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

Abstract

While virtual reality exposure (VRE) has shown effectiveness in treating social anxiety in adults, research on its efficacy for adolescents remains limited. Given that adolescence is a critical period for early intervention, this study aims to address this gap by evaluating the efficacy and acceptability of VRE compared to in vivo exposure (IVE) in a non-referred sample of socially anxious adolescents. Additionally, we seek to identify mechanisms of change—such as expectancy violation, habituation, and self-efficacy—as well as predictors of treatment response, including clinical, personality, and VR-related factors. Using a randomized controlled trial (RCT), 120 adolescents (ages 12–16) with subclinical to moderate social anxiety will be assigned to one of three conditions: VRE, IVE, or a waitlist control (WL). Participants in the active conditions will undergo a seven-session exposure-based intervention (either in VR or in vivo). Primary (SPAI-18, LSAS-avoidance) and secondary (SPWSS) measures of social anxiety, along with general well-being indicators (e.g., resilience, depression, psychosocial functioning), will be assessed at baseline, post-treatment, and 3- and 6-month follow-ups. A series of linear mixed model (LMM) analyses will be used to examine and compare the effects of the interventions. We hypothesize that both VRE and IVE will significantly reduce social anxiety symptoms compared to WL at post-assessment, with comparable long-term efficacy between the two exposure methods. Additionally, thematic analyses will be conducted to explore participants’ experiences and acceptance of VRE and IVE through qualitative interviews. The findings of this study aim to advance digital mental health research by evaluating the potential of VRE as an early intervention and identifying mechanisms and predictors to inform personalized treatments for socially anxious youth.

Trial registration

Clinicaltrials.gov: NCT06379633, registered on April, 23, 2024.

Introduction

Social anxiety in adolescents

Anxiety disorders are among the most prevalent sources of psychological suffering in youth [72]. Particularly for the emergence of social anxiety disorder (SAD), mid to late adolescence is a developmentally sensitive period [32]. Social anxiety (SA) is characterized by an intense fear of being scrutinized and negatively evaluated in social and performance situations [3]. Adolescents with SA commonly fear and avoid situations involving (in)formal public speaking, speaking up in a class, and meeting new people. Although these situations may trigger varying levels of anxiety in different individuals, it is believed that most people with SAD share similar underlying core fears, such as fear of rejection, appearing foolish or unintelligent, expressing disagreement or disapproval, and being the center of attention [40].

These fears typically lead to substantial impairments in day-to-day functioning and affect several life domains that are important in this developmental stage, including social relations (e.g., fewer friends, higher levels of loneliness), academic performance (e.g., concentration problems, test underperformance, refusal to go to school; [49, 67, 108]), and general wellbeing (e.g., depressed mood, suicidality [11, 20]). By the age of 23, about 90 percent of SAD cases have been developed [57]. The plasticity of adolescence provides a window of opportunity for early interventions to address maladaptive avoidance behaviors, preventing the onset of social anxiety disorder (SAD) and its progression to severe cases [48].

Exposure therapy

Evidence-based treatments, such as cognitive-behavioral therapy (CBT) have, generally, proven successful for treating anxiety disorders. A meta-analysis conducted by Scaini and colleagues [90], focusing on children and adolescents with SAD, showed that CBT was effective in reducing SAD symptoms, with moderate to large effect sizes observed for pre-post effects (g = 0.99), between-group effects (g = 0.71), and follow-up responses (g = 1.18 for follow-up vs. pre-test and g = 0.25 for follow-up vs. post-test). CBT typically involves elements such as cognitive restructuring and exposure to fear-evoking stimuli (e.g., social situations [56]). Exposure-based interventions are considered one of the crucial and active ingredients of effective CBT for SAD [1, 31, 53, 111].

During exposure, in the context of SAD, individuals are encouraged to confront their fears with the aim to reduce maladaptive avoidance behaviors and instead build confidence in their social interactions. The rationale is that when individuals limit or avoid engagement with anxiety-provoking stimuli, they miss out on the opportunity for corrective learning regarding the actual threat levels of specific situations and their capacity to tolerate anxiety [58]. Rather than permitting corrective learning to occur, avoidance behavior is believed to be perpetuated by the negative reinforcement resulting from the temporary reduction of anxiety and a sense of relief [65]. Specifically, for adolescents with SAD, therapy often comes in the form of in vivo exposure which includes engaging in real-life social interactions during and between therapy sessions to give them opportunities to practice confronting their feared social situations (without the use of safety and avoidance behaviors).

Notably, a large proportion of adolescents with SAD do not receive exposure-based interventions [5]. Exposure-based interventions, while effective for many anxiety disorders, often produce lower response rates in SAD than in other anxiety disorders, and its intensity can make it challenging especially for individuals with SAD to fully engage [78]. Socially anxious individuals may refuse or drop out of treatment when confronted with exercises that closely mimic their most distressing social scenarios. Direct, real-life confrontations with social fears can feel overwhelming and uncontrollable, especially for adolescents unaccustomed to managing intense social discomfort.

VR Exposure for SAD

A potentially promising way to overcome some of the barriers to exposure-based treatment is by connecting to the adolescents’ technologically driven environment and delivering exposure interventions in virtual reality (VR). In VR exposure, individuals are exposed to virtual representations of anxiety-inducing situations rather than the actual, real-life situations themselves. VR combines computer graphics, body-tracking devices, visual displays, and sensory input tools to fully immerse individuals in a digitally created virtual world. Particularly for adolescents, the game-like features of VR and its playful elements could increase treatment adherence and motivation [73]. There are already some studies to suggest that individuals with specific phobias may be more willing to engage in virtual reality than in vivo exposure [46, 91]. However, this notion has never been empirically tested in adolescents with SA.

Various studies and meta-analyses on VR exposure show that it can effectively treat SAD in adults, often proving as effective, and sometimes even superior, to traditional exposure [16, 21, 23, 37, 51, 100], but see also Wechsler et al. [110] who found in vivo exposure to be superior. However, research on VR’s effectiveness for adolescents with SAD remains limited [62]. Only a few studies have addressed this gap. Parrish et al. [79] investigated VR exposure’s feasibility for adolescents aged 13–18 with social anxiety, finding that adolescents with SAD experienced greater distress in social VR scenarios compared to non-SAD peers and felt more anxious in social environments than in neutral settings. In another study, Kahlon et al. [54] investigated the impact of a single VR session on adolescents (ages 13–16) with a fear of public speaking. The results showed a large reduction in public speaking anxiety, which was sustained over 1- and 3-month follow-ups. However, the study was limited by a lack of control conditions and a focus on public speaking, which restricts its generalizability.

