The French version (Gache et al., 2005) of the Alcohol Use Disorders Identification Test [AUDIT (Saunders et al., 1993)] includes 10 multiple-choice items measuring alcohol consumption, alcohol dependence and alcohol-related problems. High scores reflect a high probability of harmful alcohol use, with a maximum score of 40. Scores from 1 to 7 suggest low risk drinking; scores of 8–15 indicate alcohol use in excess, and scores of 16 and above imply harmful and hazardous drinking (Babor et al., 2001).

The French version (Ansseau et al., 2000) of the Obsessive Compulsive Drinking Scale (Anton et al., 1995) is a 14-item, self-administered questionnaire assessing obsessive thoughts of drinking and compulsive drinking during the last 7 days in two subscales. Scores on each subscale range from 0 to 20. Higher scores indicate stronger obsessions and compulsions towards alcohol.

This questionnaire assesses the use frequency of cannabis, cocaine, ecstasy, hallucinogens, amphetamines, opiates, and volatile solvents in the past 2 months on a 4-point likert scale (“never”—“once/twice”—“more than twice”—“regular use”).

The short version of the French Impulsive Behaviour Scale [UPPS-P (Billieux et al., 2012)] is a 20-item scale assessing impulsivity on five dimensions through a 4-point likert scale (1 = “strongly disagree” to 4 = “strongly agree”). The dimensions are positive urgency (i.e., to act rashly under conditions of positive affect), negative urgency (i.e., to act rashly under conditions of negative affect), lack of premeditation (i.e., the inability to remain focused on a task), and sensation seeking (i.e., the tendency and openness to enjoy exciting or dangerous activities).

Severity of depressive symptoms experienced during the last 7 days was assessed with the French version (Fuhrer and Rouillon, 1989) of the Center for Epidemiologic Studies—Depression [CES-D; (Radloff, 1977)]. The CES-D consists of 20 items rated on a 4-point likert scale (0 = “never” to 3 = “frequently, all the time”). The total score ranges from 0 to 60 with higher scores indicating higher symptom severity.

The Motion Sickness Susceptibility Questionnaire [MSSQ; (Golding, 1998)] assesses motion sickness experienced during various transportation-related activities and other activities. It contains 9 categories of possible experiences that elicit motion sickness (e.g., cars, buses, swings, funfairs). Each item is rated on a 5-point scale (1 = “never” to 5 = “always”).

The French version (Robillard et al., 2002) of the Immersive Tendencies Questionnaire [ITQ; (Witmer and Singer, 1998)] assesses one’s tendency to shut out external distractions in order to focus on different tasks in daily life. It contains 18 items where participants rate their level of agreement on a 7-point scale. Four dimensions are derived: focus (i.e., the tendency to maintain focus on current activities), involvement (i.e., the tendency to become involved in activities), emotion (i.e., the ease of feeling intense emotions evoked by the activity), and game (i.e., the tendency to play video games).

The Gatineau Presence Questionnaire (Laforest et al., 2016) measures the feeling of presence experienced in a virtual environment. Through 4 items rated on a scale of 0–100, this questionnaire assessed 1) the sense of being there, 2) appraising the experience as being real, 3) awareness of the virtual environment as being artificial, and 4) the feeling of being in the physical office instead of the virtual environment. The last two items are scored in reverse and an average score is computed.

We used an HTC VIVE VR headset in which we integrated eye tracking devices from Pupil Labs. The Pupil Labs code is open source and can be adapted to various VR headsets. Specifically, this device consisted of two rings fixed in the headset, one for each eye, with a camera capturing eye images at the bottom, and infrared LED placed around the ring to illuminate the eye.

Once the subject was equipped with the VR headset, the experimenter adjusted the parameters of the pupil detector: minimal and maximal value for the pupil size and coarse maximal value for the pupil luminosity. The virtual simulation started with the calibration in which the participant fixated white dots appearing around a circle on a plane P. During this time, the position of the pupils and their corresponding points on the plane P were gathered and communicated to the VR engine Unity via a communication library called ZeroMQ. Those correspondences were used at the end of the calibration to compute two homography matrices (one for each eye) [H_L, H_R] that mapped pupil positions [E_L, E_R] (pixel of the camera image) to gaze coordinates [G_L, G_R] (point on the plane P in the virtual world). At each frame, the pupil positions [E_L, E_R] were retrieved from pupil labs software through a ZMQ connection, and converted to gaze coordinates [G_L, G_R] on the plane P via the homography matrices [H_L, H_R]. The gaze point G was obtained by computing the mean between G_L and G_R (Figure 1).

During the experiment, participants were immersed in a virtual bar where they heard background music, saw people dancing and/or drinking beer. It included the following spaces: a counter, tables and a dance floor. The experimenter could bring out two light markers on the ground independently of each other (Figure 2). One was near the front door while the other was near the dance floor. These light markers allowed the standardization of participants’ positions in the environment during the immersion. This environment was originally developed by the Université du Québec en Outaouais’ Cyberpsychology lab (Loranger and Bouchard, 2017) and adapted for the alcohol-related research. In a previous study, this environment was shown to elicit alcohol craving, especially in heavy drinking individuals (Simon et al., 2020).

