Virtual Reality Therapy for Upper Limb Motor Impairments in Patients With Stroke: A Systematic Review and Meta‐Analysis

ABSTRACT

Background and Purpose

Stroke is a major cause of disability in adults. Motor recovery through conventional therapy (CT) is a fundamental approach but can sometimes face challenges related to motivation. Virtual reality (VR) rehabilitation, specifically non‐immersive VR, is an alternative therapy aimed at improving upper limb motor function and, consequently, functional independence in daily living activities. However, its effectiveness is still being evaluated. Therefore, a meta‐analysis was conducted to evaluate the effectiveness of non‐immersive VR in upper limb motor function, manual dexterity and the improvement of daily living activities in stroke patients.

Methods

The control groups included physical therapy or occupational therapy. We searched IEEE Digital Library, PubMed, SciELO, Scopus, PEDro, Web of Sciences and ScienceDirect until December 2023 and identified randomized controlled trials (RCTs). Quality and risk were assessed using the revised Cochrane Collaboration tool, PEDro scale, OCEBM, and GRADE. Publication bias and sensitivity analyses were also evaluated. The standardized mean difference (SMD) effect size was calculated to assess the effectiveness of VR therapy compared with conventional therapy. Subgroup analyses were subsequently performed to mitigate the observed heterogeneity and provide further clarity to the results.

Results

In line with previous research, using VR shows improvements in motor function and manual dexterity for stroke patients. Subgroup analyses reveal that the benefits of VR interventions are most pronounced during the acute and subacute recovery stages, particularly in motor function and manual dexterity. Furthermore, combining VR with traditional therapy seems to yield better outcomes in motor function and manual dexterity compared with VR alone. Notably, the type of VR control—whether sensory or manual—or whether the game is commercially available or rehabilitation‐specific, does not seem to influence the outcomes. VR interventions lasting less than 4 weeks are effective in improving both motor function and manual dexterity, whereas interventions of 4 weeks or longer only show significant benefits in motor function.

Discussion

These findings highlight the versatility and potential of VR as a complementary tool in neurorehabilitation.

1. Introduction

Stroke, the second leading cause of death globally, affects 17 million people annually and is a major contributor to physical disabilities (Murray et al. 2012). Two out of three people who survive a stroke experience some type of sequelae, often resulting in upper limb motor deficits that can be disabling (Crichton et al. 2016; Olsen 1990). These conditions can cause individuals to lose their independence, experience social isolation, suffer from depression, and disrupt the lives of their families (Cruz and Diogo 2009). According to recent health statistics, stroke remains a leading global health burden, accounting for 143 million disability‐adjusted life years (DALYs) and representing a major cause of healthy life expectancy (HALE) loss worldwide (GBD 2021, 2024). Given these challenges, providing individuals with physical disabilities with the necessary support is crucial to enhancing their independence and overall quality of life. Conventional physical therapy, which typically includes interventions such as task‐specific training, motor relearning techniques, and strength or coordination exercises, has been recognized as the most effective approach for addressing post‐stroke motor issues, yielding positive outcomes over time (Veerbeek et al. 2014). Recent systematic reviews and meta‐analyses have also evaluated how virtual reality (VR) can complement traditional approaches, showing significant benefits in motor function and balance in stroke patients (Wu et al. 2021).

However, the conventional rehabilitation process poses complexities on various fronts, with adherence influenced by motivational and economic factors. Confronted with these obstacles, VR emerges as a viable and complementary alternative to traditional therapy. This technology, derived from video games, consists of interactive computer software that mimics reality and provides an artificial environment and sensorial experience similar to the real world (J. H. Shin et al. 2014). VR is categorized into three primary systems based on the level of immersion: non‐immersive VR, semi‐immersive VR, and fully‐immersive VR (Mujber, Szecsi, and Hashmi 2004). Among the different types of VR, this review specifically focuses on non‐immersive VR systems, given their practical advantages in clinical settings, including cost‐effectiveness, simplified hardware and software, ease of implementation, and reduced discomfort compared to immersive alternatives (Demeco et al. 2023; Fusco and Tieri 2022; García‐López et al. 2021; Wu et al. 2021). While immersive and non‐immersive VR modalities have both demonstrated significant improvements in upper limb function and manual dexterity in stroke rehabilitation (Demeco et al. 2023; Kiper et al. 2023), non‐immersive systems stand out for their accessibility and potential for broader application in clinical practice. These systems not only enhance physical performance but also foster patient motivation, offering an effective, user‐friendly, and scalable solution for neurorehabilitation (Broeren, Rydmark, and Sunnerhagen 2004). Recent studies have demonstrated that both immersive and non‐immersive VR provide significant improvements in upper limb function and manual dexterity, with the advantages of these technologies standing out in post‐stroke rehabilitation (Demeco et al. 2023; Kiper et al. 2023).

Rehabilitation tools grounded in VR are strategically designed with an ecological and gamified approach, rendering them suitable for patient engagement and offering multimodal feedback. The key challenge in enhancing stroke recovery lies in comprehending how to optimally engage and modify surviving neuronal networks to introduce new response strategies that compensate for tissue loss due to injury (Murphy and Corbett 2009). VR plays a pivotal role in facilitating brain reorganization, particularly during the early stages of injury (acute phase) (Brunner et al. 2016; Faria et al. 2016; Laffont et al. 2020; Murphy and Corbett 2009). Notably, their effectiveness extends to subacute patients (Choi et al. 2014; Prange et al. 2015; Şimşek and Çekok 2016) and chronic cases, showcasing their impact on motor function (Rand et al. 2017; Sin and Lee 2013; Thielbar et al. 2014) and daily living activities (Kiper et al. 2023; Wu et al. 2021).

Hence, over the last decades, the use of VR in neurorehabilitation has increased significantly with a growing body of research exploring different approaches (Brunner et al. 2016; Faria et al. 2016). In this context, numerous systematic reviews and meta‐analyses have been conducted with the aim of providing an updated assessment of the effectiveness of VR in the recovery of stroke patients (Doumas et al. 2021; Fregna et al. 2023; Hao et al. 2023; H. S. Lee et al. 2019; Leong et al. 2022; Zhang et al. 2021). The majority of meta‐analyses have focused on verifying the effectiveness of various VR approaches in improving upper extremity function, as assessed by the Fugl‐Meyer Assessment. In general, a consensus has been reached that VR aids in enhancing motor function in patients across different recovery stages (Chen, Or, and Chen 2022; Fregna et al. 2023; Hao et al. 2023; Leong et al. 2022). While the number of studies is smaller, manual dexterity has also been considered by means of various scales assessing gross manual dexterity, strength, movement quality, and coordination. Similarly, it has been demonstrated that VR contributes to the recovery of manual dexterity in stroke patients (Leong et al. 2022).

