Cueing-assisted gamified augmented-reality home rehabilitation for gait and balance in people with Parkinson's disease: feasibility and effectiveness in the clinical pathway

1 Cueing-assisted gamified augmented-reality home rehabilitation for 1 gait and balance in people with Parkinson's disease: feasibility and 2 effectiveness in the clinical pathway 3 4 Eva M. Hoogendoorn1,2*, Daphne J. Geerse1, Annejet T. van Dam1, Sybren J. van Hall1, Pieter F. van 5 Doorn1,2, Lotte E.S. Hardeman1, Marco J.M. Hoozemans1, John F. Stins1, and Melvyn Roerdink1,2 6 1 Department of Human Movement Sciences, Faculty of Behavioural and Movement Sciences, Vrije Universiteit 7 Amsterdam, Amsterdam Movement Sciences, The Netherlands 8 2 Department of Nutrition and Movement Sciences, NUTRIM Institute of Nutrition and Translational Research in 9 Metabolism & MHeNs Institute of Mental Health and Neurosciences, Faculty of Health, Medicine and Life 10 Sciences, Maastricht University, Maastricht, The Netherlands 11 *Correspondence 12 Eva Hoogendoorn (e.m.hoogendoo[email protected]) 13 Abstract 14 Objective: This trial evaluated the clinical feasibility and potential effectiveness of Strolll, an augmented 15 reality (AR) neurorehabilitation platform offering gamified gait-and-balance exercises complemented 16 with assistive AR cueing for individuals with Parkinson’s disease, implemented in real-world clinical 17 practice. Additionally, we assessed how to tailor AR cues to assist gait during AR exercises. 18 Methods: In this pragmatic clinical trial, we onboarded 15 clinics in the Netherlands, trained 28 19 therapists, and included 100 individuals with Parkinson’s disease (Hoehn and Yahr stages 1-3). All 20 participants followed the T0-usual-care-control-T1-Strolll-intervention-T2 procedure. The Strolll 21 intervention consisted of a two-week supervised in-clinic training followed by 6 weeks, 5 sessions/week, 22 30 active minutes, independent home-based training. 23 Results: No serious adverse events occurred; only two non-injurious falls were reported in >60.000 24 performed exercise minutes. Adherence was high (96% session adherence and 91% active 25 minutes/session adherence). Therapists prescribed the program progressively, with significantly higher 26 game-play levels over time. Participants’ performance on exercises increased over time. Participants and 27 therapists rated user experience and technology acceptance positively. Significant improvements in 28 Timed-Up-and-Go and 10-Meter Walk Test (fast speed) scores occurred after the intervention period 29 2 only. Five-Times Sit-to-Stand, 10-Meter Walk Test (comfortable speed), and Mini Balance Evaluation 30 Systems Test scores improved after both usual-care and intervention periods. Falls Efficacy Scale 31 International scores showed no significant improvements. AR cueing was deemed beneficial for a subset 32 of participants. 33 Conclusion: Strolll is a safe, adherable, progressive, usable, and well-accepted therapist-managed, home34 based intervention for people with Parkinson’s disease, with the potential to improve gait, balance, and 35 fall-risk indicators. Findings on the integration of AR cueing highlight the importance of an 36 individualized approach. 37 Impact: Implementing AR rehabilitation technologies like Strolll in the clinical pathway is feasible, 38 offering a safe and scalable way for individuals to train independently, potentially improving accessibility 39 of care and broadening its use to physical activity promotion. 40 Trial registration 41 ClinicalTrials.gov: NCT06590987 42 43 Keywords (3-10): Parkinson’s disease, augmented reality, cueing, gait, balance, rehabilitation, gamified 44 exercise, clinical practice, telerehabilitation, blended-care 45 46 3 1. Background 47 Parkinson's disease is the fastest-growing, progressive neurological disease with a severe impact on the 48 healthcare system1. Gait-and-balance impairments are common in people with Parkinson's disease, 49 contributing to an increased risk of falls and a reduced quality of life2. Exercise and physiotherapy are 50 considered essential for people with Parkinson's disease to maintain and improve motor function, 51 decelerate further decline, and promote overall health and well-being3,4. Nevertheless, due to mobility 52 impairments and other practical limitations, individuals with Parkinson's disease often face challenges 53 attending to regular, center-based rehabilitation programs over a prolonged period5, while adherence rates 54 to prescribed exercise at home are typically low6. Various home-based rehabilitation interventions have 55 emerged for maintaining and enhancing motor function and achieving overall health benefits7,8. Despite 56 their potential, traditional home-based exercise interventions often suffer from low therapy adherence and 57 limited monitoring capabilities for therapists9. Emerging technologies with gamified exercises and 58 captured exercise data offer promising and engaging solutions to these limitations10. 