We recruited participants through convenience sampling from a youth service organization and our secondary school networks in Hong Kong using promotional leaflets and social workers’ and teachers’ referrals. We contacted individuals who expressed interest in the study and invited them to a 10-minute in-person interview to assess eligibility. Those who met the inclusion criteria were provided with detailed information about the study and completed written informed consent, as outlined in the Ethical Considerations section above.

We assessed feasibility and acceptability using participant attendance, retention, homework practice frequency, dropouts, and adverse events (eg, emotional discomfort and distress). Homework data came from participants’ homework diaries, and all other data were systematically documented by the research team in research logs.

We assessed preliminary effectiveness on anxiety, depression, and stress using both subjective and objective measures.

We used a questionnaire comprising the Chinese versions of the Depression Anxiety Stress Scale–21 (DASS-21) and the Mindful Attention Awareness Scale (MAAS). The adult versions were selected because they are more widely used in the MBI literature, thereby facilitating cross-study comparability, and have demonstrated sound psychometric properties in Chinese adolescent samples [29-31].

The DASS-21 [32,33] includes 21 items rated on a 4-point scale to assess depression, anxiety, and stress over the past week. It produces 3 subscale scores on anxiety, depression, and stress (0‐21) and a total distress score (0‐63), with higher scores indicating greater symptom severity. In this study, the DASS-21 demonstrated excellent internal consistency at baseline (α=.94) and good internal consistency at follow-up (α=.90). Subscale reliabilities at baseline and follow-up were as follows: depression (α=.88 and .79), anxiety (α=.82 and .76), and stress (α=.84 and .72).

The MAAS assesses trait mindfulness using 15 items rated on a 6-point scale [34]. Scores represent the mean across items (1-6), with higher scores reflecting greater trait mindfulness. In the study, the MAAS showed high internal consistency at baseline (α=.91) and follow-up (α=.88).

All self-report measures, including the DASS-21 and MAAS, were compiled in a paper questionnaire and administered at baseline (T0, immediately before the first session) and postintervention (T1, within one week after the final session). Questionnaires were administered on-site at the intervention location by a trained research assistant who was independent of the intervention delivery.

Heart rate variability (HRV), an objective indicator of autonomic stress regulation, was collected using the Upmood watch equipped with a 250 Hz photoplethysmogram sensor. HRV data were collected during the after-school period (4-6 PM) at both time points: at baseline (T0), prior to the completion of self-report questionnaires, and at postintervention (T1), following the final intervention session. A single recording window per time point was used to minimize participant burden, consistent with evidence that a single short-term recording yields valid estimates of standard deviation of normal-to-normal intervals (SDNN) and root-mean-square of successive differences (RMSSD) [35]. To ensure data quality and standardized device operation, research assistants completed a 5-hour structured training program prior to data collection, comprising a 2-hour session delivered by the Upmood technical team and an additional 3 hours of supervised practice.

We also administered a preintervention questionnaire (Multimedia Appendix 1) capturing age, gender, family structure, number of coresidents, parental marital status, parental education levels, housing type, monthly family income (Hong Kong Dollars; HKD), and prior mindfulness experience (yes or no).

The questionnaire was administered on-site at the intervention venue prior to the first session by a trained research assistant who was independent of intervention delivery. Participants completed it individually, with the research assistant available to clarify items as needed.

We conducted all statistical analyses in Stata 18 (StataCorp).

We summarized demographic characteristics and feasibility and acceptability outcomes using descriptive statistics (counts, percentages, means, and SDs). Attendance was calculated as the proportion of participants attending each session, with feasibility defined a priori as ≥70% of participants attending ≥80% (8/10) of sessions. Retention was reported as the proportion completing both baseline and postintervention assessments. Homework practice was summarized as the mean number of practices completed. Dropouts and adverse events were reported as counts and percentages.