In a later study, Beele et al. [10] found reduced social anxiety in school-anxious adolescents after five VR exposure sessions, though it was a pilot study without a control condition. Finally, in an adequately powered RCT, Kahlon et al. [55] showed that self-guided gamified VR exposure was efficacious in reducing fear of public speaking compared to a waitlist control condition. Taken together, these studies suggest that VR exposure may be a promising tool for the treatment of adolescents with social anxiety. However, there is a clear need for more rigorous research with adequate power (larger sample sizes), active control groups, and not limited to public speaking anxiety or school anxiety.

Underlying mechanisms and predictors of VR exposure outcomes

In addition to evaluating VR exposure’s effectiveness for adolescents with SAD, understanding the mechanisms behind its outcomes is essential. While traditional in vivo exposure has several established mechanisms, those specific to VR exposure remain less explored [92]. This study will focus on three potential mechanisms in line with three leading theoretical models on exposure: (1) emotional processing theory, (2) inhibitory learning theory, and (3) self-efficacy theory.

First, emotional processing theory has been foundational in exposure therapy [41], emphasizing fear activation and habituation within (i.e., within-session habituation) and across (i.e., between-session habituation) sessions. Effective treatment, according to this theory, requires modification of the pathological features and incorporation of new elements in a fear structure in memory [42, 86]. Second, inhibitory learning theory has more recently gained attention [29, 30]. It is rooted in associative learning models, focuses on creating new inhibitory associations to override existing fear associations, which is presumably driven by expectancy violation. Treatment, according to this model, is most effective when actual experiences during exposure defy an individual’s fear-based expectations, leading to stronger inhibitory learning. Finally, self-efficacy theory [8] has proposed another key mechanism in exposure therapy. Outcomes are enhanced by increasing a person’s belief in their capacity to face feared situations and effectively cope with them. Building self-efficacy through successful exposure allows individuals to feel more capable, directly supporting better treatment results.

While various studies suggest that these mechanisms are essential in in vivo exposure [28, 60, 81, 89], research on their demonstrated impact and relevance in VR exposure remains notably limited. At first glance, the currently dominant inhibitory learning model of exposure therapy seems less appropriate in explaining the effects observed in VR exposure, as the absence of real threats in a virtual environment can arguably limit the possibility of true expectancy violations (see [92]).

Next to answering how VR exposure works, another crucial question concerns for whom it works. Gaining knowledge about predictors for its outcome can ensure that VR exposure interventions are offered to those most likely to respond. Given that research on identifying predictors in VR treatment response is still in its infancy [80], putative predictors that are based on prior research on cognitive behavioral therapy, in vivo and VR exposure, and theoretical ideas to differentiate between those who would (not) benefit from an intervention using VR are worth investigating [14, 38, 68, 74, 76, 77, 82].

More specifically, four categories of predictors (i.e., non-specific predictors and moderators) could be of interest, namely: (1) clinical variables (e.g., pre-intervention severity of social anxiety and comorbidity); (2) personality and other individual trait variables (e.g., behavioral inhibition/activation levels, openness to experience, the value of social connectedness, attachment); (3) VR related variables (e.g., individual immersion propensity, attitude towards VR); and (4) treatment-related variables (e.g., treatment expectations, motivation, preference of treatment modality, working alliance). Gaining knowledge about these mechanisms and predictors can ensure that VR exposure interventions are optimized and offered to those who are most likely to respond.

Present study

In this study, the primary objective is to evaluate the efficacy of VR exposure in adolescents with elevated social anxiety using a three-arm randomized controlled trial: a virtual reality exposure (VRE) condition, an in vivo exposure (IVE) condition, and a waiting list (WL) control condition. We chose IVE as the active comparator given that it is the golden standard for SAD treatment [110]. Given that prior studies have found comparable effects between IVE and VRE, it is predicted that both VRE and IVE will be more successful in decreasing SA symptoms (primary outcome) than the WL condition at post-assessment and that VRE will be as effective as IVE [16, 21, 23, 37]. It is expected that there will be no differences between the two active conditions in the long term (3- month and 6-month follow up).

Similar results are expected in secondary outcome measures intended to capture the participants’ more general psychosocial functioning, anxiety, and ability to cope with social challenges. Moreover, we have the following additional objectives: (1) to elucidate the potential working mechanisms of VRE and IVE, with a focus on expectancy violation, habituation, and self-efficacy; (2) to identify predictors of adolescents’ response to VRE and IVE, including clinical and personality variables, VR-related factors, and more general training-related factors; and (3) to assess adolescents’ acceptance of VRE and IVE while gaining insight into their experience with the training.

Method

Participants and recruitment

The intended total sample is 120 adolescents (n = 40 per condition; 12–16 years old) with subclinical or mild to moderate SAD. Participants will be recruited via secondary schools in Belgium and social media platforms such as Facebook and Instagram. Other places (e.g., youth organizations and movements, extracurricular activities, GP offices, libraries) where youth and / or their legal guardians can be found will also be targeted. The estimated completion date for participant recruitment is the end of December 2025. For the (interested) adolescents to participate in the study, one of their legal guardians needs to provide their informed consent via a secure, online link. Subsequently, potential participants need to give their informed consent and complete the brief screening questionnaire, the Social Phobia Inventory (SPIN; [27]). Based on the SPIN, adolescents with sufficient SA (i.e., cut-off score, ≥ 19; [66, 84]) will be invited to the second screening phase.

During the second screening phase, a trained psychology graduate student will conduct a semi-structured interview based on the Dutch versions of the Structured Clinical Interview for DSM-V Childhood Disorders (SCID-5 Junior; [109]), the clinical severity rating (CSR) of the Anxiety Disorders Interview Schedule for Children – Child (ADIS-C), and the MINI-International Neuropsychiatric Interview for Children and Adolescents (MINI-KID, [94]). This will be done to further assess SA the study’s inclusion criteria. Participants who meet our inclusion criteria (see Table 1 for an overview of the inclusion criteria and their assessment) will be invited to participate in the study. In addition to the free exposure training provided, they will receive monetary compensation for the time spent completing assessments.

