Augmenting self-guided virtual-reality exposure therapy for social anxiety with biofeedback: a randomised controlled trial

Social Anxiety Disorder (SAD) involves marked fear of social situations under others' scrutiny and is common, underdiagnosed, and undertreated, posing a public health concern. Virtual reality exposure therapy (VRET) elicits realistic emotional responses and can be as effective as in vivo exposure. A self-guided VRET approach reduces reliance on therapists and costs, and has shown efficacy for panic disorder, specific phobias, and social anxiety/public speaking anxiety (PSA). Key mechanisms relevant to engagement and outcomes include avoidance behavior driven by uncertainty, perceived control over exposure, and sense of presence within VR. Physiological arousal biomarkers related to social anxiety include elevated heart rate, reduced heart rate variability, and frontal alpha asymmetry (FAA). Biofeedback may support anxiety regulation by increasing awareness of and control over physiological arousal, potentially enhancing perceived control and VRET responsiveness. The study aimed to evaluate whether adding continuous biofeedback to self-guided VRET enhances improvements in PSA, social anxiety, confidence, perceived arousal/anxiety during sessions, and physiological arousal (heart rate and FAA), whether improvements sustain over one month, whether perceived control explains long-term improvement, and whether presence predicts responsiveness.

VRET has demonstrated comparable efficacy to in vivo exposure for SAD, with many patients preferring VR. Self-guided VRET reduces need for trained therapists and is already implemented in healthcare settings (e.g., agoraphobia). Prior studies show self-guided VRET reduces social anxiety more than waitlist, with PSA improvements comparable to therapist-led VRET and sustained long-term, though single sessions may be unreliable and prior work often relied on self-report. Mechanisms influencing avoidance and engagement include perceived control and sense of presence; presence has been linked to outcomes in self-guided and therapist-led VR therapies, though findings vary by phobia type. Biofeedback improves interoceptive accuracy, emotion regulation, and anxiety outcomes (meta-analyses support HRV biofeedback effects on stress/anxiety, modest benefits versus active controls). In VR therapy contexts, biofeedback added to VR has produced modest reductions in self-reported anxiety and heart rate, but most studies delivered a single session. FAA (rightward asymmetry) is associated with avoidance/withdrawal; evidence on its relation to social anxiety is mixed, suggesting it may relate more to social withdrawal than anxiety per se.

Design: Randomized controlled trial (single-blind) comparing self-guided VRET with and without biofeedback across three weekly sessions with one-month follow-up. Participants: 665 screened; 397 scored ≥32 on SPIN; those scoring >19 on SPIN invited to RCT until target n≈75 reached. Final sample n=73 completed session 1; randomized to VRET+biofeedback (n=38) or VRET alone (n=35). Baseline SPIN mean=46 (SD=10), range 20–67. Groups were matched on age, sex, ethnicity, social anxiety diagnosis, and other psychiatric diagnoses. Inclusion: adults (≥18), normal/corrected vision, East Midlands community sample. Measures: Screening—Social Phobia Inventory (SPIN). PSA—Personal Report of Confidence as a Speaker (PRCS short form) and Public Speaking Anxiety Scale (PSA). Social anxiety—Liebowitz Social Anxiety Scale (LSAS: fear/avoidance in performance and social situations). Fear of negative evaluation—BFNE. Perceived control proxy—BIS-appraisal subscale (in-house reinforcement sensitivity measure). Presence—Witmer & Singer Presence Questionnaire (five subscales) post-session 1. In-VR Visual Analogue Scales (VASs) for avoidance of giving a speech (pre/post session), arousal, and anxiety at each pause. Physiological measures: Heart rate via Microsoft Band 2 wristband (sampled 10 Hz), averaged per minute/four-minute speech block; EEG via Muse headband (AF7/TP9 left; AF8/TP10 right; FPz reference) sampling 220 Hz; frontal alpha power averaged per minute and FAA computed as left minus right alpha. VR hardware/software: Samsung Gear VR with Galaxy S7 running Unity-based VRET app; Java plugin integrated Microsoft Band data. Intervention: Three sessions (≈60 minutes each). Each session involved delivering a 20-minute speech (topic: going on a holiday), segmented into five 4-minute blocks with 1-minute pauses for VAS ratings and adjusting five graded exposure elements (audience size; audience reaction; speaker distance; number of prompts; salience of self). Participants were encouraged to increase exposure at their own pace. Biofeedback group viewed two oscillating vertical bars in the VR environment indicating heart rate (red) and FAA (blue), with instructions to lower arousal. Procedure: Baseline online assessments (BFNE, LSAS, PRCS, PSA). PRCS after each session and at 4-week follow-up; Presence Questionnaire after session 1; BFNE, LSAS, PSA after session 3 and at follow-up. Randomization single-blind; sessions conducted in the same lab; ambient temperature monitored. Ethics: Nottingham Trent University School of Social Sciences (No. 2017/115). Compensation: £15 per session. Statistical analysis: Missing data—completers vs non-completers compared; multiple imputation via MCMC used where appropriate. Hypothesis tests—2×2 mixed ANOVAs (Group × Time) for PSA and social anxiety (baseline to post-treatment). 2×3×4 mixed ANOVAs (Group × Session × Block) for VAS arousal/anxiety. 2×2 ANOVAs (Group × Time: first minute of first session vs last minute of third session) for heart rate and FAA; additional ANOVA across minutes and blocks in first session for physiological measures. Quadratic contrasts for baseline/post-treatment/follow-up trajectories. ANCOVAs with BIS-appraisal covariate to test perceived control effects. Multiple regressions: Presence subscales predicting change-relative-to-baseline in PSA and social anxiety.