To examine the effects of the intervention on distress, anxiety, depression, stress, and trait mindfulness, we used paired-sample t tests for normally distributed variables and Wilcoxon signed-rank tests for nonnormally distributed variables (2-tailed, P<.05). All assumptions relevant to the statistical tests were checked.

For HRV analysis, we used the HRV analysis with the R package and followed the recommended procedures described [36]. Pulse-to-pulse intervals were extracted from a 10-minute recording window. We removed artifacts using established threshold criteria (<300 ms or >2000 ms) [37] and resampled the time series at 4 Hz using a cubic spline [38]. We then calculated standard time-domain HRV metrics, including the SDNN and the RMSSD. We focused exclusively on time-domain measures, as time-domain indices are more robust to noise and more reliable for short recordings obtained from wearable devices [39,40].

SDNN reflects overall HRV, with higher values indicating greater overall autonomic flexibility and better stress regulation, whereas RMSSD reflects short-term HRV and parasympathetic activity, with higher values indicating more adaptive stress regulation.

Pre-post changes in HRV indices (RMSSD and SDNN) were analyzed using paired t tests or Wilcoxon tests, as appropriate (2-tailed, P<.05).

A sensitivity analysis was also conducted to examine whether prior mindfulness experience influenced intervention outcomes, including self-reported outcomes (distress, anxiety, depression, stress, and trait mindfulness) and HRV indices. Participants were divided into 2 groups: those with prior mindfulness experience and those without, and a 2×2 mixed ANOVA was performed to examine time × mindfulness experience interaction effects across all outcome measures (2-tailed, P<.05). Feasibility outcomes were additionally compared descriptively between the 2 groups, including the proportion of participants meeting the ≥80% attendance threshold and mean homework practice frequency.

A research assistant transcribed all group interviews verbatim. A research team member (CTKY) who has training in qualitative methods analyzed all transcripts using NVivo software (Lumivero) and Braun and Clarke’s reflexive thematic analysis [41,42] with an inductive approach. CTKY followed the 6 phases of reflexive thematic analysis: familiarization with the data, generation of initial codes, searching for themes, iterative reviewing and refinement of themes through discussion with other project team members (PPKK and APLT), defining and naming themes, and producing the report.

To enhance reflexivity, CTKY maintained analytic memos throughout the analysis and discussed emerging interpretations with the project team. These discussions helped examine assumptions and strengthen the credibility and rigor of the analysis.

We integrated quantitative and qualitative data at the interpretation stage, following a concurrent approach [43], in which we simultaneously explored the statistical patterns and in-depth meanings, with equal priority given to both data strands. We discussed findings from both sources using data triangulation, exploring convergence and divergence to compare and synthesize results, with integrated interpretations and a joint display table presented in the Results and Discussion section [44].

The study achieved a 100% retention rate, with all 42 participants completing both the intervention phase, pre- and postintervention self-report questionnaire assessments, and wearable device measurements at both time points. Although all 42 participants completed the wearable device measurements at T0 and T1, valid HRV recordings were obtained from 33 of 42 (78.6%) participants, as 9 participants had unrecoverable recordings due to technical failure (ie, Bluetooth disconnection) of the wearable devices.

Attendance was high, with 31 of 42 (74%) participants attending at least 80% (8/10) of the sessions. Participants also demonstrated engagement outside the sessions, with a mean homework practice count of 11 (SD 8) times within the 8-week period. No dropouts or adverse events were reported during the study period.

Table 3 presents pre- to postintervention changes in DASS-21 subscale scores (depression, anxiety, and stress), total distress, self-reported trait mindfulness (MAAS), and HRV indices.

Results indicated no significant changes in self-reported psychological outcomes following the intervention. Total emotional distress remained stable from pre- to posttest (mean difference 0.88, 95% CI −5.15 to 3.39; P=.46). Similarly, no significant changes were observed in depression (mean difference 0.14, 95% CI −1.57 to 1.28; P=.56), anxiety (mean difference 0.81, 95% CI −2.21 to 0.59; P=.07), or stress (mean difference −0.07, 95% CI −1.72 to 1.86; P=.94). Self-reported trait mindfulness also did not change significantly (mean difference −2.02, 95% CI −0.65 to 4.70; P=.17). Full descriptive statistics and test statistics are presented in Table 3.

