Effects of game-based virtual reality balance training on postural stability and mobility in community-dwelling older adults: an exploratory randomized controlled trial

Abstract

Background

Falls are a major contributor to reduced independence in older adults and are associated with both physical decline and impaired cognitive–motor integration. Emerging evidence suggests that game-based virtual reality (VR) training may enhance engagement and balance performance; however, its application in community-based settings remains limited.

Objective

To examine the effects of an individualized, game-based VR balance training program on postural stability, functional mobility, and fall-related concern in community-dwelling older adults.

Methods

In this exploratory randomized controlled trial, 40 participants aged ≥ 65 years were assigned to either a VR-based balance training group or a control group receiving unsupervised home-based exercise guidance. The intervention group completed an 8-week supervised VR training program using the Biodex Balance System (2 sessions per week for 8 weeks). The control group received exercise videos and health education without direct supervision or formal adherence monitoring; however, participants received brief weekly telephone reminders to encourage engagement. Primary outcomes included postural sway indices measured by the Biodex Balance System under single- and dual-task conditions. Secondary outcomes included the Timed Up and Go (TUG) test, Single-Leg Stance (SLS), and Falls Efficacy Scale–International (FES-I). Data were analyzed using generalized estimating equations, with group × time interaction effects reported.

Results

All participants completed baseline and post-intervention assessments. Compared with the control group, the intervention group demonstrated statistically significant group × time interactions for multiple sway indices under single-task conditions and selected measures under dual-task conditions (p < 0.05), indicating differences in change over time between groups. Statistically significant interactions were also observed for functional mobility (TUG) and SLS performance (p < 0.05). FES-I scores increased in both groups, with a statistically significant group × time interaction observed, indicating an increase in self-reported concern about falling. This finding should be interpreted cautiously. Sensitivity analyses showed generally consistent results, although potential sex-related effects were observed and should be interpreted as exploratory. No adverse events were reported.

Conclusions

An 8-week game-based VR balance training program was associated with changes in postural stability and functional mobility in community-dwelling older adults. Changes in fall-related concern were also observed. These findings support the feasibility of a supervised, technology-assisted training approach, although confirmation in larger trials is needed.

Trial registration

ClinicalTrials.gov Identifier NCT05779189

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12877-026-07914-9.

Background

Falls remain a leading cause of disability, morbidity, and mortality in older adults worldwide, significantly impairing independence and quality of life [1]. Globally, approximately 30% to 35% of community-dwelling older adults experience at least one fall each year [2, 3]. This increasing incidence underscores the urgent need for multidimensional interventions targeting both physical and cognitive contributors to imbalance.

Traditional exercise interventions, such as strength or flexibility programs, have demonstrated benefits in fall prevention but often suffer from poor adherence due to their repetitive nature [4]. Recent advances in immersive technologies, particularly game-based virtual reality (VR) training, offer new opportunities to enhance engagement and effectiveness. VR-based programs integrate visual, auditory, and kinesthetic feedback, allowing simulation of real-life balance tasks within controlled environments. Evidence suggests that VR-based training, either alone or combined with conventional exercise, is as effective as or more effective than traditional programs, particularly in improving gait and balance functions in older adults [5, 6]. Additionally, immersive VR has been associated with improvements in gait and postural adaptability, especially when sufficient training exposure is achieved [7].

Beyond physical enhancement, dual-task VR training involves simultaneous cognitive and motor engagement, which is associated with fall risk and may aid fall prevention [8]. Age-related cognitive decline may reduce multitasking efficiency, thereby increasing fall risk during daily activities such as walking while conversing. Recent evidence indicates that VR-based interventions can improve dual-task performance and facilitate better integration between cognitive and motor functions in older adults [8]. Furthermore, VR-based cognitive–motor training programs have demonstrated potential benefits in executive function, reaction time, and broader aspects of cognitive health [9, 10].

Parallel to VR technology, instrumented balance assessment systems such as the Biodex Balance System have gained clinical relevance as standardized tools for quantifying postural stability. The Biodex Balance System SD is widely used in both clinical and research settings to assess balance performance and fall risk, providing objective measures of postural stability and real-time visual biofeedback during rehabilitation [11, 12]. Reliability studies have demonstrated consistent inter-rater and test–retest performance for these measures [12]. Recent developments have also expanded its application within technology-assisted and individualized rehabilitation programs for older adults [13].

Another emerging approach involves hybrid interventions that combine conventional exercise with interactive digital feedback systems. Evidence supports the use of VR-based and technology-assisted training approaches to enhance adherence and accessibility, including home-based or semi-supervised formats [13, 14]. These approaches have shown promising effects in supporting mobility and functional performance in older populations.

