The participants were enrolled via advertisements at Seoul National University Bundang Hospital and Boramae Medical Center between November 2023 and January 2024. The inclusion criteria were as follows: (1) employed adults aged 19 to 49 years; (2) individuals experiencing moderate-to-high levels of stress, defined as a Perceived Stress Scale (PSS) score of ≥14 at baseline; (3) individuals undergoing pharmacological treatment for psychiatric conditions, such as depression, anxiety disorders, or insomnia, who were eligible if their symptoms were stable and their medication dosages were not expected to change during the clinical trial; and (4) individuals who fully understood the clinical trial protocol and voluntarily consented to participate.

The exclusion criteria were as follows: (1) individuals aged <19 or >49 years; (2) those with dementia, intellectual disabilities, or other cognitive impairments; (3) individuals with seizure disorders, stroke, or other neurological conditions; (4) those with a history or current diagnosis of psychotic disorders, such as schizophrenia or bipolar disorder; (5) individuals with active suicidal ideation; (6) those using medications or having medical conditions (eg, heart or lung disease) that could affect the reliability of HRV testing; (7) individuals who received nonpharmacological psychiatric treatment or counseling within the past 6 months; and (8) individuals with physical limitations preventing them from participating in the exercise or yoga program.

The presence of psychiatric disorders was assessed using the Mini International Neuropsychiatric Interview, a short, structured psychiatric interview designed to identify a broad spectrum of psychiatric disorders as outlined in the DSM-IV (Diagnostic and Statistical Manual of Mental Disorders [Fourth Edition]), and the ICD-10 (International Statistical Classification of Diseases, Tenth Revision). The Korean version of the Mini International Neuropsychiatric Interview has been shown to have good validity and reliability. The interviews, which typically took 15 to 30 minutes to complete, were conducted by 2 qualified psychologists with master’s degrees who were familiar with the tool.

The total sample size was calculated a priori using the G*Power software (version 3.1.9.4; Heinrich-Heine-Universität Düsseldorf). On the basis of previous meta-analyses of digital mental health interventions [17], we anticipated a large effect size (f=0.40). To achieve a statistical power of 0.80 with a significance level (α) of .05 for a 3-group ANOVA, a minimum of 66 participants (22 per group) was required. This provided a conservative power estimate for the multivariate linear model (MLM) used in the primary analysis. After accounting for a potential dropout rate of 10%, 75 participants were recruited.

A fitness platform was developed to enable participants to engage in exercise or yoga programs. To improve motion recognition accuracy, usability testing was conducted in 3 phases: the first phase involved office workers, the second included medical professionals, and the third targeted the general public in home settings. The evaluation focused on key performance indicators, including user interface intuitiveness, functional accessibility, system stability, and latency. The tests identified specific areas for improvement, such as menu complexity, ambiguity in instructional text, and suboptimal user inteUser Interface intuitiveness. Recognizing that these issues could increase user-entry barriers and adversely affect adoption rates, corrective refinements were implemented prior to this clinical trial. Furthermore, the testing phases revealed that environmental variables, such as background interference and camera positioning, affected motion recognition accuracy. Consequently, the recognition algorithm was further optimized, and a comprehensive user manual was developed to ensure consistent performance across diverse environments. To ensure accuracy and quality of information, this study used Redmine, an open-source project management and defect-tracking platform, to oversee the development lifecycle. White box testing methodologies were used for logical defect detection and structural verification. We used a suite of automated testing tools, such as unit, Pytest, and SonarQube, to ensure the precision of the software’s internal logic. The version of the platform evaluated was the initial release (version 1.0, “frozen” during the trial period). No major content or functional changes, bug fixes, or downtimes occurred that might have influenced the study results. Institutional affiliations were not displayed on the digital intervention platforms and were disclosed only during the consent process.

Demographic variables and baseline psychological measurements were analyzed using ANOVA or Kruskal-Wallis tests for continuous variables, whereas chi-square or Fisher exact tests were used for categorical variables to compare the 3 groups.

