Effect of virtual reality training on dual-task performance in older adults: a systematic review and meta-analysis

Abstract

Background

Age-related decline in dual-task (DT) performance is closely associated with falls in older adults, posing a significant public health concern. Virtual reality (VR) training has emerged as a novel intervention to enhance motor-cognitive integration, yet its effects on dual-task performance require systematic evaluation.

Objective

The purpose of this systematic review and meta-analysis was to assess the impact of VR training on dual-task performance in older adults.

Methods

Following PRISMA guidelines, we searched four databases for randomized controlled trials (RCTs) evaluating VR training in adults aged ≥ 60 years. Inclusion criteria required comparisons between VR training and non-VR control groups, with outcome measures including dual-task cost (DTC), dual-task timed up-and-go (DT-TUG), DT gait parameters (speed, stride length, cadence), and DT cognitive performance. Methodological quality was assessed using the Cochrane Risk of Bias tool, and meta-analysis were conducted using RevMan 5.4.

Results

Twenty-one RCTs (935 participants) were included. Meta-analysis revealed significant improvements in VR groups for DTC of gait speed [SMD = -0.32, 95% CI (-0.57, -0.07), P = 0.01], stride length [SMD = -0.58, 95% CI: (-0.90 to -0.26), P < 0.001] and cadence [SMD = -0.32, 95% CI (-0.64, 0.00), P = 0.05]. DT-TUG time decreased significantly [SMD = -0.54, 95% CI (-0.89, -0.19), P = 0.002]. VR training also enhanced dual-task gait speed [SMD = 0.38 95% CI (0.03, 0.73), P = 0.03] and stride length [SMD = 1.15, 95% CI (0.81, 1.49), P < 0.001]. Subgroup analyses showed VR brought more notable improvements for MCI and PD patients. For VR interventions, durations over 1 h per session, more than 4 - week duration, and 3–5 sessions per week yielded better results. Yet, no significant improvements were observed in other DT aspects like cognitive reaction times and rapid gait speed.

Conclusion

VR training effectively reduces DT performance decline in older adults, particularly by lowering DTC and enhancing functional mobility, supporting its potential as a fall prevention strategy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-025-01675-z.

Introduction

The accelerating global demographic aging has markedly heightened fall risks during ambulation in elderly populations through age-related deterioration of physiological functions. Fall risk among older adults increases exponentially with advancing age. Epidemiological evidence indicates that a 20% fall incidence rate is already observed in individuals aged ≥ 60 years, highlighting the critical need for early preventive interventions [1, 2]. Additionally, falls exhibit a marked recurrent nature, with approximately 30% of fallers experiencing recurrent falls [3]. While not all falls signify underlying pathology, they frequently precipitate detrimental outcomes including psychological consequences (e.g., fear of falling, depression) and physical injuries that may progress to fractures, hospitalization, and functional disability [4, 5]. Consequently, falls have emerged as a critical public health challenge among older adults. Developing and implementing accessible, cost-effective, and engaging fall prevention strategies for individuals aged ≥ 60 years is of paramount importance.

Aging is associated with progressive declines in both motor function and cognitive efficiency, driven by the depletion of attentional resources and heightened task interference between concurrent activities [6]. Older adults exhibit heightened fall susceptibility during daily activities involving concurrent physical-cognitive task performance under the dual-task (DT) paradigm [7, 8]. When task demands exceed available cognitive capacity, dual-task cost (DTC), defined as the performance decrement in one or both tasks during concurrent execution, increase significantly [9]. This cognitive overload precipitates biomechanical compromises, such as reduced gait stability and increased stride variability, which disrupt neuromotor control mechanisms and amplify fall risks [10–12]. Critically, DT gait performance has emerged as a clinical biomarker for neurodegeneration: preserved DT capacity correlates with a 43% reduction in fall incidence and delayed dementia onset, reflecting shared neural substrate integrity in cognitive-motor networks [13–15]. Consequently, maintaining the integrity of dual-task abilities is critical for preserving gait stability and preventing falls.

Amid rapid technological advancements, emerging technologies have increasingly been applied in elderly rehabilitation, with virtual reality (VR) technology being a prominent example. VR, with its multi-sensory interactive features, offers personalized training for both physical and cognitive tasks, facilitating improvements in information processing, attention allocation, and sensory integration. These enhancements contribute to better motor performance and therapeutic outcomes by increasing participants engagement [16–18]. Compared to traditional rehabilitation methods, VR is characterized by its immersion, interactivity, personalization, cost-effectiveness, and flexibility [19, 20]. Recent research has demonstrated that VR is particularly effective in improving physical activity, cognitive function, and addressing psychological factors such as fear and depression to reduce fall risks in the elderly. Several meta-analyses have confirmed that VR is an effective measure for preventing falls by improving physical activity abilities [21–23]. For instance, VR interventions have shown significant improvements in performance on the timed up-and-go (TUG) task, a key measure of physical mobility in the elderly. Additionally, multiple studies have demonstrated that VR positively impacts cognitive functions in the elderly and individuals with cognitive impairments, including improvements in memory, executive functions, and other cognitive domains [24–27].

Recent studies have shown that, compared to traditional training methods, VR can more effectively improve DT motor-cognitive performance, reduce DTC, and enhance gait speed and cadence during DT conditions [28–31]. However, emerging evidence reveals substantial population-dependent heterogeneity in these benefits. In Parkinson’s disease patients with freezing of gait, Killane et al. demonstrated that motor-cognitive dual-task VR training significantly improved single-task gait velocity by 12.3% (P = 0.01), yet failed to reduce gait variability under DT conditions [32]. In contrast, a randomized controlled trial by Liao et al. involving older adults with mild cognitive impairment (MCI) reported a 17.6% improvement in DT gait speed (P < 0.001) following 12 weeks of integrative VR training [33]. Notably, even within healthy aging populations, the efficacy of VR remains contentious: Zukowski et al. [31] observed no significant differences in DT gait performance or DTC between VR-enhanced treadmill training and conventional protocols. These inconsistencies underscore the need for further validation of VR’s clinical utility in optimizing DT performance across heterogeneous geriatric populations. In addition current meta-analyses on VR for fall prevention predominantly assess single-task outcomes (e.g., isolated balance or cognition), neglecting DT performance, which mirrors real-world challenges requiring simultaneous motor and cognitive processing [34–36]. To address this gap, our meta-analysis specifically examines DT performance as primary outcomes, including DTC in various spatiotemporal gait parameters, dual-task timed up-and-go (DT-TUG) performance (total completion time), gait parameters under DT conditions, and task reaction time (RT).

Methods

Protocol and registration

This systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [37] and is registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42024553245).

