Introduction

Virtual reality (VR) is an interactive, real-time, entertaining simulation that can provide opportunities for active, repetitive, sensory, and motor practice1,2. Large-scale VR systems around an instrumented treadmill can emulate environments experienced in day-to-day life, providing a bridge between hospital-based rehabilitation and the real world. These systems typically combine two complementary therapeutic elements: immersive virtual environments and real-time biofeedback. The immersive component serves to increase motivation, engagement, and adherence to treatment by presenting functional tasks in gamified, dynamic contexts. In contrast, the biofeedback element delivers task-relevant extrinsic cues (visual and auditory) that guide motor performance and support motor learning3. This combination may be especially beneficial for the pediatric population, in which sustained attention and engagement are often limiting factors in rehabilitation adherence4,5.

Children and adolescents with cerebral palsy (CP) typically present with non-progressive motor impairments resulting from early brain injury, often affecting muscle tone, selective motor control, coordination, and postural stability. These impairments can disrupt gait mechanics, leading to reduced walking speed, endurance, and functional mobility6. In contrast, acquired brain injury (ABI) occurs later in development and may result from trauma (e.g. motor vehicle accident or fall from height) or a non-traumatic event (e.g. stroke, tumors, or encephalitis). Although the timing and presentation differ, children with ABI frequently experience deficits in motor planning, balance, endurance, and neuromuscular coordination, which similarly compromise gait function7. Both groups often require long-term, multidisciplinary rehabilitation, including interventions focused on improving walking capacity and efficiency. Given the need for high-repetition, goal-directed motor practice in both conditions, VR-based biofeedback training provides an engaging and adaptable platform for gait rehabilitation. Improving gait function is a critical rehabilitation goal, as it is closely associated with long-term mobility, independence, participation, and quality of life in children with neuromotor disorders.

Several studies have evaluated the influence of treadmill-based VR biofeedback training on motor function in the paediatric rehabilitation population, such as CP and ABI1,2,5,6,7,8,9,10. It was shown that a series of treatments led to an improvement in walking speed, endurance and gross motor abilities for children with bilateral CP11, and walking abilities, gait pattern and treatment engagement for children with an ABI12. Similar findings were also shown for children experiencing secondary ataxia and Rett syndrome13,14. Within these studies the quantity of treatments was variable and often in a condensed period (i.e. four weeks). Further, the sample sizes ranged from six to 16 individuals, limiting the generalizability of the findings. Moreover, the paediatric evidence base for treadmill-based VR biofeedback is still limited, consisting primarily of small single-center pre–post studies, which restricts both the certainty and generalizability of observed benefits. While controlled trials are ultimately required to establish causal efficacy and isolate the contribution of VR biofeedback from concurrent therapies, larger real-world multicentre cohorts can make a distinct contribution by characterizing pre–post change patterns across heterogeneous diagnoses, functional levels, clinical goals, and dosing schedules. This complementary evidence is essential for understanding clinical utility and feasibility under routine service conditions.

Therefore, the primary aim of this study was to describe pre-post changes in gait function following a clinical series of treadmill-based VR biofeedback training in children and adolescents with congenital or acquired brain injury, thereby evaluating its real-world clinical utility. Given the task-specific, high-repetition nature of the training, we expected to observe measurable improvements in gait-function outcomes after the training series compared with baseline. The secondary aims were to explore whether pre-post changes were associated with (1) training amount and frequency, (2) treatment goals, and (3) diagnosis (CP vs. ABI).

