Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression: A Proof-of-Concept Study

arXiv:2512.02608 · cs.HC · Submitted 2025-12-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.

Tom: I'm Tom.

Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.

Tom: We're looking at the paper: "Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression".

Jane: Here's the gist of the paper. The study proposes to design a digital intervention that integrates a therapeutic companion agent to create a therapeutic companion agent (TCA) for young adults with symptoms of depression Study Objective and Design This study aimed to develop a digital intervention prototype using the Wizard-of-Oz method, centered on a TCA that incorporates the following elements of social support: ecological momentary assessment, ecological momentary intervention, behavioral activation, and gamification The prototype was built using the messaging app KakaoTalk (a mobile social messenger service which is widely used in Korea) and Google Forms (a global online survey tool) The study was designed as a quasi-experimental trial with a nonconcurrent control group Both groups underwent pre- and post-intervention assessments The intervention group received the digital intervention for 6 weeks, guided by a companion agent to perform four daily routine activities for behavioral activation, three daily mood assessments, and meditation Intervention Components The TCA design was informed by prior research on companionship and social support and comprised the following key components: companion persona, social support (including informational, emotional, appraisal, and instrumental support), and companionship formation The companion persona chosen was a dog persona based on prior research highlighting the mental health benefits and depression-alleviating effects of dogs Social support included four types: informational support based on EMA, emotional support via EMI, appraisal support through BA, and instrumental support through gamification Informational Support involved collecting data from participants in real time during their daily lives using EMA to capture changes in emotions, behaviors, and cognitions Emotional Support consisted of EMI content that directly helps with emotional relief, comprised of mindfulness-based meditation interventions for depression Appraisal Support was implemented through a daily routine-based BA as appraisal support provided by the TCA, enabling participants to reevaluate and reconstruct their everyday lives Instrumental Support was provided through a gamification strategy including weekly missions, happy miles, levels, and donation challenges Clinical Outcomes The analysis revealed significant differences between the control and intervention groups regarding depression, anxiety, and quality of life at week 6 (P<.05) The intervention group showed a greater reduction in depressive symptoms (P=.012, partial η2=.066) and a greater improvement in quality of life (P<.001, partial η2=.139) than the control group over the 6-week study period The mean BDI-II score at baseline was 18.18 (SD=7.20), which decreased to 9.50 (SD=6.44) at week 6 in the intervention group, compared to a score of 13.76 (SD=9.50) in the passive control group A significant interaction between time and condition was observed on the alleviation of depressive symptoms (F(3,165)=3.75, P=.01, partial η2=.066) Improvements in quality of life varied meaningfully by group (F(3,165)=8.564, P<.001, partial η2=.139) Engagement and Adherence The intervention showed high adherence rates of 78% for EMA, 65% for daily routines, and 51% for EMI None of the participants in the intervention group dropped out during the 6-week intervention period The TCA-based digital intervention appears to have been accepted by the participants as a social-level treatment beyond the personal level This suggests that participants perceived the digital intervention content and messages delivered by the TCA not as a task but as a social interaction with a partner who was trying to help them Study Limitations First, the digital intervention was not a fully automated system; rather, it was provided as a WoZ prototype Second, this study did not recruit the control and intervention groups simultaneously and did not use randomization Third, most participants were not diagnosed with depression, so their depression levels were relatively low Future research will require an evaluation of the clinical effectiveness of the finalized system This PoC study found the potential for a TCA-centered digital intervention that uses a therapeutic companion concept and dog persona to positively impact the alleviation of depression and anxiety symptoms and improve the subjective quality of life in young adults with depressive symptoms Our study results suggest the possibility of using CAs in future digital interventions for depression in young adults and are expected to effectively increase the acceptability and engagement of digital interventions The TCA-based approach is expected to be less expensive than previously used methods to increase the digital intervention adherence Moreover, by using various commercially available generative AI services and applying prompt engineering techniques according to the predefined anthropomorphic style of the TCA, personalized digital intervention services can be easily concretized The TCA-based designs may increase adherence to and treatment effectiveness of digital interventions for various diseases beyond mental illnesses like depression Our study results suggest the possibility of using CAs in future digital interventions for depression in young adults and are expected to effectively increase the acceptability and engagement of digital interventions The TCA-based designs may increase adherence to and treatment effectiveness of digital interventions for various diseases beyond mental illnesses like depression References

1: Seighali, N., Abdollahi, A., Shafiee, A., Amini, M.

