Feasibility of a self-compassion-based virtual avatar therapy for patients with major depressive disorder: a pilot study

1 Introduction

Major depressive disorder (MDD) is a common mental disorder with a global 12-month prevalence of up to 4.4% (Herrman et al., 2022). It is a leading cause of disability and has been identified by the World Health Organization as a major contributor to the global burden of disease (World Health Organization, 2025). Psychotherapy and antidepressant medication are the main treatment modalities for MDD. Psychotherapy has at least moderate efficacy in improving depressive symptoms, while antidepressant medications demonstrate small-to-moderate efficacy (Simon et al., 2024). Psychotherapy is the preferred treatment for most patients and clinicians. The latest Mental Health Gap Action Programme guideline by the World Health Organization recommends psychotherapy, rather than medication, as the first-line treatment for common mental disorders (WHO, 2023). However, despite clear guideline recommendations, the utilization of mental health resources remains critically insufficient. Even in high-income countries, less than 20% of patients with MDD receive effective care; this figure drops below 10% in low- and middle-income countries. Underutilization of mental health resources has become a core treatment bottleneck (Vigo et al., 2020). The reasons include high treatment costs, lengthy intervention cycles, and a severe global shortage of mental health professionals, which collectively prevent a large number of patients with MDD from accessing standardized psychological treatment services (Wang et al., 2007). Therefore, there is an urgent need to develop innovative approaches that can enhance the accessibility, engagement, and delivery of evidence-based psychotherapy while maintaining therapeutic quality. Digital health technologies have emerged as promising tools to support, rather than replace, conventional psychotherapy (Torous et al., 2021).

Among these approaches, virtual reality (VR) has attracted increasing attention because it can create immersive, interactive therapeutic environments that complement conventional psychotherapy. Rather than replacing therapists, VR-based interventions may enhance patient engagement, facilitate experiential learning, and provide standardized therapeutic experiences (Aboujaoude et al., 2020). First applied in medical pain intervention in the early 2000s, VR technology rapidly expanded into the treatment of mental disorders (Hoffman et al., 2000a; Hoffman et al., 2000b; Emmelkamp and Meyerbröker, 2021). Compared to VR intervention research for anxiety and post-traumatic stress disorder, explorations applying VR to MDD treatment remain relatively scarce, with intervention models and efficacy validation still in preliminary stages (Baghaei et al., 2021). However, existing research has confirmed that VR technology can serve as an effective adjunct to traditional psychotherapy by optimizing the treatment experience and lowering training thresholds (Lindner et al., 2019).

Within the broader field of VR-assisted psychotherapy, virtual avatar therapy represents a particularly promising approach because it enables patients to interact with personalized virtual characters in emotionally meaningful therapeutic scenarios. It is an important branch of VR psychotherapy (Cao et al., 2026). Virtual avatars have three potential applications in psychotherapy: assessing patients’ self-representation, promoting behavioral change through alternative self-images, and training therapists’ diagnostic and assessment skills. Despite these potential values, the use of virtual avatars in psychotherapy remains limited (Ciarmoli et al., 2025).

Virtual avatar therapy was first applied in 2013 to treat auditory hallucinations in schizophrenia (Leff et al., 2013). In recent years, the evidence base for virtual avatar therapy has continuously accumulated. Research has confirmed the efficacy of this approach: by having patients drive avatars to conduct ‘counter-argument training’ against the source of hallucinations, the frequency of and distress caused by auditory hallucinations can be significantly reduced, with effects remaining stable during follow-up (Craig et al., 2018; Hsu et al., 2025). Subsequently, this therapy was extended to other mental disorders. For example, in exposure treatment for social anxiety disorder, after completing graded exposure training interacting with avatars in virtual social scenarios, patients’ social avoidance behaviors improved (Peperkorn et al., 2026).

In MDD intervention, virtual avatar therapy has developed two main technical approaches. The first is dialogic intervention, which involves interacting with avatars representing sources of depression (such as an inner critic) to achieve emotional catharsis and cognitive restructuring. For example, the ConVRSelf application combines the empty-chair technique with VR, allowing participants to alternate between two avatars and engage in self-dialogue (Zisquit et al., 2025). Results showed long-term improvements in emotion regulation abilities and reductions in depression, anxiety, and stress symptoms (Zisquit et al., 2025). Kocur et al. (2021) targeted patients’ dysfunctional beliefs by having participants face a virtual avatar expressing their personal beliefs, and required participants to refute these views and express alternative functional beliefs-an approach similar to classic hallucination counter-argument training-resulting in reduced symptom severity (Kocur et al., 2021).

The second approach is the self-compassion paradigm, pioneered by Falconer and colleagues who creatively applied VR technology to self-compassion training (Falconer et al., 2014). Through “adult-child” avatar role transitions, patients shift from “compassion giver” to “compassion receiver,” transforming abstract self-compassion experiences into immersive virtual interactions, thereby alleviating depressive symptoms by enhancing self-compassion levels.

Although these studies preliminarily confirm the potential of virtual avatar therapy, existing explorations remain mostly single-center, small-sample investigations lacking standardized intervention protocols and in-depth analysis of efficacy maintenance mechanisms and population suitability. Furthermore, technical optimization of virtual avatar therapy continues to advance. Factors such as motion capture accuracy, naturalness of avatar expressions, and personalized adaptation of virtual scenes have all been shown to be significantly correlated with treatment experience and clinical efficacy (Birk and Mandryk, 2019; Rogers et al., 2022; Piette et al., 2024), laying the technical foundation for further application of this therapy in MDD treatment.

Self-compassion, conceptualized by Neff, refers to the way individuals relate to themselves with warmth and understanding during times of suffering, perceived inadequacy, or failure, rather than engaging in self-criticism. It involves recognizing that painful experiences are part of shared human experience and holding painful thoughts and feelings in mindful awareness (Neff, 2023). Patients with MDD characteristically exhibit low self-esteem, excessive self-criticism, and prominent feelings of guilt. Their self-compassion levels are significantly lower than those in healthy populations, and low self-compassion is positively correlated with the severity of depressive symptoms (Raes, 2011; MacBeth and Gumley, 2012). Self-compassion-based psychological interventions, by cultivating self-kindness, common humanity, and mindfulness, can significantly reduce depression and anxiety levels while improving quality of life, self-efficacy, and subjective well-being, establishing them as an important direction in MDD psychotherapy (Kirby et al., 2017; López et al., 2018; Holzer et al., 2024).