The primary analysis followed the intention-to-treat principle using an MLM for repeated measures [40,41], accommodating potential missing data points. This model evaluated the effects of group, time, and their interaction across 3 time points: baseline, post-training (4 wk), and follow-up (8 wk). A first-order autoregressive (AR(1)) correlation structure was specified to account for within-subject correlations, and baseline PSS scores were included as covariates to minimize confounding effects.

To control for the family-wise error rate (FWER), a hierarchical testing strategy was adopted for the 2 co-primary outcomes (PSS and PHQ-9). First, the group×time interaction was evaluated at a Bonferroni-corrected significance level of 0.025 (0.05/2 outcomes). Regardless of the interaction significance, prespecified planned pairwise comparisons (yoga vs control, exercise vs control, and yoga vs exercise) were performed at each post-baseline time point using a Bonferroni-adjusted alpha of 0.0083 (0.025 per 3 groups) to test our a priori hypotheses. For secondary outcomes, although treated as exploratory, a Bonferroni-adjusted alpha of 0.0167 (0.05 per 3 groups) was applied for pairwise comparisons to maintain statistical stringency.

In accordance with CONSORT (Consolidated Standards of Reporting Trials) reporting standards, results of group comparisons were reported as mean differences (β), adjusted 95% CIs, and adjusted P values. These CIs were calculated to align with corrected α levels (.0083 and .0167, respectively) to ensure a transparent interpretation of both statistical significance and clinical magnitude. The overall model parameters (eg, F, df₁, and df₂) are reported in the table.

Frontal θ power was defined as the primary physiological outcome of interest. All other EEG frequency bands (α, β, and γ) were treated as secondary or exploratory outcomes. Consequently, no formal adjustment for multiple testing was applied to these exploratory measures, and the results should be interpreted as hypothesis generating.

For the user satisfaction and usability questionnaire administered at the final visit (Multimedia Appendix 1), the four 4-point Likert-scale items were compared between the 2 active intervention groups (exercise vs yoga) using 2-tailed independent samples t tests, with Cohen d as the effect size because these items referred specifically to the assigned exercise or yoga program. The 6 closed-ended items—3 single-response items and 3 multiple-response items in which participants could select up to 2 options—were summarized as frequencies and percentages, and between-group differences in the endorsement of each response option were examined using Fisher exact tests. These analyses were exploratory and were not adjusted for multiple comparisons.

All statistical analyses were performed using SAS (version 9.4).

The study protocol for this RCT was approved by the institutional review boards of Seoul National University Bundang Hospital (approval B-2309-855-301) and Boramae Medical Center in South Korea (approval 30-2023-106), and the trial was registered at ClinicalTrials.gov (NCT06620783). At their initial onsite visit, all participants were provided with comprehensive written and verbal information regarding the study purpose, procedures, and interventions and subsequently provided written informed consent in person prior to participation. Participants were informed of their right to withdraw from the study at any time without penalty, and the original consent also covered secondary analyses without the need for additional consent. All data collected, including self-reported psychological, cognitive, and physiological indices, were handled with strict confidentiality; data were deidentified and stored using unique research IDs. Participants received compensation of KRW 100,000 (US $77) for their time and travel expenses at each of the 3 required evaluation visits (baseline, post-intervention, and 4-week follow-up), amounting to a maximum total compensation of KRW 300,000 (US $231).

During the intervention period and follow-up, no significant harmful or unintended effects were reported in any of the groups (exercise, yoga, and control).

Of the 75 randomized participants (exercise, 24/75, 32%; yoga, 25/75, 33.3%; and control, 26/75, 34.7%), 71 who completed at least nine of the 12 sessions, with a minimum duration of 40 minutes per session, were included in the outcome analysis (exercise, 21/71, 29.6%; yoga, 24/71, 33.8%; and control, 26/71, 36.6%). Figure 3 presents a flow diagram of participant enrollment and retention. Specific reasons for attrition included personal time constraints (2/75, 2.67%), exercise program difficulty (1/75, 1.33%), and inability to establish follow-up contact (1/75, 1.33%). Participants’ average age was 34.14 (SD 6.18) years, and 77.3% (58/75) were women. No significant group differences were observed in demographic characteristics, such as age, sex, marital status, or education level at baseline (all P>.05; Table 1).

