This study is a systematic literature review and meta- analysis of RCTs that synthesized qualitative and quantitative data to confirm the effects of virtual reality programs utilized for patients with diabetes. The research plan for this study was registered in PROSPERO (https://www.crd.york.ac.uk/PROSPERO/) and the registration number (CRD42021236581) was obtained on March 12, 2021. Initially, the outcome variable was only HbA1c; however, upon reviewing the literature and discovering that virtual reality had a significant effect on physical function [A1,A2,A5-A7], the study was modified and re-approved by PROSPERO on July 5, 2022.

Specific questions for the systematic literature review, pertaining to PICOS (Participants, intervention, compar-ison, outcome, and study design) are as follows. Participants of these studies included patients with prediabetes, type 1 or type 2 diabetes, or gestational diabetes. The experimental intervention was a virtual reality program applied to patients with diabetes; hence, all studies using immersive virtual reality with a head-mounted display or non-immersive virtual reality with Nintendo Wii, EyeToy, Xbox, Kinect, and PlayStation were included. Comparative interventions were those that provide only routine care such as exercise and education, and periodic hospital visits that did not involve virtual reality programs. HbA1c level, static and dynamic balance, and falls efficacy were examined to validate the effect of the virtual reality programs. Study design was RCTs. The exclusion criteria were as follows: (1) studies on animals or preclinical trials; (2) studies including patients without diabetes; (3) studies whose original text could be verified; (4) studies not published in Korean, English, or Chinese; (5) non-experiment-al studies such as reviews or observations; and (6) un-published theses.

For literature search and analysis, studies up to May 2021 were investigated. English databases such as Pub-Med, EMBASE, CINAHL, Cochrane Central Register of Controlled Trials (CENTRAL), and MEDLINE, Chinese databases such as China National Knowledge Infrastructure (CNKI), Wanfang, and VIP Information, and Korean databases such as DBpia, Korean Studies Information Ser-vice System (KISS), Research Information Sharing System (RISS), and Science on were used to search for published articles or theses. In order to increase the sensitivity of literature search, gray literature such as dissertations, news, and presentation materials were manually searched. In Korean databases, search terms including ‘diabetes (당뇨),’ ‘gestational diabetes (임신성 당뇨),’ ‘prediabetes (전단계 당뇨),’ ‘type 1 diabetes (1형 당뇨),’ ‘type 2 diabetes (2형 당뇨),’ ‘virtual reality (가상현실),’ ‘Wii,’ ‘EyeToy,’ ‘Xbox,’ ‘Kinect,’ ‘PlayStation,’ and ‘Nintendo’ were used (Appendix 1-A). For an efficient search in English databases, MeSH terminology, Boolean operator, and truncation were used. MeSH and text words used in the combi-nation of P (patient) AND I (intervention) AND O (outcome) were as follows: search terms for P (patient) included ‘DM,’ ‘Diabetic,’ ‘Diabetes,’ ‘GDM,’ ‘Type 1 DM,’ ‘Type 2 DM,’ ‘Prediabetic,’ ‘Diabetes, Gestational [MeSH],’ ‘Diabetes Mellitus [MeSH],’ ‘Diabetes Mellitus, Type 1[MeSH],’ ‘Diabetes Mellitus, Type 2[MeSH],’ and ‘Prediabetic State [MeSH]’; search terms for I (intervention) included control or text words such as ‘Virtual Reality [MeSH],’ ‘Wii,’ ‘EyeToy,’ ‘Xbox,’ ‘Kinect,’ ‘PlayStation,’ and ‘Nintendo.’ No specific restrictions were set for search terms for O (outcome) (Appendix 1-B). In Chinese databases, search terms such as ‘Diabetes (糖尿病),’ ‘Diabetes* (糖尿*),’ ‘Virtual Reality (虚拟现实),’ ‘Virtual Reality (假想现实),’ ‘Remote Reality (远程现实),’ ‘Human- Computer Interaction (人机互动),’ ‘Videogame (视频游戏),’ ‘Somatosensory Game (体感游戏),’ ‘Wii,’ ‘EyeToy,’ ‘Xbox,’ ‘Kinect,’ ‘PlayStation,’ and ‘Nintendo’ were used (Appendix 1-C). For the selected literature, the search results were deduplicated. The titles and abstracts were in-dependently reviewed by three researchers: a professor of nursing, a doctor of nursing, and a doctor of Chinese medicine, according to the PICOS and exclusion criteria, for primary selection of studies. Then, secondary selection involved reviewing the full texts of the remaining studies to identify studies to be included based on the inclusion and exclusion criteria. The Chinese literature search and trans-lation were assisted by a Chinese doctoral student in oriental medicine. Any disagreement between the authors pertaining to the selection process was resolved through sufficient discussion, and the opinion of a third author, from the field of oriental medicine, was considered.

To evaluate the quality of literature used in this study, Risk of Bias the Cochrane risk-of-bias tool (RoB) was used [27], and the analysis was conducted using the Cochrane RevMan Web.

After analyzing the characteristics of studies included in the systematic literature review, data such as First Author, Country, Participants, Age, Sample size, Intervention group, Control group, Main outcome measures, Results, and Author's conclusion were extracted. In a meta-analysis, the random effects model was used, since there was a substantial difference between the selected studies [28]. The effect size was analyzed by a standard-ized mean difference (SMD), mean difference (MD) and Cohen's d. The effect of the outcome variable in this study and the 95% confidence interval (CI) were analyzed using inverse-variance. The effect size was classified as small (effect size of 0.2 or less), medium (effect size of 0.5), and large (effect size of 0.8 or more) [29]. In order to confirm heterogeneity, the forest plot was checked, and the heterogeneity between studies was analyzed using Higgins’ I² statistic. The I² value of ≤25.0%, 25.0~75.0%, and >75.0% was assessed as ‘low,’ ‘medium,’ and ‘high’ ‘heterogeneity,’ respectively [30].

The Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) were used to describe the status of the step-by-step literature selection process of this study in an easy-to-read and detailed manner [31]. In the database, 1,243 articles were searched. After excluding 171 duplicates, 1044 studies were excluded according to the PICOS and exclusion criteria. After the full texts of articles were reviewed, eight studies were finally selected. Qualitative analysis was performed on the eight selected studies, and the results of six of these studies were included in the meta-analysis (Figure 1).

All eight studies were journal articles and were published in 2010 or later. Three studies were conducted in the United States [A5-A7], four studies were conducted in Asia [A1,A2,A3,A8], two studies in South Korea [A1,A2], two studies in China [A3,A8], and one study was conducted in Germany [A4]. In terms of the study Participants, there were seven studies on type 2 diabetes [A1-A6, A8] and one study on prediabetes [A7]. Two of the studies on type 2 diabetes involved participants who were patients with a complication of diabetic neuropathy [A5]. In terms of age, six studies included participants aged ≥60 years [A1,A2,A4-A6,A8], and two studies included participants aged 34-55 years [A3,A7]. Regarding the type of virtual reality used, none of the studies used immersive virtual reality programs using head-mounted displays. All eight studies were non-immersive virtual reality program studies using 3D stereoscopic images, gaming consoles, motion capture cameras, and body-tracking sensors. Regarding the methods for executing the virtual reality programs, two studies used PlayStation and EyeToy [A1,A2], three studies used Nintendo Wii [A4,A6,A7], one study used sensor-based interactive exercise training [A5], and two studies used Kinect [A3,A8]. In terms of years and countries where certain virtual reality programs were used, PlayStation and EyeToy were used in South Korea in 2012 [A1] and 2013 [A2]. Nintendo Wii was used in Germany in 2013 [A4] and in the US in 2018 [A6] and 2019 [A7]. Sensor-based interactive exercise training was used in the US in 2015 [A5], and Kinetic was used in China in 2015 [A3] and 2020 [A8]. The application time for the virtual reality program was 30 min in two studies [A3,A4]; 35 min, five times a week in one study [A8]; 40 min, three times a week in one study [A6]; 45 min, twice a week in one study [A5]; and 50 min, twice a week in two studies [A1,A2]; and 50 min, three times a week in one study [A7]. The duration of the virtual reality intervention was 15 days in one study [A3]; four weeks in one study [A5]; six weeks in one study [A8]; 10 weeks in two studies [A1,A2]; and 12 weeks in three studies [A4,A6,A7]. The minimum number of samples in each experimental and control group ranged from 14~93, and the total number of samples ranged from 30~176. The control group received routine treatment in two studies [A4,A5]; diabetes education in two studies [A1,A2]; routine treatment and diabetes education in one study [A3]; routine treatment in one group and general exercise in one group, with two control groups in one study [A7]; general exercise in one study [A6]; and rehabilitation treatment and balance training in one study [A8]. The main outcome variables of this review were HbA1c, static body balance, dynamic body balance, and falls efficacy. Among the eight studies included in the qualitative analysis, three studies measured HbA1c levels as an outcome variable [A3,A4,A7], five studies measured static body balance [A1,A2,A5,A6,A8], two studies measured dynamic body balance [A2,A8], and two studies measured falls efficacy [A2,A5]. Regarding the effect of the application of virtual reality programs for patients with diabetes, HbA1c level was used to measure blood glucose control. Among body balance scales, static balance scales included the Postural Sway and One-Leg Standing tests, and dynamic balance scales included the Berg Balance Scale. The falls efficacy scales included the Falls Efficacy Scale-International (FES-I) and the Modified Falls Efficacy Scale (MFES) (Table 1).

The results after entering the eight selected studies into the RevMan software, according to the quality evaluation criteria are as follows (Figure 2). In terms of bias arising from the randomization process, the RoB was evaluated as ‘low’ in two studies; the allocation sequence was random and concealed until participants were enrolled and as-signed to interventions. Six studies were evaluated as ‘un-clear,’ since there was no mention of allocation conceal-ment. In terms of bias due to deviations from intended interventions, the RoB was evaluated as ‘low,’ in two studies while the remaining six studies did not mention blinding of caregivers and people delivering the intervention, and the RoB was evaluated as ‘unclear.’ As a result of analyzing bias due to missing data, one study had a ‘high’ RoB since the dropout rate of 54.0% in experimental group. In the bias in measurement of the outcome, seven studies were evaluated as ‘unclear,’ since there was no mention of blinding of the outcome evaluation. In terms of bias in selection of the reported result, two studies had a ‘high’ RoB since the results were incomplete and could not be used for meta-analysis. In the remaining six studies, the RoB was assessed as ‘unclear’ since the trial registration number was not included.

The effect size of the experimental group with the virtual reality program applied and the control group without the virtual reality program applied was -0.40 (n=236, 95% CI: -1.38~0.58), which was not statistically significant (Z=0.81, p=.42). The heterogeneity between studies was high with Higgins’ I²=98.0% (Figure 3-A).

The methods for measuring falls efficacy in this study included the FES-I and the MFES. Since the two scales op-posed each other, they were adjusted using a minus sign. The effect size of the experimental group with the virtual reality program applied and the control group without it being applied was 0.69 (n=89, 95% CI: 0.23~1.16), which was statistically significant (Z=2.95, p=.003). The heterogeneity between studies was low, with Higgins’ I²=12.0% (Figure 3-D).

The asymmetry of the funnel plot could not be con-firmed. According to Higgins, this test required at least ten studies to be included in the analysis [32].