**Pavlovian-To-Instrumental Transfer and Alcohol Consumption in Young Male Social Drinkers: Behavioral, Neural and Polygenic Correlates**

**Maria Garbusow 1,\*,**†**, Stephan Nebe 2,3,4, Christian Sommer 2, Sören Kuitunen-Paul 5,6, Miriam Sebold 1,7, Daniel J. Schad 1,7, Eva Friedel 1,8, Ilya M. Veer 1, Hans-Ulrich Wittchen 5,9, Michael A. Rapp 7, Stephan Ripke 1,10,11, Henrik Walter 1, Quentin J. M. Huys 12, Florian Schlagenhauf 1,13, Michael N. Smolka 2,3 and Andreas Heinz <sup>1</sup>**


Received: 29 June 2019; Accepted: 6 August 2019; Published: 8 August 2019

**Abstract:** In animals and humans, behavior can be influenced by irrelevant stimuli, a phenomenon called Pavlovian-to-instrumental transfer (PIT). In subjects with substance use disorder, PIT is even enhanced with functional activation in the nucleus accumbens (NAcc) and amygdala. While we observed enhanced behavioral and neural PIT effects in alcohol-dependent subjects, we here aimed to determine whether behavioral PIT is enhanced in young men with high-risk compared to low-risk drinking and subsequently related functional activation in an a-priori region of interest encompassing the NAcc and amygdala and related to polygenic risk for alcohol consumption. A representative sample of 18-year old men (*n* = 1937) was contacted: 445 were screened, 209 assessed: resulting in 191 valid behavioral, 139 imaging and 157 genetic datasets. None of the subjects fulfilled criteria for alcohol dependence according to the Diagnostic and Statistical Manual of Mental Disorders-IV-TextRevision (DSM-IV-TR). We measured how instrumental responding for rewards was influenced by background Pavlovian conditioned stimuli predicting action-independent rewards and losses. Behavioral PIT was enhanced in high-compared to low-risk drinkers (*b* = 0.09, *SE* = 0.03, *z* = 2.7, *p* < 0.009). Across all subjects, we observed PIT-related neural blood oxygen level-dependent (BOLD) signal in the right amygdala (*t* = 3.25, *p*SVC = 0.04, *x* = 26, *y* = −6, *z* = −12), but not in NAcc. The strength of the behavioral PIT effect was positively correlated with polygenic risk for alcohol consumption (*rs* = 0.17, *p* = 0.032). We conclude that behavioral PIT and polygenic risk for alcohol consumption might be a

biomarker for a subclinical phenotype of risky alcohol consumption, even if no drug-related stimulus is present. The association between behavioral PIT effects and the amygdala might point to habitual processes related to out PIT task. In non-dependent young social drinkers, the amygdala rather than the NAcc is activated during PIT; possible different involvement in association with disease trajectory should be investigated in future studies.

**Keywords:** Pavlovian-to-instrumental transfer; amygdala; alcohol; polygenic risk; high risk drinkers

#### **1. Introduction**

Problematic alcohol drinking patterns like bingeing or heavy drinking during adolescence and early adulthood are associated with severe psychological, social and health problems [1]. Therefore, elucidating mechanisms that underlie high-risk drinking in young adulthood is important. Here, we assess biological factors in relation to a behavioral phenomenon that has been associated with chronic alcohol consumption theoretically [2–4] and empirically [5,6]. Specifically, we focus on behavioral effects of Pavlovian-to-instrumental transfer and at risk alcohol consumption in young male social drinkers, neural correlations and the association to polygenic risk for alcohol consumption.

Alcohol intake has been shown to be promoted by positive and negative contexts [7,8]. One mechanism implicated in the influence of contexts on ongoing behavior is Pavlovian-to-instrumental transfer (PIT). In general PIT, appetitive Pavlovian cues promote instrumental responses while aversive Pavlovian cues reduce such responses or even promote withdrawal independent of reward types [9]. In specific PIT, Pavlovian cues promote instrumental behavior associated specifically with the same outcome [10]. In animal models of addiction, drug exposure increases general and specific behavioral PIT effects [11,12] and enhanced food-related behavioral PIT was predictive for subsequent stronger cue-induced reinstatement of alcohol seeking [13]. We have recently reported increased nondrug-related behavioral PIT in detoxified alcohol-dependent patients compared to age-and gender-matched social drinkers using monetary cues [5]. In this study, we ask whether similar differences in PIT are measurable in an independent and much younger cohort of male high-versus low-risk social drinkers. Previous studies have examined alcohol-specific behavioral PIT effects in social drinkers but did not assess the association between behavioral PIT effects and individual drinking patterns [14], nor did they find an association with subclinical alcohol dependence [15,16] or neural PIT correlates using electroencephalography (EEG) [16]. In contrast to these studies, we investigate nondrug-related PIT effects in young high-versus low-risk [17] social drinkers on a behavioral and neural level using functional magnetic resonance imaging (fMRI).

On a neural level, animal studies showed that the amygdala is a core region associated with behavioral PIT [10,18–20]. Moreover, the strength of behavioral PIT is positively correlated with dopaminergic neurotransmission in the ventral striatum [21], which in turn is known to be modulated by alcohol intake [22–24]. In humans, both the nucleus accumbens (NAcc) and amygdala are activated during PIT [25–27], and amygdala activation by alcohol cues has been positively correlated with craving in alcohol-dependent patients during an alcohol-approach bias task [28]. Interestingly, PIT-related activation of the NAcc, but not the amygdala, predicted relapse after detoxification in alcohol-dependent patients [5].

Many genes can be involved in phenotypes such as alcohol use with respectively small effect sizes [29]. Therefore, we used a polygenic risk approach to investigate the genetic influence on alcohol consumption and behavioral PIT in our sample. It has been shown that higher polygenic predisposition for alcohol problems predicts earlier initial alcohol consumption and early heavy drinking patterns, as well as more alcohol-related problems in independent samples [30–32]. We therefore aimed to investigate how a polygenic risk score (PRS) for alcohol consumption derived from an independent

large genome-wide association study [33] is associated with alcohol consumption and behavioral PIT in our sample.

