A stepped randomized controlled trial was conducted with two groups randomly assigned to one of two conditions: an experimental group that received treatment according to the VRET protocol for fear of public speaking, and a control group placed on a waitlist.

Inclusion criteria were: (a) individuals over 18 years old who reported fear of public speaking, as indicated by a Personal Report of Confidence as a Speaker-12 (PRCS-12) score equal to or greater than 48 points; and (b) individuals available to attend weekly treatment sessions for two months. Exclusion criteria were based on adverse effects associated with the use of VR headsets, as outlined by the Amelia Virtual Care software (n.d.), a platform that provides controlled virtual environments for mental health interventions. Excluded were individuals with: (a) heart disease; (b) severe respiratory illnesses; (c) vertigo syndrome; (d) schizophrenia or other active psychotic disorders; (e) epilepsy; (f) severe visual impairment or blindness; (g) deafness; and (h) pregnancy. Additionally, individuals undergoing any type of psychological treatment were excluded due to potential interference with the study. All criteria were assessed using a screening form provided alongside the PRCS-12 at recruitment.

A total of 26 participants were recruited and randomly assigned to the experimental and control conditions. Data from participants who did not attend or complete the treatment and/or failed to respond to the assessments were excluded. One participant was excluded due to outlier data linked to risk factors identified in the Clinical Outcomes in Routine Evaluation-Outcome Measure (CORE-OM; Feixas et al., 2012) and the participant form. The final sample consisted of 14 participants (7 in each group), with 9 identifying as female (64.29%) and 5 as male (35.71%). Participants ranged in age from 19 to 28 years, with a mean age of 22.6 years (SD = 2.31). A priori power analysis was conducted using G*Power (Faul et al., 2007), assuming a medium effect size of 0.30 (Cohen’s f) based on Anderson et al. (2013), an alpha level of 0.05, and a power of 0.80, indicating a required sample size of 18 participants.

Participants were recruited through an open call by posting a digital flyer on the U-Cursos platform (the official platform for students and faculty at the University of Chile) as well as on social media. Individuals who met the inclusion criteria were selected. Afterward, they signed an informed consent form prior to beginning treatment, which had been approved by the Ethics Committee (CEI) of the Faculty of Social Sciences at the University of Chile (No. 26-34/2023). Both control and treatment group participants were assessed through repeated measures. While the treatment group received the intervention, the control group was monitored using the same assessments administered to the experimental group. Subsequently, control group participants received the same treatment.

The procedures were carried out by ten therapists, all of whom were either licensed psychologists or psychology graduates who were currently enrolled in or had completed the Postgraduate Certificate in CBT at the University of Chile. They received training that included detailed explanations of the therapy protocol and safety procedures, therapeutic materials, and the VR instruments to be used, in addition to biweekly supervision sessions. Before beginning the process, the therapists signed informed consent forms. Each participant was assigned to one therapist, who guided them through the entire treatment process. Additionally, a lead researcher assisted the therapist as a technical aide during the VR exposure sessions, managing the Amelia Virtual Care platform.

The treatment protocol consisted of 7 to 8 sessions distributed across four modules (Psychoeducation, Cognitive Flexibility, Exposure, and Closure), with each session lasting approximately one hour. Sessions were held at the Center for Applied Psychology (CAPs) and the Faculty of Social Sciences at the University of Chile. The protocol used was developed by Ayala (2022) and was adapted according to the improvement suggestions provided by the author, especially regarding its clinical applicability and module structure (see Table 1). Furthermore, adjustments were made to modify the delivery format to enable in-person application with participants.

To assess the efficacy of the protocol, evaluations were conducted at four stages, with the aim of assessing the different modules of the therapy. In Stage 1, participants were assessed before starting treatment; in Stage 2, they were evaluated after completing the psychoeducation and cognitive flexibility modules; in Stage 3, assessment followed the exposure module; and in Stage 4, participants were assessed one week after completing the intervention. The only measure administered differently was the SUDS, which was recorded during each exposure session.

