Effectiveness of Virtual Reality in the Rehabilitation of Motor Function of Patients With Subacute Stroke: A Meta-Analysis

Stroke is a major cause of death and disability in adults. Conventional therapy (CT) has limited effectiveness, and therefore, various virtual reality (VR) rehabilitation programs have been designed. However, their efficacy in regaining motor function in patients with subacute stroke is questionable. Therefore, we conducted this meta-analysis to determine the efficacy of VR, compared to CT, in restoring motor function in this patient population. Up to October 10, 2020, nine electronic databases were searched for relevant articles reporting the effectiveness of VR in regaining motor function in patients with subacute stroke. This search was updated on March 7, 2021, with no additional added articles. The control group included CT, physical therapy, occupational therapy, or a combination of them. Effectiveness is defined as the positive change from baseline values to the last follow-up point. The Cochrane's revised risk-of-bias tool was used to determine the quality of included trials. A metaregression analysis was conducted to determine the effect of “time since last stroke” on reported outcomes. Publication bias and sensitivity analyses were also carried out. A total of 19 studies (17 randomized controlled trials, 1 cohort study, and 1 crossover trial) were included in the qualitative analysis, whereas 16 trials were meta-analyzed. A great improvement in motor function was noted in the VR group, when compared to preintervention values [standardized mean difference (SMD) = 1.14; 95% confidence interval (CI) = 0.77–1.52; I2 = 82%; P < 0.001]. When compared to CT, VR resulted in mild improvement in motor function (SMD = 0.47; 95% CI = 0.22–0.72; I2 = 75%; P < 0.001). However, upon trim-and-fill adjustment, this finding was deemed insignificant (SMD = 0.08; 95% CI = −0.16 to 0.33; I2 = 82.6%; P < 0.001). Ten studies had low risk, five had some concerns, three had high risk, and one had a moderate risk of bias. VR programs can be used jointly with CT for the rehabilitation of the motor function of patients with subacute stroke. However, more studies are still warranted to determine the effectiveness of these interventions in retaining the cognitive function and physical performance of such patients.

