Is emotion regulation a mechanism of change in virtual reality-assisted therapy for auditory hallucinations? Secondary analyses of the CHALLENGE randomized clinical trial Authors: Louise Birkedal Glenthøj, Ditte Lammers Vernal, Lise Sandvig Mariegaard, Lisa Charlotte Smith, Mads Juul Christensen, Sara Breivik Soleim, Tina Dam Kristensen, Neil Thomas, Merete Nordentoft, Carsten Hjorthøj Published: 2026 ABSTRACT Virtual reality (VR)–assisted avatar therapy has shown promising effects for reducing voice-related distress and overall severity of persistent auditory verbal hallucinations (AVH) in psychosis, yet the mechanisms underlying these improvements remain unclear. Emotional processes may play a role during therapeutic encounters with simulated voices, but it is unknown whether treatment effects involve broader changes in emotion regulation. We conducted secondary analyses of the CHALLENGE randomized clinical trial (N = 270), comparing a seven-session immersive VR-based avatar therapy with enhanced treatment as usual (standard care plus additional supportive counseling). Mediation models tested whether changes in cognitive reappraisal or expressive suppression, assessed using the Emotion Regulation Questionnaire (ERQ), mediated treatment effects on AVH severity and frequency at treatment end. Randomization to VR-assisted therapy did not significantly change ERQ reappraisal or suppression. Although higher reappraisal at follow-up was associated with lower AVH severity, no significant indirect effects were observed. These findings suggest that improvements following VR-assisted avatar therapy are unlikely to be explained by changes in self-reported trait emotion regulation strategies and may instead reflect more context-specific processes during voice confrontation. Introduction Auditory verbal hallucinations (AVH) are a core symptom of schizophrenia, affecting approximately two thirds of individuals with the diagnosis1. Relational psychotherapies that conceptualize the voice-hearer relationship as a social interaction have been developed and have proven clinical utility2–5. Avatar therapy operationalizes this approach by enabling patients to engage in therapist-guided, real-time dialog with a digital avatar representing their dominant, distressing voice6,7. The therapeutic aim is to transform the patient’s relationship with the voice from one of fear and submission to one of greater power, control, and confidence8. Through repeated interactions, patients are encouraged to confront the avatar, express their own perspective, and challenge the authority or hostility of the voice. Randomized controlled trials of avatar therapy have reported significant reductions in hallucination severity and distress6,7. Immersive virtual reality (VR) implementations may further extend this approach by creating a more realistic and potentially more emotionally engaging encounter with the simulated voice9–11. Despite promising clinical outcomes, the psychological mechanisms through which avatar therapy produces symptom improvement remain poorly understood. Several candidate mechanisms have been proposed, although empirical evidence for these mechanisms remains limited. First, the intervention may alter relational dynamics between the voice and the hearer, reducing perceived voice power and increasing the individual’s sense of agency during interactions6,8. Second, the intervention may function as a form of graded exposure to feared voice stimuli, enabling patients to confront a persecutory voice in a controlled environment and potentially facilitating inhibitory learning processes that weaken fear responses over time12. Third, engaging in dialog with the avatar may promote changes in behavioral responses to voices, including reduced avoidance and greater assertiveness. In addition to these mechanisms, emotional processes may influence how individuals engage with and respond to voices during therapy13,14. It is important, however, to distinguish between different levels of emotional processes. First, avatar-based interventions are likely to elicit in-the-moment emotional activation during direct engagement with the simulated voice15. Second, individuals differ in their trait-level emotion regulation capacities, such as the habitual use of cognitive reappraisal or expressive suppression, which can be assessed using self-report measures such as the Emotion Regulation Questionnaire (ERQ)16,17. Third, broader relational and behavioral processes, including changes in perceived voice power, coping strategies, and patterns of engagement with voices, may also play an important role in therapeutic change6,8. Emotion regulation refers to the processes through which individuals influence the onset, intensity, duration, and expression of emotional responses16,17. Contemporary models conceptualize emotion regulation as a dynamic and