Data sources and searches

This systematic review followed PRISMA guidelines to evaluate the effects of VR on DT performance in older adults. We searched four databases-PubMed, Embase, Cochrane Library, Web of Science from their inception to April 2025. The search strategy combined Medical Subject Headings (MeSH) and free-text terms targeting title, abstract, and keyword fields, with no restrictions on language or publication type (e.g., preprint, conference abstracts). Two independent reviewers (XYW and CH) conducted a two-stage screening process: initial title/abstract screening followed by full-text assessment against predefined eligibility criteria. Disagreements were resolved through consensus or consultation with a third reviewer. To minimize selection bias, we performed backward citation tracking of included studies and relevant reviews. The complete search strategy, including PubMed’s syntax, is publicly accessible in Supplementary Material Table S1.

Inclusion/exclusion criteria

The eligible studies for our study were identified using the PICOS (population, intervention, comparison, outcome, study design) approach. The studies were included if they met the following inclusion criteria: (1) Population: Elderly individuals aged 60 and above who are functionally independent and able to walk normally and independently, including healthy elderly individuals, those with mild cognitive impairment, and patients with early-stage Parkinson’s disease(PD); (2) Intervention: The intervention group was provided with training programs based on VR systems (excluding those combined with other exercise groups), such as Nintendo Wii games, Microsoft Kinect sports games, custom-developed interactive video games, dual-task operation programs, or biofeedback training in virtual specific scenarios; (3) Comparison: Interventions not involving VR, such as traditional fall prevention training, health education, or usual daily activities; (4) Outcome: Measurable DT performance, including gait speed, stride length, and cadence under DT conditions, DTC, cognitive task RT, and performance on the DT-TUG test; (5) Study Design: Randomized Controlled Trials (RCTs).

The exclusion criteria for this review were as follows: (1) Severe neurological disorders [e.g., advanced Parkinson’s disease (Hoehn-Yahr stage ≥ III, freezing of gait, wheelchair dependency), post-stroke hemiplegia, multiple sclerosis resulting in functional dependence]; (2) Other non-neurological conditions compromising ambulatory ability (e.g., severe arthritis, limb amputation); (3) Articles not meeting the inclusion criteria, such as books, letters, conference abstracts, experimental reports, registration protocols, reviews, systematic reviews, and meta-analyses; (4) Experimental designs other than randomized controlled trials, including self-comparison before and after studies, case reports, and cross-sectional studies; (5) Outcome measures that do not include DT performance indicators; (6) Studies for which full text was unavailable or data were incomplete.; (7) Publications in languages other than English; (8) Duplicate publications.

Study selection and data extraction

Two researchers (XYW and CH) independently conducted the literature search and screening process strictly following the predefined search strategy and inclusion and exclusion criteria. The search results were recorded using EndNote™ 20 (Clarivate Analytics, USA), and eligible studies were selected through this application. In the first round of screening, duplicate articles, as well as those identified by automated tools as not meeting the inclusion criteria (e.g., conference abstracts and review articles), were excluded. Subsequently, the researchers independently reviewed the titles, abstracts, and full texts of each article to identify those that met the inclusion criteria. For articles with missing full texts or incomplete data, the researchers reached out to the study authors via email for clarification. Disagreements between the two reviewers were resolved through discussion, and when consensus could not be achieved, a third researcher (JJL) was consulted to make the final decision.

Data analysis

Using RevMan 5.4 software for statistical analysis, the differences in outcome measures before and after the intervention were treated as continuous variables, with mean deviation (MD), standardized mean difference (SMD), and 95% confidence interval (CI) used as effect indicators. The chi-square test was employed to assess heterogeneity among study results. If P ≥ 0.5 and I²< 50%, the studies were considered homogeneous, and a fixed-effect model was applied for meta-analysis. Conversely, when P < 0.5 or I² ≥ 50%, the sources of clinical heterogeneity were analyzed through sensitivity or subgroup analyses. If significant clinical heterogeneity was identified, a random-effects model was employed for the meta-analysis [38].

Risk of bias and certainty of evidence

To assess potential publication bias, we conducted a search of trial registries (ClinicalTrials.gov and the International Clinical Trials Registry Platform, ICTRP) to identify completed but unpublished trials. For completed trials where published articles could not be retrieved, we contacted the principal investigators to obtain additional information. Using the Cochrane Risk of Bias 2.0 (RoB2) tool [39], two independent assessors (XYW and CH) evaluated the risk of bias in the included studies. The Cochrane Risk of Bias checklist consists of six key domains: (1) generation of random sequences and allocation concealment; (2) blinding of participants and intervention implementers; (3) blinding of outcome assessors; (4) incomplete outcome data; (5) selective reporting of outcomes; (6) other potential sources of bias. In cases of disagreements, consensus was reached through discussions between the assessors or by consulting a third author (JJL). According to the Cochrane Handbook for Systematic Reviews of Interventions [40], Possible publication bias and small-study effects were assessed with visual inspection of funnel plots of comparisons with 10 or more trial studies and statistically using Egger’s test and Begg’s test (P < 0.05). The certainty of the evidence for effect estimates was assessed by 2 independent reviewers (XYD and CH) using the Grading of Recommendations Assessments, Development and Evaluation (GRADE) [41–43]. A third reviewer (XYW) was engaged to resolve any discrepancies when consensus was not reached. The GRADE evaluation system rates the quality of evidence as high (i.e., further research is very unlikely to change our confidence in the estimate of effect), moderate (i.e., further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate), low (i.e., further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate), or very low quality (i.e., any estimate of effect is very uncertain).

Results

Study selection

Searches of 4 databases up to April 2025 yielded a total of 3316 studies. After removing duplicates and excluding articles that did not meet the inclusion criteria (such as letters, books, experimental reports, conference abstracts, and reviews), 1,582 studies remained. A subsequent review of titles and abstracts led to the exclusion of 1,532 unrelated articles, leaving 50 studies for full-text review. Of these, 3 studies did not meet the control group requirements, 8 studies were not randomized controlled trials, and 18 studies lacked relevant DT performance outcome measures. After a rigorous selection process, 21 studies were ultimately included [28–30, 33, 44–54]. The flow diagram of the literature selection process is shown in Fig. 1. Good consistency was demonstrated at the title/abstract screening stage [K = 0.72, 95% CI (0.68, 0.76)], and excellent consistency was achieved at the full-text screening stage (K = 0.85). Consensus was reached by a third reviewer for all 375 discrepancies at the title screening stage and 30 discrepancies at the full-text screening stage.

Fig. 1.