Several items in the environment were considered as ROI. Some of these were associated with alcohol consumption while others were more neutral and used as controls (Figure 2 and Table 2). Alcohol-related cues were selected based on a pilot study in which ten psychology students were asked to identify cues that they believed are associated with alcohol. These ROI were recorded and labelled using colliders. A collider is an invisible volume, usually surrounding a virtual item, which allows the virtual environment to react to an interaction with the item. At each frame, an invisible ray was cast from the centre of the head of the user’s avatar through the point G. The ray stopped when it reached a collider or a distance DMAX. If the ray hit a collider tagging an item in a tracked category, a timer started. It ended when the ray leaves the item and a record was then logged. Else, the ray stopped after a distance DMAX to avoid hitting and recording irrelevant items in the background. Since the whole process was live, no data filtering was done during acquisition.

A fixation was defined as holding one’s gaze on a single ROI over a period of at least 100 ms. A threshold of 100 ms therefore discriminated fixations from other oculomotor activity and was consistent with current physiological and visuo-cognitive models (Manor and Gordon, 2003). For each ROI, a dwell time (i.e., overall fixation time for each ROI) and a number of fixations (i.e., the number of times a fixation was made in this ROI) were calculated. A sum for the duration and the number of fixations was then calculated for all items in the same category (bottle of alcohol, glasses, etc.) (Porras-Garcia et al., 2019).

Although the present study successfully shown that alcohol attentional biases can be assessed in a virtual environment, some limitations must be acknowledged. Firstly, the current study was restricted to a non-clinical population of young social alcohol drinkers. It will be important to replicate the current study in a clinical sample of people with alcohol use disorders, who hold stronger alcohol attentional biases according to the theoretical model of Field and Cox (2008) and the empirical literature (Lusher et al., 2004; Fadardi and Cox, 2006). Secondly, the present study was exploratory in nature. As one of the exploratory goals was to test which predefined cues might or might not relate to alcohol craving and consumption, a stricter control of statistical type I error rate was not applied to the multiple correlations. Additionally, the composite score of alcohol attentional bias was created a posteriori using an inductive procedure from individual correlations. Although the results clearly show significant correlations with predefined alcohol cues, but not with predefined neutral cues, a validation of the composite score will require a replication study.

Another limitation of the present study pertains to the tradeoffs between experimental control and ecological validity. Dot probe tasks and other cue exposure computer tasks heavily rely on contrasting alcohol cues with control cues. The strength of these techniques lies in the experimental control over the stimuli, as they usually oppose very comparable alcohol and control cues, with a strict control over their properties, such as color, brightness, saliency, size and so on. However, such a strict experimental control is obtained at the expense of the ecological validity of the tasks (for a critical review see Christiansen et al., 2015a; Field et al., 2016; Rodebaugh et al., 2016). The use of virtual reality to study alcohol-related attentional biases presents the opposite strengths and weaknesses. It is much closer to how attentional biases work in a natural setting, especially because attention may be tested while the participant is immersed in an alcohol-related context. The tradeoff, however, is that the technique does not allow a strict control on the visual cues seen by the participants. A participant may even miss a significant stimulus in the environment if he never watches in its direction. This why in the present study, alcohol-related cues were not matched one-to-one with control cues similar in color, size or saliency, as it is generally the case with dot probe tasks.

Additionally, it should be noted that the data was collected during the COVID-19 pandemic. During this particular period, dramatic changes in the pattern of alcohol consumption were reported (Pabst et al., 2021). In general, alcohol consumption decreased in Europe due to the closure of bars, restaurants and nightclubs (Kilian et al., 2021), while other studies suggested a slight increase in consumption (Vanderbruggen et al., 2020). This might have affected some of the results in the present study, and especially the AUDIT questionnaire. Indeed, this questionnaire assesses harmful alcohol use during the current year. The AUDIT scores of the participants might therefore capture their unusual patterns of alcohol consumption during the COVID-19 pandemic rather than their more general alcohol consumption tendencies. This could have affected the correlations with alcohol attentional biases and may help to explain why stronger correlations were found with OCDS subscales.

It is postulated that attentional biases for drug-related cues show a fair degree of stability across contexts and time. Once attentional biases have developed, they are not easily affected by the availability of the substance in the environment (Wilcockson et al., 2019). However, the magnitude of attentional biases could fluctuate with current motivational (Hobson et al., 2013; Ramirez et al., 2015; Bollen et al., 2020; Soleymani et al., 2020), emotional (Grant et al., 2007; Bollen et al., 2020) and stress states (Field and Quigley, 2009). Attentional biases might therefore fluctuate according to transitional changes in the appetitive and/or aversive motivational states of the participants (Field et al., 2016). In future studies, it might be interesting to test for correlations of the attentional biases in a virtual environment with phasic rather than tonic measures of craving. These technical limitations should soon be overcome as many researchers are now working to improve the gaze tracking data collection in 3D space (Clay et al., 2019).

