Virtual Reality-Assisted, Single-Session Exposure for Public Speaking Anxiety: Improved Self-Reports and Heart Rate but No Significant Change in Heart Rate Variability

Abstract

Introduction: This study examines the combined use of objective physiological measures (heart rate [HR], heart rate variability [HRV]) and subjective self-reports to gain a comprehensive understanding of anxiety reduction mechanisms—specifically, habituation—in the context of Virtual Reality Exposure (VRE) for public speaking anxiety (PSA). The present study evaluated whether a single-session, personalized VRE intervention could effectively reduce PSA. Methods: A total of 39 university students (mean age = 20.97, SD = 3.05) with clinically significant PSA were randomly assigned to a VRE group or a control group. Participants completed a 2 min speech task before and after the intervention and reported subjective distress (Subjective Units of Distress, SUDs), public speaking confidence (Personal Report of Confidence as a Speaker, PRCS), and willingness to speak in public. Heart rate (HR) and heart rate variability (HRV; RMSSD) were recorded at baseline and during speech tasks. The VRE protocol used personalized, hierarchical exposure to virtual audiences, with repeated trials until a criterion reduction in SUDs was achieved. Non-parametric analyses assessed group and time effects. Results: VRE participants showed significant reductions in subjective distress (p < 0.001) and HR (p < 0.001), with HR returning to baseline post-intervention. No such reductions were observed in the control group. Willingness to speak improved significantly only in the VRE group (p = 0.001). HRV did not differ significantly across time or groups. Conclusions: A single, personalized VRE session can produce measurable reductions in PSA, particularly in subjective distress and autonomic arousal, supporting habituation as a primary mechanism of change, even after one session. The lack of HRV change suggests that emotion regulation may require more prolonged interventions. These findings support VRE’s potential as an efficient and scalable treatment option for PSA.

1. Introduction

Effective communication, including public speaking, remains an important skill in many professional, educational, and other settings; however, it often causes significant anxiety [1] and avoidance of public speaking opportunities. Public speaking anxiety (PSA) represents a sub-category of Social Anxiety Disorder (SAD), the fear of negative evaluation in social or performance situations, meeting the ‘performance only’ specifier of DSM-5 SAD criteria [2]. Clinical or subclinical PSA is common in the general population, including among university students [3,4,5].

Cognitive–behavioral therapy (CBT) interventions for SAD and PSA, and other anxiety disorders, involve exposure to phobic stimuli, delivered in vivo, in virtuo, or using a hybrid approach (e.g., Virtual Reality Exposure—VRE) [6,7]. The present study’s focus on single-session VRE aims to provide support for a cost-effective and time-efficient alternative to traditional multi-session protocols. VRE effectively provokes anxiety and autonomic arousal [8,9,10] and yields favorable outcomes for SAD (for meta-analyses, see [11,12,13,14]), being superior to waitlist control but comparable to traditional modalities [7,15]. Similar conclusions hold for PSA [16,17,18] (for meta-analyses, see [19,20]). Daniels et al. [20] and Boetje and van Ginkel [21] conclude that VRE is an effective alternative to other modalities, both manageable and cost-effective. Findings are encouraging for the wider adoption of VRE for PSA, especially considering the increasing affordability of VRE technologies and in relation to initial evidence that even very brief VRE can be impactful, something that might boost accessibility, acceptability, and cost-effectiveness. In this direction, the present study aims to verify that even a single exposure, when conducted in a systematic and personalized manner, focused on bolstering the effects of a central mechanism of action, habituation, can yield substantial reductions in subjective distress and objective measures of arousal. However, while brief VRE is effective for phobias [22,23], evidence for PSA is inconsistent. Takac et al. [24] found that a single VRE session was too brief to produce significant improvements, a conclusion also supported by the meta-analysis of Lim, Aryadoust, & Esposito [11], who found that successful interventions for PSA included multiple sessions. However, Linder et al. [25] and Banakou et al. [26] reported promising results in a single session, especially when it included psychoeducation and personalized exposures. Therefore, a gap remains in determining whether and under what circumstances single-session VRE protocols for PSA are effective, to reconcile mixed existing evidence. In this study, we address this gap and incorporate elements that are anticipated to maximize habituation-driven effects on anxiety, specifically personalized hierarchies and multiple exposures to the same hierarchy level, until a criterion is reached, predicting that this would result in measurable reductions in PSA.