PRISMA flow diagram of the screening process

Study characteristics

Population characteristics

The included trials, published between 2007 and 2024, and their characteristics are summarized in Table 1. The studies were conducted across various countries, including Switzerland, Brazil, France, Canada, Hong Kong, Portugal, Taiwan, Spain, the United States (Boston), Thailand, Hungary, the Netherlands, Australia, Japan, and New Zealand. A total of 935 participants were included in the analysis. Sixteen of the studies reported on healthy older adults, three studies had PD patients as their subjects, and two studies focused on patients with MCI. The average age ranged from 65.0 to 85.5. Among them, there were 70 people aged between 60 and 65 years old, accounting for 7% of the total number of included participants. One study included only male participants [50], while another did not report gender distribution [55].

Table 1.

Description of reports and characteristics of included samples

Study	Year	Region	Sample size	Male/Female	Age Mean ± SD	Participants HO、MCI、PD	Intervention	VR Presentation Devices	Duration, Frequency and cycle	Control	Outcome
Altorfer [45]	2021	Switzerland	EG:19 CG:20	EG:8/11 CG:10/10	EG: 73.0 ± 8.8 CG: 72.2 ± 9.8	HO	Dividat Senso motor-cognitive training by playing exergames	The Dividat Senso: a pressure-sensitive platform connected to a computer and screen	10–15 min/ session; 5 sessions/week; 2–3 weeks	Standard rehabilitation treatment	④
Alves [68]	2018	Brazil	NW:9 XB:9 CG:9	NW:9/0 XB:8/1 CG:8/1	NW:58.89 ± 11.16 XB:62.67 ± 13.81 CG:61.67 ± 10.74	PD	NW, Nintendo Wii™ group; XB, Xbox Kinect™ group	Motion-sensing controllers (Wii) or infrared cameras (Kinect) without head-mounted displays	45–60 min/ session; 2 sessions/week; 5 weeks	No training of any type	③
Béraud Peigné[30]	2023	France	EG:19 CG:15	8/26	EG: 69.63 ± 5.31 CG: 70.27 ± 5.82	HO	Multidomain training using Immersive and Interactive Wall Exergames (I2WE)	The Neo One device is equipped with a projector system and wall-mounted infrared sensors	60 min/session; 2 sessions/week; 3months; 24 sessions	A Walking and Muscle-Strengthening (WMS) program	②
Bisson [46]	2007	Canada	EG:12 CG:20	EG:7/5 CG:11/9	EG: 74.4 ± 3.65 CG: 74.4 ± 4.92	HO	Virtual reality (VR)	The participants viewed themselves in a virtual environment on a 29-inch television monitor.	30 min/session; 2 sessions/week; 10 weeks	Computer-based biofeedback (BF) training	①
Chan [29]	2024	Hong Kong	EG:24 CG:21	EG:9/15(29%:71%) CG:4/17(19%:81%)	EG: 68.2 ± 6.0 CG: 71.4 ± 5.1	HO	A Nintendo Ring Fit AdventureTM-based balance and muscle strengthening exercise program	A ring-shaped controller and a leg sensor to detect body movements	60 min/sessions; 2 sessions /week; 8 weeks; 16 sessions	Usual care	②⑩
Eggenberger [47]	2015	Zurich	DANCE:24 MEMORY:22 PHYS:25	DANCE:10/14 MEMORY:6/16 PHYS:9/16	DANCE:77.3 ± 6.3 MEMORY:78.5 ± 5.1 PHYS:80.8 ± 4.7	HO	DANCE, virtual reality video game dancing MEMORY, treadmill walking with simultaneous verbal memory training	A screen-based virtual reality video game dancing platform	60 min/session; 2 sessions/week; 6 months	PHYS, Treadmill walking	④⑤⑦⑫⑮⑯⑱
Ferreira [28]	2024	Portugal	EG2:21 EG1:21 CG:22	37.5%: 62.5%	EG2: 72.14 ± 6.1 EG1: 73.60 ± 7.6 CG: 71.48 ± 4.1	HO	EG2, a multimodal exercise program with augmented reality exergames	Augmented reality (AR) exergames projected in the real environment	60 min/session; 3 sessions/week; 12 weeks	EG1, a multimodal exercise-only intervention program; Usual activities	①⑨
Jäggi [67]	2023	Switzerland	EG:19 CG:21	EG:12/7 CG:15/6	EG: 71.89 ± 9.09 CG: 72.86 ± 10.14	PD	Cognitive–motor training on the exergaming device Dividat Senso + conventional rehabilitation program	A pressure-sensitive platform connected to a computer and a large screen	15 min/session; 5 sessions/week; 2–4 weeks	The standard rehabilitation treatment only	④
Liao [74]	2019	Taiwan	EG:18 CG:16	EG:7/11 CG:4/12	EG: 75.5 ± 5.2 CG: 73.1 ± 6.8	MCI	A VR-based physical and cognitive training (VR) group	A Head-Mounted Display (HMD)	60 min/session; 3 sessions/week; 12 weeks; 36 sessions	A combined traditional physical and cognitive training (CPC) group	④⑤⑥⑫ ⑬ ⑭
Liu [49]	2022	Taiwan	ETC:16 TC:17 CG:17	ETC:4/12 TC:5/12 CG:6/11	ETC:74.6 ± 6.1 TC:73.2 ± 6.3 CG:73.4 ± 6.5	MCI	ETC, exergaming-based Tai Chi	A Head-Mounted Display (HMD)	50 min/session; 3 sessions per week; 12weeks; 36 sessions	Traditional Tai Chi (TC); Usual daily physical activities;	④⑤⑥⑫ ⑬ ⑭
Nobari [62]	2021	Spain	EG:20 CG:20	all male	EG: 71.40 ± 2.64 CG: 71.70 ± 2.40	HO	Virtual driving	A computer equipped with a steering wheel and pedals for virtual driving exercises	20 min/session; 3 sessions/week; 6 weeks	Daily activities	②
OGAWA [51]	2019	Boston	EG:15 CG:14	EG:4/11 CG:1/13	EG: 75.20 ± 7.31 CG: 78.85 ± 7.13	HO (community elderly)	Microsoft Kinect-based exergames	A Microsoft Kinect-based system, which includes a TV and a Kinect sensor	30 min/session; 2 sessions/week; 8 weeks	Traditional physical exercise (TPE) program	④⑥⑲
Phirom [52]	2020	Thailand	EG:20 CG:20	EG:3/17 CG:4/16	EG: 70.21 ± 4.18 CG: 69.40 ± 3.38	HO (community elderly)	Interactive physical-cognitive game-based training	A Microsoft® Xbox 360 Kinect sensor V2, which is a motion-sensing input device that projects a virtual game onto a rubber mat on the floor	60 min/session; 3 sessions/week; 12 weeks	Health education	②
Peng [57]	2020	Taiwan	EG:56 CG:54	EG:17/39 CG:6/48	EG: 70.7 ± 4.6 CG: 72.0 ± 5.7	HO (community elderly)	An interactive exergame mat system to develop a novel cognitivephysical training program	An interactive exergame mat system that incorporates cognitive–physical training	2 h/session; 1 session/week; 3 months	A multicomponent exercise intervention	②
Sapi [58]	2019	Hungary	KB:30 CB:23 CG:22	KBTG:1/29 CB:1/23 CG:4/18	KB:69.57 ± 4.66 CB:69.12 ± 4.19 CG:67.18 ± 5.56	HO	The Microsoft Xbox 360 Kinect videogames	A Microsoft Xbox 360 Kinect, which is a motion-sensing input device that uses a camera to capture movements	30 min/session; 3 sessions/week; 6 weeks	CB, the conventional balance training; CG, no-intervention	②
Schättin [54]	2016	Netherlands	EG:13 CG:14	EG:5/8 CG:7/7	EG: 80.00 ± 7.41 CG: 80.00 ± 7.04	HO	Video game-based physical exercise	A pressure-sensitive dance platform connected to a computer and a beamer for projecting video games on a wall	30 min/session; 3 sessions/week;. 8 weeks; 24 sessions	Conventional balance training	④⑤⑥⑯
Schoene [55]	2013	Australia	EG:15 CG:17	NM	EG: 77.5 ± 4.5 CG: 78.4 ± 4.5	HO	A step game on a computerized step pad system connected to the TV	A step pad system connected to a TV for displaying the game	15–20 min/session; 2–3 sessions/week; 8 weeks	Usual activities	②
Uematsu [60]	2023	Japan	EG:15 CG:9	EG:6/3 CG:8/7	EG: 76.0 ± 3.3 CG: 74.9 ± 2.8	HO	Dual-tasking standing balance training comprised the accurate control of a ping-pong ball on a tray held with both hands, while standing on one leg (analog training) and three modules of Wii FitÔ exergaming (digital training)	A Wii Fit exergaming system, which is a type of digital exergaming device	15 min/day, 2 days/week; 8 weeks; 16 sessions	Daily activities	③
Wang [61]	2021	Taiwan	EG:10 CG:10	EG:3/7 CG:4/6	EG: 71.30 ± 5.33 CG: 73.50 ± 5.66	HO (community elderly)	Exergame-based dual-task training: Walking on a treadmill as a primary task, while performing videogames as the secondary task	A Kinect sensor with a projector and screen setup for exergame-based dual-task training	60 min/session; 3 sessions/week; 12 weeks; 36 sessions	Home-based multicomponent exercise training	②⑪
Yen [59]	2011	Taiwan	EG:14 CB:14 CG:14	EG:12/2 CB:12/2 CG:9/5	EG:70.4 ± 6.5 CB:70.1 ± 6.9 CG:71.6 ± 5.8	PD	Virtual reality (VR)–augmented balance training	A dynamic balance board with a tilting function, an LCD screen, and a personal computer setup for virtual reality training	30 min/session; 2 sessions/week 6weeks	CB, balance training program; CG, daily activities	⑧
Zukowski [31]	2022	New Zealand	EG:30 CG:30	EG:8/22 CG:9/21	EG: 71.20 ± 6.50 CG: 72.00 ± 7.70	HO	Virtual reality treadmill training (VRTT)	A semi-immersive virtual reality system with a 180-degree curved projection screen and a dual-belt treadmill	a single 30-minute session	Conventional treadmill training(CTT)	①④⑪⑫⑰