Methods

Participants

This was an observational study, without a control or comparator group, based on a retrospective clinical dataset (Center 1) and previously collected prospective research data (Center 2). The absence of a comparator arm reflects the real-world clinical or research contexts in which data were collected, and the study was not designed as a controlled trial. Inclusion criteria were: (1) a confirmed diagnosis of CP or ABI, the latter due to trauma, brain tumour, stroke, encephalitis, anoxia, or arteriovenous malformation, (2) aged between 6 and 21 years of age, and (3) a minimum of six training sessions on the MOTEK Gait Realtime Analysis Interactive Lab (GRAIL). The data from center 1 (ALYN Hospital, Jerusalem, Israel) was collected between 2019 and 2024, retrospectively, as part of the routine clinical care (Prot. 075 − 23, date of approval: 13/02/2023). The retrospective protocol was approved by the local Ethics Committee in line with the Declaration of Helsinki. The data from center 2 (Scientific Institute IRCCS Eugenio Medea, Lecco, Italy) were collected as part of prospective research studies between 2014 and 2020 (Prot. 108, date of approval: 18/02/2014; Prot. 10-15CE, date of approval: 10/02/2015; Prot. 355, date of approval: 22/09/206; Prot. 454, date of approval:20/04/2017), and were included in the current study as anonymised. No a priori sample size calculation was performed, and the cohort represents all eligible participants available from the two centres during the study periods. All protocols were approved by the local Ethics Committee in line with the Declaration of Helsinki, and patients or their parents provided written informed consent. All methods were carried out in accordance with relevant guidelines and regulations, including institutional and national standards for research involving human participants. The context and rationale for treatment planning at each center, clinical discretion at Center 1 versus predefined protocols at Center 2, are described in detail in the Treatment Protocol section.

VR based biofeedback training

The Motek GRAIL is based around a split-belt treadmill with integrated force plates, incline/decline and medial/lateral sway. VR environments are projected on an immersive 180° semicircular screen in front of the treadmill, as well as on the treadmill itself, and synchronized with the treadmill movements. This has been shown to engender the perception of an experience similar to over-ground walking15. For the patient’s safety, the system is also equipped with two handrails and a harness. The system incorporates an optoelectronic motion capture system (Vicon, Oxford, UK) and three high-speed video cameras (Basler, Germany). The integration of all these features provides multi-sensorial feedback (visual, proprioceptive, and auditory) during training.

Treatment protocol

At both centers, the treatment aims were categorized according to the following: (1) gait quality, including gait pattern and symmetry, (2) gait function, including walking speed, walking endurance and dual tasking performance, and (3) balance, including balance function in standing and walking and falling prevention. Each patient was assigned one or several of these treatment goals (Table 1). Each training session lasted 45 min, with 40–45 min of active training using two to five training applications, increasing in difficulty according to the patient’s progress. Applications were selected to match each participant’s primary treatment aim (gait function, gait quality, or balance). For example, tasks targeting symmetry used visual feedback on step length or weight distribution, whilst balance-focused tasks involved perturbation. Some applications included tasks such as obstacle negotiation, walking on simulated uneven terrain or with treadmill perturbations (pitch/sway and belt acceleration/deceleration), and dual-task scenarios, particularly for participants whose treatment goals focused on balance. These elements were not isolated for analysis, and outcome measures were not stratified based on the specific content of the training applications. Most applications involved an element of gamification with a points-based feedback approach to encourage motivation and progression throughout the series. In both centers, biofeedback was externally focused, projected onto the screen in front of the treadmill and/or onto the treadmill surface itself, together with auditory cues. Although the number and frequency of sessions varied across participants, these were included as continuous variables in our analyses to examine whether treatment dosage was associated with outcome changes. An example of the GRAIL training can be seen in Fig. 1.

At Center 1, patients were referred by a clinical multidisciplinary team (physiotherapists, neurologists or orthopaedic consultants/surgeons), or as a recommendation following a clinical gait analysis. The treatments aim(s) were then discussed by the biofeedback team (five physiotherapists) in a weekly meeting, and if deemed suitable, an individually tailored protocol was created. The total number of sessions, number of treatments per week and length of the treatment series varied according to the clinical needs of the patient. Participants continued to receive their usual conventional clinical treatment during this time, including physical therapy, occupational therapy, hydrotherapy, and other relevant non-surgical interventions. GRAIL sessions were typically scheduled once or twice per week as part of this broader therapeutic program. However, the proportion of GRAIL relative to other therapies was not systematically documented in clinical records.