Understanding mechanisms of action is a critical aspect of research for improving VRE and anxiety interventions more broadly. For example, face-to-face therapies involve personalized fear hierarchies, graduated exposure, and documentation of progress at each step [27]; VRE protocols may require comparable customization to boost their effectiveness. Furthermore, to elucidate mechanisms of action, treatment evaluation should extend beyond pre and post measures to include changes during the intervention process. It should also include indices of change beyond self-reports and their limitations [28], for example, physiological [29] and behavioral indicators of treatment effects. Specifically, evidence of reduction in autonomic responses, e.g., heart rate (HR), during challenges using VRE, paralleling improvements in subjective anxiety, would suggest that the mechanism of habituation, i.e., anxiety, decrease in the presence of feared stimuli after repeated exposures has occurred [30,31,32]. Most prior studies of VRE for PSA have not offered comprehensive assessments of anxiety change in order to isolate active mechanisms. We address this gap by incorporating subjective and physiological outcome measures to assess the effects of our single-session VRE protocol, which relies on habituation as a mechanism of action.

Another mechanism that may contribute to the effects of VRE on PSA is emotion regulation. Consistent with modern conceptualizations of exposure therapy that emphasize inhibitory learning, a change in expectancies, and the ability to tolerate distress, along with or instead of habituation (e.g., [33]), improvement in PSA may occur even without a significant decrease in arousal and subjective anxiety. These changes could be related to improved emotion regulation, reflected in the ability to remain in distressing situations. Therefore, more work is needed to fill gaps in our understanding of mechanisms of change in VRE and their similarity to those considered central for in vivo interventions. We test the hypothesis that a single-session, individualized VRE intervention can reduce subjective and physiological indices of PSA, primarily through habituation. In parallel, however, we explore whether heart rate variability (HRV), a proxy of emotion regulation, also changes after this intervention, to address the possibility that change occurs via improved emotion regulation.

By incorporating both heart rate (HR) and heart rate variability (HRV) as objective measures, this study provides novel insights into the autonomic underpinnings of VRE’s effects, contributing to a more comprehensive understanding of mechanisms of action in operation during brief but intensive VRE for PSA. Heart rate variability (HRV), the fluctuation in time intervals between heartbeats [34], reflects the interaction between the Sympathetic and Parasympathetic Nervous Systems [35]. High HRV indicates healthy heart function [36] and correlates with emotion regulation [37,38]. Conversely, low HRV is associated with anxiety, depression, emotional dysregulation, SAD, and PSA [39,40]. HRV is rarely assessed in VRE. Kothgassner et al. [41] showed reduced HRV during feared situations in people with PSA. Biesmans et al. [42], however, did not find a significant HRV change by VRE, suggesting that more research is needed.

Present Study

This study evaluates the effects of a very brief VRE-assisted intervention for PSA on measures of habituation and emotion regulation. Specifically, we examined a single-session graduated VRE procedure using a personalized fear hierarchy and repeated exposures to each stimulus, until anxiety reduction is reported. This design mirrors the principles of face-to-face personalized exposure protocols, especially those that emphasize habituation. Outcome measures tap into all aspects of the anxiety response, i.e., subjective distress, avoidance, and physiological arousal. We also included HRV as an index of emotion regulation, specifically Root Mean Square of Successive Differences (RMSSD), a common and reliable marker of parasympathetic control, which is less affected by respiration patterns compared to other HRV measures, making it a preferred selection over other markers as respiration was not assessed. We evaluated anxiety reduction not only between the treatment and control groups but also within participants over the course of the session. Specifically, we examined (a) whether virtual reality-assisted exposure (VRE) can reduce subjective, behavioral/avoidance, and physiological symptoms of PSA in a single session and (b) whether VRE results in anxiety habitation, as a mechanism of action, and/or whether it affects emotion regulation.

We expanded on van Dis et al. [43] by using more personalized exposures, asking participants to construct a hierarchy of the available VR environments, to achieve high levels of experienced anxiety and by repeating exposures until habituation was achieved. Our sample included university students scoring in the clinical range of PSA. Although this is not a treatment-seeking population, students frequently need to speak in public (e.g., class presentations, graduate school, and job interviews), a task that many dread and avoid [44,45]. Showing that brief VRE can reduce PSA in this population has implications for its wide implementation, both for clinical samples and for those with subclinical symptoms, who experience impairment, lost opportunities, and distress.

We expected participants receiving VRE to show anxiety reduction (subjective, physiological) and increased willingness for public speaking, in comparison to a no-exposure control group (CG). We also expected that the VRE group (VREG) would evidence increased HRV post exposure compared to controls, indicating improved emotion regulation, alongside habituation, indexed by reduced self-reported distress and HR progressively within exposure repetitions.