As we previously observed stronger nondrug-related behavioral PIT in alcohol-dependent patients compared to controls as well as a stronger PIT-related NAcc activation predicting relapse [5], we wanted to assess whether there are comparable differences in nondrug-related behavioral PIT between the two groups of young male high-versus low-risk drinkers. Therefore, we examined a non-clinical sample of young males and hypothesized (1) stronger nondrug-related behavioral PIT effects in high-compared to low-risk drinkers [17]; (2) PIT-related blood oxygen level-dependent (BOLD) activity in an a-priori region of interest (ROI) encompassing amygdala and NAcc; and (3) a positive association between alcohol-related polygenic risk [33] and both the strength of nondrug-related behavioral PIT and alcohol consumption in our sample.

#### **2. Materials and Methods**

#### *2.1. Participants and Procedure*

1974 males were randomly drawn from local registration offices in two sites (Berlin & Dresden, Germany [34]) shortly after their 18th birthday representing their local legal adult age. We screened 445 respondents via telephone. Exclusion criteria were left-handedness, history of major neurological or psychiatric disorders (except for nicotine dependence and alcohol abuse), current alcohol abstinence and MRI-specific contraindications. In total, 209 subjects were included and tested. After quality control, 191 behavioral, 139 imaging and 157 genetic datasets could be analyzed (see Figure 1). Subjects were descriptively comparable to similar cohorts drawn from the German general population (see Supplementary Table S1).

All participants were assessed with the Composite International Diagnostic Interview (CIDI) [35,36] according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR) [37] and completed a neuropsychological test battery. On a second appointment (mean = 8.5 (SD = 16.2) days later), participants performed a task battery during a functional magnetic resonance imaging (fMRI) scan. The experimental procedure comprised a two-step Markov decision making task [38,39] and the PIT task with nondrug-and drug-related contexts [40]. Blood samples for genetic analyses were taken at first (Berlin) or second (Dresden, after MRI scan) appointment. The study procedures (clinical trials identifier: NCT01744834) adhered to the Declaration of Helsinki and were approved by local ethics committees of Charité Universitätsmedizin Berlin (EA/1/157/11) and Technische Universität Dresden (EK 227062011). All participants gave written informed consent prior to participation.

#### *2.2. Experimental Design*

The PIT task consisted of four parts:

Instrumental training. Participants collected shells by repeated button presses receiving probabilistic feedback (see Figure 2A). To control for instrumental performance, participants trained until they reached a criterion of 80% correct choices over 16 trials (for a minimum of 60 or a maximum of 120 trials).

Pavlovian conditioning. Trials began with presenting for3sa compound stimulus consisting of fractal-like pictures and pure tones (conditioned stimulus, CS); followed by a 3 s delay, and finally an unconditioned stimulus (US: picture of a coin) for 3 s (see Figure 2B). Participants were instructed to memorize the pairings. All participants completed 80 trials.

**Figure 1.** Recruiting and exclusion procedure leading to the final behavioral, genetic and imaging datasets. MRI: magnetic resonance imaging; PIT: Pavlovian-to-instrumental transfer.

Pavlovian-to-instrumental transfer. Participants performed the instrumental task now with CS tiling the background (see Figure 2C). Note that the instrumental task was independent of the value of the background stimulus. No outcomes were presented, but participants were instructed that their choices still counted towards the final monetary outcome. The pairings of CS in background und shell in foreground were counterbalanced with each combination showing three times, resulting in 90 trials.

Forced choice task. Finally, participants chose one of two CSs (Figure 2D). All possible CS pairings were presented three times in randomized order.

**Figure 2.** Pavlovian-to-instrumental Transfer (PIT) task. (**A**): Instrumental training: collecting a 'good' shell was rewarded in 80% while not collecting a 'good' shell punished in 80%. The opposite reinforcement contingencies applied to 'bad' shells. Red arrows indicate the five or more button presses required to approach and collect the presented shell. By trial and error, subjects learned to collect or not to collect three out of six shells. (**B**): Pavlovian conditioning: subjects passively viewed a conditioned stimulus (CS), which was deterministically followed by an unconditioned stimulus (US). As CS, a compound of a tone and five fractal-like visual stimulus was used. USs were pictures of a coin (−2€, −1€, 0€, +1€, +2€). (**C**): Transfer: subjects were asked for the instrumental response, while the background was tiled with the CS. Trials with drink-related background stimuli are not displayed. (**D**): Query trials: Subjects were asked to choose the better (i.e., that was associated with the highest reward or lowest punishment during Pavlovian conditioning) between sequentially presented CSs.

#### *2.3. Self-Reported Questionnaires*

We used self-reported measures for sample description measuring alcohol dependence severity (ADS) [41], alcohol craving (Obsessive Compulsive Drinking Scale, OCDS) [42] and nicotine dependence severity (Fagerström Test for Nicotine Dependence, FTND) [43].

#### *2.4. Measures of Alcohol Consumption*

In accordance with previous analyses of nondrug-related PIT group differences between alcohol-dependent patients and matched social drinkers [5], we used the World Health Organization (WHO) definition for risk of acute alcohol-related problems [17] based on average ethanol consumption on a drinking occasion within the last year. Accordingly, subjects qualified as low-risk drinkers (≤60 g of alcohol on a single occasion) or high-risk drinkers (>60 g), respectively. To further characterize participants' drinking behavior and how this relates to polygenic risk, we calculated a sum score of drinking variables (henceforth referred to as drink score) from the z-scaled CIDI items with higher values indicating higher or more risky alcohol consumption [44] (see supplementary materials for calculations "measures of alcohol consumption").

#### *2.5. MRI Data Acquisition*

Functional imaging was performed on a Siemens Trio 3 Tesla MRI scanner with an Echo Planar Imaging (EPI) sequences (repetition time, 2410 ms; echo time, 25 ms; flip angle, 80◦; field of view, 192 <sup>×</sup> 192 mm2; voxel size, 3 <sup>×</sup> 3 <sup>×</sup> 2 mm3, 1 mm gap; 480 volumes) comprising 42 slices acquired in descending order and rotated approximately −25◦ to the bicommissural plane. For coregistration and normalization during pre-processing, a three-dimensional magnetization-prepared rapid gradient echo image was acquired (repetition time, 1900 ms; echo time, 2.52 ms; flip angle, 9◦; field of view, <sup>256</sup> <sup>×</sup> 256 mm2; 192 sagittal slices; voxel size, 1 <sup>×</sup> <sup>1</sup> <sup>×</sup> 1 mm3). A field map was recorded to account for individual homogeneity differences of the magnetic field.

The PIT task was programmed using Matlab with the Psychophysics Toolbox Version 3 (PTB-3) extension [45]. Responses during PIT were made using a current-design MRI-compatible response box with the right index finger.