59 Augmented reality (AR) glasses, such as Magic Leap 2 or Microsoft HoloLens 2, represent a 60 promising innovation in this domain, allowing for an interactive experience within an individual’s 61 environment. This technology is embodied in the development of Strolll, an AR neurorehabilitation 62 platform with gamified gait-and-balance exercises specifically designed for people with Parkinson's 63 disease for use in the clinic or at home (see Figure 1A-F, H-I and Supplementary Video 1), which can be 64 managed by therapists for supervised in-clinic therapy as well as for asynchronous telerehabilitation 65 independently at home11. A recent controlled feasibility study in a science-led setting demonstrated that 66 the earlier version of Strolll (previously called Reality DTx®) is safe, usable, adherable, and well67 accepted for independent use at home, with targeted intervention effects for improving gait, balance, and 68 fall risk in people with Parkinson's disease12. A key innovation of the Strolll platform is the integration of 69 AR cueing, involving visual or auditory stimuli to assist and modify gait during the gamified exercises13. 70 These AR cues may be used by individuals with more severe gait impairments, like freezing of gait and 71 4 festination, to assist them with AR exercises that require walking and turning. Since cueing is not a one72 size-fits-all solution14-16, these AR cues are probably most beneficial when tailored to the individual (i.e., 73 type of cue and settings). 74 This pragmatic clinical trial is designed to investigate the clinical feasibility and potential 75 effectiveness of Strolll implemented in real-world clinical practice. We expected that Strolll is i) clinically 76 feasible in terms of safety, adherence, performance, user experience, and acceptability (for individuals 77 with Parkinson’s disease and their therapists) and ii) potentially effective for improving gait, balance, and 78 fall-risk indicators. Additionally, we evaluated procedures to i) identify participants deemed to benefit 79 from complementary AR cues in AR exercises and ii) individually tailor AR cue characteristics. We 80 expected that a subset of participants, particularly those with severe mobility impairments such as 81 freezing of gait and festination, could benefit from adding complementary AR cues and that the selected 82 type of cue and its settings varied over participants (i.e., cueing is not a one-size-fits-all solution15,16). 83 5 84 Figure 1 Strolll gait-and-balance exercises (A-F, H-I) implemented in the clinical pathway in the 85 Netherlands (G). Specific exercises were Smash! (without [A] and with [B] assistive AR cues), Mole 86 Patrolll (C), Puzzle Walk (D), Cue Challenge (E), Basketballl (F), Wobbly Waiter (H), and Hot Buttons 87 (I). 88 2. Methods 89 The trial was registered on Clinicaltrials.gov (NCT06590987). The methods are concisely described here, 90 with a fully detailed study protocol previously published17. Deviations from the published study protocol 91 (i.e., required adjustment of the statistical analysis18) are documented and justified in the Transparent 92 Changes document (Supplementary Material 1). 93 6 2.1 Study setting, design, and procedures 94 This study was implemented in the clinical pathway at 15 healthcare practices affiliated with 95 ParkinsonNet, Dutch society for Parkinson’s therapists19, across the Netherlands (Figure 1G): Pieter van 96 Foreest, Houding & Bewegen, Verheul & Weerman Fysiotherapeuten, BeweegBewust, PACA, Motion 97 Fysiotherapie & Preventie, SFR Beweegt, Fysiotherapie Zwaansvliet, Careyn, Norschoten, Parkinson 98 TrainingsCentrum, Kr8, Leidsche Rijn Julius Gezondheidscentra, Fysiotherapeuten Maatschap Woerden 99 and Parkinson Centraal. All therapists involved (n=28) were trained in using Strolll and familiarized with 100 the study procedure, including administering the clinical assessments17. 101 This pragmatic clinical trial included three in-clinic assessments performed by the participant’s 102 therapist, which we aimed to schedule at the same time of day to minimize the influence of medication 103 and ON/OFF motor fluctuations. After the baseline assessment (T0), participants underwent a 6-week 104 usual-care period, followed by a pre-intervention assessment (T1). During the usual-care period, 105 participants received no additional instructions or therapy related to Strolll and continued their regular 106 weekly therapy schedule, generally consisting of in-clinic sessions adhering to physiotherapy guidelines 107 for Parkinson’s disease3,4, possibly supplemented by prescribed conventional home exercises (e.g., paper108 listed or video clips demonstrating the exercises). This was followed by two weeks of supervised in-clinic 109 training with Strolll, after which participants used Strolll independently at home for 6 weeks, 110 asynchronously managed by their therapists, and integrated with their regular in-clinic usual-care 111 sessions. Participants finished the study with a post-intervention assessment (T2). This design, where all 112 participants followed the same T0-control-T1-intervention-T2 procedure, streamlined implementation in a 113 real-world clinical setting and facilitated the assessment of both clinical feasibility and potential 114 effectiveness. 