Tom: J., Teymouri Athar, M.

Jane: M., Safari, O., and Bakhtiyari, M.

Tom: The global prevalence of depression

5: Capili, B., and Anastasi, J.

Title and authors: Tom: So we're looking at this paper now from arXiv titled "Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression: A Proof-of-Concept Study". It's about seeing if putting a companion agent into a digital intervention actually helps young adults who are depressed.

Jane: Exactly. The core idea here is moving beyond just a standard app and creating this therapeutic companion agent, or TCA, that tries to act like a real friend or support system.

Lu: The study proposes designing this digital intervention prototype using the Wizard-of-Oz method, which means they built it up manually with a human guiding it through the messaging app KakaoTalk and Google Forms.

Meng: A Wizard of Oz setup is interesting because it lets them test the interaction flow without needing to have a fully automated AI system ready right away, which is smart for a proof-of-concept.

Lalam: It’s a proof of concept study, so they're not claiming massive clinical certainty yet, but they are testing the idea that this TCA approach could work.

Tom: Right. So what exactly did they build in this prototype? What was the agent supposed to do for these young adults?

Jane: They focused on integrating several types of social support into the agent's design, including ecological momentary assessment, or EMA, and ecological momentary intervention, or EMI.

Lu: The TCA design itself was built around a dog persona based on prior research showing that dogs can have mental health benefits and help with depression.

Meng: A dog persona sounds really engaging for building rapport initially, which is important for getting people to stick with a digital tool.

Lalam: They included four types of social support: informational support through EMA, emotional support through EMI, appraisal support via behavioral activation, and instrumental support through gamification.

Tom: Informational support means the agent collects data in real time using EMA to see how the user's emotions and behaviors are changing day by day.

Jane: And emotional support comes from EMI content that gives direct emotional relief, like mindfulness-based meditation interventions for depression.

Lu: Appraisal support is done through a daily routine-based behavioral activation strategy, where the TCA helps participants reevaluate and reconstruct their everyday lives based on their current mood.

Meng: Behavioral activation sounds like it’s about getting people to do small, manageable things every day, like washing or writing a to-do list.

Lalam: Instrumental support is handled by gamification elements, such as weekly missions, happy miles, levels, and donation challenges to keep people motivated.

Tom: So the intervention group received this digital package for six weeks—four daily routine activities guided by the agent—three daily mood assessments, and meditation sessions <ref:2512.02608#pg1>.

Jane: And in terms of the trial setup, it was a quasi-experimental trial with a nonconcurrent control group, so they had a comparison group getting no intervention at all.

Lu: Both groups underwent pre- and post-intervention assessments to see if there were any shifts in their depression scores or quality of life.

Title and authors: Meng: And the results showed significant differences at week six between the two groups regarding depression, anxiety, and overall quality of life <ref:2512.02608#pg1>.

Tom: Specifically, the intervention group had a greater reduction in depressive symptoms at week six compared to the control group over that six-week period <ref:2512.02608#pg1>.

Jane: They saw a reduction in their BDI-II score from an average of eighteen point eighteen at baseline down to nine point fifty by week six, which is quite a drop.

Lu: Compared to the control group, who had a baseline BDI-II score of thirteen point seventy-six the intervention group showed a much larger improvement <ref:2512.02608#pg2,the intervention group showed a>.

Meng: But we also saw some important engagement numbers here—they had high adherence rates for EMA at seventy-eight percent and daily routines at sixty-five percent.

Lalam: And they even had fifty-one percent adherence for the emotional support component through EMI, which shows people were actually using the emotional content.

Tom: That's good, because engagement is often the biggest hurdle in digital mental health treatments, so high adherence is a big win here.

Jane: The paper suggests that participants felt this TCA-based intervention wasn't just a task to complete, but more like an actual social interaction with someone trying to help them.

Lu: That perception of social interaction seems key; it changes how people engage with the system, moving it from being just a piece of software to something they connect with.

Meng: So what does this mean for us when we think about deploying these kinds of companion agents at scale? Is this proof-of-concept scalable?

Tom: The study suggests that the TCA concept has potential for using conversational agents in future digital interventions for depression in young adults, which is a big area.