Stroke is a major cause of death and disability in adults. Conventional therapy (CT) has limited effectiveness, and therefore, various virtual reality (VR) rehabilitation programs have been designed. However, their efficacy in regaining motor function in patients with subacute stroke is questionable. Therefore, we conducted this meta-analysis to determine the efficacy of VR, compared to CT, in restoring motor function in this patient population. Up to October 10, 2020, nine electronic databases were searched for relevant articles reporting the effectiveness of VR in regaining motor function in patients with subacute stroke. This search was updated on March 7, 2021, with no additional added articles. The control group included CT, physical therapy, occupational therapy, or a combination of them. Effectiveness is defined as the positive change from baseline values to the last follow-up point. The Cochrane's revised risk-of-bias tool was used to determine the quality of included trials. A metaregression analysis was conducted to determine the effect of "time since last stroke" on reported outcomes. Publication bias and sensitivity analyses were also carried out. A total of 19 studies (17 randomized controlled trials, 1 cohort study, and 1 crossover trial) were included in the qualitative analysis, whereas 16 trials were meta-analyzed. A great improvement in motor function was noted in the VR group, when compared to preintervention values [standardized mean difference (SMD) = 1.14; 95% confidence interval (CI) = 0.77-1.52; I 2 = 82%; P < 0.001]. When compared to CT, VR resulted in mild improvement in motor function (SMD = 0.47; 95% CI = 0.22-0.72; I 2 = 75%; P < 0.001). However, upon trim-and-fill adjustment, this finding was deemed insignificant (SMD = 0.08; 95% CI = −0.16 to 0.33; I 2 = 82.6%; P INTRODUCTION Stroke is a major cause of mortality and acquired disabilities in the adult population (1,2). Recently, the risk of stroke-related mortality has been greatly reduced because of the improved accessibility to healthcare and stroke management protocols, which include recanalization therapy (3), decompression therapy (4), and stroke unit management (5). However, there is a remarkable increase in the number of neurologically impaired patients with significant disabilities (6). Of whom, only a few may regain some functionality in affected upper limb (UL) or lower limb (LL) (7,8). This, in turn, will greatly impact the affected individuals' ability to self-care and their engagement in social activities. As most activities of daily living (ADLs) involve the use of ULs and LLs, it is of great importance to improve their functional state in post stroke patients.
In clinical practice, post stroke rehabilitation currently depends mainly on promoting neuroplasticity after brain injury (9,10). In an attempt to maximize the effect of neuroplasticity, training must be based on learning repetitive, challenging, and motivating, as well as intense tasks (11,12). In this context, conventional therapies (CTs), in the form of occupational or physical therapy, are commonly used to improve the motor function of affected limbs following brain injury (13)(14)(15). However, researchers have been studying other treatment options because conventional rehabilitation programs are often timeconsuming and resource-intensive, and their outcomes rely mainly on the ability and prior training of the interventionist. Therefore, virtual reality (VR) has gained attention in the past decades for its potential benefits in promoting motor recovery in stroke survivors. Moreover, it has been reported that repetition, intensity, and dose in CT settings are not sufficient to reach plasticity-based optimal motor recovery (16). The aforementioned limitations drove the introduction of new options of potential benefit in regaining the motor function in affected individuals, such as VR.
VR-based interventions are now used as therapeutic options for promoting neurorehabilitation in patients with stroke, enabling patients to perform their daily activities, which are difficult to be carried out in a rehabilitative facility. Also, compared to the standard CT, VR therapy is designed to offer entertainment and joy, and therefore, it encourages patients to participate more in the rehabilitative program (17,18). Moreover, VR programs can be carried out in clinical settings and at a low cost, and therefore, many patients can comply with them.
VR training therapy has been increasingly used in order to facilitate motor recovery in stroke survivors, and the majority of published articles investigated patients in the chronic stage. They also have shown that the use of VR is beneficial in this regard. For example, a recent systematic review and meta-analysis concluded that VR therapy could moderately improve the motor function of both ULs and LLs in patients with chronic stroke in comparison with CTs (19). Moreover, compared to CT, VR has been shown to result in a positive impact on balance as well (20).
That being said, the number of studies investigating the effect of VR-based interventions in patients with subacute stroke remains scarce (9,(21)(22)(23), and no conclusions have been reached in this regard. This topic is of great interest because (1) the plasticity of the brain remains good during the acute and subacute phase (24,25) and (2) the effectiveness of rehabilitation therapy-induced neuroplasticity is limited in chronic stroke patients, particularly those who missed the window of opportunity that is present during the subacute phase (when the brain plasticity peaks) (26).
Therefore, we conducted the current investigation to systematically review the available high-quality evidence in the literature regarding the use of VR rehabilitation interventions, compared to CTs, in regaining the functional state of affected limbs among patients with subacute stroke.

Search Strategy and Study Selection
The study process was conducted following the accepted methodology recommendations of the PRISMA checklist for systematic review and meta-analysis where registration of the protocol is not mandated (27). An electronic database search was conducted for relevant studies published from inception until October 10, 2020, in nine databases: PubMed, Google Scholar, Scopus, Web of Science, The New York Academy of Medicine, Virtual health library, the System for Information on Gray Literature in Europe, ClinicalTrials.gov, and meta Register of Controlled Trials. Of note, this search was updated on March 7, 2021, to include any recently published relevant articles. However, no additional articles were found eligible.
For the purposes of conducting this research, we used the definition of the subacute phase of stroke as the period from 7 days to 6 months following the occurrence of stroke, as highlighted in the recent consensus of the Stroke Recovery and Rehabilitation Roundtable Taskforce (28). The search was conducted using the following keywords: (subacute stroke OR stroke) AND (virtual reality) AND (trial OR RCT or random OR randomized) and/or medical subject (MeSH) terms, as appropriate. We further did a manual search of references in our included articles to avoid missing relevant studies (29,30). The search process was done based on the PICO framework: participants were any patient with subacute stroke (as defined above) affecting the motor function of upper or lower extremities, the interventions were VR-based therapies, the comparison was CTs, and treatment effectiveness was the outcome of interest. "Conventional therapy" could consist of usual CT, physical therapy, occupational therapy or a combination of any of them. "Effectiveness" was defined as the positive change from baseline values to the last follow-up point. The change in motor function was our primary outcome of interest; in the case of multiple scales, effectiveness was measured by the change in the scale that was reported as the primary outcome or most relevant to other studies (to maintain homogeneity). We also aimed to highlight the feasibility of using different VR programs in patients with subacute stroke.
We included all original studies that assessed the effectiveness of different VR programs in subacute stroke patients. However, articles were excluded according to the following exclusion criteria: (1) nonoriginal studies or nonhuman (in vitro or animal) studies; (2) duplicate records, overlapped data, or when data could not be reliably extracted; (3) incomplete reports; and (4) abstract-only articles, reviews, thesis, books, conference papers, or articles without available full texts.
The title and abstract screening were performed by four independent reviewers. Then, three independent reviewers performed a full-text screening to ensure the inclusion of relevant articles in our systematic review. Any disagreement was resolved by discussion and referring to the senior author when necessary.

