This study was a follow-up to a randomized controlled trial of ANZEILAX, a digital therapeutic designed to alleviate anxiety symptoms in young adults. This 8-week exploratory single-group study examined the real-world applicability of interventions that integrate EMA/EMI-based digital therapeutics with wearable-derived digital phenotypic data and evaluated the operation of intervention routines, user acceptance, and potential changes in clinical outcome measures before a controlled clinical evaluation of the EMA/EMI-based DTx.

This study used the ANZEILAX-Green smartphone app (HAII Corp) with integrated EMA/EMI functions in conjunction with a smartwatch-based tool that collects health data to comprehensively explore real-time emotion assessment, momentary intervention delivery, and changes in behavioral and physiological responses in everyday contexts. The study was conducted from September to November 2025. An overview of the study design and procedures is presented in Figure 1.

This study was reviewed and approved by the Institutional Review Board of Yonsei University (IRB number 7001988-202509-HR-2892-03). All participants provided written informed consent after receiving a full explanation of the study purpose and procedures from the research team. Participant confidentiality was maintained throughout the study, with each individual being assigned a numerical code by their order of participation. All collected data were anonymized to ensure participant privacy. Participants who completed the 8-week study were compensated with KRW 130,000 (approximately US $88) for their time and effort, and those who completed the postintervention interview received an additional KRW 20,000 (approximately US $14) as supplementary compensation.

The participants were Korean young adults aged between 19 and 39 years, with a Beck Anxiety Inventory (BAI) score of 8 to 25 (indicative of mild-to-moderate anxiety), who could communicate in Korean and use smartphones, and included individuals who had not been prescribed psychiatric medication by a psychiatric department within 30 days before the start of the study or those without changes in medication if they had been taking any. Individuals with current or past psychiatric disorders (including schizophrenia, psychosis, bipolar disorder, and epilepsy), brain injury, cognitive impairment, neurological disorders, intellectual disability, or substance or alcohol use disorders, as well as those who had exhibited suicidal intent, suicidal impulses, or self-harming behavior within the past 6 months, were excluded. High-risk individuals with BAI scores of 26 or higher were also excluded.

Purposive sampling was used to ensure sufficient representation across age groups and anxiety severity levels (mild and moderate). Participants were approached by a member of the research team who had not been involved in quantitative data collection so as to minimize response bias. A total of 30 participants were enrolled after the research team confirmed their eligibility through online recruitment. After providing informed consent, the participants completed baseline assessments covering their sociodemographic characteristics, anxiety, and depression and then downloaded the ANZEILAX-Green app (HAII Corp). They used the app for 8 weeks in combination with a Galaxy Watch 7 to be worn throughout the study period. On the enrollment day, all participants received hands-on onboarding instructing them in the use of the app and the wearable device and were instructed to contact the research team at any time if they encountered technical difficulties.

Participants completed survey evaluations at weeks 2, 4, and 6, followed by a postintervention assessment at week 8, and those who had volunteered were invited to take part in in-depth interviews following the intervention; all 29 participants who had completed the 8-week intervention voluntarily agreed to participate in the interviews. A research team experienced in qualitative interviews conducted semistructured interviews to explore their experiences with the intervention and elicit suggestions for improvement. The key topics included participants’ overall experience with the app, their perceptions of its effectiveness and satisfaction, and any facilitating and hindering factors they had noted. In particular, an interview guide was developed a priori based on the study objectives and reviewed by the research team for content validity, and the following topics were set in advance: (1) overall experience with the ANZEILAX-Green app and wearable device, (2) perceived effectiveness and satisfaction with the intervention, (3) factors facilitating and hindering engagement, (4) perceptions of the personalized weekly feedback reports, and (5) suggestions for improvement.

All interviews were conducted individually via video call after the intervention period had ended and were audio-recorded with participants’ prior written consent. Interviews lasted approximately 30-60 minutes. The interviewer maintained a reflexive stance throughout, bracketing their assumptions about the intervention and remaining open to participants’ perspectives. The audio recordings were transcribed verbatim by a research assistant and reviewed for accuracy before analysis. Transcripts were not returned to participants to be checked at this stage. However, participant quotations representative of each theme were verified against the original transcripts to ensure accuracy.

This study used the following measures. The Generalized Anxiety Disorder 7-item scale (GAD-7) was administered to assess anxiety symptom relief [65] and the Patient Health Questionnaire-9 (PHQ-9) to evaluate depressive symptoms, which frequently co-occur with and are closely associated with anxiety [66]. Self-esteem, a psychological construct associated with anxiety levels, was assessed using the Rosenberg Self-Esteem Scale (RSES) [67]. Finally, app usability was evaluated using the mHealth App Usability Questionnaire (MAUQ) [68].