However, traditional self-compassion training is often constrained by the techniques used, typically relying on imagination, meditation, empty-chair work, and role-playing. These approaches typically require individuals to internally generate vivid compassionate mental imagery and sustain attention throughout the exercise. Patients with MDD commonly experience cognitive impairments, including deficits in attention, slowed information processing, and impaired executive functioning (Varghese et al., 2022; Rhee et al., 2024). Together with evidence suggesting reduced vividness and emotional engagement in positive mental imagery (Holmes et al., 2016; Ji et al., 2017), these characteristics may make it more challenging for some individuals to engage in imagery-based self-compassion exercises.

Virtual reality (VR) offers a potential solution by externalizing compassionate interactions into immersive and embodied experiences. Rather than requiring patients to generate vivid compassionate imagery entirely through imagination, VR provides concrete sensory and interpersonal experiences that may facilitate the acquisition of self-compassion skills. Importantly, the purpose of VR is not to permanently replace mental imagery but to serve as an experiential scaffold, enabling patients to establish vivid compassionate memories that can subsequently be recalled and applied in everyday life. In this way, VR may reduce reliance on spontaneous imagery generation during the initial learning process while facilitating the later transfer of self-compassion skills beyond the virtual environment. Accordingly, the goal of subsequent homework practice is not to generate entirely new compassionate imagery but to recall and generalize the compassionate experiences established during VR training.

Building upon this rationale, Falconer et al. (2014) first developed a VR-based self-compassion paradigm based on adult-child avatar role transitions. Participants first expressed compassion toward a child avatar from the perspective of an adult avatar and subsequently embodied the child avatar to receive the compassion they had previously delivered, thereby transforming self-compassion into an immersive embodied experience. In 2016, the team applied this paradigm in treating 15 patients with MDD, showing significant remission of depressive symptoms and increased self-compassion levels, providing preliminary evidence supporting the clinical potential of VR-based self-compassion training (Falconer et al., 2016).

Nevertheless, existing studies have primarily evaluated VR self-compassion training as a standalone experiential technique. Relatively little attention has been paid to integrating VR into a structured psychotherapeutic framework that systematically combines psychoeducation, therapist guidance, experiential learning, and skill transfer to daily life. Without such integration, opportunities to consolidate compassionate experiences through therapist-guided reflection and behavioral practice may be reduced, which may limit the potential transfer of VR experiences into daily self-compassion practices. This possibility, however, has not yet been directly examined and warrants further empirical investigation.

To address these limitations, we developed the Computer Avatar Therapy System (CATS), which integrates VR-assisted self-compassion training into a structured therapist-guided psychotherapeutic framework designed to facilitate experiential learning during VR sessions and support the subsequent application of self-compassion skills in everyday life.

2 Methods

2.1 Study design

This study was a prospective single-arm feasibility pilot study. Feasibility was conceptualized as a multidimensional construct encompassing recruitment, retention, intervention completion, adherence, participant acceptability, and practical implementation. Because the primary aim of this study was to evaluate feasibility and acceptability rather than to establish treatment efficacy, a formal a priori sample size calculation was not performed. Instead, the target sample size was determined with reference to similar domestic and international VR psychotherapy pilot studies (Colombo et al., 2022; Paul et al., 2022; Kouijzer et al., 2026) and was considered appropriate for obtaining preliminary feasibility estimates and informing the design of future randomized controlled trials. Participants underwent a 6-week stepped virtual avatar therapy intervention using the self-developed CATS system, consisting of one 60-min session per week. Follow-up assessments were conducted 3 months after treatment completion. During the follow-up period, participants continued their original antidepressant medication regimens, and no additional psychotherapy or study-related psychological intervention was provided. The study was approved by the Ethics Committee of the First Hospital of Shanxi Medical University ([2022] Ethics Review No. K039). All participants provided written informed consent. This pilot feasibility study was conducted independently from a subsequent randomized controlled trial evaluating the same intervention framework. The pilot study was not prospectively registered. The subsequent randomized controlled trial was prospectively registered at the Chinese Clinical Trial Registry (ChiCTR2200063483; registered 8 September 2022). Although the two studies used the same intervention protocol and eligibility criteria, they involved independent recruitment procedures, participant samples, and datasets.

The reporting of this study was informed, where applicable, by the CONSORT extension for randomized pilot and feasibility trials, while recognizing that the present study used a prospective single-arm design.

Participants were recruited between September 2022 and February 2023 at the First Hospital of Shanxi Medical University.

2.2 Feasibility assessment

Feasibility was evaluated using multiple implementation indicators, including recruitment, retention, treatment completion, adherence, participant acceptability, and practical implementation. Twelve participants were enrolled. Intervention completion was defined as completion of all six intervention sessions, whereas post-treatment assessment completion was defined as completion of the assessment immediately after intervention. Follow-up retention was defined as completion of the 3-month follow-up assessment. These indicators were calculated separately to distinguish intervention adherence from follow-up retention. Intervention adherence was evaluated according to session completion and scheduling adherence.

All participants who completed the study attended all six scheduled intervention sessions. Session rescheduling was permitted when necessary; three participants rescheduled one session and two participants rescheduled two sessions, with all schedule adjustments occurring within 1 week and without disrupting the overall intervention process.

Participant acceptability of the VR component was assessed using the Virtual Reality Experience Questionnaire (VREQ), which evaluates subjective VR experiences including presence, embodiment, agency, and perceived comfort. The VREQ was considered an indicator of VR-related acceptability and user experience rather than a standalone measure of overall feasibility. Homework completion was discussed qualitatively during therapist sessions but was not formally quantified. No technical problems requiring interruption, modification, or discontinuation of the intervention were encountered during the study period. Given the exploratory nature of this pilot study and the small sample size, internal consistency analyses were not performed.