Data Extraction
Two authors developed the data extraction sheet using the Microsoft Excel software. Data extraction was performed by three independent reviewers using the Excel sheet. The fourth independent reviewer performed data checking to ensure the accuracy of extracted data. All disagreements and discrepancies were resolved by discussion and consultation with the senior author when necessary.

Risk of Bias
Three independent reviewers evaluated the risk of bias of included studies. For randomized controlled trials, we decided to use Cochrane's revised quality assessment tool (ROB-II) (31). For non-randomized studies, the risk of bias in non-randomized studies of interventions (ROBINS-I) was used to assess the quality of included studies (32). Any discrepancy between the reviewers was solved by discussion.

Statistical Analysis
All data were analyzed using the R software, version 4.0.2 (33). The standardized mean difference (SMD) effect size and its variance were calculated using the preintervention and postintervention data for both intervention and control groups (34,35). Using a "meta" package, changes from baseline for both intervention and control were analyzed to calculate the pooled SMD and the corresponding standard errors (36). The SMD was used because of the difference in measurement methodology among included studies, and it is more generalizable than the mean difference (37).
The related 95% confidence intervals (CIs) of the computed effect size were calculated using a fixed-or random-effects model based on the extent of heterogeneity. To assess heterogeneity, Q statistics and I 2 test were used, with I 2 value higher than 50% or P < 0.05 considered significant (38). Moreover, we conducted Egger regression test to assess publication bias, which was considered significant when P < 0.10 (39,40). Whenever publication bias was found, the trim-and-fill method of Duvall and Tweedie was applied to add studies that appeared to be missing (41) to enhance the symmetry. Furthermore, metaregression was used to explore the effect of "time passed since the last stroke" on the VR effectiveness (42). In the case of statistically significant results, a leave-one-out sensitivity analysis was performed by iteratively removing one study at a time to confirm that our findings were not driven by any single study (43).

Search Results
We identified 861 records after excluding 202 duplicates using the Endnote X9 software. Title and abstract screening resulted in 88 records for further full-text screening. No articles were added after performing a manual search. Finally, a total of 19 studies were included in the systematic review, and 16 articles were eligible for meta-analysis (Figure 1).

Study Characteristics
Nineteen studies compared VR vs. controls. According to the countries in which the included trials were conducted, four studies were conducted in Turkey, four in Korea, two in Norway, two in Italy, two in Australia, two in Canada, two in China, and one in Germany. Based on the design of included studies, 17 articles were randomized controlled trials, one was a cohort study, and the last one had a crossover design ( Table 1).

Risk of Bias
The overall risk of bias was low in nine studies, whereas five studies had some concerns, and three had a high risk of bias (Figure 2). For non-randomized trials, the risk of bias was moderate in one of them and low in the other (Figure 3).