Despite these advancements, important gaps remain. Although a growing number of studies have included community-dwelling older adults [5, 13, 14], many interventions are conducted under structured or supervised conditions, which may limit generalizability to real-world settings. In addition, relatively few randomized controlled trials have examined individualized, game-based VR interventions that incorporate adaptive dual-task progression within a structured framework [8, 13]. Furthermore, psychological outcomes such as fall-related self-efficacy may be influenced by cultural and linguistic factors, including variations in how concern about falling is interpreted across different versions of the Falls Efficacy Scale. These considerations highlight the need for cautious and context-sensitive interpretation of such outcomes.

This study therefore aimed to explore the effects of an individualized, game-based VR balance training program incorporating physical, cognitive, and technology-assisted components on postural control, functional mobility, and fall-related self-efficacy in community-dwelling older adults.

The novelty of this study lies in the structured integration of three key components within a single intervention framework: (1) real-time quantitative balance assessment using the Biodex system; (2) interactive VR-based training incorporating dual-task demands with progressive adaptation of task difficulty (e.g., adjustment of task speed, visual complexity, and cognitive load based on participant performance); and (3) supervised implementation within a community-dwelling older adult population.

This integrated design extends prior work in which these components have often been examined separately. Rather than focusing on a single modality, the present study combines quantitative assessment, interactive training, and supervised delivery within a unified protocol. From a methodological perspective, this approach provides a pragmatic framework that reflects a structured, technology-assisted intervention under supervised conditions, rather than an isolated evaluation of VR efficacy alone. To our knowledge, relatively few randomized trials have implemented such an integrated and structured approach in community-dwelling older adults.

Methods

Study design

This exploratory randomized, parallel-group interventional study was conducted between July 1, 2022, and February 28, 2023. Participants were randomly assigned (1:1) to the experimental or control group using a computer-generated randomization sequence produced by an independent statistician. Allocation assignments were placed in sequentially numbered, opaque, sealed envelopes and were opened by the intervention coordinator at the time of group assignment, ensuring allocation concealment.

Stratified randomization by sex had been planned; however, due to the small number of male participants (n = 4), all male participants were allocated to the experimental group by chance, despite appropriate allocation concealment. This random imbalance is recognized as a potential occurrence in small-sample trials and was addressed through prespecified sensitivity analyses that included sex as a covariate.

Because of the nature of the intervention, neither participants nor outcome assessors were blinded; therefore, the study was conducted as an open-label trial. This lack of blinding may introduce potential performance and assessment bias and should be considered when interpreting the findings.

All study procedures were conducted in accordance with the Declaration of Helsinki and were approved by the Institutional Review Board of Taipei Medical University (IRB No. 202204048). The trial was registered at ClinicalTrials.gov (identifier: NCT05779189). All participants provided written informed consent prior to enrollment. This study is reported in accordance with the CONSORT 2025 statement [15]. The study flow and randomization process are presented in Fig. 1.

Fig. 1.

Study flow chart of participant recruitment, allocation, follow-up, and analysis

Participants

Community-dwelling older adults (≥ 65 years) were recruited from the Taipei Medical University Active Aging University and affiliated community centers. A total of 40 participants were enrolled: 20 in the experimental group (4 men, 16 women) and 20 in the control group (all women).

Although randomization was properly implemented, all four male participants were allocated to the experimental group by chance, resulting in a sex imbalance. Fisher’s exact test indicated that this difference was not statistically significant (p = 0.106). All participants were retained in the primary analyses, and additional sex-adjusted sensitivity analyses were conducted (see Supplementary Materials).

Inclusion criteria were: (1) age ≥ 65 years; (2) ability to ambulate independently or with an assistive device; and (3) willingness to complete the intervention and assessments. Exclusion criteria included: (1) cognitive impairment, assessed based on self-reported medical history and the participant’s ability to comprehend and follow study instructions; (2) major neurological disorders (e.g., stroke, Parkinson’s disease); (3) severe osteoporosis or major lower-limb joint surgery; and (4) contraindications to exercise participation.

Intervention

The intervention was delivered under supervised conditions by the study investigator, who had formal training in gerontology, exercise science, and research design related to older adults. Investigators were present during all sessions to ensure participant safety and protocol adherence. Standardized, minimal guidance was provided when necessary (e.g., correcting posture or providing brief encouragement), while participants actively performed the training tasks themselves. Participants were not instructed to modify their habitual physical activity and were asked to attend scheduled sessions and adhere to the assigned protocol.