The number of individuals approached, screened, or excluded before enrollment was not prospectively recorded; therefore, recruitment feasibility was evaluated based on the number of enrolled participants and study completion indicators. Because this was an exploratory pilot study, no prespecified progression criteria for feasibility were established.

2.3 Participants

Participants were recruited from the outpatient clinic of the Department of Mental Health, First Hospital of Shanxi Medical University. The recruitment process was as follows:

Interested participants completed the Beck Depression Inventory-II (BDI-II) for initial screening; those with scores ≥14 proceeded to the next step.

Two standardized trained psychiatrists (attending physician level or above) performed diagnostic screening using the Structured Clinical Interview for DSM-5 (SCID-5) to confirm diagnosis and ensure consistency.

Researchers thoroughly informed potential participants about the risks, benefits, and their right to voluntarily withdraw. Participants were enrolled after signing the informed consent form. Clinical outcome assessments were conducted at baseline, post-treatment, and 3-month follow-up.

Inclusion criteria: age 18–65, right-handed, diagnosis confirmed by two psychiatrists using the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), and either no antidepressant treatment or a stable SSRI regimen for at least 4 weeks before enrollment. Exclusion criteria: severe suicidality (BDI-II suicide item score ≥2), comorbid psychiatric disorders, serious physical illness, visual/auditory impairment, recent psychotherapy/physical treatments, substance abuse, pregnancy.

2.4 Intervention

A total of six intervention sessions were delivered once weekly, with each session lasting approximately 60 min. The intervention followed a structured and progressively therapeutic framework. Session 1 focused on establishing the therapeutic relationship, assessing participants’ depressive symptoms and self-critical patterns, introducing the concept of self-compassion, and defining treatment goals. Sessions 2–5 constituted the core virtual avatar self-compassion training. Although the VR procedure remained consistent across these sessions, including adult-avatar compassion expression followed by child-avatar compassion receiving, the therapeutic focus progressively developed across sessions. Session 2 focused on identifying and challenging self-critical beliefs; Session 3 emphasized cultivating a compassionate inner voice and self-compassionate responses; Session 4 focused on processing difficult emotions and exploring personal values; and Session 5 emphasized applying self-compassion strategies in daily life. Session 6 focused on reviewing the treatment experience and developing a long-term self-compassion maintenance plan. A summary of the therapeutic focus, main components, VR activities, and homework assignments for each session is provided in Supplementary Table S1.

Each of the four core intervention sessions (Sessions 2–5) followed a standardized three-stage structure. The first stage (approximately 20 min) consisted of therapist-guided review and preparation. During this stage, the therapist reviewed the participant’s progress since the previous session, discussed homework completion, and introduced concepts and exercises tailored to the therapeutic objectives of the current session. Before each VR exercise, the therapist and participant rehearsed a standardized three-step compassionate response framework comprising validation, attentional redirection, and activation of a supportive interpersonal memory. Representative therapist prompts and compassionate statements are provided in Supplementary Methods S1.

The second stage (approximately 20 min) comprised the VR-based virtual avatar intervention using the Computer Avatar Therapy System (CATS). This stage included two sequential components: (1) an adult-avatar phase, during which participants practiced self-compassion from the perspective of their adult self, and (2) a child-avatar phase, during which participants experienced receiving compassion from their adult avatar while embodying their childhood self.

The final stage (approximately 20 min) consisted of therapist-guided debriefing and integration. Participants discussed their experiences during the VR intervention, reflected on the emotions and insights elicited during the session, and received homework assignments designed to facilitate the transfer of self-compassion skills into daily life. Specifically, homework assignments encouraged participants to recall and apply the compassionate experiences established during the VR sessions to real-life situations. Therefore, the VR intervention represented one structured component within a broader therapist-guided psychotherapeutic session rather than the entirety of the intervention.

The overall structure of the intervention is illustrated in Figure 1.

Throughout the intervention period, participants continued routine outpatient psychiatric care. Treating psychiatrists independently evaluated participants’ clinical conditions and determined the need for pharmacological treatment. When clinically indicated, SSRI antidepressants (paroxetine, sertraline, fluvoxamine, fluoxetine, citalopram, and escitalopram) could be initiated or adjusted according to clinical judgment and approved prescribing guidelines. Short-term use of benzodiazepine sedative-hypnotics was permitted for insomnia symptoms for no longer than 2 weeks.

To minimize potential confounding effects of concurrent treatments, the use of other psychotropic medications, including antipsychotics, traditional Chinese medicine, health products, nutritional supplements, as well as additional psychotherapy or physical interventions, was prohibited during the intervention period. All medication initiation or adjustments were documented by the research team. Participants requiring medication changes outside the predefined protocol (e.g., initiation of antipsychotic medication) were withdrawn from the study.

The intervention was delivered by three psychotherapists with more than 5 years of clinical psychotherapy experience. Before study initiation, all therapists received standardized training based on the intervention manual, including the theoretical framework of the self-compassion-based virtual avatar intervention, operation of the Computer Avatar Therapy System (CATS), session procedures, standardized therapeutic procedures, and management of emotional responses during VR exposure. Participants were randomly assigned to therapists, with Therapist A, Therapist B, and Therapist C treating 3, 3, and 4 participants, respectively.

Throughout the study, therapists delivered the intervention according to the standardized treatment manual and predefined intervention procedures. Regular case discussions were conducted among therapists to promote consistency in treatment delivery and address potential discrepancies during implementation. Formal treatment fidelity assessments, independent fidelity ratings, or systematic review of session recordings were not conducted in this pilot study.

To ensure participant safety throughout the intervention, therapists assessed participants’ recent clinical status at the beginning of each session by asking about their experiences during the preceding week, including changes in mood, suicidal ideation, and psychological distress. Before entering the VR environment, participants were informed about the emotional content of the virtual scenario, including the presence of a distressed child avatar, and practiced compassionate responses with therapists to enhance familiarity with the task. During VR exposure, participants were informed that they could request interruption at any time by raising their right hand. Therapists continuously observed participants’ responses and could pause or terminate the VR procedure if significant emotional distress or difficulty continuing the exercise was observed.