multi-level system involving both deliberate cognitive strategies, such as cognitive reappraisal, and more automatic processes, including attentional, behavioral, and interpersonal regulation17,18. Difficulties in emotion regulation are commonly reported across the psychosis spectrum and have been associated with greater symptom severity, distress, and functional impairment19,20. At the level of emotional activation, responses to voices often involve fear, anger, or perceived threat. These emotional reactions may reinforce maladaptive patterns of responding, such as avoidance or submission to the voice. Emotion regulation processes may therefore shape whether individuals are able to tolerate voice-related distress and experiment with alternative responses during therapeutic interventions. Empirical studies have shown that maladaptive regulation strategies such as expressive suppression and rumination are associated with greater hallucination severity and distress21, whereas greater use of cognitive reappraisal has been linked to more adaptive emotional experiences in individuals with schizophrenia22. Within avatar therapy, emotional engagement is central to the therapeutic process. Interacting with a hostile or critical avatar often elicits strong emotional reactions, particularly during early sessions. Patients are encouraged to remain engaged with the avatar while experimenting with new responses to the voice, which may involve tolerating emotional arousal and reinterpreting the meaning or authority of the voice. Observational studies provide preliminary evidence that emotional responses may change over the course of treatment, with reductions in anxiety and increases in positive emotional engagement across sessions15. Similarly, a greater sense of presence and reductions in anxiety during avatar interactions have been associated with stronger clinical outcomes in one study23. However, it remains unclear whether emotional processes involved in avatar therapy reflect broader changes in general emotion regulation capacity or more context-specific emotional responses occurring during voice confrontation. Most commonly used measures of emotion regulation, including the ERQ, assess relatively stable tendencies to employ particular regulation strategies such as cognitive reappraisal and expressive suppression16. As a result, these measures may capture broader changes in regulatory style but may not fully reflect moment-to-moment emotional processes occurring during therapeutic interactions with the avatar. The present study aimed to investigate whether changes in trait-level emotion regulation represent a pathway of therapeutic change in immersive VR-assisted avatar therapy for treatment-resistant AVH. Using secondary analyses from the CHALLENGE randomized clinical trial, we tested whether changes in cognitive reappraisal and expressive suppression mediated treatment effects on hallucination severity and frequency at treatment end. We hypothesized that increases in cognitive reappraisal and reductions in expressive suppression would mediate reductions in AVH severity and frequency following the VR intervention. Results Of the 270 participants in the study, 61% were females, and participants had a mean age of 32 years (SD = 11.9). The majority (˃90%) had a diagnosis of schizophrenia and had been experiencing auditory hallucinations for an average of 15 years. Baseline characteristics for each intervention group are presented in Table 1.Table 1Participant demographic and clinical characteristics (N = 270).VariablesChallenge VR-group (N = 140)Enhanced TAU group (N = 130)Age32.5 (11.62)33.2 (12.18)Sex, assigned at birthFemale87 (62%)78 (60%)Male53 (38%)52 (40%)EducationPrimary59 (42%)46 (35%)Secondary35 (25%)43 (33%)Vocational education28 (20%)23 (18%)University degree13 (9%)11 (8%)Other5 (4%)6 (5%)EmploymentFull-time2 (1%)5 (4%)Part-time9 (6%)11 (8%)Student12 (9%)20 (15%)Unemployed59 (42%)39 (30%)Retirement48 (34%)45 (35%)Other10 (7%)10 (8%)Diagnosis (ICD-10)Schizophrenia131 (94%)118 (91%)Persistent delusional disorder2 (1%)2 (2%)Schizoaffective disorder2 (1%)4 (3%)Other non-organic psychotic disorder5 (4%)6 (%%)TreatmentsCurrently treated with antipsychotic medication126 (90%)118 (91%)Previously, but not currently treated with antipsychotic medication14 (10%)12 (9%)Dosage of olanzapine equivalent antipsychotic medication /mg/day)19.2 (14.85)18.8 (12.8%)Clozapine31 (22%)31 (24%)Electroconvulsive therapy within the past 12 months8 (6%)5 (4%)Voice duration, years15.0 (10.9)16.1 (11.3)Emotion Regulation QuestionnaireCognitive reappraisal22.3 (8.17)23.8 (8,13)Expressive suppression16.5 (5.68)16.1 (5.41)Values are presented as number and percentage (N (%)) for categorical variables and as mean and standard deviation (M (SD) for continuous variables. Diagnostic categories are based on ICD-10 criteria. Antipsychotic