Jane: It points toward making these digital interventions more acceptable and engaging than what we’ve seen before because of this companionship aspect.

Lu: The paper also explores how these companion agent designs might increase the adherence to and treatment effectiveness of digital interventions for various diseases beyond just mental illnesses like depression.

Meng: That’s ambitious, extending the applicability beyond just clinical depression symptoms, which is something I’m interested in seeing explored further.

Lalam: The authors suggest that by using various commercially available generative AI services and applying prompt engineering techniques based on the TCA's style, personalized digital intervention services can be made very concrete.

Tom: So the idea is that we can use existing AI tools to quickly build these personalized companion experiences for different needs.

Jane: And they think this TCA-based approach could potentially be less expensive than older methods for boosting adherence in a digital setting.

Lu: If they can leverage generative AI effectively, it makes the creation of these personalized therapeutic companions much more accessible and cost-effective for researchers and developers.

Meng: From my side, I’m thinking about the operational reality; if we use these generative AI services to personalize the content, how hard is it to maintain that level of personalization at scale?

Title and authors: Lalam: The paper implies that this personalized approach could significantly improve both adherence and treatment effectiveness for a wide range of conditions.

Tom: So, to wrap up, the study on "Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression: A Proof-of-Concept Study" shows that integrating social support via a TCA can lead to better clinical outcomes and higher user engagement.

Jane: It’s showing that giving users an agent that provides emotional, informational, and instrumental support through routine activities makes a real difference in reducing depressive symptoms.

Lu: The key finding is the significant improvement in depressive symptoms at week six when compared to the control group, with a partial eta squared of zero point sixty-six for that effect.

Meng: But we have to remember their limitation here: this was only a proof-of-concept study, and they also noted that most participants were not clinically diagnosed with depression, so their depression levels were relatively low.

Lalam: And they also pointed out the limitations of the setup itself—it wasn't a fully automated system, but a Wizard of Oz prototype, and they didn't recruit the control and intervention groups at the same time.

Tom: So while it’s promising for showing potential, we still need more rigorous clinical evaluations to confirm these results in a broader setting.

Jane: True. The next step is definitely refining this TCA design and doing those deeper, more comprehensive evaluations they mentioned in their conclusions.

Lu: I think the real implication here is the framework itself—the combination of EMA, EMI, behavioral activation, and gamification structured around a persona like a dog seems to be a viable blueprint.

Meng: If we look at deployment costs for something like this, having high adherence rates means lower churn and less effort spent on re-engagement campaigns down the line.

Lalam: And the idea of using prompt engineering with generative AI to easily realize these personalized designs is a practical path forward for getting these concepts into production quickly.

Tom: So we’ve seen how this TCA approach impacts depression outcomes, but the real takeaway is that personalization through agent design and leveraging current generative AI can make these digital tools much more effective and engaging.

Jane: It really suggests that the therapeutic alliance formed with a companion agent is a powerful mechanism for improving health outcomes in young adults dealing with depression.

Lu: The paper on "Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression: A Proof-of-Concept Study" demonstrates how integrating multiple support types can lead to measurable reductions in depressive symptoms and quality of life.

Meng: And from an engineering standpoint, the fact that adherence was high despite the demanding nature of the intervention is something we should keep focused on for scaling these things.

Lalam: It’s a solid proof-of-concept showing that this TCA-centered digital intervention has potential to positively impact depression and anxiety symptoms in young adults.

The paper's summary: Tom: So, to recap, this paper is looking at a proof of concept where they designed an integrated digital intervention using this therapeutic companion agent for young adults dealing with depression symptoms.

Jane: Right, and what they found in that test was that combining several kinds of social support—like real-time emotional feedback through a dog persona—actually led to better results than just doing the standard treatment alone.

Tom: Exactly. They weren't just testing one thing; they had this setup with ecological momentary assessment, which is tracking mood moment by moment, and behavioral activation, which makes them do daily routines.

Jane: And the numbers on that were pretty compelling at week six. They saw a significant drop in depression scores for the group using the agent compared to the control group who got nothing but passive support.

Tom: The baseline depression score went from eighteen point one eight down to nine point five zero in their intervention group, which is a substantial change over six weeks.

Jane: But what I found most interesting wasn't just the drop in scores; it was how they felt about using the agent. They reported high adherence rates for things like tracking their daily routines, even though it was a demanding intervention.