115 2.2 Study population 116 Participants (N=100, based on an a-priori power analysis17) were recruited by their physiotherapist or 117 exercise therapist in the participating healthcare practices. Eligibility was established through telephone 118 7 screening by the researchers. Participants were included if they had been diagnosed with Parkinson's 119 disease (stages 1-3 on the Hoehn and Yahr scale), experienced self-reported gait and/or balance 120 impairments, were aged 21 years or over, and proficient in the Dutch language. Exclusion criteria 121 included neurological and/or orthopedic conditions seriously affecting gait, severe cognitive, visual, 122 and/or hearing impairments (after corrective aids), insufficient physical capacity (e.g., frequent faller 123 and/or the inability to walk independently for 30 minutes), severe visual hallucinations and/or illusions, 124 and unstable dosage of medication at the start of the study. All participants provided written informed 125 consent before participating in the study. 126 2.3 Intervention 127 The Strolll intervention, a certified neurorehabilitation software platform developed for AR glasses in 128 collaboration with Strolll Limited (www.strolll.co), was delivered in two stages. In the first stage, 129 therapists introduced participants to Strolll during a two-week supervised in-clinic training period, 130 allowing participants to familiarize themselves with the technology and evaluate the potential benefits of 131 AR cueing. Strolll comprises seven complementary gamified exercises: Mole Patrolll, Smash!, 132 Basketballl, Hot Buttons, Puzzle Walk, Wobbly Waiter, and Cue Challenge, delivered via state-of-the-art 133 AR glasses (i.e., HoloLens 2 and Magic Leap 2), see Figure 1A-F,H-I and Supplementary Video 1, each 134 designed to improve aspects of gait, balance, and fall risk. AR cues could be added to Smash! and Cue 135 Challenge (Supplementary Video 2) to support or improve the participant’s gait (e.g., increasing step 136 length, alleviating freezing). During the in-clinic training period, all types of cues were explored 137 (Supplementary Material 2) and tailored to the participant by their therapist (e.g., step length, step height, 138 color). When deemed beneficial, the most suitable AR cue was selected through a shared decision-making 139 process involving the participant and therapist during the in-clinic training period and added to Smash! 140 and Cue Challenge. The other walking-based exercises, Mole Patrolll, Puzzle Walk, and Wobbly Waiter, 141 already incorporated AR-mediated goal-directed elements acting as spatial location cues (moles, puzzle 142 pieces) or spatiotemporal cues (moving waiter). A detailed description of the gamified exercises and cues 143 can be found in Supplementary Material 2. At the end of the in-clinic period, therapists, in consultation 144 8 with the participant, decided whether the Strolll intervention at home was deemed safe and suitable for 145 the participants; if not, participants discontinued the study. The second stage comprised six weeks of 146 training with Strolll independently at home, consisting of 30-minute sessions five days per week by 147 default, following the recommended 150 exercise minutes per week20, with adjustments based on 148 individual capacity as determined by the therapist. The therapists were instructed to remotely prescribe, 149 evaluate, and adjust the exercise program weekly, with the option to modify each gamified exercise 150 separately in terms of duration, level, and settings, through the Strolll web portal. By default, all exercises 151 were prescribed for each participant, although therapists could adjust the selection if needed. This process 152 was guided by shared decision-making between participant and therapist using participant’s feedback and 153 therapist’s clinical expertise, and potentially driven by adherence and game scores (e.g., number of moles 154 spawned, meters walked; see Supplementary Material 2) from the Strolll web portal as input. 155 2.4 Outcomes 156 The clinical feasibility of Strolll was assessed in terms of safety (i.e., adverse events while using Strolll, 157 such as falls or dizziness, and weekly fall rate in daily life), adherence (i.e., adherence to the prescribed 158 number of sessions and the prescribed active minutes/session), performance (i.e., game-related and 159 functional performance scores), user experience, and acceptability. User experience and acceptability 160 were evaluated using the User Experience Questionnaire (UEQ)21,22, a technology acceptance and use 161 questionnaire (based on the model of Unified Theory of Acceptance and Use of Technology23), a top-3 162 advantages and disadvantages of Strolll for future clinical implementation, and a Strolll intervention163 specific questionnaire including items on the integration of AR cues. Additionally, therapists' experiences 164 were assessed with comparable questionnaires, followed by a post-study process evaluation in focus 165 groups, which will be reported in a separate qualitative paper. A final feasibility outcome was the 166 proportion of participants deemed eligible for training independently at home and the number of 167 intervention-related dropouts. 