At Center 2, the total number of sessions and length of the treatment series varied according to the specific research project, although typically five times per week. Participants received a combined protocol of GRAIL training and conventional physical therapy, whilst according to patient’s needs, neuropsychological, speech, behavioural and occupational therapy was added. As in Center 1, detailed quantification of therapy distribution across disciplines was not consistently recorded.

Data harmonization and between-center variance

This study pooled an observational retrospective clinical dataset from Center 1 and prospective research datasets from Center 2. At both centers, the 6MWT, SSWS, and FAQ-WS were routinely used to evaluate gait-related function before and after GRAIL training and were administered using comparable clinical procedures (Outcome measures section). Prior to pooling, variable definitions and units were checked for consistency. Because protocols differed by site (clinically tailored practice at Center 1 vs. predefined research schedules at Center 2), center-level variability was treated as an inherent aspect of the real-world design, and diagnostic subgrouping and goal-stratified analyses were therefore performed across the combined cohort to examine heterogeneity of pre–post change.

Outcome measures

The following outcome measures were chosen to evaluate gait function for their established clinical utility in assessing functional mobility and walking performance in paediatric populations with neuromotor disorders, and are well-suited to evaluating the type of repetitive, task-oriented gait practice facilitated by the GRAIL system.

Six-minute walk test (6MWT)

The 6MWT is a self-paced walking test generally used to assess submaximal functional capacity and endurance. The main value is the distance that an individual can walk in six minutes when instructed to “walk as far as possible”16,17. The 6MWT was performed over-ground pre- and post-training.

Self-selected walking speed (SSWS)

Following a minimum familiarization period of six minutes, a self-selected comfortable fixed speed was determined according to feedback from the patient and their families. This was performed in the first and last session of the training.

Gillette functional assessment questionnaire walking scale (FAQ-WS)

The FAQ-WS is a parent/caregiver reported outcome measure to identify the patient’s usual level of function18. It assesses the level of a functional mobility on an ordinal scale by describing various levels of mobility. The scale ranges from one (the child cannot take any steps at all) to ten (the child walks, runs, and climbs on level and uneven terrain without difficulty). The FAQ was scored in the first and last session, following a short interview with the patient and primary caregiver.

Descriptive characteristics (gender, age, Gross Motor Function Classification System (GMFCS)), diagnosis, treatment information (i.e. training objective) were also extracted from each participant’s medical file. The GMFCS is a classification system which assesses the patient’s movements capabilities and use of mobility aids. It provides a rough description of the patient’s motor function, and a score relative to the type of equipment or mobility devices that he/she may need (e.g. wheelchairs, walking frames or crutches)19. Although the GMFCS was originally developed for children with CP, it has been applied in paediatric ABI populations to provide a coarse classification of motor function, particularly in studies aiming to compare across diagnostic groups20,21. It was used in this study to allow gross stratification of baseline function, whilst acknowledging that it is not formally validated for ABI.

Statistical analysis

Parametric and non-parametric statistics were used to describe the participants, based on skewness and kurtosis. The change in the outcome measures were evaluated using Wilcoxon test or t-test, as needed. Effect sizes were calculated using Cohen’s d for parametric data and r for the nonparametric data22. The difference between the study groups was assessed using repeated measure ANOVA for parametric variables and Mann Whitney U test for non-parametric variables. The relationship between outcome measures was assessed using a Spearman or Pearson correlation coefficient as needed. Statistical significance was set at p < 0.025 to account for two study hypothesis. Analyses were performed using SPSS version 29 (SPSS Inc., Chicago, IL, USA).

Results

Participants

Across both centres, 201 total records were screened. At Centre 1, four participants were excluded because they had longer than one series of biofeedback treatment or a duration of more than 120 days between the first and last session, and 54 were excluded for non-CP/ABI diagnoses. At Centre 2, eight participants were excluded due to age outside 6–21 years. Of the remaining 135 participants who were included, 51 were from Centre 1 and 84 from Centre 2. Sixty-one participants were diagnosed with CP and 74 with an ABI. The causes of an ABI included traumatic brain injuries (n = 20), brain tumors (n = 31), and stroke (n = 13). Ten participants had other diagnoses (encephalitis, Chiari malformation, Acute disseminated encephalomyelitis, and epilepsy). Cohort characteristics are presented in Table 1.