Another relevant component to evaluate in stroke patients is how this improvement in upper limb motor function translates into daily living activities. An assessment of the effectiveness of motor therapy should not be deemed complete without appraising autonomy as a direct consequence of motor enhancement. Meta‐analyses have addressed this aspect with self‐report questionnaires or observation, assessing autonomy in daily living activities and showing improvements when VR is used (Larson et al. 2005; Laver et al. 2015, 2017; Leong et al. 2022). However, it is noteworthy that the limited number of studies included in the assessment of daily living activities does not allow generalization of the obtained results (Laver et al. 2015).

While many of these studies encourage the ongoing use of updated reviews given the increasing prevalence of this therapeutic technique, they point out certain limitations, including significant heterogeneity in the results (Doumas et al. 2021; H. S. Lee et al. 2019; Leong et al. 2022; Zhang et al. 2021). Variability in the results is attributed to various factors, with the primary one being the wide diversity of applications, games, and VR modalities employed. The lack of homogeneity in interventions makes it difficult to obtain generalizable results. Additionally, differences attributable to individual variances stemming from the post‐stroke recovery state should be taken into consideration. Similarly, methodological differences in selecting how experimental groups are formed in order to contrast the effectiveness of the intervention also need to be considered. The present review aims to address these gaps by narrowing its focus to non‐immersive VR interventions and exploring their specific contributions to stroke rehabilitation outcomes.

Recent meta‐analyses have taken an approach to subgroup analysis to unravel the influence of variables such as recovery stage (Leong et al. 2022), comparing immersive VR with non‐immersive VR (Hao et al. 2023), or examining differences between VR, augmented reality, or a mix of various types of VR (Leong et al. 2022). In an attempt to control the variability attributable to the VR intervention modality, the present review specifically focuses on non‐immersive VR systems, given their numerous advantages over immersive alternatives. Non‐immersive VR systems, supported by previous reviews for their practical and economic advantages (Demeco et al. 2023; Wu et al. 2021), offer benefits such as cost‐effectiveness, simplified components, user‐friendly deployment, and reduced sensation of sickness (García‐López et al. 2021; Saposnik et al. 2016). Moreover, as highlighted by Garay‐Sánchez et al. (2021), non‐immersive VR is more accessible for routine clinical use, especially for patients who may struggle to adapt to immersive environments, and it reduces adverse effects such as motion sickness. These systems typically present the virtual environment on a standard monitor, with user interaction through devices like a mouse, joystick, or remote controller (Fusco and Tieri 2022). Furthermore, motion‐based user interface games are used enabling users to interact with electronic devices through physical movements and gestures, exemplified by products such as the Nintendo Wii and the Xbox Kinect (Afsar et al. 2018; Choi et al. 2014; Ikbali Afsar et al. 2018; Kim et al. 2018; Kong et al. 2016).

In order to address the limitations highlighted in recent meta‐analyses (Doumas et al. 2021; Zhang et al. 2021), the main objective of this review and meta‐analysis is to provide an updated assessment of the effectiveness of VR rehabilitation tools in stroke patients, considering the three dimensions of upper limb rehabilitation described in the literature: (a) motor function, defined as an improvement in the functional dimension of gross motor skills; (b) manual dexterity, which consists of the activity dimension that enhances fine motor skills; and (c) daily living activities, referring to functional independence evaluated using scales that measure the level of autonomy. This targeted approach aims to provide a clearer, clinically relevant framework for the application of non‐immersive VR systems in stroke rehabilitation.

In addition, the analysis includes a subgroup assessment of non‐immersive VR modalities, distinguishing between different stage recoveries, manual and sensor‐based virtual reality games, exploring commercial usage, and investigating potential disparities in outcomes when VR is employed on its own or in conjunction with conventional therapy.

The present review and meta‐analysis aim to address variables influencing heterogeneity and its role in prescribing VR in clinical settings, complementing recent works by Kiper et al. (2023) and Wu et al. (2021) that explore optimizing VR implementation and reducing variability in outcomes.

2. Methods

2.1. Data Sources and Searches/Search Strategy

A systematic search was conducted using the IEEE Digital Library, PubMed, SciELO, Scopus, PEDro, Web of Sciences and ScienceDirect databases, covering studies published up to December 2023, with the exclusion of studies published more than 10 years ago. The exploration techniques integrated controlled vocabulary terms and title/abstract keyword terms in the search topics of virtual reality, upper limb, stroke, and randomized controlled trials; the following keywords were finally used: (“stroke” OR “brain injury” OR “ictus” OR “ischemic ictus” OR “post stroke”) AND (“Neurorehabilitation” OR “motor improve” OR “motor recovery” OR “motor rehabilitation” OR “neuroplasticity” OR “stroke recovery” OR “upper limb rehabilitation”) AND (“Virtual reality” OR “Computer game” OR “serious games” OR “videogame”). In addition, a snowballing procedure was applied in this systematic review to find more articles by examining the references of previous systematic reviews (Sayers 2007).

The present systematic review and meta‐analysis were performed following the PRISMA guidelines (Page et al. 2021), and used the five elements of the Population, Intervention, Comparison, Outcome, and Study design (PICOS) framework (Liberati et al. 2009) to conduct the search. Participants included any patients with post‐stroke upper limb motor function impairment, with assessments conducted on the affected limb, regardless of whether it was dominant or non‐dominant. The intervention was non‐immersive VR rehabilitation, and the comparison was conventional therapy (occupational therapy, physical therapy or similar). The outcome was defined by the baseline values before the intervention and the values in the first measurement after treatment in the domains of motor function, manual dexterity, and daily living activities. The study had a randomized controlled trial design.

2.2. Study Selection and Data Extraction

All studies were imported into Parsifal, an online tool designed to facilitate the planning and execution of systematic reviews (Kitchenham 2007). During the database research phase, two independent reviewers conducted the screening process to determine eligibility. Initially, researchers evaluated the abstracts, titles, and study types to assess eligibility, followed by a thorough examination of the full text of relevant publications. In instances where reviewers had differing opinions, a consensus was reached through discussion. In cases of disagreement between reviewers, consensus was reached.

The following inclusion criteria were applied: (1) patients undergoing post‐stroke rehabilitation, (2) randomized controlled trials (RCTs) or pilot RCTs, (3) a non‐immersive VR‐based motor intervention, (4) treatment as usual (physical therapy or occupational therapy) as the comparison, (5) reporting upper limb motor outcomes (motor function, manual dexterity and/or living daily activities), (6) parallel design, and (7) language restrictions were applied (limited to studies published in English). The primary outcome of interest assessed was upper extremity motor function, while the secondary outcomes of interest included manual dexterity and daily living activities.

On the other hand, the following exclusion criteria were applied: (1) animal studies, (2) subjects under 18 years old, (3) articles with insufficient data for analysis, (4) studies that are not randomized controlled trials (RCTs), including observational studies, cohort studies, and non‐randomized trials and (5) those that have not been published in the last 10 years were excluded (in order to achieve the most up‐to‐date information, considering the burgeoning advancements in new technology), (6) robotic devices, (7) augmented virtual reality, (8) immersive virtual reality, (9) studies with fewer than five subjects per group, (10) control groups that included other interventions (i.e., proprioceptive neuromuscular facilitation (PNF) or alternative interventions involving different types of virtual reality). Table 1 summarizes the inclusion and exclusion criteria. The reviewers extracted data from each included trial about: (a) authors, (b) number of patients randomized in the treatment groups, (c) age group, (d) diagnosis, (e) assessment scales, (f) VR intervention system, and (g) duration of intervention. All outcome data were carefully extracted from the assessment scales in the included trials (Table 2).