Effectiveness of VR Therapy
The details of VR-based interventions and the control groups are provided in Table 2. Sixteen studies evaluating the effectiveness of VR therapy (preintervention vs. post intervention) were included in the analysis. There was a significant improvement in motor function score following VR therapy in patients with subacute stroke, compared to their preintervention score (SMD = 1.14; 95% CI = 0.77-1.52; P < 0.001). According to Egger regression test, there was no significant risk of bias among included studies (P = 0.275); however, significant heterogeneity was present (I 2 = 82% and P < 0.001) (Figure 4). The contribution of each study to the overall heterogeneity is presented in Supplementary Figure 1. Moreover, the leaveone-out sensitivity analysis did not affect the significance of the overall effect size, indicating that the findings were not driven by any single study Supplementary Figure 2. The meta regression analysis did not show any significant effect of the post stroke duration (days) on the treatment effects (P = 0.230) (Supplementary Figure 3).
In the same context, 15 studies evaluating the comparative effectiveness of VR therapy and CTs were included in the analysis. There was a significant improvement in motor function following VR therapy in patients with subacute stroke, compared to those undergoing CT (SMD = 0.47; 95% CI = 0.22-0.72; P < 0.001). Nevertheless, there was significant heterogeneity among the included studies (I 2 = 75% and P < 0.001) ( Figure 5). The contribution of each study to the overall heterogeneity is presented in Supplementary Figure 4. The leave-one-out sensitivity analysis did not affect the significance of the overall effects indicating that the findings were not driven by a single study (Supplementary Figure 5).
Noteworthy, according to Egger regression test, there was a significant risk of bias among the included studies (P = 0.001). Therefore, we used the trim-and-fill method to "trim" the studies that caused the asymmetry of the funnel plot so that the overall effect estimate would be minimally affected by publication bias and then to "fill" imputed missing studies in the plot according to the bias-corrected overall effect estimate. Unfortunately, upon using this method, the overall effect estimate was deemed insignificant (SMD = 0.08; 95% CI = −0.16-0.33; P = 0.507), with a higher heterogeneity (I 2 = 82.6% and P < 0.001) (Figure 6). This indicates that VR therapy is not associated with significant improvement in motor function as compared with CT. The meta regression analysis did not show any significant effect of the post stroke duration (days) on the treatment effects (P = 0.413) (Supplementary Figure 6).

DISCUSSION
The field of neurorehabilitation after stroke, especially during the subacute phase, is still evolving. The current paradigms of To prove the efficacy of the low-cost Kinect-based virtual rehabilitation (VR) system for UE recovery among patients with subacute stroke Low-cost Kinect-based UE rehabilitation system was not more efficacious compared with sham VR. However, the compliance in VR was good and VR system induced more arm motion than control and similar activity compared with the conventional therapy, which suggests its utility as an adjuvant additional therapy during inpatient stroke rehabilitation. Lee To evaluate the acceptability of robot-assisted gait training (RAGT) with and without VR and the feasibility of potential outcome measures to guide the planning of a larger randomized controlled trial (RCT).
VR-augmented RAGT resulted in high acceptability and motivation, and in a reduced drop-out rate and an extended training time compared to standard RAGT. This pilot trial provides guidance for a prospective RCT on the effectiveness of VR-augmented RAGT.

(Continued)
Frontiers in Neurology | www.frontiersin.org To compare the efficacy of novel interactive, motion capture-rehabilitation software to usual care stroke rehabilitation on physical function No differences between the rehabilitation units were seen except in lateral reach (less affected side) (P = 0.04). No adverse events were recorded during therapy.

(Continued)
Frontiers in Neurology | www.frontiersin.org Xbox KinectTM training appears feasible and safe in upper extremity rehabilitation after stroke. It could enhance motor and functional recovery of the affected upper extremity as an adjunctive method.

(Continued)
Frontiers in Neurology | www.frontiersin.org To investigate the efficacy of balance training using video game-based intervention on functional balance and disability in individuals with hemiparesis due to stroke in subacute phase.
Balance training performed with a Wii Fit as an add on to the conventional therapy was found to be more effective than conventional therapy alone in improving balance and reducing disability in patients with subacute stroke.

(Continued)
Frontiers in Neurology | www.frontiersin.org  Frontiers in Neurology | www.frontiersin.org neurorehabilitation strategies in order to enhance the motor function of affected extremities are focused on high-intensity, task-specific interventions with the main character of repetition (58,59). The repetition of such programs or tasks potentiates the acquisition of new along with the retrieval of previous motor functions of affected limbs, leading to long-term potentiation. VR game-based programs are a group of interventions that are directed to incorporate affected populations in a computersimulated environment while giving them almost real-time feedback on their performance. According to previous evidence (59,60), the degree to which VR programs can aid or facilitate current CT programs is still to be determined. We conducted this meta-analysis to determine the comparative effectiveness of VR-based systems in the rehabilitation of the motor function of affected limbs in patients with subacute stroke. We aimed to investigate the effect of VR on restoring the motor function of both UL and LL; however, this was not possible because the majority of included studies investigated the effect of VR on the motor function of the upper limbs, whereas only two studies recruited patients with hemiparesis with no description of which limb was investigated. Therefore, we reported the change in the motor function of affected limbs, with no discrimination, following VR-based therapy as compared to CT. The CT group included various programs: physical therapy, occupational therapy, functional electrical stimulation, or a combination of them.
In our study, we noted that VR interventions resulted in a great improvement in motor function of affected limbs in patients with subacute stroke compared to preintervention values (SMD = 1.14; 95% CI = 0.77-1.52). However, we detected significant considerable heterogeneity, which on further analysis revealed that the study of Afsar et al. (45) was the major contributor to the resultant heterogeneity. This could be related to their heterogeneous population compared to other studies. They included patients with mild to moderate upper limb motor deficits, and they were further subdivided into two groups (VR vs. This group received: (1) occupational therapy (0.5 hours /5 days/2 weeks) and (2) kinetic-based VR rehabilitation system. This system three types of programs: "Push Museum," "Apple Run," and "Fruit Market." These programs were made using the Unity three-dimensional (3D) game engine (Unity Technology Inc., San Francisco, CA). This system induces arm motions important during rehabilitation (reaching, wrist extension, hand grasping, and releasing).