Note: EG = VR experimental group, CG = control group, NW = Nintendo Wii™ group, HO = Healthy older, PD = Parkinson’s disease, MCI = mild cognitive impairment, XB = Xbox Kinect™ group, CB = conventional balance group, DANCE = virtual dance group, MEMORY = virtual memory group, PHYS = treadmill exercise group; Outcome indicators: ①Dual-task reaction time ②Dual task timed up-and-go(DT-TUG) ③Dual-Task Walking Distance ④Dual-task gait speed ⑤Dual-Task Cadence ⑥Dual-task stride length ⑦Dual-task step length ⑧Sensory Organization Test (SOT) Score ⑨Dual-Task Calculation Accuracy Count ⑩ Dual-task cognitive throughput ⑪Cognitive throughput ⑫ Dual-task cost on gait speed ⑬Dual-task cost on stride length ⑭ Dual-task cost on cadence ⑮Dua-task cost on step length ⑯Fast-Paced Gait Speed ⑰Dual-task gait speed variability ⑱Dual-task step length variability ⑲Dual-task stride length variability

Intervention characteristics

VR interventions demonstrated significant heterogeneity in protocol design, encompassing variations in content modalities, frequency, duration, and system configurations. Thirteen studies employed commercial exergaming platforms (e.g., Wii, Kinect) utilizing predefined gaming modules with fixed-task paradigms, primarily focusing on gamified interactions for lower extremity strengthening, balance coordination, multimodal physical training, or somatosensory-cognitive challenges [28, 30, 31, 49, 51, 52, 54–59]. Six studies implemented DT integration training through VR systems, combining motor-cognitive challenges such as “ball control on a tray + single-leg stance + Wii Fit” or “VR treadmill walking + concurrent video gaming” [33, 45, 47, 48, 60, 61]. One study employed context-specific functional simulation (e.g., driving scenarios) with emphasis on ecological transferability [62]. One study incorporated biofeedback training utilizing real-time physiological parameter monitoring (e.g., postural sway metrics) for self-regulation [46]. In the design of experimental groups, 12 studies compared VR training with conventional training [28, 30, 33,45–49, 51, 53, 54, 57, 59, 61]. Nine studies compared VR-based training with a blank control group receiving conventional care or health education [28, 44, 50, 52, 53, 55, 59, 63, 64]. Among these, three studies used control groups that included both conventional training and a blank control group maintaining daily activities without any intervention [28, 53, 59]. In addition, regarding intervention duration and frequency, the included studies implemented training sessions ranging from 15 minutes to 1 hour per session, administered 2–5 times weekly over a sustained intervention period of ≥ 2 weeks. Only one study investigated the immediate effects following a single 30-minute session [31].