168 9 Standardized clinical gait-and-balance assessments, which are indicators of fall risk24-27, were 169 administered to evaluate the potential effectiveness of the intervention at baseline, pre-, and post170 intervention (T0, T1, T2). The Timed Up-and-Go test (TUG) was the primary outcome used for the 171 sample-size calculation, with as secondary outcomes the Five Times Sit-to-Stand test (FTSTS), 10-Meter 172 Walk Test at comfortable and fast speeds (10MWT), Mini Balance Evaluation Scale Test (Mini173 BESTest), and Falls Efficacy Scale International (FES-I). 174 To evaluate the integration of AR cues within the gamified exercises, we documented whether 175 participants used AR cues during the Strolll intervention, along with the specific type of AR cue, its 176 settings (e.g., length, height, width), and the rationale for using the AR cue (e.g., alleviating freezing of 177 gait or increasing step length). 178 2.5 Statistical analysis 179 Safety was described by the number of adverse events, including a paired-samples t-test comparison of 180 the average weekly number of falls experienced during the usual-care and intervention periods. 181 Adherence to the Strolll intervention was assessed with a repeated-measures ANOVA with the within182 subject factor Time (three levels: first, second, and third part, so approximately two-week intervals, of 183 Strolll intervention at home). In case of significant main effects of Time, post-hoc paired-samples t-tests 184 were conducted, with Bonferroni-adjusted p-values. With one-sample t-tests against 100%, we evaluated 185 possible systematic deviations in adherence from what was prescribed. To assess whether Strolll 186 integrated into the clinical pathway was prescribed in a progressive yet achievable manner, we subjected 187 game levels and performance scores (see Supplementary Material 2) to repeated-measures ANOVAs or 188 their non-parametric equivalents with Time as a within-subject factor (three levels: first, second, and third 189 part), using paired-samples t-tests (or non-parametric equivalents) for post-hoc analyses of significant 190 main effects, with p-values adjusted using the Bonferroni correction. Since analyses were conducted 191 independently for each gamified exercise, we did not apply an additional correction for multiple exercises 192 beyond the within-analysis adjustments. The UEQ outcomes were described descriptively and interpreted 193 16 space at home, and (3) limited variation in exercises. Before the intervention, the top-3 disadvantages 299 reported by therapists were (1) technical issues (see Supplementary Material 6), (2) reduced supervision, 300 and in shared third place (3) unsuitability for every Parkinson's disease patient and financial costs. After 301 delivering the intervention, the reported barriers shifted to (1) technical issues, (2) financial costs, and (3) 302 the need for specific technical skills. These top-3 rankings were determined based on how frequently and 303 prominently participants and therapists mentioned each factor (see Supplementary Material 7 for a 304 comprehensive overview of reported advantages and disadvantages). 305 306 Figure 4. Visualization of the User Experience Questionnaire (UEQ) scores across the six subscales for 307 participants (A) and therapists (B), relative to benchmark scores of the questionnaire. Error bars indicate 308 the 95% confidence intervals. 309 17 310 Figure 5. Technology acceptance and use questionnaire and the Strolll evaluation questionnaire for the 5311 point Likert-scale questions. 312 313 18 3.3 Potential effectiveness 314 A significant main effect of Time was observed for the LME analyses for all outcome measures (TUG, 315 Mini-BESTest, FTSTS, 10MWT at comfortable walking speed, and FES-I; all p≤0.013), except for the 316 10MWT at fast walking speed, showing a borderline significant effect of Time (p=0.053), see Table. For 317 the primary outcome measure, TUG, and the 10MWT at fast walking speed, only the second reverse 318 Helmert contrast (T2-mean(T1, T0)) was significant (t(165.006)=-2.39, p=0.018 and t(163.992)=-2.36, 319 p=0.020, respectively), indicating that improvements occurred only after the intervention. For the 320 secondary outcome measures Mini-BESTest, FTSTS, and 10MWT at comfortable walking speed, 321 significant improvements were observed after both the first (T1-T0) and second reverse Helmert contrasts 322 (T2-mean(T1, T0)); p≤0.04), indicating performance improvements after usual care, with further 323 improvement following the intervention. Regarding the FES-I, only the first reverse Helmert contrast 324 reached significance (t(164.450)= 4.23, p<0.001), showing higher scores after the usual-care period 325 compared to baseline, suggesting an increased concern about falling following usual care only. 326 327 19 Table. Main effects of Time from the linear mixed-effects analyses with their reverse Helmert contrasts for the potential effectiveness 328 concerning clinical test outcomes 329 T0 T1 T2 Main effect of Time 1st reverse Helmert contrast 2nd reverse Helmert contrast EMM (SE) EMM (SE) EMM (SE) F(df) p t(df) p ΔT1-T0 (SE) t(df) p ΔT2-T1,T0 (SE) TUG (s) 8.56 (0.22) 8.30 (0.22) 8.09 (0.23) F(2,163.59)=4.45 0.013 t(162.188)= -1.80 0.075 -0.27 (0.15) t(165.006)= -2.39 0.018 -0.34 (0.14) FTSTS (s) 13.79 (0.35) 12.73 (0.35) 11.99 (0.36) F(2,164.09)=21.87 <0.001 t(162.431)= -4.15 <0.001 -1.06 (0.26) t(165.763)= -5.17 <0.001 -1.27 (0.25) 10MWT comf (s) 8.23 (0.17) 7.99 (0.17) 7.74 (0.18) F(2,163.27)=7.53 <0.001 t(161.885)= -2.07 0.040 -0.24 (0.12) t(164.664)= -3.29 0.001 -0.36 (0.11) 10MWT fast (s) 6.42 (0.16) 6.36 (0.16) 6.18 (0.16) F(2,163.00)=2.99 0.053 t(162.003)= -0.66 0.513 -0.06 (0.09) t(163.992)= -2.36 0.020 -0.20 (0.09) MiniBESTest 23.09 (0.30) 23.80 (0.30) 24.41 (0.32) F(2,164.56)=13.17 <0.001 t(162.418)= 3.01 0.003 0.72 (0.24) t(166.714)= 4.16 <0.001 0.96 (0.23) FES-I 24.20 (0.78) 26.67 (0.78) 25.70 (0.80) F(2,165.44)=9.02 <0.001 t(164.450)= 4.23 <0.001 2.47 (0.59) t(166.437)= 0.49 0.625 0.27 (0.54) EMM = Estimated Marginal Means, SE = Standard error 330 331 20 3.4 AR cues 332 AR cues were deemed beneficial for 53 out of 92 participants. Participant characteristics (Hoehn & Yahr, 333 Freezers, MoCA score [cognitive dysfunctions], and baseline TUG) did not differ between participants 334 for whom AR cueing was and was not deemed beneficial (all p>0.05). The preferred type of cue varied 335 across participants: 2D lines (n=22), 3D obstacles (n=15), auditory rhythm (n=12), and dinosaur 336 footprints (n=4). Cue settings were personalized, resulting in a broad range of configurations 337 (Supplementary Material 8). Mentioned reasons for the decision to add AR cues to gamified exercises 338 were alleviating freezing of gait (n=6), improving step length (n=29), walking stability (n=15), foot 339 clearance (n=28), and other purposes (n=11), with multiple reasons possible per participant. These 340 findings support the hypothesis that most participants were deemed to benefit from adding AR cues and 341 that the selected type of cue and its settings varied among participants. As an example of an 342 individualized cueing approach, we highlight two participants: one participant preferred AR lines to 343 alleviate freezing and take big steps, with an intercue distance of 70 cm and a line width of 50 cm, while 344 another participant chose AR obstacles with an intercue distance of 66 cm and an obstacle height of 20 cm 345 to better lift the feet during walking. Eventually, 46 participants used personalized AR cues at home 346 during either Cue Challenge, Smash!, or both. Both participants and therapists generally found AR cues 347 valuable in Cue Challenge, while for Smash! their opinions varied (see Figure 5). 348 21 4. Discussion 349 This pragmatic clinical trial evaluated the clinical feasibility and potential effectiveness of Strolll, an AR 350 neurorehabilitation platform offering gamified gait-and-balance exercises for individuals with Parkinson's 351 disease, implemented in real-world clinical practice. Furthermore, the study explored the potential of 352 integrating personalized AR cueing into AR exercises. 353 4.1 Feasibility 354 This study was conducted under real-world conditions within clinical practice in the Netherlands, which 355 allowed us to evaluate Strolll beyond prior research-controlled conditions12. Overall, the intervention 356 proved to be clinically feasible in terms of safety, adherence, performance, user experience, and 357 acceptability. 358 A large majority of participants (93.9%) started the independent home training with Strolll, nine 359 of whom withdrew for intervention-related reasons. No serious adverse events were reported during the 360 intervention period. A few participants reported symptoms associated with cyber sickness, like headache, 361 nausea, and dizziness (see Supplementary Material 5). Some of these symptoms may also have a different 362 origin than cyber sickness, such as exercise-induced dizziness through squatting or turning. Nevertheless, 363 prior research indicated that symptoms associated with cyber sickness can occur in augmented-reality 364 applications, but that their prevalence is generally lower than in virtual reality contexts30,31, which seems 365 supported by the limited reports in the current study. Although two participants experienced a non366 injurious fall (out of >60.000 performed exercise minutes in total, one supervised in clinic and one 367 independent at home), the incidence rate was low relative to the study size, study duration, and nature of 368 the target population32. Furthermore, participants generally did not report fear of falling. This perception 369 supports the notion that the trajectory from in-clinic sessions to independent home-based exercise with 370 Strolll, under therapist prescription, is safe. 371 22 Therapists were able to prescribe, evaluate, and adjust the exercise program of the participants, 372 based on shared decision-making with the participants, throughout the intervention in terms of exercise 373 type (i.e., prescribed gamified exercises), duration (e.g., for a subset of participants [n=25] the prescribed 374 active-minutes was adjusted to align with their physical capacity), game-play level, and frequency. This 375 individualized approach presumably contributed to a high adherence compared to similar interventions 376 targeting individuals with Parkinson’s disease33, which was consistent with findings from our previous 377 research-led controlled study12. This highlights the importance of individually tailoring prescribed 378 treatment programs in promoting long-term engagement and adherence and supports its feasibility when 379 implemented in the clinical pathway. 