h x 5 days x 2 weeks
This group received: (1) occupational therapy (0.5 hour/5 days/2 weeks) and (2) Sham VR rehabilitation. A similar strategy as for real VR was applied and subjects were instructed to use the hemiparetic upper limb. In this group patients reach to the button and pushing the selected button during the cognitive task.

h x 5 days x 2 weeks
Lee, 2016 This group received: (1) conventional rehabilitation program (same as control) and (2) canoe game-based VR rehabilitation therapy. The canoe game-based VR training program was conducted using the Nintendo Wii Sports Resort package. Participants paddled by grasping the motion controller, alternating between hands while sitting on the springboard. They also adjusted their trunk to maintain balance on the springboard during paddling.

h x 3 days x 4 weeks
This group received: conventional rehab program alone. The conventional program consists of: physical therapy for gait training and lower limb strengthening (0.5 h twice daily/5 days/4 weeks), occupational therapy to improve performance in activities of daily living (0.5 hour twice daily/5 days/4 weeks), and FES applied to both UL and LL (15 min/5 days/4 weeks).
Variable Bergmann, 2018 (46) This group received: (1) RAGT (standard protocol) and (2) VR training: using two VR scenarios (the coin scenario and the dog scenario)-both scenarios took place in a forest, where subjects had to walk along a straight alley in the middle of the screen, and solve different tasks by controlling the avatar's speed by adapting their motor activity. In the dog scenario, patients' activity was displayed as a red dot on the path, where the patient is instructed to place the red dot underneath the dog. In the coin scenario, coin scenario, patients had to collect coins, and avoid rocks that were placed on the path. The dog scenario was applied in sessions 1&2, while the coin scenario was applied in sessions 3&4. Furthermore, this group received physiotherapy sessions (1 hour x 2 days x 4 weeks).

h x 3 days x 4 weeks
Standard RAGT using the robotic-driven gait orthosis lokomat. Patients were fixed into the gait orthosis with a harness, which was attached to a body-weight support system, and had cuffs placed around the legs. Furthermore, this group received physiotherapy sessions (1 h x 2 days x 4 weeks).

(Continued)
Frontiers in Neurology | www.frontiersin.org This group received: (1) conventional physical therapy (same as control) + (2) game-based VR canoe paddling training using the Nintendo Wii Sports Resort game. Patients performed a paddling movement with both hands grasping the motion controller that was inserted in a separate canoe paddle accessory. Participants operated the paddle in the direction of the virtual character displayed on an LED TV 42LN549C screen. They were also instructed to focus on trunk control to maintain their balance on top of the springboard, while canoe paddling. This group received: (1) conventional rehab program (same as control) and (2) Leap Motion-based VR training, which consisted of a computer and a leap motion controller. The controller can track, with sub-millimeter accuracy, the movement of multiple hands and fingers. These games were designed to focus on the development of the pinching, grasping, and individuating motor skills of fingers; the improvement of the dexterity and coordination of the digits; the improvement of the ability to flex and extend the hand, the pronation and supination of the forearm; the increase in the joint range of motion of the hand, elbow, shoulder and wrist; the improvement of the movement speed, muscle strength, and motor control.

