A convergent, parallel, mixed method study was used to gain an in-depth understanding of feasibility and acceptability of applying immersive VR on our selected population [26]. Quantitative and qualitative data were conducted concurrently in a 12-week, 2-arm, pilot randomized controlled trial design (CHiCTR2100046313). This study was conducted in Hong Kong from June 1, 2021, to March 18, 2022, in 3 separate batches. All collected quantitative and qualitative data were equally weighted, independently analyzed, and then integrated to generate results [26].

This study complied with the Declaration of Helsinki, and ethical approval was obtained from the Joint Chinese University of Hong Kong–New Territories East Cluster Clinical Research Ethics Committee (2021.052). Written informed consent was obtained from all participants. All data were deidentified and kept confidential and were disposed on completion of this study.

Participants were recruited from an existing community project titled “CUHK-Jockey Club Pain Relief Project for Seniors,” a charity program that offered nonpharmacological pain management to older adults with chronic musculoskeletal pain [27]. The eligibility criteria were screened by a trained research assistant and confirmed by a primary care physician. The inclusion criteria were a diagnosis of knee osteoarthritis based on clinical and radiographic criteria, as defined by the American Rheumatology College; moderate to severe knee pain for at least 3 months; pain intensity score ≥4 on a numeric rating scale (NRS) of 10; and use of a smartphone [28]. The exclusion criteria were decompensated organic and psychiatric disease; contraindications to VR therapy due to history of epilepsy or severe myopia (>−3.5 diopters); and comorbidities that may impede active participation in this study. Participants were recruited in 3 separate batches based on the availability of lower limb sensors.

All participants attended a 30-minute health talk led by a registered physiotherapist to explain and demonstrate the home-based exercises. Further, 2 sets of lower limbs strengthening exercise were selected from the Ottawa panel clinical practice guideline and validated by a physiotherapist and a primary care physician [29]. This included repeated knee-extension exercise for concentric quadriceps training and squeezing a fitness ring between the thighs for vastus medialis oblique isometric training, both to be performed in a sitting position.

Participants in the VRiKnee group were instructed to perform lower limb exercises for 12 weeks using an immersive VR platform developed by our team. A smartphone app (iOS or Android) which captured outdoor garden scenes was used to create an immersive environment and participants were immersed using a head-mounted display (HMD) device (VR Shinecon 5.0). The app provided a virtual environment with ambient audio, an amateur as a coach, and interactive animations with visual feedback that encourage practice, that is, virtual flowers that would blossom with successful exercise moves and a timer for unlocking the 3 levels of difficulty. The app also recorded participants’ accumulated moves, total exercise time, and game level in a virtual scoreboard (Multimedia Appendix 1).

Participants in the control group were instructed to perform the same set of exercises as guided by paper-based education pamphlets. Both groups were advised to perform the exercise 5 days per week, with an expected duration of 30 min/d for 12 weeks. An exercise diary was given to the control group to record exercise participation.

We applied the stepped rule of thumb in this pilot sample size calculation [30]. With a proposed future main trial design of 90% power and 2-sided 5% significance, a total of 30 participants (15 at each arm) would be able to detect an assumed effect size of 0.5.

An off-site statistician performed 1:1 randomization using Random Allocation Software (version 1.0) to allocate and control groups. The allocation sequence was concealed from investigators and participants using sequentially numbered, opaque, and sealed envelopes. The corresponding envelopes were opened at the time of intervention assignment after all the enrolled participants had undergone all baseline assessments. It was not possible to blind participants and research assistants implementing interventions in this open-label study; however, data collectors and outcome adjudicators were blinded to the allocation status.

The outcome measures were collected through face-to-face interviews at baseline, 6 weeks, and 12 weeks. Demographic data such as age, gender, the number, and type of comorbid illnesses were recorded.

A concurrent approach was used to understand the acceptability toward VRiKnee. All participants were recruited to focus group interviews at 12-week using convenience sampling and qualitative data was collected until theoretical saturation [42]. A licensed counseling assistant (HHKL) with experience in qualitative interviews and a registered nurse (HHML) held the semistructured interviews with open-ended questions (Multimedia Appendix 2) to assess user experience and technology acceptance [43,44]. To engage participants in a dialogue, follow-up questions were asked according to individual responses. Since this study was conducted during the COVID-19 pandemic, both face-to-face and Zoom (Zoom Video Communications, Qumu Corporation) videoconferencing were used for focus group interviews [45].

Quantitative data were statistically analyzed with IBM SPSS Statistics software (version 28.0.1.1, IBM Corp). Between-group differences were compared with the Mann-Whitney test for continuous variables and the Fisher exact test for categorical variables [46]. Prior to the intention-to-treat (ITT) analysis, missing values were imputed using a Markov chain Monte Carlo model, in which 10 completed data sets were imputed under the assumption that data were missing at random [47-49]. Imputed dependent variables included the NRS, WOMAC, EQ-5D, and MET; imputed independent variables or covariates included gender, number of chronic diseases, and baseline MET. The parameters were combined according to the Rubin rule, and a linear mixed model was used to analyze results. Analysis of covariance was used to assess the intervention effects on secondary outcomes at week 6 and 12 following the ITT analysis, adjusting baseline measurements at randomization [50].