Analyses were conducted using SPSS version 29.0 statistical software (IBM Corp). Outcome measures were analyzed using repeated-measures ANOVA to examine changes over time. When the assumption of sphericity was violated, the Greenhouse-Geisser correction was applied, and pairwise differences across time were examined using post hoc comparisons with Bonferroni adjustment. Associations between digital phenotype indicators and clinical outcome measures were examined using ordinary least squares (OLS) regression implemented in Python (version 3.12; Python Software Foundation).

The qualitative data from the semistructured interviews were analyzed independently by 2 researchers using reflexive thematic analysis following the 6-phase framework of Braun and Clarke [69]. This approach was selected for its flexibility and suitability for exploratory research aimed at identifying experiential patterns across participants. The analysis proceeded inductively without a predetermined coding framework, allowing themes to emerge from the data.

The six phases were as follows: (1) familiarization, in which both researchers read and reread the transcripts and recorded initial impressions; (2) systematic coding, in which meaningful units of data were assigned descriptive codes; (3) searching for themes, in which codes were collated into candidate themes; (4) reviewing themes, in which the coherence and distinctiveness of the themes were evaluated in relation to the full dataset; (5) defining and naming themes, in which each theme was clearly delineated and given a descriptive name; and (6) writing up, in which themes were reported with supporting participant quotations.

To enhance the rigor and dependability of the analysis, each researcher coded the transcripts independently, and discrepancies were resolved through iterative discussion until consensus was reached. Initial coding was compared to ensure consistency. Credibility was supported through the use of verbatim participant quotations illustrating each theme and by grounding interpretations in the data. Transferability was addressed through a thick description of the study context, setting, and participant characteristics. Confirmability was maintained by documenting analytic decisions throughout the process. Reflexivity was practiced throughout the analysis, with both researchers reflecting on how their professional backgrounds and prior knowledge of the intervention might have shaped their interpretation of the data. All qualitative data were managed and organized using a structured coding matrix in Microsoft Excel.

A total of 61 individuals were screened for eligibility in this study. Of these, 31 were excluded because their BAI scores fell outside the range of 8-25. The other 30 participants met all the inclusion criteria and were enrolled.

Table 1 provides an overview of the demographics and baseline clinical characteristics. The mean age was 29.83 (SD 5.81) years, with participants evenly distributed between the 19-29 (15, 50%) and 30-39 years age groups (15, 50%). The majority of the participants were women (21/30, 70%), and the remaining 9/30 (30%) were men. Most participants had completed an undergraduate degree or higher, and about half were employed. The reported duration of anxiety symptoms varied, most commonly falling in the range of 1-3 years. The mean BAI score, which was used for screening, was 15.03 (SD 4.12), with 17/30 (57%) participants classified as having mild anxiety and 13/30 (43%) moderate anxiety.

Baseline outcome scores indicated moderate levels of anxiety and depressive symptoms, with mean scores of 10.70 (SD 2.85) on the GAD-7 and 10.50 (SD 4.47) on the PHQ-9. The mean baseline score on the RSES was 29.13 (SD 7.22). Based on the baseline GAD-7 scores, 10/30 (33%) participants were classified as having mild, 17/30 (57%) moderate, and 3/30 (10%) severe anxiety. Based on their PHQ-9 scores, 3/30 (10%) had none-to-minimal, 11/30 (37%) mild, 9/30 (30%) moderate, and 7/30 (23%) moderately severe depressive symptoms, with none classified as severe (Table 1).

A total of 29 participants were included in the final analysis. One participant was excluded because they did not respond to contact after the baseline assessment.

The main findings are as follows. Compared with the baseline, the GAD-7 and PHQ-9 scores showed significant within-group reductions in anxiety (GAD-7) and depression (PHQ-9), and a significant increase was found in self-esteem (RSES) scores. Please note that the following are within-group changes from a single-arm design without a control group. The mean GAD-7 score decreased from 10.66 (SD 2.89) at baseline to 5.86 (SD 4.34) at the completion of the intervention, with an average change of –4.79 points (95% CI –6.56 to –3.02; Cohen d=1.03; P<.001). The mean PHQ-9 score decreased from 10.31 (SD 4.43) to 6.45 (SD 4.99) over the same period, with an average change of –3.86 points (95% CI –5.79 to –1.93; Cohen d=0.76; P<.001), while the mean RSES score increased from 29.34 (SD 7.25) to 34.55 (SD 7.31) over this period, with an average change of 5.21 points (95% CI 3.47 to 6.95; Cohen d=1.14; P<.001), indicating a significant improvement in self-esteem over the course of the intervention (Table 2).