2. Materials and Methods

2.1. Participants

Greek-Cypriot college students (N = 254) were screened for PSA using the Personal Report of Confidence as a Speaker (PRCS; Paul, 1966) [46], with a clinical cutoff of >17 (Harris et al., 2002) [1]. Clinical cutoffs were used as a selection criterion to ensure that PSA was significant enough to respond to treatment and to increase sample homogeneity. A total of 40 students (VREG = 20; CG = 20) meeting the cutoff (M = 22, SD = 1.51 for males, M = 21, SD = 3.17 for females) consented to participate in exchange for extra credit. Participants with a history of PSA or SAD treatment were excluded. The final sample (N = 39, after excluding one dropout from the control group) consisted of 34 females and 5 males, primarily undergraduate psychology students. Participants were between 18 and 30 years old. All participants were fluent Greek speakers and had no prior experience with virtual reality- or exposure-based therapy. Power analysis indicated that 54 participants were needed for adequate power (0.95) in a mixed-factorial ANOVA. However, previous studies (e.g., Culver et al., 2012) [47] suggest that a sample of >34 would achieve power for a moderate-to-large effect size. Finally, 20 and 19 participants in the VREG (5 males, 15 females) and CG, respectively (19 females), participated, while 19 VREG and 16 CG participants were included in HR/HRV analyses, due to the loss of some noisy signals. Recruitment and group assignment were conducted on a rolling basis: each new consenting participant was assigned to an alternate group.

2.2. Measures

The Personal Report of Confidence as a Speaker (PRCS; [46]) is a 30-item self-report scale, answered true or false, that assesses behavioral and emotional reactions to public speaking. Scores range from 0 (no fear) to 30 (very high fear). The scale has demonstrated good internal consistency, with Cronbach’s alpha ranging from 0.62 to 0.94 [48,49].

Subjective Units of Distress (SUD) is a single-item scale rated 0–100, commonly used in CBT during exposure tasks [50,51]. It serves to measure the intensity of distress during various tasks.

The willingness-to-speak-in-public rating is a single item rated 1 (=lowest willingness) to 5 (=highest willingness), asking participants to rate their readiness to give a speech in front of an audience in the future. This measure was taken after each of two speeches to gauge participants’ confidence in future public speaking.

Psychophysiological Measures: Electrocardiographic activity was recorded using a Lead I configuration (Lead I EKG; BIOPAC Systems Inc., Goleta, CA, USA) for two 5 min baselines at the beginning of experiment, before each of the two speeches, and during the speeches, sampled continuously at 1000 Hz using BIOPAC MP150 for Windows and Acknowledge 3.9.0 software and two Ag/AgCl disposable electrodes placed on each forearm, filtered by the ECG100C bio-amplifier (BIOPAC Systems, Inc., Goleta, CA, USA) and converted to beats per minute (BPM). For HRV, the EKG was examined visually for artifacts that were corrected manually. Then, data were converted to inter-beat intervals and processed with ARTiiFACT, 2.13 [52], which extracted RMSSD.

Procedure

Participants initially completed an online battery, which included the PRCS, to assess PSA. Those who met inclusion criteria attended a laboratory session, where they were given an explanation of the study, provided written consent, and were fitted with electrodes. Next, a 5 min baseline was recorded during which participants engaged in a standardized relaxation protocol involving ‘silent repetitive counting to clear the mind’ [53]. Subsequently, they were instructed to prepare a 2 min semi-structured speech (e.g., [54]), incorporating their “name, age, field of study, interests, three personal strengths, three weaknesses, and future goals”, to enhance self-awareness and self-evaluation.

Once ready, participants delivered their speech to a female graduate student (SN) acting as the audience. The speech was terminated after 2 min, after which participants rated SUDs during the speech and their willingness to give a speech in the future. The same procedure (baseline, speech, ratings) was repeated after the exposure phase for the VREG and questionnaire rating phase for the CG. Lastly, participants were debriefed, received information on PSA and its treatment, and were given a list of psychological service providers, if they requested it.

This study followed a structured three-phase procedure (Scheme 1):

• Screening and baseline assessment: Online PRCS screening—all eligible participants completed baseline HR/HRV recordings and a pre-exposure speech task with SUD, PRCS, and willingness-to-speak ratings.
• Intervention or control condition: The VRE group underwent a personalized, graduated VRE session based on individual fear hierarchies and repeated exposures until SUD reduction; the CG completed neutral questionnaires.
• Post-intervention assessment: All participants completed the same 2 min speech task, post-exposure HR/HRV recordings, and repeated self-report measures (SUDs, PRCS, willingness to speak).

VRE Group: Following the first speech, participants in the VREG underwent graduated exposure to seven virtual audiences () of varied aversiveness: completely static, mostly neutral, somewhat interested, disinterested, bored, rude/impatient, and extremely rude. These are pre-prepared standard VR scenes, created using the Vizard VR Toolkit developed by WorldViz LLC (Santa Barbara, CA, USA), integrated with BIOPAC’s data acquisition systems MP150 system (BIOPAC Systems Inc., Goleta, CA, USA) from BIOPAC.com (accessed on 17 June 2025). The seven audiences varied in the movements and expressions of the audience members, who consisted of men and women characters, sitting in a classroom-type setting. Participants rated their SUDs after viewing each audience for 20 s. Based on these ratings, they constructed a hierarchy of their five most distressing audiences, in order from least to most fearful, not including their two least-feared audiences to keep the task challenging.