Table 1 Cohort characteristics.
Full size table

The amount and frequency of treatments throughout the series and the treatment aims are presented in Table 1. All participants received at least one treatment per week (median of 4 treatment per week, minimum 1-maximum 6). Fifty-five participants had one treatment aim, 50 participants had two treatment aims, and 30 participants had three treatment aims.

Clinical evaluation pre-post training

The 6MWT, SSWS and FAQ-WS values before and after the GRAIL treatment series are presented in Table 2. Missing outcome data were present at both centres. At Center 1 (retrospective clinical cohort), missing values were 6MWT Pre n = 2 / Post n = 8, SSWS Pre n = 1 / Post n = 1, and FAQ-WS Pre n = 0/ Post n = 0. At Center 2 (prospective datasets), missing values were 6MWT Pre n = 4 / Post n = 5, SSWS Pre n = 0 / Post n = 0, and FAQ-WS Pre n = 20 / Post n = 21. A significant improvement in the 6MWT distance was observed across the entire group with a mean improvement of 27.5 m (95% CI 20–41 m, p < 0.001), above the minimum clinical important difference23. Most participants showed improvements in the 6MWT (n = 84). When comparing the CP and ABI groups, both demonstrated significant improvement: the first improved by 20 m (95% CI 6–34 m, p = 0.005) and the second improved by 40 m (95% CI 25–56 m, p < 0.001), whilst there was a trend for an interaction effect with the ABI group presenting greater improvement (F1,120 = 3.748, p = 0.055) (Fig. 2). When assessing the influence of treatment aims on the changes in 6MWT, participants with a primary goal of improving gait function (n = 62) presented significantly greater improvements with respect to those without a goal of improving gait function (mean change of 45 m, 95% CI 29–62 m in comparison to 16 m, 95% CI 4–28 m; F1,120 = 8.280, p = 0.005). The treatment aims of gait quality and balance did not influence the changes in 6MWT. In addition, the amount and frequency of treatments through the series did not present a significant correlation with the changes in 6MWT.

Most participants improved their SSWS (n = 97), whilst the entire group demonstrated a significant improvement in the SSWS (mean increase of 0.11 m/s, 95% CI 0.09–0.14 m/s, p < 0.001). To our knowledge robust MCID benchmarks for SSWS have not been established for mixed pediatric CP/ABI cohorts, and SSWS change is therefore interpreted based on magnitude and 95% confidence intervals. When comparing the CP and ABI groups, both groups had significant improvements; the CP group improved by 0.12 m/s (95% CI 0.08–0.15 m/s, p < 0.001) and the ABI group improved by 0.11 m/s (95% CI 0.08–0.14 m/s, p < 0.001). No interaction effect was found (F1,132 = 0.089, p = 0.766) (Fig. 2). None of the treatment aims influenced the SSWS changes. The amount and frequency of treatments through the series presented weak relationships with the changes in SSWS (rs= 0.27, p = 0.002 and rs= − 0.25, p = 0.015, respectively).

A significant difference was presented in the FAQ-WS score across the entire group (median = 0, IQR= − 1 to 2), p < 0.001), although most participants did not experience a change in their FAQ-WS score (n = 93, 82%); 20 participants improved, and one participant showed a decline. When comparing groups, only the ABI group demonstrated a significant improvement (median = 0, IQR= − 1 to 1; p < 0.001), with a close to significant improvement in the CP group (median = 0, IQR = 0–2; p = 0.059). A significant interaction effect was approached (Z = − 2.208, p = 0.027), with the ABI group presenting a larger effect size. However, most participants in both the CP group (n = 51, 88%) and the ABI group (n = 42, 75%) showed no change in FAQ-WS score. Treatment aims and the amount and frequency of treatments through the series did not influence changes in FAQ-WS score.