Table 1 Overview of inclusion criteria per screening phase

Study design and procedure

The current study will consist of a randomized controlled trial (RCT) with three arms: VRE, IVE, and a WL condition. It will follow a mixed-subjects design with condition as a between factor and measurement time-points as a within factor. An independent assessor, blind to the participants’ condition, will do baseline, post- and follow-up assessments. Due to the nature of the study, the participants and the trainers will be aware of the condition (VRE or IVE). The study design is depicted in Fig. 1 while the participant flowchart is depicted in Fig. 2.

Fig. 1

Overview of study design

Fig. 2

Participant flowchart

Outcome measures will be assessed at baseline, post-assessment, a 3-month follow-up, and a 6-month follow-up. Semi-structured, qualitative interviews will be conducted between the post-assessment and the 3-month follow-up. Participants in the two active conditions will undergo the training sessions right after the baseline assessment. For the WL condition, all outcome measures will be administered at the same instances as in the active conditions. After the waiting period, participants in the WL condition will have the opportunity to be randomized into IVE and VRE. The present research protocol has been approved by the Medical Ethics Committee Research of UZ/KU Leuven (S67010 / B3222022001079) and has been registered on Clinicaltrials.gov (on 2024–04–23 with identifier: NCT06379633). Any potential amendments to the study protocol will be submitted to the abovementioned ethics committee.

Randomization

An external research assistant (unaffiliated with the project) will generate a blocked randomization list from the Sealed Envelope website, ensuring participants are randomly assigned to one of three conditions. Allocation will follow a 1:1:1 ratio within blocks of 12 (four per group) to maintain balance and minimize bias. The assistant will prepare sealed envelopes for each participant, which a central researcher will open after baseline assessment to manage logistics, including informing trainers and participants.

Interventions

The intervention will take place either at participants’ schools or at a location within the university, depending on participants’ preference and availability. Due to the study’s goal to unravel the effects of VRE, the focus of the intervention will be on exposure in both active conditions (without explicitly including other treatment components such as cognitive restructuring). Overall, participants will undergo 7 training sessions, each lasting approximately 60–90 min.

During session 1, the trainer will introduce the treatment rationale and provide psychoeducation. Moreover, a list will be made of participants’ most feared situations (e.g., ordering food in a restaurant, asking a question in class, engaging in a group conversation), based on their individual case conceptualization and functional analysis. This list will be used throughout the exposure sessions to determine the specific nature and content of the exercises. Participants will finish session 1 with a first brief exposure exercise of 10 min. Sessions 2 to 6 will be fully dedicated to exposure. Each session will comprise of approximately 2 sets of exposure exercises, with each set lasting about 20 min. Prior to each exposure set, there will be a pre-discussion regarding the participants threat expectancies (and likelihood) related to the exercises. Similarly, after each set there will be a post-discussion reflecting on how the exercises went (e.g., threat occurrence, surprise). Finally, in session 7, participants will complete a 20-min exposure exercise, reflect on their progress, learning and achievements during the training, and discuss relapse prevention strategies.

To standardize across participants, we will reproduce situations during exposure that adolescents with social anxiety typically find threatening, such as those involving performance, intimacy, scrutiny, and assertiveness [59]. Notably, to keep the two conditions comparable, exposure duration will be held constant in both conditions and homework assignments will not be actively encouraged in either condition. Training sessions will be provided by trainers holding a master’s degree in psychology. The trainers are trained to adhere to the protocol and will be supervised by experienced CBT therapists (and supervisors). Sessions will be audio-recorded when possible and regularly discussed during supervision in order to monitor intervention adherence and competence.

Virtual reality exposure (VRE)

VR sessions

Participants in the VRE condition will undergo exposure sessions designed to recreate scenarios that typically trigger social anxiety in their daily life. These sessions, tailored to each participant’s unique needs, will involve VR-based one-on-one and group interactions simulating common social anxiety triggers such as speaking up in class, saying no to a friend, ordering food, or presenting in front of a group. Each session will focus on a specific theme, such as being the center of attention or engaging with strangers. This individualized approach ensures that the exposure exercises align closely with each participant’s case conceptualization and targeted avoidance behaviors.

VR-Software

To conduct the VRE, we will use Social Worlds (version 4.1), a versatile VR software developed by CleVR (see Fig. 3). This program provides a wide range of virtual environments relevant to social anxiety, including classrooms, playgrounds, sports courts, shopping areas, and social gatherings. Social Worlds features dynamic tools such as role-playing, walking around, dialogue variations, and adjustable group sizes, enabling highly personalized and realistic interactions. Trainers can further customize sessions in real time by taking on the role of avatars and using voice modulators to represent different characters.

Fig. 3

Screenshots from the software of the different environments and functionalities (e.g., avatars, reactions)

VR-Hardware

The VR system consists of a gaming laptop (an Alienware laptop with high-performance speeds, substantial memory capacity, triple-A graphics, and fast processing power), a Meta Quest 3 VR headset along with the corresponding VR controllers, noise-cancelling headphones for the user, a Touch laptop, and a PELTOR headset with a microphone for the therapist/trainer (see Fig. 4).

Fig. 4

Display of VR-hardware

In vivo exposure (IVE)

Participants in the IVE condition will be exposed in vivo to various one-on-one and group interactions (e.g., asking direction from strangers at the street, returning clothes at a store, ordering and / or eating in front of others) which tend to elicit anxiety. Similarly to VRE, the exposure exercises will be tailored to each participant in consultation with them, with each session focusing on a specific theme. Depending on the given exposure exercise, exposure will either take place at the training location or in its neighborhood (e.g., nearby supermarkets, subway stations, cafes, buses).

Measures

Screening measures

SPIN

The Dutch Social Phobia Inventory (SPIN; [27]) will be administered to potentially eligible adolescents. Based on prior studies [66, 84], we will invite adolescents who score ≥ 19 on the SPIN to the second screening phase to further assess their eligibility. The SPIN has a demonstrated ability to differentiate adolescents with (sub-)clinical social anxiety from adolescents without social anxiety complains [84].

