Active-Control group. Randomization was accomplished using REDCap-generated sequences of equal sizes between groups^64,65. The allocation list was concealed from researchers, and group allocation was fully automated through REDCap^64,65 randomization module. All clinical assessments (i.e., baseline, 6-week, 8-week) were performed by research physical therapists (NW, TCB) who were blinded to participant group assignment. All intervention-related procedures were managed separately by an unblinded interventionist/physical therapist (JAZ). Participants were not blinded to the intervention and were advised to avoid discussing or disclosing their group assignment with blinded assessors to ensure overall fidelity of blinding.

Intervention of experimental arm: Amped-PD intervention

Participants in the Amped PD group were instructed to use MR-005 during the 6-week walking intervention period (Fig. 8A–C; informed consent obtained for use of participant images in the figure). MR-005 was designed to leverage auditory-motor entrainment to deliver rhythmic auditory stimulation (RAS) and improve walking performance in people with PD. System components included a pair of shoe-worn inertial sensors, a bone-conducting headset, and a hand-held touchscreen device that housed proprietary software including the auditory stimuli (Fig. 8A). The system played popular and familiar music based on each user’s preferred music genre (e.g. classic rock, country, oldies, etc.), with songs proprietarily screened for therapeutic utility by the manufacturer. Participants allocated to the Amped-PD intervention received instruction on how to independently operate the MR-005 system on their first study visit, immediately following their baseline assessment (Fig. 8B). Instructions included proper wear of shoe sensors and headset, device initialization, and charging of system components. Participants were instructed to walk with the MR-005 system on generally level surface suitable for long straight ahead walking, in environments with minimal obstacles as possible (e.g. pedestrian traffic, vehicles, pets, and other safety hazards).

Fig. 8. Overview of MR-005 closed-loop digital RAS system and the Amped-PD Intervention.

a Components of MR-005 autonomous rhythmic auditory stimulation system. b Instructions on independent use of MR-005 system. c Self-managed real-world walking program. Informed consent was obtained for the use of participant images in the figure.

Each session with MR-005 began with an autonomous baseline assessment of the participant’s walking using approximately 20 viable strides as reference to define at what tempo the music should start. Upon successful system calibration, music-based rhythmic cues were introduced, and participants were prompted to “walk to the beat of the music”. The closed-loop rhythmic stimulation (Fig. 8C) was then achieved through continuous relay of real-time gait metric data from the shoe-worn sensors to the software to evaluate user’s entrainment (i.e., alignment of walking to music tempo), gait variability, and gait symmetry. A participant’s successful achievement of entrainment and gait quality criteria referenced to proprietary thresholds prompted the algorithm to progress music tempos by 5% at two-minute intervals. If the system determined that user’s gait did not meet performance criteria, it superimposed an additional rhythmic cue over the music until performance improved. As needed, the system also reduced the tempo to better align with participant’s cadence, thereby enhancing the likelihood that they could “re-entrain”. Once gait performance improved, system algorithms would again progress the music tempo. Walking with rhythmic cueing provided by MR-005 application was automatically capped at 30 minutes

Intervention of control arm: active-control intervention

Participants in the Active-Control intervention were instructed to engage in walking in the community as the Amped PD group, but without using MR-005 or any RAS. They were instructed to walk at self-selected speeds slightly faster than their comfortable speed. Participants were asked to monitor the duration of their walking sessions using their personal watch or digital timer.

Walking intervention and follow-on period

For the first six weeks following baseline activity monitoring, all participants were instructed to engage in one 30-minute walking session/day, five times per week (Fig. 2). The unblinded physical therapist scheduled a phone call once a week for a general check-in on any questions related to the walking intervention, including practical and health-related considerations (e.g., adverse events if any). For the Amped-PD group, any technical questions were addressed and resolved during weekly check-in phone calls with the research team. These were infrequent and did not interfere with the independent use of the device. Following the 6-week walking intervention period and after completing a clinical reassessment, all participants were instructed to engage in the walking intervention as desired for another 2 weeks. During this follow-on period, participants chose their walking doses, (both amount and frequency) without weekly physical therapy phone check-ins. Notably, the Amped-PD group could use MR-005 during this period, if desired. During the walking intervention and follow-on periods, participants in both groups logged their walking sessions (date, time of day) on a paper form, the data from which was used to assess adherence to the walking intervention.