Experimental group

Participants received game-based virtual reality (VR) balance training using the Biodex Balance System (BIODEX Medical Systems Inc., Shirley, NY, USA) (Fig. 2). Each session lasted 30 min, with 1–2 sessions per week over 8 weeks, for a total of 16 sessions, followed by 10 min of standardized verbal health education. While scheduling constraints occasionally resulted in one session per week, all missed sessions were rescheduled, and all participants completed the full 16-session protocol (100% adherence). Training tasks included ball catching, river rowing, marble maze, and word spelling (Fig. 3). These tasks incorporated both motor and cognitive (dual-task) components, with difficulty progressively adjusted based on individual performance. Specifically, task parameters such as speed, visual complexity, and cognitive load were modified in real time to provide adaptive progression.

Fig. 2.

Photographs of a participant using the Biodex Balance System during VR training. (a) Back view with weight shifted to the left. (b) Side view with weight shifted to the right. (c) Side view with weight shifted backward

Fig. 3.

Game-based training tasks in the Biodex module. (a) Ball catching. (b) River rowing. (c) Marble maze. (d) Word spelling

Control group

Participants received 10 min of verbal health education and were provided with five YouTube videos demonstrating balance exercises (7–15 min each). They were encouraged to practice independently according to their physical condition, defined as their perceived physical capacity, including fatigue level, balance confidence, and minor physical discomfort. No direct supervision, adherence tracking, or monitoring of training intensity was implemented in the control group. However, participants received brief weekly telephone reminders during the 8-week intervention period to encourage continued engagement with the prescribed exercises. These calls were intended to provide minimal engagement support, although no adherence data were collected. Despite this, the actual training dose (e.g., frequency, duration, and intensity) could not be quantified. This design reflects a pragmatic comparison with usual, self-directed exercise behavior in community settings.

Outcome measures

Assessments were conducted at baseline and within 2–3 days after completion of the intervention, on a separate day from the final training session to minimize the effects of acute fatigue.

Primary outcomes

Balance function: Postural stability was assessed using the Biodex Balance System, including Overall (OA), Anterior–Posterior (AP), and Medial–Lateral (ML) stability indices. These indices represent the variability of platform displacement and are expressed in degrees (°), with higher values indicating poorer balance performance. The modified Clinical Test of Sensory Integration of Balance (m-CTSIB) was also conducted under four conditions (firm/foam surface × eyes open/closed).

Postural control was evaluated under both single-task and dual-task conditions. In the dual-task condition, participants performed concurrent visual and auditory discrimination tasks while maintaining balance. Colored squares (red, blue, green, yellow) appeared every 2 s, and participants verbally identified the color. Simultaneously, high- and low-pitched tones were presented, and participants responded verbally (“high” or “low”).

Secondary outcomes

Physical function: Timed Up and Go (TUG) test and Single-Leg Stance (SLS; right and left, eyes open, maximum 30 s). The TUG test was performed at a self-selected comfortable walking speed, and trials were repeated if participants deviated from their habitual pace to ensure consistency.

Concern about falling: Assessed using the Falls Efficacy Scale-International (FES-I) [16], Taiwan Chinese version [17] (16 items, total score range 16-64), with higher scores indicating greater concern about falling. Assessed using the Taiwan Chinese version of the Falls Efficacy Scale–International (FES-I; 16 items, total score range 16–64), with higher scores indicating greater concern about falling [17].

Sample size considerations

No formal a priori sample size calculation was performed, as this study was designed as an exploratory randomized controlled trial. Observed effect sizes were calculated to describe the magnitude of treatment effects (e.g., Cohen’s d for TUG = 1.05). Effect sizes and 95% confidence intervals are presented to indicate the precision of the estimates and to inform sample size planning for future definitive trials.

Statistical analysis

Analyses were performed using IBM SPSS Statistics (version 31.0; IBM Corp., Armonk, NY, USA). Baseline differences between groups were assessed using independent-samples t-tests for continuous variables and Fisher’s exact test for categorical variables.

Intervention effects were evaluated using generalized estimating equations (GEE), including terms for group, time, and the group × time interaction. Group was coded as 0: control and 1: experimental, whereas time was coded as 1: pre-intervention and 2: post-intervention. The “estimate” reported in the Results represents the regression coefficient of this interaction term, reflecting the between-group difference in change over time.

Each model specified a normal distribution with an identity link function, an exchangeable working correlation structure, and robust (sandwich) covariance estimators. GEE was selected to account for within-subject correlations and to provide population-averaged estimates.

Given the exploratory nature and small sample size of this study, GEE estimates may be less stable and should be interpreted with caution. No adjustment for multiple comparisons was applied given the exploratory nature of the study; therefore, findings should be interpreted as hypothesis-generating. This approach may increase the risk of Type I error and requires cautious interpretation.