Participants were then exposed to each audience for 1 min while delivering their speech aloud. After each exposure, they provided an SUD rating. If their distress did not decrease by at least 20% after three exposures, they received a fourth exposure, and at most, a fifth. The threshold of a 20% decrease in SUDs was chosen to ensure a meaningful reduction in distress, reflecting a change of approximately 1SD from their mean SUDs during the first speech (M = 57.80, SD = 18.9), in line with Cohen’s (1988 [55]) rule of a minimally important difference.

Control Group: CG participants were asked to complete, instead of VRE, a questionnaire irrelevant to the experiment (personality and psychopathology measures) of approximately equal duration. Otherwise, they underwent the same procedures as the VREG: Initially they completed the baseline recordings, 2 min speech task, and ratings and, after questionnaire completion, the second speech and ratings. Participants were offered VRE later, if interested, for ethical reasons, but data are not included.

3. Results

3.1. Preliminary Analyses

Variables (HR, HRV-RMSSD, PRCS, SUDs) were checked for outliers through boxplots. Outliers > 3 times the interquartile range were removed. The exclusion of extreme outliers greater than three times the interquartile range (IQR) was necessary to maintain data integrity, as these values were likely attributable to measurement artifacts rather than genuine physiological responses, thereby reducing error variance and preserving the robustness of the findings. The removal of these outliers did not change the overall results. The number of outliers was too small to conduct formal sensitivity analysis. Dependent variables were examined for normality through the Kolmogorov–Smirnov test, the Shapiro–Wilk test, and the z-scores of the skewness and kurtosis. Significant violations of normality were identified in some of the variables. Taking this into consideration, all analyses were conducted with non-parametric methods. However, parametric repeated measures ANOVA results for the PRCS, SUDs, heart rate, and RMSSD are also provided in the Supplementary Materials (RM ANOVA Results) for further reference.

First, the VREG and CG were tested for equivalence at the pre-exposure baseline, to verify random assignment. The equivalent to a t-test non-parametric analysis is the Mann–Whitney test for two independent groups The Mann–Whitney tests indicated non-significant group differences for all variables: PRCS (VREG: Mdn = 19 (IQR = 5); CG: Mdn = 20 (IQR = 3), U = 205, Z = 428, p = 0.687, r = 0.006), pre-exposure SUDs (VREG: Mdn = 53 (IQR = 38.75); CG: Mdn = 70 (IQR = 30), U = 224, Z = 0.96, p = 0.35, r = 0.15), pre-exposure willingness to speak (VREG: Mdn = 2 (IQR = 2); CG: Mdn = 2 (IQR = 2), U = 184, Z = −0.17, p = 0.88, r = −0.02), baseline (5 min period before speech 1) HR (VREG: Mdn = 87 (IQR = 15.75); CG: Mdn = 83 (IQR = 22), U = 187.5, Z = −0.21, p = 0.82, r = 0.03), and baseline RMSSD (VREG: Mdn = 33.49 (IQR = 24.31); CG: Mdn = 29.06 (IQR = 26.98), U = 141.00, z = −0.39, p = 0.71, r = −0.06).

A Wilcoxon signed-rank test (with Bonferroni adjustment, p = 0.01) was conducted to examine whether VRE induces anxiety. SUDs were compared between speech 1 (speech without virtual audiences) and the first exposure to each of the five VRE audiences selected by participants (i.e., speech plus VRE). Significant comparisons were found only between speech 1 (Mdn = 58, IQR = 40) and audience 5 (Mdn = 90, IQR = 27.50), with Z = −3.26 and p = 0.001. However, mean SUDs for audiences 3, 4, and 5 were higher than mean SUDs for speech 1, suggesting that VRE to audiences at higher steps of the personalized hierarchies induces anxiety over and above repeated speech practice alone.

To verify that the virtual audiences included as anxiety-provoking stimuli indeed induce more anxiety as they increase in ranking on participants’ (VRE group only) fear hierarchy, we conducted a repeated measures ANOVA on Subjective Units of Distress, with audience x exposure as repeated variables. Audience represents the audience scene, 1–5, as ranked by participants (such that content of each level may be different for each participant), while exposure represents the fear rating at each time the participant was exposed to the same scene. Of interest was the main effect of audience and, in particular, the linear trend contrast for this factor, which was significant, with F(1, 4) = 23.73 and p = 0.008, showing that, indeed, virtual audiences ranked at higher ranks by participants (irrespective of the content each participant perceived as most fearful) induced progressively more fear.