Tom: That adherence is key for any digital health product, right? And they also noted that participants didn't feel like they were doing a chore; instead, it felt more like interacting with a partner who was actively trying to help them.

Jane: That shift in perception—from task to social interaction—is what the authors are really highlighting here. It suggests that designing an AI agent not just for information delivery but for companionship is what makes these digital tools stick with people.

Tom: I agree, and that opens up a lot of possibilities. If we can get this level of engagement through a well-designed persona, it changes how we think about scaling these things beyond just clinical depression treatment.

Jane: It really pushes us to think about how much personalized social scaffolding an intervention needs to be effective for young adults who are struggling with mental health challenges.

Tom: And that leads us right into the mechanics of building that agent itself, because if you want high engagement at scale, you need a model that can handle those nuanced interactions seamlessly.

The paper's improvements: Tom: So we're moving on to how they suggest making this prototype better, and what that means for future development of these companion agents.

Jane: The authors are suggesting several ways to refine the design, starting with making the companion agent even more human-like so it builds a better therapeutic alliance.

Tom: They propose using persona-based models that can mimic humans more closely, like the dog persona they used, to foster a stronger connection with the user.

Jane: That's smart because if people feel like they are talking to something genuinely responsive, they’ll stay engaged much longer than if it’s just delivering static information.

Tom: Then there’s this idea for personalization using advanced AI, specifically large language models like GPT-four or GPT-four-o.

Jane: They suggest using those generative models to create totally customized intervention content, tailoring it not just to the user's age but even their specific depression level and gender.

Tom: So instead of one generic meditation script, the agent could generate a five minute audio intervention or a short video tailored exactly to what that person needs right now.

Jane: And they want to integrate behavioral activation more deeply into the daily structure, prompting users for specific routine activities like washing or tracking their step count.

Tom: That way, instead of just being told to do something, the agent actively guides them through reconstructing their everyday life based on what the user is currently experiencing.

Jane: They also talked about upgrading the emotional feedback loop using AI to analyze that real-time mood data from the assessment and then immediately offering tailored interventions, like ACT based journaling, when it sees a depressive fluctuation.

Tom: That level of immediacy sounds powerful, but I have to ask—how much computational power is needed for that kind of instant analysis across thousands of users?

Jane: Well, they're looking at making these personalized reports visual too. They suggest using generative AI to create illustrated diary images based on the user’s text entries, adding a word of praise and color-coded results right there.

Tom: That moves it from just text and numbers into something much more rich for self-perception, which is what we need for real healing.

Jane: The implication here is that the next generation of these agents won't just be chatbots; they’ll be dynamic, personalized therapists that use AI to provide support in a way that feels deeply personal and immediately relevant.

Conclusion: Tom: So we've covered the mechanics of this proof of concept study on "Investigating the Integrated Digital Interventions Delivered by a Therapeutic Companion Agent for Young Adults with Symptoms of Depression".

Jane: Exactly, and what we see is that when you combine behavioral activation, real-time mood tracking, and a persona like a dog agent, you get measurable drops in depression scores.

Tom: It shows that this TCA approach can actually work for improving quality of life in young adults with depressive symptoms over six weeks.

Jane: The big implication here is that we are looking at a new way to deliver support digitally, moving away from just static information delivery toward something that feels like a sustained social presence.

Tom: If we can nail the personalization and engagement aspect they showed, this could be a viable model for scaling digital mental health tools across different conditions.

Jane: It really shifts the focus from just making an app functional to designing an AI agent that builds a therapeutic relationship with the user.

Lu: I think what’s really exciting is that this framework shows how well distinct social support types—informational, emotional, appraisal, and instrumental—can be layered together effectively.

Meng: From my side, the scale is the question. If we try to deploy this kind of personalized agent system across millions of users at once, we have to worry about the computational cost of that real-time analysis.

Lalam: I’m confident in the direction they are pointing; based on all the literature, this combination of tailored interaction and structured support is exactly what's needed to make these agents effective for people.

Tom: It’s a solid proof-of-concept showing that this TCA-centered digital intervention has real potential to positively impact depression and anxiety symptoms in young adults.

Jane: And it suggests that using conversational agents can significantly boost both the acceptability and the actual effectiveness of digital interventions for various diseases beyond just mental illnesses.