All statistical tests were two-tailed, with significance set at p < 0.05. Effect estimates are reported with 95% confidence intervals. Directionality of estimates should be interpreted according to each outcome (e.g., negative values indicate improvement when lower scores represent better performance, such as in TUG).

Results

Participants (Table 1)

Table 1.

Participant Characteristics (n = 40)

Characteristic	Experimental (n = 20)	Control (n = 20)	p value
Male sex, n (%)	4 (20%)	0 (0%)	0.106a
Age, years (mean ± SD)	69.9 ± 4.3	70.8 ± 4.7	0.55
Height, cm (mean ± SD)	158.8 ± 6.5	156.4 ± 4.8	0.20
Weight, kg (mean ± SD)	58.7 ± 7.7	56.4 ± 8.9	0.39
BMI, kg/m² (mean ± SD)	23.3 ± 2.8	23.1 ± 3.6	0.84

Data are presented as mean ± SD for continuous variables and n (%) for categorical variables. Independent-samples t-tests were used for continuous variables and Fisher’s exact test for categorical variables

^a Fisher’s exact test

All 40 participants completed both baseline and post-intervention assessments (experimental = 20; control = 20). Adherence in the control group was not formally monitored; therefore, the actual training dose in this group could not be quantified, although weekly telephone reminders were provided to encourage exercise participation.

The experimental group comprised 4 men (20%) and 16 women (80%), whereas all participants in the control group were women (100%). Mean age was 69.9 ± 4.3 years in the experimental group and 70.8 ± 4.7 years in the control group.

Except for two m-CTSIB indices (EOFOS(S) and ECFOS(S)), no statistically significant differences were observed between groups at baseline. These two variables showed statistically significant differences (see Supplementary Table S1), with effect sizes in the small-to-moderate range (Cohen’s d = − 0.65 and − 0.69). Given their proximity to the null boundary and the exploratory nature of the study, these findings were interpreted with caution in subsequent analyses.

Following the 8-week intervention, several outcomes demonstrated statistically significant group × time interactions, indicating greater improvement in the intervention group. Sensitivity analyses incorporating sex as a covariate yielded generally consistent results, although these findings should be interpreted with caution due to the small sample size (Supplementary Table S2). No adverse events, injuries, or intervention-related discomfort were reported.

Primary outcomes

Balance function under single-task condition (Table 2)

Table 2.

Balance performance under single-task conditions (group × time interaction effects, GEE; n = 40)

Variable	Time	Experimental (mean ± SD)	Control (mean ± SD)	Estimate	SE	95% CI	Wald χ²	p value
OA(S)_EO	Pre	0.7 ± 0.6	0.8 ± 0.3	0.22	0.10	0.02 to 0.42	4.86	0.028
	Post	0.5 ± 0.2	0.7 ± 0.3
AP(S)_EO	Pre	0.6 ± 0.5	0.6 ± 0.2	0.19	0.09	0.02 to 0.37	4.58	0.032
	Post	0.4 ± 0.1	0.6 ± 0.2
ML(S)_EO	Pre	0.3 ± 0.3	0.4 ± 0.3	0.11	0.04	0.03 to 0.19	6.70	0.010
	Post	0.2 ± 0.2	0.3 ± 0.2
OA(S)_EC	Pre	1.9 ± 0.8	2.1 ± 0.9	0.46	0.16	0.13 to 0.78	7.70	0.006
	Post	1.4 ± 0.6	1.8 ± 0.6
AP(S)_EC	Pre	1.6 ± 0.7	1.7 ± 0.8	0.38	0.15	0.08 to 0.68	6.15	0.013
	Post	1.2 ± 0.6	1.5 ± 0.6
ML(S)_EC	Pre	0.8 ± 0.5	0.8 ± 0.5	0.25	0.10	0.06 to 0.44	6.80	0.009
	Post	0.5 ± 0.3	0.6 ± 0.3
EOFIS(S)	Pre	0.8 ± 0.2	0.9 ± 0.3	0.07	0.06	-0.06 to 0.20	1.13	0.288
	Post	0.7 ± 0.3	0.8 ± 0.3
EOFOS(S)	Pre	0.9 ± 0.3	1.4 ± 1.0	0.06	0.05	-0.04 to 0.16	1.26	0.262
	Post	0.8 ± 0.3	0.9 ± 0.2
ECFIS(S)	Pre	1.2 ± 0.4	1.3 ± 0.5	0.06	0.09	-0.12 to 0.24	0.45	0.504
	Post	1.1 ± 0.3	1.3 ± 0.4
ECFOS(S)	Pre	2.4 ± 0.6	3.0 ± 1.0	0.39	0.14	0.12 to 0.66	8.14	0.004
	Post	2.0 ± 0.5	2.6 ± 0.7