TABLE 1.

Inclusion and exclusion criteria.

Inclusion criteria

Exclusion criteria

Patients undergoing post‐stroke rehabilitation Randomized controlled trials (RCTs) or pilot RCTs VR‐based motor intervention (Kinematic or with assisted movements) Treatment as usual (physical therapy or occupational therapy) as the comparison Parallel design Reporting upper limb motor outcomes (motor function, manual dexterity and/or living daily activities) Studies publish in English

Animal studies Subjects under 18 years old Studies that are not randomized controlled trials (RCTs), including observational studies, cohort studies, and non‐randomized trials Articles with insufficient data for analysis Published in the last 10 years Articles that have not been published in the last 10 years Robotic devices Augmented virtual reality Immersive virtual reality Studies with fewer than 5 subjects per group Control groups that included other interventions (i.e., proprioceptive neuromuscular facilitation (PNF) or alternative interventions involving different types of virtual reality)

TABLE 2.

Summary table.

Author and year	NVR a	NCT b	Age c	Diagnosis d	Assessment scales e	VR intervention	Intervention period	GRADE	PEDro	Rehabilitation strategies
Afsar et al. (2018)	19	16	69,42 (8,55)	SSP	FMA‐UE/BBT/FIM	X‐box kinect	4 weeks	⨁⨁⨁◯	6	VR + CON
			63,44 (15,73)					Moderate
Ain et al. (2021)	25	25	57,48 (10,60)	CSP	FMA‐UE/BBT	X‐box kinect	6 weeks	⨁⨁⨁◯	7	VR
			57,68 (10,43)					Moderate
Anwar et al. (2022)	34	34	51,56 (7199)	CSP	FMA‐UE	Nintendo Wii	6 weeks	⨁⨁⨁◯	7	VR
			51,35 (5783)					Moderate
Aşkın et al. (2018)	18	20	53,27 (11,19)	CSP	FMA‐UE/BBT	Xbox kinect	4 weeks	⨁⨁⨁◯	7	VR + CON
			56,55 (9,85)					Moderate
Brunner et al. (2017)	57	55	62 (16,5)	SSP	BBT/FIM	VR YouGrabber system	4 weeks	⨁⨁⨁◯	7	VR + CON
			62 (11,5)					Moderate
Cannell et al. (2018)	35	38	72,8 (10,4)	SSP	BBT	Jintronix (JRS WAVE)	8 weeks	⨁⨁◯◯	9	VR + CON
			74,8 (11,9)					Low
Cho et al. (2021)	12	12	71,00 (8,36)	ASP	FMA‐UE/MBI	Nintendo Wii	4 weeks	⨁⨁⨁◯	8	VR
			67,25 (10,33)					Moderate
Choi et al. (2014)	10	10	64,3 (10,3)	SSP	FMA‐UE/BBT/MBI	Nintendo Wii	4 weeks	⨁⨁⨁◯	8	VR + CON
			64,7 (11,3)					Moderate
da Silva Ribeiro et al. (2015)	15	15	53,7 (6,1)	CSP	FMA‐UE	Nintendo Wii	8 weeks	⨁⨁⨁◯	9	VR
			52,8 (8,6)					Moderate
Kim JH et al. (2018)	12	12	50,91 (9,57)	CSP	FMA‐UE/MFT	Nintendo Wii	12 weeks	⨁⨁⨁◯	5	VR + CON
			57,23 (14,63)					Moderate
Kim WS et al. (2018)	11	9	54,7 (17,3)	SSP	FMA‐UE/BBT/MBI	X‐box kinect	2 weeks	⨁⨁◯◯	10	VR + CON
			53,5 (16,0)					Low
Kong et al. (2016)	33	34	58,1 (9,1)	SSP	FMA‐UE/ARAT/FIM	Nintendo Wii	3 weeks	⨁⨁◯◯	8	VR + CON
			59 (13,6)					Low
Kottink et al. (2014)	8	10	65,3 (6,5)	CSP	FMA‐UE/ARAT	Game ‘‘FurballHunt’’	6 weeks	⨁⨁⨁◯	7	VR
			58,4 (14,8)					Moderate
Kuo et al. (2023)	19	18	57,47 (6,99)	CSP	BBT	PABLO virtual reality system	9 weeks	⨁⨁◯◯	9	VR + CON
			59,50 (10,65)					Low
Laffont et al. (2020)	25	26	60,8 (14,1)	SSP	FMA‐UE/BBT/BI	Nintendo Wii	6 weeks	⨁⨁◯◯	9	VR + CON
			55,8 (14,0)					Low
Lee M et al. (2016)	13	13	66,46 (7,26)	CSP	FMA‐UE/BBT/MBI	X‐box kinect	8 weeks	⨁⨁⨁◯	6	VR + CON
			69,92 (7,18)					Moderate
Lee MM et al. (2016)	5	5	65,2 (5,0)	SSP	FMA‐UE	VR canoe game	4 weeks	⨁⨁⨁◯	9	VR + CON
			66,2 (3,4)					Moderate
Lee MM et al. (2018)	15	15	61,8 (6,8)	SSP	MFT	Nintendo Wii	5 weeks	⨁⨁⨁◯	7	VR + CON
			61,3 (8,4)					Moderate
Leng et al. (2022)	31	26	59,2 (10,7)	SSP	FMA‐UE/BI	Microsoft Xbox 360 kinect	3 weeks	⨁⨁⨁◯	7	VR + CON
			59,1 (11,6)					Moderate
Marques‐Sule et al. (2021)	15	14	61,5 (8,4)	CSP	FMA‐UE/BI	Nintendo Wii	4 weeks	⨁⨁⨁◯	9	VR + CON
			58,2 (7,4)					Moderate
McNulty et al. (2015)	20	20	59,9 (13,8)	CSP	FMA‐UE/BBT	Nintendo Wii	2 weeks	⨁⨁◯◯	9	VR
			56,1 (17)					Low
Norouzi‐Gheidari et al. (2019)	9	9	42,2 (9,5)	SSP/CSP	FMA‐UE/BBT	Rehabilitation‐based virtual reality	4 weeks	⨁⨁⨁◯	5	VR + CON
			57,6 (10,5)					Moderate
Oh et al. (2019)	17	14	57,4 (12,2)	CSP	FMA‐UE/BBT	VR “real” instruments	6 weeks	⨁⨁⨁◯	8	VR
			52,6 (10,7)					Moderate
Ozen et al. (2021)	15	15	62 (13,12)	SSP/CSP	FMA‐UE	Games which are played with a joystick	4 weeks	⨁⨁⨁◯	7	VR + CON
								Moderate
			69,8 (8,41)
M. Park, et al. (2019)	12	13	53,5 (13)	CSP	FMA‐UE/WMFT/BI	Smart board intervention	4 weeks	⨁⨁⨁◯	8	VR + CON
			51,5 (16,7)					Moderate
Y. S. Park, et al. (2021)	22	22	60,59 (18,1)	ASP	MBI	RAPAEL smart Glove digital system	4 weeks	⨁⨁⨁◯	7	VR + CON
			62,29 (13,9)					Moderate
Rand et al. (2016)	13	11	59,1 (10,5)	CSP	BBT	Videogames with videocapture	5 weeks	⨁⨁◯◯	9	VR
			64,9 (6,9)					Low
Rodríguez‐hernández et al. (2023)	23	20	62,6 (13,5)	SSP	FMA‐UE/ARAT	HandTutor glove	3 weeks	⨁⨁⨁◯	8	VR + CON
			63,6 (12,2)					Moderate
Rodríguez‐Hernández et al. (2021)	23	20	62,6 (13,5)	SSP	FMA‐UE	HandTutor glove	3 weeks	⨁⨁⨁◯	7	VR + CON
			63,6 (12,2)					Moderate
Saposnik et al. (2016)	59	62	62 (13)	ASP	BBT/BI	Nintendo Wii	2 weeks	⨁⨁◯◯	8	VR
			62 (12)					Low
S. Shin, et al. (2022)	20	16	57,00 (12,78)	SSP	FMA‐UE	RAPAEL smart Glove digital system	4 weeks	⨁⨁⨁◯	6	VR + CON
			63,69 (8,58)					Moderate
J. H. Shin, et al. (2014)	9	7	52 (11,9)	ASP/SSP	FMA‐UE/MBI	Rehabmaster	2 weeks	⨁⨁⨁◯	7	VR + CON
			46,6 (5,8)					Moderate
J. Shin, et al. (2016)	24	22	57,2 (10,3)	CSP	FMA‐UE	RAPAEL smart Glove digital system	4 weeks	⨁⨁◯◯	8	VR + CON
			59,8 (13,0)					Low
Şimşek and Çekok (2015)	20	22	54,1 (20,2)	SSP	FIM	Nintendo Wii	10 weeks	⨁⨁⨁◯	8	VR
			61,5 (11,6)					Moderate
Sin and Lee (2013)	18	17	71,7 (9,4)	CSP	FMA‐UE/BBT	X‐box kinect	6 weeks	⨁⨁⨁◯	6	VR + CON
			75,5 (5,5)					Moderate
Türkbey Kutlay, and Gök. (2017)	10	9	61,7 (10,2)	SSP	BBT/FIM	Xbox kinect	4 weeks	⨁⨁⨁◯	7	VR + CON
			62,4 (8,1)					Moderate
Velmurugan Viswanath, and Andrews. (2023)	20	20	52,4 (6,2)	SSP/CSP	FMA‐UE	Nintendo Wii	6 weeks	⨁⨁⨁◯	7	VR
			53,6 (5,8)					Moderate
Wang et al. (2017)	13	13	55,3 (8,40)	SSP	WMFT	Leap motion‐based VR training	4 weeks	⨁⨁⨁◯	8	VR + CON
			53,3 (7,65)					Moderate
Xie et al. (2021)	6	6	48,8 (9,6)	ASP/SSP/CSP	FMA‐UE	Intelligent glasses‐free VR	3 weeks	⨁⨁⨁◯	7	VR + CON
			62,4 (8,1)					Moderate
Zondervan et al. (2016)	9	8	60 (7,3)	CSP	BBT	Music glove	3 weeks	⨁⨁⨁◯	7	VR
			59 (9,94)					Moderate

