We followed the PSD model (see below) to specify our system requirements and guide the implementation of related persuasive design features [40]. Building on our theoretical analysis of the persuasive intent, event, and strategy, we conducted 3 focus groups with experts to cocreate the IVR (Figure 1). For this, we focused on the (1) IVE, (2) persuasive ECA, and (3) dialog. Based on the derived requirements, we built our first IVR prototype. We then conducted another focus group with the same experts to evaluate the IVR for clinical suitability and to discuss its potential uses for treatment, considering our ethical responsibility for patients with MBID.

Participants were recruited from the staff of a Dutch addiction clinic, specialized in treating individuals with MBID and substance use disorder. Exclusion criteria for our expert evaluation were a history of migraine, epilepsy, motion sickness, severe visual or motor impairment, or being unable to wear the head-mounted display (HMD).

In total, 2 psychologists, 1 psychiatrist, 1 nurse specialist, and 1 psychomotor therapist participated. Experts were all female, had a mean age of 37 (SD 10.3) years, and worked on average since 9.8 (SD 7.2) years in addiction care. The technology experience rating (using a 7-point Likert scale) indicated a high computer (mean 5.20, SD 1.1) know-how compared with a medium videogame (mean 3.20, SD 1.5), as well as low IVR (mean 2.40, SD 1.1) and dialog system (mean 1.80, SD 0.8) knowledge. Most (n=3) had previous experience with IVR for addictive disorders from prior research projects of our group.

For our cocreation, we formulated brainstorming goals based on the 3 major components found in the PSD template: (1) IVE, (2) persuasive ECA(s), and (3) dialog. For (1) IVE, we asked “In what contexts do patients with MBID and AUD face persuasive attempts by another person to drink together?” to identify the 5 most significant contexts. For our (2) persuasive ECA, we asked “How should the persuader in the context of [outcome (1)] look like?” to identify appearance (eg, look, age, gender, body shape, clothing, and properties) and character trait requirements. For this, experts were asked to rapid prototype an ECA using a paper-based template. Lastly, for the (3) dialog, we asked “What are typical arguments of the persuader?,” “What are typical appeals to emotion of the persuader?,” and “What are good, fair, or bad coping reactions by the patient based on the identified appeals” to create ECA interactions. We used the collaboration platform Miro to collect our data.

For our expert evaluation, we created a semistructured focus group with 6 open-ended questions, elucidating the (1) experts’ first impression of the IVR and ECA interaction, that is, (2) user experience, (3) persuasive mechanisms, and (4) persuasive power; (5) procedures for patients with MBID and AUD; as well as (6) experts’ intention to use our IVR for treatment. To support recall, we provided paper-based templates to note down experiences and observations during IVR try out.

Ethics approval was obtained from the University of Twente’s ethics committee (approval number RP 2021-154) and the care institution’s scientific board.

For the cocreation and evaluation, all participants were thoroughly informed, signed the informed consent, and completed a demographic questionnaire. After welcoming participants to the focus group sessions, the audio, screen, and video recordings were started. For cocreation, the researcher explained brainstorming rules and introduced a specific goal for each session. For insights into the research background, a presentation was shown during the first focus group. In subsequent sessions, experts were reminded to adhere to rules and the research background was briefly repeated.

For the expert evaluation, another focus group with the same experts was held 16 weeks later. We invited participants in groups of 2 to evaluate and observe the prototype with the IVR apparatus alternately. First, 1 participant was introduced to the HMD and controllers, while the other one received a paper-based template for taking notes by observing the participant and IVR. Then, a tutorial was conducted to train interactions, by letting the participant use teleport locomotion, customize the virtual hand’s skin tone, and experience the dialog UI without ECA and alcohol cues present. Upon completion, experts were immersed into the enriched IVE via verbal storytelling and asked to engage in a dialog with the ECA by selecting (refusal) responses on the related UI. Finally, experts were asked 5 questions using a Visual Analog Scale (VAS; ie, cravings, ECA persuasiveness, presence, anthropomorphism, and perceived safety), implemented for future patient evaluations. Then, experts alternated, and the procedure was repeated. Upon completion, a focus group was held with all experts using Microsoft Teams to evaluate the IVR prototype toward clinical suitability. The experts received no compensation besides their regular salary.