The changes over time also exhibited consistent improvement, as reflected by ongoing reductions in the symptom severity scores. Gradual improvements from the midpoint assessment were noted for anxiety and depression, with a significant reduction observed at week 4 during the early phase of the intervention. The downward trend continued steadily through weeks 6 and 8, with the most pronounced improvement observed postintervention. Anxiety symptoms, measured by GAD-7, showed a steady and sustained decrease from week 4 onward, while depressive symptoms, measured by PHQ-9, followed a similar overall trajectory. Although the PHQ-9 scores in week 8 were numerically slightly higher than in week 6 (not formally tested), they remained significantly lower than baseline at postintervention (Table 3; Figure 4).

These patterns of change suggest consistent improvements in symptoms throughout the intervention, with early changes significantly contributing to the overall effect. The effect sizes and temporal change patterns observed in this study are consistent with those reported in previous ANZEILAX clinical studies [58]. In particular, rapid responses and sustained effects on anxiety symptoms may be of clinical interest, although this requires confirmation by research using controlled designs.

Adherence was calculated based on the protocol requirement of 3 daily EMA/EMI completions over the 8-week intervention period. The participants maintained an average compliance frequency of 80%, thus adhering to the EMA/EMI protocols at a consistently high level.

The heatmap presented in Figure 5 displays the adherence patterns of the 29 participants along the vertical axis over 56 days. Colors closer to green indicate higher adherence, with 1.0 representing full compliance, whereas those closer to gray indicate lower adherence. Blue indicates additional EMA/EMI completions beyond the prescribed 3 daily completions.

As the heatmap illustrates, most participants consistently completed 2 to 3 EMA/EMI reports per day throughout the study, thus maintaining stable use patterns. The average adherence to EMA/EMI remained stable throughout the study period, with no sharp declines at any specific time points or weeks, despite variations in completion frequency among participants.

After the 8-week intervention period, participants completed the MAUQ to assess the usability of the app. The MAUQ results showed an average score of 5.03 (SD 1.32) on a 7-point Likert scale. Among the subscales, ease of use and satisfaction scored the highest at 5.28 (SD 1.28), followed by information organization at 4.98 (SD 1.31) and usefulness at 4.84 (SD 1.36). These results indicate that participants rated the overall usability of the app as favorable, while also making suggestions to further optimize the user experience.

In this study, several digital phenotypes, including heart rate, GPS-derived mobility indicators, sleep patterns, and activity levels measured by step count, were collected and analyzed to comprehensively detect changes in participants’ mental health status. The procedures for preprocessing and analysis were applied based on the characteristics of each type of data.

Heart rate data were categorized into 3 ranges: high, ≥100 bpm; medium, 60 bpm to <100 bpm; and low, <60 bpm. The total distance and movement radius were calculated from the GPS latitude and longitude information and normalized from 0 to 1 for comparison between individuals.

To analyze behavioral patterns by time of day, the total distance was classified into daytime and nighttime intervals, with 8 PM as the cutoff. Sleep data were analyzed using Samsung Health’s 4-stage classification, including awake, light, REM, and deep, based on sleep records from 6 PM of the preceding day to noon of the following day. Step count data were also normalized from 0 to 1, as with the GPS-derived indicators.

Weekly averages were calculated for data analysis, and trends over the 8-week period were examined using the ordinary least squares method. Statistically significant changes were observed in several physiological and behavioral indicators. The proportion of measurements within the low heart rate range significantly increased over time (β=0.008, P=.02), while GPS-based movement data exhibited a significant decrease in distance over the 8-week period (β=–0.3029, P=.03). Among sleep-related indicators, deep sleep duration showed a significant increase (β=0.2624, P=.02), whereas light sleep duration showed a decreasing trend. Although changes in step count did not reach statistical significance, an upward trend was observed (β=0.1475, P=.11). These wearable-derived indicators were not adjusted for physical activity. In addition, heart rate ranges were defined using fixed rather than individualized thresholds, and consumer-wearable sleep-stage estimates are less accurate than polysomnography. These exploratory findings should thus be interpreted with caution.