2.5 CATS and training paradigm

2.5.1 CATS system design

The CATS system was developed based on the Unity engine, programmed in C#, and compiled using the Microsoft. NET Framework. It runs on the Windows operating system and is a VR psychotherapy system integrating interactive avatar control, display, and management. The system includes standardized 3D avatar models (incorporating skeletal system assembly and facial expression BlendShape creation). It receives real-time axial motion data from inertial motion capture devices via TCP/IP protocol to drive the virtual character’s skeleton for synchronized movement. It is compatible with the OpenVR interface, enabling full 3D scene viewing and experience through VR head-mounted displays. The system features a visual interface design, hiding program execution logic and compatible with multiple data interfaces to simplify user operation. The motion capture device used in this study is the Noitom Perception Neuron 3 Pro, and the VR headset is the HTC Vive Cosmos. Some operational interfaces of the CATS system are shown in Figure 2. The embodied role-transition process used during each core VR session is illustrated in Supplementary Figure S2.

The figure illustrates the main functional modules of CATS, including virtual environment selection, avatar customization, and interaction settings used during therapist-guided virtual avatar self-compassion training.

2.5.2 Localization optimization of the CATS system

The CATS system incorporates multi-dimensional optimizations based on Falconer and colleagues’ VR self-compassion paradigm:

Support for personalized avatar customization, allowing patients to choose a virtual model resembling themselves or have one customized by engineers based on their requirements.

Virtual scenes relevant to daily life in China (living room, classroom, street), which patients can select independently.

Addition of a movable virtual mirror to enhance visuomotor synchrony, improving patients’ sense of immersion and body agency.

Equipping a virtual stool with adjustable height and position to accommodate the postural needs of seated training, enhancing patient comfort during sessions.

2.5.3 CATS self-compassion training paradigm

The training consists of four stages, all conducted in a seated position:

• Adult Avatar Adaptation Stage: The patient enters the chosen virtual environment and controls the adult avatar to perform simple actions (e.g., raising a hand, turning around). They observe the avatar’s movements in sync with their own via the virtual mirror, becoming familiar with system operation and establishing a sense of body agency and immersion.
• Compassion Giving Stage: A crying child avatar of the same gender as the patient appears opposite them (height 1.2 m, initial distance 1.5 m from the adult avatar). The patient is guided to comfort the child avatar using compassionate language and gestures. The child avatar has three pre-set emotional states (curled up and crying heavily, crying while bent over, looking at the patient and sobbing). The therapist manually switches between these states according to predefined intervention procedures and the participant’s progression in expressing compassionate responses. Simultaneously, the actions and speech of the adult avatar are recorded.
• Child Avatar Adaptation Stage: The patient switches to control the child avatar, performing simple actions and observing themselves in the virtual mirror to become familiar with the child avatar’s operation and perspective.
• Compassion Receiving Stage: The patient, controlling the child avatar, sits on the virtual stool. The recording from the second stage is played back. The patient, now from the child avatar’s perspective, receives the compassion they themselves (as the adult avatar) had just expressed. This facilitates the role transition from “compassion giver” to “compassion receiver,” completing the immersive self-compassion experience.

Avatar selection was standardized but allowed limited personalization. Participants selected their adult avatar according to their personal preference from the available avatar options. For the child-avatar phase, participants selected one of two standardized child avatars (male or female), with female participants selecting the female child avatar and male participants selecting the male child avatar. The child avatar was not designed to replicate participants’ actual childhood appearance but served as a symbolic representation of their younger self during the compassion-receiving phase. The same child avatar was used throughout Sessions 2–5, while emotional responses of the child avatar were adjusted according to predefined system settings and participants’ compassionate interactions.

2.6 Measures

2.6.1 Clinical outcome measures

Beck Depression Inventory-II (BDI-II) (Beck et al., 1996): A self-report scale assessing the severity of depressive symptoms. It comprises 21 items rated on a 4-point scale from 0 to 3. Higher scores indicate more severe depressive symptoms. Scores of 0–13 are generally interpreted as indicating minimal depressive symptom severity.

Self-Compassion Scale (SCS) (Neff, 2003): Reflects the participant’s level of self-compassion. It consists of 26 items across six subscales: Self-Kindness, Self-Judgment, Common Humanity, Isolation, Mindfulness, and Over-Identification. Items are rated on a 5-point Likert scale from 1 to 5. Subscale scores are calculated, and a total score is derived, with Self-Judgment, Isolation, and Over-Identification items reverse-scored. The total score represents an individual’s overall level of self-compassion. The total SCS score ranges from 26 to 130, with higher scores indicating greater self-compassion. For the negative subscales (self-judgment, isolation, and over-identification), items were reverse-scored before calculating the total score.

Depressive Experience Questionnaire-Introjective (DEQ-I) (Yao et al., 2009): Reflects the patient’s level of self-criticism. It comprises 15 items rated on a 7-point scale from “strongly disagree” to “strongly agree.” Higher scores indicate a greater degree of self-criticism.

2.6.2 Virtual reality experience measure

VR-related subjective experiences were assessed using a multidimensional VR experience assessment questionnaire (VREQ) adapted from the virtual embodiment paradigm developed by Falconer et al. The measure was used to explore participants’ subjective experiences during the immersive VR procedure, including presence, embodiment-related experiences, avatar identification, and emotional experiences associated with receiving compassion.

The assessment included five experiential domains: presence, adult-avatar body ownership and agency, child-avatar body ownership and agency, recognition of the adult avatar as oneself during child-avatar embodiment, and perceived comfort while embodied as the child avatar. The body ownership and agency-related items were adapted from previous VR embodiment studies, whereas the remaining items were developed to capture intervention-specific experiential components of the self-compassion-based virtual avatar paradigm.

Items were rated on a 7-point scale ranging from −3 (“not at all”) to +3 (“very much so”). Scores for each domain were calculated by averaging the corresponding items, with higher scores indicating stronger subjective experiences.

The original items were presented in Chinese for participant administration. The wording was reviewed by clinical researchers with experience in VR-based psychotherapy to improve clarity and contextual appropriateness. However, no formal forward translation, back-translation, linguistic validation, or psychometric validation of the Chinese version was conducted in this pilot study.