#### *2.6. Polygenic Risk Score*

To genotype our sample, DNA was extracted semi-automatically with a Chemagen Magnetic Separation Module (Perkin Elmer) from whole blood drawn in EDTA tubes before fMRI assessment. All samples were genotyped with the Illumina Infinium Psych Array Bead Chip [46]. Content for the PsychArray includes 265,000 proven tag SNPs found on the HumanCoreBeadChip, 245,000 markers from the Human Exome Bead Chip and 50,000 additional markers.

For calculating the polygenic risk score (PRS), we used a standard approach [47]. Our training data set derived from a large genome-wide association study (GWAS) investigating the genetic basis of alcohol consumption in *n* > 105,000 healthy social drinkers [33]. To calculate a polygenic risk score for each individual in our independent sample we summed up the number of alleles for each single nucleotide polymorphism (SNP) weighted by the effect size (association between each SNP and alcohol consumption) drawn from GWAS from the training data set. The score was computed at different *p*-value thresholds (*p* = 1, *p*= 0.5, *p* = 0.2, *p* = 0.1, *p* = 0.05, *p* = 0.01) representing the composite additive effect of all SNPs (*p* = 1, *n* = 100,000 SNPs) or the number of SNPs above the respective threshold. This gives the SNPs with higher significance automatically more weight than SNPs with lower significance.

#### *2.7. Statistical Analysis*

Data were analyzed in Matlab 2011a [48] and the R System for Statistical Computing Version 3.3.3 [49]. Functional magnetic resonance imaging (fMRI) data were analyzed using Statistical Parametric Mapping (SPM 12) software package [50]. All analyses refer to the transfer part of the PIT task (Figure 2C).

#### *2.8. Behavioral Analysis*

We conducted a generalized linear mixed-effects model implemented in the lme4 package (version 1.1-12). In order to assess the individual contribution of Pavlovian values on behavior, we built a Poisson distributed model where the number of button presses in each trial was predicted by the value of the background CS (−2, −1, 0, +1, +2; linear effect) and the instrumental condition (collect/not collect; coded as +0.5/−0.5). The within-subject factors (intercept, main effect of CS value, instrumental condition, and their interaction) were taken as random effects across subjects. Instrumental stimuli (shells) and Pavlovian CSs were taken as separate crossed random effects with varying intercepts in order to control for potential item effects. We included group (high-versus low-risk drinkers; coded as +0.5/−0.5) as between-subject factor to this model, performing two-tailed statistical tests on the a-priori hypothesis that behavioral PIT effects are stronger in high-compared to low-risk drinkers. Furthermore, we extracted individual regression slopes from the original generalized linear mixed-effects model as a measure of individual strength of behavioral PIT for further testing the association between the strength of the behavioral PIT effect and polygenic risk for alcohol consumption.

#### *2.9. Imaging Analysis*

Preprocessing. For preprocessing information see supplementary materials.

First-level analysis. The influence of Pavlovian stimulus values on instrumental responses (PIT effect) was measured by constructing a linear contrast, which weighted the parametric modulator of each condition (i.e., trial-by-trial number of button presses) by their associated Pavlovian values (−2, −1, 0, +1, +2) [5], i.e., the neural PIT effect was modeled by number of button presses times value of background stimulus (onset: appearance of shell in foreground). To account for variance caused by motor responses, button presses for all trials together were modeled with a regressor of no interest. Regressors were then convolved with the canonical hemodynamic response function. The six realignment parameters and their first derivatives were included as regressors of no interest. For a measure of the neural PIT effect a linear contrast was constructed, which weighted the parametric modulators for each condition by the related Pavlovian background value. The neural CS value effect was measured with a similar linear contrast on the CS event regressors.

Second-level analysis. Linear contrast images for neural PIT and neural CS contrasts were taken to the second level. To test for the neural PIT effect, we conducted a one-sample *t*-test. Study site was included as covariate. Using the wake Forest University (WFU) Pick Atlas software [51], we computed one ROI for a small volume correction (SVC) approach including both the bilateral NAcc and bilateral amygdala to avoid multiple testing. Next, we extracted individual mean beta values of the observed neural PIT effect to test the association of neural and individual behavioral PIT effect. We expected a positive association, yet conducted a two-tailed test.

#### *2.10. Polygenic Analyses*

We computed a PRS (see methods above), to verify genetic risk for alcohol consumption computed at threshold *p* = 1, thus including all genetic signal. To present the full picture, we also report results at other *p*-levels. Spearman's correlation coefficient was used to compute the respective association between PRS and (i) the continuous composite drink score, and (ii) behavioral PIT slope extracted from the glmer model described above. We expected a positive association between these measures and tested two-tailed. While the first analysis (i) provides evidence of whether the PRS is associated with drinking in our sample (replications see [30,31]), the second analysis (ii) explores a direct association between PRS and behavioral PIT (*p*-values for descriptive reasons only).

#### **3. Results**

#### *3.1. Sample Characteristics by Drinking Group*

Supplementary Table S2 summarizes sample characteristics comparing high-risk drinkers (*n* = 94) to low-risk drinkers (*n* = 97) according to WHO stratification [17]. Pure alcohol consumed in life in kg, ADS and OCDS are for clinical description of severity of alcohol use problems. According to that, high-risk drinkers reported higher lifetime alcohol intake, stronger craving in the past seven days and more problems associated with alcohol dependence. Groups did not differ significantly in terms of smoking severity, age, socio economic status and verbal intelligence.

#### *3.2. Behavioral Results*

Behavioral PIT effects were significantly stronger in high-compared to low-risk drinkers (for PIT effect in whole sample see supplementary material Figure S1). Specifically, the regression analyses showed an interaction effect between Pavlovian background and group on instrumental response rate (*b* = 0.09, *SE* = 0.03, *z* = 2.7, *p* < 0.009, *n* = 191, two-tailed; see Figure 3 and Supplementary Table S3) in the way that with higher value of the background stimulus the instrumental response rate increases. Crucially, this was not due to smoking severity (see Supplementary Table S4), or differences in instrumental performance (*p* = 0.54, see Supplementary Table S3).

**Figure 3.** Behavioral PIT effect in low-versus high-risk drinkers (*n* = 191). Number of button presses for each Pavlovian background condition. The behavioral PIT effect is stronger in high-risk drinkers (as indicated by a steeper group regression slope).

