Recruitment occurred between April 2019 and June 2025. Participants were either referred by their health care team or self-referred. Active efforts were made to inform health care providers about the project in 4 psychiatric institutions across the province, including the major psychiatric institutions on the island of Montreal, as well as in other urban and suburban areas, and online via videoconferencing. Presentations were also offered to patients and their support network through community organizations such as the Société Québécoise de la Schizophrénie and the Association des parents et amis de la personne atteinte de maladie mentale. Information about the project was available on a dedicated website, which included a self-referral form and contact information (phone number and email address) for more details on study eligibility. However, participants could come from anywhere in the province. The therapies were delivered at the Institut universitaire en santé mentale de Montréal located in Montreal, Canada. Participants were responsible for traveling to Montreal for therapy sessions. When needed, compensation for public transportation (eg, bus tickets) or taxi transportation was provided to facilitate attendance.

Participants were eligible if they were aged 18 years or older; diagnosed with either schizophrenia or schizoaffective disorder with persistent AVH; nonresponsive to 2 or more antipsychotic trials, each lasting more than 6 weeks at a therapeutic dosage equivalent to at least 600 mg of chlorpromazine daily; and on stable doses of medication during the 2 months before study enrollment with good adherence (defined as 80% or more of the doses being taken). Participants were excluded if they presented a neurological disorder, had an unstable and serious physical illness, had a substance use disorder in the past year, had received any form of CBT for psychosis in the past year, lacked the capacity to consent, or were currently experiencing an acute psychotic episode. These criteria were similar to those of other trials in the field of treatment-resistant schizophrenia [21-23]. Participants with insufficient French language proficiency to complete the study protocol were excluded. With the participant’s consent, attempts were made to contact each participant’s treating psychiatrist to confirm eligibility, inform them of their participation in the clinical trial, and ensure that no changes in psychiatric medication were planned in the near future. Moreover, participants could be excluded during therapy if their mental state deteriorated and prevented them from continuing the treatment.

All participants continued to receive standard psychiatric care throughout the study, which involved antipsychotic medication and regular outpatient psychiatric appointments. Eligible participants were randomly assigned in a 1:1 ratio to 1 of 2 parallel arms to receive either VRT or a short course of CBT.

The trial was conducted in accordance with the Declaration of Helsinki. It was approved by the institutional ethics committee of the Centre intégré universitaire de santé et de services sociaux (CIUSSS) de l’Est-de-l’Île-de-Montréal (project MP-12-2019-1726) in April 2019. Prior to their inclusion in the study, the information and consent form was carefully presented and explained to potential participants by a research assistant or psychiatric nurse. The paper form was also provided, along with time and space to make an informed decision. All questions were answered as needed, whether during the presentation or later by phone. Participants’ capacity to consent was also assessed; for example, participants were asked to summarize different sections of the form. If they agreed to participate and had the capacity to consent, they were allowed to provide written informed consent, and basic information was collected during the same visit, including sociodemographic data, contact information, and current use of clozapine (for randomization purposes). As compensation for their time, participants received CAD $20 (CAD $1=US $0.73 as of September 8, 2026) per clinical assessment.

Protocol deviations and undesirable treatment effects were compiled, analyzed, and presented by a graduate student (MB). Four categories of deviation were documented, including assessment timing, evaluator blinding, changes in antipsychotic medications, and treatment fidelity. The study was then overseen regularly by a steering committee that included the involved researchers, clinicians, a graduate student (MB), the president of the ethics committee, a biostatistician (CÉG), and one patient partner. After each meeting, the committee had to vote to continue the study, and the vote had to be unanimous to allow continuation. Moreover, the committee’s mandate was to recommend future adjustments to the study, notably to minimize and manage protocol deviations. If serious adverse events had occurred, a protocol was in place to ensure that they would have been handled seriously, including careful revision of the event by an independent committee to determine whether it was linked to therapy and what could have been done to prevent it. Moreover, the treatment fidelity was evaluated across 120 randomly selected sessions from both conditions, comprising 60 sessions from the CBT condition and 60 sessions from the VRT condition. Adherence to the respective treatment manuals was assessed using structured fidelity checklists developed specifically for each intervention. All items were rated dichotomously by an independent and trained evaluator. Fidelity ratings were conducted by 7 evaluators, comprising research assistants with clinical training, psychiatry residents, or graduate-level students. Evaluators received standardized training from the CBT and VRT experts, respectively, based on the corresponding treatment manuals. Finally, potential between-group differences in dropouts, exclusions, losses to follow-up, and overall completion rates were also examined.