We used the filled-in digital and paper-based templates, along with ranking and agreements during our focus group sessions, to design our IVR. The audio recording from our expert evaluation was transcribed verbatim for subsequent thematic analysis, with the aim of identifying requirements for our IVR development. We applied the reflexive protocol by Braun and Clarke [46,47], by using the researcher’s theoretical sensitivity and reflexivity, to analyze experiences of experts when interacting with our IVR. For this, the researcher SL followed the 6-step protocol: For (1) data familiarization, SL listened to the recording and (re)read the transcript. Subsequently, (2) relevant segments in our data were identified, coded, and collated. Afterward, (3) initial (sub)themes were generated using our coded data and (4) reviewed by revisiting the themes with respect to data set and goal. Lastly, (5) final themes were defined and our (6) report was produced. For this, we used the qualitative data analysis program ATLAS.ti (version 9.1.6; ATLAS.ti Scientific Software), and themes were discussed among participating researchers (ie, SL, JV, and RK) during refinement (steps 4 and 5) to eliminate discrepancies within the analysis.

In the following section, we describe our (1) findings from the cocreation with experts in addiction care, and (2) prototype development with subsequent expert evaluation to tailor IVR and procedures toward clinical applications.

This work reports on the cocreation of an IVR alcohol refusal training for patients with MBID and AUD. Following our theoretical PSD, we conducted 3 focus groups with experts from a specialized addiction clinic for patients with MBID to understand the persuasion context and establish design guidelines. Based on the identified requirements, a first peer pressure simulation was developed. Then, we evaluated the IVR with the same experts to tailor prototype and procedures toward clinical usage. For our IVR design, experts described the risk situation of visiting a friend at home with multiple friends. Based on the data, we developed a social-housing apartment with alcohol-related objects and intoxicated virtual humans present. To simulate peer pressure, we embedded a virtual man with generic appearance who persuades patients with MBID to drink alcohol. Patients can practice alcohol refusal skills with the ECA by selecting responses from a dialog UI. Following the playtesting, experts described the implemented multimodality, interactivity, and narratives as factors for deep immersion. For peer pressure, the ECA’s insulting dialog and nonverbal animations (eg, offering alcohol) were described as contributing design features. To advance the ECA’s persuasiveness, however, improving voice vividness (eg, paralinguistics) and adding group dynamics were suggested. Still, the IVR should remain customizable (ie, IVR, training difficulty) based on the patient’s personal goals. For behavior change support, experts preferred therapist over system-delivered interventions to avoid a perilous try-and-error in patients with MBID. Lastly, facilitators (ie, embodied interactions, tutorial, VAS) and barriers (ie, text-based procedures, complex terminology) for IVR accessibility were described.

For our IVE design, experts described multimodality, interactivity, and narratives as important factors to immerse patients with MBID. Here, we used our prior experience in developing IVR interactions for this group, by designing a small but realistic IVE with embodied interactions, anchor-based teleportation, and controller scaffolding [48]. Previous work reported such error-free and positive user experience as a major factor leading to user satisfaction [41,49]. Similar to our prototype, other scholars built alcohol-related IVEs for exposure purposes [50]; however, to elicit realistic behavioral responses, plausible interactions are required [51]. Hence, compared with others, we advanced IVR interactions on the object/conversation level and integrated a narrative to engage users realistically with our tutorial and subsequent refusal skills training. Interestingly, narratives might act as persuasive feature [52], presumably as part of simulation [41]. Previous work showed that narratives can influence behavioral intentions for health behaviors [52]. Hence, influencing and experiencing the own narrative appear interesting for IVR therapy.

For our ECA, we established a first PSD to avoid blackboxing [53], allowing to identify active PSD elements and replicate findings in diverse user and use areas. For our cocreation, we looked into the ECA and persuasive dialog to connect specific design elements to our PSD. Here, we systematically designed for the social role of a virtual friend to transfer emotional appeals that induce realistic reactions (eg, guilt) in patients. Previous studies on IVR for AUD embedded mostly “nonintelligent” agents as social cues [54-56]. Yet, as learning paradigms appear promising for IVR addiction therapy [54], interactable and persuasive ECAs are needed to simulate peer pressure, a factor well-known to cause relapse [2]. Using our IVR enables novel treatments using direct peer pressure, but also adds realism to indirect peer pressure during virtual reality therapy.