^a NVR, number of participants randomized to the VR intervention.

^b NCT, number of participants randomized to the control intervention.

^c Mean (SD).

^d CSP (Chronic Stroke patients); SSP (Subacute stroke patients); ASP (Acute stroke patients).

^e FMA‐UE (Fugl Meyer Assessment for Upper Extremity); ARAT (Action research arm test); BBT (Box and blocks test) WFMT(Wolf motor function test) FIM (Functional independence measure); BI (The Barthel Index and modified versions).

2.3. Quality and Risk Assessment

The methodological quality of individual studies was independently assessed by two reviewers using different scales. The first was the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system (Guyatt et al. 2008); this system considers the balance of desirable and undesirable outcomes among alternative management strategies, and makes a strong or weak recommendation for or against an intervention based on four domains (quality of evidence, balance between benefits and risk, patient's preferences and cost and resources) (Andrews et al. 2013). The Oxford Centre for Evidence‐Based Medicine (OCEBM) (Howick 2011) (available at: http://www.cebm.net) can also be used as a heuristic for clinicians and patients to answer clinical questions quickly, without relying on pre‐appraised evidence or resources and the PEDro scale, which is confidently used to assess the methodological quality of clinical trials of physiotherapy interventions (PEDro scale, available at: https://pedro.org.au) (de Morton 2009). The most commonly used minimum score on the PEDro scale to define a high‐quality RCT is 6/10, as scores of six or above typically indicate studies of high methodological quality. In the case of this meta‐analysis, the 6/10 threshold was applied, with the exception of studies rated 5/10 only when deemed strictly necessary due to their relevance.

The Revised Cochrane Collaboration Tool for Assessing Risk of Bias (RoB 2.0, available at: http://www.riskofbias.info/welcome/rob‐2‐0‐tool) (Sterne et al. 2019) was used to assess bias arising from five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. The overall risk of bias judgment was rated as low risk of bias, with some concerns or high risk of bias mentioned by the two reviewers. Disagreements were resolved through consensus. In cases of conflicting assessments, a third investigator was consulted to ensure an objective resolution, allowing for a thorough discussion of discrepancies and a final decision based on the best available evidence.

2.4. Planned Methods of Analysis

All data were analyzed using Review Manager (RevMan 5.3), a software developed by Cochrane that assists in creating and managing reviews, and analyzing study data for meta‐analyses (Cochrane RevMan, available at https://training.cochrane.org/es/online‐learning/core‐software‐cochrane‐%20reviews/revman/download). The effectiveness of VR therapy versus conventional therapy was measured using the mean difference (MD) when the outcome measurements were made on the same scale and the standardized mean difference (SMD) when the outcome measurements were made with different scales; 95% confidence intervals (CIs) were used to calculate the pooled estimates (Hedges, Pustejovsky, and Shadish 2013). For example, in studies assessing manual dexterity, some utilized the Box and Blocks Test (BBT), while others employed the Manual Function Test (MFT). Using the SMD enabled the results to be combined and compared by standardizing the scores across different measurement tools, thus facilitating a unified interpretation of the effect size.

As recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022), in articles where standard deviation (SD) is not reported but the confidence interval (CI) is available for an absolute effect measure (e.g., standardized mean difference), the SD can be calculated as √N x (upper limit‐lower limit)/3.92.