Qualitative data were collected and analyzed with our quantitative data. Each semistructured interview was transcribed verbatim and analyzed using thematic analysis [51]. Further, 2 researchers (HHML and HHKL) read the transcripts line by line to familiarize with the data and initial codes were then formulated. The coding structure was then validated by 2 authors (MN and RWSS) independently. Codes sharing similar meanings were consolidated and organized into potential themes. Thematic maps were generated for review of themes in relation to the coded extract [52]. Similar codes were grouped into categories and themes and were discussed within the research team until consensus was met. We initiated coding and analysis of data early after batch one to identify possible patterns and themes, and qualitative data was collected until theoretical saturation was reached [53]. The researcher reflexivity of this study was enhanced with several strategies [54]. First, the interviews were led by HHKL, a counseling assistant with experience in qualitative interviews; HHML is a registered nurse with experience in older adult care, who participated in observation note taking during the interviews. Second, a reflexive journal was kept throughout the research process to increase transparency. To minimize individual biases, reflexive team discussions were carried out among the research team throughout the interpretation phase. The work was supported by RWSS, a family medicine physician with clinical and research experience, and MN who is an expert on mixed method studies.

Finally, data-mixing occurred when both quantitative and qualitative data were analyzed using side-by-side comparisons and joint displays [55].

In total, 30 participants with a median age of 63.5 (IQR 61.8‐66.3) years, 76% (n=23) of whom were female, were recruited. No statistical differences were found between the groups in terms of baseline measurements of variables (all P values were >.05). Participants’ baseline characteristics are summarized in Table 1.

Recruitment was conducted between May 27, 2022, to November 3, 2022, in 3 separated batches; the total recruitment time was 7.57 weeks. A total of 100 registrations were received; 29 did not have knee osteoarthritis, 30 did not have smartphones that operate on Android or iOS, and 11 had comorbidities that precluded their participation. The remaining 30 eligible participants were all enrolled to this study and successfully randomized, with a recruitment rate of 100%. The response rates were 80% and 93% for the VR group and control group, respectively. The median exercise adherence for VR and control groups was 78.89% (IQR 47.75%-142%) and 68.75% (IQR 38.57%-188%) during the initial 6 weeks, respectively; both groups had reduced exercise adherence, with rates decreasing to 77.22% (IQR 36.78%-104%) for the VR group and 62.08% (IQR 40.43%-166%) for the control group over the 12 weeks. The adherence rates in terms of bout time are shown in Table 2.

In total, 2 participants dropped out in the VRiKnee group and 1 in the control group, with a dropout rate of 13.3% and 6.7%, respectively. Reasons for dropout included hearing impairment (n=1) and worries for personal health (n=1) in the VR group and emotional distress (n=1) in the control group. Adverse events included 5 (33.3%) participants reporting cybersickness in the VR group. This study workflow is summarized in Figure 1.

At 12 weeks, no statistically significant between-group differences were detected for all secondary outcomes in this small pilot (Table 3). Yet, we detected a trend favoring the use of VRiKnee versus control in reducing the NRS (Cohen d=−0.084, P=.64) and WOMAC pain subscale (Cohen d=−0.089, P=.62) and improving the WOMAC stiffness subscale (Cohen d=−0.190, P=.29).

Quantitative and qualitative findings are integrated and displayed in Tables 5-8.

In summary, VRiKnee was well-received by participants and provided positive user experience. The qualitative analysis revealed that VRiKnee showed promising potential for managing knee osteoarthritis symptoms and improving exercise adherence, which could explain the positive trends observed in some of the quantitative outcomes. However, the quantitative results indicated a decline in exercise adherence over time. The qualitative findings highlighted various challenges faced by the participants during the VR intervention, emphasizing the need to address these issues in order to enhance the overall user experience and foster better adherence to VR-based interventions in future applications.

Recently, a study conducted by Özlü et al [24] investigated the impact of disease-specific gamification through HMD on pain, physical function, and balance in participants with knee osteoarthritis. While this study population consisted of younger participants, with a mean age of 53 (SD 10.19) years, the researchers documented similar adverse effects such as cybersickness, nausea, and headache. Consequently, if immersive VR technology is to be employed in the management of knee osteoarthritis or other chronic diseases, the level of immersion must be carefully considered [56,57]. Interestingly, some participants suggested that using 2D screens may offer a more comfortable alternative, thereby prompting the exploration of the cave automatic virtual environment. This technology entails a cube-shaped room where screens project computer-generated images onto the walls, floor, and ceiling, allowing users to interact with the virtual environment via handheld controllers or trackers. Previous application of the cave automatic virtual environment system in training patients with Parkinson disease has yielded promising preliminary results [58].

Furthermore, Özlü et al [24] also observed that the beneficial effects of VR intervention diminished over time. Therefore, to sustain the positive effects of VR interventions, ensuring compliance is of utmost importance. Overcoming several obstacles during the design phase can facilitate this process. For instance, regular content updates can maintain player engagement and interest by introducing new levels, characters, game modes, or features that expand the gameplay experience. Additionally, community engagement is essential in fostering a strong player community. This can be achieved through encouraging player interaction, sharing of experiences, and providing feedback via online platforms such as forums and social media, as well as through in-game features such as leaderboards and multiplayer modes. Finally, continuous improvements based on player feedback will play a critical role in enhancing the intervention.

This study’s strengths lie in its use of a mixed methods design, which contributes to a comprehensive understanding of VRiKnee, enhances validity through triangulation, provides valuable contextual insights, and generates practical implications on the design of future VR interventions. Further, 1 limitation of this study is the small sample size, which may restrict the trial’s ability to detect meaningful effect sizes.

The feasibility and acceptability of VRiKnee in managing knee osteoarthritis have been demonstrated, suggesting its potential clinical efficacy. However, further confirmation through larger scale trials is necessary. VRiKnee has shown promise in enhancing exercise adherence, although a decline was observed over time. Participants faced challenges during the VR intervention, underscoring the need to address these barriers to improve adherence in future implementations.