Nonetheless, experts described our ECA as low in (persuasive) power due to a lack in voice vividness. Similar findings were observed when persuading with anger [31], though this was dependent on power relations. This is fascinating, given that power dimensions appear similar to group dynamics in peer pressure (ie, group leaders exhibit greater power [31]). Thus, presumably ECAs must be seen as a cognitive entity with social power to convey appeals to emotion/habit (normative influence) in IVR [57]. Our findings suggest that this may be achieved through the body’s (non)verbal communication channels [58], such as voice vividness (ie, paralinguistics) and animations (eg, drinking). Further, to establish social power, narratives may be useful to build rapport with ECAs, while multiagent interactions (eg, 2 against 1) could establish social hierarchies.

Lastly, for clinical usage, experts preferred therapist- as opposed to system-delivered interventions to explore symptoms, stimulate self-reflection, and train coping skills [59,60]. Similar findings were reported by Skeva et al [59], who described the gain of self-efficacy during relapse prevention trainings as the key factor for control in reality. For this, we assume that rehearsal may be refined by letting individuals express utterance themselves instead via a narrator voice, while keeping the feature for accessibility. Lastly, we found that Skeva et al [59] emphasized cognitive-behavioral practitioners to engage with IVR, while particularly vulnerable groups could benefit from body-centered paradigms. In our evaluation, we found that disciplines would use this tool differently [59], for instance, with focus on bodily signals during psychomotor therapy. Thus, future research should create IVR therapy protocols using interdisciplinary teams to establish a taxonomy with learning goals [54,61]. Further, system-delivered approaches should not be neglected, for example, as homework [59], as IVR therapy is often seen as adjunct treatment.

The findings of this study have to be seen in light of some limitations: First, we cocreated our IVR prototype by involving specialized professionals from an addiction clinic, therefore, perception in patients with MBID and AUD might differ, given the preconditioned behavioral patterns in AUD. Hence, patient evaluations are required to validate the ECA’s persuasiveness, including replication studies using our template. Our PSD enables scholars to develop similar experiences through an analogous user-centered design using our ECA architecture. Further, scholars should learn from our unimplemented requirements, such as (intra)group dynamics (eg, multiagent scenarios) and natural voices with paralinguistics, to induce peer pressure. Second, we have not yet explored ECA animations in a systematic manner. Instead, we used animations for a natural interaction. Hence, nonverbal communication forms a promising factor for normative influence that should be addressed in future work, although it was not missed by our experts. For this, scholars should consider facial expressions [31], gaze, and gestures, presumably combined with natural speech to create plausible interactions [51]. Lastly, we implemented a traffic light feedback as part of the dialog UI to make self-monitoring discussable. However, future research should systematically explore error-free training approaches that make use of formative and summative feedback.

Future work should clarify factors that enable ECAs to elicit peer pressure in IVR. Our findings suggest that paralinguistic and nonverbal information form vital factors for persuasion toward alcohol use in patients with MBID and AUD. Hence, subtle emotional expressions, along the verbal appeals, seem to take a crucial role for persuasion using virtual humans, which connotes to further explore interpersonal (eg, rapport, group dynamics, interpersonal stance) and empathetic (eg, facial expressions, social touch) behaviors along the aforementioned information using natural speech. For rapport building and social hierarchies, integrating narratives seems auspicious, for instance by playing short games with ECAs before starting refusal trainings. Thus, future research with patients should be performed to further investigate specific PSD elements and validate the IVR for future work, in particular to develop therapy protocols with interdisciplinary teams. Lastly, peer pressure via ECAs should be studied more comprehensively to understand the power in influencing humans toward negative behavior, including potential usage for therapy, ethics to adhere to, and possible negative effects on therapy adherence (within outpatient settings).

Our findings establish a PSD for IVR peer pressure simulations in patients with MBID and AUD. Using an analogous design approach, researchers can replicate findings and explore effective PSD features for peer pressure ECAs. Throughout our cocreation with experts, we learned that appeals to emotion/habit should be conveyed via natural human speech channels (ie, verbal and nonverbal). For this, subtle emotional information must be provided via paralinguistics in the ECA’s speech (eg, intonation), gestures (eg, handing over beer), and interpersonal dynamics (eg, group formation, rapport effects). For behavior change support, experts favored therapist-delivered interventions to avoid a perilous trial and error. Future work should further investigate factors contributing to persuasive power in virtual humans, such as emotional information in speech and nonverbal expressions to induce realistic peer pressure. Further, research with interdisciplinary teams should be conducted to establish IVR therapy protocols with specific pedagogical learning goals for our patients.