Data was extracted using a standardized extraction sheet using Microsoft excel. We included study characteristics (author, year, country, age, total participants, number and percentage of male and female participants in case and control group, case group and control group), near point of convergence (NPC), positive fusional vergence (PFV), convergence insufficiency symptoms survey (CISS), near phoria before therapy and after therapy.20,52-55 Although other parameters like positive fusional vergence, near phoria and convergence insufficiency symptoms survey (CISS) are also used for CI diagnosis, they are not reported or available in all our selected studies.18,56,57 We found a standard parameter NPC in all our selected study. Therefore, we used NPC during our analysis. Discrepancies in extraction were resolved by discussion with the co-author.

We used the Newcastle-Ottawa quality assessment scale (NOS) to assess the quality of each selected study.58,59 NOS has three categories: selection(representativeness of the exposed cohort, selection of the non-exposed cohort, ascertainment of exposure, demonstration that the outcome of interest was not present at the start of the study), comparability (comparability of cohorts/cases and controls based on design or analysis), outcome (assessment of outcome, was follow-up long enough for outcomes to occur?, adequacy of follow-up) with a total of eight items.58 Here, the selection has 0 to 4 points, comparability has 0–2 points, and outcome has 0–3 points. By adding points from each section, we quantify the quality of the articles.

We used Cochrane risk of bias (RoB) tool which is widely used in systematic review and meta-analysis to evaluate the internal validity of trial results and to ensure that results are accurate and reliable.60,61 In short, it is a way of checking whether clinical trials are trustworthy and fair. This tool identifies and evaluates the biases that may arise during study design, data analysis, outcome assessment, reporting of data and other stages. Potential biases can overestimate or underestimate the results which may significantly impact the study outcome. This tool has six domains for assessment: selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants), detection bias (blinding of outcome assessment), attrition bias (handling of incomplete data), reporting bias (selective outcome reporting) and other bias. Each domain is assigned a judgment such as: low risk, high risk, unclear risk. The assessment was done by the authors independently and discrepancies were solved by discussion.

R studio was used for statistical analysis in this study.62 We used the mean and standard deviation of near point of convergence before and after VR therapy for statistical analysis as it was only available in all our selected studies.20,52,53 We used a random effect REML model and created a forest plot.63,64 The heterogeneity of the studies was calculated by the Q test and I2 statistics.65,66 We also calculated the sensitivity of the studies to ensure that findings are reliable and stable using the “Leave-One-Out Sensitivity Analysis” method.67 We created a funnel plot to visualize the publication bias.68,69 As the number of selected studies was low, we used Begg's test and Egger's test to get reliable publication bias.70-72

We initially identified 649 potential records from four databases (Google scholar, Scopus, PubMed, and Science Direct). Then, after titles and abstracts, 636 records were excluded as they did not match inclusion criteria or meet exclusion criteria. Afterward, 13 records were fully assessed for eligibility. However, 10 records were excluded after being fully assessed as they did not match the inclusion criteria or meet the exclusion criteria. Finally, only 3 records were included in the study. The number of records is small as there are few records relevant to this topic, and it is a new emerging therapy for CI. The flow diagram for study selection is displayed in Fig. 1.

Fig. 1.

Flow diagram of the study selection process in this study showing the number of records retrieved, number of records excluded, and number of records included.

(0.13MB).

The three studies included in this study had a total of 60 patients from 3 different countries (Australia, USA and China). The participants were primarily young adults, and their average age ranged from 18 to 26. The studies included both male and female participants. Two studies had case and control group and one study had only case group. Number of total participants, number of male and female are reported in the Table 1. In addition, percentage of male and female participants are also reported in Table 1.

We found near point of convergence (NPC) value before and after therapy for all the selected study in cases group. Therefore, these data were used for meta-analysis in this study. Extracted NPC values are reported in Table 2. All the studies used centimeters (cm) as the unit of NPC. One study did not have a control group. However, as we found data for case groups in all selected study, it did not affect our study.