For the main analyses, Kruskal–Wallis’s test and Friedman’s ANOVAs were used to assess the effect of the two independent variables (time: pre-exposure speech, post-exposure speech; group: VREG, CG) on HR and RMSSD. The Wilcoxon signed-rank test and Mann–Whitney tests (equivalent to the paired-samples t-test) were used for the subjective variables (PRCS, SUDs, willingness-to-speak rating).

3.2. Effects of Group and Time

Subjective PSA: A Wilcoxon signed-rank test revealed no significant change for the PRCS from pre- to post-exposure (pre Mdn = 19 (IQR = 4), post Mdn = 21 (IQR = 5), T = 200.5, z = −1.20, p = 0.22, r = 0.19), and this was true for both groups (VREG: pre Mdn = 19 (IQR = 5), post Mdn = 20.5 (IQR = 6.75), z = −1.90, p = 0.057, r = 0.42; CG: pre Mdn = 20 (IQR = 3), post Mdn = 21 (IQR = 4), z = −0.23, p = 0.81, r = 0.05). A Mann–Whitney test compared groups at post-exposure: the results showed that the PRCS scores of the two groups did not differ significantly (VREG: Mdn post = 20.5 (IQR = 6.75); CG: Mdn post = 21 (IQR = 4), U = 205, Z = 0.42, p = 0.68, r = 0.06).

A Wilcoxon signed-rank test revealed an overall effect of time (pre to post speech), with SUDs being significantly lower at post-exposure compared to pre-exposure (overall pre Mdn = 60 (IQR = 40), overall post Mdn = 50 (IQR = 40), T = 49.5, z = −3.9, p < 0.001, r = 0.62). When running the analysis by group, a significant decrease in SUDs was observed only for the VREG (VREG Mdn pre = 58 (IQR = 38.75), Mdn post = 30 (IQR = 23.75), z = −3.94, p < 0.001, r = 0.88) but not the CG (Mdn pre = 70 (IQR = 30), Mdn post = 70 (IQR = 30), z = −0.650, p = 0.51, r = 0.14). At post-exposure, SUDs of the VREG were significantly lower than those of the CG, with U = 344, Z = 4.36, p = < 0.001, and r = 0.69.

Willingness to speak in public: For this rating, medians remained the same from pre to post exposure (pre Mdn = 2 (IQR = 2), post Mdn = 2 (IQR = 1)) (but note that means differed: M pre = 1.97 (SD = 0.84), M post = 2.35 (SD = 0.84); T = 91, z = 3.41, p < 0.001, r = 0.54). When examining the effect of time separately for each group, the VREG median (Mdn pre = 2 (IQR = 2), Mdn post = 3 (IQR = 1), z = −3.27, p = 0.001, r = 0.73) but not the CG median (Mdn pre = 2 (IQR = 2), Mdn post = 2 (IQR = 2), z = −1, p = 0.31, r = 0.22) significantly increased from pre- to post-exposure. At post-exposure, the VREG had significantly higher ratings than the CG, with U = 106, Z = −2.52, p = 018, and r = 0.40.

HR: Changes in HR were assessed by comparing three timepoints (5 min initial baseline, speech 1, speech 2) using Friedman’s ANOVA. HR showed an overall statistically significant change across time, with χ² (2, n = 38) = 41.67, p < 0.001, and W = 0.54 (baseline Mdn = 83.0 (IQR = 49), speech 1 Μdn = 95.5 (IQR = 22.25), speech 2 Mdn = 86.0, IQR = 15.25). When assessed separately for each group, these effects were significant for both groups: for VREG, χ² (3, n = 20) = 50.74, p < 0.001, and W = 0.60 (baseline Mdn = 85 (IQR = 15.75), speech 1 Mdn = 95 (IQR = 20.25), speech 2 Mdn = 84, IQR = 10.25), and for CG, χ² (3, n = 18) = 43.59, p < 0.001, and W = 0.50 (baseline Mdn = 82.5, IQR = 22, speech 1 Μdn = 96, IQR = 26.25, speech 2 Mdn = 90, IQR = 12.25).

Wilcoxon signed-rank post hoc tests with Bonferroni correction and significance set to p = 0.016 were used to follow up on this finding, comparing pairs of timepoints within each group. These showed that the VREG showed a significant increase in HR from baseline to the first speech (z = −3.82, p < 0.001, r = −0.85), a significant reduction from the first to second speech (z = −3.78, p < 0.001, r = −0.084), and a non-significant difference from baseline to the second speech (z = −0.705, p = 0.48, r = −0.15), indicating that after VRE, HR returned to baseline. For the CG, there was a significant increase from baseline to the first speech (z = −3.55, p < 0.001, r = −0.81), a significant decrease from the first to second speech (z = −3.008, p = 0.003, r = −0.69), and a significant increase from baseline to the second speech (z = −2.89, p = 0.004, r = −0.66), suggesting that HR failed to return to baseline (Figure 1). The results indicate an appropriate increase in arousal from baseline to speech 1 for both groups, a reduction in arousal after VRE for participants receiving the intervention and after the wait period of completing questionnaires for the control group, and a return to baseline arousal by speech 2 for the VREG only.