A random‐effects model was used for data syntheses, taking into consideration the sample and methodological diversity among the included epidemiological studies and clinical trials, and the inverse variance weighted method was employed to minimize the imprecision (uncertainty) of the pooled effect estimate. Subgroup analysis was conducted to explore variables of interest.

Heterogeneity for each meta‐analysis was estimated using the Thau² statistic (appropriate for small samples) and the I ² statistic. An I ² value greater than 50% or a Thau² p‐value > 0.05 was considered significant (Higgins 2022). Publication bias was assessed using Egger's weighted regression and was visually estimated based on funnel plots.

3. Results

3.1. Search Results

Two thousand three hundred sixteen (2316) studies were initially identified from the online databases. Removal of one hundred eight (108) duplicates using the Parsifal online tool (Kitchenham 2007) left two thousand two hundred eight (2208) articles. Through screening of abstracts, a total of two hundred and seventy‐three (273) studies were assessed for eligibility for full text review. Studies were excluded if they did not have randomized groups, control groups, VR treatment, upper limb recovery as an outcome, and other reasons in line with the exclusion criteria (Table 1). Some studies met more than one exclusion criteria. Finally, a total of 40 studies were selected, including 1596 subjects (Figure 1), representing the inclusion process flowchart following the PRISMA guidelines (Page et al. 2021).

FIGURE 1.

Flow diagram of the study identification and selection process, following Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines.

3.2. Study Characteristics

Forty studies compared VR of upper limb therapy versus CT (control therapy operationalized as physical therapy or occupational therapy). All articles were randomized controlled trials and parallel designs; eleven of them were pilot studies. Sixteen studies selected patients with chronic stroke, sixteen had subacute stroke, three had acute stroke phase and five were included in different stages of injury, three combined subacute and chronic patients, one combined acute and subacute patients, and another combined acute, subacute, and chronic patients (Table 2). It is important to note that the definition of each phase (acute, subacute, chronic) was based on the criteria used by the authors of the studies included in this meta‐analysis. Generally speaking, the acute phase refers to the first month post‐stroke, the subacute phase spans from one to 6 months, and the chronic phase includes individuals more than 6 months post‐stroke. However, these timeframes may vary slightly across studies. To account for these variations between the acute and subacute phases, the corresponding groups were combined in the subgroup analysis.

The outcome was defined as the change in three dimensions: (a) motor function, assessed by the Fugl‐Meyer Assessment of upper extremities (FMA‐UE), the most frequently used scale providing measures in post‐stroke patients (Gladstone, Danells, and Black 2002); (b) manual dexterity was also assessed through measures comprising of: the Box and Blocks test (BBT), employed to evaluate unilateral manual dexterity (Mathiowetz et al. 1985); the Manual Function Test (MFT), developed to evaluate unilateral manual performance (Miyamoto et al. 2009); the Wolf Motor Function Test (WMFT), which quantifies upper extremity motor ability through time in functional tasks, and quality of movement (Wolf et al. 2001); the Action Research Arm Test (ARAT), was used to evaluate upper extremity performance in four subscales (grasp, grip, pinch, and gross movement) (Lyle 1981), (c) the daily living activities were assessed using two questionnaires: the Barthel Index (BI), which measures the execution of 10 basic daily living activities and obtains a quantitative estimation of the subject's level of dependency, including modified or adapted versions of the Barthel Index (Mahoney and Barthel 1965; Quinn, Langhorne, and Stott 2011); the Functional Independence Measure (FIM), which evaluates the level of a patient's disability and indicates the amount of assistance required for the individual to perform daily living activities (Keith et al. 1987). The scales included in motor function and manual dexterity are recommended for their high level of psychometric qualities and clinical utility across different practice settings and stages of stroke recovery. The selection of these measures helped ensure the validity and reliability of the results (Alt Murphy et al. 2015; Sullivan et al. 2013). In studies involving multiple scales, selection was prioritized based on the clinical utility of each measure and its frequency of use in the included articles. For example, in the manual dexterity domain, the Box and Blocks Test (BBT) was selected as the primary measure due to its clinical relevance and high frequency of use in the studies included in the meta‐analysis, contributing to greater homogeneity in the scale selection process.

3.3. Quality of Studies

All studies conducted randomized trials and received a score of two in OCEBM. This indicates that all included studies are randomized with adequate design and execution. The assessment of the certainty of evidence by the GRADE tool graded the studies as 31 being moderate and 9 low due to the risk of bias. This suggests that the majority of the included studies provide evidence of sufficient quality to support their conclusions. In the PEDRO scale, 38 studies had a minimum score of 6, and 2 had a score of 5, which has been considered here to be of acceptable methodological quality (Table 2).

3.4. Risk of Bias

The overall risk of bias was low in nine studies, whereas 31 studies had some concerns (Figure 2). Random sequence generation was rated as some concern in 30 trials and low risk in 10 trials. For deviations from intended interventions, nine studies were rated as low risk, and 31 studies were rated as some concerns. Concerning missing outcome data, six studies were rated as having some concerns, one as high risk and 33 as a low risk. For bias related to outcome measurement, based on the trial reports, three studies were rated as high risk, one as having some concerns and 36 as low risk. Finally, 40 were rated as low risk in the domain of the selected results.

FIGURE 2.

Cochrane Risk‐of‐Bias Assessment. (A) Review authors' judgments on each risk of bias item for each included study. (B) Review authors' judgments on each risk of bias item presented as percentages across all included studies.

3.5. Effectiveness of VR Therapy

3.5.1. Motor Function

Twenty‐eight studies including 903 subjects were included in the meta‐analysis to compare the effectiveness of VR therapy and a control group on motor function. There was a significant improvement in motor function score following VR therapy in stroke patients compared with those undergoing control treatments (Z = 4.13; P < 0.0001; SMD = 5.24; 95% CI = 2.76–7.73; P < 0.05) (Figure 3). The results suggested moderate and significant heterogeneity among the included studies, indicating that the observed variation in effects is partly due to real differences among the studies rather than just chance (I ² = 75%; Tau² = 24.43; Chi² = 106.47; P < 0.00001).

FIGURE 3.

Global results for motor function.

A funnel plot analysis was conducted to assess the potential of publication bias. Figure 4 displays the funnel plot, which shows a slight asymmetric distribution of studies included in the meta‐analysis. After inspection, one outlier study effect was identified and excluded (J. Shin et al. 2016) to ensure the robustness and reliability of the results. Additionally, the Egger regression test did not show any significant risk of bias among the included studies (P > 0.05). Furthermore, the leave‐one‐out sensitivity analysis did not affect the significance of the overall effect size, indicating that the findings were not driven by any single study (SMD = 5.24; 95% CI = 2.76–7.73; P < 0.05).

FIGURE 4.

Funnel plot for motor function.