medication dosage is reported in olanzapine equivalents (mg/day). Voice duration refers to the self-reported number of years since the onset of auditory verbal hallucinations. Table 1Participant demographic and clinical characteristics (N = 270).VariablesChallenge VR-group (N = 140)Enhanced TAU group (N = 130)Age32.5 (11.62)33.2 (12.18)Sex, assigned at birthFemale87 (62%)78 (60%)Male53 (38%)52 (40%)EducationPrimary59 (42%)46 (35%)Secondary35 (25%)43 (33%)Vocational education28 (20%)23 (18%)University degree13 (9%)11 (8%)Other5 (4%)6 (5%)EmploymentFull-time2 (1%)5 (4%)Part-time9 (6%)11 (8%)Student12 (9%)20 (15%)Unemployed59 (42%)39 (30%)Retirement48 (34%)45 (35%)Other10 (7%)10 (8%)Diagnosis (ICD-10)Schizophrenia131 (94%)118 (91%)Persistent delusional disorder2 (1%)2 (2%)Schizoaffective disorder2 (1%)4 (3%)Other non-organic psychotic disorder5 (4%)6 (%%)TreatmentsCurrently treated with antipsychotic medication126 (90%)118 (91%)Previously, but not currently treated with antipsychotic medication14 (10%)12 (9%)Dosage of olanzapine equivalent antipsychotic medication /mg/day)19.2 (14.85)18.8 (12.8%)Clozapine31 (22%)31 (24%)Electroconvulsive therapy within the past 12 months8 (6%)5 (4%)Voice duration, years15.0 (10.9)16.1 (11.3)Emotion Regulation QuestionnaireCognitive reappraisal22.3 (8.17)23.8 (8,13)Expressive suppression16.5 (5.68)16.1 (5.41)Values are presented as number and percentage (N (%)) for categorical variables and as mean and standard deviation (M (SD) for continuous variables. Diagnostic categories are based on ICD-10 criteria. Antipsychotic medication dosage is reported in olanzapine equivalents (mg/day). Voice duration refers to the self-reported number of years since the onset of auditory verbal hallucinations. Participant demographic and clinical characteristics (N = 270). Values are presented as number and percentage (N (%)) for categorical variables and as mean and standard deviation (M (SD) for continuous variables. Diagnostic categories are based on ICD-10 criteria. Antipsychotic medication dosage is reported in olanzapine equivalents (mg/day). Voice duration refers to the self-reported number of years since the onset of auditory verbal hallucinations. Mediation models of cognitive reappraisal and expressive suppression mediating the effect of therapy on AVH severity at treatment cessation In the mediation model assessing cognitive reappraisal as a mediator, randomization was not associated with a change in reappraisal scores, although higher reappraisal at follow-up was associated with lower AVH severity. The indirect effect was negligible and not statistically significant (B = −0.11.95% CI [−0.49 to 0.27], p = 0.57), indicating no evidence of mediation on AVH severity. Similarly, for expressive suppression, randomization was not associated with changes in suppression, and suppression was not significantly associated with AVH severity (B = −0.03, 95% CI [−0.25 to 0.18], p = 0.75), providing no evidence of mediation (Table 2).Table 2Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination severity at treatment cessation (12 weeks). N = 270.PathB95% CIp valueProportion mediatedRandomization → Reappraisal0.64[−1.49, 2.76]0.56Reappraisal → Severity−0.17[−0.31, −0.03]0.01Indirect effect−0.11[−0.49, 0.27]0.57Proportion mediated (Reappraisal)0.675.7% [0%, 32.5%]Randomization → Suppression−0.94[−2.35, 0.48]0.20Suppression → Severity−0.03[−0.25, 0.18]0.75Indirect effect0.03[−0.17, 0.24]0.75Proportion mediated (Suppression)1.000% [0%, 17.8%]Bold text indicates a statistically significant association between the variables. Table 2Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination severity at treatment cessation (12 weeks). N = 270.PathB95% CIp valueProportion mediatedRandomization → Reappraisal0.64[−1.49, 2.76]0.56Reappraisal → Severity−0.17[−0.31, −0.03]0.01Indirect effect−0.11[−0.49, 0.27]0.57Proportion mediated (Reappraisal)0.675.7% [0%, 32.5%]Randomization → Suppression−0.94[−2.35, 0.48]0.20Suppression → Severity−0.03[−0.25, 0.18]0.75Indirect effect0.03[−0.17, 0.24]0.75Proportion mediated (Suppression)1.000% [0%, 17.8%]Bold text indicates a statistically significant association between the variables. Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination severity at treatment cessation (12 weeks). N = 270. Bold text indicates a statistically significant association between the variables. Mediation models of cognitive reappraisal and expressive suppression mediating the effect of therapy on AVH frequency at treatment cessation For AVH frequency, randomization was not associated with changes in reappraisal, and reappraisal was not significantly associated with AVH frequency. The indirect effect was negligible and not statistically significant (B = −0.001, 95% CI [−0.02 to 0.01], p = 0.91), with no evidence of mediation. Similarly, for expressive suppression, randomization was not associated with changes in suppression, and suppression was not significantly associated with AVH frequency. The indirect effect was negligible and not statistically significant (B = 0.002, 95% CI [−0.03 to 0.03], p = 0.99), indicating no evidence of mediation (Table 3).Table 3Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination frequency at treatment cessation (12 weeks). N = 270.PathB95% CIp valueProportion mediatedRandomization → Reappraisal0.68[−1.44, 2.81]0.53Reappraisal → Frequency−0.01[−0.02, 0.02]0.91Indirect effect−0.001[−0.02, 0.01]0.91Proportion mediated (Reappraisal)0.930.3% [0%, 8.4%]Randomization → Suppression−0.93[−2.35, 0.49]0.20Suppression → Frequency−0.0003[−0.03, 0.03]0.99Indirect effect0.002[−0.03, 0.03]0.99Proportion mediated (Suppression)1.000% [0%, 11.4%] Table 3Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination frequency at treatment cessation (12 weeks). N = 270.PathB95% CIp valueProportion mediatedRandomization → Reappraisal0.68[−1.44, 2.81]0.53Reappraisal → Frequency−0.01[−0.02, 0.02]0.91Indirect effect−0.001[−0.02, 0.01]0.91Proportion mediated (Reappraisal)0.930.3% [0%, 8.4%]Randomization → Suppression−0.93[−2.35, 0.49]0.20Suppression → Frequency−0.0003[−0.03, 0.03]0.99Indirect effect0.002[−0.03, 0.03]0.99Proportion mediated (Suppression)1.000% [0%, 11.4%] Mediation models of cognitive reappraisal and expressive suppression mediating the effect of VR-assisted therapy on auditory verbal hallucination frequency at treatment cessation (12 weeks). N = 270. Discussion This study is the first to examine global emotion regulation as a potential mediator of treatment response in immersive VR-assisted therapy for treatment-resistant auditory verbal hallucinations (AVH). We hypothesized that increases in cognitive reappraisal and reductions in expressive suppression would partially account for reductions in AVH severity and frequency. Contrary to this hypothesis, neither aspect of emotion regulation, as measured by the ERQ, mediated treatment effects. While higher levels of cognitive reappraisal were associated with lower AVH severity at treatment end, this association did not appear to reflect systematic changes induced by the intervention. Previous research suggests that individuals with schizophrenia report greater use of expressive suppression and lower use of cognitive reappraisal compared to healthy controls, with maladaptive strategies such as suppression and rumination associated with more severe and distressing hallucinations21,24. Although findings are based primarily on self-report measures. From this perspective, enhancing adaptive emotion-regulation strategies could be expected to contribute to symptom improvement. However, the absence of treatment effects on ERQ subscales indicates that the VR intervention did not alter participants’ broader, trait-like emotion regulation styles. Importantly, this should not be interpreted as evidence that emotional processes more generally are not involved in treatment response. Rather, the findings suggest that the intervention may operate through more context-specific emotional and interpersonal processes that are not captured by trait-level measures such as the ERQ. This aligns with meta-analytic evidence showing that trait-level ERQ reappraisal scores are weakly related to psychotic symptom severity, whereas maladaptive strategies tend to show stronger associations20. It is important to note that contemporary models conceptualize emotion regulation as a multi-level, dynamic process encompassing both top-down (deliberate) and bottom-up (automatic, physiological, behavioral, and interpersonal) mechanisms, rather than purely reflective cognitive control25,26. Trait measures like the ERQ capture only a subset of these processes, focusing on general, reflective strategies. In the context of VR-assisted avatar therapy, it is plausible that participants developed context-specific regulatory skills, such as reduced avoidance, increased approach behavior, or more assertive responding toward the avatar, without altering habitual emotion regulation patterns detectable by the ERQ. These processes involve tolerating distress during voice confrontation and engaging more actively with the voice, which are central therapeutic targets but not directly assessed by the ERQ. Qualitative data from the Challenge trial support this interpretation, showing that participants reported shifts in their approach to voices and engagement during avatar encounters, which were highly situation-specific and not framed as changes in general emotion regulation27. The intervention’s emphasis on direct engagement and dialog with the avatar also suggests that treatment effects may involve mechanisms such as exposure, habituation, and inhibitory learning, as well as shifts in perceived control and relational dynamics with the voice6,8,28. However, these