This review was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) CRD: 42018102759. This review is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (The PRISMA Statement).89

We included randomized controlled trials that consisted of individuals with CP. CP was defined as a group of non-progressive neurological disorders of motor and postural development that occurs in fetal or infant brain development.1 It can be diagnosed by clinical examination and/or imaging tests. There were no exclusions regarding age, CP types, location of motor change, and severity.

We included the following types of VR: video games commercially available (e.g. Nintendo Wii) that were non-immersive or immersive (custom interface); applied in isolation or combined with other treatments. We considered any other type of treatments, no treatment, waiting list, usual care, or placebo as control groups.

The primary outcomes analyzed were upper limb function, lower limb function (divided into gait and strength of the lower limbs), and postural control and balance. The secondary outcomes included were global motor function, spatial functions, perception and cognition, motivation, motor learning, and adverse events. All outcomes needed to be assessed by a validated instrument. We included four time-points for each outcome: post intervention, short (> two weeks but ≤ three months), intermediate (> three months but ≤12 months), and long term follow-up (> 12 months).

The searches were performed in the following electronic databases: EMBASE (Ovid), MEDLINE (Ovid), LILACS, Cochrane library, PEDro, AMED (Ovid), and PsycINFO (Ovid) in March 2020. There were no restrictions regarding year of publication or language. The search strategy used in the databases is described in the Supplemental online material. Conference proceedings were excluded.

We also performed searches on clinical trial site registries such as ClinicalTrials.gov and WHO International Clinical Trials Registry Platform (www.who.int/ictrp/en/) and manual searches on reference lists of eligible trials. Two authors independently screened the study titles, abstracts, and full texts of all identified studies. Any disagreement among the reviewers was resolved by consensus or arbitration by a third author.

One author performed data extraction using a standardized form. Extracted data included name of the authors, country, year of publication, type of PC, outcome measures, age, number of participants, VR type, description of interventions, duration of sessions, frequency of intervention, duration of the intervention, and place of recruitment. Finally, data on treatment effects and number of participants in each group were extracted. Attempts to contact the authors via e-mail were made when the information from the studies was insufficient or unclear. We extracted data from graphs using the WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). Data reported as mean and range of values were converted to mean and standard deviation data.33 If possible, cross-over studies, when identified, had their data extracted preferably at the end of the first phase, if not possible the data were extracted at the end of the last phase and provided in descriptive form.34

We used the 7-criteria Cochrane risk of bias tool.35 Two independent reviewers (JF and RS) analyzed the risk of bias of all included trials. Disagreements between the reviewers were resolved by consensus or arbitration by a third author.

We used mean differences (MD) with 95% confidence intervals (CI) when studies used continuous outcomes variables with the same scale. When different scales were used for continuous outcomes, we opted to use standardized mean differences (SMD) with 95% CI. We followed a system commonly used by Cohen 198836 for interpretation of effect sizes: d = 0.2 represents a small; d = 0.5 a medium; and d = 0.8 a large effect, and used in previous studies.37,38Odds ratios with 95% CI were used to calculate treatment effects or adverse events for dichotomous variables. Meta-analyses were performed using Review Manager 5.3.39 When data could not be included in a meta-analysis, data are presented descriptively.

Heterogeneity of data was evaluated by visual inspections of the forest plot, observing the overlap of the 95% CIs, followed by a Chi-square test and I2 statistic as recommended by the Cochrane Handbook.34 We pooled the data using a random effect model. When we observed substantial data heterogeneity, I2> 50% and p < 0.10, quality of evidence was downgraded due to inconsistency. Where possible, we stratified the analyses into subgroups to verify if the effects varied based on the intervention duration (i.e. ≤ 30 minutes, 30–60 minutes, >60 minutes) or frequency of intervention (≤ 2× week, 3× week, ≥ 3× week). We performed sensitivity analyses for VR type (commercial game or personalized game) for comparisons that presented with moderate level of heterogeneity.