Completion rates: VRET+biofeedback completers n=34 (89%), 29 (76%), 25 (66%), 19 (45%) at sessions 1–3 and follow-up; VRET-alone completers n=29 (81%), 26 (72%), 25 (69%), 23 (64%). No group differences in dropout (χ²<1.25, p>0.228). Completers scored higher on presence self-evaluation than non-completers at session 1. End-of-treatment: Significant main effect of time on PSA (F(1,71)=42.23, p<0.001), PRCS (F(1,71)=53.39, p<0.001), and avoidance of giving a speech (F(1,71)=116.24, p<0.001), with clinically meaningful Reliable Change Indices (RCI): PSA -4.4, PRCS -5.4, avoidance -7.6. BFNE and LSAS did not show significant time effects at post-treatment (F(1,71)<2.66, p>0.108). Trend-level group differences favored VRET-alone for PSA at end-of-treatment (RCI_PSA -4.5 VRET-alone vs -2.1 VRET+biofeedback). In-VR arousal/anxiety: Group × Session × Block interaction for VAS arousal (F(5.6,130)=2.24, p=0.046); the biofeedback group exhibited lower and more steadily declining self-reported arousal in sessions 2–3, while the VRET-alone group showed haphazard changes. No comparable interaction for VAS anxiety (F(4,91.8)=0.66, p=0.620). One-month follow-up: Significant time effects across all PSA and social anxiety measures (e.g., PRCS, avoidance, PSA, BFNE, LSAS subscales; F>10, p<0.001), with clinically meaningful RCI values (e.g., PRCS -5.2; avoidance -6.3; PSA -6.3; BFNE -3.4; LSAS fear of performance -7.3; LSAS fear of social situations -6.7; LSAS avoidance of performance -6.5; LSAS avoidance of social situations -6.4). Group × time interactions at follow-up for avoidance of giving a presentation, PSA, LSAS Fear of Performance, and LSAS Fear of Social Situations (medium effect size for LSAS Fear of Performance; small for BFNE and LSAS avoidance; negligible for PRCS, avoidance, PSA). VRET-alone showed greater end-of-treatment improvement for avoidance and PSA; differences diminished by follow-up, while VRET-alone gains from post-treatment to follow-up on LSAS Fear subscales exceeded those in biofeedback. Physiological arousal: Across all sessions/minutes, heart rate decreased over time (F(1,16)=6.24, p=0.024). In session 1, heart rate declines appeared steadier block-to-block in biofeedback, though Group × Block × Minute contrasts were not significant (F(3,48.3)=0.32, p=0.81). FAA across all sessions showed no main effect of time (F(1,20)=0.33, p=0.573); however, in session 1, a Group × Block interaction (cubic) was significant (F=5.86, p=0.021), indicating a steady FAA decline in biofeedback versus haphazard changes in VRET-alone. Mechanisms: Covarying BIS-appraisal eliminated time effects across PSA/social anxiety, suggesting perceived control/risk appraisal explains improvement. Presence: Greater sense of presence (particularly Possibility to Examine) at session 1 predicted improvement in PSA by end of treatment (Regression: F(5,55)=5.58, p<0.001; standardized beta for Possibility to Examine=0.53, p<0.001).

Self-guided VRET produced clinically meaningful improvements in PSA by the end of treatment and broader social anxiety improvements at one-month follow-up in a community sample with high social anxiety. Biofeedback did not enhance overall symptom reductions relative to VRET-alone but did steady physiological (FAA, heart rate) and perceived arousal within sessions, consistent with biofeedback improving interoceptive awareness and emotion regulation. The association between presence (ability to examine the VR environment) and PSA improvement underscores the role of presence in eliciting anxiety that can be therapeutically reduced, and in facilitating engagement with graded exposure. Perceived control, indexed by BIS-appraisal, accounted for improvements, suggesting that individuals with greater capacity for risk appraisal and control may derive more benefit from self-guided exposure. At follow-up, VRET-alone sometimes showed greater gains (e.g., LSAS fear subscales), implying that biofeedback as implemented may yield temporary arousal regulation benefits without additional long-term symptom improvement, potentially due to challenges in sustaining attention to feedback and lack of explicit training (e.g., breathing) or reinforcement. Integrating structured biofeedback training (operant learning, gamification) may strengthen enduring benefits. Overall, findings support self-guided VRET as an accessible therapy that can reduce reliance on therapists, while highlighting the importance of presence and perceived control mechanisms and optimizing biofeedback delivery.

Three weekly sessions of self-guided VRET yielded clinically meaningful reductions in public speaking anxiety and broader social anxiety, sustained up to one month post-treatment, accompanied by objective reductions in heart rate. Adding biofeedback stabilized physiological and perceived arousal during exposure and showed marginal advantages for social avoidance in performance situations, but did not clearly outperform VRET-alone on long-term symptom reductions. Presence within the VR environment and perceived control (risk appraisal) were key correlates of improvement. Future work should enhance biofeedback with training (e.g., breathing/relaxation), operant reinforcement and gamification, and assess generalization to real audiences to increase ecological validity. Self-guided VRET offers practical benefits for service delivery, potentially reducing wait times and therapist burden while supporting engagement with in vivo therapies.

High attrition by follow-up (biofeedback: 55% dropout; VRET-alone: 36%), with non-completers reporting lower presence; limited realism/immersion of the virtual audience noted by participants; adverse effects were not routinely monitored and included distress, anxiety, and physical discomfort; absence of guidance on how to lower heart rate during biofeedback may have undermined coping confidence and sustained improvement; small subsamples for physiological analyses reduce power; potential gender bias in completion of in-session VAS arousal/anxiety; single-site laboratory setting; reliance on virtual audience may limit real-world generalization; continuous biofeedback without structured training or reinforcement may have led to reliance on momentary feedback rather than durable regulation skills.