Both the exercise and yoga interventions led to improved mental health outcomes over time. The yoga group demonstrated the most pronounced and sustained benefits, while the exercise group showed immediate gains in some cognitive measures.

The group×time interaction for PHQ-9 did not reach the prespecified Bonferroni-corrected significance level of 0.025 (F_4,136=2.71; P=.03). However, planned pairwise comparisons revealed a significant reduction in depressive symptoms for the yoga group compared to the control group at the 4-week follow-up (β=−3.67, 95% CI −5.64 to −1.69; adjusted P<.001). No significant differences were observed between the exercise and control groups at any time point (Figure 4A).

We also analyzed changes in another primary outcome: the PSS. The MLM showed that the group×time interaction was not statistically significant (F_4,136=1.27; P=.29), indicating that the trajectory of stress reduction did not differ significantly between the yoga, exercise, and control groups. However, a significant main effect of time was observed (F_2,136=38.56; P<.001), confirming a general reduction in perceived stress across all participants from baseline to the 4-week follow-up. The main effect of group was not significant (F_2,68=1.13; P=.33). The results, including the F statistics and P values, are summarized in Table 2, while Table 3 presents between-group differences and within-group changes in items from baseline (adjusted 95% CI).

This study aimed to assess whether a 12-session mobile-delivered exercise and yoga programs could effectively reduce stress and depressive symptoms among employees. An RCT demonstrated that both an adaptive moderate-intensity exercise program and a yoga program delivered via a digital platform significantly improved mental health outcomes among employees with moderate-to-high stress compared with a CBT-based self-care control condition. Both interventions decreased depressive symptoms, hyperarousal, and maladaptive CSRS scores. In particular, the yoga intervention showed the most pronounced and sustained benefits across a range of psychological measures. The significant group×time interactions for depressive symptoms, hyperarousal, and CSRS indicated that the trajectories of improvement differed by group, with the yoga group often showing the greatest improvement over time. This is consistent with research on the mental health benefits of mindfulness-based practices, such as yoga, which have been shown to alleviate anxiety and depressive symptoms [10]. A significant main effect of time was observed on frontal θ power; however, no significant group or group×time interaction effects were found. The exercise group also showed improvements in most outcomes, although in some cases, these were not as large or sustained as those in the yoga group. The yoga group demonstrated the greatest and most sustained improvements, whereas the exercise group showed immediate but less sustained gains in secondary cognitive measures. These results suggest differential efficacy patterns among the intervention modalities.

Furthermore, CSRS scores, particularly for extreme negative, aggressive hostile, and self-depreciative thoughts, decreased more prominently in the yoga and exercise groups than in the control group. These findings suggest that physical activity and mind-body interventions can effectively reduce cognitive distortions associated with stress. Such decreases in negative thought patterns may have contributed to the overall improvement in mental health observed in this study.

In addition to psychological outcomes, this study explored brainwave activity, focusing specifically on θ power in the prefrontal cortex, which is commonly associated with cognitive processes, such as attention and memory. We observed a significant decrease in θ power after both interventions across all groups. Elevated θ power in resting states has been associated with poor cognitive functioning [42], including deficits in attention and executive function. The overall reduction in resting frontal θ power over time may be consistent with changes in neural markers associated with stress and arousal regulation. However, given the absence of cognitive performance measures, this interpretation is exploratory and should be viewed with caution [43]. This interpretation is supported by the results of previous studies. Thornberry et al [43] demonstrated that decreases in θ power are associated with improved spatial learning. Brzezicka et al [44] reported that reduced θ power in the dorsolateral prefrontal cortex, especially under a high cognitive load, predicts better task performance, which allows the brain to manage cognitive demands more effectively. The overall reduction in θ power observed across all groups may reflect improved arousal and attention regulation, possibly enhanced by the mindfulness components inherent in the yoga sessions. However, the absence of a significant group×time interaction precludes attributing this effect specifically to yoga.