Regarding the exposure sessions, the duration of each exposure ranged between 15 and 25 minutes, becoming longer toward the final sessions. Additionally, repeated intra-session SUDS measurements were implemented every two minutes during these sessions to monitor anxiety fluctuations throughout the session and allow the therapist to take precautions or intervene in case of emotional destabilization.

To analyze the collected data, version 2.3 of the statistical software Jamovi (Sydney, New South Wales, Australia) (The jamovi project, 2022) was used. To evaluate the efficacy of the protocol, a 4 $\times$ 2 mixed repeated measures ANOVA was conducted. The within-subject factor was the four measurement stages, and the between-subject factor corresponded to the experimental group and the waitlist control group. Mean scores from the PRCS-12, SSPS, LSAS, and CORE-OM, collected before, during, and after treatment, were analyzed. Both within-subject and between-subject comparisons were made for the PRCS-12 and SSPS scales. Accordingly, a reduction in symptomatology was indicated by increased public speaking confidence (lower PRCS-12 scores), more positive self-statements (higher scores on the SSPS positive subscale), and fewer negative self-statements (lower scores on the SSPS negative subscale), along with decreased scores on the other distress measures.

In the case of the CORE-OM, as described by Feixas et al. (2012), the total score is calculated as the average of the scores from all dimensions, excluding the risk dimension. Additionally, a clinical change analysis specific to the instrument is included. A clinical cutoff score of 10 is used to identify clinically significant levels of distress. To detect a reliable clinical change, the authors state that a minimum difference of 5 points between questionnaire administrations must be observed. This significance is reflected in a decrease in scores, meaning that lower scores indicate greater clinical improvement following treatment.

To assess SUDS, a one-way repeated measures ANOVA was conducted. The within-subject factor was the mean SUDS score of the experimental group for each of the four exposure sessions. For participants who requested a fifth exposure session, the data from that session were used in place of the fourth session in the exposure module. This instrument is scored on a scale from 1 (no anxiety) to 10 (intolerable anxiety), with lower scores indicating reduced anxiety.

For main effects and interactions, partial eta squared ($\eta$²p) was used to report effect size, and the significance level was set at 0.05. In cases where the assumption of sphericity was not met (p 0.05), corrections were applied: Greenhouse-Geisser (when $\epsilon$ $\le$ 0.75) or Huynh-Feldt (when $\epsilon$ $>$ 0.75). For post hoc analyses, if the assumption of sphericity was satisfied, Tukey’s correction was applied; if it was violated, Bonferroni correction was used. In addition, the mean difference (MD) was reported for each significant comparison.

The 4 $\times$ 2 ANOVA (Stage $\times$ Condition) for the PRCS-12 was conducted using the Greenhouse-Geisser correction ($\epsilon$ = 0.52). Results revealed a significant effect of Stage (F[3, 36] = 11.99, p 0.001, $\eta$²p = 0.50), Condition (F[1, 12] = 24.01, p 0.001, $\eta$²p = 0.67), and the Stage $\times$ Condition interaction (F[3, 36] = 11.57, p = 0.001, $\eta$²p = 0.49). These results indicate an increase in public speaking confidence over the stages in the experimental group, but not in the control group, as reflected in a decrease in PRCS-12 scores (see Fig. 1). Post hoc analysis using the Bonferroni test showed a significant decrease in the experimental group between Stage 1 and Stage 3 (p = 0.02, MD = 18), between Stage 1 and Stage 4 (p = 0.001, MD = 20.71), between Stage 2 and Stage 3 (p 0.05, MD = 11.86), and between Stage 2 and Stage 4 (p = 0.002, MD = 14.57). A significant difference was also found between Stage 4 of the experimental group and Stage 4 of the control group (p 0.001, MD = –19.14). No other between-group comparisons across stages showed significant differences (p $>$ 0.05).