RMSSD: A significant overall change in RMSSD was seen across the three timepoints (baseline Mdn = 30.35, IQR = 22.67, speech 1 Mdn = 28.01, IQR = 19.64, speech 2 Mdn = 36.01, IQR = 18.38, χ² (2, n = 41) = 11.90, p = 0.003, W = 0.14). The main effect of time was significant for the CG only (baseline Mdn = 37.32, IQR = 26.98, speech 1 = 30.61, IQR = 30.69, speech 2 Mdn = 35.55, IQR = 24.91, χ² (2, n = 17) = 9.88, p = 0.007) and not for the VREG (baseline Md = 37.67, IQR = 24.31, speech 1 Mdn = 31.87, IQR = 18.65, speech 2 Mdn = 37.10, IQR = 17.88, χ² (2, n = 18) = 4.11, p = 0.13).

Post hoc tests with Bonferroni correction (p set at 0.016) compared medians at each timepoint within each group. Comparisons were not significant for either group. When looking at mean RMSSD between baseline and the second speech, the change was smaller for the VREG, showing a fuller return to initial values after VRE (Figure 2), an effect which did not reach significance.

Given our relatively small sample size, which was below the size recommended by our initial power calculations, we computed post hoc observed power analyses to ensure that we had enough power for the observed effects of the main analyses of each outcome. The results showed excellent power for SUDS, HR, RMSSD and willingness-to-speak-in-public measures (observed power > 0.98). The only variable for which power to detect medium effects was limited (0.64) was the PRCS, which may in part explain the absence of significant post-intervention improvements in this measure.

3.3. Within-Subject Effects of Audience Type

A Friedman’s ANOVA assessed SUD differences between the five hierarchy levels, averaged across VRE participants. This showed a statistically significant effect of level, with χ² (4, n = 19) = 44.25 and p < 0.001. Post hoc analyses with Wilcoxon signed-rank tests and Bonferroni correction (p set at < 0.005) showed significant differences between most audience pairs, indicating a stepwise distress increase along the hierarchy.

Friedman’s tests were additionally conducted to assess within-subject changes in SUDs between exposures within the same hierarchy level. For all audience levels, significant changes (p-value set at 0.016) were shown across the first three compulsory exposures, except during exposure to audience 5 (i.e., the most fearful), for which the three exposures did not significantly differ (audience 1: χ² (2, n = 20) = 27.16, p < 0.001, W = 0.68; audience 2: χ² (2, n = 20) = 14.901, p = 0.001, W = 0.37; audience 3: χ² (2, n = 20) = 8.86, p < 0.001, W = 0.47; audience 4: χ² (2, n = 20) = 15.31, p < 0.001, W = 0.38; audience 5: χ² (2, n = 20) = 7.000, p = 0.030, W = 0.17). Means and standard deviations are shown in

Table 1. SUD means (M) and standard deviations (SD) for exposures within each audience level.

Audience Level
	Exp1	Exp2	Exp3	Exp4	Exp5
1	M = 55 (SD = 22.47)	M = 48 (SD = 22.14)	M = 37.7 (SD = 21.48)	M = 43.75 (SD = 22.87) N = 12	M = 41.11 (SD = 17.81) N = 9
2	M = 52.75 (SD = 18.60)	M = 45 (SD = 16.85)	M = 36.5 (SD = 17.99)	M = 33.21 (SD = 15.14) N = 14	M = 30 (SD = 12.90) N = 10
3	M = 61.35 (SD = 18.62)	M = 54.15 (SD = 20.12)	M = 45.75 (SD = 19.07)	M = 43.12 (SD = 16.62) N = 16	M = 38.46 (SD = 16.88) N = 13
4	M = 63.75 (SD = 16.13)	M = 59.50 (SD = 19.86)	M = 52.95 (SD = 16.91)	M = 45.79 (SD = 17.50), N = 19	M = 43.52 (SD = 12.59) N = 17
5	M = 84.90 (SD = 17.29)	M = 80.95 (SD = 17.46)	M = 79.60 (SD = 19.60)	M = 72.26 (SD = 20.25), N = 19	M = 70.15 (SD = 21.62), N = 19

Note. SUD change after each exposure within the same audience scene representing a level of idiographic fear hierarchies. All participants underwent exposures 1–3 (Ν = 20); if 20% decreases in SUDs were shown, they moved to the next audience. If not, they underwent exposure 4 and/or 5. N values during exposures 4 and 5 vary.