We found positive fusional vergence (PFV) value before and after therapy two selected study in cases group. One study did not report PFV in their study. Therefore, we did not perform meta-analysis using PFV. All the studies used prism diopters (∆) as the unit of PFV. Extracted PFV values are reported in Table 3.

We found convergence insufficiency symptoms survey (CISS) for one study only in case groups before and after therapy. The other two studies did not report CISS score. Therefore, we did not perform meta-analysis using CISS. Extracted CISS values are reported in Table 4.

We found near phoria in two selected studies. One study did not report near phoria. Therefore, we did not perform meta-analysis using near phoria. Negative sign in the reported near phoria indicates exophoria. Extracted near phoria values are reported in Table 5.

We found that one study scored 5 points out of 9 in quality assessment. One study scored 6 points out of 9 in the quality assessment. Another study scored 7 points out of 9 in the quality assessment. All 3 studies scored fewer points in the comparability section and scored well in the selection section. Quality assessment values for the selected studies are reported in Table 6.

We also performed Risk of bias assessment using Cochrane risk of bias (RoB) tool for selected three studies. A graphical representation of the result was used to present the results for facile comprehension in Fig. 2.

Fig. 2.

Risk of bias assessment using the Cochrane risk of bias tool.

(0.21MB).

The mean difference with 95% confidence intervals is shown for each included study in Fig. 3. The mean difference and 95% CI intervals were 3.10 [95% CI, 1.33–4.87], 1.97 [95% CI: −0.16–4.10], 5.30 [95% CI: 2.96–7.64] for the included 3 studies. The weight of each study is also shown, and we found Mei ying boon's study had a maximum weight of 44.28% and Chang yaramothu's study had lowest weight of 25.22%. The diamond represents the overall effect estimate at the bottom. The random-effects model indicates an overall mean difference of 3.38 (95% CI: 1.61–5.16).

Fig. 3.

Forest plot of mean differences in near point of convergence (NPC) before and after virtual reality (VR) intervention for each study.

(0.07MB).

We used “The Leave-One-Out” Sensitivity Analysis to show how the overall effect size and error change when we remove each study, one by one, from our analysis.67,76 We can learn from this sensitivity analysis how much each individual study influences the overall findings. For instance, when we removed Mei Ying Boon's study, the average effect size became 3.64 with an average error of 3.96. Then we removed Shinjin Li's study, the effect size increased to 4.20, and the error decreased to 3.14. Eventually, we also removed Chang Yaramothu's study resulted in an effect size of 2.54 with an error of 3.52.

We observed that the effect sizes and error changes are noticeable, but they are not extreme. Thus, it implies that each study contributes differently to the overall outcome, but no single study seems to extremely change the conclusions. Therefore, it indicates that the overall findings of our analysis are relatively stable and not excessively dependent on any single selected study.

Publication bias is defined as when only studies with positive or interesting results get published, while studies with no effect or negative results are often not published.77 This creates a problem because we only see studies that show positive results and do not see negative results, which can make something more effective compared to actual effectiveness.

Funnel plot

A funnel plot is a graphical representation used to evaluate publication bias. It shows whether the results of studies are skewed, or any study tends to show different outcomes compared to other studies.78,79 If publication bias is not present, the plot will look like an inverted funnel or triangle. Selected studies will be symmetrically distributed around the average effect within the inverted funnel. If there is a publication, the plot may look asymmetrical. This lack of symmetry in the funnel plot indicated that selected studies only partially represent all potential findings. Therefore, the conclusions drawn from the meta-analysis might be less reliable.

As the number of studies selected is low and it is tough to draw a reliable conclusion from this limited number of studies using funnel plots in Fig. 4. The somewhat asymmetry in the plot could be due to the small number of selected studies rather than actual publication bias. Therefore, we also conducted Begg's test and Egger's test to evaluate the publication bias.

Fig. 4.

Funnel plot for publication bias. The plot shows the relationship between the effect size (mean difference) and the standard error.

(0.06MB).