Fig. 1.

Distribution of Personal Report of Confidence as a Speaker-12 (PRCS-12) scores. Note: The mean of each score obtained is plotted across treatment stages, with standard error bars shown for each group at each stage. The distribution of PRCS-12 scores shows an increase in confidence when speaking in public throughout the stages in the experimental group (●), compared to the control group ($\mathrm{\square }$). Source: Own elaboration.

Regarding the positive self-statements subscale of the SSPS, the 4 $\times$ 2 ANOVA (Stage $\times$ Condition) with Greenhouse-Geisser correction ($\epsilon$= 0.52) revealed a significant main effect of Stage (F[3, 36] = 9.68, p = 0.002, $\eta$²p = 0.45), but no significant effect of Condition (F[1, 12] = 2.63, p = 0.13, $\eta$²p = 0.18), nor a significant Stage $\times$ Condition interaction (F[3, 36] = 3.38, p = 0.07, $\eta$²p = 0.22). These results indicate an increase in positive self-statements across the stages regardless of group assignment (see Fig. 2A).

Fig. 2.

Distribution of Self-Statements during Public Speaking (SSPS) scores according to the positive self-verbalizations subscale (A) and the negative self-verbalizations subscale (B). Note: The mean of each score obtained is plotted across treatment stages, with standard error bars shown for each condition at each stage. (A) shows the distribution of scores for positive self-verbalizations, where a slight increase can be observed across the stages in the experimental group (●), while a smaller increase is observed in the control group ($\mathrm{\square }$). (B) shows the distribution of scores for negative self-verbalizations. In Stage 1, the experimental group (●) had a higher mean score than the control group ($\mathrm{\square }$). Subsequently, participants in the experimental group showed a decrease in their scores, eventually becoming lower than those of the control group, which tended to remain stable across the stages. Source: Own elaboration.

For the negative self-statements subscale, the 4 $\times$ 2 ANOVA (Stage $\times$ Condition) showed a significant main effect of Stage (F[3, 36] = 7.37, p 0.001, $\eta$²p = 0.38) and a significant Stage $\times$ Condition interaction (F[3, 36] = 8.35, p 0.001, $\eta$²p = 0.41); however, there was no significant effect for Condition (F[1, 12] = 0.62, p = 0.45, $\eta$²p = 0.05). These results suggest a decrease in negative self-statements over time in the experimental group, while the control group showed no differences across stages (see Fig. 2B). Post hoc analysis using the Bonferroni test revealed a significant reduction in the experimental group between Stage 1 and Stage 3 (p = 0.001, MD = 6.71), and between Stage 1 and Stage 4 (p = 0.005, MD = 7.86). No other comparisons were statistically significant (p $>$ 0.05).

For the total CORE-OM score, the 4 $\times$ 2 ANOVA (Stage $\times$ Condition) with Greenhouse-Geisser correction ($\epsilon$= 0.51) revealed a significant main effect of Stage (F[3, 36] = 5.17, p = 0.02, $\eta$²p = 0.30) and a significant Stage $\times$ Condition interaction (F[3, 36] = 4.20, p 0.05, $\eta$²p = 0.26), but no significant effect of Condition (F[1, 12] = 2.59, p = 0.13, $\eta$²p = 0.18). These results indicate a difference between the experimental and control groups, reflected in lower scores over time in the experimental group (see Fig. 3). Post hoc analysis using the Bonferroni test showed a significant decrease in the experimental group between Stage 2 and Stage 3 (p = 0.001, MD = 0.63), and between Stage 2 and Stage 4 (p = 0.005, MD = 0.65). No other comparisons were statistically significant (p $>$ 0.05).

Fig. 3.