In contrast, analyses of HRV indices revealed significant physiological changes following the intervention. Results showed a significant increase in SDNN from baseline to postintervention (mean difference 17.60 ms, 95% CI −33.88 to −1.32; t₃₂=2.202; P=.04; Cohen d=.383). The RMSSD also increased significantly, from baseline to postintervention (mean difference 20.20 ms, 95% CI −38.76 to −1.65; t₃₂=2.218; P=.03; Cohen d=.387). These findings may suggest enhanced parasympathetic activity and improved physiological stress regulation following the intervention.

To examine whether prior mindfulness experience influenced intervention outcomes, a sensitivity analysis was conducted by dividing participants into those with (27/42, 64.3%) and without (15/42, 35.7%) prior mindfulness experience.

A series of 2×2 mixed ANOVAs was conducted with time (pre- vs postintervention, within-subjects) and mindfulness experience (between-subjects) as factors across all outcomes (DASS-21 total and subscale scores, MAAS, and HRV indices). No significant time×mindfulness experience interaction effects were observed across any outcome, suggesting that prior mindfulness experience did not meaningfully affect pre-to-post changes, lending confidence that the observed outcomes were not attributable to participants’ prior familiarity with mindfulness practice. Detailed results are presented in Multimedia Appendix 2.

For feasibility outcomes, attendance rates were comparable between the two groups, with 87% (13/15) of participants without prior experience and 67% (18/27) of participants with prior experience meeting the ≥80% attendance threshold. Homework practice frequency was similarly comparable across groups (without prior experience: mean 13.27, SD 7.38; with prior experience: mean 10.19, SD 7.71).

Participants described mindfulness as a practical strategy for regulating negative emotions, particularly stress and anger, by creating a pause that allowed them to respond more calmly in challenging situations.

Several participants emphasized how mindfulness helped them manage interpersonal conflicts by reducing impulsive reactions. One participant reflected on family arguments, noting a shift from reacting defensively to listening and letting go:

The same participant described applying mindfulness when feeling irritated while teaching her younger sister, using the practice to calm herself before responding:

This capacity to pause and reflect before acting was also evident in school settings. One participant involved in a school committee noted that mindfulness helped prevent impulsive decisions during conflict:

Others described reduced anger and aggressive urges in everyday situations, such as gaming frustration:

and

Participants described mindfulness as enhancing their awareness of emotions, thoughts, and bodily sensations, which in turn supported greater self-understanding and emotional clarity. One participant contrasted mindfulness with earlier stress-management approaches taught in school:

Others reflected on how this increased awareness reduced feelings of confusion and pessimism about their daily lives. One participant described gaining a clearer sense of self and a more positive outlook over the course of the program:

Mindfulness practices also enhanced awareness of bodily sensations, helping participants recognize physical discomfort and needs more clearly. One participant highlighted the impact of body scan exercises:

Participants frequently reported improved focus following mindfulness practice, describing feeling more attentive and less distracted during and after the sessions. Some expressed this briefly:

and

Others highlighted that mindfulness created a mental space that allowed them to disengage from academic pressures:

In addition to improved focus, many participants reported better sleep quality after learning mindfulness techniques. Several described intentionally practicing mindfulness before bedtime to help them fall asleep, including one participant who shared:

And another who noted:

Participants identified the immersive CAVE environment as a key facilitator of engagement in mindfulness practice. The CAVE was frequently described as a distraction-free space that supported sustained attention. The simulated projected nature scenes, together with the sound, made the experience more immersive.