#### *3.3. Imaging Results*

The ROI analysis (encompassing bilateral amygdalae and NAcc) for the whole sample revealed a significant PIT-related activation in the right amygdala (*t*(137) = 3.25, *p*SVC = 0.04, *x* = 26, *y* = −6, *z* = −12, *k* = 29, *n* = 139, see Figure 4A), which could not be explained by a pure CS effect (see Supplementary Figure S2). Extracted mean beta-values within the right amygdala showed a positive association with the behavioral PIT effect (*b* = 0.07, *SE* = 0.014, *z* = 4.7, *p* < 0.001, two-tailed, *n* = 139). High-versus low-risk drinkers according to the WHO did not differ significantly in neural activation during PIT. Within our single ROI, there was no significant activation in the NAcc. For exploratory whole brain analyses at *puncorr* < 0.001 and *k* = 10 see Supplementary Table S5.

**Figure 4.** (**A**). Neural PIT effect in the right amygdala for the whole group (*n* = 139). For illustrational purposes, this effect was masked for the bilateral amygdala (region of interest (ROI) derived from wake Forest University (WFU) Pick Atlas). (**B**). The PIT-related activation in the right amygdala positively correlated with the behavioral PIT effect.

#### *3.4. Polygenic Risk in Association with Behavioral PIT*

We found a significant positive correlation between the PRS and the composite drink score in our sample (*rs* = 0.17, *p* = 0.032, *n* = 157, two-tailed see Figure 5A). Figure 5B illustrates this association between polygenic risk and drink score in our sample using PRS computed at different thresholds ranging from *p* = 0.01 to *p* = 1. Furthermore, there was a significant positive correlation between the strength of the behavioral PIT effect and the PRS (*rs* = 0.17, *p* = 0.032, *n* = 157, two-tailed, see Figure 5C). Figure 5D illustrates this association between polygenic risk and behavioral PIT using PRS computed at different thresholds ranging from *p* = 0.01 to *p* = 1. For a multi-level approach using multi-modal information (behavioral PIT, neural PIT effect and PRS), see Supplementary Table S6.

**Figure 5.** Polygenic risk score (PRS) for alcohol consumption in association with alcohol intake and PIT in our sample (*n* = 157). (**A**): Association between PRS and alcohol intake as measured by the drink score in our sample. (**B**): Explained variance of the association between PRS and drink score as indicated by rs <sup>2</sup> for each threshold. Values at each bar represent the p-values, tested two-tailed. (**C**): Association between PRS and behavioral PIT effect slope. (**D**): Explained variance of the association between PRS and behavioral PIT slope as indicated by rs <sup>2</sup> for each threshold. Values at each bar represent the *p*-values, tested two-tailed.

#### **4. Discussion**

We aimed to investigate the nondrug-related behavioral PIT effect in a cohort of young male high-versus low-risk [17] drinkers and the neural correlate of this PIT effect. We further explored the association of behavioral PIT with polygenic risk for alcohol consumption. Our main finding of enhanced behavioral PIT in high-risk drinkers suggests that strong effects of Pavlovian cues on instrumental behavior could be a core behavioral signature of risky alcohol consumption. We further observed PIT-related amygdala activation and a positive association between the strength of the behavioral PIT effect and polygenic risk related to alcohol consumption. A similarly assessed behavioral PIT effect was previously enhanced in alcohol-dependent patients compared to matched controls [5]. However, whether this is a vulnerability marker for developing alcohol use disorder or a consequence of substance exposure requests investigation in future longitudinal studies.

On a neurobiological level, we observed PIT-related activation in the right amygdala, which is in line with animal [52] and human studies [25,26,53], supporting previous reports of a key role for the amygdala in PIT [10]. Lesions of the basolateral amygdala abolished the selective excitatory effects of reward-related Pavlovian stimuli, while lesions of the central amygdala abolish general motivational effects of such cues [10]. A human study by Prevost et al. [26] confirmed the dissociation within the amygdala and general and specific PIT. Although our task has repeatedly identified Pavlovian modulation of instrumental responses [5,6,9], we cannot distinguish between general and specific PIT effects, as the reward used for Pavlovian and instrumental conditioning was both monetary, preventing conclusions about the specificity of the Pavlovian influence on instrumental behavior. Moreover, the amygdala was involved in cue-elicited habitual rather than goal-directed reward seeking in healthy humans [54]. Thus, our results on PIT effects in the amygdala might point to habitual processes that are related to enhanced PIT effects in our task.

Interestingly, PIT-related NAcc activity was related to poor treatment outcome in alcohol-dependent patients [5] supporting the known role of chronic alcohol intake on striatal dopaminergic neurotransmission [55], while in this sample of young at-risk drinkers, we could not observe PIT-related NAcc activity. This may suggest that the neurobiology by which CSs come to guide behavior is different in at risk alcohol consumption versus alcohol-dependent patients. The amygdala may be involved in cue-induced modulation of instrumental behavior among young adults and the NAcc might be associated with the nondrug-related PIT effect in alcohol dependence only. These biological differences may be associated with the shift from impulsive to habitual drug intake during the development of substance use disorders [3,56–58]. Further studies should assess whether these PIT activation differences are correlated with loss of control over alcohol intake.

Moreover, we observed alcohol-related polygenic risk to be associated with higher alcohol consumption in our sample. This effect is in line with numerous studies that emphasize a polygenic risk for risky alcohol consumption [30,31]. In addition, our study reveals that this polygenic risk is also related to a possible underlying mechanism associated with risky alcohol consumption, namely PIT. Thus, the strength of contextual stimuli in influencing ongoing instrumental behavior might be modulated by an underlying genotype for alcohol consumption, strengthening PIT to be a relevant mechanism to understand alcohol intake.

In terms of clinical implications, Ostafin et al. [59] showed that prevention programs boosting the explicit motivation to reduce alcohol consumption are only effective in hazardous drinkers with low automatic approach tendencies towards alcohol, and can even result in increased alcohol consumption if they fail to address implicit motivational processes [60]. Our data suggest that high-risk drinkers may be more susceptible to cue-triggered processes; this group may particularly profit from interventions focusing on implicit motivational processes to reduce hazardous alcohol consumption [61].