380 In terms of performance, Strolll was expected to be prescribed as a progressive but achievable 381 training. Participants demonstrated improvements over time, not only by executing higher game-play 382 levels but also by achieving higher game-play and functional performance scores. Notably, where our 383 prior research focused primarily on movement quantity scores (e.g., number of squats, meters walked)12, 384 performance scores were refined by tempo metrics (e.g., number of squats per minute, meters walked per 385 minute; see Figure 3 and Supplementary Material 2). As a result, we observed improvements not only in 386 game-play performance scores, but also in functional performance metrics (i.e., number of sit-to387 stands/squats, functional reaches, and meters walked per minute). These findings confirm the intended 388 progressive nature of Strolll, also when implemented in the clinical pathway. 389 The overall user experience was positive among both participants and therapists. UEQ results 390 revealed that across domains, Strolll was evaluated higher than the benchmarks, showing excellent ratings 391 for being able to offer a novel, attractive, and stimulating technology. A notable exception was the 392 (borderline) below-average rating for efficiency by therapists (see Figure 4), which may be due to being 393 relatively inexperienced in using this novel advanced form of rehabilitation technology and technical 394 inefficiencies (e.g., within the web portal for managing programs). The technology acceptance 395 questionnaire indicated a positive behavioral intention towards the use of Strolll, with the participants and 396 23 therapists demonstrating similar ratings across most subscales. Participants described the training as 397 enjoyable and appreciated the flexibility of being able to train at home and at their own convenience, 398 which is in line with our previous qualitative study34. Therapists particularly valued the ability to monitor 399 their patients’ progress remotely. Reported disadvantages included technical issues, which are not 400 uncommon in studies involving innovative digital interventions35,36, limited training space at home (which 401 hindered the use of Wobbly Waiter and Cue Challenge), and perceived lack of variation in exercises. 402 These insights were shared with Strolll Limited for future optimization of the platform. 403 4.2 Potential effectiveness 404 The primary outcome measure, TUG, and the 10MWT at fast speed showed significant improvements 405 after the Strolll intervention. The other secondary outcome measures, including the Mini-BESTest, 406 FTSTS, and 10MWT at comfortable walking speed, showed significant improvements after usual care, 407 which limited the identification of subsequent intervention effects. Nevertheless, significant 408 improvements in those secondary clinical outcome measures were still observed after the Strolll 409 intervention. Although significant improvements in both primary and secondary outcome measures 410 following the intervention did not reach the minimal clinically important difference thresholds37-40 (i.e., no 411 clinically meaningful effect), they were consistently observed in the expected positive direction, 412 suggesting the potential effectiveness of Strolll in enhancing gait, balance, and fall-risk indicators. Thus, 413 although prior improvements after usual care may have attenuated effects, these improvements were 414 maintained and further enhanced, highlighting the potential added value of Strolll. Additionally, 415 participants and therapists also rated the perceived expectancy for improving gait and balance positively 416 (see Figure 5). 417 To conclusively validate the effectiveness of Strolll, a randomized controlled trial on clinical 418 outcomes is warranted41 and initiated42. Additionally, the effectiveness findings raise concerns about the 419 sensitivity of sparsely sampled clinical outcome measures (T0, T1, T2), due to ceiling or floor effects in 420 relatively high-functioning groups (e.g., M=8.53, SD=2.33 sec of TUG at baseline), and symptom 421 24 fluctuations common in Parkinson’s disease. In that sense, the observed significant improvements in more 422 frequently sampled game-play and functional performance scores, such as increased walking distance per 423 minute, a higher number of sit-to-stands/squats per minute, and more functional reaches per minute 424 (Figure 3), suggest that Strolll targets aspects of gait and balance in a task-specific manner. However, 425 similar to most of the standard clinical outcome measures, these game-play and functional performance 426 scores primarily reflect aspects of movement quantity and tempo, while aspects of movement quality 427 remain less explored. The rich data captured by AR glasses offers a promising opportunity to align game428 play performance, functional performance, and movement-quality outcome measures more closely with 429 the task-specific rehabilitation goals of the gamified exercises, tentatively offering a greater 430 responsiveness to change than standard clinical assessments. Prior work