Outcomes measures and intervention effects on dual-task performance

Table 1 provides a summary of the outcome measures. Among the included studies, four employed DTC as the outcome measure [31, 33, 47, 49]. Seven studies have reported using DT-TUG as an outcome measure [29, 50, 52, 53, 55, 57, 61]. Various spatiotemporal gait parameters during DT walking are also frequently used as outcome measures for DT performance, with eight studies focusing on DT gait speed. Various spatiotemporal gait parameters during DT walking are also commonly used as outcome measures for DT performance. Among these, eight studies investigated DT gait speed [31, 33, 45, 48, 49, 51, 54, 65], five examined DT stride length [33, 49, 51, 54], and three focused on DT step frequency [28, 31, 46]. Three studies have reported cognitive task reaction times under DT conditions. Other metrics for assessing DT performance include gait speed under DT fast walking conditions, stride length, and step length [31, 46, 54]. Gait variability under DT conditions at a normal comfortable gait speed, the number of correct responses in the continuous subtraction task during walking [31], cognitive throughput during the clock-drawing task while walking [61], the number of words recalled during DT walking [29], SOT sensory balance integration test scores [59], and gait distance under DT conditions [29, 30].

Risk of bias and certainty of evidence

The risk of bias and the quality of evidence can be assessed in Figs. 2 and 3. Green, yellow, and red correspond to low risk, uncertain risk, and high risk of bias, respectively. Most studies exhibited a low risk of bias (selection bias) in the random sequence generation process. Four studies used computer-generated random number sequences or random number Tables [31, 47, 50, 54, 55], four studies employed group randomization methods [45, 48, 59, 61], one study used a drawing method [29], two studies used sealed envelope methods for randomization [33, 49], and two studies used random grouping but did not specify the randomization methods [30, 64]. Six studies lacked randomization [28, 51, 52, 57, 58, 66], and one study did not specify the allocation concealment method [46].

Fig. 2.

Risk of bias graph of the included studies

Fig. 3.

Risk of bias summary: Methodological quality of each included study

Regarding the evaluation of blinding in the implementation of interventions and outcome assessments, three studies provided insufficient information on blinding for participants and personnel [28, 60, 62]. Six studies either did not blind the outcome assessors or failed to provide information on assessor blinding [30, 31, 47, 51, 54, 58]. employed a structured protocol and objective measurements of cognitive function and gait assessments to minimize the potential for bias in the evaluations. Two studies could not implement blinding at later stages because both the measurements and interventions were conducted by the same research team, rendering pre-test and post-test blinding impossible [45, 48]. In other studies, assessors were unaware of the participants’ group assignments, which could lead to detection bias.

Regarding incomplete data reporting and selective result reporting, only one study [50] inadequately reported missing outcome data (attrition bias). The risk of selective reporting is very low across all studies (reporting bias). Furthermore, all studies disclosed information regarding funding sources or conflicts of interest (other biases). Overall, according to the Cochrane Collaboration Bias Risk Tool, fourteen studies were classified as “low risk” for bias in all domains [30, 31, 33, 45–48, 51, 52, 58–62], while two studies were classified as having “uncertain risk” [49, 57] and five studies were “high risk” [28, 54, 55, 63, 66]. Furthermore, we assessed the quality of evidence and strength of recommendation, finding that virtual reality demonstrates moderate-to-low certainty evidence for improving DT gait outcomes. Despite observed imprecision and inconsistency, most outcome measures demonstrated significant improvements in DT gait performance in the intervention group compared to controls (Fig. 4).

Fig. 4.

Certainty of the evidence (GRADE) about dual‑task performance

The funnel plot for DT gait speed shows that most points are concentrated in the middle of the funnel and are roughly symmetrically distributed, but there is one point far off to the right, which may suggest the presence of publication bias (Fig. 5). To further quantify this possibility, we conducted Egger’s regression test and Begg’s rank correlation test. The P-value for Egger’s test was 0.019, indicating the presence of statistically significant publication bias. However, the P-value for Begg’s test was 0.640, indicating insufficient evidence to support the presence of publication bias. The inconsistency between these two test results may reflect the different sensitivities of the test methods to the data. We further explored this by using the trim-and-fill method, which estimated that three missing negative studies were needed to be added. After correction, the combined effect size was SMD = 0.18 [95% CI (0.01, 0.35)], which is a 35.7% reduction from the original result (SMD = 0.28). However, the findings still support that VR technology may improve DT gait speed.

Fig. 5.

Funnel plot for dual‑task gait speed with standardized mean difference on the X-axis and the standard error on the Y-axis

Results of syntheses

Dual‑task cost on gait speed

Four studies [31, 33, 47, 49] provided reports on six control group comparisons evaluating the impact of VR interventions on the DTC of gait speed, as shown in Fig. 6a. The SMD was chosen as the combined effect size. Studies included in the analysis showed no significant heterogeneity (P = 0.24, I² = 26%). A fixed-effects model was applied, which revealed that the experimental group using VR technology exhibited superior improvement in gait speed DTC compared to the control group, with a statistically significant difference [SMD = -0.32, 95% CI (-0.57, -0.07), P = 0.01]. These results suggest that VR training significantly reduces the gait speed DTC in older adults compared to conventional training.

Fig. 6n.

Forest plot for dual-task fast gait speed comparison

Fig. 6a.

Forest plot for dual‑task cost on gait speed comparison

Dual‑task cost on Stride length

Two studies with four outcomes compared the impact of VR interventions on stride length DTC [33, 49]. The SMD was chosen as the combined effect size, and no heterogeneity was observed among the included studies (P = 0.50, I² = 0%). A fixed-effects model meta-analysis demonstrated that the experimental group using VR technology showed superior improvement in stride DTC compared to the control group, with a statistically significant difference [SMD = -0.58, 95% CI (-0.90, -0.26), P < 0.001]. These findings suggest that VR training significantly reduces stride length DTC in older adults compared to conventional training (see Fig. 6b).

Fig. 6b.

Forest plot for dual‑task cost on stride length comparison

Dual‑task cost on Cadence

Two studies with four comparison groups examined the impact of VR interventions on DTC in cadence [33, 49].The SMD was selected as the combined effect size, and the included studies showed no significant heterogeneity (P = 0.337, I² = 4%). A fixed-effects model meta-analysis revealed that the experimental group using VR technology outperformed the control group in improving DTC in cadence, with a statistically significant difference [SMD = -0.32, 95% CI (-0.64, 0.00), P = 0.05]. These findings indicate that VR training significantly reduces the cadence DTC in older adults compared to conventional training (see Fig. 6c).

Fig. 6c.