processes were not directly assessed in the present study and should therefore be considered as theoretically plausible explanations rather than empirically established mechanisms. Previous avatar trials have highlighted reductions in perceived omnipotence and negative beliefs about voices as key outcomes6,7. In the Challenge trial, several voice-specific cognitive variables (e.g., perceived power, acceptance) did not show robust change at treatment end29, suggesting that symptom improvements may instead reflect processes such as procedural learning: patients may have learned that confronting the voice is safe, diminishing its emotional salience and behavioral impact independently of deliberate cognitive reappraisal, although this was not directly assessed in the present study Clinically, these findings caution against assuming that enhancing trait-level ERQ-defined emotion regulation strategies will necessarily strengthen therapeutic effects in AVH. The ERQ may be a suboptimal measure in schizophrenia, as it presupposes intact higher-order reflective control over emotions, which may be broadly impaired in psychotic disorders. Broader, more sensitive measures of dysregulation, such as the Difficulties in Emotion Regulation Scale (DERS)30, may better capture clinically relevant regulation difficulties in psychosis, including global dysregulation and negative-symptom-related impairment. Nevertheless, the observed association between higher cognitive reappraisal and lower AVH severity at treatment end suggests that explicit training in reappraisal could augment outcomes if systematically integrated. Extending the intervention beyond seven sessions may allow more sustained and focused work on reappraisal within the VR context, potentially enhancing therapeutic efficacy. This is consistent with findings from the AVATAR2 trial, where a more extended, personalized intervention yielded larger and more durable benefits7. By contrast, expressive suppression did not appear to play a meaningful role in symptom change, indicating that targeting suppression is unlikely to constitute a central therapeutic pathway. Several limitations warrant consideration. First, as a secondary analysis, mediation effects are expected to be small. Although point estimates for indirect effects were near zero, confidence intervals were wide, leaving some uncertainty regarding very small effects. Second, the ERQ primarily assesses trait-like, reflective regulation strategies and may not capture context-specific emotional processes engaged during direct avatar interactions. Third, mediators and outcomes were assessed at the same follow-up time point, which limits the ability to establish temporal ordering between changes in emotion regulation and AVH outcomes. As a result, the analyses do not permit causal inference regarding mediation, and the findings should be interpreted as associations consistent with, but not confirming, underlying mechanisms. This limitation is particularly relevant given that mediation models assume a temporal sequence in which changes in the mediator precede changes in the outcome. Future research should employ repeated, temporally ordered, voice-specific, and state-level assessments (e.g., ecological momentary assessment), ideally complemented by objective and process-sensitive measures such as physiological markers or behavioral indices, to better capture dynamic emotional and interpersonal processes during voice encounters31. Finally, the analyses focused on AVH severity and frequency, as these were the outcomes demonstrating significant intervention effects in the trial. However, voice-related distress may be more closely linked to emotional processes, including emotion regulation. The lack of mediation effects observed in the present study may therefore partly reflect the choice of outcome measure. Future studies should examine whether emotion regulation plays a more prominent role in changes in distress specifically. In summary, changes in emotion regulation as measured by the ERQ did not mediate the effects of VR-assisted avatar therapy on AVH. Instead, improvements in AVH severity and frequency may be more closely linked to voice-specific, context-dependent mechanisms, including graded confrontation, habituation, procedural learning, and altered relational dynamics. Future studies should prioritize testing and refining these mechanisms using context-sensitive, state-level measures to optimize clinical benefit. Explicitly incorporating cognitive reappraisal training within VR sessions also represents a promising avenue for future investigation. Methods Study design and participants The study is based on data from the Challenge, a randomized, assessor-blinded, clinical trial conducted in Denmark between 2020 and 202311 (ClinicalTrials.gov NCT04661163). Patients were recruited from psychiatric in- and outpatient