Pre-registered three-arm RCT: CBT with VR exposure was superior to in-vivo exposure on the primary social-anxiety outcome - and more practical for therapists

Background

Social anxiety disorder (SAD) is among the most common psychiatric conditions (lifetime prevalence ~12.1%), yet remains substantially undertreated. Treatment-seeking is itself limited by the disorder: patients with SAD often avoid healthcare services as they do other social interactions, find psychotherapy threatening to their need for privacy, and may wait many years before consulting. The consensus first-line psychotherapy for SAD is cognitive-behavioral therapy (CBT) with exposure - but in-vivo exposure exercises raise practical problems for therapists (logistically demanding, scheduling-dependent, hard to grade systematically, privacy concerns when conducted in public) and arousal/avoidance issues for patients. Virtual reality (in virtuo) exposure has been proposed as an alternative medium that gives the therapist controllable, repeatable, confidential exposure scenarios delivered from the office.

By the time of this trial, prior work had supported the efficacy of VR exposure for specific phobias and for public-speaking anxiety, with two pilot studies addressing both performance and non-performance social situations in SAD (Klinger et al. 2005; Kampmann et al. 2016). Bouchard and colleagues set out to conduct a more methodologically rigorous test: a three-arm RCT comparing 14 weekly sessions of individual CBT with in-virtuo exposure, the same CBT with in-vivo exposure, and a waitlist control, with a pre-specified superiority hypothesis (VR more effective than in vivo) and a measure of the practical burden of conducting each modality (SWEAT scale).

What the researchers did

Patients were recruited through referrals from practitioners and advertisements in local newspapers and university networks in the Outaouais region of Quebec. Of 90 contacted, 80 underwent the Structured Clinical Interview for DSM-IV (SCID); 21 were excluded, leaving 59 randomized to one of three conditions: CBT with in-virtuo exposure (n=17), CBT with in-vivo exposure (n=22), or waitlist control (n=20). Random assignments were generated with a random numbers table prior to recruitment; assignments were concealed until the first therapy session began. The trial was registered as ISRCTN99747069.

Inclusion criteria: French-speaking adults aged 18-65 with a primary DSM-5 diagnosis of SAD for at least 2 years. Stable medication (>=6 months) permitted if unchanged throughout the study.

Exclusion: dementia, intellectual disability, amnesia, schizophrenia, psychosis, bipolar disorder, SAD secondary to another DSM-IV diagnosis, concurrent psychotherapy, history of seizures.

Treatment. Both active arms received 14 weekly 60-minute individual CBT sessions adapted from Clark & Wells’s model. Therapists were graduate students experienced in CBT for anxiety disorders with at least one year of practical experience in in-vivo or in-virtuo exposure. The two arms differed only in the modality of exposure (sessions 7-14, 20-30 minutes per session):

• In-vivo arm: role-playing and guided exposure inside or outside the therapist’s office (e.g., asking for the time in a coffee shop, making mistakes in a public place, being video recorded, wearing mismatched socks in public, asking strangers on a date, awkward impromptu speech to a mock audience of staff members, improper requests in stores). Laboratory staff were called upon for mock audience exposure.

• In-virtuo arm: 8 virtual scenarios using environments from Virtually Better (Decatur, GA) and Klinger et al.: speaking in front of an audience in a meeting room (two scenarios), having a job interview (two scenarios), introducing oneself and having a talk with relatives in an apartment, acting under the scrutiny of strangers on a coffee shop patio, and facing criticism or insistence in two assertiveness situations (unfriendly neighbours; refusing a persistent seller). Hardware: eMagin z800 head-mounted display + InterSense Inertia Cube motion tracker. Patients navigated using the HMD and a wireless computer mouse, interacting with virtual characters via preformatted dialogue triggered by the therapist. Participants in the in-virtuo arm were instructed not to engage in any in-vivo exposure between sessions.

In both arms, exposure followed the inhibitory learning approach (Craske et al. 2014) - focused on developing new non-threatening and adaptive interpretations of feared social situations rather than habituation. Exposure to the same situation was NOT necessarily repeated frequently. The first six and last therapy sessions delivered cognitive components: alliance-building, personal case conceptualization, cognitive restructuring of dysfunctional beliefs, and relapse prevention.

Primary outcome: total score on the Liebowitz Social Anxiety Scale-Self Report (LSAS-SR), pre-specified before the study began.

Secondary outcomes: Social Phobia Scale (SPS), Social Interaction Anxiety Scale (SIAS), Fear of Negative Evaluation (FNE), Beck Depression Inventory-II (BDI-II), and a Behavioral Assessment Task (BAT) in which patients gave an impromptu speech for up to 6 minutes, scored by three masked independent assessors using the Social Performance Rating Scale.

Process measures: the Specific Work for Exposure Applied in Therapy (SWEAT) scale was completed by therapists after every exposure session (294 sessions total) to measure the practical effort/cost of delivering exposure. Treatment credibility, working alliance, Simulator Sickness Questionnaire (pre/post each immersion), and Presence Questionnaires (Witmer & Singer PQ; Gatineau PQ) were also collected.

Hypothesis (pre-specified): that CBT with in-virtuo exposure would be more effective AND more practical than CBT with in-vivo exposure. This was a superiority test, not a non-inferiority test. A power analysis indicated that a sample of 78 was sufficient to detect Cohen’s f = 0.4 with alpha 0.05 and power 0.80 for the gold-standard comparison.