. Post hoc Wilcoxon signed-rank tests were conducted within each audience level, across exposures (Figure 3).

4. Discussion

This study examined the effects of single-session VRE, compared to a control group, on anxiety-related outcomes, subjective and physiological, among high-PSA university students, contributing to the validation of VRE for PSA and the verification of mechanisms of change. The study addressed a key gap in the literature by investigating whether a single-session, criterion-based exposure protocol using individualized hierarchies could induce significant change—something not yet reliably demonstrated in prior VRE studies. Increased subjective and physiological arousal from baseline to required speech tasks was documented, even when the experimenter was the sole audience member. Higher SUDs during exposure to virtual audiences compared to speech without VRE suggest that virtual reality-assisted exposure exacerbated PSA.

The VREG showed significant reductions in SUDs across all hierarchy levels over the three compulsory exposures, with the most challenging audiences requiring four (and five) additional exposure trials to reach a 20% SUD reduction. These findings are consistent with the habituation hypothesis as a mechanism of change in exposure therapy [55], as supported in previous virtual reality-assisted interventions [24,25,32,56]. Pre- to post-VRE comparisons were consistent with this conclusion: the VREG’s SUDs were reduced from the first to second speech, something not observed for the CG, suggesting that improvement can be attributed to exposure and not just to the passage of time. Similarly, HR decreased from speech 1 to speech 2 for the VREG, returning to baseline, whereas the CG exhibited no significant recovery. Although an active non-VRE control group was not included, evidence suggests that habituation observed in the VREG was unlikely to be due to speech repetition alone, as SUDs were higher for speeches given to virtual audiences compared to speech alone. The findings contradict those of Takac et al. [24], who found no significant PSA improvement in a single VRE session. This is a novel contribution, as previous findings suggested that single-session VRE may be insufficient. The individualized and criterion-based habituation design of the present protocol appears to be a critical innovation that may account for the stronger results. The current study supports that VRE can yield significant habituation, even for highly stressful stimuli, in one session. Key differences in this study compared to prior work included the use of individualized hierarchies and repeated exposures until a criterion was met, suggesting that these parameters may be critical for maximizing impact.

Significant improvement was also observed in VREG participants’ willingness to give a speech. Change in this outcome indicates that new learning may have been acquired, as proposed in newer conceptualizations of exposure therapy (e.g., [30,33]). In the future, more objective behavioral measures of readiness for public speaking, anxiety, and performance on speaking tasks should complement self-rated willingness, to demonstrate more conclusively that VRE has real impact on the behavior of those with PSA. Future research should also test fully remote or self-guided VRE platforms to gauge feasibility in non-lab settings, which could broaden access and reduce costs, particularly for individuals with subclinical PSA.

Despite hypothesizing that changes in RMSSD would indicate the operation of additional mechanisms, such as improved emotion regulation, no significant differences were observed between speech 1 and speech 2 for either group. However, RMSSD, a reliable index of parasympathetic activity, relatively free of respiratory influences [57] that might have confounded results during different phases of the experiment, effectively tracked task-related changes. It decreased from baseline to speech 1 and returned to baseline during speech 2, indicating adequate autonomic flexibility for all participants [58]. The lack of RMSSD improvement specifically for the VREG is inconsistent with an emotion regulation mechanism operating in VRE. This may reflect the insufficiency of a single-session intervention to induce measurable shifts in parasympathetic regulation, particularly in a nonclinical, young adult population with relatively adequate baseline autonomic function. Moreover, improvements in self-report measures like SUDs and willingness to speak may partially stem from expectancy or novelty effects associated with immersive VR, though the consistent stepwise distress during exposure and HR recovery in the VRE group support the argument for an exposure-related effect. We also acknowledge that the use of university students limits the generalizability of results to broader or treatment-seeking populations. Future studies should aim to replicate findings in more diverse samples, over longer treatment durations, and with control conditions to isolate potential expectancy effects. Perhaps a single session was insufficient to yield such an improvement. Alternatively, the inclusion of nonclinical participants with adequate baseline regulation likely constrained HRV improvements, as greater gains are typically observed in populations with more baseline dysfunction. However, the possibility that RMSSD was influenced by the small sample size and variability across participants should also be considered as an alternative possibility.

No significant changes in the PRCS from speech 1 to 2 for the VRE group, or between the two groups, were observed, despite this being considered a tool that can capture cognitive, behavior, and emotional changes due to exposure interventions [48,56]. This finding may pertain to the nature of the tool, which may assess more durable, trait-like cognitive and behavioral tendencies toward public speaking. These may be less likely to change after very brief exposure treatments. Repeated sessions may be needed to produce significant and lasting change in this dimension. Studies finding significant improvement in the PRCS typically involved multiple sessions and included additional CBT components [1,59,60] that target, for example, aspects of PSA like anticipatory anxiety and post-event processing, which were not explicitly addressed here. These aspects may require longer interventions or real-world behavioral experiments to challenge core beliefs [61]. At the same time, we note that we had limited statistical power to observe effects on this specific measure. Although it is unclear that a larger sample size would have changed this non-significant finding, it would have provided increased confidence in this result.