Distribution of Clinical Outcomes in Routine Evaluation-Outcome Measure (CORE-OM) total scores.Note. The mean of each score obtained is plotted across treatment stages, with standard error bars shown for each condition at each stage. The distribution of total CORE-OM scores shows that, in Stage 1, the experimental group (●) had a higher mean score than the control group ($\mathrm{\square }$). Subsequently, the mean score of the experimental group decreased, becoming similar to that of the control group, which tended to remain stable throughout the stages. Source: Own elaboration.

When analyzing clinical change, the experimental group initially showed a moderate to severe level of distress (20.40 pts.). During Stage 2, this distress slightly increased (20.61 pts.), but then considerably decreased in Stage 3 (14.28 pts.) and Stage 4 (14.13 pts.), reaching a moderate distress level. This reflects a reliable clinical change with a decrease of 6.12 points between Stage 1 and Stage 3, 6.27 points between Stage 1 and Stage 4, 6.33 points between Stage 2 and Stage 3, and 6.48 points between Stage 2 and Stage 4. On the other hand, the control group started with a moderate clinical level of distress (15.05 pts.), which did not significantly decrease throughout Stages 2 (14.49 pts.), 3 (14.59 pts.), and 4 (14.29 pts.), indicating no clinically reliable change.

No statistically significant results were found in the subscales. However, reliable clinical changes were observed in all dimensions except for the risk dimension. In the well-being subscale, the experimental group initially scored 19.64 and showed a clinically significant reduction by Stage 4, scoring 12.85 (reliable change index = 6.79). In contrast, the control group started at 17.50 and maintained similar scores across all measurements, indicating no clinically reliable change.

In the symptoms/problems subscale, the experimental group initially reported clinical distress at 22.14, which decreased to 16.42 by Stage 4 (reliable clinical change index = 5.73). The control group, which started at a clinical level of 14.05 pts., did not reach a reliable clinical change in any later stage.

In the general functioning subscale, the experimental group started with a score of 18.92, which significantly decreased to 12.26 by Stage 4 (reliable clinical change index = 6.66). In contrast, the control group began with a score of 15.24, which slightly decreased to 13.81 by Stage 4, not reaching a clinically reliable change.

Performance subscale: Regarding the anxiety score on this subscale, the 4 $\times$ 2 ANOVA (Stage $\times$ Condition), using Greenhouse-Geisser correction ($\epsilon$ = 0.54), revealed a significant main effect of Stage (F[3, 36] = 14.75, p 0.001, $\eta$²p = 0.55), but no significant effect of Condition (F[1, 12] = 0.07, p = 0.80, $\eta$²p = 0.01), nor of the Stage $\times$ Condition interaction (F[3, 36] = 3.38, p = 0.06, $\eta$²p = 0.22). On the other hand, avoidance score in the Performance Subscale: the 4 $\times$ 2 ANOVA (Stage $\times$ Condition), using Greenhouse-Geisser correction ($\epsilon$ = 0.47), revealed a significant main effect of Stage (F[3, 36] = 8.63, p = 0.005, $\eta$²p = 0.42), but no significant effect of Condition (F[1, 12] = 0.16, p = 0.69, $\eta$²p = 0.01), nor of the Stage $\times$ Condition interaction (F[3, 36] = 1.22, p $>$ 0.31, $\eta$²p = 0.09).

Social Interaction Subscale: Regarding the anxiety score on this subscale, the 4 $\times$ 2 ANOVA (Stage $\times$ Condition), using Greenhouse-Geisser correction ($\epsilon$ = 0.48), revealed a significant main effect of Stage (F[3, 36] = 4.56, p = 0.03, $\eta$²p = 0.28), but no significant effect of Condition (F[1, 12] = 0.28, p = 0.61, $\eta$²p = 0.02), nor of the Stage $\times$ Condition interaction (F[3, 36] = 3.17, p = 0.08, $\eta$²p = 0.21). On the other hand, avoidance score in the Social Interaction Subscale: the 4 $\times$ 2 ANOVA (Stage $\times$ Condition), using Greenhouse-Geisser correction ($\epsilon$ = 0.59), revealed a significant main effect of Stage (F[3, 36] = 5.30, p = 0.02, $\eta$²p = 0.31), but no significant effect of Condition (F[1, 12] = 0.58, p = 0.46, $\eta$²p = 0.05), nor of the Stage $\times$ Condition interaction (F[3, 36] = 2.84, p = 0.08, $\eta$²p = 0.19).