Forest plot for dual‑task cost on cadence comparison

Dual-task timed up-and-go (DT-TUG)

Seven RCT studies used the DT-TUG test to evaluate the impact of VR interventions on total completion time during the DT-TUG process [29, 50, 52, 53, 55, 57, 61]. Given the variability in dual-task paradigms across studies, SMD was selected as the pooled effect size. To ensure analytical robustness, two high-risk studies were excluded. Subsequent heterogeneity testing revealed non-significant between-study heterogeneity (P = 0.02, I² = 62%), therefore justifying the use of a random-effects model. Meta-analysis demonstrated that the experimental group exhibited significantly shorter DT-TUG completion time compared to controls [SMD = -0.54, 95% CI (-0.89, -0.19), P = 0.002, Fig. 6d]. To address substantial heterogeneity, subgroup analyses stratified by control group types revealed that VR interventions demonstrated more pronounced efficacy improvements in studies with blank controls, achieving statistical significance [SMD = -0.58, 95% CI (-1.05, -0.10), P = 0.02; Fig. 6e]. These findings suggest that VR training significantly reduced the DT-TUG time in older adults compared to conventional training.

Fig. 6d.

Forest plot for dual-task timed up and go comparison

Fig. 6e.

Forest plot for subgroup analysis according to the control group conditions

Dual‑task gait speed

Eight studies using gait speed as an indicator under dual-task conditions were included [31, 33, 45, 47–49, 51]. Due to inconsistencies in DT conditions, the SMD was chosen as the combined effect size. A heterogeneity analysis was conducted on the 11 comparison groups included, revealing significant heterogeneity between studies (P = 0.004, I² = 61%). Meta-analysis demonstrated superior improvements in DT gait speed for the VR intervention group compared to conventional training [SMD = 0.34, 95% CI (0.02, 0.66), P = 0.04,]. To validate robustness, sensitivity analysis excluding high-risk studies retained elevated heterogeneity (P = 0.003, I² = 64%). Subsequent random-effects modeling confirmed the persistent advantage of VR training [SMD = 0.38, 95% CI (0.03, 0.73), P = 0.03, Fig. 6f]. To elucidate sources of heterogeneity and identify moderators of VR-induced improvements in DT gait speed, subgroup analyses were conducted across populations, session duration, weekly frequency, and intervention period. VR interventions demonstrated superior efficacy in individuals with MCI and PD compared to cognitively healthy older adults [SMD = 0.35, 95% CI (0.06, 0.63), P = 0.02, Fig. 6g]. In terms of the weekly intervention duration of VR, it was found that the intervention effect of more than 1 h was significantly better than that of 1 h or less [SMD = 0.30, 95% CI (0.06, 0.54), P = 0.01, Fig. 6h]. Regarding training frequency, protocols delivering 3–5 sessions per week demonstrated significant improvements in DT gait speed [SMD = 0.31, 95% CI (0.04, 0.58), P = 0.02, Fig. 6i]. For intervention duration, VR programs lasting ≥ 4 weeks yielded clinically meaningful enhancements [SMD = 0.48, 95% CI (0.05, 0.91), P = 0.03, Fig. 6j; Table 2].

Table 2.

Subgroup analysis of the meta-analysis of the effect of virtual reality intervention on dual tasks gait speed

Subgruop analysis	N	Sample			Heterogeneity		SMD(95%CI)	Meta analysis
		E	C		I2(%)	P		Z	P
Intervention population
HO	5	110	91		82	0.0002	0.49(-0.21,1.20)	1.37	0.17
MCI/PD	5	87	121		0	0.54	0.80(0.06,0.63)	2.40	0.02*
Intervention duration(h/week)
<1	3	60	56		91	<0.0001	0.80(-0.61,2.20)	1.11	0.27
≥ 1	7	137	156		0	0.74	0.30(0.06,0.54)	2.47	0.01*
Intervention frequency(n)
<3	5	110	91		82	0.0002	0.49(-0.21,1.20)	1.37	0.17
3–5	5	87	121		0	0.54	0.31(0.04,0.58)	2.29	0.02*
Intervention period(week)
<4	2	48	50		0	0.41	0.04(-0.35,0.44)	0.21	0.83
≥ 4	8	149	162		69	0.002	0.48(0.05,0.91)	2.20	0.03*

Note: N, number; E, VR experimental group; C, control group; HO, Healthy older; PD, Parkinson’s disease; MCI, mild cognitive impairment; * represents statistically significant difference

Fig. 6f.

Forest plot for dual-task gait speed comparison

Fig. 6g.

Forest plot for subgroup analysis according to the intervention population

Fig. 6h.

Forest plot for subgroup analysis according to the intervention session duration

Fig. 6i.

Forest plot for subgroup analysis according to the intervention frequency

Fig. 6j.

Forest plot for subgroup analysis according to the intervention period

Dual‑task Cadence

Three studies with five comparison groups used step cadence as an indicator under DT conditions [33, 49, 54], employing the SMD as the combined effect size. After excluding one high-risk study, heterogeneity analysis of the remaining studies revealed no significant between-study heterogeneity (P = 0.63, I² = 0%). A fixed-effects model was subsequently applied, demonstrating no statistically significant difference in DT cadence improvement between VR intervention and control groups [SMD = -0.09, 95% CI (-0.38, 0.20), P = 0.54]. These findings suggest that there is no significant change in DT cadence in older adults when comparing VR intervention to conventional training (see Fig. 6k).

Fig. 6k.

Forest plot for dual-task cadence comparison

Dual‑task Stride length

Four studies with seven comparison groups focused on stride length as an indicator under DT conditions [33, 51, 54], using the SMD as the combined effect size. After excluding one high-risk study, heterogeneity analysis of the included studies revealed substantial between-study heterogeneity (P < 0.001, I² = 91%). A random-effects model was subsequently employed, demonstrating superior DT stride length improvements in the VR group compared to controls [SMD = 2.54, 95% CI (1.35, 3.73), P < 0.001]. Further sensitivity analysis excluding two comparative groups from Ogawa et al. [51]substantially reduced heterogeneity (P = 0.68, I² = 0%). Using a fixed-effects model for the meta-analysis, the results showed that VR training significantly improved DT gait in the elderly compared to conventional training [SMD = 1.15, 95% CI (0.81, 1.49), P < 0.001, Fig. 6l].

Fig. 6L.

Forest plot for dual-task stride length comparison

Dual‑task reaction time

Three RCT studies compared the effects of VR training on cognitive task reaction times in DT scenarios between an experimental group and a control group receiving conventional training [28, 31, 46]. After excluding one high-risk study, pooled SMD were calculated. No significant between-study heterogeneity was observed (P = 0.43, I² = 0%), warranting the use of a fixed-effects model. A fixed-effects model was used for analysis, and the results indicated no statistically significant difference in DT cognitive reaction times between VR training and conventional training for older adults [SMD = 0.17, 95% CI (-0.26, 0.60), P = 0.43]. These findings suggest that VR training may not significantly improve cognitive task response ability in the elderly under DT conditions compared to conventional training (see Fig. 6m).