All sway indices are expressed in degrees (°). Between-group differences were estimated using generalized estimating equations (GEE) with group × time interaction. Lower values indicate better balance performance. The estimate represents the group × time interaction effect. Confidence intervals are presented after rounding; statistical significance was determined based on full-precision estimates from the GEE models. OA overal, AP anterior–posterior, ML medial–lateral, EO eyes open, EC eyes closed, FIS firm surface, FOS foam surface

Under single-task conditions, statistically significant group × time interactions were observed for several sway indices. Significant effects were identified for overall sway with eyes open (OA(S)_EO; estimate = 0.22, 95% CI 0.02 to 0.42, p = 0.028), anterior–posterior sway with eyes open (AP(S)_EO; estimate = 0.19, 95% CI 0.02 to 0.37, p = 0.032), and medial–lateral sway with eyes open (ML(S)_EO; estimate = 0.11, 95% CI 0.03 to 0.19, p = 0.010).

Significant interactions were also observed under eyes-closed conditions, including overall sway (OA(S)_EC; estimate = 0.46, 95% CI 0.13 to 0.78, p = 0.006), anterior–posterior sway (AP(S)_EC; estimate = 0.38, 95% CI 0.08 to 0.68, p = 0.013), and medial–lateral sway (ML(S)_EC; estimate = 0.25, 95% CI 0.06 to 0.44, p = 0.009). In addition, a significant interaction was observed for the eyes-closed foam-surface condition (ECFOS(S); estimate = 0.39, 95% CI 0.12 to 0.66, p = 0.004). No statistically significant group × time interactions were observed for EOFIS(S), EOFOS(S), or ECFIS(S) (all p > 0.05).

Balance function under dual-task condition (Table 3)

Table 3.

Balance performance under dual-task conditions (group × time interaction effects, GEE; n = 40)

Variable	Time	Experimental (mean ± SD)	Control (mean ± SD)	Estimate	SE	95% CI	Wald χ²	p value
OA(D)_EO	Pre	1.6 ± 1.0	1.4 ± 0.7	0.57	0.23	0.12 to 1.02	6.07	0.014
	Post	1.0 ± 0.4	1.3 ± 0.4
AP(D)_EO	Pre	1.3 ± 0.9	1.0 ± 0.6	0.50	0.21	0.09 to 0.91	5.79	0.016
	Post	0.8 ± 0.3	0.9 ± 0.5
ML(D)_EO	Pre	0.7 ± 0.7	0.7 ± 0.6	0.21	0.15	-0.08 to 0.49	1.98	0.160
	Post	0.5 ± 0.4	0.6 ± 0.3
OA(D)_EC	Pre	1.8 ± 1.1	1.7 ± 1.1	0.52	0.31	-0.09 to 1.12	2.82	0.093
	Post	1.3 ± 0.6	1.4 ± 0.5
AP(D)_EC	Pre	1.6 ± 1.1	1.4 ± 1.0	0.49	0.30	-0.10 to 1.07	2.69	0.101
	Post	1.1 ± 0.6	1.2 ± 0.4
ML(D)_EC	Pre	0.6 ± 0.5	0.7 ± 0.6	0.13	0.11	-0.08 to 0.35	1.43	0.232
	Post	0.5 ± 0.3	0.6 ± 0.4

All sway indices are expressed in degrees (°). Between-group differences were estimated using generalized estimating equations (GEE) with group × time interaction. Lower values indicate better balance performance. The estimate represents the group × time interaction effect. Confidence intervals are presented after rounding; statistical significance was determined based on full-precision estimates from the GEE models. OA overall, AP anterior–posterior, ML medial–lateral, EO eyes open, EC eyes closed, FIS firm surface, FOS foam surface, S single-task, D dual-task

Under dual-task conditions, statistically significant group × time interactions were observed for overall sway with eyes open (OA(D)_EO; estimate = 0.57, 95% CI 0.12 to 1.02, p = 0.014) and anterior–posterior sway with eyes open (AP(D)_EO; estimate = 0.50, 95% CI 0.09 to 0.91, p = 0.016). No statistically significant interactions were observed for medial–lateral sway with eyes open (ML(D)_EO; estimate = 0.21, 95% CI − 0.08 to 0.49, p = 0.160) or for eyes-closed conditions, including OA(D)_EC (estimate = 0.51, 95% CI − 0.09 to 1.12), AP(D)_EC (estimate = 0.49, 95% CI − 0.10 to 1.07), and ML(D)_EC (estimate = 0.14, 95% CI − 0.08 to 0.35) (all p > 0.05). These findings indicate that several dual-task outcomes did not reach statistical significance, particularly where confidence intervals crossed zero, and should therefore be interpreted with caution.