3.5.2. Manual Dexterity

Twenty‐seven studies with 1031 subjects were included in the analysis to assess the effectiveness of VR therapy compared with control groups in improving manual dexterity in stroke patients. The results showed that VR therapy led to a significant improvement in manual dexterity scores compared with conventional therapy (Z = 2.36; P = 0.02; SMD = 0.23; 95% CI = 0.04–0.43; P < 0.05) (Figure 5). There was moderate and significant heterogeneity among the included studies, suggesting that the observed variation in effects is partly due to real differences among the studies rather than just chance (I² = 55%; Tau² = 0.13; Chi² = 57.16; P = 0.0004).

FIGURE 5.

Global results for manual dexterity.

A funnel plot analysis was performed to evaluate the possibility of publication bias. Figure 6 shows the funnel plot, revealing a minor asymmetrical distribution of the studies that were incorporated in the meta‐analysis. Subsequently, effects of two outliers' studies were identified and excluded from the dataset (Cannell et al. 2018; J. Shin et al. 2016) to ensure the robustness and reliability of the results. Moreover, the Egger regression test did not show any significant risk of bias among the included studies (P > 0.05). Moreover, the leave‐one‐out sensitivity analysis did not affect the significance of the overall effect size, indicating that the findings were not driven by any single study (SMD = 0.23; 95% CI = 0.04–0.43; P < 0.05).

FIGURE 6.

Funnel plot for manual dexterity.

3.5.3. Daily Living Activities

Sixteen studies with 708 subjects were included in the analysis to compare the effectiveness of VR therapy and control groups on daily living activities in stroke patients. The results showed no significant improvement in daily living activities scores following VR therapy compared with conventional therapies (Z = 1.65; P = 0.10; SMD = 0.15; 95%CI = −0.03–0.34; P < 0.05) (Figure 7). There was moderate heterogeneity among the included studies (I ² = 27%; Tau² = 0.04; Chi² = 20.59; P = 0.15).

FIGURE 7.

Global results for daily living activities.

A funnel plot analysis was conducted to evaluate the potential presence of publication bias in the meta‐analysis. Figure 8 shows the funnel plot, which does not have a substantial asymmetric distribution of studies included in the meta‐analysis. The asymmetry of the funnel plot does not suggest the possibility of publication bias. Likewise, the Egger regression test indicated no significant risk of bias among the included studies (P > 0.05). Furthermore, the leave‐one‐out sensitivity analysis did not affect the significance of the overall effect size, indicating that the findings were not driven by any single study (SMD = 0.15; 95%CI = −0.03–0.34; P < 0.05).

FIGURE 8.

Funnel plot for daily living activities.

3.6. Subgroup Analysis

Regarding the moderate heterogeneity, subgroup analysis was performed by dividing the studies based on different categories: (a) recovery stage, defined as the duration following a stroke, categorized into acute‐subacute (up to 6 months) or chronic (more than 6 months). Studies including participants across different recovery stages were excluded; (b) rehabilitation strategies: defined as therapies included in the experimental groups, categorized as integrated VR rehabilitation when the experimental group also receives conventional therapy, or exclusive VR rehabilitation when the experimental group only undergoes VR; (c) intervention period (IP): refers to the specific timeframe during which the interventions were implemented in the studies, categorized into groups of less than or equal to 4 weeks or more than 4 weeks (IP ≤ 4 Weeks or IP > 4 Weeks); (d) VR interaction approaches: referred to as the interface in games, sensor interactive games include studies using gaming consoles such as Wii and Kinect, whereas manual interactive games encompass studies employing games controlled via mouse or joystick; (d) games diversity: this term pertains to the diversity of applications and games used in the research. Within this subcategory, the studies were classified into commercial games, widely available for purchase and not specifically developed for rehabilitation, or non‐commercial games, typically created for rehabilitation by researchers with experimental or clinical purposes.

3.6.1. Recovery Stage

In motor function, the patients at the acute‐subacute stage showed significant effects (13 studies, Z = 3.36; P = 0.0008) (Figure 9A), whereas patients with chronic stroke stage did not show significant effects (12 studies, Z = 1.68; P = 0.09) (Figure 9A). However, there were no significant differences between the groups (P = 0.84).

FIGURE 9.

Subgroup results by recovery stage. Panel (A) presents the impact on motor function, whereas panel (B) focuses on manual dexterity, and panel (C) on daily living activities.

Concerning manual dexterity, the patients at the acute‐subacute stage showed significant effects (14 studies, Z = 2.77; P = 0.006) (Figure 9B), whereas patients with chronic stroke stage did not show significant effects (12 studies, Z = 0.54; P = 0.59) (Figure 9B). There was a marginal significance difference between the groups (P = 0.07).

In terms of daily living activities, the patients at the acute‐subacute stage did not show significant effects (13 studies, Z = 1.48; P = 0.14) (Figure 9C), and patients with chronic stroke stage did not show significant effects either (3 studies, Z = 0.56; P = 0.58) (Figure 9C). There were no significant differences between the groups (P = 0.92).

3.6.2. Rehabilitation Strategies

Regarding motor function, studies that are categorized in the integrated VR rehabilitation group (VR combined to conventional intervention) showed a significant effect of VR compared to the control group (20 studies, Z = 4.35; P < 0.0001) (Figure 10A), while studies in the exclusive VR rehabilitation group (Only VR) showed a marginal effect compared to the control group (8 studies, Z = 1.87; P = 0.06) (Figure 10A). However, there were no significant differences between the groups (P = 0.94).

FIGURE 10.

Subgroup results by rehabilitation strategies. Panel (A) presents the impact on motor function, whereas panel (B) focuses on manual dexterity, and panel (C) on daily living activities.

With respect to manual dexterity, studies that are categorized in the group integrated VR rehabilitation (VR combined to conventional intervention) showed a significant effect of VR compared to the control group (19 studies, Z = 2.87; P = 0.004) (Figure 10B), while studies in the exclusive VR rehabilitation group (Only VR) did not show a significant effect compared to the control group (6 studies, Z = 0.06; P = 0.95) (Figure 10B). There was a marginal significance difference between the groups (P = 0.07).

Regarding daily living activities, studies that are categorized in the integrated VR rehabilitation group (VR combined to conventional intervention) showed a no significant effect of VR compared to the control group (13 studies, Z = 1.41; P = 0.16) (Figure 10C), and those in the exclusive VR rehabilitation group (only VR) also showed a no significant effect compared to the control group (3 studies, Z = 1.16; P = 0.25) (Figure 10C). There were no significant differences between the groups (P = 0.99).

3.6.3. Intervention Period

This period refers to the specific timeframe during which the interventions were implemented in the studies, categorized into groups of less than or equal to 4 weeks or more than 4 weeks (IP ≤ 4 Weeks or IP > 4 Weeks).

Concerning motor function, the studies classified in the group of less than or equal to four weeks showed significant effects (18 studies, Z = 3.19; P = 0.001) (Figure 11A), and studies that were classified in the group of more than four weeks also showed significant effects (8 studies, Z = 2.64; P = 0.008) (Figure 11A). There were no significant differences between the groups (P = 0.35).

FIGURE 11.