min x 5 days x 4 weeks
This group received conventional rehab program only, which included stretches, strength, balance, gait, and functional training.     conventional). Furthermore, patients in both groups received CT. Also, patients in the VR group received mild training of 30 min per day 5 days a week for 4 weeks using the Xbox Kinect gaming console in addition to CT of 60 min five times a week for 4 weeks. The aforementioned factors could contribute to the observed heterogeneity. Furthermore, the sensitivity analysis revealed that the effect size "great improvement" was not driven by a single study. The variable "time since last stroke" had no significant effect on the outcomes.
Upon comparing VR programs to CT, our analysis revealed a significant mild improvement (0.5 ≤ SMD ≤ 0.8) in the motor function of the affected limb in favor of VR-based rehabilitation programs. We observed considerable heterogeneity, and the study of (61) was the major contributor. This could be explained by the fact that both groups (VR and control group) in the previous study had VR training of the upper limb in addition to their assigned interventions. Surprisingly, following the trimand-fill adjustment of risk of bias, the observed results were deemed insignificant (SMD = 0.08; 95% CI = −0.16 to 0.33), indicating no significant difference between both interventions on restoring the motor function of affected limbs. Our findings are comparable to the systematic review and meta-analysis of Laver et al. (62), who assessed the efficacy of VR in improving upper limb motor function in patients with stroke. A significant but mild improvement (SMD = 0.28) in motor function was noted in favor of VR-based interventions. Noteworthy, only 2 of the 12 analyzed studies recruited patients with subacute stroke (9,12). Therefore, our study provides stronger evidence in regard to patients with subacute stroke, indicating that VR does not result in significant improvement in motor function. In 2019, an umbrella review of meta-analyses was conducted to determine the impact of various neurorehabilitation interventions on changes in ADLs in patients with subacute stroke (63). A total of 55 meta-analyses were investigated, reporting 21 subacute rehabilitation interventions. Of investigated interventions, VR was reported by two meta-analyses (62,64) and resulted in mild to moderate improvement of ADLs in the subacute phase of stroke rehabilitation. However, the authors highlighted that the lack of high-quality evidence in analyzed meta-analyses highlights the need for more research. In our review, the majority of studied interventions were designed mainly to improve motor function rather than other outcomes, including cognitive function, activity performance, ADLs, or feasibility. As for rehabilitation of motor function, it was reported that constraint-induced motor therapy is considered, by far, the most promising intervention in patients with stroke, in general, based on the findings of a recent systematic review (59).
The feasibility of VR-based rehabilitative interventions was assessed in three trials. In the crossover trial of Iosa et al. (57), four elderly patients with subacute stroke were allocated to receive six sessions of 30 min of leap motion controller-based intervention in addition to CT. Participation in these sessions was excellent in three patients and very good in the remaining patient. This highlights the feasibility of this intervention for neurorehabilitation in this patient group due to the easiness of its use without the need for the subject to stand alone. However, because of the limited number of included participants, these conclusions still need confirmation by larger, well-conducted trials. In the pilot, randomized controlled trial of (12), patients were allocated to receive Nintendo Wii gaming (nine patients) or recreational therapy in the form of playing cards, bingo, or Jenga (eight patients). Feasibility was reflected by the total time of intervention receipt. The mean total time of VR was comparable to that of recreational therapy (388 vs. 364 min; P = 0.75). Meanwhile, Brunner et al. (56) recorded 50 videos of patients with subacute stroke who were allocated to receive either VR or CT, and the authors reported higher feasibility in the VR group, with higher mean time of active practice (77.6 min) compared to the CT group (67.3 min). Of note, the fact that patients knew they were being recorded affects the validity and interpretation of this finding. Therefore, despite the promising feasibility of VR-based interventions for the neurorehabilitation in patients with subacute stroke, more robust trials of larger sample sizes are still warranted to reach a more definitive conclusion.
In view of the paucity of well-designed randomized controlled trials, the limited funding for stroke rehabilitation programs and research, and the limitations of CT, VR interventions can be used adjunctly with CT for the following reasons: (1) VRbased interventions are accessible for all patients with subacute stroke, (2) the low cost of VR programs, and (3) it requires no special resources or assistance. Therefore, we hypothesize that VR interventions might potentiate motor rehabilitation in subacute stroke patients when added to CT. However, this finding remains inconclusive for the following reasons.