Therefore, the VREQ was used as an exploratory characterization measure of participants’ subjective VR experiences rather than as a validated outcome instrument. Detailed descriptions of the assessed domains, example items, scoring procedures, and conceptual sources are provided in Supplementary Table S2.

2.6.3 Tolerability assessment tool

Simulator sickness was assessed using the validated Chinese version of the Simulator Sickness Questionnaire (SSQ; Kennedy et al., 1993). The questionnaire consists of 16 symptoms, with each symptom rated on a 4-point scale ranging from 0 (none) to 3 (severe).

In this pilot feasibility study, SSQ responses were summarized using an item-level descriptive approach to characterize the tolerability of the VR intervention. Although the standard SSQ provides composite scores for Nausea, Oculomotor, Disorientation, and Total Severity Score, these scores were not calculated in the present study because of the small sample size and the exploratory feasibility focus. Instead, analyses focused on the frequency and proportion of participants reporting individual symptoms.

SSQ symptoms were summarized at the participant level across the four VR sessions. For each symptom, the percentage was calculated as the proportion of participants (N = 10) who reported that symptom at least once during the intervention. If the same participant reported the same symptom during multiple VR sessions, it was counted only once. For each symptom, severity classification was based on the maximum severity level reported by the participant across all VR sessions.

2.6.4 Outcome measures

Outcome assessments were conducted at three time points: baseline (before the intervention), immediately after completion of the six-session intervention, and at the 3-month follow-up. The Beck Depression Inventory-II (BDI-II), Self-Compassion Scale (SCS), and Depressive Experiences Questionnaire (DEQ-I) were administered at all three assessment points. In contrast, the Virtual Reality Experience Questionnaire (VREQ), including measures of embodiment and presence, and the Simulator Sickness Questionnaire (SSQ) were administered immediately after each of the four VR intervention sessions (Sessions 2–5) to evaluate participants’ subjective VR experiences and potential simulator-related adverse effects during VR training.

2.7 Statistical analysis

All statistical analyses were performed using SPSS version 26.0 software (IBM Corp., Armonk, NY, United States). Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range), as appropriate. The Shapiro–Wilk test was used to assess distributional characteristics. Given the exploratory nature of this pilot study, normality testing was used as a pragmatic guide for selecting the statistical model rather than as a definitive criterion for statistical inference.

Normally distributed variables were analyzed using repeated-measures analysis of variance (ANOVA). For repeated-measures ANOVA, the assumption of sphericity was assessed using Mauchly’s test, and the Greenhouse–Geisser correction was applied when the assumption of sphericity was violated. Effect sizes were reported as partial eta squared (ηp²) (small effect: 0.01 ≤ ηp² < 0.06; medium effect: 0.06 ≤ ηp² < 0.14; large effect: ηp² ≥ 0.14). Non-normally distributed variables were analyzed using the Friedman test. The specific statistical test used for each outcome is indicated in the corresponding tables.

A two-sided P < 0.05 was considered statistically significant. Bonferroni correction was applied to pairwise comparisons among assessment time points within each outcome and among repeated measurements when post hoc comparisons were performed. No additional multiplicity adjustment was performed across the BDI-II, SCS, and DEQ-I because these measures were not analyzed as independent confirmatory primary endpoints. Instead, results were interpreted collectively and descriptively within the exploratory framework of this pilot study.

Exploratory Spearman rank correlation analyses were performed to examine associations between treatment-related changes and VR experience variables. Improvement in depressive symptoms was defined as the reduction in BDI-II scores from baseline to post-treatment (baseline minus post-treatment), whereas improvement in self-compassion was defined as the increase in SCS scores (post-treatment minus baseline). Because the Virtual Reality Experience Questionnaire consists of conceptually distinct dimensions rather than a validated total embodiment score, exploratory analyses focused on Adult Body Ownership and Agency and Child Body Ownership and Agency. Mean scores across the four VR training sessions were calculated for each dimension and correlated with improvement in depressive symptoms. These correlation analyses were considered hypothesis-generating and were interpreted cautiously.

3 Results

3.1 Participant characteristics and flow

A total of 12 patients with MDD meeting the inclusion and exclusion criteria were recruited. Of these, 11 participants completed the 6-week self-compassion-based virtual avatar therapy intervention and post-treatment assessment. Ten participants completed the additional 3-month follow-up assessment and were included in the final analysis. Two participants did not complete the full study protocol. One participant discontinued participation after the second intervention session because of a clinically significant change in mental status occurring during the week following the session, which required additional pharmacological management after psychiatric evaluation. Quetiapine was subsequently initiated for clinical management. As antipsychotic medication was not permitted according to the study protocol, the participant was withdrawn from the study. The second participant completed all six intervention sessions but was unable to complete the 3-month follow-up assessment because of relocation for education. Detailed information regarding the characteristics, intervention completion status, available outcome data, and reasons for incomplete study participation is provided in Supplementary Table S3. Participant flow and reasons for incomplete study participation are summarized in Supplementary Figure S1.

Among the 10 patients who completed the study, there were 2 males and 8 females. Mean age was 27.90 (SD 5.99) years, mean years of education was 15.10 (SD 1.29) years, mean illness duration was 10.70 (SD 10.82) months, and mean age of onset was 26.90 (SD 5.82) years. Regarding marital status, 6 were unmarried, 3 were married, and 1 was divorced. All were first-episode patients with MDD.

Regarding medication use, 7 patients did not use any psychotropic medications throughout the intervention. Two patients took a stable dose of sertraline 50 mg/day throughout, and 1 patient took escitalopram oxalate 20 mg/day combined with oxazepam 15–7.5 mg (oxazepam was discontinued after 10 days). No patient adjusted their core antidepressant dosage or used other psychological or physical interventions during the study period. Specific demographic and clinical characteristics of the participants are shown in Table 1.

TABLE 1. Characteristics of the study sample (N = 10).

3.2 Feasibility and VR experience

A total of 12 participants were enrolled. Eleven participants completed the six-session intervention and post-treatment assessment, corresponding to an intervention completion rate of 91.7% (95% CI, 64.6%–98.5%). Ten participants completed the 3-month follow-up assessment, corresponding to a follow-up retention rate of 83.3% (95% CI, 55.2%–95.3%). All participants who completed the intervention attended all six scheduled sessions.