Treatment efficacy and comparisons between treatments for primary and secondary outcomes were assessed using linear mixed models with restricted maximum likelihood (REML) estimation for missing data. Fixed effects included time (pretherapy, posttherapy, and 3-month follow-up), group (CBT and VRT), sex, clozapine use, and the interaction between time and group. A random intercept within subject was included, and residual variance was heterogeneous across time. Analyses were conducted according to the intention-to-treat principle, including all randomly assigned participants. Both between-group (treatment) and within-group (time) comparisons were examined, and time-by-treatment interactions were used to examine whether there was a significant change between VRT and CBT over time. Linear mixed models were chosen because they are well established and recommended for analyzing psychotherapy RCTs with repeated measures. They handle unequal variances and correlated data, which are very common in treatment trials, and allow unequal numbers of follow-ups, making them suitable for clinical trial data. Furthermore, linear mixed models are based on maximum likelihood methods rather than the usual ANOVA methods, which require data sphericity and are not robust with small to moderate sample sizes [40,41]. Post hoc false discovery rate corrections were applied separately within each questionnaire across all of its outcomes (eg, total score and subscale scores), excluding the prespecified primary outcome (PSYRATS-AH total score), to the between-group contrasts post therapy and at the 3-month follow-up, as well as to the within-group analyses [42]. These adjusted analyses were considered exploratory, as no multiplicity adjustment strategy was prespecified.

To evaluate whether there were significant differences in antipsychotic use between the 2 groups at baseline, posttherapy, and 3-month follow-up, doses were converted to olanzapine equivalents and analyzed using the Mann-Whitney U test. The number of participants with a medication change in each group was compared using the chi-square test. Moreover, intervals between assessments and therapy sessions were also compared between the 2 groups using the Mann-Whitney U test. Chi-square or Fisher tests (as appropriate) were performed to assess whether there were systematic differences between groups in dropout rates, exclusion, and missing data. Potentially serious adverse events were compared between treatment groups using Fisher exact test. Between-group differences in the NEQ total score and subscale scores were assessed using 2-tailed independent-samples t tests, and mean differences with 95% CIs were reported.

Drawing on findings from comparable studies, pilot data on the effects of VRT on AVH, and the small to moderate effect sizes typically observed for CBT, the between-group statistical power for the PSYRATS-AH outcome was estimated at f=0.25 [15,16,19,43]. Sample size calculations were performed using G*Power. With a 2-sided α=.05 and 80% power, a sample of 52 participants per arm was required to detect differences using pretreatment and posttreatment assessments. To account for an anticipated 20% attrition rate (based on our pilot trial) and 5 df for potential confounders, the final recruitment target was set at 136 participants [16].

Effect sizes (Cohen d) were categorized as small (0.2), medium (0.5), or large (>0.8) [44]. Effect sizes were estimated using R software (version 4.5.2; R Foundation for Statistical Computing) emmeans package (version 2.0.3), using the square root of the sum of the variances (random intercept + residuals) as sigma and the residuals df for the calculation [45]. Two-tailed P values were reported using .05 as the significance threshold. Analyses were performed independently by a doctoral student (MB) and an independent biostatistician with 15 years of experience (CÉG), using IBM SPSS Statistics for Windows (version 31) and R Statistics, (version 4.5.2; R Foundation for Statistical Computing), respectively, to ensure consistent results [46,47].

Participant recruitment occurred between April 2019 and June 2025. A total of 356 individuals were referred and assessed for eligibility across geographically and demographically diverse locations in Quebec, Canada. Of these, 136 participants met the inclusion criteria and were recruited and randomly assigned, yielding a recruitment rate of 38.2% (136/356; see Figure 1). Participants were recruited from 10 health regions across Quebec, representing both urban and mixed urban–rural catchment areas. Most participants were recruited from the 5 CIUSSSs on the island of Montreal (Centre-Sud, Centre-Ouest, Est, Ouest, and Nord; n=107), while additional participants were referred from Laval (n=3), Montérégie (n=6), Lanaudière (n=12), Laurentides (n=1), Estrie (n=2), and Quebec (n=1). Three participants had no documented referral region. Referring regions ranged from the Montreal area (approximately 15‐50 km) to more distant regions located up to approximately 260 km away.

