This study was approved by the Institutional Review Board of Yonsei University Mirae Campus (Approval No. 1041849-202508-BM-171-03), and participant recruitment began after approval. Written informed consent was obtained from the legal guardians, and assent was obtained from the children when appropriate. Data were de-identified using study codes and managed according to institutional regulations. This study was conducted in accordance with the principles of the Declaration of Helsinki.

Participants were children and adolescents (8--15 years) with developmental disabilities (e.g., autism spectrum disorder, cerebral palsy, and chromosomal disorders). Eligible participants were those classified as standard GMFCS levels I--III (individuals with levels IV and V were excluded). These criteria targeted a population capable of walking independently or with minimal assistance, following multi-step instructions, and safely participating in ~30-minute exercise sessions. Furthermore, participants must have received rehabilitation therapy within the past 6 months and have a caregiver available to support home participation.

The exclusion criteria were as follows: medical instability limiting safe exercise, recent surgery (<6 months) or botulinum toxin injection (<4 months), moderate-to-severe sensory impairment, severe behavioural/emotional problems affecting safety, concurrent participation in similar programs, or caregiver inability to participate.

The target sample comprised 16 participants (8 per group). As this was a low-risk QI exploratory pilot study, the sample size was determined pragmatically based on feasible recruitment and the clinical capacity to deliver safe small-group rehabilitation sessions, rather than through a formal power calculation. Because the study was not designed or powered for definitive hypothesis testing, this pragmatic approach was consistent with its dual aims: to assess feasibility (recruitment/retention, implementation fidelity, and safety) and to generate preliminary estimates of intervention effects (Leon et al., 2011). Accordingly, this sample should be regarded as appropriate for pilot feasibility purposes, but not for drawing confirmatory conclusions regarding efficacy.

The GMFCS level was recorded to describe the functional status at baseline. Although GMFCS was developed for cerebral palsy, it was used conservatively as a functional descriptor across heterogeneous diagnoses, and interpretations were made accordingly (Palisano et al., 1997).

This was an 8-week, exploratory randomised pilot study (parallel-group) conducted in a single tertiary hospital rehabilitation setting in Seoul, South Korea. Participant recruitment occurred in 2025. Participants were individually randomised (1:1) to a hospital-based SRP or a HEP. Randomisation was stratified according to the GMFCS level (I--III) and implemented using a prespecified permuted-block sequence. Allocation concealment was ensured using sequentially numbered, opaque, sealed envelopes prepared and administered by an independent third party not involved in recruitment, intervention delivery, or outcome assessment.

The blinding of participants and intervention therapists was not feasible. Outcome assessments were conducted by assessors blinded to group allocation. Assessments were performed at baseline (T0) and 8 weeks after the intervention (T1) using standardised procedures (scripted instructions, assessor training, and predefined scoring criteria).

New concurrent therapies during the intervention period were discouraged to reduce confounding from co-interventions. Usual care was requested to remain stable when possible, and any changes (type, frequency, and duration) were recorded using a standardised log. Figure 1 shows the flow of the participants.

Outcomes were prespecified as primary (motor function and balance), secondary (participation and engagement), and exploratory (goal attainment) measures based on clinical relevance. Primary and secondary outcomes were assessed at T0 and T1, whereas the exploratory GAS T-scores were derived from goal attainment ratings only at T1, reflecting the inherent design of the tool. COPM and GAS were administered using researcher-defined standardised common goals aligned with the intervention tasks and instrument requirements rather than individualised goals to enable between-group comparison using common sport-related functional task demands (Kiresuk & Sherman, 1968; Law et al., 1990). The same prespecified goal set and standardised rating anchors (detailed in Supplementary File S1) were applied in both groups according to the study protocol.

Primary outcomes: GMFM (0--100%) and PBS (0--56) were assessed. GMFM summarised gross motor performance as a percentage score based on standardised item performance (Russell et al., 1989). PBS quantified functional balance performance across sitting, standing, and weight-shift tasks with a total score ranging from 0 to 56 (Franjoine et al., 2003).

Secondary outcomes: COPM--Performance and COPM--Satisfaction (1--10) captured caregiver ratings for the prespecified common goal tasks (Law et al., 1990). The PVQ (15--60) was used to assess the observed volition and engagement during the usual-care therapy sessions using standardised anchors. The usual-care therapist, who was blinded to group allocation, rated the PVQ, and the rater received prespecified training on anchor use and scoring procedures (Andersen et al., 2005).