Primary assessments and outcomes

Real-world walking activity was captured using a research-grade step activity monitor (StepWatch 4 Activity Monitor; SAM, Orthocare, WA, USA). The monitor, which captured the number of steps taken during each minute of a 24-hour day, was worn for 4 days at each assessment period as follows: (i) during baseline customary daily activity immediately preceding the start of the community-based walking intervention, (ii) during the first four days of the walking intervention, (iii) during the first four days of the follow-on period, and (iv) during customary activity after completion of the final clinical reassessment (Fig. 2). Manufacturer software for the step activity monitor was used to calculate average daily values for two outcome measures: the number of minutes of moderate intensity walking and total number of steps accumulated. A minute of moderate intensity walking was defined using an established threshold as a minute containing 100 steps or greater^66,67. In addition, the number of moderate intensity minutes occurring during structured and unstructured walking intervention sessions was extracted by mapping SAM data to walking session data contained in participant training logs. The SAM has been validated for use in PD and other neurological disorders^46,68, and 4 days of step monitoring is sufficient to reliably capture naturalistic walking in mild-to-moderate PD^43,69.

Gait variability was assessed based on stride time variability (STV) during in-lab clinical assessments of the 6-Minute Walk Test at baseline, 6 weeks, and 8 weeks. Participants did not receive RAS cues during testing. Participants were instructed to “cover as much distance as possible” by walking back and forth along a 30-m straight walkway for 6 minutes. STV was calculated using the coefficient variation (CoV), expressed as a percentage (%), i.e., 100 X (SD of stride time/average stride time)^20,37. We measured STV with thigh-worn inertial sensors (Dot, Xsens, The Netherlands) using validated procedures and algorithms^70,71 to estimate spatiotemporal metrics of walking. Turns during the 6-Minute Walk Test were excluded in the analyses in order to assess the intrinsic dynamics of straight-ahead walking. This was accomplished by applying a median filter to eliminate strides during turns, which were outliers relative to the median¹⁰.

The strength of a “walking for exercise” habit was assessed using the Self-Report Habit Index (SRHI)⁷². The SRHI is the most-used outcome measure for assessing habits on health behaviors⁵⁸. The SRHI is comprised of 12 items, with statements that follow a stem (i.e., ‘Walking for exercise is something…’), spanning features of habit including repetition (e.g., “…I do frequently”), automaticity (e.g. ‘…I do without thinking’), and expressing identity (e.g. ‘…that’s typically “me”’). Participants designated their agreement to these statements by using response scales from 0 – 10 (strongly disagree to strongly agree). A total habit strength score was calculated as the mean item score, expressed as a percentage (0–100%: low to high habit strength).

We tracked adherence to the walking intervention and engagement in the follow-on period by asking participants to record the date, start time, and end time of each walking session. We validated the accuracy of self-reported adherence by examining the concurrence of paper-based session data with sensor-derived gait data from MR-005 used by the Amped PD group. Only negligible differences were found. Data availability for outcomes are reported in Supplementary Table 3.

Secondary outcomes

Global Rating of Change (GRoC) Scale: We used the GRoC Scale to examine participants’ self-perceived impact of the intervention on their ‘overall physical activity’ and ‘walking’. The GRoC Scale is an outcome measure commonly used as an anchor in the assessment of responsiveness or meaningfulness of an intervention⁷³. We calculated the frequency of participants who reported improvements in overall physical activity and walking (i.e., minimally, much, very much improved) at the completion of the walking intervention during the 8-week clinical assessment.