Table 2 Outcome measures before and after the GRAIL treatment series.
Full size table

Discussion

The primary aim of this study was to describe pre-post change in gait function following a clinical series of treadmill-based VR biofeedback sessions in children and adolescents with CP or ABI. We observed improvements in the 6MWT and SSWS after the training series compared with baseline. These findings should be interpreted as real-world pre-post change rather than causal evidence of effectiveness, given the observational design and concurrent therapies. Whilst this has been previously reported, the current study was based on a larger sample size of individuals with data from two unrelated rehabilitation centers in different countries, employing different intensities and frequencies of training and improving the generalizability of these finding. Lastly, both individuals with CP and ABI were included in the same study, providing relative insight into how both groups respond to treadmill-based VR biofeedback training.

With respect to the 6MWT, a previous study on 17 children with CP diplegia found a mean increase in 6MWT score of 11.8 m following 18 training sessions on the GRAIL11. In the current study, the individuals with a diagnosis of CP diplegia presented with a slightly larger increase in 6MWT of 18.2 m, whilst the individuals with CP hemiplegia presented a mean increase of 26.8 m. This may relate to the lower mean GMFCS level of the individuals with hemiplegia with respect to diplegia, suggesting more potential for improving submaximal functional capacity and endurance. With respect to the ABI cohort, a previous study on 12 individuals with an ABI (11 hemiplegia/1 ataxic) found a mean increase in 6MWT score of 168 m following ten 30-minute training sessions four times a week12. This is far larger than the mean increase of 40 m identified in the current study. However, another study in six individuals with an ABI found an increase in mean 6MWT score of 20 m following twenty 45-minute training sessions over a period of four weeks7. The discrepancy between the three studies may relate to the heterogeneity of the diagnosis of ABI that encompasses a diverse group of brain injuries, of which each sub-type may respond differently to VR-based biofeedback training. In addition, it may also relate to the different number of treatment sessions. However, all studies reported significant improvement in the 6MWT which emphasis the benefit of using treadmill-based VR biofeedback training to improve walking endurance.

With respect to the SSWS, a previous study on 16 children with CP diplegia found a mean increase in SSWS of 0.17 m/s following 18 training sessions on the GRAIL over a period of four weeks11. This is larger than the mean increases in SSWS for the CP cohort in the current study (0.12 m/s), whilst there was no meaningful difference in mean walking speed change when comparing sub-cohorts of diplegia and hemiplegia (0.11 m/s and 0.12 m/s, respectively). With respect to the ABI cohort, a previous study in six individuals with an ABI increased their mean SSWS by roughly 0.15 m/s following twenty 45-minute training sessions over a period of one month7, larger than the 0.11 m/s increase in the current study. The larger increases in SSWS in both previous studies may be related to the larger number of treatment sessions that were performed over a shorter period, suggesting a greater influence on gains in SSWS. This is further supported by the weak but statistically significant correlation between the amount and frequency of treatments and the SSWS observed in the current study.

Whilst there was a significant improvement in the FAQ-WS following the training, only 18% of the participants had a change in their FAQ-WS score, indicating limited responsiveness in this cohort. This was more noticeable in the ABI cohort, in line with a previous study where individuals with an ABI presented a mean increase in FAQ-WS score from eight to nine12. The median baseline FAQ-WS of eight for the CP cohort in the current study is the same as was reported in a previous study in individuals with bilateral CP11. The FAQ-WS differs with respect to the 6MWT and SSWS as it is based on the subjective opinion of the parent/caregiver of the child/adolescent, having a high correlation with a more robust parent reported measure of gait function, the gait outcomes assessment list24. However, with respect to the current study, most individuals were highly functioning at baseline, suggesting a potential celling effect and limited sensitivity of the FAQ-WS for detecting change in this population. Other scales such as the FAQ-22-item skill set, or gait outcomes assessment list may be more appropriate in future studies24,25.