facilities in the Mental Health Services in the Capital Region of Denmark, North Denmark Region, and the Region of Southern Denmark. The study protocol was approved by the Danish Committee on Health Research Ethics (H-19086621). Assessments were carried out at baseline, 12- (treatment cessation) and 24-weeks post baseline. All participants provided written informed consent prior to inclusion in the study. Participants met the following inclusion criteria: age ≥18 years; a diagnosis of a schizophrenia spectrum disorder according to ICD-10; presence of auditory verbal hallucinations for at least three months, defined as a score ≥3 on the auditory hallucinations item of the Scale for the Assessment of Positive Symptoms32); ability to provide informed consent; no changes in antipsychotic medication within the past four weeks; and either current insufficient response to antipsychotic treatment or prior insufficient response to at least two antipsychotic agents. Exclusion criteria were: inability to identify a dominant voice; diagnosis of organic brain disease; substance abuse severe enough to impede engagement in therapy; auditory hallucinations occurring in a language not spoken by the therapists; insufficient proficiency in spoken Danish or English to engage in therapy; inability to tolerate the assessment process; or severe visual impairment. The present analyses include the 270 participants who were randomized to receive the experimental VR-based intervention or the control condition of enhanced TAU. Experimental intervention (Challenge-VRT) Challenge-VRT is a manualized, phase-based psychotherapeutic intervention co-developed in collaboration with individuals with lived experience of hearing voices. The intervention expands upon the original avatar therapy protocol8 and is structured into three sequential phases: (1) fostering empowerment in relation to the voice, (2) strengthening self-worth, and (3) supporting personal recovery. In the initial session, patients and therapists jointly create a digital avatar that visually and auditorily represents the patient’s dominant or most distressing voice. Using specialized software, the avatar’s appearance and voice characteristics (e.g., tone, pitch, timbre) are adjusted to closely match the actual perceptual qualities of the voice. Once finalized, patients engage in immersive, therapist-guided 3D dialogs with the avatar using a head-mounted display. The therapist provides the avatar’s speech in real time via voice-modulation software and may switch to their natural voice to deliver supportive or reflective comments. The intervention comprised seven sessions. The first session (approximately 90 minutes) was dedicated to constructing the personalized avatar. The following six sessions (approximately 60 minutes each) included 2–20 min of VR-based dialog, with the remaining time devoted to collaboratively preparing the dialog content and conducting reflection and debriefing after the VR interaction. Two booster sessions were offered during the follow-up period (weeks 12–24). A central aim of the intervention was to strengthen patients’ ability to manage emotional responses during voice interactions. Patients were encouraged to tolerate emotional arousal during exposure to hostile or demeaning voice content, reduce avoidance and fear-driven responses, and practice calm, firm, and assertive responding to the avatar. Therapists also supported cognitive reappraisal of the voice’s intentions and meaning, helping patients reinterpret the voice as less omnipotent and less threatening. Although the intervention targeted emotion regulation within the specific context of voice-hearing, it was hypothesized that these skills could generalize to broader regulatory capacities. The VR platform was developed by Heka-VR. Challenge-VRT was delivered by a multidisciplinary team of psychologists, medical doctors, and nurses, all (except one) with prior clinical experience in psychosis treatment. Therapists received biweekly supervision throughout the intervention. Independent fidelity assessments indicated strong adherence to the treatment manual29. Control condition Enhanced treatment as usual (TAU) was delivered by participants’ outpatient psychiatric services. Standard care in Denmark typically involves the early intervention facility (termed OPUS), with weekly to biweekly sessions for up to two years, followed by lower-intensity follow-up in Flexible Assertive Community Treatment (F-ACT). As most participants were in F-ACT, clinics provided additional supportive counseling to ensure that control-group participants were offered at least seven sessions within 12 weeks after enrollment, matching the minimum session dose in the VR condition. The control sessions were not manualized and were delivered as routine care. Measures Auditory hallucinations Severity and frequency of AVH