Clinical Measures: We explored the effects of the intervention on motor impairment severity (UPDRS Part 3 score), balance (Mini-BESTest⁷⁴), walking capacity (10-Meter Walk Test⁴⁴, 6-Minute Walk Test⁴⁴), walking confidence (Self-Efficacy of Walking-Duration⁴¹), spatiotemporal gait metrics (stride velocity, stride length), functional lower extremity strength (Five-Times Sit-to-Stand Test⁷⁵), quality of life (PDQ39⁷⁶), and depression (Geriatric Depression Scale⁷⁷). The outcome measures were administered during the baseline, 6-Week, and 8-week clinical assessments.

User Experience: Critical to the widespread adoption of new technologies, we sought to understand the user experience with the novel, autonomous closed-loop, music-based RAS system. During the final clinical assessment, participants in the Amped-PD group participated in interview implemented by the interventionist. They were asked open-ended questions about perceived changes in walking, as well as barriers to and facilitators of walking related to the walking intervention and/or technology. However, no follow-up questions or probing were administered. A team of 5 researchers systematically reviewed the interview transcripts from 20 participants using Clarke and Braun’s six phase approach for thematic qualitative analysis⁷⁸. This procedure generated defined themes and supported with participant quotes.

Statistical analysis

Data analyses was performed using SPSS Statistics (IBM, version 29). Study data were collected and managed using REDCap electronic data capture tools hosted at Boston University Medical Center^55,56. All measurements were taken from distinct participant samples. Normality of distribution was examined based on skewness and kurtosis, and the Shapiro-Wilk test. We used parametric and non-parametric statistics depending on normality of data distribution and other related test assumptions. To examine between-group differences for all primary outcomes, we ran a one-way ANOVA or Kruskall-Wallis test to compare change scores at the 6-Week and 8-Week clinical assessments relative to baseline. To examine within-group differences, we ran a Friedman’s ANOVA on absolute values of all primary outcomes across their respective timepoints, with multiple post-hoc comparisons corrected using Bonferroni adjustments. As exploratory analyses on the relationship between habit strength and adherence, we ran a linear regression between SRHI at the 8-Week clinical assessment and total walking sessions completed for the study. For all secondary outcomes, we ran between-group analyses using non-parametric statistics to compare change scores at the 6-Week and 8-Week clinical assessments relative to baseline, using two-tailed tests with alpha set at 0.05. To fully convey the effect the intervention as intended, the study results reported above were based on a completer analysis. The reported results were consistent with an intention-to-treat analysis using “last-observation-carried-forward” data imputation (see Supplementary Tables 4 and 5). All statistical tests were two-sided, with central tendencies reported as mean and standard deviation.

The study’s sample size was based on a priori power analysis using G*Power. With ANOVA set as statistical test, a total sample size of 36 across 2 groups was adequately powered (power = 0.80) with alpha of 0.05 to detect an effect size (f = 0.49) for the effects of intervention on the primary outcome of change in moderate intensity minutes. For the purpose of the power analyses, we focused on moderate intensity minutes as this directly hinges on the fundamental premise of the intervention. Thus, conceptually, the effects of other outcomes rely on successful engagement in moderate intensity walking enabled by the RAS intervention. To account for at least 20% attrition across groups, a total sample of 44 was set as the target sample size (22 per group).

Author contributions

F.P. contributed to study concept, data collection, data analysis, data interpretation, first and subsequent drafts, and critical review of manuscript. J.T.C. contributed to study concept, data interpretation, first and subsequent drafts, and critical review of manuscript. J.A.Z. contributed to study concept and data collection. N.W. contributed to data collection. T.B. contributed to data collection. D.A.R. contributed to data collection and technical support. N.E. contributed to data analysis and interpretation. M.B.H. contributed to data analysis and interpretation. L.N.A. contributed to study concept, data interpretation, and critical review of manuscript. T.D.E. contributed to study concept, data analysis, data interpretation, first and subsequent drafts, and critical review of manuscript. All authors reviewed and approved the final manuscript.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

L.N.A. is a paid advisor to MedRhythms Inc. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-025-00952-x.