has already demonstrated the 431 utility of AR data for valid and reliable parameterization of various gait and balance aspects during 432 standardized tests43-46. Accordingly, the potential of remotely collected gait-and-balance data during at433 home training will be further explored using the rich AR dataset of this trial (>60,000 minutes). At the 434 same time, further optimization between the gamified-exercise mechanics and their task-specific training 435 goals may strengthen the intended effects of the intervention, ultimately aiming to achieve an optimal 436 alignment among gamified-exercise mechanics (e.g., adding obstacles to Mole Patrolll), task-specific 437 goals (e.g., including obstacle avoidance under walking adaptability), and their functional or movement438 quality outcome measures (e.g., obstacle-avoidance success rate). 439 4.3 Integration of AR cueing 440 AR cues were deemed beneficial for a slight majority of participants. Contrary to our expectations, these 441 participants did not differ from non-users in baseline characteristics. This may be due to the study’s 442 broader focus on various gait impairments, rather than exclusively targeting freezing of gait. These 443 findings highlight the importance of personalized decision-making when integrating AR cueing into 444 rehabilitation exercises, rather than focusing on predefined subgroups such as individuals with freezing of 445 25 gait. Furthermore, both the type of cue and its settings varied over participants, supporting our expectation 446 that cueing is not a one-size-fits-all solution. 447 Some participants who initially opted to use AR cueing ultimately did not incorporate it into their 448 home training. This may have been due to either a lack of perceived benefits or practical barriers, such as 449 insufficient space at home to effectively apply spatial AR cues. For Smash!, some participants reported 450 that AR cues offered little or no added value, possibly because the exercise was already inherently goal451 directed (i.e., requiring participants to walk towards a pillar to punch an object from that pillar), making 452 the cues feel redundant or even distracting. However, exercises designed specifically for cueing, like Cue 453 Challenge and Wobbly Waiter, may hold promise for future research and clinical application for 454 improving aspects of gait, such as increasing step length, reducing shuffling, alleviating freezing of gait, 455 and improving gait speed in people with Parkinson’s disease. 456 4.4 Limitations and considerations 457 The pragmatic nature of this study holds a potential bias regarding the potential effectiveness. Although 458 all participating therapists were trained prior to the study, assessments were conducted by different 459 clinicians at multiple sites, and minor deviations in clinical test execution cannot be excluded. This 460 variability could have affected both the validity and the inter-rater reliability of the measurements. 461 Nevertheless, this reflects the intended use of the platform in clinical practice, where such variability is 462 expected. Besides that, therapists had no prior experience with the use of AR technology in clinical 463 practice before this study. Increased familiarity with the technology may enhance their ability to 464 prescribe, evaluate, and adjust the intervention, individualized to the participant, which may optimize 465 clinical outcomes. In the current study, exercise prescription relied on therapists’ clinical expertise in 466 combination with shared decision-making with participants and was potentially driven by adherence and 467 game scores available in the online web portal, where exercise prescription and its tailoring were 468 managed. Future implementation may benefit from more structured guidance (e.g., decision trees or 469 formats informed by baseline assessments of participants), which may support or guide individualization 470 32 25. Duncan RP, Leddy AL, Earhart GM. Five times sit-to-stand test performance in 628 Parkinson's disease. Archives of Physical Medicine and Rehabilitation. 2011;92(9):1431-6. 629 doi:10.1016/j.apmr.2011.04.008 630 26. Nocera JR, Stegemöller EL, Malaty IA, Okun MS, Marsiske M, Hass CJ. Using the 631 Timed Up & Go Test in a Clinical Setting to Predict Falling in Parkinson's Disease. Archives of 632 Physical Medicine and Rehabilitation. 2013;94(7):1300-1305. doi:10.1016/j.apmr.2013.02.020 633 27. Lopes LKR, Scianni AA, Lima LO, de Carvalho Lana R, Rodrigues-De-Paula F. The 634 Mini-BESTest is an independent predictor of falls in Parkinson Disease. Brazilian Journal of 635 Physical Therapy. 2020;24(5):433-440. doi:10.1016/j.bjpt.2019.07.006 636 28. JASP Team. version 0.19.0, 2024. 637 29. Nieuwboer A RL, Herman T, Vandenberghe W, Emil GE, Thomaes T, et al. Reliability 638 of the new freezing of gait questionnaire: agreement between patients with Parkinson's disease 639 and their carers. Gait and posture. 2009;30(4):459-463. 640 30. Kaufeld M, Mundt M, Forst S, Hecht H. Optical see-through augmented reality can 641 induce severe motion sickness. Displays. 2022/09/01/ 2022;74:102283. 642 doi:10.1016/j.displa.2022.102283 643 31. Zhang J, Che X, Chang E, et al. How different text display patterns affect cybersickness 644 in augmented reality. Scientific Reports. 