Our results are consistent with those of a growing body of literature highlighting the mental health benefits of exercise and yoga interventions. A recent systematic review by Singh et al [45] found that physical activity interventions can significantly reduce depression, anxiety, and psychological distress in various populations. Similarly, our exercise intervention reduced depressive symptoms and stress-related outcomes. The observed improvements in the exercise group align with the findings of Arida and Teixeira-Machado [46], who noted that exercise contributes to neuroplasticity and cognitive reserve, which are crucial for brain resilience to cognitive impairment.

Our findings on cognitive resilience and stress responses complement those of Shields et al [47], who found that collegiate athletes exhibited cognitive resilience through adaptations in mood and stress responses during a competitive season, suggesting that regular physical activity supports mental health and cognitive flexibility. Furthermore, Brush et al [48] showed that 8 weeks of moderate-intensity aerobic exercise can reduce depressive symptoms and influence neural indicators of reward and cognitive control in young adults with major depression, indicating potential neurobiological pathways for exercise effects and paralleling our EEG findings.

Bouthillier and Asmundson [49], Tseng et al [50], and Soteriades et al [51] demonstrated the efficacy of physical exercise (including aerobic and resistance training) in reducing anxiety, PTSD symptoms, and occupational stress in patients with anxiety disorders, middle-aged adults with poor sleep, and firefighters, respectively. By including physiological stress measures (HRV and EEG), our study provides additional evidence that such interventions can induce favorable changes in stress-related biomarkers, although our HRV results did not reach significance, possibly because of the relatively short intervention period.

Previous studies conducted on trauma-exposed and high-stress populations have supported the benefits of combining mind-body practices with exercise. Goldstein et al [52] reported that an integrative exercise program for veterans with PTSD reduced symptoms, whereas Rosenbaum et al [53] observed that an exercise program tailored to police officers with PTSD improved psychological outcomes. Similarly, our yoga program, which included mindfulness and relaxation components, produced notable improvements in hyperarousal and negative thought patterns, reflecting outcomes particularly relevant to stress and trauma-related conditions. McKeon et al [54] also showed that a mental health–informed physical activity intervention delivered via social media was feasible for first responders, suggesting that digital delivery can be effective, a finding echoed by the high adherence and engagement observed via our digital platform.

Our results on physical activity and stress align with epidemiological findings such as those of LeardMann et al [55], who observed an association between regular vigorous exercise and a lower incidence of PTSD symptoms in military personnel. In our study, participants in the exercise and yoga groups engaged in structured physical activity and showed corresponding decreases in stress and hyperarousal, supporting the protective and ameliorative effects of exercise on stress-related outcomes. Schuch et al [4] and Mikkelsen et al [5] found that physical exercise reduces anxiety and depression, likely through endorphin release and enhanced neuroplasticity. The decreases in depressive symptoms and improvements in hypervigilance and cognitive responses observed in our study align with these physiological mechanisms.

Our findings emphasize the importance of exercise intensity and enjoyment. Paolucci et al [56] noted that both high-intensity interval training and moderate continuous training could reduce depression and inflammation, although intensity matters and high intensity may increase perceived stress. In the exercise program, we chose moderate-intensity exercise and observed mental health benefits without evidence of increased stress, suggesting that a tolerable intensity level is crucial for psychological benefits. Similarly, the yoga program, which was generally low-to-moderate intensity, yielded strong benefits, consistent with the idea that the optimal intensity for mental health improvements might not necessarily be maximal.

The general improvements in mental health observed in this study are supported by large-scale data, such as those of Chekroud et al [57], who found that individuals who exercise have a significantly lower mental health burden. In that study, all exercise types were associated with better mental health, with decreases in mental health burden ranging from 12% to 22% compared to nonexercisers. We observed robust improvements in the exercise and yoga groups, reinforcing the idea that regular physical activity, whether traditional exercise or yoga, can substantially improve mental well-being. Notably, Chekroud et al [57] also found that certain group-based exercises and mindfulness practices (eg, yoga and tai chi) had larger effects than some other forms of activity, a trend mirrored in our study in which the yoga group often experienced equal or greater benefits than the conventional exercise group.