Exploratory outcome: GAS T-scores at T1 were derived from goal attainment ratings for the standardised common goals using a 5-point scale from -2 to +2 and converted to T-scores based on the original GAS framework (Kiresuk & Sherman, 1968). The findings of the GAS were interpreted in the context of evidence from pediatric rehabilitation (Steenbeek et al., 2007).

SRP: Participants in the SRP group attended supervised sessions delivered by a physical therapist who also held a certified sports coaching qualification, in a designated hospital space, twice weekly for 45 minutes over 8 weeks (16 sessions). Each session included a warm-up, task-oriented core activities, and a cool-down. The core activities included static/dynamic balance, coordination/agility, functional mobility, and small-group game-based sports skills. Motor learning principles were followed for function-based pediatric interventions (Levac et al., 2009). The exercise intensity was individualised to an OMNI Rating of Perceived Exertion (OMNI-RPE) of 5--7 (0--10 scale) (Utter et al., 2002). The task difficulty was progressed using a standardised performance-based rule. The next difficulty level was introduced when the target quality of performance was achieved on the same task in two consecutive opportunities under standardised conditions consistent with the principles of progressive overload (ACSM, 2009).

HEP: The HEP group completed caregiver-led home sessions three times weekly for approximately 30 minutes over 8 weeks (24 sessions). Before initiation, caregivers received standardised instruction (~30 minutes) on task execution, safety, and recording procedures. During the program, weekly telephone coaching (10--15 minutes) was provided to monitor adherence, provide feedback on performance errors, and adjust task difficulty when needed. The intervention content and intensity targets were aligned with the SRP (OMNI-RPE 5--7) (Utter et al., 2002), and the same progression rule was applied (ACSM, 2009).

Implementation fidelity and safety monitoring: Implementation fidelity and feasibility were assessed using prespecified adherence metrics, including session completion (SRP: attendance-based; HEP: completion-based), coaching participation (HEP), weekly exercise time (min/week), and total exercise time over 8 weeks (min). Feasibility success was defined a priori as achieving a ≥75% threshold for each metric. This pragmatic clinical benchmark accommodates the population's logistical challenges and aligns with the methodological guidelines for pilot progression criteria (Thabane et al., 2010). Safety was monitored by documenting adverse events, defined as any untoward medical occurrence during participation requiring task modification, temporary suspension, or medical evaluation (e.g., falls, musculoskeletal injury, marked pain exacerbation, dizziness, or cardiopulmonary symptoms). When an adverse event occurred, the intervention was modified or discontinued as appropriate, and events were documented and reported according to institutional procedures.

All analyses were performed using IBM SPSS Statistics version 28 (IBM Corp., Armonk, NY, USA). Categorical variables are presented as n (%), and continuous variables as mean ± SD or median (IQR), as appropriate. The normality and homogeneity of variance were assessed using the Shapiro--Wilk and Levene's tests, respectively. Since all primary outcome variables met these assumptions, parametric mixed ANOVAs were employed for the repeated-measures analyses, obviating the need for nonparametric alternatives. Baseline characteristics were compared using the independent-samples t-test or Mann--Whitney U test for continuous variables and Fisher's exact test for categorical variables.

For outcomes measured at T0 and T1, a two-way mixed ANOVA was conducted with group (SRP vs. HEP) and time (T0 vs. T1) as the between-subject and within-subject factors, respectively. The effect sizes for the ANOVA results were reported as partial η². The GAS T-scores measured at T1 were compared between groups using an independent-samples t-test, with Cohen's d reported as the effect size. The analyses followed the intention-to-treat principle. There were no missing outcome data; therefore, no imputation was required. All tests were two-sided with an alpha level of 0.05.

Several limitations should be noted. As this is a single-center exploratory study with a small sample (n = 16), the results are not generalisable, and session frequency could not be included as a covariate in the statistical model. Additional constraints include diagnostic heterogeneity, conservative use of the GMFCS, PVQ context dependence, reduced responsiveness due to the common-goal GAS approach, and the absence of long-term follow-up (Andersen et al., 2005; Steenbeek et al., 2007).

Future studies should recruit larger samples with functional level stratification and incorporate 3--6-month follow-up assessments. Measurement rigor should be strengthened using PVQ scoring based on recorded sessions to improve standardization. A mixed GAS approach that combines individualised goals with standardised scoring rules may better balance clinical meaning and comparability. Finally, family-level outcomes should be included, and models to identify which families benefit the most from HEP may improve targeting.