Fig. 6m.

Forest plot for dual-task reaction time comparison

Dual‑task fast gait speed

Two RCT [47, 54] studies explored the effects of VR training in the experimental group compared to conventional training in the control group on gait speed under DT conditions. The SMD combined effect size was chosen, and there was no statistical heterogeneity between the studies (P = 0.92, I² = 0%). A fixed-effect model was used for analysis, and the results showed that there was no statistically significant difference in DT gait speed between the VR training and the control group for the elderly [I² = 0%, SMD = -0.08, 95% CI (-0.54, 0.38), P = 0.74]. Indicates that VR training did not significantly improve the fast gait speed of elderly individuals under DT conditions compared to conventional training (see Fig.6n).

Discussion

In this meta-analysis, a comparison was made between the efficacy of VR intervention and that of traditional training, routine care, or health education in terms of enhancing DT performance among older adults. As the first systematic evaluation of VR’s effects on DT outcomes, we focused on three domains: (1) DTC; (2) DT-TUG time; and (3) spatiotemporal gait metrics (e.g., stride length, cadence) under DT conditions. Our findings indicate that VR interventions significantly outperformed control group in reducing DTC for gait speed, stride length, and cadence. Meanwhile, VR intervention reduced DT-TUG time compared to controls. Among the various performances of DT gait, only the improvement effects on gait speed and stride length were better than those of the control group. The long-term and multi-frequency intervention plan of VR can improve gait speed under DT conditions more significantly. Specifically, the more significant effects were achieved when the weekly intervention duration of VR exceeded 1 h, the weekly intervention frequency was 3–5 times, and the intervention lasted for 4 weeks. In addition, this meta-analysis also compared the reaction time of cognitive tasks in DT and the gait speed in DT fast walking, but no positive improvement effects were shown.

Our meta-analysis demonstrates that VR training significantly reduces DTC in gait speed, cadence, and stride length compared to conventional interventions—a finding consistent with evidence that motor-cognitive dual-task paradigms outperform traditional approaches in improving DT gait parameters and associated cost [67]. The superiority of VR may arise from its capacity to integrate three core elements: (1) cognitively demanding tasks requiring attentional shifting, sensory integration, and motor planning [68]; (2) real-time biofeedback to reinforce motor learning [16]; and (3) reduced resource competition between concurrent tasks through enhanced cognitive-motor integration [36]. Gait, as a complex cognitive-motor behavior, relies on both automatic locomotor control and higher-order cognitive functions (e.g., executive function, spatial awareness) [31]. Under DT conditions, older adults experience neurocognitive overload, particularly in prefrontal cortical regions responsible for task prioritization and resource allocation [69, 70]. VR-induced neuroplasticity may mitigate these limitations by engaging multisensory networks—notably the prefrontal cortex—through immersive simulations that concurrently activate motor control and cognitive processing circuits [71, 72]. This aligns with capacity-sharing theory, which posits that performance degradation during dual-tasking reflects finite attentional resources [73]. VR’s ecological complexity may train users to optimize resource allocation, as evidenced by improved Trail Making Test (TMT) scores and task-switching efficiency post-intervention [25]. Such enhancements likely reduce cognitive-motor interference, enabling faster processing speeds and prioritized gait stability during DT walking [74].

Our findings further demonstrate that VR training significantly improves DT gait speed and stride length in older adults. These findings suggest that VR enhances gait adaptability in ecologically complex environments, potentially mitigating fall risks through multimodal sensorimotor integration. Specifically, VR immersive environments deliver spatially contextualized stimuli (visual, auditory, proprioceptive), enabling repetitive practice of adaptive gait patterns that strengthen lower-limb neuromuscular control and dynamic postural stability [75]. Whether it is through interactive game training or DT integration exercises, VR training provides motor tasks under different task conditions. It requires not only coordinated movements of the legs but also the completion of cognitive task challenges based on the constantly changing cue information in the virtual scene, which involves perceiving stimuli, concentrating attention and making quick decisions [76]. In addition, the VR system also offers a pure physical exercise training mode. Through the motor feedback mechanism, subjects can spontaneously integrate multimodal inputs (vision, hearing, proprioception). This implicit integration may optimize motor automatization (such as gait fluency) through the cerebellum-basal ganglia loop [77]. Therefore, when the level of motor automatization of patients is improved, their gait during DT walking will be more natural and fluent, with reduced hesitation, dragging and other situations [78, 79].

Despite epidemiological evidence linking cadence increases to reduced fall incidence (e.g., a 10-step/minute increment correlates with a 13.2% lower fall rate), our meta-analysis found no significant improvement in DT cadence following VR interventions. Prolonged training facilitates neuromuscular adaptations (e.g., joint flexibility, muscle strength) necessary for cadence enhancement [80], but age-related biomechanical constraints—particularly postural stability thresholds—may impose upper safety limits on achievable cadence progression [81]. Importantly, temporal-spatial gait parameters like speed and stride length demonstrate > 70% sensitivity in fall prediction [82], with DT assessments providing superior predictive validity for fall risk [83, 84]. The observed improvements in DT speed and stride length therefore suggest VR’s potential utility in fall prevention strategies.

Dose-response subgroup analysis revealed that interventions exceeding 1 h/week and ≥ 4 weeks produced clinically meaningful DT gait speed improvements. While traditional cognitive-motor training requires ≥ 12 weeks for comparable effects [85, 86]. VR accelerates therapeutic gains through neuroplastic reorganization—functional MRI studies confirm strengthened connectivity between prefrontal cortices and supplementary motor areas post-VR training [87]. Prolonged VR intervention (> 4 weeks) promotes automation of gait-cognitive integration via neurofeedback mechanisms, reducing attentional demand while optimizing motor efficiency [88]. Subgroup analyses of VR intervention frequency revealed that training sessions administered 3–5 times per week significantly improved DT gait speed. This finding aligns with the neuroplasticity window critical for motor skill consolidation. Animal studies have demonstrated that repetitive motor tasks induce long-term potentiation (LTP) in the primary motor cortex (M1), with effects gradually decaying over 48–72 h [89]. A weekly frequency of ≥ 3 sessions may sustain synaptic efficacy to optimize gait automation [90]. Regarding intervention populations, VR demonstrates superior efficacy in individuals with MCI and PD compared to healthy older adults. Both PD and MCI share common cognitive deficits, including impairments in attention, executive function, memory, and visuospatial abilities [91]. VR facilitates neuroplasticity through repetitive motor-cognitive task stimulation, enhancing functional connectivity in the motor cortex and prefrontal cortex, thereby improving motor coordination and cognitive load management [87]. Evidence further indicates that VR strengthens cross-modal remodeling through real-time audiovisual feedback, activating multisensory neurons in the superior temporal sulcus (STS) [92]. Prolonged training enables older adults to progressively integrate gait control and cognitive tasks into automated patterns, reducing cognitive burden while enhancing motor performance through neural feedback mechanisms [88].