The main reasons for exclusion were having unstable medication use during the 2 months before enrollment (n=63), not meeting the diagnostic criteria for schizophrenia or schizoaffective disorder with persistent AVH (n=26), having a substance use disorder within the past year (n=19), having insufficient French language proficiency to complete the study protocol (n=18), and experiencing an acute psychotic episode at the time of screening (n=11).

Following randomization, 67 participants were allocated to the VRT condition and 69 to the short-course CBT condition. Baseline characteristics were well balanced across both treatment arms. Randomization was stratified by sex and clozapine use to ensure comparability between groups.

Therapy sessions were delivered between April 2019 and October 2025, and the last 3-month follow-up assessment was completed in January 2026. A total of 51 participants completed 9 VRT sessions for a total of 459 therapy sessions. In comparison, 53 participants completed 9 CBT sessions, for a total of 477 therapy sessions.

No significant between-group differences were found in overall completion rate, dropouts, exclusions, and losses to follow-up at the posttherapy and 3-month follow-up assessments; please see Table S2 in Multimedia Appendix.

The initial demographic and medical profiles of study participants appeared to be well balanced between the 2 groups (see Table 1). Participants had a mean age of 40.3 (SD 12.8) years (range 18‐69 years) and were predominantly male (63.2%, 86/136), White (59.6%, 81/136), single (82.4%, 112/136), and without children (83.1%, 113/136). Social assistance was the primary source of income for 67.6% (92/136) of the participants, and 75.7% (103/136) reported annual incomes below CAD $20,000 (CAD $1=US $0.73 as of September 8, 2026), which is below the poverty threshold. Among participants, 65.4% (89/136) had a primary diagnosis of schizophrenia, and 34.6% (47/136) had schizoaffective disorder. The mean duration since first contact with psychiatry was 16.7 (SD 11) years, and 56.6% (77/136) of the participants were receiving clozapine.

For auditory hallucination severity, the intention-to-treat analysis using linear mixed models revealed a significant time-by-treatment interaction measured by PSYRATS-AH total score (P=.013), demonstrating VRT’s superiority over a short course of CBT in the primary outcome. Regarding specific time points, no differences were seen at the posttherapy evaluation (P=.809), whereas VRT was significantly more effective than a short course of CBT targeting AVH at the 3-month follow-up (Cohen d=0.614, 95% CI 0.085‐1.143, P=.022).

When examining the different subscales of the PSYRATS-AH scale, it was observed that only the frequency subscale showed greater improvements with VRT than with the short-course CBT (P=.031), with again no differences at the posttherapy evaluation (P=.142) but a significant difference at the 3-month follow-up visit (Cohen d=0.513, 95% CI 0.080‐0.947, P=.020). No significant between-group differences were observed in AVH-related distress, attribution, and loudness.

Regarding the delusions subscale of the PSYRATS, there was a significant time-by-treatment interaction in the distress subscale, demonstrating overall superiority of VRT over a short course of CBT over time (P=.011), although no differences were seen when examining specific time points.

Finally, no significant between-group differences were observed for all the other secondary outcomes, including overall psychotic symptomatology (measured using the PANSS, including the positive, negative, disorganized, excitation/hostility, and anxio-depressive subscales), beliefs about the voices (measured using the BAVQR, including the malevolence, benevolence, omnipotence, resistance, and engagement subscales), voices acceptance and action (measured using the VAAS), emotional regulation (measured using the DERS), self-esteem (measured using the RSES), and quality of life (measured using the QLESQ-SF).

Detailed between-group results for all linear mixed models, including time-by-treatment interactions, can be found in Table 2.

Within-group effects of both therapies were also examined using linear mixed models. These results are visually presented in Figure 2, and detailed statistics can be found in Table S3 in Multimedia Appendix 1.

