VIRTUS: virtual reality exposure training for adolescents with social anxiety – a randomized controlled trial

Social anxiety disorder (SAD) commonly emerges in mid-to-late adolescence and is associated with significant impairments across social, academic, and general well-being domains. Avoidance of feared social situations (e.g., public speaking, speaking up in class, meeting new people) maintains anxiety by preventing corrective learning. Exposure-based interventions are a key component of CBT for SAD, yet many adolescents do not receive or fully engage in exposure due to distress and perceived uncontrollability of real-life exercises. Virtual reality exposure (VRE) may address barriers by offering immersive, controllable, and engaging environments that align with adolescents’ technological interests. Adult studies and meta-analyses support VRE’s efficacy for SAD, but rigorous adolescent-focused trials are scarce and often limited to public speaking or lack controls. This study proposes a three-arm RCT to test whether VRE is efficacious and acceptable in adolescents with elevated social anxiety, compared to IVE and WL, and to investigate working mechanisms (expectancy violation, habituation, self-efficacy) and predictors of treatment response (clinical, personality, VR-related, treatment-related factors). The primary hypothesis is that both VRE and IVE will reduce social anxiety relative to WL at post-assessment and that VRE will be as effective as IVE over the long term (3- and 6-month follow-ups).

Prior research indicates CBT is effective for adolescent SAD, with moderate to large effects across pre-post, between-group, and follow-up comparisons. Exposure is a crucial active component but can be difficult for adolescents to engage with. VRE has shown efficacy in adults with SAD, sometimes comparable or superior to IVE, though some studies favor IVE. Adolescent-focused VRE studies suggest feasibility and reductions in public speaking anxiety and school-related anxiety, but limitations include small samples, lack of active controls, and focus on narrow social fears (public speaking). A well-powered RCT (Kahlon et al., 2023) demonstrated efficacy of self-guided gamified VRE for public speaking compared to WL. Theoretical models of exposure—emotional processing theory (habituation), inhibitory learning theory (expectancy violation), and self-efficacy theory—are well supported in IVE but less explored in VRE; concerns exist about whether virtual environments allow for true expectancy violation. Predictors of treatment outcome in youth anxiety include clinical severity, comorbidity, personality traits (e.g., BIS/BAS), attachment, immersion propensity, attitudes toward technology, treatment preference, working alliance, motivation, and expectations. This trial will address gaps by including active comparators, larger samples, broader social fears, mechanism assessment, and predictor analysis.