Our study adds to existing evidence on the efficacy of yoga in occupational contexts. Rhodes et al [58] demonstrated the benefits of yoga in women with treatment-resistant PTSD, and Shohani et al [59] demonstrated its positive effects on stress, anxiety, and depression in women. Jeter et al [60] found that a yoga program for police trainees improved mood and stress, whereas Weathers-Meyer et al [61] reported that firefighters who participated in an 8-week yoga program experienced reduced PTSD symptoms, negative affect, and trait anxiety. Similarly, our yoga program improved mood and stress (depressive symptoms and CSRS). Fang and Li [62] found that nurses who regularly practiced yoga reported better sleep quality and lower work stress, highlighting the broad benefits of yoga.

The comparable effectiveness of the CBT-based self-care control and yoga intervention groups may reflect the active psychological mechanisms embedded in both interventions. Self-guided CBT and psychoeducation, even in brief, have significantly reduced depressive symptoms in occupational populations [63]. This suggests that the control group did not serve as a passive comparator but instead offered evidence-based psychological tools with known antidepressant efficacy. Furthermore, stress regulation and mindfulness-based practices in yoga may overlap with CBT techniques through shared mechanisms of emotion regulation, attentional control, and cognitive reappraisal [64]. These parallels could account for the similar symptom reduction observed in both groups.

In contrast, the relatively weaker performance of the exercise group may be attributable to several pragmatic and methodological factors. First, the intervention was delivered remotely, without live supervision. Prior meta-analyses suggest that structured and professionally supervised exercise interventions tend to be associated with greater improvements in depressive symptoms compared to structured formats [65]. Second, real-world barriers, such as fatigue after long workdays and time constraints, may adversely affect engagement with exercise sessions [66]. Finally, the 4-week intervention duration may not have been sufficient to produce robust antidepressant effects from exercise [67], whereas the cognitive and emotional benefits of CBT-based self-care or yoga may manifest more quickly.

The user satisfaction findings converged with the primary outcome results. The yoga group, which showed the most pronounced and sustained improvements in depressive symptoms and cognitive stress responses, also rated their program as significantly more helpful for both stress relief and mental health management than the exercise group. This concordance between subjective perceptions of benefit and objective changes in psychological outcomes supports the internal consistency of our findings and suggests that the superior efficacy of the yoga intervention was accompanied by a correspondingly favorable user experience. The usability feedback also carries practical implications for future platform development. Across both groups, the most frequently endorsed request was for compatibility across multiple devices, and the most frequently reported barrier to sustained engagement was insufficient personal motivation to continue. These observations indicate that broadening device support and incorporating motivation-enhancing features (eg, progress reports, reminders, and social or gamified elements) may further improve adherence and engagement when such digitally delivered programs are scaled to routine workplace settings.

These findings have important clinical implications. First, yoga is a viable and accessible intervention, and exercise may complement it in reducing workplace stress and improving employee mental health. Given that these programs were delivered via a mobile platform, our study supports the potential of digital health solutions to reach employees with time or accessibility constraints. Employers and health professionals should consider incorporating such programs into workplace wellness initiatives to address stress and mental health proactively.

Second, the distinct advantages of yoga suggest that mind-body exercises may be particularly useful for individuals who struggle with stress-related cognitive distortions or heightened arousal. However, the exercise program showed meaningful improvements in vitality and energy, consistent with findings indicating that exercise boosts energy levels. A combined approach that offers both traditional exercises and yoga can cater to personal preferences and target a broader range of outcomes.