Analysis. Intent-to-treat with last-observation-carried-forward for non-completers. Repeated-measures ANOVAs with planned orthogonal contrasts: first contrast = waitlist vs the two active conditions (one-tailed); second contrast = the two active conditions (two-tailed). Family-wise Bonferroni correction applied to the five secondary measures (alpha = 0.05/5).

What they found

Compared with waitlist. Both active treatments produced significant improvements on the LSAS-SR (primary outcome) and on all five secondary outcomes (BAT, SPS, SIAS, FNE, BDI-II), with the planned orthogonal contrasts confirming both active conditions significantly outperformed waitlist on every measure.

Compared with each other (the pre-specified superiority test). CBT with in-virtuo exposure was significantly more effective than CBT with in-vivo exposure on:

• LSAS-SR (primary outcome) at post-treatment: planned orthogonal contrast t(56)=2.02, p<.05 (two-tailed). At 6-month follow-up: condition-by-time interaction F(1,37)=4.78, p<.05.
• SPS (secondary) at post-treatment: t(56)=2.69, p<.01 under Bonferroni correction.

The two active conditions did NOT differ significantly on BAT, SIAS, FNE, or BDI-II at post-treatment, and did not differ on the secondary outcomes at follow-up. The reliable change index from pre- to post-treatment was 76.5% in CBT with in-virtuo exposure, 68.3% in CBT with in-vivo exposure, and 30.0% in the waitlist condition (chi-square(2)=9.78, p<.01); the difference between active conditions did not approach significance (chi-square(1)=0.33, ns).

Practical burden (SWEAT). Conducting exposure was significantly less cumbersome and effortful in VR than in vivo (15.24 vs 24.46, t(22.83)=3.66, p<.001) - a large effect.

Process measures. Working alliance (measured after session 7) was a strong significant predictor of LSAS-SR change in both conditions (CBT+VR: t=-2.52, p<.05, sr=-.52; CBT+in vivo: t=-2.8, p<.05, sr=-.42). Treatment credibility was NOT a significant predictor. Multivariate regression showed treatment modality, treatment credibility (combined), and working alliance jointly explained 55% of the variance in LSAS-SR change (F(4,27)=9.22, p<.001).

Tolerability. Presence Questionnaire scores were strong (M=78.3 at session 1, rising to M=93.7 at session 8) and Simulator Sickness Questionnaire scores did not significantly increase pre/post any of the 8 exposure sessions. Attrition rates did not differ significantly between active conditions (Fisher’s exact test p=.67). The only attrition reasons explicitly cited were one participant who did not want to be exposed (in-vivo arm) and two who were no longer interested in therapy (in-vivo arm).

Treatment durability. All gains were maintained at the 6-month follow-up; pre-vs-follow-up time effects were significant across all measures with no significant post-vs-follow-up changes.

Why this matters

This trial provides one of the most methodologically rigorous head-to-head comparisons of VR-based and traditional in-vivo exposure for SAD in the published literature. The pre-registered superiority hypothesis - that VR would be more effective than in vivo - was confirmed on the LSAS-SR primary outcome at both post-treatment and 6-month follow-up. This is a stronger conclusion than “VR is non-inferior” or “VR is equivalent”: it is “VR can be more effective” under the conditions of this trial.

The practical-burden finding (SWEAT) is equally important for clinical translation: the Bouchard team operationalized and measured the effort required to conduct exposure, finding VR significantly less effortful to deliver. This is the kind of evidence clinicians and service managers ask for when deciding whether to adopt new technology.

The working-alliance finding is a useful reminder for any team introducing VR into clinical practice: the modality of exposure does not replace the therapist relationship. Working alliance predicted LSAS-SR change strongly in BOTH conditions; treatment credibility (whether patients found the rationale plausible) did NOT predict response. The active ingredient remains the therapeutic relationship; VR is the medium of exposure delivery.

For Therapy withVR specifically: this trial did not use, evaluate, or compare against Therapy withVR. The hardware was an eMagin z800 head-mounted display and the environments were developed by Virtually Better and Klinger et al. The Bouchard trial is included in the Evidence Hub because it adds to the broader evidence base on immersive VR for social anxiety in adults, not because it relates to Therapy withVR. The Bouchard team’s broader program of work - including the more recent Delangle et al. 2026 (also in this Evidence Hub) - has extended their VR exposure approach into communication-adjacent populations.

Limitations

The authors explicitly flag the following:

• No independent clinical assessor for diagnostic interviews. Clinical outcomes were measured via self-report questionnaires; the authors note this raises the issue of objectivity. The behavioral assessment task with three independent masked raters partially addressed this.
• Individual format only. The trial tested individual CBT; group CBT is also widely used and equally effective. The design choice was made to isolate the effect of exposure modality.
• Modest total sample (n=59 across three arms). Replication with a larger sample, addition of physiological measures, and detailed analyzes of presence remain valuable for future work.
• In-vivo and in-virtuo scenarios did not match exactly. The differences between conditions might be explained in part by subtle differences in stimuli used, not only the exposure modality. The authors note that pairing participants on specific stimuli would have limited how therapists could tailor interventions, but acknowledge that exact-matched scenarios would increase trial validity.
• Single-site context. The trial was conducted at one laboratory at Universite du Quebec en Outaouais, which limits external validity. Multi-site replication is warranted.
• Bouchard/Robillard COI to disclose. Co-authors Bouchard and Robillard own equity in Cliniques et Developpement In Virtuo, a VR-development company - flagged in the paper’s Declaration of Interest, and reiterated in our funding section. The paper explicitly states that company did not create the environments used in this study, but the equity relationship is relevant background context. This is the same Bouchard equity relationship disclosed in the Delangle et al. 2026 paper (also in this Evidence Hub).