The results must be evaluated taking into consideration some limitations. The sample size, though based on power analysis and prior research, was small and violated parametric assumptions, which necessitated the use of non-parametric tests, potentially reducing sensitivity to group differences [62], (pp. 132) and the generalizability of results. It also did not provide us with enough power to examine any effects of SAD, given that the numbers of males were small but differed between the VRE and control groups. Unfortunately, the presence of five male participants in the VRE group only raises the possibility that gender-related physiological or emotional reactivity differences (e.g., in HR or HRV responses) may have influenced the observed effects. While our sample size did not permit gender-based analysis, this remains an important direction for future research using gender-balanced designs. Despite the smaller-than-planned sample size following initial power analysis, observed effects were of substantial magnitude and we had adequate power to observe them. These effect sizes, combined with prior research supporting the idea that our sample size was adequate to observe medium-to-large effects [47], lend support to the robustness of results, despite the use of conservative, non-parametric methods.

However, the small number of participants prevented us from conducting various sensitivity analyses, e.g., based on gender and age, and from introducing different experimenters as “audiences” to the speech task and assessing potential differences due to their characteristics. Yet, with the present sample, significant effects were observed in expected directions. Our findings necessitate replication with larger, more clinical, culturally diverse, and treatment-seeking samples.

A second limitation is that changes in physiological arousal were inferred from pre- and post-speech task measurements only, as no psychophysiological data were collected during exposure. Although this was carried out to avoid signal noise from heavy VR wearables, it would have been valuable to track arousal in real time alongside SUD reductions. Moreover, respiration rate was not measured or controlled during HRV recordings, which limits the interpretability of RMSSD differences, given its potential confounding by breathing patterns. While RMSSD is considered less sensitive to respiratory variation than other HRV indices, this remains a limitation to conclusions regarding this outcome measure.

Third, differences in task engagement between groups (i.e., passive questionnaire completion by the CG versus active VR exposure in the VRE group) may have introduced attentional and time-on-task effects that could influence between-group differences in HRV or other autonomic measures. It should be noted, however, that active engagement is not an extraneous factor in exposure; it is, rather, a component that may represent an active ingredient of exposure treatments (e.g., [33]) and cannot be separated from other treatment properties and mechanisms. Nevertheless, future studies should consider including an active control condition matched for engagement and cognitive load.

An additional limitation was the absence of other active control groups. It may have been insightful to include a neutral VR control scene, to isolate the effects of exposure versus VR itself. However, ample research already supports the effectiveness of exposure-based interventions [9,12,20] and VRE, and therefore, this was deemed unnecessary and beyond the scope of this study. A potentially more important omission was that of exposure without VRE, to document that virtually assisted exposure has added value over repeated speech practice alone. Some evidence speaking to this is the fact that SUDs were higher during exposure to medium and high hierarchy levels of audiences, compared to the first speech without VRE, suggesting that VRE intensified the exposure experience. In defense of our design, waitlist or minimal intervention controls are common practice in psychosocial interventions (Mohr et al., 2009) [63] as it is challenging to identify active control groups that enable the simultaneous control of all non-specific variables that can influence results.

Also, the use of the PRCS, which may describe more dispositional fear of public speaking and may be less sensitive to short-term changes due to interventions, in addition to being a relatively old measure, can be considered a limitation, and more sensitive indices should be examined in the future.

Despite limitations, this study contributes valuable insights into VRE’s utility for PSA treatment and highlights its potential for wide accessibility. Virtual reality offers a novel means of exposure and skill practice [64], with a rapidly expanding body of research supporting its effectiveness [61]. VRE has benefited individuals who struggle to imagine phobic scenarios or who are not ready for in vivo exposure [65], as a preparatory step in CBT protocols. The present evidence suggests, additionally, that VRE can be effective on its own, even in a single session. As such, it offers a promising solution for those with difficulties performing in public, whether or not their anxiety reaches clinical levels. This study demonstrates that virtually assisted exposure can reduce all aspects of PSA quickly, encouraging the development of brief VRE tools for widespread application. In summary, this study provides novel evidence that a single, criterion-driven VRE session, tailored to individual fear hierarchies, can lead to significant reductions in subjective, behavioral, and physiological indices of public speaking anxiety. These findings support the potential of flexible, efficient, and accessible digital interventions to meet the mental health needs of high-anxiety individuals in nonclinical and clinical populations alike.