This study evaluated the efficacy of VR training on DT-TUG performance, a validated clinical tool that combines gait assessment (e.g., standing, walking, turning) with concurrent cognitive tasks (e.g., arithmetic, category naming) to quantify fall risk in older adults [93, 94]. Our meta-analysis of seven randomized controlled trials showed that, compared with the control group, VR intervention reduced the completion time of the DT-TUG test by 0.54. The observed improvements may stem from VR ability to provide real-time visual biofeedback during task execution, which fosters self-regulation mechanisms and promotes the formation of automated gait patterns through repetitive task-specific practice in immersive environments [95, 96]. Emerging neurophysiological evidence indicates that VR training activates mirror neuron systems, engaging neural pathways involved in balance control and functional mobility through action observation and imitation [97]. Furthermore, VR-mediated interventions appear to optimize cognitive resource allocation during dual-tasking, as evidenced by reduced prefrontal cortical hyperactivity—a biomarker of attentional conflict resolution [98]. Subgroup analyses stratified by control conditions revealed that comparisons with blank controls may overestimate VR’s therapeutic effects. The absence of active interventions in blank control groups allows for positive psychological expectations stemming from novel technology exposure (e.g., VR devices). Studies indicate that VR’s immersive nature induces placebo effects, with approximately 25% of motor improvements attributable to psychological factors (e.g., enhanced motivation, focused attention) rather than specific neuroplasticity mechanisms [99]. Conventional training may partially replicate the cognitive-motor coupling demands inherent in VR protocols (e.g., dynamic balance challenges), thereby attenuating VR’s unique therapeutic advantages. For instance, Tai Chi training—which similarly requires DT coordination (movement sequence memorization + postural control)—demonstrates efficacy overlapping with VR-based dual-task paradigms [100]. Future investigations should employ three-arm trials (VR vs. active control vs. blank control) to quantify VR-specific effects, complemented by multimodal mechanistic analyses integrating neuroimaging (e.g., fNIRS monitoring of prefrontal activation) and biomechanical assessments (e.g., inertial sensor-derived gait cycle parameters). Previous studies propose that reductions in DT-TUG time below the minimal clinically important difference (MCID) of 1.5 s may not translate to meaningful reductions in fall incidence among frail older adults [101]. This discrepancy highlights ongoing controversies regarding the interpretation of dual-task performance metrics, with some researchers questioning their ecological validity in predicting real-world fall risks [102, 103]. Thus, while the 0.54-second improvement observed in our study demonstrates statistical significance, its clinical meaningfulness in fall prevention requires confirmation through longitudinal outcome studies with extended follow-up periods.

Although this meta-analysis provides valuable insights into the effects of VR training on dual gait performance in older adults, it is important to acknowledge a few limitations. First, there is a paucity of research investigating the effects of VR on DT gait cost in older adults. Our meta-analysis incorporated only four studies for evaluation, which may limit the generalizability of findings regarding VR-mediated improvements in DTC to the broader elderly population. Second, the included studies exhibited methodological limitations, with some demonstrating higher risk of bias in specific domains due to insufficient randomization grouping and lack of blinding in interventions, thereby constraining the robustness of conclusions. Current VR interventions often combine non-VR elements (e.g., muscle training), potentially enhancing effects through synergy. While effective versus controls, this mix may overstate VR’s independent role. Future trials should use three-arm designs (VR-only, non-VR exercise, passive control) to clarify VR-specific benefits. Additionally, only 7% of participants were aged 60–64, with no age-specific outcome data available, limiting VR efficacy evaluation in younger seniors. Age-stratified studies (60–64, 65–74, ≥ 75 years) are needed to explore age-related impact variations. Current evidence primarily reflects short-term intervention outcomes, potentially overlooking the longitudinal trajectory of VR training benefits. Subsequent investigations should employ extended follow-up periods to examine the sustained effects on DT performance and fall prevention efficacy. Additionally, personalized VR training regimens tailored to individual cognitive-motor profiles and fall risk stratification warrant systematic exploration to optimize clinical implementation.

Conclusion

In conclusion, our systematic review and meta-analysis indicate that VR training is a useful intervention for increasing DT performance in older persons. According to the findings of this study, VR technology may minimize dual-task gait cost while also improving gait performance in older persons, including DT gait speed and stride length. In terms of VR intervention duration, we discovered that durations over 1 h per session, more than 4 - week duration, and 3–5 sessions per week were more effective in improving gait speed, and these findings have significant implications for the development of fall prevention strategies and the improvement of functioning in older adults. Although our findings are valuable, more research is needed to properly understand the potential benefits of VR. In the future, VR therapies should be standardized, validated in larger-scale, multicenter, high-quality trials, and tailored for usage in clinical and community contexts.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

VR

Virtual reality

DT

Dual-task

RCTs

Randomized controlled trials

DTC

Dual-task cost

DT-TUG

Dual-task timed up-and-go

SMD

Standardized mean difference

TUG

Timed-up-and-go test

RT

Reaction time

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

MD

Mean difference

CI

Confidence interval

HO

Healthy older

MCI

Mild cognitive impairment

PD

Parkinson’s disease

Author contributions

XW: Data curation, Formal analysis, Project administration, Writing-original draft, Writing-review & editing, Data curation. CH: Investigation, Formal analysis, Supervision, Writing-original draft. XD: Investigation, Methodology, Software, Writing-original draft. ZZ: Project administration, Resources, Supervision, Validation, Writing-original draft. YZ: Investigation, Methodology, Software, Writing-original draft. XF: administration, Resources, Supervision, Validation, Writing-original draft. SZ: Investigation, Methodology, Software, Writing-original draft. TL: Investigation, Resources, Supervision, Validation, JL: Investigation, Project administration, Resources, Supervision, Validation, Visualization, Writing-original draft and funding acquisition. Each author contributed to the article and approved the version that was submitted.

Funding

This research received funding from the National Natural Science Foundation of China (grant numbers: 12302418), Shanghai Municipality-level Key Courses in Higher Education Institutions (Shanghai Municipal Education Commission [2024] No. 38), Pathogenic Biology (grant numbers: A020201.24.602), and College Students’ Innovation and Entrepreneurship Training Program (grant numbers: A1-0202-25-0170-7).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All authors have approved this manuscript for publication. This manuscript has not previously been published and is nor pending publication elsewhere.

Competing interests

The authors declare no competing interests.