We evaluated the quality of evidence for each outcome using GRADE as recommended by the Cochrane Handbook.34 The quality of evidence was based upon the performance of the studies for five domains: 1) study design and risk of bias (downgraded by one level if 25% or more of the participants in the comparison are from studies with high risk of bias defined as one or more criteria classified as high risk of bias in the study); 2) Inconsistency (downgrade by one level if by visual inspection the presence of wide variation of effect estimates is verified or the I2 test is greater than 50%); 3) Indirectness (downgrade by one level if more than 50% of participants vary from the population of interest (e.g. mixed populations or multiple disorders)); 4) Inaccuracy (downgrade by one level if, in dichotomous outcomes, the sample size is less than 300 participants or for continuous outcomes, less than 400 participants40); and 5) Publication bias (downgrade by one level if publication bias was identified by visual inspection of funnel plots if more than 10 studies are included in the comparison). The quality of evidence was defined as follows:

High quality of evidence: New research is unlikely to change confidence in the effect estimates;

Moderate quality of evidence: Further research is likely to have an important impact on the confidence in the effect estimates and may change the estimate;

Low quality of evidence: More research is very likely to have impact on the confidence in the effect of the estimate and is likely to change the estimate;

Very low quality of evidence: There is much uncertainty about the effect of the estimate;

No evidence: No randomized controlled trials were identified to evaluate this outcome.

The primary findings of this review are included in a "Summary of findings" custom table using GRADEpro GDT.41

In general, the trials used VR for training upper limb function,26,30,31,42–54 lower limbs,19,26,27,50,51,54–59 postural control and balance,24,27,28,45,50,55,58,60–65 global motor function,19,24,27,30,46,47,51,54–59,62,63,66 spatial functions, cognition, and perceptions,23,29,50,52,67–69 and motivational influence.43 The use of VR ranged from custom technological resources27,31,44,52,53,57,65,67–69 to consoles commercially available as Nintendo Wii.42,45,46,48,50,51,55,56,58,60–64 Session duration ranged from 2051,27,29,49,52,53,55,56,60,61,64,67,68 to more than 60 minutes26,28,31,43,44,46–48,67,70with some trials with unspecified time.45,65 The frequency of treatment sessions ranged from two20,23,24,30,43,46,49,53,56,58,62,67 to more than three times per week,26–28,31,50–52,61,63,65,66,68 with two trials not reporting the frequency.60,70 In general, interventions ranged from four to twenty weeks in duration. The use of VR combined with conventional versus conventional rehabilitation was analyzed in 25 trials,19,20,23,24,26,27,29,30,42–50,54,55,59,62,63,66,67,70 VR versus conventional rehabilitation in eight trials,25,28,31,53,56,58,60,64 and VR versus waiting list in five trials.51,52,61,65,68

Risk of bias of the 38 eligible trials are described in Fig. 2. A total of 26 trials performed adequate randomization 23,24,26,27,29–31,42–45,47,48,50,52,53,55,57,58,60,62–65,69,71 Only 11 trials performed allocation concealment.23,24,30,48,50,52,56–58,60,69 All trials were classified as having a high risk of bias for blinding of participants and therapists. A total of 29 trials provided complete information on the outcomes analyzed.19,23,24,26–31,42,44–46,48–50,52,55–60,63,64,68–70 A total of 35 trials reported all of the outcomes analyzed.19,23,24,26–28,30,31,42–53,55–59,61–65,68–71

Fig. 2.

Summary of risk of bias.

(0.71MB).

A total of nine trials analyzed global motor function,19,24,27,30,46,47,55,62,63 three trials spatial function, perception, and cognition29,50,67 for the effect of VR combined to a conventional rehabilitation versus conventional rehabilitation alone. Two trials analyzed global motor function56,58 for the effect of VR versus conventional rehabilitation. Two trials analyzed global motor function49,58 and two trials spatial and executive functions68,69 for the effect of VR versus waiting list or no treatment. Most of the comparisons demonstrated low quality of evidence. The benefits of VR added to a conventional rehabilitation program or used alone is unclear and may result in a slight improvement in spatial perception and executive functions. The evidence suggests that adding VR to conventional rehabilitation or used alone results in little to no difference in global motor function. Further details can be found in the Supplemental online material.

The majority of the subgroup and sensitivity analyzes did not demonstrate significant differences.

We did not assess publication bias because few studies were included in the review.