Tom: We've seen how this TCA approach impacts depression outcomes, but the real takeaway is that personalization through agent design and leveraging current generative AI can make these digital tools much more effective and engaging.

Jane: It really suggests that the therapeutic alliance formed with a companion agent is a powerful mechanism for improving health outcomes in young adults dealing with depression.

Lu: And the framework they built seems like a good starting point for understanding how to structure complex, multi-faceted interventions using AI companionship.

Youngjae Yoo, Minuk Kim, Soyoung Kim, Gayeon Lee, Jinwoo Kim

Yonsei University · HAII Corp

cs.HC

Submitted: 2025-12-02

Updated: 2025-12-02

DOI: 10.1177/20552076261464801

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 74/100

The gist: The study proposes to design a digital intervention that integrates a therapeutic companion agent to create a therapeutic companion agent (TCA) for young adults with symptoms of depression Study…

Key concepts

Therapeutic Companion Agent (TCA)
A digital agent designed to offer social support using elements like a companion persona (e.g., a dog), emotional support, and instrumental help. It aims to create a therapeutic alliance by acting as a supportive partner for users struggling with depression.
Ecological Momentary Assessment (EMA)
A method where participants use their mobile devices to report on their emotions, behaviors, and cognitions in real-time throughout their daily lives. This data helps the agent understand immediate changes in a user's state.
Behavioral Activation (BA)
A technique used to improve mood by encouraging individuals to engage in structured, routine activities. The TCA prompts users to perform these activities, helping them reevaluate and reconstruct their daily lives positively.
Gamification
The use of game design elements like missions, levels, and rewards (e.g., happy miles) within the intervention. This strategy is used here to provide instrumental support and maintain user motivation through challenges.

Terminology

Summary

The study proposes to design a digital intervention that integrates a therapeutic companion agent to create a therapeutic companion agent (TCA) for young adults with symptoms of depression

Study Objective and Design

This study aimed to develop a digital intervention prototype using the Wizard-of-Oz method, centered on a TCA that incorporates the following elements of social support: ecological momentary assessment, ecological momentary intervention, behavioral activation, and gamification The prototype was built using the messaging app KakaoTalk (a mobile social messenger service which is widely used in Korea) and Google Forms (a global online survey tool) The study was designed as a quasi-experimental trial with a nonconcurrent control group Both groups underwent pre- and post-intervention assessments The intervention group received the digital intervention for 6 weeks, guided by a companion agent to perform four daily routine activities for behavioral activation, three daily mood assessments, and meditation

Intervention Components

The TCA design was informed by prior research on companionship and social support and comprised the following key components: companion persona, social support (including informational, emotional, appraisal, and instrumental support), and companionship formation The companion persona chosen was a dog persona based on prior research highlighting the mental health benefits and depression-alleviating effects of dogs Social support included four types: informational support based on EMA, emotional support via EMI, appraisal support through BA, and instrumental support through gamification Informational Support involved collecting data from participants in real time during their daily lives using EMA to capture changes in emotions, behaviors, and cognitions Emotional Support consisted of EMI content that directly helps with emotional relief, comprised of mindfulness-based meditation interventions for depression Appraisal Support was implemented through a daily routine-based BA as appraisal support provided by the TCA, enabling participants to reevaluate and reconstruct their everyday lives Instrumental Support was provided through a gamification strategy including weekly missions, happy miles, levels, and donation challenges

Clinical Outcomes

The analysis revealed significant differences between the control and intervention groups regarding depression, anxiety, and quality of life at week 6 (P<.05) The intervention group showed a greater reduction in depressive symptoms (P=. p. 1

Where the paper says it

The analysis revealed significant differences between the control and intervention groups regarding depression, anxiety, and quality of life at week 6 (P<. The intervention group showed a greater reduction in depressive symptoms (P=.

012, partial η2=. p. 1
Where the paper says it

The intervention group showed a greater reduction in depressive symptoms (P=. 012, partial η2=.

066) and a greater improvement in quality of life (P<. p. 1
Where the paper says it

012, partial η2=. 066) and a greater improvement in quality of life (P<.

001, partial η2=. p. 1
Where the paper says it

The intervention group showed a greater reduction in depressive symptoms (P=. 012, partial η2=.

p. 1
Where the paper says it

066) and a greater improvement in quality of life (P<. 001, partial η2=.

p. 1
Where the paper says it

066) and a greater improvement in quality of life (P<. 001, partial η2=.