First, a wide variety of VR programs were reported. Each of these programs included different activities aiming to improve the function of certain parts of the affected limb. For example, the VR program in the study of (45) included the active abduction, adduction, flexion, and extension of the shoulder in addition to flexion and extension of the elbow. Meanwhile, Choi et al. (47) focused on regaining the motor function of affected ULs by promoting certain activities in their VR program, including extension and internal and external rotation of the shoulder; flexion and extension of the elbow; and pronation and supination of the forearm. On the other hand, another trial focused on regaining more delicate motor functions, and therefore the VR programs were more focused on improving the delicate movements of the fingers (promoting the development of pinching and grasping), digits (improving their dexterity and coordination), and hand (enhancing flexion and extension) (51). Other VR programs were primarily used to regain arm motions, such as reaching, wrist extension, and hand grasping and releasing (22).
Another point worth mentioning is the differences in VR programs among included trials. Typically, VR interventions include a head-mounted display, which allows users to experience 3D content (either videos or games) in an immersive virtual environment. In the case of VR interventions in stroke patients with affected ULs, leap motions are usually used in order to track the user's movements. Three of the included trials in our review used this method in their treatment protocol to regain the motor function of affected limbs in patients with subacute stroke (44,51,57). However, the remaining studies used the terms "virtual reality intervention" and "game-based intervention" interchangeably. For instance, their treatment protocols incorporated the use of different gaming-based engines or systems, such as the Xbox Kinect system (45,52), Unity 3D game engine (22), Nintendo Wii Sports Resort package (12,21,23,47,53), the YouGrabber system (56), Jintronix Rehabilitation system (49), and PlayStation console (50).
Of note, the intensity of the VR intervention was widely variable among included studies. For example, in the study of Brunner et al. (56), a total of 62 patients with subacute stroke underwent VR therapy using the YouGrabber gaming system. The VR therapy lasted for 1 h each and was performed four to five times a week for 12 consecutive weeks. On the other hand, the intensity of the VR intervention was low in the study of (55), where patients with subacute stroke underwent video game-based VR therapy sessions that lasted 20 min each. These sessions were repeated three times per week for a total of 4 consecutive weeks. Therefore, we could not determine the effect of VR intensity on our outcome of interest.
The aforementioned points reflect the heterogeneity of studied populations at baseline, which could also explain the reason for encountering the highly significant considerable heterogeneity in our analyses.
Although our study is the first study to investigate the efficacy of VR interventions in restoring the motor function of patients with subacute stroke, several limitations were encountered. First, most included studies compared VR programs in addition to CT compared to CT alone, which subsequently accounted for more rehabilitation time in the experimental group. This could potentially account for a bias in favor of VR therapy because both the intensity and frequency of rehabilitation, per se, are known to be directly correlated with beneficial functional outcomes. Second, the number of participants in included studies was small, ranging from 4 to 120 patients, and thus, the generalizability of our findings could be affected. Third, eight randomized controlled trials had some concerns and a high risk of bias. Based on these limitations, we recommend conducting more robust, well-designed trials of larger sample sizes and longer follow-up periods to reach more definitive conclusions. Fourth, outcomes were assessed directly after the application of the VR interventions with a short follow-up period, ranging from 2 to 12 weeks (44,56). Therefore, we are not confident that the improvement in motor function following the implementation of VR protocols is long-lasting. Eventually, our systematic review was conducted mainly to determine the effect of VR intervention on the motor function of patients with subacute stroke. Only two articles reported outcomes related to the cognitive function of subacute stroke patients following VR therapy; however, these studies used different measurement tools with different cutoff values for defining cognitive impairment (47,54). Therefore, it was inapplicable to conduct a meta-analysis on this outcome, and thus, more research is needed to determine the impact of these VR interventions on other outcomes, such as cognitive function, ADLs, and quality of life.
VR provides a great improvement in motor function in patients with subacute stroke, compared to the preintervention state. However, when compared to CT, mild to no significant improvement in motor function was noted. Therefore, VR can be used as an adjuvant to CT in restoring the motor function of affected limbs. That being said, more studies are still warranted to reach a more definitive conclusion and to investigate the effect of VR on the cognitive function and physical performance of affected patients.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

AUTHOR CONTRIBUTIONS
Q-cP and LY wrote the manuscript. Q-cP, LY, and YC collected and analyzed the data. All authors contributed to the article and approved the submitted version.