Analyses of the correlations between digital phenotyping indicators and clinical measures revealed several significant associations. A significant correlation was identified between PHQ-9 scores and deep sleep duration (r=0.655, P=.04), indicating an association between depressive symptom severity and deep sleep characteristics. In addition, low heart-rate indicators showed significant positive correlations with both anxiety and depressive symptom severity, as measured by GAD-7 (r=0.578) and PHQ-9 (r=0.836).

Exploratory analyses found that several digital phenotyping indicators showed concurrent changes during the intervention period. Given the small sample and wide CIs, these associations should be interpreted cautiously (Table 4; Figure 6).

This study examined the real-world applicability and effectiveness of an EMA/EMI-based digital therapeutic, ANZEILAX-Green, in anxiety reduction among young adults and investigated how the personalized feedback generated by the combined EMA/EMI data and wearable-based digital phenotypes supported symptom improvement, self-management, and user engagement. The results of this 8-week exploratory study indicate that personalized feedback integrating the EMA/EMI system with wearable-based digital phenotype data was associated with improvements in both clinical outcomes and participants’ capacity for autonomous symptom management, although causality cannot be inferred due to the single-arm design of this study.

The intervention was associated with significant improvements in psychological outcomes. GAD-7 scores showed a marked reduction with a large effect size (Cohen d=1.03) and PHQ-9 scores a medium-to-large effect size (Cohen d=0.76), while RSES scores showed a significant increase with a large effect size (Cohen d=1.14). The intervention was thus associated with reductions in anxiety symptoms and concurrent improvements in depressive symptoms and self-esteem, suggesting a broader positive impact on overall mental health. GAD-7 scores began to decline from week 2, showed statistically significant reductions from week 4 onward, and continued to improve through postintervention. These effect sizes and temporal change patterns are generally comparable to those reported elsewhere [58].

The magnitude of the observed effects suggests a meaningful therapeutic signal in real-world use. The relatively large effect sizes may in part reflect the single-arm study design, but they also appear to be related to participants’ repeated engagement with the EMA/EMI system in their daily lives. Repeated use of the app likely supported continuous self-monitoring and the timely application of therapeutic strategies. In addition, the study sample primarily consisted of young adults with mild-to-moderate anxiety symptoms and relatively short symptom histories, who might have been more responsive to structured self-management support.

This pattern of early improvement and sustained reduction in symptoms suggests that EMA/EMI-based interventions can help attenuate symptom exacerbation in everyday contexts and support ongoing self-management through the timely delivery of therapeutic content. The EMA/EMI system was designed for frequent use while imposing a low perceived burden through personalized, context-appropriate intervention tasks. This balance might have supported sustained engagement and enabled repeated exposure to therapeutic strategies feasible for use in daily life, which might have contributed to gradual symptom regulation and the reinforcement of adaptive coping processes.

Participants maintained a high compliance rate of 80% over the 8 weeks, consistently adhering to the EMA/EMI protocol. Most participants completed 2 to 3 EMA/EMI per day throughout the study, demonstrating stable usage patterns without sharp declines at any time point. This suggests that EMA/EMI use was naturally integrated into the participants’ daily routines. The provision of weekly personalized feedback reports might have further supported sustained engagement by reinforcing a sense of individualized care and ongoing monitoring. Given that user attrition has been identified as a major challenge in digital mental health intervention research, the stable engagement patterns observed in this study demonstrate that the EMA/EMI approach can be an acceptable and sustainable intervention modality for young adults. These improvements are broadly consistent with those obtained from digital therapeutics and app-based interventions that reported reductions in anxiety and depressive symptoms [71,72].

At the same time, although the frequent EMA prompts and repeated self-reporting were generally well tolerated, they were perceived as cognitively or emotionally demanding in certain situations. In particular, during periods of heightened stress, users’ receptivity to intensive assessment may vary, highlighting the importance of flexibly balancing assessment frequency with user burden in the design of EMA/EMI systems. The streamlined nature of the EMA/EMI tasks might have contributed to the high levels of engagement observed in this study. However, the therapeutic effects demonstrated in this population with mild-to-moderate anxiety symptoms may not readily generalize to individuals with chronic anxiety or more severe psychiatric conditions, for whom more intensive or clinically supervised interventions may be necessary.

Significant changes in several physiological and behavioral indicators were also observed in the digital phenotype analysis. Deep sleep duration increased significantly while light sleep decreased, and GPS-based movement distance decreased significantly. In addition, the proportion of measurements falling within the low-heart-rate range increased significantly over time, which is to some extent consistent with lower physiological arousal. However, heart rate alone is a weak index of relaxation. HRV, which reflects parasympathetic modulation, would be a more informative physiological marker. Furthermore, these analyses were not adjusted for physical activity. A significant correlation was identified between PHQ-9 and deep sleep duration, indicating a possible association between depressive symptoms and deep sleep. This interpretation assumes reliable wearable sleep-stage detection and, given the small sample, requires cautious reading.