Design: Three-arm randomized controlled trial (VRE, IVE, WL) with mixed-subjects design (condition between-subjects; time within-subjects). Assessments at baseline (T1), mid-training (T2), post-training (T3), 3-month (T4), and 6-month (T5) follow-ups. Independent assessors blinded to condition conduct baseline, post, and follow-ups; participants and trainers are aware of assigned condition. Participants: Target N=120 adolescents (n=40 per arm), aged 12–16, with subclinical to mild/moderate SAD. Recruitment via Belgian secondary schools, social media, youth organizations, GP offices, libraries. Guardian informed consent and adolescent assent obtained. Screening phase 1: SPIN (cut-off ≥19) identifies elevated social anxiety. Screening phase 2: semi-structured diagnostic assessment with SCID-5 Junior (Dutch), ADIS-C CSR ratings, and MINI-KID to confirm inclusion criteria and assess severity/comorbidities. Eligible participants receive free training and monetary compensation for assessments. Randomization: External assistant generates blocked randomization (Sealed Envelope), 1:1:1 allocation in blocks of 12 (4 per group). Sealed envelopes opened post-baseline by central researcher to inform logistics. Interventions: Seven sessions (60–90 min). Session 1: psychoeducation, treatment rationale, individualized hierarchy of feared situations, brief 10-min exposure. Sessions 2–6: two exposure sets (~20 min each) per session with pre-discussion (threat expectancies) and post-discussion (outcomes, surprise). Session 7: 20-min exposure, reflection, relapse prevention. Themes target common adolescent social fears (performance, intimacy, scrutiny, assertiveness). Exposure duration standardized across conditions; no homework encouraged. Trainers hold master’s degrees in psychology; supervised by experienced CBT therapists. Sessions audio-recorded when possible; adherence monitored. VRE: Tailored VR scenarios simulating class participation, assertiveness, ordering food, group presentations, engagement with strangers. Software: Social Worlds v4.1 (CleVR) offering classrooms, playgrounds, sports courts, shopping areas, social gatherings; customizable interactions (avatars, dialogue, group size), trainer-controlled in real time (voice modulation, avatar role). Hardware: high-performance Alienware laptop; Meta Quest 3 headset and controllers; noise-cancelling headphones (user); Touch laptop; PELTOR headset with microphone (trainer). IVE: Tailored in vivo exercises (e.g., asking directions, returning items, ordering/eating in public) at training location or nearby settings (supermarkets, transit, cafes, buses), aligned with participant case conceptualization and session themes. Primary outcomes: Social anxiety (verbal): SPAI-18; LSAS-CA avoidance subscale. Social anxiety (behavioral): BAT peak SUDS (0–100) following a standardized 3–5 min online conversation with a confederate (MS Teams), with controlled confederate behavior and timing signals. Secondary outcomes: LSAS-CA fear subscale; SPWSS (weekly); BRS; MAFS (peer relationships, overall functioning); RCADS (anxiety, depression); SEQ-C social efficacy subscale; ASCQ (frequency and believability of social cognitions). BAT-related measures: Anticipatory SUDS before BAT; total task duration; confederate ratings of anxiety (1–5) and skill (1–5); psychophysiology during BAT—HR, HRV, SCL using MP160 (EDA 100D, ECG 100D), standard electrode placements, 5-min neutral baseline, AcqKnowledge v5.0 analysis. Mechanisms (Objective 2): Expectancy violation and change via pre/post VAS (0–100): threat expectancy likelihood (pre), threat occurrence (post), adjusted likelihood (pre vs post). Violation = pre likelihood – post occurrence; change = pre – post likelihood. Learning rate estimated per participant (0–1) per Pittig et al. to quantify translation of violation into change. Post-exercise VAS for surprise, relief, and relief pleasantness. Habituation via SUDS: within-session (peak – end SUDS per exercise) and between-session (change in peak SUDS across sessions). Self-efficacy VAS alongside SEQ-C. Predictors (Objective 3): Clinical—SPIN at screening; SCID-5 Junior SAD diagnosis and ADIS-C CSR; RCADS depression; other comorbidities (global anxiety, substance use, suicidality, self-harm via SCID-5). Personality/traits—BIS/BAS; ECR-RS (attachment avoidance/anxiety); VAS for motivation, need for relatedness, openness. VR-related—ITQ involvement; UTAUT Attitude and Outcome Expectancy subscales; self-reported VR/game experience frequency. Treatment-related—WAI-S (Session 1); preferred modality (binary VRE vs IVE); CEQ after Session 1. Acceptability (Objective 4): Post-treatment semi-structured exit interviews (randomized subsample ~10–16 per condition) covering experience, pros/cons, preferences, improvements; independent interviewer. CEQ after training rationale; CSQ-8 at post-assessment; weekly session VAS for difficulty, usefulness/relevance, safety behaviors, motivation; single-item VR presence in VRE; refusal and drop-out (<5 sessions) recorded. Monitoring and data management: GDPR-compliant data handling; anonymized Qualtrics assessments stored on secure Utrecht University server; pseudo-anonymization via participant codes; restricted access via MFA. No interim analyses; collection continues to target sample size. Safety: exclusion of acute suicidality/self-harm; SAE procedures and ethics reporting; annual ethics progress reports. Analyses: Power estimates via G*Power: for 3-group comparison (pre/mid/post), α=0.05, power=0.80, effect size f=0.25 → N=108; for 2-group longitudinal (T1–T5), α=0.05, power=0.80, f=0.25 → N=78; planned N=120 (40/arm) to account for attrition. Primary verbal outcomes: LMM (lme4 in R) with time nested in individuals; fixed effects group, time, group×time; random intercepts and tested random slopes for time. Follow-up LMM for active arms including WL participants later randomized. Behavioral BAT peak SUDS: 2×3 mixed ANOVA (time: baseline vs post; condition: VRE, IVE, WL). Secondary outcomes: LMMs analogous to primary. ITT analyses: include all randomized participants; missing data treated under MAR; multiple imputation for those with ≥5 sessions. Mechanisms: LMM mediation framework with bootstrap indirect effects; moderation by condition via interaction terms; model fit via likelihood ratio tests and AIC. Predictors: multiple linear regressions predicting post (and exploratory follow-up) primary outcomes, controlling for baseline severity and treatment-related variables. Acceptability: thematic analysis of interviews (Braun & Clarke); t-tests and chi-squared tests compare session-related quantitative variables between VRE and IVE.