Regarding intervention implementation, session scheduling adjustments occurred in five participants (three participants rescheduled one session and two participants rescheduled two sessions). All rescheduled sessions were completed within 1 week and did not affect the planned intervention schedule. No technical problems leading to session interruption or discontinuation were reported.

Regarding VR tolerability, six of the ten participants who completed the intervention (60.0%; 95% CI, 31.3%–83.2%) reported at least one simulator sickness symptom across the four VR sessions. All reported symptoms were mild to moderate, and no participant discontinued the intervention because of VR-related discomfort. Detailed simulator sickness symptoms and severity are presented in Table 2.

TABLE 2. Participant-Level Frequency and severity of simulator sickness Symptoms Reported Across VR Sessions (N = 10, %).

Symptoms were summarized at the participant level across the four four VR sessions. Repeated reports of the same symptom by the same participant were counted once. Severity classification was based on the maximum severity level reported for each symptom by each participant.

Homework adherence was encouraged and reviewed during therapist sessions; however, it was not systematically quantified in this pilot study.

Exploratory analyses of the Virtual Reality Experience Questionnaire scores collected after the four virtual avatar self-compassion training sessions (Sessions 2–5) revealed variations in participants’ subjective VR experiences across sessions. Several experiential dimensions showed higher scores in later sessions compared with earlier sessions; however, these differences should be interpreted cautiously, as they may reflect not only changes in the quality of the VR experience but also increased familiarity with the virtual environment, repeated exposure to the intervention procedures, and adaptation to the avatar-based tasks.

Exploratory pairwise comparisons indicated that scores for Child Body Ownership and Agency, and Feels comforted as child were higher in Session 4 than in Session 2 (all P < 0.05). In addition, Adult Body Ownership and Agency and Feels comforted as child showed higher scores in Session 5 compared with Session 3 (P < 0.05). No statistically significant change was observed in the Presence and Recognise Adult as Self dimension across sessions (P > 0.05). Given that these analyses were exploratory and intended to characterize participants’ VR experiences during the intervention rather than to test predefined hypotheses, the findings are presented descriptively. Detailed Virtual Reality Experience Questionnaire scores and statistical results are provided in Table 3.

TABLE 3. Means (SD) of virtual reality experience scores across four training sessions (N = 10).

Presence was calculated as the mean score of 1 item; Adult Body Ownership and Agency and Child Body Ownership and Agency were each calculated as the mean score of 2 items; Recognise Adult as Self was calculated as the mean score of 3 items; and Feels comforted as child was calculated as the mean score of 3 items. Superscript letters indicate significant differences based on Bonferroni-adjusted pairwise comparisons: a indicates a significant difference compared with Session 2, and b indicates a significant difference compared with Session 3. Exact adjusted P values for pairwise comparisons are provided in Supplementary Table S4.

3.3 Tolerability

During the intervention period, therapists monitored participants’ clinical status at each session by assessing changes in mood, suicidal ideation, psychological distress, and other clinically relevant concerns since the previous visit. One participant experienced a clinically significant change in mental status during the week following Session 2 and required additional pharmacological management after psychiatric evaluation. This event was documented and reviewed by the treating psychiatrists and research team as part of the safety-monitoring procedures. Because quetiapine was initiated and was not permitted according to the predefined medication protocol, the participant was withdrawn from the study. Based on clinical evaluation and the absence of VR-related distress or intolerance during the session, the event was not considered related to VR exposure. No participant required interruption or termination of a VR session because of emotional distress or safety concerns.

Regarding VR tolerability, SSQ responses from the 10 participants across the four VR sessions were summarized at the participant level. A total of 19 participant-level symptom endorsements were identified, representing distinct VR-related discomfort symptoms reported by participants during the intervention. Six of the 10 participants (60.0%; 95% CI, 31.3%–83.2%) reported at least one simulator sickness symptom. Repeated reports of the same symptom by the same participant across different VR sessions were counted only once. All reported symptoms were transient and resolved after a brief rest, and no participant discontinued or withdrew from the intervention because of VR-related discomfort.

Among the 19 participant-level symptom endorsements, 16 were classified as mild (84.2%) and 3 were classified as moderate (15.8%) based on the maximum severity level reported for each symptom. The most frequently reported mild symptoms were fatigue (3 participants), dizziness with eyes open (2 participants), dizziness with eyes closed (2 participants), and general discomfort (2 participants). Other mild symptoms, including eyestrain, difficulty focusing, nausea, difficulty concentrating, blurred vision, stomach awareness, and vertigo, were reported less frequently. Moderate symptoms included eyestrain, sweating, and vertigo (one participant each). No severe symptoms were reported.

At the participant level, four participants reported no simulator sickness symptoms throughout the intervention, five participants reported one to four symptom types, and one participant reported nine symptom types. The participant reporting multiple symptoms gradually adapted to the VR environment during subsequent sessions, and these symptoms did not interfere with treatment completion. Overall, VR-related discomfort was transient and did not result in treatment discontinuation. Detailed symptom frequencies and participant-level reporting are presented in Table 3.

3.4 Preliminary clinical outcomes

Descriptive analyses were conducted for BDI-II, SCS, and DEQ-I scores at baseline, post-treatment, and the 3-month follow-up. Scores on all three clinical outcome measures showed changes in the expected direction over time. Exploratory analyses identified significant within-participant differences between assessment points. Post-treatment and follow-up scores did not differ significantly, although DEQ-I scores showed a further descriptive decline during follow-up. Given the uncontrolled exploratory design, these findings should be interpreted as preliminary within-participant changes rather than evidence of treatment efficacy. Detailed results are presented in Table 4.

TABLE 4. Means (SD) of clinical outcome scores at baseline, post-treatment, and 3-month follow-up (N = 10).