Limitations of our study include potential selection bias during recruitment: due to ethical guidelines, we were required to state in the invitation letter that we wish to recruit for a study on alcohol consumption. Subjects with high alcohol consumption might have been more reluctant to participate. Moreover, we excluded alcohol-dependent subjects, as we aimed to investigate PIT in social drinkers only. These two reasons might explain why most of our sample reported low life-time alcohol consumption (kg pure alcohol), a comparably small number of dependence problems as measured by the ADS as well as little numbers of DSM-IV-TR alcohol abuse diagnoses. Moreover, we focused on high risk for acute alcohol-related problems as defined by the WHO. Therefore, our data cannot make conclusions about chronic alcohol problems. In line with DSM 5, a dimensional approach to AUD has been applied, which needs to also identify risky variants at the lower end which might convert into high risk drinkers. Here, the reported study design is cross-sectional, which limits conclusions about

how PIT effects change over the course of alcohol consumption and it limits mechanistic statements and explanations. Moreover, we included male subjects exclusively, thus limiting the generalization to female alcohol consumers. However, female drinking has been reported to be manifested in a different way that lack some of the very public and overt ways that male drinking presents [62]. We only assessed males to avoid loss of power due to potential gender differences; findings need to be replicated for women. Furthermore, as expected, especially the genetic effects sizes are rather small, which limits the predictive power on an individual level. Finally, we cannot draw conclusions about other substance-related disorders, as we focused our analyses on alcohol only.

In summary, we observed enhanced behavioral PIT effects in young high-risk drinkers, which were associated with functional activation in the right amygdala and correlated with an alcohol-related polygenic risk. Together with previous findings on behavioral PIT effects in alcohol-dependent patients [5], these data suggest that stronger behavioral PIT effects are a trait marker for high-risk alcohol use. How this is related to the risk of developing alcohol dependence should be further explored in longitudinal studies. Although the behavioral effects reported here were similar to the previously reported effects in patients, the neural correlates involved the amygdala rather than the NAcc [5], suggesting a differential involvement of these structures at different time points in the disease trajectory. Given the well-documented effects of alcohol on striatal dopaminergic neurotransmission [55], future studies should explore alcohol effects on striatal versus amygdala function and their malleability in longitudinal settings.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2077-0383/8/8/1188/s1, Figure S1: PIT effect in the whole sample (*n* = 191). Number of button presses are displayed relative to the zero (0€) condition. Grey line reflects the group regression slope, Figure S2: Pure CS effect during the PIT task. This effect was significant for the whole imaging group of *n* = 139 subjects in the right NAcc only (pSVC = 0.041). Table S1: Comparing our study cohort to a reference population, Table S2: Sample characteristics for the two groups of low-and high-risk drinkers, Table S3: Regression coefficients indicating the influence of Pavlovian background conditions (−2€, −1€, 0€, +1€, +2€), instrumental conditions (collect versus not-collect) and drinking group, Table S4: Regression coefficients indicating the influence of Pavlovian background conditions (−2€, −1€, 0€, +1€, +2€), instrumental conditions (collect versus not-collect) and drinking group controlled for smoking severity, Table S5: Exploratory whole brain analysis of PIT-related activations, Table S6: Generalized linear mixed-effects model for predicting number of button presses as a function of instrumental task (collect versus not collect) and Pavlovian background stimulus value (−2€, −1€, 0€, +1€, +2€). Moreover, this analysis includes additional predictors, namely alcohol risk group, neural PIT effect and polygenic risk (*n* = 115).

**Author Contributions:** Conceptualization, M.G., E.F., I.M.V., H.-U.W., M.A.R., S.R., H.W., Q.J.M.H., F.S., M.N.S. and A.H.; Data curation, M.G., S.N., C.S., S.K.-P., M.S. and I.M.V.; Formal analysis, M.G., D.J.S. and F.S.; Funding acquisition, H.-U.W., M.A.R., Q.J.M.H., M.N.S. and A.H.; Investigation, M.G., S.N., S.K.-P. and M.S.; Methodology, M.G., S.K.-P., D.J.S., I.M.V., H.-U.W., M.A.R., S.R., Q.J.M.H. and F.S.; Project administration, M.G., S.N., S.K.-P. and E.F.; Resources, H.-U.W., M.A.R., H.W., M.N.S. and A.H.; Software, Q.J.M.H.; Supervision, H.-U.W., M.A.R., H.W., F.S., M.N.S. and A.H.; Validation, S.R., Q.J.M.H., F.S. and A.H.; Visualization, M.G.; Writing—original draft, M.G. and A.H.; Writing—review & editing, M.G., S.N., C.S., S.K.-P., M.S., D.J.S., E.F., I.M.V., H.-U.W., M.A.R., S.R., H.W., Q.J.M.H., F.S., M.N.S. and A.H.

**Funding:** This work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, FOR 1617 grants FR 3572/1-1, HE 2597/13-1, HE 2597/13-2, HE 2597/15-1, HE 2597/15-2, RA 1047/2-1, SCHL 1969/2-2/4-1, SM 80/7-1, SM 80/7-2, WA 1539/7-1, WI 709/10-1, WI 709/10-2, as well as DFG grants SFB 940/1 and SFB 940/2). Eva Friedel is a participant in the BIH Charité Clinician Scientist Program funded by the Charité—Universitätsmedizin and the Berlin Institute of Health.

**Acknowledgments:** MR-imaging for this study was performed at the Berlin Center for Advanced Neuroimaging (BCAN) and Neuroimaging Centre (NIC) at TU Dresden. We acknowledge support from the German Research Foundation (DFG) and the Open Access Publication Fund of Charité—Universitätsmedizin Berlin.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

1. Maggs, J.L.; Schulenberg, J.E. Initiation and course of alcohol consumption among adolescents and young adults. In *Recent Developments in Alcoholism: An O*ffi*cial Publication of the American Medical Society on Alcoholism, the Research Society on Alcoholism, and the National Council on Alcoholism*; National Institute on Alcohol Abuse and Alcoholism: Bethesda, MD, USA, 2005; Volume 17, pp. 29–47.