139) than the control group over the 6-week study period The mean BDI-II score at baseline was 18. p. 1
Where the paper says it

001, partial η2=. 139) than the control group over the 6-week study period.

18 (SD=7.20), which decreased to 9.50 (SD=6.44) at week 6 in the intervention group, compared to a score of 13. p. 1
Where the paper says it

One participant from the intervention group dropped out; thus, data from 57 participants were analyzed. The analysis revealed significant differences between the control and intervention groups regarding depression, anxiety, and quality of life at week 6 (P<.

76 (SD=9.50) in the passive control group A significant interaction between time and condition was observed on the alleviation of depressive symptoms (F(3,165)=3. p. 1
Where the paper says it

The analysis revealed significant differences between the control and intervention groups regarding depression, anxiety, and quality of life at week 6 (P<. The intervention group showed a greater reduction in depressive symptoms (P=.

75, P=.01, partial η2=.066) Improvements in quality of life varied meaningfully by group (F(3,165)=8. p. 1
Where the paper says it

012, partial η2=. 066) and a greater improvement in quality of life (P<.

564, P<.001, partial η2=.139)

Engagement and Adherence

The intervention showed high adherence rates of 78% for EMA, 65% for daily routines, and 51% for EMI None of the participants in the intervention group dropped out during the 6-week intervention period The TCA-based digital intervention appears to have been accepted by the participants as a social-level treatment beyond the personal level This suggests that participants perceived the digital intervention content and messages delivered by the TCA not as a task but as a social interaction with a partner who was trying to help them

Study Limitations

First, the digital intervention was not a fully automated system; rather, it was provided as a WoZ prototype Second, this study did not recruit the control and intervention groups simultaneously and did not use randomization Third, most participants were not diagnosed with depression, so their depression levels were relatively low Future research will require an evaluation of the clinical effectiveness of the finalized system

Conclusion

This PoC study found the potential for a TCA-centered digital intervention that uses a therapeutic companion concept and dog persona to positively impact the alleviation of depression and anxiety symptoms and improve the subjective quality of life in young adults with depressive symptoms Our study results suggest the possibility of using CAs in future digital interventions for depression in young adults and are expected to effectively increase the acceptability and engagement of digital interventions The TCA-based approach is expected to be less expensive than previously used methods to increase the digital intervention adherence Moreover, by using various commercially available generative AI services and applying prompt engineering techniques according to the predefined anthropomorphic style of the TCA, personalized digital intervention services can be easily concretized The TCA-based designs may increase adherence to and treatment effectiveness of digital interventions for various diseases beyond mental illnesses like depression Our study results suggest the possibility of using CAs in future digital interventions for depression in young adults and are expected to effectively increase the acceptability and engagement of digital interventions The TCA-based designs may increase adherence to and treatment effectiveness of digital interventions for various diseases beyond mental illnesses like depression

References

[1] Seighali, N., Abdollahi, A., Shafiee, A., Amini, M. J., Teymouri Athar, M. M., Safari, O., and Bakhtiyari, M. The global prevalence of depression

[5] Capili, B., and Anastasi, J. An Introduction to Types of Quasi-Experimental Designs

[18] Dahlbäck, N., Jönsson, A., and Ahrenberg, L. Wizard of Oz studies: why and how

[20] Seighali, N., Abdollahi, A., Shafiee, A., Amini, M. J., Teymouri Athar, M. M., Safari, O., and Bakhtiyari, M. The global prevalence of depression

[23] Tremain, H., McEnery, C., Fletcher, K., and Murray, G. The therapeutic alliance in digital mental health interventions for serious mental illnesses

[26] Henson, P., Wisniewski, H., Hollis, C., Keshavan, M., and Torous, J. Digital mental health apps and the therapeutic alliance: initial review

[54] Schulz, K. F., Altman, D. G., and Moher, D. CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials

[84] Beck, A. T., Steer, R. A., and Brown, G. Beck Depression Inventory–II

[13] Janssen, N. P., Hendriks, G. J., Sens, R., Lucassen, P., Oude Voshaar, R. C., Ekers, D., and Bosmans, J. E. Cost-effectiveness of behavioral activation compared to treatment as usual for depressed older adults in primary care