Aside the larger sample size of individuals, this study had three innovative components with respect to earlier studies. The first is the inclusion of participants from two different rehabilitation centres in different countries, in which the amount and frequency of treatments sessions differed. While pooling data from two centres enhances generalisability, it also introduces real-world heterogeneity in treatment context and data collection (retrospective vs. prospective), and outcome-specific missingness. Therefore, the observed pre–post changes should be interpreted as descriptive clinical-utility findings rather than centre-controlled estimates. Notably, centre-related variability was not meaningfully associated with changes in 6MWT or FAQ-WS, and showed only weak associations with SSWS change. The latter finding likely relates to earlier studies where larger increases in SSWS were identified in smaller cohorts of individuals with CP and ABI undergoing higher frequencies and intensities of treatments with respect to the current study7,11. Whilst that may be the optimal paradigm, it may not be clinically sustainable or feasible for individuals to arrive at a rehabilitation centre on an almost daily basis for treadmill-based VR biofeedback training. Therefore, it is encouraging to learn that it is possible to improve measures of the 6MWT, SSWS and FAQ-WS with fewer treatment sessions over a longer period, such as once or twice per week.

The second innovative component is related to the delineation of treatment goals for the participants of this study into gait function, gait quality and/or balance. Participants who had a primary treatment goal of improving gait function demonstrated additional improvement in endurance as measured by the 6MWT. This could be explained by the relatively extended periods of time those participants practiced in the goal-specific training applications. However, whilst this study focuses on outcome measures related to gait function, that was not the primary aim for all participants. Treatment goals of improving gait quality or balance still require being active in standing and walking during the treatment sessions and so may also positively influence measures of gait function. This appeared to be the case for the improvements in walking speed and walking ability, where treatment goals did not influence SSWS and FAQ-WS changes following the VR-based biofeedback training.

The third innovative component is the inclusion of individuals with CP and ABI in the same study, reflecting the diagnostic heterogeneity commonly encountered in paediatric neurorehabilitation. Although these conditions differ in aetiology, lesion timing, and neurodevelopmental trajectory, they share persistent gait impairments that are typically addressed using similar therapeutic strategies, including task-specific motor training and extrinsically focused feedback. The integration of both cohorts within a single therapeutic framework was therefore intended to reflect real-world clinical implementation of treadmill-based VR biofeedback. To account for potential diagnostic differences, we stratified key parts of our analysis by group. Accordingly, the CP–ABI comparisons are presented to describe heterogeneity in observed pre–post change under routine care, rather than to imply identical premorbid gait trajectories or response mechanisms. Notably, the ABI cohort demonstrated a tendency for greater improvements in the 6MWT and the FAQ-WS with respect to the CP cohort. This may relate to the timing of the brain lesion that occurred in the ABI cohort at a later stage of development, and as such may provide a window of greater modifiability with additional capabilities for neuroplasticity. Additionally, individuals with CP and ABI have also been shown to present with impaired function of the cardiac autonomic control system26,27, which may be attributed to their endurance and the 6MWT performance. Whilst direct autonomic measurements were not conducted, the underlying neurological differences between these conditions may influence cardiovascular responses to exercise. The intact neural pathways in ABI patients could be related to more preserved cardiac autonomic nervous system regulation, potentially allowing for more efficient cardiovascular adaptation and improved exercise tolerance during training. This may consequently enhance performance in the 6MWT. Further studies should investigate this hypothesis in more detail. Nonetheless, combining CP and ABI participants can introduce potential confounding due to underlying pathophysiological differences. Future studies would benefit from diagnosis-specific designs or stratified protocols to more precisely delineate treatment responsiveness across paediatric neurorehabilitation populations.

Whilst this study focused on evaluating functional outcomes associated with treadmill-based VR biofeedback training, the clinical utility of such systems extends beyond performance metrics. Large-scale immersive systems like the GRAIL offer unique opportunities for targeted feedback, gamified engagement, and real-time modulation of training environments. Future investigations should examine how specific system capabilities are used to support individualized therapy goals, how these align with therapist decision-making, and how they compare to alternative VR delivery methods, such as head-mounted displays or mobile platforms. Understanding these implementation dynamics would provide valuable insights for clinicians and developers seeking to optimize the integration of VR in rehabilitation.