Secondary outcomes (Table 4)

Table 4.

Physical performance and concern about falling outcomes (group × time interaction effects, GEE; n = 40)

Variable	Time	Experimental (mean ± SD)	Control (mean ± SD)	Estimate	SE	95% CI	Wald χ²	p value
TUG	Pre	8.8 ± 1.9	9.1 ± 3.0	2.34	0.34	1.68 to 3.00	48.37	< 0.001
	Post	6.5 ± 1.0	8.2 ± 2.3
SLS-R	Pre	19.3 ± 11.6	20.5 ± 10.1	-6.40	2.31	-10.93 to -1.87	7.67	0.006
	Post	25.4 ± 8.8	23.9 ± 8.4
SLS-L	Pre	20.7 ± 11.9	22.9 ± 9.8	-2.85	1.25	-5.30 to -0.40	5.20	0.023
	Post	23.4 ± 10.3	24.8 ± 8.0
FES-I	Pre	40.5 ± 10.4	42.7 ± 14.5	-6.74	2.29	-11.22 to -2.25	8.68	0.003
	Post	46.8 ± 9.8	47.7 ± 15.4

TUG is measured in seconds (s). SLS is measured in seconds (s). FES-I scores range from 16–64, with higher scores indicating greater concern about falling. Between-group differences were estimated using generalized estimating equations (GEE) with group × time interaction. Lower TUG and higher SLS indicate better performance. Higher FES-I scores indicate greater concern about falling. The negative estimate reflects coding direction and does not indicate improvement. The estimate represents the group × time interaction effect. TUG Timed Up and Go test, SLS single-leg stance test, R Right, L Left, FES-I Falls Efficacy Scale–International

Physical fitness

The mean TUG time in the experimental group decreased from 8.8 ± 1.9 s to 6.5 ± 1.0 s, whereas the control group decreased from 9.1 ± 3.0 s to 8.2 ± 2.3 s. The GEE model demonstrated a statistically significant group × time interaction (estimate = 2.34, 95% CI 1.68 to 3.00, p < 0.001), indicating a greater change over time in the experimental group compared with the control group.

Single-leg stance performance also showed statistically significant interactions. For the right leg (SLS-R), the estimate was − 6.40 (95% CI − 10.93 to − 1.87, p = 0.006), and for the left leg (SLS-L), the estimate was − 2.85 (95% CI − 5.30 to − 0.40, p = 0.023).

Concern about falling

FES-I scores increased in both groups following the intervention. The GEE model showed a statistically significant group × time interaction (estimate = − 6.74, 95% CI − 11.22 to − 2.25, p = 0.003), indicating a greater increase in FES-I scores in the experimental group. Because higher FES-I scores indicate greater concern about falling, this result reflects an increase in reported concern about falling.

Sensitivity analyses (Supplementary Table S2)

Sensitivity analyses incorporating sex as a covariate showed that most group × time interactions remained statistically significant. However, several group × time × sex interactions were observed, including ML(S)_EC (p < 0.001), EOFIS(S) (p = 0.001), EOFOS(S) (p = 0.035), OA(D)_EO (p = 0.006), ML(D)_EO (p = 0.001), OA(D)_EC (p < 0.001), and AP(D)_EC (p < 0.001). Given the small sample size and sex imbalance, these subgroup findings should be considered exploratory and interpreted with substantial caution.

Discussion

Given the exploratory nature of this study and the small sample size, the findings should be interpreted as preliminary and hypothesis-generating rather than definitive. Compared with unsupervised exercise conditions, this exploratory study suggests that an 8-week game-based VR balance training program may be associated with improved performance in postural stability and functional mobility outcomes in community-dwelling older adults; however, these findings should be interpreted with caution.

Statistically significant group × time interactions were observed across several balance and mobility outcomes under both single- and dual-task conditions, as well as for TUG and SLS performance. In contrast, FES-I scores increased, indicating greater reported concern about falling. Overall, these findings support the feasibility of a supervised, technology-assisted training approach, while remaining preliminary in nature.

Balance under single-task condition

Statistically significant group × time interactions were observed for multiple postural sway indices under single-task conditions, indicating differential change trajectories between groups. These findings are consistent with previous studies reporting improvements in balance performance following VR- or game-based training [6, 18]. The present study extends existing literature by examining these patterns within a structured and supervised VR-based training protocol applied to relatively healthy, community-dwelling older adults. The use of objective Biodex Balance System parameters, including overall, anterior–posterior, and medial–lateral stability indices, provides a quantitative assessment of postural control. These findings suggest that interactive VR-based balance training may be associated with improved performance in controlled static balance tasks; however, given the relatively high baseline function of the participants, the clinical significance of these changes remains uncertain.