SCID-5 Junior

The Dutch version of the Structured Clinical Interview for DSM-V Childhood Disorders (SCID-5 Junior; [109]) will be used to assess social anxiety diagnosis and the inclusion criteria. It is widely used to for the classification of mental disorders in children and adolescents and demonstrates moderate to good interrater reliability and internal consistency [87]. It will be administered by trained clinical psychology master students. When the response is YES to the screening questions of the modules in question (see Table 1, the complete module will be used. Any SCID-5 Junior diagnosis will be complemented by the Clinical Severity Rating (CSR from the Anxiety Disorders Interview Schedule for Children – Child (ADIS-C; [96]) to quantify the severity and interference of the diagnosis, following the procedure outlined by Telman et al. [101]. The CSR obtained via the ADIS-C encompasses factors such as number of confirmed symptoms, level of interference caused by the disorder, and overall judgment of the clinician.

In the following section, we will outline our outcome measures for each research objective (for an overview of the measures see Table 2).

Table 2 Overview of data collection

Objective 1

To compare the effectiveness of VRE and IVE in reducing social anxiety symptoms in a non-referred sample of adolescents with elevated social anxiety, and to assess their impact on general wellbeing.

Primary outcomes

Social anxiety (verbal)

The Social Phobia and Anxiety Inventory-18 (SPAI-18; [33, 102]) and the avoidance subscale of the Liebowitz Social Anxiety Scale for Children & Adolescents (LSAS-CA; [69]) will serve as our primary (verbal) outcome measures. The SPAI-18 is an 18-item questionnaire measured on a 7-point Likert scale (1 = never, 7 = always). It serves as an encompassing measure of SAD symptoms. In addition, the LSAS-CA avoidance subscale is included to specifically measure changes in avoidance behavior triggered by social situations. This questionnaire consists of 24 items rated on a 4-point Likert scale (0 = never, 3 = almost always). The SPAI-18 and the LSAS-CA avoidance subscale demonstrate strong psychometric properties, with the SPAI-18 exhibiting high internal consistency (α = 0.94) and good construct validity ([33, 102]), while the LSAS-CA avoidance subscale shows strong reliability (α = 0.87) and validity in assessing social avoidance [69]. Both instruments are sensitive to treatment effects [6, 7].

Social anxiety (behavioral) – BAT peak anxiety

As a behavioral measure of social anxiety, a behavioral assessment task (BAT) will be done where peak anxiety levels (i.e., subjective units of distress [SUDS]; [113]) will be measured [12, 34, 45, 85]. More specifically, peak SUDS will be measured on a scale from 0 (not anxious at all) to 100 (extremely anxious) immediately after the task. The task will consist of a brief social exercise (e.g., 3–5 min) in which the adolescent is asked to interact online (during an MS Team video call) with a (role-playing) confederate. The participant will be given the following instructions: “Pretend that the person you will be having the online conversation with is a student who is new to your school, and you have been assigned to be his/her ‘buddy’. Get to know him/her.” Participants will also be asked to initiate and keep the conversation going for at least 3 min, but if they can, they may go up to 5 min. To standardize this task, confederates are provided with the same brief backstory (i.e., 16 years old, hobbies: sports & music, due to relocation start going to this new school before two weeks). They are instructed to follow the participant’s lead and to not initiate new topics (except for two questions that they may ask from a standardized list of questions, only if necessary). Moreover, the confederate will be using two emojis during the conversation to signal when 3 min (thumbs up emoji) and 5 min (clapping emoji) have passed.

Secondary outcomes

Measures related to self-reported social anxiety

Liebowitz Social Anxiety Scale – Children and Adolescents (LSAS-CA), fear subscale

The fear subscale of the LSAS-CA is a 24-item questionnaire assessing the level of fear experienced during specific social situations [69] and is rated on a 4-point Likert scale (0 = never, 3 = almost always). The LSAS and its subscales are widely used and have demonstrated excellent internal consistency, as well as convergent validity [50].

Social Phobia Weekly Summary Scale (SPWSS)

The SPWSS is a brief 6-item questionnaire assessing the severity and frequency of symptoms related to social anxiety experienced by individuals over a weekly period [25]. The scale has demonstrated good internal consistency, reliability, and validity [15, 25, 26].

Measures related to the Behavioral Assessment Task (BAT)

Anticipatory levels of fear

Subjective units of distress (SUDs), on a scale from 0 (not anxious at all) to 100 (extremely anxious), will be rated prior to the BAT.

Total Task duration

The total duration for which the participant engages in the conversation will be documented.

Confederate ratings

After the task, (role-playing) confederates will provide the following ratings [13]: (1) overall anxiety (how anxious the participant appeared during the task) on a scale from 1 (animated – spontaneous expression of emotions, very engaging, clearly comfortable in control, effective interactions no overt signs of anxiety) to 5 (severe anxiety – clearly uncomfortable, overt signs of anxiety [hand wringing, sweating, flushing, turning, fidgeting, unable to speak at all]); and an (2) overall skill (how well the participant performed during the task) on a scale from 1 (not effective at all – extremely awkward, barely responds if at all; does not ask questions.) to 5 (very effective – good interpersonal skill, carries part of conversation, self discloses, uses appropriate transitioning, enjoys interaction).

Psychophysiological measures

During the Behavioral Avoidance Test (BAT), psychophysiological measures will be recorded to assess participants’ physiological responses. Heart rate (HR), heart-rate variability (HRV), and skin conductance levels (SCL) will be measured using the MP160 System, equipped with the EDA 100D and ECG 100D amplifiers. Electrodes appropriate for each measure will be used: disposable Ag/AgCl electrodes for ECG to ensure accurate cardiac signal acquisition, and conductive adhesive electrodes for EDA to monitor skin conductance. EDA electrodes will be attached to the volar surfaces of the distal phalanges (index and middle fingers), while ECG electrodes will follow a standard three-lead configuration to maximize signal quality. Data will be collected and analyzed using AcqKnowledge software (version 5.0), which allows for real-time visualization and post-recording analysis of physiological signals. Baseline measurements will be taken for 5 min before the BAT while participants watch a neutral video to establish a physiological baseline. All recordings will be conducted in a controlled environment to minimize artifacts, and participants will be instructed to minimize movement.