This is a study protocol; no empirical findings are reported. Hypotheses: (1) Both VRE and IVE will significantly reduce social anxiety symptoms versus WL at post-assessment; (2) VRE will be as effective as IVE, with comparable outcomes at 3- and 6-month follow-ups; (3) Mechanisms will include expectancy violation, habituation, and increased self-efficacy; (4) Predictors of response will span clinical severity, comorbidity, personality/traits, VR-related factors, and treatment-related variables. Planned sample size: N=120 (40 per arm). Power analyses indicated N=108 sufficient for 3-group comparisons and N=78 for 2-group longitudinal comparisons.

The protocol outlines the first large-scale RCT in adolescents with generalized social anxiety to directly compare VRE with IVE and a WL control. By isolating exposure (without adding cognitive restructuring), the study aims to clarify the specific contribution of exposure as an early, accessible intervention suited to adolescent schedules. It integrates assessment of three leading mechanisms (expectancy violation, habituation, self-efficacy) without enforcing mechanism-driven criteria during exercises, providing ecological insight into how exposure operates in practice. Identifying predictors of outcome may support personalized assignment to VRE or IVE, optimizing clinical effectiveness and resource allocation. Qualitative interviews and quantitative acceptability metrics will address the assumption that VRE is more appealing and potentially enhances adherence among adolescents. Utilizing dynamic, trainer-controlled virtual environments addresses limitations of static or scripted VR content in social anxiety, potentially improving ecological validity and responsiveness to participants’ behavior.

This RCT will evaluate whether VR-based exposure can serve as an effective, acceptable early intervention for adolescents with subclinical to moderate social anxiety, benchmarked against in vivo exposure and waitlist control. The study’s contributions include isolating exposure effects, elucidating mechanisms of change, and identifying predictors to inform personalized treatment. Findings may broaden treatment options by validating VRE as a viable, engaging modality for youth, and guide refinement of exposure-based interventions through mechanism and moderator insights. Future research can build on these results to tailor VR environments and protocols to individual profiles and to test scalable implementations in school and community settings.

As a protocol, no outcomes are yet available. Anticipated and acknowledged limitations include: (1) exclusion of adolescents with ASD, which enhances internal validity for expectancy measures but limits generalizability; (2) participants and trainers cannot be blinded to condition; (3) exposure is delivered without cognitive restructuring or homework, which may differ from standard CBT packages; (4) mechanisms are measured rather than experimentally manipulated, constraining causal inference; (5) reliance on self-report VAS and SUDS for mechanism indices; (6) recruitment of a non-referred sample may limit generalizability to clinical referrals; (7) potential attrition and missing data, mitigated via ITT and multiple imputation; (8) WL participants later randomized to active arms may introduce variability in follow-up analyses.