Balance under dual-task condition

Under dual-task conditions, statistically significant effects were observed for selected sway indices, while several outcomes did not reach statistical significance, particularly where confidence intervals included zero. These mixed findings should therefore be interpreted with caution. Dual-task performance requires the integration of cognitive and motor processes, including attentional allocation and executive function, which are known to decline with aging. The observed patterns in overall and anterior–posterior sway suggest that VR-based training may be associated with changes in cognitive–motor interaction [8, 19, 20]. Previous studies have reported that cognitive–motor or dual-task training is associated with improvements in functional performance and gait automaticity in older adults [19–21]. In this context, the interactive nature of VR-based tasks, which require simultaneous visual processing and coordinated movement, may contribute to engagement of both cognitive and motor domains during training. However, given the variability of the results and the wide confidence intervals observed in several measures, these findings should be considered preliminary.

Physical fitness

Statistically significant group × time interactions were observed for TUG and SLS performance, indicating differences in change over time between groups. These findings are consistent with previous studies reporting that VR-based or exergaming interventions may be associated with improvements in lower-limb function, coordination, and reaction time in older adults [6, 22]. The interactive and goal-oriented nature of VR-based training may contribute to differences in participant engagement compared with unsupervised exercise; however, given the absence of adherence and training intensity data in the control group, these effects cannot be attributed solely to the VR component. Although improvements in TUG performance may reflect enhanced functional mobility [4, 6], the relatively high baseline function of the study population suggests the possibility of a ceiling effect. Therefore, the clinical relevance of these findings in terms of fall risk reduction remains uncertain, particularly given that the participants were relatively healthy, community-dwelling older adults.

Concern about falling

Both groups exhibited increases in FES-I scores following the intervention period, with a greater increase observed in the experimental group. As higher FES-I scores indicate greater concern about falling, this finding reflects an increase in reported concern. The current study was not designed to determine the underlying mechanism of this change. While educational components of the intervention may have influenced participants’ perceptions of fall risk, it is not possible to distinguish whether the observed increase reflects heightened fear, a more realistic appraisal of risk, or other psychological responses. In addition, cultural and linguistic factors may influence how FES-I items are interpreted. Previous studies have noted that the translation of “concern” in the Chinese version of the FES-I may carry connotations closer to “worry” or “fear,” which may differ from the intended construct of the original scale [17, 23]. Such differences highlight the need for cautious interpretation when applying this measure across cultural contexts. Taken together, the increase in FES-I scores should be interpreted descriptively as a rise in reported concern about falling. The clinical implication of this finding remains uncertain. Future research incorporating qualitative approaches or additional validated psychological measures is needed to better understand the underlying mechanisms.

Sex differences

Sensitivity analyses suggested potential sex-related moderation effects for certain outcomes; however, these findings should be interpreted with substantial caution due to the small number of male participants and the resulting imbalance between groups. Although some group × time × sex interactions were observed, the current study is not adequately powered to draw definitive conclusions regarding sex differences. These results are therefore considered exploratory and hypothesis-generating. Future studies with more balanced samples are needed to further investigate potential sex-specific responses to VR-based balance training.

Study limitations

This study has several limitations. First, the sample size was relatively small, which limits statistical power and generalizability. Second, although participants were randomly assigned to groups, blinding of participants and outcome assessors was not feasible due to the nature of the intervention. This open-label design may have introduced performance and assessment bias, and the lack of blinded outcome assessment may have resulted in overestimation of intervention effects. Future studies should incorporate blinded assessors to minimize potential bias.

Third, a random sex imbalance occurred, with all male participants (n = 4) allocated to the experimental group by chance. Although baseline differences were observed in two balance measures, their effect sizes were modest, and these variables were interpreted with caution in subsequent analyses. This imbalance reflects random variation that may occur in small-sample randomized trials rather than procedural error. Sensitivity analyses adjusting for sex showed generally consistent findings; however, these results should be interpreted cautiously given the limited sample size. In addition, the predominantly female sample may limit generalizability to older men.