We used a random effect model with our selected 3 studies. Our estimated heterogeneity(tau2) is 1.3465, which indicates some variation between the selected studies. We also found moderate heterogeneity as I2 is 54.85%. Then, we conducted a Q test for heterogeneity, and it did not show significant evidence for heterogeneity as P value = 0.1131(P > 0.05). We found that the overall estimated effect size is 3.3837, which indicates that VR-based treatment has a positive effect on the outcome. The confidence interval varies from 1.6096 to 5.01578, which indicates that true effect size lies between them. The p value is 0.0002, which means that there are statistically significant effects of the treatment. However, statistical significance does not always indicate clinical significance. NPC improvement of at least 4–5 cm is generally considered clinically significant. Since the effect size in this study is 3.38 cm, it is approaching towards clinical significance. From “Leave-One-Out” Sensitivity Analysis, we observed that although there are some changes in mean difference but it is not excessive. Therefore, we can conclude that the overall result is reliable as result is relatively stable when the individual study is removed. Initially, we created a funnel plot, but publication bias was not clear. Therefore, we conducted Begg's test and Egger's test for further evidence. We did not find significant evidence of publication bias for our analysis as Begg's test (p-value = 0.33) and Egger's test (p-value = 0.40).

We found that different studies used different parameters for outcome assessment. We only found NPC as a common parameter in our selected three studies. We would find more reliable and robust data if we have more common parameters. However, VR-based CI therapy also showed improvement in other parameters (PFV, CISS, near phoria). The amount of therapy is not same for all included studies. The amount of therapy can significantly affect the outcome assessment. One study used three 20 mins per week session over 6 weeks.53 However, the other two study therapy times were same. One study used 12 one-hour sessions of therapy and another study 1 hour per week for 12 weeks.20,52 Included studies performed two types of outcome assessment: clinical (NPC, PFV, Near phoria) and questionaries (CISS). They are subjective assessment as they depend on patients’ reporting and clinicians ability to observe. This subjective assessment can be error prone. Although, we found positive effect from VR-based CI therapy, its initial cost can be a barrier to its application. Therefore, cost effectiveness analysis is important, which was not performed by any of the included studies. All the included studies were limited to young adults (ages 18–26) which leaves an uncertainty about the effectiveness of CI in children and older adults. However, the studies included participants from Australia, the USA, and China, suggesting cross-cultural feasibility.

Higher patient engagement in VR based CI therapy was mentioned in the study conducted by Li et al.72 On the other hand, real-time tracking and feedback was used by the study conducted by Chang et al.20 which may influence higher compliance level. However, Boon et al.53 study did not explicitly mention compliance data and we did not find anything which may help us to comment on compliance. We believe that this study also has higher compliance as it also used VR. Boon et al.53 used a VR snake game as a therapy modality. On the other hand, Chang et al.20 developed VERVE (virtual eye rotation vision exercise) and used in their study. Li et al.72 used a customized VR training module. A commercially available VR headset (oculus rift) was used in Boon et al.53 study. An eye tracking enabled VR headset was used in chang et al.20 study. On the other hand, Li et al.72 used a high resolution VR display.

VR-based therapy for CI showed positive overall results. VR-based therapy has several advantages compared to others, such as more engagement, personalized, and virtual environment, which is not feasible with other traditional therapies. VR-based therapy may mitigate the drawbacks of other conventional therapy. For example. Office-based vision therapy requires expert clinicians, time, and resources. VR therapy may solve this dilemma by providing therapy at home without expert clinicians. On the other hand, home-based vision therapy is monotonous, has low patient compliance, no real-time feedback, and is less effective. VR therapy may also solve this problem and may provide more effective, real-time feedback and engaging therapy. However, the effectiveness of VR-based CI therapy is comparatively low with other traditional therapies. Total NPC changes after 12 weeks for home-based pencil push-up therapy is −6.4, [95% CI: −7.8, −5.0], for office based vergence accommodative therapy with home reinforcement is −10.4 [95% CI: −11.7, −9.0].80 For VR-based therapy the total average NPC changes for included studies was 3.38 (95% CI: 1.61, 5.16). The reason for less effectiveness may be different therapy periods and less sample size in VR therapy groups.