VRT produced a large effect on auditory hallucination severity immediately post therapy (Cohen d=0.796, P<.001) and at the 3-month follow-up (Cohen d=1.160, P<.001). Regarding specific auditory hallucination subscales, VRT demonstrated substantial and enduring improvements in hallucination-related distress, frequency, attribution, and loudness. For the delusions subscale, improvements were seen in the total score both post therapy and at the 3-month follow-up. Improvements were also seen in general psychotic psychopathology, as measured with the PANSS, at both time points, including the positive (both time points), negative (3-month follow-up only), and disorganized (posttherapy evaluation only) subscales. As for beliefs about the voices, significant improvements were also seen in the total score (both time points) as well as in the malevolence (3-month follow-up only) and omnipotence (both time points) subscales. Statistically significant improvements were also seen in voices acceptance and action (3-month follow-up only), emotional regulation (both time points), and self-esteem (both time points). No significant improvements were seen in quality of life enjoyment and satisfaction measured using the QLESQ-SF.

Finally, the short course of CBT targeting AVH showed significant improvements regarding auditory hallucination severity at the posttherapy visit only (Cohen d=0.572, P=.004), whereas no significant improvements were seen at the 3-month follow-up (P=.080). Nonetheless, significant improvements were seen in some subscales, including AVH-related distress (posttherapy only) and frequency (both time points). No significant effects were seen on attribution and loudness. Moreover, no improvements were seen in the delusions subscale of the PSYRATS. CBT had significant within-group impacts on overall psychotic psychopathology at all time points, including improvements in positive, negative, and disorganized symptoms. Regarding beliefs about the voices, although no significant changes were seen in the total BAVQR score, significant improvements were observed in the malevolence subscale only (both time points). Improvements were also seen in voices acceptance and action (all time points), emotional regulation (3-month follow-up only), and self-esteem (posttherapy evaluation only). No changes were seen in quality of life.

The NEQ was completed by 100 of the 136 participants (73.5% completion rate among those who underwent at least one therapy session). Missing data mainly occurred when participants were unreachable, which was also common among excluded participants and those lost to follow-up. Although efforts were made to contact excluded participants once their mental state stabilized, the questionnaire was not administered if they were unable to properly recall therapy. Reported negative effects were low and comparable across study conditions. The between-group comparison of the NEQ total score and subscale scores showed a significant difference only for the perceived treatment quality subscale, with participants reporting lower satisfaction with CBT than with VRT. Results by subscale are presented in Table 3. Additional information is presented in Figure S2 in Multimedia Appendix 1.

Other potentially serious adverse events were thoroughly examined. Hospital admissions were infrequent but more prevalent in the CBT group (6 admissions for psychological health events and 1 admission for a physical health event in the CBT group [the latter occurred in a participant who also had a psychiatric admission] vs only 2 psychiatric admissions in the VRT group). No temporal pattern was observed, as hospital admissions occurred after different therapy sessions, and hospitalization rates did not differ significantly between the 2 groups (P=.275). One detox admission occurred in the CBT group. Finally, one death occurred in the CBT group due to medical causes unrelated to the trial protocol.

Protocol deviations were prospectively recorded and reviewed during steering committee meetings.

First, deviations in assessment timing were generally low. The interval between the initial session and the baseline assessment exceeded the 7-day threshold for a greater proportion of participants in the VRT group; similarly, a greater number of individuals in the CBT group required more time to complete the 3-month follow-up assessment. The other intervals were comparable, without statistically significant differences. Full details are provided in Table S4 in Multimedia Appendix 1.

Second, evaluator blinding was compromised in 3 posttreatment assessments, in which case outcomes requiring clinical judgment were rerated from audio recordings by an independent blinded evaluator for those participants.

Third, changes in antipsychotic medication were common, occurring in 24.8% of participants during therapy and in 50% of participants by the 3-month follow-up visit. However, no between-group differences were observed (please refer to Table S5 in Multimedia Appendix 1 for details). In addition, the olanzapine-equivalent dosage in antipsychotic medication did not differ between the 2 treatment arms at any of the 3 assessments, as detailed in Table S6 in Multimedia Appendix 1. Although the analyses followed an intention-to-treat approach, follow-up data, including outcome measures, could not be obtained for participants who were excluded, despite repeated attempts to recontact them at the time of exclusion. In most cases, exclusion occurred because participants were deemed too clinically unstable to continue therapy, which also generally prevented them from providing informed consent for a research evaluation. All available information, including baseline assessments, was nevertheless incorporated into the linear mixed model analyses.

Results indicated good adherence to the study protocol across both conditions. In the VRT group, therapists achieved a mean fidelity rate of 96.8% (SD 3.8%; range 90%‐100%), while therapists in the CBT condition demonstrated a mean fidelity rate of 93.2% (SD 6.5%; range 81%‐100%).