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Cue-Elicited Anxiety and Alcohol Craving as Indicators of the Validity of ALCO-VR Software: A Virtual Reality Study**

**Alexandra Ghi¸tă 1, Olga Hernández-Serrano 2, Yolanda Fernández-Ruiz 1, Miquel Monras 3, Lluisa Ortega 3, Silvia Mondon 3, Lidia Teixidor 3, Antoni Gual 3, Bruno Porras-García 1, Marta Ferrer-García <sup>1</sup> and José Gutiérrez-Maldonado 1,\***


Received: 17 July 2019; Accepted: 30 July 2019; Published: 2 August 2019

**Abstract:** Background: This study is part of a larger project aiming to develop a virtual reality (VR) software to be implemented as a clinical tool for patients diagnosed with alcohol use disorder (AUD). The study is based on previous research in which we identified factors that elicit craving for alcohol in a sample of AUD patients, and which led to the development of a virtual reality software to be used in cue exposure treatments of alcohol use disorder (ALCO-VR). The main objective of this study was to test the effectiveness of ALCO-VR to elicit cue-induced craving and anxiety responses among social drinkers (SD) and AUD patients. Our secondary objective was to explore which responses (cue-induced craving or anxiety) can best differentiate between AUD patients and the SD group. Method: Twenty-seven individuals (13 AUD patients and 14 SD) participated in this study after giving written informed consent. Their anxiety and alcohol craving levels were measured by different instruments at different stages of the procedure. The VR equipment consisted of Oculus Rift technology, and the software consisted of the ALCO-VR platform. Results: Our data indicate that the ALCO-VR software can elicit responses of anxiety and alcohol craving, especially in the group of AUD patients. The cue-induced anxiety response differentiated AUD patients and the SD group better than the cue-induced craving response. Conclusions: The general interest in applying new technologies to the assessment and treatment of mental health disorders has led to the development of immersive real-life simulations based on the advantages of VR technology. Our study concluded that the ALCO-VR software can elicit anxiety and craving responses and that cue-induced anxiety responses can distinguish between AUD and SD groups better than cue-induced craving. The data on craving and anxiety were assessed consistently by different instruments. In addition, we consider that ALCO-VR is able to ecologically assess cue-induced anxiety and alcohol craving levels during exposure to VR alcohol-related environments.

**Keywords:** ALCO-VR; virtual reality; cue-exposure; alcohol use disorder; alcohol craving; anxiety; social drinkers

#### **1. Introduction**

Alcohol misuse is considered to be a serious condition with important negative consequences at a personal and societal level [1–4]. Numerous studies have emphasized that excessive alcohol use may establish a solid basis for general heavy drinking patterns, binge-drinking episodes, and further development of alcohol use disorder (AUD) in adulthood [5–8]. In AUD, a substantial number of patients respond to treatment by accomplishing individual goals such as controlled drinking or total abstinence [9]. However, many individuals struggle with maintaining long-term abstinence; despite treatment, they experience relapses, especially in the first three months [10,11]. AUD is a complex disorder and many factors contribute to its development and maintenance. The current study focuses on the interplay between anxiety and alcohol craving as factors interfering in long-term abstinence.

Anxiety, ranging from momentary distress to a psychopathological diagnosis, is common in the majority of patients with AUD, and is considered a facilitating factor in the onset of alcohol consumption [12]. Anxiety is involved not only in the pre-phase of drinking behaviors, but is also a predominant symptom experienced during alcohol withdrawal in patients with AUD [13]. A growing and solid body of research has focused on understanding the phenomena underlying anxiety disorders and AUD, since individuals with dual diagnoses are more vulnerable in terms of recovery and maintaining abstinence [14–16]. A network modeling analysis revealed that anxiety-related states such as social anxiety and stress are directly engaged in craving elicitation in patients diagnosed with AUD [14]; thus, a strong causal relationship was established between anxiety and craving for alcohol, which further precipitates drinking behaviors and, implicitly, relapse. This new model explains the critical involvement of this causative relationship between anxiety and alcohol craving in patients diagnosed with AUD.

Alcohol craving, described in the literature as a "pathological appetite", is an unbearable desire to use alcohol [17] and has a crucial role in alcohol misuse behaviors [18]. Explained by the classical conditioning theory [19], alcohol craving is acquired through repetitive drinking behaviors accompanied by positive emotional valence [20]. Over the past two decades, laboratory research [21,22] and clinical research [23–25] have emphasized the strong relationship between proximity of alcohol-related stimuli that trigger alcohol craving. This context dependency theory indicates that alcohol-related stimuli gain incentive salience over time [26,27] and therefore, cues and contexts related to alcohol consumption emerge as high-risk stimuli for individuals who misuse alcohol [28]. A common method for exploring alcohol craving responses is based on the cue-exposure paradigm, which involves in vivo alcohol cue presentation, or exposure to photographic alcohol-related stimuli in non-realistic experimental or clinical settings [29]. Over the past years, an increasing interest in new technologies like virtual reality (VR) has led to the development of more ecologically valid methods to explore alcohol craving [30]. VR may add effectiveness to the classical cue-exposure paradigm by including multiple inputs like visual, auditory, olfactory or tactile stimuli, thus creating an immersive experience based on naturalistic daily-life contexts [31]. This three-dimensional (3D) system allows a high degree of interaction, which further increases the individual's level of momentary presence within the VR environment [32]. These are fundamental variables for developing ecological momentary assessment (EMA) instruments, particularly for its use in clinical settings, as most assessment instruments in clinical psychology rely on self-reported scales [33,34]. In drug cue reactivity, VR has been included in many protocols, primarily studying craving for alcohol [35,36], tobacco [37–39], methamphetamine [40] or cocaine [41]. In AUD, VR has also been implemented in the treatment of patients with the final aim of reducing craving levels and, implicitly, of preventing further relapses. A comprehensive review of the existing studies indicated that all studies focused on the applications of VR as a therapy tool showed consistent results in terms of craving reduction, thus favoring the inclusion of this technology in the treatment of substance use disorders [30].

The present study is based on the results of a previous study, in which we identified triggering factors for alcohol craving, aiming to develop significant VR alcohol-related environments. Our main objective was to validate the ALCO-VR platform (a virtual reality software to be used in cue exposure treatments of alcohol use disorder) and to test its effectiveness in terms of the elicitation of cue-induced alcohol craving and anxiety responses in a sample of social drinkers (SD) and patients diagnosed with AUD. A secondary objective of our study was to explore which cue-elicited response (alcohol craving or anxiety) differentiates better between AUD patients and SD. We expected to obtain significant differences between the neutral VR condition and the VR alcohol-related environments regarding cue-induced anxiety and alcohol craving responses in the two groups, AUD patients and SD. Similarly, we expected to find statistically significant differences between AUD and SD groups in terms of their alcohol craving and anxiety responses. Based on an earlier study in bulimia nervosa [42], we expected cue-induced anxiety responses to differentiate between AUD patients and SD better than cue-induced alcohol craving responses.