This study has several limitations. Most importantly, there was no control group of individuals with CP or an ABI undergoing only their routine rehabilitation, and it is likely that improvements in gait function could also observed. Whilst the variation in session number, frequency, and treatment goals introduces heterogeneity, this reflects individualized clinical practice. We attempted to account for this by including these variables in secondary analyses and discussing their influence on outcomes. Total training dose in hours was not systematically recorded, so session count and frequency were used as dose proxies, limiting dose–response conclusions. In addition, treatment duration varied widely (Table 1), and we did not statistically control for elapsed duration. Thus, longer training series may have allowed greater consolidation of motor learning independent of session count, which could confound pre–post change. Anecdotally, the general feedback from patients at both centres regarding treadmill-based VR biofeedback was very positive and favourable, more engaging than traditional treatment and increased motivation and treatment adherence. However, these impressions should be interpreted cautiously. Our cohort includes only children who completed a training series and had pre–post assessments, it may under-represent individuals for whom GRAIL training was not acceptable or feasible from the outset; thus, anecdotal acceptability feedback may be subject to selection bias. In addition, the amount and frequency of conventional therapies (e.g., physiotherapy, hydrotherapy) delivered alongside GRAIL training were not systematically documented, which limits the ability to isolate the contribution of the VR intervention within the broader rehabilitation context. Whilst the 6MWT was performed overground, the SSWS was performed on the treadmill. Nevertheless, therapists asked the patient to define the comfortable self-selected walking speed during the adaptation period on the treadmill. The chosen measures of gait function have been validated for use in children and adolescents with CP and an ABI. However, this still only provides a partial picture of gait function. Future studies should also incorporate measures of physiological parameters, such as energy expenditure, heart rate variability, as well as more robust patient reported outcome measures. Lastly, whilst our study categorized treatment aims and demonstrated their association with some outcomes, we did not formally evaluate how individual GRAIL applications or feedback modalities were selected in relation to specific therapy goals. As such, we are unable to comment on the fidelity of alignment between treatment content and the system’s unique capabilities. Future work should incorporate structured documentation of therapist decision-making and system usage to better understand implementation strategies and optimize goal-directed application of VR-based interventions, along with long-term follow-up to evaluate carry-over.

Conclusion

Children and adolescents with CP and ABI demonstrated higher gait-function measures after the treadmill-based VR biofeedback training series than before. These findings are consistent with prior reports and extend the literature by describing pre–post changes in a larger multicenter cohort with clinically variable dosing schedules. Improvements were greater among participants with gait-function-related treatment goals, whereas associations with the amount and frequency of sessions were weak. Participants with ABI showed larger gains in walking endurance (6MWT) than those with CP, while changes in self-selected walking speed were comparable between diagnostic groups. Taken together, the results indicate real-world clinical utility and acceptability of treadmill-based VR biofeedback training, which appears to provide an engaging and motivating rehabilitation environment for gait-focused practice. Controlled trials with systematic dose quantification and responsive patient-reported outcomes are now needed to establish causal effectiveness, define optimal training paradigms, and refine knowledge translation into routine paediatric neurorehabilitation.

Fig. 1
Fig. 1
Full size image

Visual examples of GRAIL training. Each application promotes one of the specified treatment aims through gamification and a point-based feedback system to encourage progression. The three aims are (1) gait quality, including gait pattern and symmetry, (2) gait function, including walking speed, walking endurance and dual tasking performance, and (3) balance, including balance function in standing and walking and falling prevention.

Fig. 2
Fig. 2
Full size image

The distribution of the 6MWT and SSWS change in both the CP and the ABI groups. 6MWT, six-minute walk test; SSWS, self selected walking speed; ABI, acquired brain injury; CP, cerebral palsy.