Values are presented as mean (SD). Repeated-measures ANOVA, results are reported with Greenhouse–Geisser corrected degrees of freedom when the assumption of sphericity was violated. ηp² indicates partial eta-squared. Friedman test effect size is reported as Kendall’s W. Superscript letters indicate significant differences based on Bonferroni-adjusted pairwise comparisons: a indicates a significant difference compared with baseline, and b indicates a significant difference compared with post-treatment. Exact adjusted P values for pairwise comparisons are provided in Supplementary Table S5.

Changes in Depressive Symptoms: The mean BDI-II score at baseline was 26.60 (SD 6.04). This significantly decreased to 12.70 (SD 6.33) post-treatment (P < 0.001), with a mean reduction of 13.90 points. At the 3-month follow-up, the mean score was 12.30 (SD 8.78), showing no significant difference from the post-treatment score (P > 0.05), with symptom levels remaining comparable to those observed at post-treatment. Six of the ten participants (60%) had BDI-II scores at or below the threshold for minimal depressive symptoms at the post-treatment assessment. Individual trajectories of BDI-II scores across the three assessment time points are presented in Figure 3, illustrating individual patterns of symptom change in depressive symptoms and the variability in individual treatment responses. The repeated-measures ANOVA indicated a large exploratory within-participant time effect (ηp² = 0.848), indicating substantial within-participant changes across the three assessment time points.

Each line represents one participant’s BDI-II score trajectory across the three assessment points, illustrating individual patterns of symptom change over time.

Changes in Self-Compassion Levels: The mean total SCS score at baseline was 55.20 (SD 14.99). This significantly increased to 92.40 (SD 10.87) post-treatment (P < 0.001) and further increased to 94.20 (SD 17.19) at follow-up, with no significant difference compared with post-treatment (P > 0.05). Examining the subscales, scores for Self-Kindness, Self-Judgment (reverse-scored), Common Humanity, Isolation (reverse-scored), Mindfulness, and Over-Identification (reverse-scored) all showed significant increases post-treatment compared with baseline (P < 0.05). All subscale scores remained comparable to post-treatment levels at follow-up, with no significant differences observed. Particularly notable improvements were observed in the Isolation and Over-Identification subscales.

Changes in Self-Criticism: The mean DEQ-I score at baseline was 79.60 (SD 9.61). This significantly decreased to 61.20 (SD 8.31) post-treatment (P < 0.001). At follow-up, the mean score was 55.60 (SD 10.62), showing a further numerical decrease compared with post-treatment. The repeated-measures ANOVA yielded a large exploratory effect of time (ηp² = 0.855).

3.5 Exploratory correlation analysis

Exploratory Spearman correlation analyses were conducted to examine associations between changes in clinical outcomes and VR-related experiential measures. Changes in self-compassion were positively associated with changes in depressive symptoms (ΔSCS vs. ΔBDI-II, Spearman’s ρ = 0.875, P = 0.001). In addition, the mean Adult Body Ownership and Agency score across the four VR sessions was positively associated with changes in depressive symptoms (ρ = 0.746, P = 0.013). The mean Child Body Ownership and Agency score showed a moderate positive association with changes in depressive symptoms; however, this association did not reach statistical significance (ρ = 0.564, P = 0.089). Given the exploratory nature of these analyses, the small sample size, and the potential influence of shared method variance due to the use of concurrent self-report measures, these associations should be interpreted cautiously. They are considered hypothesis-generating findings rather than evidence of causal mechanisms. Detailed results of the exploratory correlation analyses are presented in Table 5, and the corresponding individual-level scatterplots are provided in Supplementary Figure S3.

TABLE 5. Exploratory correlations between changes in depressive symptoms and self-compassion/embodiment measures.

ΔBDI-II was calculated as baseline BDI-II minus post-treatment BDI-II. ΔSCS was calculated as post-treatment SCS minus baseline SCS. Adult Body Ownership and Agency and Child Body Ownership and Agency represent the mean scores across the four VR training sessions (Sessions 2–5). Correlations were analyzed using Spearman’s rank correlation coefficient.

4 Discussion

The present pilot study provides preliminary support for the feasibility and tolerability of the self-compassion-based virtual avatar intervention in this small sample of patients with MDD. Most participants completed the intervention, and the majority completed the 3-month follow-up assessment, no participant discontinued treatment because of VR-related adverse events, and preliminary within-participant changes were observed in depressive symptoms, self-compassion, and self-criticism. However, because this was an uncontrolled single-arm feasibility pilot study, these findings should therefore be regarded as hypothesis-generating rather than confirmatory evidence of treatment efficacy.

The feasibility of intervention delivery was supported by a high intervention completion rate (91.7%) and a 3-month follow-up retention rate of 83.3%, which is comparable to previous VR-based interventions for depression (Paul et al., 2022; Liu et al., 2025; Zhang et al., 2025) and higher than that reported for many internet-based interventions (Christensen et al., 2009). Most VR-related discomfort was mild and transient, with no serious adverse events or treatment discontinuation due to simulator-related symptoms, suggesting that the intervention was generally well tolerated in this sample. Across the four VR sessions, participants also reported progressively higher ratings on several dimensions of presence, body ownership and agency, and perceived comfort. These exploratory patterns may reflect the cumulative effects of repeated VR-assisted experiential learning throughout the intervention. Compared with previous studies that typically employed two or three avatar-training sessions (Falconer et al., 2016; Halim et al., 2023), the present study incorporated four VR training sessions, which may have provided additional opportunities for participants to engage in avatar-based self-compassion practice. A recent review likewise suggested that longer durations of VR intervention are generally associated with greater improvements in depressive symptoms (Wang et al., 2025). Nevertheless, because these analyses were exploratory and no comparison condition was included, the observed changes should not be interpreted as evidence that increasing the number of VR sessions alone improves feasibility or produces specific therapeutic effects.

Beyond the observed completion and tolerability indicators, several design characteristics of the present intervention may have contributed to its feasibility and may inform future implementation. First, the intervention was developed as a structured six-session program with predefined therapeutic objectives, standardized procedures, and a consistent session framework combining preparation, VR experiential practice, and post-session reflection. This structured format may facilitate treatment delivery consistency and reduce variability across therapists and clinical settings. Second, the VR component was embedded within therapist-guided psychotherapy rather than implemented as a standalone digital intervention. Therapist involvement before, during, and after VR exposure provided opportunities for emotional preparation, in-session support, and integration of virtual experiences into participants’ daily lives, which may improve acceptability and clinical applicability. Third, the avatar-based paradigm followed a relatively simple and reproducible experiential sequence, including adult-avatar compassion expression, perspective transition, and child-avatar compassion receiving. These clearly defined experiential components may facilitate replication and adaptation in future clinical studies.