Several participants contrasted practicing mindfulness in the CAVE with doing so at home or at school, noting that the quiet environment made it easier to concentrate:

Others similarly noted:

Another participant compared phone-based mindfulness practice with the CAVE, noting that:

In addition to the physical environment of the CAVE, participants emphasized the motivational benefits of practicing mindfulness in a group in the CAVE. Many reported that the presence of others helped them stay engaged:

and

The group setting also fostered social connection and stress relief, with participants describing shared practice as an opportunity to:

and

Participants reported that certain technological aspects of the CAVE occasionally disrupted their mindfulness experience. Several described the on-screen visuals as unnatural or lacking realism, which reduced immersion and distracted them during practice. One participant explained:

Another noted:

Brightness and visual intensity were also reported as barriers. Some participants found the screens uncomfortably glaring, which interfered with relaxation:

and

Some participants described challenges in applying mindfulness when emotions, particularly anger, were intense or long-lasting. In such moments, emotional reactions felt automatic and difficult to regulate. One participant acknowledged reverting to previous coping patterns when emotions became overwhelming:

Others noted that while mindfulness helped reduce distress in the moment, its effects were sometimes temporary. One participant reflected:

Participants also described practical barriers to integrating mindfulness into daily routines, particularly for practices such as mindful eating. Time constraints were frequently cited, with participants feeling that mindfulness practices were difficult to sustain amid busy schedules. One participant explained simply:

while another elaborated:

Family expectations further constrained practice, especially when mindfulness behaviors conflicted with household routines. One participant noted:

To the best of our knowledge, this study represents the first local evaluation of an immersive virtual reality MBI for adolescents in Hong Kong. Our findings indicate high feasibility and acceptability of VR-MBI for adolescents, which aligns with prior research conducted in other countries. For example, the “Relaxing Environment for Stress in Teens” (RESeT) program in the United States reported strong engagement and good usability, along with high acceptability, feasibility, and appropriateness ratings among adolescents. Similarly, adolescents in the RESeT program also highlighted the immersive VR environment as a key factor supporting engagement and temporary disengagement from distressing thoughts and emotions [24].

Our VR-MBI appears to demonstrate better feasibility and acceptability than traditional mindfulness interventions previously evaluated in Hong Kong. A local Hong Kong study reported moderate acceptability, with only about half of participants rating the program positively and very low home practice rates [12]. Similarly, a school-based MBI local program observed high dropout rates (41%), low attendance (50%), behavioral issues during sessions, limited engagement, and minimal home practice [46]. The immersive CAVE environment may have been an important factor underlying this discrepancy. VR has been empirically demonstrated to enhance immersion and sustained attention during practice [17-19], and our qualitative findings corroborate this—participants consistently identified the CAVE environment as a key facilitator of engagement, describing it as a comfortable, distraction-free space that was qualitatively distinct from other mindfulness delivery modes such as home-based practice, smartphone apps, and classroom-based delivery. That said, we acknowledge that other contextual and design factors likely also contributed. The challenges reported in prior local studies, including large class sizes, compulsory participation, restrictive classroom environments, and monotonous activities [46], were largely addressed through deliberate program design choices in this study, including small group sizes, interactive and creative activities, and flexible scheduling arranged collaboratively with participants. The improved feasibility and acceptability observed in this study, therefore, likely reflects the combined contribution of the immersive VR environment and these broader design considerations.