Measures related to general wellbeing

Brief Resilience Scale (BRS)

The BRS [97] consists of 6 items and is utilized to assess participants’ ability to bounce back or recover from stress. Each item is rated on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). The BRS has demonstrated construct validity and has shown strong internal consistency (Cronbach’s alpha values above 0.80) and good test–retest reliability [97, 98].

Multidimensional Adolescent Functioning Scale (MAFS)

The MAFS [70] is employed to evaluate psychosocial functioning in adolescents. For the current study, the subscales peer relationships and overall functioning will be used. These two subscales consist of 16 items which are rated on a 4-point Likert scale (1 = strongly disagree, 4 = strongly agree). The MAFS has demonstrated good internal consistency and construct validity [70].

Self-Efficacy Questionnaire for Children (SEQ-C) – Social Efficacy subscale

The social efficacy subscale of the SEQ-C [75] assesses individuals’ beliefs in their ability to perform social tasks and interact with others effectively and consists of 8 items. Responses are rated on a 4-point Likert scale (1 = not at all true, 4 = very true). The scale has indicated good to excellent internal consistency and validity [31, 75].

Revised Children’s Anxiety and Depression Scale (RCADS)

The RCADS [24] is a 46-item questionnaire used to assess the levels of anxiety and depression experienced by adolescents. Each item is rated on a 4-point Likert scale (0 = never, 3 = always). It measures various symptoms related to anxiety disorders and depressive disorders. The RCADS is widely used and has demonstrated excellent internal consistency and convergent validity [24, 61].

Adolescents Social Cognitions Questionnaire (ASCQ)

The ASCQ [64] is a 27-item questionnaire used to assess commonly experienced cognitions in stressful social situations (during the last week). First, each item’s frequency is rated on a 5-point Likert scale (1 = thought never occurs, 5 = thought always occurs) and subsequently each item’s believability is rated on a scale ranging from 0–100. The ASCQ has shown strong internal consistency and good construct validity [64].

Objective 2

To elucidate potential working mechanisms of VRE.

Mechanisms

Inhibitory Learning Theory: Expectancy violation and change

In line with recommendations by Craske et al. [29, 30], before each exercise, the trainer and participants will identify the participants’ feared outcome expectancy (e.g., what do you fear the most will happen?). Pre- and post- exposure Visual Analogue Scales (VASs; on a scale from 0 to 100) adapted from Craske et al. [29, 30] and [81] will be used to measure violation and change of threat expectancies. To operationalize expectancy violation, we will measure threat expectancy likelihood (How likely is it that your greatest worry will come true on a scale from 0 [not at all] to 100 [definitely]) and threat occurrence (To what extent did your expected threat / feared outcome actually come true on a scale from 0 [did not occur at all] to 100 [occurred exactly as I expected it].

To operationalize expectancy change, threat expectancy likelihood will be measured before and after exposure. Expectancy violation will be operationalized as the difference score between threat expectancy likelihood before the exercise and threat occurrence after exposure. Expectancy change will be operationalized as the difference score between threat expectancy likelihood before and after exposure (i.e., adjusted threat expectancy). As per Pittig et al. [81], a single learning rate value per participant will be estimated (e.g., using a Gaussian likelihood distribution), representing how much expectancy violation was translated into expectancy change. A learning rate of 1 will indicate that the full extent of the violation was reflected in the expectancy change, while a learning rate of 0 means that no expectancy change occurred.

Finally, surprise and relief (pleasantness) will also be measured using VASs after each exercise to capture more subjective (gut-feeling) types of expectancy violation. Surprise will be assessed by asking participants how surprised they were by what actually happened during the exercise, using a scale from 0 (not at all surprised) to 100 (extremely surprised). In cases where the participants’ feared expectancy does not occur, relief will be measured by asking them how relieved they felt by the outcome, specifically that their expectation was not realized, using a scale from 0 (not relieved at all) to 100 (extremely relieved). Additionally, the pleasantness of the relief will be evaluated by asking participants how pleasant the relief they experienced felt, using a scale from 0 (not pleasant at all) to 100 (extremely pleasant).

Emotional Processing Theory: Habituation

Furthermore, as proposed by Foa and Kozak [41], within- and between sessions fear reduction will be measured as indicators of within- and between session habituation. Both measures will be operationalized using subjective units of distress (SUDs; self-ratings of distress ranging from 0 “complete relaxation’’ to 100 “maximum distress’’; [113]). To measure within-session habituation (WSH), peak SUDs and end SUDs will be collected at the end of each exercise. More specifically, WSH will be operationalized as the difference score between peak SUDs (maximum anxiety experienced during the exercise) and end SUDs (referring to anxiety experienced during the last couple of minutes before the end of the exercise). Moreover, between-session habituation (BSH), will be calculated as the difference score (in peak SUDs) between two subsequent exposure sessions.

Self-efficacy theory

Finally, in addition to the SEQ-C mentioned above, participants’ self-efficacy (i.e., How confident do you feel that you can cope well in an anxiety provoking social situation?) will be acquired using a VAS [9].

Objective 3

To identify predictors of intervention outcome.

Predictors

Clinical variables

Pre-intervention severity of social anxiety

During the screening phase, social anxiety severity will be assessed in two ways: (1) self-reported severity using the SPIN [27] and (2) clinician-rated severity, including SAD diagnosis and functional impairment level (mild or severe), evaluated with the social anxiety module of the SCID-5 Junior [109] and CSR ratings from the ADIS-C [96].

Symptomatic depressive comorbidity

At baseline, this will be assessed using the depression scale of the RCADS [24].

Other comorbidities

These will be explored as predictors (e.g., symptomatic comorbidity of global anxiety levels as measured by RCADS, and substance use, suicidal thoughts, and self-harm as measured by SCID-5).

Personality and individual difference variables

Temperament

The Behavioural Inhibition System and Behavioural Activation System Scale (BIS/BAS Scale; [105]) is a 24-item scale which is widely used to measure individual differences in sensitivity to punishment (BIS) and reward (BAS). Responses are recorded on a 4-point Likert scale ranging from 1 (not true for me) to 4 (true for me). The BIS/BAS Scale exhibits strong psychometric properties, including high internal consistency and construct validity [22, 105].