Several theoretical mechanisms may potentially contribute to the observed changes; however, these mechanisms were not directly tested in the present study. Through repeated role switching between the adult and child avatars, participants alternated between expressing compassion and receiving compassion, thereby transforming abstract self-compassion into an embodied interpersonal experience. Rather than relying entirely on internally generated imagery, participants experienced compassion through direct interaction with their own virtual avatar, allowing VR to function as an experiential scaffold for later recall and application in daily life. Previous studies suggest that body ownership, visuomotor synchrony, and embodied perspective taking may facilitate emotional processing and self-compassion (Cebolla et al., 2019; dos Santos Kawata et al., 2025). Consistent with these theoretical accounts, our exploratory correlation analyses identified associations between changes in self-compassion, adult-avatar embodiment experiences, and changes in depressive symptoms. However, these findings should be interpreted cautiously because they do not establish causal mechanisms. Furthermore, the association between changes in SCS and BDI-II may partly reflect shared method variance because both measures were self-reported and assessed concurrently.

Beyond these theoretical mechanisms, several intervention components may represent important active ingredients that contribute to the feasibility and potential clinical value of this approach. First, the intervention combines a structured avatar-based experiential procedure with established self-compassion principles. Rather than using VR as an isolated technological exposure, the intervention embeds avatar interaction within a therapist-guided framework that includes preparation, reflection, and behavioral application exercises. Second, the standardized sequence of adult-avatar compassion expression, perspective transition, and child-avatar compassion receiving provides a reproducible therapeutic structure that can potentially be adapted across populations and cultural contexts. Third, the integration of immersive experience with therapist-guided reflection may facilitate the translation of virtual experiences into personally meaningful self-compassion practices. Importantly, the innovation of this approach lies not simply in the use of VR technology, but in leveraging immersive embodiment to enhance experiential learning within an established psychological framework. These components may provide adaptable principles for future culturally tailored interventions while preserving the underlying therapeutic processes.

Despite these encouraging preliminary findings, several limitations should be considered. First, the uncontrolled single-arm design precludes causal inference, and alternative explanations—including spontaneous symptom fluctuation, regression to the mean, expectancy effects associated with immersive VR, and nonspecific therapeutic effects arising from therapist contact—cannot be excluded. In addition, feasibility was not evaluated against prespecified progression criteria, and the observed increases in VREQ scores may partly reflect increasing familiarity with the VR environment rather than improvements in intervention feasibility alone. Second, the study included a small and relatively homogeneous sample, predominantly consisting of young female patients with first-episode MDD, which limits generalizability. Although medication regimens remained stable throughout the intervention and follow-up, the potential influence of concurrent pharmacotherapy cannot be excluded. Furthermore, therapist-related factors may also have contributed to treatment outcomes despite standardized training and treatment manuals. In addition, although therapists received standardized training and followed a treatment manual, formal fidelity assessment was not conducted. Third, all clinical outcomes were assessed using self-report measures, making the findings susceptible to expectancy effects, demand characteristics, and response-shift bias. Cognitive functioning, imagery ability, and transfer of self-compassion skills were not directly assessed, preventing evaluation of their potential influence on treatment response. Furthermore, because outcome analyses were based on treatment completers, the potential for attrition bias cannot be excluded. Moreover, increases in VREQ scores across sessions may have been influenced by familiarity with the VR environment and repeated exposure to the intervention procedures. Finally, conventional frequentist statistical methods were used throughout the study. Although recent uncontrolled VR pilot studies have incorporated Bayesian estimation to better characterize uncertainty in small samples (Kostadinov et al., 2026), such analyses were not prespecified in the present study. Future research may benefit from combining Bayesian and conventional statistical approaches to provide complementary evidence regarding intervention effects.

Despite these limitations, the present study contributes to the growing literature on VR-assisted psychotherapy for depression by integrating immersive virtual avatar technology within a structured self-compassion-based psychotherapeutic framework. Rather than evaluating VR as an isolated technological intervention, the present study emphasizes the complementary roles of immersive experiential learning and therapist-guided psychotherapy. Future technological developments, including standalone VR headsets, marker-free motion tracking, artificial intelligence-assisted avatar interaction, and remote therapist supervision, may further improve accessibility and scalability. However, larger, adequately powered randomized controlled trials with longer follow-up periods are required to determine treatment efficacy, clarify therapeutic mechanisms, establish progression criteria for implementation, and evaluate the long-term clinical value and cost-effectiveness of this intervention.

5 Conclusion

The present pilot study provides preliminary support for the feasibility and tolerability of the intervention. Changes in depressive symptoms, self-compassion, and self-criticism were observed following treatment; however, because of the uncontrolled design, these findings should be considered preliminary and hypothesis-generating. Larger randomized controlled trials are required to determine treatment efficacy.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative AI was used in the preparation of Supplementary Figure S2 for illustrative purposes. The AI-generated graphical elements were reviewed and edited by the authors to accurately represent the intervention procedures. Generative AI was not used for data generation, analysis, or interpretation of the study findings.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frvir.2026.1889087/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Rick Yiu Cho Kwan, Tung Wah College, Hong Kong SAR, China

Reviewed by: Yun-Mi Park, CHA University, Republic of Korea

Marit Hidding, University Medical Center Groningen, Netherlands

Footnotes

BDI-II, Beck Depression Inventory-II; CATS, Computer Avatar Therapy System; DEQ-I, Depressive Experience Questionnaire-Introjective; DSM-5, Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition; MDD, Major Depressive Disorder; PMID, PubMed Identifier; RCT, Randomized Controlled Trial; SCS, Self-Compassion Scale; SD, Standard Deviation; SSQ, Simulator Sickness Questionnaire; SSRI, Selective Serotonin Reuptake Inhibitor; VR, Virtual Reality.