The concurrent increase in deep sleep and reduction in light sleep may indicate an overall improvement in sleep quality, which has been associated with reduced physiological arousal and greater emotional stability. Similarly, the increasing proportion of low heart rate measurements may reflect a more stable pattern of autonomic regulation, consistent with reduced sustained stress or anxiety-related arousal in daily life. Naturalistic studies combining wearable sensing with EMA have similarly examined the physiological correlates of affect and stress in everyday settings, albeit with varying predictive accuracy [42,54,73]. Although the decrease in movement distance may appear negative at first glance, when considered alongside the increasing trend in step count, it suggests that regular activity patterns within primary living areas were established rather than more extensive movement.

The interview analysis yielded 5 major themes in participants’ experiences with the intervention: enhanced self-awareness, promotion of data-driven reflection, support for behavioral change, maintenance of intrinsic motivation, and expansion of active treatment. These themes suggest that the integration of EMA/EMI systems with digital phenotype data contributed to meaningful changes in clinical outcomes, behavioral patterns, and participants’ capacity for self-management.

The combination of real-time self-reported data and objective physiological indicators helped participants develop a more coherent understanding of their emotional and physical states and appears to have been particularly beneficial for individuals with elevated anxiety, who often experience discrepancies between their physical symptoms and subjective perceptions. By providing objective reference points alongside subjective assessments, the system helped participants recognize patterns that might otherwise have remained unnoticed and supported the development of more targeted self-management strategies. Furthermore, the finding that participants independently explored additional therapeutic modules beyond the core app functions suggests that the intervention successfully fostered user autonomy. The therapeutic benefits were thus not confined to moments of active app use but extended into participants’ daily lives through the self-initiated exploration and application of learned strategies.

This study has several limitations that should be borne in mind. First, the exploratory single-arm design limits causal inference regarding the observed changes. Without a control group, natural course, regression to the mean, recruitment bias, and expectancy effects cannot be excluded. In addition, because participants could optionally use other therapeutic modules within ANZEILAX-Green, the observed changes cannot be attributed specifically to the EMA/EMI or wearable components.

Second, the wearable-derived indicators were treated as contextual physiological and behavioral signals rather than validated objective measures of anxiety. Therefore, these findings should be interpreted cautiously. Repeated self-monitoring may also have affected participants differently, as a small number of participants reported heightened awareness of anxiety during periods of elevated distress. Future EMA-based interventions should therefore consider individual differences in response to repeated self-assessment and balance monitoring intensity with user burden [74].

Third, although the sample size was sufficient for this exploratory assessment of feasibility and acceptability, it limited the analysis of individual differences and the generalizability of the findings. In addition, because no follow-up assessment was conducted, it remains unknown whether the observed improvements were sustained beyond the intervention period. Future studies should use larger and more diverse samples and include longer follow-up periods to examine the durability of the observed effects.

In conclusion, this study provides preliminary evidence for the feasibility, acceptability, and potential clinical benefits of ANZEILAX-Green, an EMA/EMI-based mobile intervention, in reducing anxiety symptoms among young adults. The integration of EMA/EMI systems with wearable-derived digital phenotype data enables a comprehensive assessment of subjective symptom changes and objective indicators, including physiological and behavioral data, demonstrating how personalized feedback can operate within everyday contexts. This integrative approach extends the clinical applicability of mobile mental health interventions by allowing the multidimensional emotional changes associated with anxiety symptoms in daily life to be captured. A randomized controlled trial is thus needed to confirm its effectiveness.

Beyond symptom reduction, the findings suggest that EMA/EMI-based interventions integrated with digital phenotyping can support ongoing self-management by mitigating symptom exacerbation in everyday contexts and facilitating the application of therapeutic strategies outside moments of active app use. In addition, the observed clinical effects appear to be supported by an integrated feedback structure that leverages digital phenotype data to enhance the acceptability and enactment of the intervention in daily life.

Furthermore, the findings provide empirical support for the design and advancement of automated digital mental health interventions that reflect individuals’ ongoing states and contextual conditions in the feedback they provide. By showing that the integration of EMA/EMI systems with digital phenotyping extends and supports self-awareness, reflection, and self-management, this study highlights a scalable and clinically meaningful direction for future context-responsive digital mental health interventions.