As the present study developed a new paradigm for the measurement of attentional focus in a virtual environment, it was not easy to transpose usual features of eye tracking techniques, such as measures of ocular saccades or first ROI visited. Another limitation intrinsically related to the naturalness of the paradigm is the fact that participants were free to blink during the whole experiment, which might have reduced the number and duration of fixations. Finally, as Roy-Charland et al. (2017) study showed, in order to most effectively measure attentional biases in complex scenes, it is important to ensure full attention to the scene. As the main benefit of VR is to allow participants to freely explore an immersive environment, we chose to minimize the instructions given to the participants about their task in the virtual environment. From the debriefing, it was apparent that participants developed various hypotheses about the goal of the study that might have interfered with their attentional focus on specific cues. In future studies, it might be useful to provide participants with more specific instructions.

The experimental paradigm developed in the present study will be useful to further investigate alcohol-induced attentional biases in subclinical and clinical populations of alcohol consumers.

In particular, future studies might use the present paradigm to test the correlation between attentional biases for alcohol-related cues and phasic craving concurrently induced by those cues during the immersion in the virtual environment. In that context, questions remain about what substance-related items to include in virtual environments to induce high levels of phasic craving. According to Pericot-Valverde et al. (2016), substance-related cues can be divided into three categories: proximal, contextual, and complex cues. Proximal cues include all specific objects that are typically associated with the substance. Contextual cues include the physical settings or situations where drugs are used, such as a party or a bar. The settings typically associated with the highest levels of craving are party venues, restaurants, bars or pubs, nightclubs, or being at home (Monk and Heim, 2014; Ghiţă et al., 2019). Finally, complex cues are a combination of contextual and proximal cues, such as situations involving social interactions, where people are dancing, drinking alcoholic beverages, etc. Relative to the more traditional eye tracking paradigm, ET-VR allows studying attentional biases for various alcohol proximal cues while the participant is immersed in an alcohol-related virtual environment (i.e., complex cues). The choice of the proximal and contextual cues incorporated into the virtual environment is therefore a parameter that can strongly affect the intensity of craving and attentional biases, as well the time course of their correlation (Drummond, 2001; Fadardi et al., 2006; Bordnick et al., 2008; Field and Cox, 2008). Additionally, VR allows the personalization of alcohol proximal cues, for example by taking into account the preferred alcoholic beverage(s) of the participant. Therefore, the features of the virtual environment that allow the expression of high levels of alcohol craving and attentional biases are important parameters to investigate. Several authors suggest that a sufficient level of presence during the immersion is necessary for the expression of emotional responses, such as craving (Riva et al., 2007; Diemer et al., 2015). The different dimensions of the sense of presence could therefore mediate or moderate the expression of phasic alcohol craving and attentional biases, as well as their relationships. Based on previous results, the degree of realism (Hernández-Serrano et al., 2020; Simon et al., 2020) and the social nature of the virtual environment (Bordnick et al., 2008) could be two interesting parameters to explore. Embodiment (which is defined as the total or partial representation of the user’s body in a virtual body) is also another parameter that could help the user to immerse himself in the environment (Kilteni et al., 2012) and increase the craving and the attentional bias.

Technological advances, and in particular ET-VR devices, offer new diagnostic and clinical tools for patients with alcohol use disorders. With these techniques, an evaluation of the level of craving combined with attentional biases could be integrated into the diagnostic assessment of the patients. In some cases, patients (and their relatives) may adopt a defensive position by communicating erroneous, incomplete or unsatisfactory information, especially if they are in denial about the disease. Using less explicit measures would therefore improve the assessment of the severity of the disorder. Similarly, it is expected that effective treatments reduce both craving and attentional biases. Their assessment through VR could then be used as a measure of treatment effectiveness (Martingano and Persky, 2021).

Several tasks based on classical paradigms were developed to increase the inhibitory control of eye movements to help patients reduce their attentional biases. Such training was shown to reduce the magnitude of attentional biases for alcohol-related cues (Schoenmakers et al., 2010; Wiers et al., 2011; Lee and Lee, 2015; Clerkin et al., 2016; Rinck et al., 2018). However, while many studies still consider attentional bias modification as a promising therapeutic tool, its clinical relevance has been recently questioned, especially due to the low internal reliability of the tasks (Christiansen et al., 2015b). ET-VR devices might be useful to make patients aware of their biases in psycho-educational session and could even be used to train disengagement strategies to decrease the attentional attractiveness of alcohol cues. Finally, ET-VR devices could be integrated into Virtual Reality-based Exposure Therapies (VRETs). This should enhance patient self-regulation and self-efficacy to promote sustainable behaviour changes and improve their management of cue reactivity (Segawa et al., 2020).