Subgroup results by intervention Period. Panel (A) presents the impact on motor function, whereas panel (B) focuses on manual dexterity, and panel (C) on daily living activities.

Regarding manual dexterity, the studies classified in the group of less than or equal to four weeks showed significant effects (17 studies, Z = 2.05; P = 0.04) (Figure 11B), whereas studies that were classified in the group of more than four weeks showed no significant effect (10 studies, Z = 1.26; P = 0.21) (Figure 11B). There were no significant differences between the groups (P = 0.62).

With respect to daily living activities, the studies classified in the group of less than or equal to four weeks did not show significant effects (13 studies, Z = 1.47; P = 0.14) (Figure 11C), and studies that used manual interactive games also showed no significant effect (3 studies, Z = 0.95; P = 0.34) (Figure 11C). There were no significant differences between the groups (P = 0.97).

3.6.4. VR Interaction Approaches

Concerning motor function, the studies that use sensor interactive games showed significant effects (21 studies, Z = 2.93; P = 0.003) (Figure 12A), and studies that use manual interactive games also showed significant effects (7 studies, Z = 3.22; P = 0.001) (Figure 12A). There were no significant differences between the groups (P = 0.72).

FIGURE 12.

Subgroup results by VR Interaction approaches. Panel (A) presents the impact on motor function, whereas panel (B) focuses on manual dexterity, and panel (C) on daily living activities.

Regarding manual dexterity, the studies using sensor interactive games showed no significant effects (18 studies, Z = 1.77; P = 0.08) (Figure 12B), and studies using manual interactive games also showed no significant effect (9 studies, Z = 1.46; P = 0.14) (Figure 12B). There were no significant differences between the groups (P = 0.56).

With respect to daily living activities, the studies using sensor interactive games showed no significant effects (13 studies, Z = 1.72; P = 0.09) (Figure 12C), and studies using manual interactive games also showed no significant effect (3 studies, Z = 0.17; P = 0.87) (Figure 12C). There were no significant differences between the groups (P = 0.35).

3.6.5. Games Diversity

Regarding motor function, the studies that include commercial games showed significant effects (15 studies, Z = 2.44; P = 0.01) (Figure 13A), and studies that include no commercial games also showed significant effects (13 studies, Z = 4.37; P < 0.0001) (Figure 13A). There were no significant differences between the groups (P = 0.70).

FIGURE 13.

Subgroup results by game diversity. Panel (A) presents the impact on motor function, whereas panel (B) focuses on manual dexterity, and panel (C) on daily living activities.

Concerning manual dexterity, the studies that include commercial games showed no significant effects (14 studies, Z = 1.70; P = 0.09) (Figure 13B), and studies that include no commercial games also showed no significant effects (11 studies, Z = 0.72; P = 0.47) (Figure 13B). There were no significant differences between the groups (P = 0.71).

With respect to daily living activities, the studies that include commercial games showed no significant effects (10 studies, Z = 1.21; P = 0.22) (Figure 13C), and studies that include no commercial games also showed no significant effects (6 studies, Z = 1.30; P = 0.19) (Figure 13C). There were no significant differences between the groups (P = 0.46).

4. Discussion

The strength of VR software lies in its ability to incorporate motivational components and provide multimodal feedback. These features can facilitate cerebral reorganization during the phases of highest plasticity, ultimately leading to the recovery of motor skills (Murphy and Corbett 2009; Yin et al. 2014). However, despite its potential benefits, the efficacy of VR rehabilitation is still being studied (Brunner et al. 2016).

A systematic review and a meta‐analysis was conducted here to compare the effectiveness of no immersive VR‐based systems with conventional therapy in the motor recovery rehabilitation of upper limbs after stroke. Initially, 40 randomized studies were identified and included in the review process. Subsequently, in the meta‐analysis, two outlier studies were excluded from the dataset (Cannell et al. 2018; J. Shin et al. 2016) to ensure the robustness and reliability of the results. As a result, 38 randomized studies were finally included, with the aim of presenting a comprehensive view, considering VR technology as a rehabilitation tool, and offering insights into its potential for improving the quality of life for stroke survivors. Subgroups were explored to provide a theoretical framework supporting future VR rehabilitation approaches in the clinical setting to examine the multidimensionality of the variables influencing the heterogeneity of VR rehabilitation studies.

4.1. Quality of the Evidence and Risk of Bias

All included studies in the present meta‐analysis were randomized trials with moderate certainty of evidence according to GRADE and acceptable methodological quality as assessed by the PEDRO scale. The risk of bias analysis revealed concerns in the randomization process and deviations from intended interventions. Issues include inadequate details in the randomization process, potential differences in baseline characteristics, and allocation sequences, which may compromise participant characteristic balance across treatment groups. “Some concerns" classification for deviations stems from challenges in achieving double‐blindness, especially in VR studies where therapists and participants may be aware of treatment group allocation due to the nature of VR therapy. Despite potential validity compromises, the certainty of evidence and methodological quality should be considered for study inclusion, recognizing these challenges as inherent characteristics of VR clinical research.

4.2. Effectiveness of VR

Given the context of the selected virtual reality studies in terms of evidence quality and risk of bias, the meta‐analysis results revealed, consistent with other meta‐analyses, a noteworthy and statistically significant effect in favor of virtual reality across two dimensions assessed for upper limb improvement: (a) motor function (Doumas et al. 2021; Laver et al. 2011 2015; H. S. Lee et al. 2019; Leong et al. 2022; Zhang et al. 2021); and (b) manual dexterity (Doumas et al. 2021; H. S. Lee et al. 2019; Zhang et al. 2021). However, no significant effect was found in (c) daily living activities (Laver et al. 2015; Leong et al. 2022). Some meta‐analyses do not find significant effects of VR on various measures associated with the manual dexterity dimension. This could be attributed to factors such as the inclusion of a limited number of studies (Laver et al. 2017) or the approach of conducting the analysis by dividing it into scales (Leong et al. 2022). Additionally, fine motor control, which involves distal segments of the upper limb, is inherently more challenging to train using VR. Devices such as remote controllers or joysticks may not effectively replicate the complex and precise movements required for these tasks (Hao et al. 2023). The variability in measurement tools and the limited focus on manual dexterity in the included studies may also contribute to the observed discrepancies in outcomes. These factors highlight the need for more targeted research to optimize VR‐based interventions for improving fine motor skills.

In this setting, a wide variety of approaches used by different studies were found in the literature review for the present systematic review and meta‐analysis. Thus, research in VR has intrinsic heterogeneity as a characteristic. Each study used a different VR device, software and tasks with different physical and cognitive demands. Regarding heterogeneity in the study, the authors have attempted to minimize this by restricting studies that employ non‐immersive VR. In the control group, the authors ensured that no other tools involving VR were used to standardize the intervention. Additionally, for the outcomes, selection was limited to scales that the authors consider having higher validity due to their extensive clinical use. However, moderate and significant effects of heterogeneity were observed when analyzing the effectiveness of VR in two of the three evaluated dimensions. Furthermore, the studies included in the meta‐analysis involve patients in various stages of stroke recovery, with significantly varying sample sizes and diverse durations, modalities, and types of games. These factors could potentially contribute to the observed heterogeneity (Peng, Yin, and Cao 2021).