Fourth, adherence to the exercise program in the control group was not systematically or quantitatively monitored. Although weekly telephone reminders were provided to encourage engagement, the actual exercise frequency, duration, and intensity remain unknown. In contrast, the experimental group received a structured and supervised intervention with a fixed number of sessions. Therefore, observed between-group differences may reflect not only the intervention itself but also differences in supervision, adherence support, and overall training exposure. Accordingly, the present study should be interpreted as a pragmatic comparison between a supervised, structured, technology-assisted training program and usual, unsupervised exercise behavior in community settings. The findings therefore do not allow attribution of effects to the VR component alone, and the independent effect of VR cannot be isolated from other contributing factors.

Fifth, logistical constraints related to facility access and scheduling influenced the frequency of sessions, although the total number of sessions was standardized. In addition, formal cognitive assessment was not conducted; cognitive status was screened based on self-report and ability to follow study procedures, which may have limited the detection of subtle impairment. Finally, no long-term follow-up was conducted, precluding conclusions regarding the durability of the observed changes.

Despite these limitations, this exploratory randomized trial provides preliminary evidence supporting the feasibility of a supervised, technology-assisted balance training approach in community-dwelling older adults. The consistency observed across several outcome measures may suggest potential patterns; however, confirmation in larger, well-controlled studies is required.

Recommendations for follow-up studies

Future studies should recruit larger and more diverse samples, including higher-risk populations (e.g., frail older adults and those aged > 80 years), and incorporate longitudinal follow-up to evaluate the sustainability of intervention effects. Implementation of blinded outcome assessment and more robust allocation strategies (e.g., stratified randomization) would help reduce potential bias.

In addition, future studies should include an active control group receiving supervised conventional balance training with comparable intensity and duration. This would allow clearer differentiation between effects related to VR-based training and those attributable to supervision, training dose, and participant engagement. Objective and standardized monitoring of adherence in both groups (e.g., exercise logs or wearable devices) is also needed to ensure comparability of intervention exposure.

Clinically relevant outcomes, including fall incidence, frequency, and severity, should be incorporated to determine whether observed changes in balance and mobility translate into real-world reductions in fall risk. The observed increase in FES-I scores warrants further investigation. Future studies incorporating qualitative approaches (e.g., post-intervention interviews) and complementary validated measures may help clarify whether this reflects increased concern, changes in perceived risk, or other psychological responses.

Furthermore, incorporating physiological and neurocognitive measures (e.g., electromyography, center-of-pressure dynamics, or neuroimaging techniques) may help elucidate potential mechanisms underlying changes associated with VR-based training. Exploration of optimal training intensity, frequency, and task design across different older adult populations will further support the development of targeted, evidence-informed fall-prevention strategies.

Conclusion

In summary, this exploratory randomized controlled trial suggests that an 8-week game-based VR balance training program was associated with improvements in several postural stability indices and functional mobility in community-dwelling older adults. Changes in fall-related concern, as measured by the FES-I, were also observed but should be interpreted cautiously. These findings provide preliminary support for the feasibility of a supervised, structured, technology-assisted training approach in community settings. However, given the exploratory design, small sample size, and methodological limitations, the results should be interpreted as hypothesis-generating rather than definitive. Further large-scale trials with improved control conditions, balanced samples, and long-term follow-up are warranted.

Abbreviations

VR

Virtual reality

BBS

Biodex Balance System

m-CTSIB

Modified Clinical Test of Sensory Integration of Balance

OA

Overall sway index

AP

Anterior–posterior sway index

ML

Medial–lateral sway index

EO

Eyes open

EC

Eyes closed

FIS

Firm surface

FOS

Foam surface

S

Single-task condition

D

Dual-task condition

TUG

Timed Up and Go test

SLS

Single-leg stance test

R

Right

L

Left

FES-I

Falls Efficacy Scale–International

GEE

Generalized estimating equations

CI

Confidence interval

SE

Standard error

SD

Standard deviation

BMI

Body mass index

IRB

Institutional Review Board

NSTC

National Science and Technology Council (Taiwan)

Authors’ contributions

C.-J.H. and L.-F.L. conceived and designed the study. C.-J.H. performed the intervention, collected the data, conducted the statistical analyses, and drafted the manuscript. H.-C.C., L.O., S.-W.H., and L.-F.L. critically reviewed and revised the manuscript, and L.-F.L. supervised the overall research process. All authors read and approved the final manuscript and agree to be accountable for all aspects of the work.

Funding

This study was supported by the National Science and Technology Council of Taiwan (grant numbers: NSTC 11-2813-C-038-046-B and NSTC 113-2221-E-038-018.

Data availability

The datasets generated and analyzed during the current study are not publicly available but are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All study procedures were conducted in accordance with the Declaration of Helsinki and were approved by the Institutional Review Board of Taipei Medical University (IRB No. 202204048). All participants provided written informed consent prior to participation.

Competing interests

The authors declare no competing interests.