Attachment-related avoidance and anxiety

The Experiences in Close Relationship Scale – Relationship Structures (ECR-RS; [44]) will be used to measure attachment style for the person they are most attached to, consisting of 9 items rated on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). It has good internal consistency and validity in assessing different dimensions of avoidance and attachment [44, 95].

Visual analogue scales (VASs)

Three VASs (ranging from 0 to 100) will be used to measure participants’ motivation to participate in the training, need for connection/relatedness, and openness to experiences, with higher scores reflecting greater levels of each construct. VASs are well-regarded for simplicity and reliability in capturing subjective experiences [39, 71].

VR-related variables

Immersion

The involvement scale of the Immersive Tendency Questionnaire (ITQ; [88, 112]) will be used to measure an individual’s propensity to become immersed in various experiences, such as books, movies, and computer games, and consists of 6 items on a 7-point Likert scale ranging from 1 (never) to 7 (often). The ITQ has demonstrated strong reliability and validity [112, 93].

Attitude towards technology & experience with computer game playing and VR

The Attitude towards Technology (AT) subscale of the Unified Theory of Acceptance and Use of Technology (UTAUT) questionnaire [103] will be used to measure attitude toward VR. The AT consists of 4 items and uses a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The UTAUT has shown good psychometric properties, including high internal consistency and construct validity, in assessing technology acceptance [103, 107]. Furthermore, experience with VR and computer games will be assessed through self-reported frequency of use (e.g., self-reported multiple-choice responses to questions such as how often do play computer games, how much experience do you have with VR, etc.,).

Training-related variables

Working alliance

The 12-item Working Alliance Inventory-Short Form [99] will be used to measure the quality and strength of the therapeutic alliance between the trainer and the participant, using a 5-point Likert scale ranging from 1 (seldom) to 5 (always). The scale is well-established, with high reliability and validity in measuring therapeutic alliance [52, 99].

Preferred treatment

Preferred treatment (IVE vs VRE) will be measured prior to the sessions, using a binary scale.

Treatment expectancy

The Credibility/Expectancy Questionnaire (CEQ; [35]) and the 4-item subscale Outcome Expectancy of the Unified Theory of Acceptance and Use of Technology (UTAUT; [36, 103]) will measure expectancies regarding the treatment and the expected benefit on personal mental health by using VR, after session 1. The CEQ consists of 6 items in total, primarily using a 9-point Likert scale ranging from 1 (not at all) to 9 (very much), with the exception of two items, which are rated on a scale from 0 to 100%.

Objective 4

To assess adolescents’ acceptance of VRE and IVE while gaining insight into their experience with the training.

Exit interview

A semi-structured interview will be conducted in a randomized sample of participants from both active conditions to investigate VRE’s acceptability and to compare it to IVE. The sample will consist of approximately 10–16 participants per condition, based on general recommendations as to when theoretical data saturation can be reached [47, 106]. The interview will focus on the following themes: general experience, perceived advantages and disadvantages of VRE/IVE, personal preference between IVE/VRE, potential improvements to IVE/VRE. To decrease demand effects, the (online) interview will be conducted by an independent interviewer (e.g., not the trainer).

Credibility and Expectancy Questionnaire (CEQ)

The CEQ [35] will be used to assess outcome expectancy and credibility of the intervention after the training rationale is provided (end of session 1).

Client Satisfaction Questionnaire (CSQ)

The CSQ [63] will be administered at post-assessment to measure (quantitative) feedback regarding the training. It consists of 10-items rated on a 4-point Likert scale ranging from 1 (very dissatisfied) to 4 (very satisfied) and has demonstrated strong psychometric properties, including internal consistency and validity in measuring client satisfaction [63],Attkisson & Greenfield, 2004).

Participant feedback during weekly sessions

VASs will be administered to measure exercise difficulty, exercise usefulness/relevance, use of safety behaviors, and motivation to participate in the next session, during the weekly sessions. In the VRE condition, during each exposure exercise, measures of presence will be obtained by a single item: “To which extent did you feel present in the virtual environment(s), as if you were really there?” [17]. Finally, refusal and drop-out rates (i.e., participants who have completed fewer than 5 training sessions) will also be recorded.

Monitoring

Data collection and storage

Data collection will be conducted in compliance with the requirements of the EU General Data Protection Regulation 2016/679 (GDPR), the relevant Belgian laws implementing GDPR in relation to the processing of personal data. A data management plan has been developed following the Faculty of Psychology and Educational Sciences of KU Leuven guidelines and it has been approved by KU Leuven’s Research Data Management. To ensure participants confidentiality, all online assessments will be pseudo-anonymized and conducted via a secure online survey software called Qualtrics, where anonymized data (i.e., via a unique participant code) will be temporarily stored in a secure server of Utrecht University. Only the researchers directly involved in the project will have access to any linking information (i.e., identifiers connecting unique participant code to personal information). Access will be granted through their university accounts using a multi-factor authenticator. No interim analyses will be conducted as data collection will be continued until the desired sample size is achieved.

Safety and adherence

To monitor intervention adherence and competence, the sessions will be audio-recorded when possible and regularly discussed during supervision. Based on similar prior studies, we do not anticipate heightened risks for participants undergoing the current interventions. To mitigate any potential risks of harm, we have explicitly defined acute suicidality and / or self-harm behavior as exclusion criteria. In the unlikely event of any serious adverse events (SAEs) immediate actions will be taken by the involved researchers and reports will be submitted to the ethical committee of UZ/KU Leuven. In any case, yearly progress reports will be submitted to the ethical committee of UZ/KU Leuven.

Analyses

Sample size calculations

Given the lack of a clear consensus on power analyses for (linear mixed models) LMMs, researchers often use simulation-based approaches. For the current study, given time constraints, we consulted statisticians and it was determined that the required sample size should not exceed that of traditional repeated measures ANOVA. This expectation is supported by evidence that multilevel modeling accounts for the sampling hierarchy, effectively handles missing data, and generally offers greater statistical power, as demonstrated in simulation studies [83]. To obtain an estimation of the required sample size, a power analysis was conducted using G*Power. Since repeated measures ANOVA (within-between interactions) has been argued to yield overly liberal estimates, a more conservative "between-factors ANOVA" approach was used.