Among participants in the experimental group, 3 requested a fifth exposure session (42.86%). The one-way repeated measures ANOVA revealed no significant differences across stages (F[3, 18] = 0.09, p = 0.96, $\eta$²p = 0.02). Therefore, there was no observed change in SUDS values across the sessions in the exposure module.

One of the main limitations of this pilot study was the small sample size (14 participants, 7 in each group). This reduced number limits the generalizability of the results and decreases the statistical power of the analyses. While the design allowed for the exploration of preliminary differences between groups, future studies should consider larger samples to validate and extend the findings.

Another important limitation to consider is that the data may be influenced by the sample’s homogeneity, as it predominantly consisted of individuals with higher education levels. This can be explained by the fact that fear of public speaking is one of the most common problems among university students, as the ability to communicate effectively before an audience is a key factor in many professions (Piñeiro, 2015). Although the number of individuals who meet clinical criteria is low, this fear is still impactful enough that 76% of people avoid at least one public speaking situation each year, and 45% experience significant or extreme difficulty when facing such situations, with university students being the most affected (Piñeiro, 2015). Therefore, to improve the external validity and overall applicability of future research, it is recommended to include a more diverse sample in terms of educational background. This approach will allow for a more comprehensive and nuanced understanding of the phenomenon and the ability to more robustly extrapolate results to the general population.

Moreover, the treatment was administered by different therapists, which could have influenced the outcomes. Individual differences in therapeutic style, clinical experience, or communication skills may affect the effectiveness of the treatment, even when using a standardized protocol. In this regard, the therapist can represent a significant source of variability in psychotherapy outcomes (Johns et al., 2019; Lutz and Barkham, 2015; Mahon et al., 2024), highlighting the importance of considering this factor in future studies.

Lastly, this study only compared the effectiveness of the treatment against a no-treatment condition. Therefore, future research should ideally compare the efficacy of this protocol with other types of treatments, as well as with in vivo exposure therapy. Several studies have indicated that VR demonstrates comparable efficacy to in vivo exposure therapy (Anderson et al., 2013; Chesham et al., 2018; Reeves et al., 2021). Additionally, it would be important to include follow-up assessments to evaluate long-term effects and determine whether the efficacy of the protocol is sustained over time (Hill et al., 2016).

Ultimately, although these findings should be interpreted with caution due to the exploratory nature of the study design and the small sample size, the results reveal promising implications for the field of mental health, as effects suggesting the efficacy of the protocol in treating public speaking anxiety were observed. In this way, the study contributes to the development and refinement of treatments in clinical settings.

One potential advantage of VRET for SAD is its ability to reduce treatment dropout rates, since exposure to social situations takes place in a virtual environment rather than in real-life situations, as occurs with in vivo exposure (Emmelkamp et al., 2020).

Another implication of this pilot study is its contribution to the development of treatments that incorporate new technologies. The well-documented technological gap in Latin America, evidenced by access barriers and lack of digital skills (Economic Commission for Latin America and the Caribbean, 2021) makes this especially relevant. This is particularly important considering the WHO (2023) has stated that diverse technological innovations should be adopted to positively impact public health and improve people’s quality of life. Given this context, it is essential to promote access to treatment and raise awareness of these issues in the region to support and encourage help-seeking among those in need.

In summary, while this is a pilot study and results should be interpreted with caution, considering the absence of effective VR-based protocols in Chile, the findings of this research represent a relevant starting point for the development, evaluation, and implementation of VR interventions aimed at various anxiety disorders in the region.