2024;14(1):11693. doi:10.1038/s41598-024-62338-y 645 32. Hiorth YH, Alves G, Larsen JP, Schulz J, Tysnes O-B, Pedersen KF. Long-term risk of 646 falls in an incident Parkinson’s disease cohort: the Norwegian ParkWest study. Journal of 647 Neurology. 2017;264(2):364-372. doi:10.1007/s00415-016-8365-z 648 33. Flynn A, Allen NE, Dennis S, Canning CG, Preston E. Home-based prescribed exercise 649 improves balance-related activities in people with Parkinson's disease and has benefits similar to 650 centre-based exercise: a systematic review. Journal of Physiotherapy. 2019;65(4):189-199. 651 doi:10.1016/j.jphys.2019.08.003 652 34. Hardeman LES, van Benten E, Hoogendoorn EM, et al. Home-Based Augmented Reality 653 Exercise For People With Parkinson Disease: Qualitative Acceptability Study. JMIR Rehabil 654 Assist Technol. 2025;12:e70802. doi:10.2196/70802 655 35. Glegg SMN, Liisa H, Sue S, et al. Evaluating change in virtual reality adoption for brain 656 injury rehabilitation following knowledge translation. Disability and Rehabilitation: Assistive 657 Technology. 2017;12(3):217-226. doi:10.3109/17483107.2015.1111944 658 33 36. Levac D, Glegg SMN, Sveistrup H, et al. A knowledge translation intervention to 659 enhance clinical application of a virtual reality system in stroke rehabilitation. BMC Health 660 Services Research. 2016;16(1):557. doi:10.1186/s12913-016-1807-6 661 37. Godi M, Arcolin I, Giardini M, Corna S, Schieppati M. Responsiveness and minimal 662 clinically important difference of the Mini-BESTest in patients with Parkinson’s disease. Gait & 663 Posture. 2020;80:14-19. doi:10.1016/j.gaitpost.2020.05.004 664 38. Spagnuolo G, Faria CDCM, da Silva BA, Ovando AC, Gomes-Osman J, Swarowsky A. 665 Are functional mobility tests responsive to group physical therapy intervention in individuals 666 with Parkinson’s disease? NeuroRehabilitation. 2018;42(4):465-472. doi:10.3233/nre-172379 667 39. Hass CJ, Bishop M, Moscovich M, et al. Defining the Clinically Meaningful Difference 668 in Gait Speed in Persons With Parkinson Disease. Journal of Neurologic Physical Therapy. 669 2014;38(4):233-238. doi:10.1097/npt.0000000000000055 670 40. Taghizadeh G, Eissazade N, Fereshtehnejad S-M, et al. Minimal clinically important 671 difference and substantial clinical benefits for singleand dual-task timed up and go test 672 following motor-cognitive training in Parkinson’s disease. Age and Ageing. 673 2025;54(9)doi:10.1093/ageing/afaf241 674 41. Skivington K, Matthews L, Simpson SA, et al. A new framework for developing and 675 evaluating complex interventions: update of Medical Research Council guidance. BMJ. 676 2021;374. doi:10.1136/bmj.n2061 677 42. Arundel C. Augmented reality (Reality DTx®) compared to routine physiotherapy for 678 patients with Parkinson's. ISRCTN. 2025;doi:10.1186/ISRCTN41469859 679 43. Sun R, Aldunate RG, Sosnoff JJ. The Validity of a Mixed Reality-Based Automated 680 Functional Mobility Assessment. Sensors. 2019;19(9):2183. 681 44. van Bergem JS, van Doorn PF, Hoogendoorn EM, Geerse DJ, Roerdink M. Gait and 682 Balance Assessments with Augmented Reality Glasses in People with Parkinson’s Disease: 683 Concurrent Validity and Test–Retest Reliability. Sensors. 2024;24(17):5485. 684 45. van Doorn PF, Geerse DJ, van Bergem JS, Hoogendoorn EM, Nyman E, Roerdink M. 685 Gait Parameters Can Be Derived Reliably and Validly from Augmented Reality Glasses in 686 People with Parkinson’s Disease Performing 10-m Walk Tests at Comfortable and Fast Speeds. 687 Sensors. 2025;25(4):1230. 688 34 46. Miller Koop M, Rosenfeldt AB, Owen K, et al. The Microsoft HoloLens 2 Provides 689 Accurate Measures of Gait, Turning, and Functional Mobility in Healthy Adults. Sensors. 690 2022;22(5):2009. 691 47. Batenburg RS, Vis EBA, van der Velden LFJ. Capaciteitsraming Fysiotherapie. Nivel; 692 2018. Accessed 27-05-2025. 693 48. Ministerie van Volksgezondheid Welzijn en Sport. Missies Gezondheid & Zorg 2024– 694 2027. 2023. Accessed 27-05-2025. 695 49. van Nimwegen M, Speelman AD, Hofman-van Rossum EJ, et al. Physical inactivity in 696 Parkinson's disease. Journal of Neurology. 2011;258(12):2214-2221. doi:10.1007/s00415-011697 6097-7 698 50. Yoon SY, Suh JH, Yang SN, Han K, Kim YW. Association of Physical Activity, 699 Including Amount and Maintenance, With All-Cause Mortality in Parkinson Disease. JAMA 700 Neurology. 2021;78(12):1446-1453. doi:10.1001/jamaneurol.2021.3926 701 51. Tsukita K, Sakamaki-Tsukita H, Takahashi R. Long-term Effect of Regular Physical 702 Activity and Exercise Habits in Patients With Early Parkinson Disease. Neurology. 703 2022;98(8):e859-e871. doi:10.1212/WNL.0000000000013218 704 52. Johansson ME, Cameron IGM, Van der Kolk NM, et al. Aerobic Exercise Alters Brain 705 Function and Structure in Parkinson's Disease: A Randomized Controlled Trial. Annals of 706 Neurology. 2022;91(2):203-216. doi:10.1002/ana.26291 707 53. van der Kolk NM, de Vries NM, Kessels RPC, et al. Effectiveness of home-based and 708 remotely supervised aerobic exercise in Parkinson's disease: a double-blind, randomised 709 controlled trial. The Lancet Neurology. 2019;18(11):998-1008. doi:10.1016/S1474710 4422(19)30285-6 711 712