For the comparison between the active conditions and the waitlist (WL) condition, we conducted a power analysis based on three measurement points (pre-, mid-, and post-assessments) across three groups (VRE, IVE, and WL). We set an alpha level of 0.05 and a power level of 0.80 to detect a moderate effect size of 0.25, informed by prior studies investigating VR for social anxiety [2, 16, 54]). The results indicated that a total sample size of 108 participants is required. For the comparison between the two active conditions, we conducted a power analysis with 5 measurement points (pre-, mid, post-, follow-up 1-, follow-up 2- assessments) across two groups (VRE, IVE). We used an alpha level of 0.05 and a power level of 0.80 to detect a moderate effect size of 0.25. The results indicated that a total sample size of 78 participants is required. Combining the results of both power analyses, the proposed study with a total sample size of 108 participants (36 per condition, prior to anticipated attrition) should be adequately powered (≥ 80%) to detect the hypothesized effects. To account for anticipated drop-out, we aim to recruit 40 participants per condition, resulting in a total planned sample size of 120 participants.

Primary outcomes analyses (Objective 1)

Social anxiety (verbal measures)

A series of linear mixed models (LMM) will be conducted with two levels (time points [Level 1] nested in individuals [Level 2]) to assess the effectiveness of virtual reality exposure (VRE) compared to in vivo exposure (IVE) and the waitlist control group in reducing social anxiety symptoms. Specifically, we will analyze social anxiety symptom scores (one for SPAI-18 and one for LSAS-avoidance subscale) as dependent variables, measured at three time points during the training period: pre-training (T1), mid-training (T2), and post-training (T3). Fixed effects in the LMM will include training group (VRE, IVE, WL), time (T1, T2, T3), and the interaction between training group and time. The main effect of training group will test for overall differences among the three groups, while the main effect of time will examine symptom changes over time. The interaction term (training group × time) will assess whether changes over time differ between groups. A random intercept will be included to account for individual differences in baseline social anxiety symptoms, and we will test whether adding a random slope for time improves model fit using likelihood ratio tests and AIC.

We will also run another analysis including the follow up period but this comparison will only involve the two active conditions (VRE, IVE) at T1 (baseline), T2 (mid), T3 (post), T4 (follow-up 1), and T5 (follow-up 2). Notably, this will also include participants who were randomized into one of the two active conditions after being in the waitlist control. The dependent variables will remain social anxiety symptom scores (SPAI-18 and LSAS-avoidance subscale). Fixed effects will include training group (VRE, IVE), time (T1, T2, T3, T4, T5), and the interaction between training group and time. Specifically, this analysis will focus on determining whether VRE and IVE are equally effective (main effect of training group) and whether their effectiveness differs across time points (training group × time interaction). A random intercept will again account for individual differences in symptom levels, and a random slope for time will be tested for inclusion based on model fit. The analyses will be conducted using R (package:lme4).

Social anxiety (behavioral measure)

To evaluate the effects of the intervention on peak anxiety levels during the BAT, a 2 × 3 mixed-design ANOVA will be conducted. This analysis will include two factors: a within-subjects factor (measurement point: baseline and post-assessment) and a between-subjects factor (condition: VRE, IVE, WL). The ANOVA will assess the main effects of measurement point and condition, as well as their interaction, to determine whether changes in peak anxiety levels over time vary across conditions.

Secondary outcomes analyses (Objective 1)

Social anxiety measures

To evaluate the effects of the intervention on the fear subscale of the LSAS-CA and the SPWSS, we will run linear mixed models (LMM) similarly to the primary outcome measures.

General wellbeing measures

To assess the broader impact of the intervention on general wellbeing, we will also conduct linear mixed model (LMM) analyses for each of our secondary outcome measures: resilience, psychosocial functioning, social self-efficacy, (general) anxiety and depression, and social cognitions. These models will follow the same structure as those used for the primary social anxiety measures, with each outcome treated as the dependent variable. Fixed effects will include training group, time, and their interaction, while random effects will account for individual differences in baseline scores (random intercepts) and, where appropriate, variability in change over time (random slopes). This approach allows us to examine changes in each wellbeing construct across time and between training groups, while accommodating the nested structure of the data and any missing values.

Intent-to-treat analyses

All analyses will be conducted using an Intent-to-Treat (ITT) approach, including all participants according to their initial group assignments. Missing data from drop-outs will be treated as missing values under the assumption that data are Missing at Random (MAR; unless the data show otherwise), while missing data from participants who have completed at least five training sessions will be handled using multiple imputation. This approach provides a robust estimate of the intervention’s effectiveness in a real-world context, preserving the benefits of randomization. By using ITT for all primary and secondary outcomes, we aim to capture the true impact of the intervention across all conditions while accounting for potential challenges such as missing data and adherence.

Mechanisms (Objective 2)

To investigate the mechanisms underlying the intervention’s effect on social anxiety symptoms, we will conduct Linear Mixed Models (LMM) to examine how changes in key mechanisms—threat expectancies (violation and change), fear levels (SUDs), and self-efficacy—assessed at each of the seven weekly sessions, relate to changes in social anxiety symptoms at the three main assessment time points (T1, T2, T3). To test for mediation, we will use a combination of LMMs for the mechanisms and outcome models, along with bootstrap resampling to estimate indirect effects and obtain confidence intervals. Additionally, we will explore potential moderation effects, testing whether the mechanisms differ in their role between the two active treatment conditions, VRE and IVE, by including interaction terms between treatment condition and mechanisms in the models. Model fit will be evaluated using likelihood ratio tests and AIC, ensuring that the best-fitting models are selected for interpretation. All analyses will be conducted with the assumption that the data are missing at random, and missing data will be handled appropriately using maximum likelihood estimation (MLE) or multiple imputation.

Predictors (Objective 3)

To test the effects of the predictors on VRE and IVE outcome (Objective 3), multiple linear regression analyses will be conducted with primary outcome(s) at post-assessment as dependent variable(s). Exploratory, primary outcome measures at follow-up assessment will be used as dependent variables. Analyses will be run by controlling for pre-intervention severity and related variables (e.g., include all treatment-related factors as predictor variables).