were assessed with the clinician-rated Psychotic Symptoms Rating Scale, Auditory Hallucinations subscale (PSYRATS-AH)33 severity and its frequency scale. The PSYRATS-AH is widely used in psychosis trials and has demonstrated strong reliability and sensitivity to change in clinical trials34. Emotion regulation questionnaire (ERQ) The Emotion Regulation Questionnaire (ERQ)16 is a self-report measure comprising ten questions: four that focus on suppression (“I control my emotions by not expressing them”), and six which focus on cognitive reappraisal (“I control my emotions by changing the way I think about the situation I’m in”). Each emotion regulation strategy is rated on a seven-point Likert scale from 1, strongly disagree, to 7, strongly agree. The ERQ has proven sound psychometric properties16,35, and has been used extensively to measure emotion regulation in studies within the psychosis population22,24,36–39. Statistical analysis Statistical analyses were performed using STATA version SE 18.5. Descriptive statistics were used to summarize demographic and clinical characteristics at baseline. Means and standard deviations were calculated for continuous variables, and proportions for categorical variables. Primary analyses consisted of mediation models examining whether changes in emotion regulation, operationalized as cognitive reappraisal and expressive suppression, mediated the effect of the VR-based intervention on auditory verbal hallucinations (AVH). Two outcomes were analysed separately: AVH severity, the primary outcome in the Challenge trial, and AVH frequency, a predefined secondary outcome, both assessed at 12-week follow-up using the PSYRATS-AH subscales. These outcomes were selected as they demonstrated statistically significant intervention effects in the primary trial analyses. Mediation analyses were therefore restricted to outcomes with evidence of treatment effects, as mediation is not meaningful in the absence of a total effect. Associations between the mediator and outcome reflect relationships at the 12-week follow-up, adjusted for baseline levels. Treatment allocation (VR-assisted therapy vs enhanced TAU) was included as a binary independent variable in all mediation models, which were estimated across the full sample. Each SEM followed a path-analytic framework, decomposing the total effect of treatment into direct and indirect (mediated) effects. Separate models were estimated for each ERQ subscale (reappraisal and suppression). The indirect effect was calculated as the product of the path from treatment allocation to change in ERQ score (path a) and the path from change in ERQ score to the AVH outcome (path b). Standardized regression coefficients and 95% bias-corrected confidence intervals were estimated using 5000 bootstrap resamples. Statistical significance of indirect effects was defined as confidence intervals not crossing zero (Fig. 1).Fig. 1Conceptual model of the proposed relationships between treatment, emotion regulation, and auditory hallucination outcomes.The figure illustrates the hypothesized mediation model in which treatment condition (VR vs. TAU) influences auditory verbal hallucination (AVH) outcomes via emotion regulation strategies measured by the Emotion Regulation Questionnaire (ERQ; reappraisal and suppression). Path a represents the association between treatment and the mediator, path b the association between the mediator and outcome, and c′ the direct effect of treatment on outcome. Outcomes were assessed using the Psychotic Symptom Rating Scales – Auditory Hallucinations subscale (PSYRATS-AH). VR = virtual reality; TAU = treatment as usual. Conceptual model of the proposed relationships between treatment, emotion regulation, and auditory hallucination outcomes. The figure illustrates the hypothesized mediation model in which treatment condition (VR vs. TAU) influences auditory verbal hallucination (AVH) outcomes via emotion regulation strategies measured by the Emotion Regulation Questionnaire (ERQ; reappraisal and suppression). Path a represents the association between treatment and the mediator, path b the association between the mediator and outcome, and c′ the direct effect of treatment on outcome. Outcomes were assessed using the Psychotic Symptom Rating Scales – Auditory Hallucinations subscale (PSYRATS-AH). VR = virtual reality; TAU = treatment as usual. All models controlled for baseline values of both the mediator and outcome variables. Mediation effects are presented both as absolute indirect effects and as proportions of the total effect. Statistical significance was set at p ≤ 0.05 (two-tailed). Ethical standards The authors assert that all procedures contributing to this work comply with the ethical standards of the national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. Supplementary information NPJSCHZ-02149-s01 NPJSCHZ-02149-s01