#### **2. Method**

#### *2.1. Participants*

Twenty-seven individuals participated in this study, 13 patients from the Addictive Behaviors Unit, Hospital Clinic of Barcelona and 14 SD, students at the University of Barcelona. Ethical approval was obtained from the Ethics Committees at the University of Barcelona and Hospital Clinic of Barcelona. All patients and students participated in this study after providing written informed consent. The clinical sample consisted of 13 outpatients, 8 men and 5 women (Mage = 48, SD = 4.8), diagnosed with AUD according to the *Diagnostic and Statistical Manual of Mental Disorders* (5th ed.) [43]. Patients presented comorbid diagnoses of borderline personality disorder, anxiety, and attention deficit hyperactivity disorder (ADHD). Their pharmacotherapy included anxiolytic, antidepressant, disulfiram, and antipsychotic medication. Seven patients reported occasional tobacco, cannabis, and cocaine use in the month prior to the experiment, but none reported alcohol consumption in the last month as they were under treatment for AUD at Hospital Clinic of Barcelona. The mean abstinence period was 68 days, ranging from one month to one year. Self-reports of substance use and abstinence data were supported by results of urine analyses performed in all patients. The inclusion criteria were having an AUD diagnosis and receiving outpatient treatment for AUD. Individuals with comorbid disorders like personality disorders or anxiety disorders and occasional use of illicit drugs (e.g., cannabis or cocaine) or tobacco were included, but those with severe comorbid psychopathology (e.g., psychosis or dementia), severe cognitive impairment that might interfere with the task completion, or anti-craving medication (e.g., naltrexone) were excluded. Pregnant women were also excluded.

The control group consisted of 14 SD, 2 men and 12 women (Mage = 23, SD = 5.6), all students at the University of Barcelona. Among this group, there were self-reports of one diagnosis of depression and one diagnosis of ADHD. One student reported using anxiolytic medication, and one student antidepressant medication. Tobacco and cannabis use were self-reported by six students in the last month prior to the experiment. SD also reported their monthly consumed standard drink units (SDU). A Spanish SDU is a single consumption of 10 g of ethanol (the standard quantity of wine or beer and half the standard quantity of liquor) [44]. Students consumed approximately 9 SDU (M = 9.2, SD = 9.9) per month, ranging from 3 to 24 SDU/month. Abstinence data (Mdays = 15, SD = 18) in this group were based on self-reports from the students. The inclusion criterion was a social drinking pattern (i.e., individuals who use alcohol on a social basis and are not identified as having a problematic pattern of alcohol use). As in the clinical group, comorbid disorders (personality disorders or anxiety) and occasional use of illicit drugs (e.g., cannabis or cocaine) or tobacco were not considered exclusion criteria. However, any participants with severe comorbid psychopathology or severe cognitive impairment were excluded.

#### *2.2. Measures*

Alcohol consumption was assessed with the Alcohol Use Disorder Identification Test (AUDIT) [45]. The Spanish version of AUDIT [46] is a 10-item scale assessing alcohol misuse and severity of alcohol-related problems. Responses are scored from 0 to 4 with a maximum score of 40.

Alcohol craving was assessed with the Multidimensional Alcohol Craving Scale (MACS) [47]. The MACS is a Spanish self-reported scale aiming to assess the "intensity of alcohol craving experienced by the participant in the previous week". Scores on each item range from 1 ("Strongly disagree") to 5 ("Strongly agree"). There are three MACS outcome scores: "desire to drink," "behavioral

disinhibition," and the total score, which were each graded as nonexistent, mild, moderate or intense. Moreover, a modified version of MACS (MACS-VR) was introduced in the experimental procedure to explore alcohol craving immediately after VR exposure to alcohol-related contexts and cues. The items, scores, and outcomes of the MACS-VR remained the same as in the original MACS, although the instructions were now to assess the "intensity of alcohol craving experienced during VR exposure".

Anxiety was explored with the Spanish version of the State-Trait Anxiety Inventory (STAI) [48]. The STAI is a self-reported questionnaire with two subscales assessing how a person feels at the moment (STAI-state, state anxiety) and how a person feels in general (STAI-trait, trait anxiety). Each subscale consists of 20 items, and scores of each item range from 0 ("Not at all") to 3 ("Very much so"). Higher scores on each subscale indicate higher levels of trait and/or state anxiety.

Alcohol craving and anxiety experienced during VR exposure were explored with visual analogue scales (VAS). These scales are widely used as instruments to measure craving (VAS-C) and anxiety (VAS-A) during VR exposure [35]. The VAS-C is a self-reported virtual craving scale, with scores ranging from 0 to 100, where 0 is interpreted as "no craving" and 100 indicates "intense craving". Similarly, VAS-A is a self-reported scale aiming to explore anxiety levels, with scores ranging from 0 to 100, where 0 is interpreted as "no anxiety" and 100 as "intense anxiety". VAS-C and VAS-A are ecological scales used during exposure to VR environments.

#### *2.3. Instruments*

#### 2.3.1. Hardware

The VR equipment consisted of Oculus Rift head-mounted display (HMD), 1080 × 1200 resolution per eye, a 90 Hz refresh rate, and 110◦ field of view, sensors, touch controllers, and a computer compatible with the VR technology (INTEL(R) Core(TM) i7-2600 CPU, 16.0 GB RAM, Operating System 64 bits, processor ×64, graphic card NVIDIA GeForce GTX 1080 Ti).

#### 2.3.2. "ALCO-VR" Software

"ALCO-VR" software was developed based on the results of a previous study [49], in which we identified factors that contribute to the elicitation of alcohol craving in a sample of patients diagnosed with AUD. Based on that study, the ALCO-VR software was created, consisting of four VR alcohol-related environments: a restaurant, a bar, a pub, and at-home environments (Figure 1). These VR environments were created considering multiple variables, as in our previous research, such as social interactions (including avatars in the environments), different alcohol-related cues (a menu of 22 alcoholic drinks), or different times of day (daytime or nighttime). Hence, there were two environments during daylight (bar and restaurant) and two environments during nighttime (pub and at-home), one environment with no social interaction (the at-home environment) and three with social interaction (bar, pub, restaurant). All environments were created to simulate real-life scenarios based on patients' experiences. The ALCO-VR platform consisted of two parts, assessment and therapy. The assessment part was the focus of the current study and the ALCO-VR created a hierarchy of exposure from the lowest rated environment with the lowest rated alcoholic drink to the highest rated environment and the highest rated alcoholic drink. On the VAS-A and VAS-C, users were asked to rate how much cue-induced anxiety and craving they considered the environments and alcoholic drinks triggered on a scale from 0 to 100. The assessment part of the ALCO-VR centered on the first five rated alcoholic beverages, the favorite drinks of each participant, which were presented in each VR environment. In addition, the software consisted of a neutral environment (a room with a white background and a glass of water), where the participants could familiarize themselves with the VR technology. A high interaction level between the user and the VR platform was considered fundamental and individuals could approach their alcoholic beverages, hold them and observe them from all angles with Oculus Touch controllers.