In light of this, the slight asymmetry observed in two of the three funnel plots can be attributed more to the heterogeneity of the included studies rather than publication bias. This is evident as the studies deviate from the center on both sides of the mean line, indicating a wide dispersion of effect sizes. The presence of significant heterogeneity in the analysis further emphasizes the need to explore potential sources of variation among the studies. For this purpose, additional subgroup analyses were conducted to determine the effects of recovery stage, rehabilitation strategies, intervention period, VR interaction approaches and game diversity.

No significant differences were identified in any of the subgroup analyses, as reported by other studies (Laver et al. 2017; Leong et al. 2022). However, there are some findings that are worth mentioning. In the case of motor function and manual dexterity, significant effects were only found for acute and subacute patients, which aligns with other studies discussing neuronal plasticity in the early stages (Huynh et al. 2013). It is also evident that when conventional therapy is combined with VR, the effects are significant in motor function and manual dexterity, whereas this is not the case when VR is used exclusively. This reinforces the premise that VR should be utilized as a complementary therapy alongside physiotherapy and occupational therapy rather than as a standalone intervention strategy. Furthermore, when considering intervention period (IP ≤ 4 Weeks or IP > 4 Weeks), there was no discernible difference in motor function. This is similar to that reported in other meta‐analyses (Laver et al. 2017; Leong et al. 2022).

In the context of motor function, both VR interaction approaches employing motion sensors and those utilizing manual control prove to be equally effective, suggesting that either modality can be beneficial. Additionally, whether the games are commercial or non‐commercial (games diversity), the impact remains consistent in enhancing upper limb motor function. This leads us to the conclusion that the modality and type of game are irrelevant when it comes to improving motor function. This is consistent with what has been reported (Laver et al. 2017), where the type of therapy has a marginal effect on the effectiveness in the recovery of upper limb motor skills, and only becomes relevant when rehabilitation programs are personalized.

With regard to manual dexterity, no effect was found in the VR interaction approach groups (Sensors vs. Manual), this may be because the skills that require fine motor control, involving distal upper extremity segments, cannot be effectively practiced with remote controllers or manual controllers such as a mouse or joystick (Hao et al. 2023).

One noteworthy finding is that found in the results of daily living activities, where no significant pro‐VR effect is observed when all studies are included, and this lack of effect persists when subgroup analyses are conducted. But what is the reason for this? The authors suspect that this could be due to selection bias in measurement, considering that these are self‐report scales that depend on the subjectivity of the respondent. Furthermore, it should be remembered that these scales attempt to capture complex skills that involve a sequence of coordinated movements, as well as integrate social and motivational aspects. The absence of significant differences in the analysis of daily living activities could be due to the fact that changes in this type of scale respond to complex functional skills involving social, motivational, and behavioral components. In other words, the patient may improve their grip strength or decrease pain or spasticity in the affected limb, but this improvement may not necessarily translate into improved functionality and autonomy, such as being able to dress or brush their teeth independently.

4.3. Relevance of Evidence

In line with the rest of the research, improvements in motor function and manual dexterity attributable to the complementary use of VR were observed. A large number of updated studies have been included to assess improvements in these three main dimensions, which are measured to evaluate enhancement in upper limbs. The results here align with recent meta‐analyses, indicating that VR therapy contributes to the recovery of upper limb motor function in stroke patients, across various recovery stages, particularly when the intervention duration exceeds 15 h (Laver et al. 2017).

Furthermore, it has been evidenced that the type of control (VR interaction approach) used in the game (motion sensors vs. manual control) and whether the game used is commercially available, not specifically designed for rehabilitation, or if, on the contrary, it has been designed with the purpose of improving motor skills, does not influence the recovery of patients (Laver et al. 2017).

The above uniform effect across subgroups could be explained by the fact that, although different tools were used to train upper limb movement in VR therapy studies, the common goal was to improve motor abilities and encourage patients to move their affected limbs in a gamified environment. Encouraging movement in patients promotes the brain to establish new connections and utilize its plasticity to recover impaired functions (Franceschini, Mammi, and Ercolini 2001). Additionally, the role of motivation in patient recovery has been studied, and introducing VR as a therapeutic play complement aids treatment adherence (Gil‐Gómez et al. 2011).

Although measures were implemented to mitigate heterogeneity, including the use of a random‐effects model (Deeks, Higgins, and Altman 2019) and controlling for treatment type and measurement scales, heterogeneity was not entirely eliminated, as indicated by indices exceeding 50% in the overall analyses. However, the I² values were lower compared to those obtained in previous studies on motor function (e.g.,Leong et al. 2022).

As mentioned above, obtaining indices that speak to the homogeneity of research in this field of study is complicated by the nature of the measures and interventions. In the clinical setting, participants have injuries located in various regions of the brain and differing degrees of functionality. In addition, each therapist uses the resources available for the intervention of their patients. Thus, a crucial decision is made regarding prioritizing research in real‐world conditions rather than aligning strictly with the demanding standards of scientific research (Laver et al. 2015, 2017).

5. Implications of Physiotherapy Practice

Non‐immersive VR can serve as a valuable complementary tool in physiotherapy for post‐stroke motor recovery. Physiotherapists should consider incorporating this technology into rehabilitation protocols to enhance motor function, manual dexterity, and daily living activities. By targeting these key areas, VR therapy enhances the overall effectiveness of treatment plans. Personalizing VR therapies to address the specific needs and challenges of each patient may lead to more precise and effective interventions. Additionally, identifying the most effective types of VR interventions according to the stages and the types of stroke could optimize rehabilitation outcomes.

Implementing these strategies could complement traditional approaches, potentially providing significant benefits to patients in their recovery process. While VR may not currently be a fully effective standalone option, its integration into rehabilitation protocols has been shown to have significant benefits. However, further research is needed to fully understand the extent of its impact and to optimize its integration into rehabilitation practices. Ongoing studies should examine the specific mechanisms through which VR facilitates motor recovery, as well as identify the most effective types of VR interventions for different stages and types of stroke.

Moreover, as technology continues to advance, personalized approaches can be developed to better suit the individual needs and preferences of stroke survivors. Tailoring VR therapy to address specific impairments and challenges faced by each patient can lead to more targeted and effective interventions. Regardless of the wide range of VR possibilities available for clinical practice, the most important aspect remains to adjust the interventions to the needs and characteristics of each patient. By implementing these personalized strategies, VR therapy offers promising potential as a complementary tool alongside traditional methods for post‐stroke motor recovery.

Author Contributions

Searched literature: R.V., B.R.G.R. Extracted data from the selected studies: R.V., B.R.G.R. Pooled the data: R.V., B.R.G.R., C.M. Analyzed the data: R.V., B.R.G.R. Contributed reagents/materials/analysis tools: J.L.G.M., C.M. Wrote the paper: R.V.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon request.