To be best of authors knowledge, there are no systematic reviews and meta-analysis for VR-based therapy for CI patients. Our analysis will fill this critical gap. In addition, we are providing quantitative data from the multiple existing study, which renders the overall outcome of the VR-based therapy which was not available before. Sometimes, only positive results are published by some individual study, which may mislead the clinicians or researchers. Therefore, data from individual studies may sometimes be less reliable and less robust. However, quantitative data from multiple study by meta-analysis is more reliable because if there is any publication bias that can be identified.

The most important limitation of this review and meta-analysis is the small sample size, with only three studies included. Since it is an emerging therapy, the number of studies is low. Some included study quality was medium, and particularly for comparison, it was low. This may affect the robustness of our findings from the analysis. Although the analysis was conducted on 3 studies, significant evidence of publication bias was not found. We performed meta-analysis only on near point of convergence as it was the only common parameters and other parameters were not available for all selected studies. Additionally, we found variability in therapy protocols among the included studies which complicates drawing robust conclusions. For instance, one study employed three 20-minute sessions per week over six weeks, while others used 12 one-hour sessions spread across different timelines. The populations studied were often homogeneous, primarily involving young adults, which limits applicability to other age groups, such as children or older adults, who might respond differently. While VR-based therapies were found effective, the high cost and technical requirements (e.g., VR headsets, specialized software) limit accessibility, particularly in resource-limited settings. Lifestyle habits (e.g., screen time, occupation, near-work intensity) were not adequately controlled or quantified which may affect the outcome.

Assessing the quality and bias of studies is important for systematic reviews or meta-analysis. It ensures that the findings are trustworthy and reliable. We used Newcastle-Ottawa quality assessment scale (NOS) to assess the quality of our selected study. Our selected 3 studies scored 5, 6 and 7 points out of 9.20,52,53 All three selected studies scored 3 in the selection section which indicates that studies adequately addressed most aspects of participant selection. Two studies scored 1 in compatibility section as they had control group. However, one study did not have any control group. Thus, it scored 0 in comparability section.20 In addition, selected studies did not mention controlling for other confounding variables such as age, gender, or baseline severity of CI. One study scored 3 in exposure section as it had strong methods in exposure/outcome assessment and follow-up methods.52 However, other studies did not have clear follow-up methods.20,53

We used Cochrane risk of bias tool for risk of bias assessment. It has 7 domains for bias assessment. For random sequence generation domain, two studies had unclear risk, and one study had low risk as it mentioned a randomization process clearly. For allocation domain, 3 selected studies had unclear risk as they did not clearly mention the methods to conceal allocation. For blinding participant and researchers all 3 selected studies had high risk as it was not performed in the studies. For blinding of outcome assessment domain, 3 selected studies had low risk as they assessed outcome by masked examiner. For Incomplete outcome data domain and selective reporting domain, 3 studies had low risk as they satisfy this domain. For other bias domain, 3 selected studies had high risk due to the low number of participants.

We did not find large-scale randomized trials which are required to get more reliable results. It may include diverse populations to see whether it is genuinely effective in diverse populations or not. While assessing the quality of the included study, comparison points were low for each study. Therefore, more study should be done with direct comparisons with other conventional therapy procedures. In addition, we did not find where therapy was provided on normal binocular people. It will also be interesting to include normal binocular people and provide VR-based therapy and see the outcome. Comparison with the normal group and CI group after providing VR-based therapy will provide more reliable data. We did not find the long-term effects of the therapy. It will be interesting if future study includes long-term follow-up to observe how VR-based therapy performs in the long-term. Future studies should include follow-ups at intervals of 6 months, 1 year, and beyond to assess whether the intervention's effectiveness is sustained over time. It may evaluate whether effectiveness is sustained over time or not, and if not, how it changes over time. Future studies should incorporate advanced technologies, such as functional MRI, to obtain objective metrics of therapeutic efficacy, thereby complementing subjective assessments. In future, studies should perform cost-effective analyses to evaluate the feasibility of integrating VR therapy into clinical practice, especially in economically challenged areas. In future studies, a similar amount of therapy should be provided to each therapy group to obtain actual comparison results. Future studies may also increase the quality of the research and make the VR-based therapy more reliable.