Despite these strengths, our study identified several VR technologies–related challenges that warrant further refinement to enhance feasibility and acceptability. Although the CAVE was deliberately designed as a headset-free VR system to minimize physical discomfort such as dizziness, participants still reported limitations related to the visual design and display characteristics of the system. Specifically, some noted that the visuals appeared unnatural and that the screens were overly bright, which reduced immersion and interfered with relaxation during mindfulness practice. Similar issues have been reported in other VR-based mindfulness studies, where sensory discomfort and image quality concerns have emerged when VR environments are not optimally designed [47,48]. Our participants also reported that they faced challenges in using mindfulness when emotions were very intense, noting that the practice offered limited benefit in such moments. This aligns with prior research highlighting similar challenges [49,50], and it is suggested that integrating complementary strategies (eg, cognitive reappraisal, grounding) or offering brief “mini practices” may provide additional support. Our participants also described difficulty sustaining certain practices outside sessions, such as mindful eating, due to time constraints and limited family support. Similar barriers have been reported among adolescents more broadly, who often struggle to maintain regular practice without external support [12]. To address this, prior studies recommend embedding short practices into daily routines and involving family members in the intervention to help sustain motivation and continuity [28,51]. Addressing these challenges in future iterations is important for optimizing the feasibility and acceptability of VR-MBI for adolescents.

Preliminary improvements in physiological stress regulation (SDNN and RMSSD) were observed post intervention. These positive changes were complemented by qualitative findings, in which participants described reduced stress, increased calmness, and improved emotion regulation. This finding is consistent with a review reporting improvement in SDNN and RMSSD following 8 weeks of mindfulness training [40], as well as with similar findings observed in MBIs’ randomized controlled trials targeting adolescents [52]. One proposed mechanism underlying these HRV improvements is that mindfulness practice often involves slow, attentive breathing, which promotes parasympathetic (vagal) activation, enhances physiological relaxation, and improves the body’s capacity to regulate stress [53].

However, contrary to expectations, the intervention did not result in improvement in self-reported anxiety, depression, stress, or trait mindfulness. One plausible explanation is a potential floor effect [54]. Our post hoc examination indicated that only 38% (16/42) of participants reported clinically elevated anxiety on the DASS-21 anxiety subscale at baseline (mean: 9.19, SD: 3.29), whereas the remaining 62% (26/42) scored below the clinical threshold (mean: 2.23, SD: 1.61), indicating limited room for measurable improvement at the group level. This interpretation is further supported by qualitative accounts in which participants described applying mindfulness as an active strategy to manage acute stress responses in emotionally challenging situations (theme 1), consistent with the notion that those with greater baseline distress may derive more measurable benefit, while those with lower baseline scores have less capacity for detectable change. We acknowledge, however, that this remains a post hoc interpretation and should be examined more rigorously in future studies with a larger sample size stratified by baseline anxiety severity.

The mixed findings, where the VR-MBI was effective in improving objective HRV indices but not subjective self-report measures, may be plausibly explained by differences in the dimensions of psychological functioning captured by these outcomes. HRV indices are sensitive to short-term changes in autonomic regulation and may therefore detect early physiological effects of mindfulness practice [39,55]. In contrast, the DASS-21 and MAAS primarily assess broader and more stable mood states and trait-like characteristics, which may require longer intervention exposure or sustained practice before significant changes become detectable. This interpretation aligns with prior evidence. A meta-analysis of MBIs in children and adolescents reported a trend toward reductions in self-reported anxiety symptoms, but effect sizes were small and nonsignificant [56]. Mechanistic accounts further propose that early gains from mindfulness are most evident in domains of emotion regulation and stress reactivity, while more stable improvements in depression and anxiety typically require longer duration or greater intensity of practice [57]. From this perspective, our 8-week VR-MBI may have facilitated short-term physiological and subjective improvements in well-being, while more enduring shifts in mood states would likely require extended or repeated engagement with practice. However, this interpretation requires further investigation in future studies.

Taken together, these findings suggest that our participants may have experienced meaningful early benefits, evident in physiological regulation and qualitative reports, that were not yet sufficiently robust to be reflected in self-reported mood measures. Future programs may consider extended duration or booster sessions. Our study also highlighted the importance of using a mixed methods design by integrating quantitative self-report measures, objective physiological biomarkers of stress regulation, and qualitative findings. Future studies should adopt similar approaches, as this can provide a nuanced understanding of efficacy, including capturing changes not fully captured by subjective measures or objective measures alone.