[14] Paul, M., Bullock, K., Bailenson, J., and Burns, D. Examining the Efficacy of Extended Reality–Enhanced Behavioral Activation for Adults With Major Depressive Disorder: Randomized Controlled Trial

[20] Six, S. G., Byrne, K. A., Tibbett, T. P., and Pericot-Valverde, I. Examining the effectiveness of gamification in mental health apps for depression

[23] Tremain, H., McEnery, C., Fletcher, K., and Murray, G. The therapeutic alliance in digital mental health interventions for serious mental illnesses

[27] Tong, F., Lederman, R., D'Alfonso, S., Berry, K., and Bucci, S. Conceptualizing the digital therapeutic alliance in the context of fully automated mental health apps: A thematic analysis

[48] Inkster, B., Sarda, S., and Subramanian, V. An empathy-driven, conversational artificial intelligence agent (Wysa) for digital mental well-being

[109] Dahlbäck, N., Jönsson, A., and Ahrenberg, L. Wizard of Oz studies: why and how

[112] Pérez, J. C., Fernández, O., Cáceres, C., Carrasco, Á. E., Moessner, M., Bauer, S., and Krause, M.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Digital Therapeutic Alliance formation through Conversational Agents (TCA) can be improved by designing persona-based models that can mimic humans to foster a better DTA as suggested in the text. The AI system can engage users by adopting specific personas, such as the dog persona Happy, which is chosen for its ability to provide emotional support and connections.

  2. Just-in-time interventions can be enhanced by using LLM-based generative AI like GPT-4o to create personalized content that is tailored to a user's specific characteristics, including their gender, age, [and] depression level. The system can generate 5-minute or one-minute intervention texts and videos customized for each participant.

  3. Behavioral activation (BA) can be made more effective by integrating routine-based BA into the daily structure of the digital intervention. The AI system can prompt users to complete daily routine activities such as washing or showering, writing a daily to-do list, and tracking step counts, thereby helping participants reevaluate and reconstruct their everyday lives.

  4. Gamification elements should be structured with differentiated challenges based on user levels to maintain motivation. The system can implement a leveling-up system where missions progress from quantitative goals in early levels to qualitative goals centered on BA activities in later levels, rewarding successful completion with points like Happy Mileage.

  5. Ecological momentary assessment (EMA) and ecological momentary intervention (EMI) should be refined by using AI to analyze real-time data for more precise interventions. The system can use EMA data to capture real-time fluctuations in depressive symptoms and then immediately offer EMI content, such as mindfulness or ACT-based rumination journaling, based on the participant's current state.

  6. Personalized daily reports can be enhanced by using LLM-based generative AI to create dynamic visual content. The system can generate an illustrated diary image generated by ChatGPT-4o based on their diary entries that includes a word of praise to promote positive self-perception and color-coded EMA severity results for immediate, tailored feedback.

Abstract

Background: Despite the clinical effectiveness of digital interventions for young adults with depression, low engagement and adherence remain persistent challenges. Building a strong digital therapeutic alliance has been proposed to address these barriers. This study highlights the need for a conversational therapeutic companion agent (TCA)-based intervention design. Objective: This study aimed to develop a Wizard-of-Oz TCA-centered prototype integrating social-support-based ecological momentary assessment (EMA), ecological momentary intervention (EMI), behavioral activation, and gamification. We evaluated the six-week proof-of-concept efficacy of this intervention among young adults with depressive symptoms. Methods: Korean young adults aged 20--39 years with mild-to-moderate depressive symptoms (PHQ-9) were recruited online. The intervention group (n = 29) received a six-week TCA-based digital intervention, while the control group (n = 29), recruited four weeks later, continued their usual routines. The TCA guided four daily behavioral-activation tasks, three mood assessments, meditation, daily summaries, and weekly mission feedback. Both groups were assessed at baseline and at weeks 2, 4, and 6 using the BDI-II, GAD-7, and Q-LES-Q-SF. Results: Of 58 participants, 57 completed the study (one dropout in the intervention group). At week 6, the intervention group showed significantly greater reductions in depressive symptoms and improvements in quality of life than controls. Adherence was 78% for EMA, 51% for EMI, and 65% for daily routines. Conclusions: The TCA-based digital intervention improved depressive symptoms and quality of life with adherence levels comparable to previous digital health interventions. Future studies should refine the TCA design and conduct larger-scale evaluations.

Sources

Related papers