Despite these promising results, this study had several limitations. First, the relatively short intervention period (12 sessions over 4 wk) may limit the generalizability of the findings. Although this study demonstrated short-term improvements in mental health outcomes, it remains unclear whether these benefits can be sustained in the long term. Future studies should include longer follow-up periods to assess the durability of the observed effects. Second, our sample size (N=75, three groups) was calculated to detect large effects (f=0.40). A larger sample size would increase statistical power. Third, the study relied on self-reported measures of psychological outcomes, which could be subject to response bias. Objective measures of stress, such as cortisol levels, were not obtained in this study but could strengthen future research. Fourth, although we attempted to control for participant behavior outside the interventions (eg, advising against extra exercise), we cannot fully rule out external factors influencing the outcomes (eg, changes in work stress or personal life events during the study). Fifth, the control group received a CBT-based self-care program that, while intended to be minimal, might have provided some benefits (eg, stress management education). This may have diluted between-group differences. Nevertheless, the inclusion of basic psychoeducation in the control condition made our findings more applicable to real-world settings in which doing nothing is often unethical or impractical. More importantly, by comparing the interventions to active digital control, the study successfully isolated the differential impact of physical and mind-body components (yoga and exercise) from the nonspecific effects of general engagement and digital access. The observed significant group×time interactions validate that the specific mechanisms of the active interventions produced differential trajectories of mental health improvement compared to CBT-based self-care control. Sixth, the homogeneity of the sample—predominantly female, young, urban, and well-educated employees—may limit the applicability of the findings to other populations, such as men, older adults, rural workers, and those with lower digital literacy. The Korean cultural context may also influence intervention uptake. Seventh, the stringent exclusion criteria ensured participant safety but may limit applicability to employees with comorbid psychiatric or physical conditions. Eighth, several elements of this RCT may differ from routine application. In-person assessments and follow-ups conducted at the hospital provided a level of accountability and reduced loss to follow-up, which might not exist in purely remote routine applications. Additionally, the use of unique IDs and concealed allocations, when necessary, would be streamlined in a real-world scenario. The omission of in-person assessments in a routine setting might affect retention rates, underscoring the need for future studies on effective remote engagement strategies (eg, gamification or peer support). Ninth, EEG analyses were exploratory, and we did not apply formal corrections for multiple comparisons across frequency bands. As such, the observed time-related reductions in θ power should be considered hypothesis generating and require replication in future studies with preregistered EEG outcomes and appropriate controls for multiple testing. Finally, although the study successfully used a digital platform to deliver interventions, technical issues or user proficiency may have influenced engagement and outcomes. In addition, the analyses of the user satisfaction and usability questionnaire were exploratory and were not corrected for multiple comparisons; given the small subgroup sizes, these results should be interpreted with caution and regarded as hypothesis generating.

Despite these limitations, this study highlighted several unmet needs and areas for future research. First, further investigations are required to explore the long-term effects of digital mental health interventions, particularly in diverse workplaces. Understanding how different populations engage with digital platforms and whether the benefits of yoga and exercise persist beyond the initial intervention period is crucial for upscaling these programs.

Future research should focus on refining the digital delivery of mental health interventions. For example, incorporating gamification or social support elements may motivate users to adhere to the programs. Finally, studies comparing these interventions across different psychological distress levels are required. Although this study primarily focused on individuals with moderate-to-high stress, future research should investigate whether these findings hold true for individuals with lower baseline stress or severe mental health conditions. This could help tailor interventions for specific mental health needs, ensuring that the most appropriate treatment is provided.

This study provides evidence that digitally delivered adaptive yoga programs demonstrate superior and sustained improvements in depressive symptoms and CSRS compared to an active CBT-based self-care control. However, the exercise program showed more modest and less sustained effects, which warrant further investigation using larger samples. The use of a motion-detecting digital platform demonstrated high feasibility and engagement, suggesting that mobile health interventions can overcome common barriers to exercise (eg, time and accessibility) in employee populations. Our findings support the integration of regular physical activity, including mind-body practices such as yoga, into workplace wellness programs to promote mental well-being. Large-scale studies with longer follow-up periods are warranted to confirm these results, examine the long-term maintenance of benefits, and evaluate the cost-effectiveness of such interventions. Employers and health care providers should consider leveraging digital health platforms to deliver structured mind-body programs, particularly yoga, to reduce workplace stress and enhance overall mental health among employees; however, the role of exercise warrants further investigation.