Acceptability & attitudes towards VRE and IVE (Objective 4)

Qualitative data (post-treatment exit interviews) will be analyzed using thematic analysis as described by Braun and Clarke [18, 19]. This qualitative methodology sets out to discover emergent themes in the available data. This approach ensures to reveal theoretically sound information about the way the intervention is received and appreciated by the adolescents. Furthermore, a number of session-related variables (e.g., quantitative data such as outcome expectancies and credibility, anticipatory anxiety, willingness to engage in exposure and refusal and drop-out rates) will be compared between the VRE and IVE conditions by using t-tests and chi-squared tests.

Discussion

We present a study protocol for a randomized controlled trial (RCT) where the primary objective is to investigate the efficacy of virtual reality exposure (VRE) for adolescents with social anxiety (SA) and compare this to in vivo exposure (IVE). To our knowledge, this is the first large-scale RCT comparing the effects of VRE to IVE and a waitlist (WL) condition for adolescents with generalized SA. Although VRE has been extensively investigated in adults with SA, high-quality studies focusing on adolescents are notably lacking.

This study also represents one of the few efforts to isolate the specific effects of exposure for SA. In most previous research, exposure is typically combined with cognitive interventions, such as cognitive-behavioral therapy (CBT), making it difficult to disentangle their unique contributions. By primarily focusing on exposure, our protocol provides a valuable opportunity to evaluate its isolated effects as an (early) intervention for adolescents with SA. Given the focus on early intervention, our approach (e.g., 7 weekly sessions instead of 10–12 which is found in CBT) is designed to be practical and accessible for adolescents who may not yet have fully developed severe SA. This design also aims to accommodate the unique challenges faced by this age group, including long school hours and various extracurricular commitments, making the intervention a better fit for their busy schedules.

Moreover, to optimize any intervention, it is crucial to obtain a comprehensive understanding of the mechanisms at play. In light of this, during this study we will explore the assumption that similar mechanisms are at play in VRE and IVE. Traditionally, fear habituation was considered the ‘active ingredient’ of exposure [42, 86]. However, in recent years, the consensus has shifted towards viewing a mismatch between expected and actual outcomes as the most significant driver of symptom reduction and as such exposure is often conducted in ways where this mismatch can be maximized [29, 30, 81]. Notably, in this study we will not design the exposure exercises based on the inhibitory learning theory or the emotional processing theory. In other words, we will not repeat a given exposure exercise until we have reached a specific criterion, such as a reduced threat expectancy likelihood or fear levels. Instead, we will deliver exposure exercises tailored to the needs of the participants, based on their case conceptualization, and will measure proxies such as expectancy violation, fear reduction, and self-efficacy without effectively manipulating or inducing either construct. This study is among the first to investigate these three mechanisms together, which is critical for advancing our understanding of how (VR) exposure-based treatments achieve their effects and for refining therapeutic interventions.

In addition to the mechanisms, we aim to examine various predictors with an ambition to identify overarching predictors of outcome as well as differential predictors tailored to the effects of VRE. Even if VRE is, as hypothesized, effective in adolescents with SA at the group level, there might still be individual differences in the outcome, with the intervention being less or more effective for at least some adolescents. Therefore, we aim to address an important, yet under-investigated question in research on VRE – namely, which characteristics are related to less or more successful outcomes [80]. This knowledge will help to enhance intervention assignment and identify those individuals who will profit from VRE as compared to those who might benefit more from IVE.

Finally, our goal is to obtain more in-depth insights into the experiences of participants and the extent to which they accept VRE compared to IVE by integrating qualitative interviews into our methodology. It has generally been considered that VRE may have a potential advantage over IVE due to its more appealing nature to adolescents which in turn may boost acceptability and motivation for an exposure intervention (Katharina & Morina, 2021). However, this belief lacks empirical support in the context of socially anxious adolescents. With this study, we will integrate quantitative and qualitative data to address this gap. More specifically, we will quantitatively examine outcome expectancies and credibility of the intervention, while also qualitatively exploring the experiences of a sub-sample of participants, in both VRE and IVE for an active comparison. If the findings suggest a greater favorability towards VRE and a greater willingness to participate in VRE compared to IVE, it could substantially improve the treatment options for adolescents with SA and underscore the importance of utilizing VR-based exposure for this population (provided VRE proves to be effective).

In defining our sample, we made careful decisions aligned with our research objectives and procedures. One significant decision was to exclude participants with an autism spectrum disorder (ASD) diagnosis, as individuals with ASD typically benefit from exposure interventions specifically adapted to their unique needs, an accommodation our current protocol could not provide. More specifically, although they may profit from CBT, the extent of improvement tends to be less pronounced (than with children without ASD; [104]). We also excluded adolescents with ASD because measuring expectancies can be more challenging in this population. They often struggle to anticipate social expectations [43], which could bias our measurements of expectancies. Although this exclusion could be considered a limitation in terms of generalizability, it was necessary to preserve the internal validity of our study. By doing so, we ensured that our exposure interventions would accurately target the social anxiety characteristics of our intended population without introducing variability related to ASD-specific social experiences and skills. One of our priorities, nonetheless, was to minimize exclusion criteria as much as possible to improve the generalizability of our findings. Therefore, we included participants with conditions that did not require protocol adaptations and could be handled by the trainers under the supervision of certified CBT therapists (e.g., non-severe suicidality, self-harm, PTSD).

On a more practical note, one of the strengths of this study is the type of software it utilizes to deliver VRE. It has been argued that it may be more difficult to reproduce social situations in VR environments, compared to other phobic stimuli (e.g., heights, flying), due to the complexity of human interactions [4]. Hence, arguably the use of VR with 360-degrees film content or scripted environments with avatars may not render the most optimal results for social fears. In the present study, we will be using dynamic, interactive virtual worlds. It is interactive in the sense that the environment can react directly to the participant’s behavior and dynamic since the trainer has the flexibility to modify the virtual environment as needed, at any point during the session. As such, this software appears to be an ideal candidate for the use of VRE for SA.

In conclusion, this RCT will be the first large-scale study to investigate the efficacy of a promising intervention, namely exposure by means of VR, for adolescents with subclinical to mild and moderate SAD. Considering how debilitating this condition can be, it is vital to improve adolescents’ access to treatment, especially given their hesitancy to seek help despite effective options like exposure-based treatments being available. Findings from this study will help determine if VR-based exposure may be a more viable and appealing option for adolescents with SA. Furthermore, the results of this study may lead to a better understanding of “how” and “for whom” VRE works, thereby facilitating the development of more tailored interventions.