**Figure 1.** Pictures of the VR environments.

#### *2.4. Procedure*

Participants from the control group (SD group) were recruited through social media platforms at the University of Barcelona. Patients from the clinical group (AUD group) were informed about the study during one of their appointments with their psychiatrists or clinical psychologists at the Addictive Behaviors Unit at the Hospital Clinic of Barcelona. Once they had agreed to participate, they were referred to the researcher in charge of the study. Participants were asked to sign the informed consent document after a short explanatory introduction regarding the technology used. Subsequently, clinical data were collected from the patients such as dual diagnoses, medication, abstinence data, alcohol consumption, and other substance use (illicit or licit) during the month prior to the experiment. These data were confirmed by their clinical psychologist. Participants were then asked to complete the AUDIT, STAI (the trait part), and MACS. Upon completion of these self-reported scales, the researcher instructed participants how to use the touch controllers, and they were given water to drink in order to not interfere with alcohol consumption patterns. The ALCO-VR software started by assessing how much craving and anxiety was triggered by 2D images of each VR environment (pub, at home, restaurant, and bar) and each alcoholic drink on a VAS from 0 to 100. Based on these responses, the individualized exposure hierarchy consisted of the interplay between the first five chosen alcoholic beverages, and the four contexts, gradually increasing from the lowest rated drink and the lowest rated VR environment to the highest rated alcoholic beverage and highest rated environment. When this initial part of the procedure was completed, the Oculus Rift HMD was attached to the participant's head. This represented the start of the 3D virtual experience and each participant was exposed to the VR environments following his/her hierarchy. The first instruction was to approach the beverage and observe it from all angles for approximately 20 s, but without attempting to virtually drink from it. The VR exposure started with a neutral environment, exposure to a glass of water in a white room,

aiming to familiarize participants with the VR technology. This environment served as a neutral condition in the analyses and as training for the participants. The hierarchy created by the ALCO-VR platform consisted of exposure to each alcoholic drink, which appeared in each environment. Hence, participants self-reported their cue-induced alcohol craving and anxiety levels on the VAS-C and VAS-A throughout exposure to the ALCO-VR software (4 environments × 5 drinks = 20 ratings × 2 (craving and anxiety) = 40 ratings). Olfactory stimuli were introduced during the exposure procedure and corresponded to each drink. Previously prepared alcoholic beverages were transferred onto cotton pads and placed on the table, close to the participant each time a new alcoholic drink appeared during the exposure procedure. Overall, the exposure lasted for approximately 10–15 min. After the experiment, participants were asked to complete the MACS-VR, and STAI (the state part). In addition, they were also asked to rate several variables of the ALCO-VR platform on a scale from 0 to 10, with 0 considered "very low" and 10 was considered "excellent"; these variables were user-friendliness, overall quality of the software, realism of the environments, and realism of the alcoholic beverages. Finally, for the AUD group, a debriefing session was carried out at the end of the procedure with the aim of reducing momentary craving and anxiety levels and to minimize any further possible alcohol use. Each session lasted for approximately one hour. The cross-sectional sessions were delivered by an experienced clinician-scientist at VR-Psy Lab, University of Barcelona.

#### *2.5. Statistical Analysis*

As the assumption of normality of data was not satisfied according to the Shapiro–Wilks test *p-*value of ≤0.05, non-parametric tests were applied in this study. First, Friedman tests were run separately for each group to compare participants' levels of craving and anxiety experienced during VR exposure to the neutral environment and to the four alcohol-related environments. Post-hoc Wilcoxon signed-rank tests were used to determine specific significant differences in craving and anxiety between the conditions in both groups. In addition, Mann–Whitney U tests were applied to explore differences between the AUD group and SD in terms of anxiety and alcohol craving levels experienced during exposure to the VR alcohol-related environments. Spearman correlations were conducted to explore the relationship between self-reported craving and anxiety responses on VAS-C/VAS-A and the results of AUDIT, STAI, and MACS. Finally, due to the gender and age imbalance, we performed analysis of covariance (ANCOVA) after converting our data into rank scores. All statistical analyses were carried out using IBM SPSS version 24.

#### **3. Results**

Table 1 indicates the data of the self-reported questionnaires in terms of alcohol misuse (AUDIT), trait-anxiety (STAI, trait part), alcohol craving during the last week prior to the experiment (MACS), state anxiety (STAI, state part), and momentary cue-induced alcohol craving immediately after exposure to the ALCO-VR software (MACS-VR).


**Table 1.** Data of the self-reported scales.

#### *Data of the Self-Reported Questionnaires*

Significant differences were found in self-reported levels of anxiety and alcohol craving across exposure to the VR environments. The Friedman test indicated statistically significant differences in self-reported anxiety levels reported on the VAS-A across the VR environments in both AUD patients (χ<sup>2</sup> (4) = 11.26, *p* = 0.02) and the SD group (χ<sup>2</sup> (4) = 15.78, *p* = 0.003). In the AUD group, post-hoc Wilcoxon rank test confirmed that there were significant differences in cue-induced anxiety reported on the VAS-A between the neutral and at-home environments (*Z* = −2.621, *p* = 0.009), between neutral and bar environments (*Z* = −2.447, *p* = 0.01), between neutral and restaurant scenarios (*Z* = −2.762, *p* = 0.006) and between neutral and pub environments (*Z* = −2.691, *p* = 0.007). No statistically significant differences were found between the four VR alcohol-related environments on self-reported anxiety on the VAS-A (*p* > 0.05). Figure 2 confirms that self-reported levels of anxiety were higher during exposure to the VR alcohol-related environments than during exposure to the neutral environment. However, in the SD group, we found the opposite pattern. Although there were statistically significant differences between neutral and at-home environments (*Z* = −2.707, *p* = 0.007), between neutral and bar environments (*Z* = −2.621, *p* = 0.009), between neutral and restaurant environments (*Z* = −2.589, *p* = 0.01), and between neutral and pub environments (*Z* = −2.135, *p* = 0.03), the scores of self-reported anxiety were higher in the neutral environment than in the four VR alcohol-related environments, as Figure 2 shows.

**Figure 2.** Self-reported mean levels of anxiety (A) on visual analogue scales (VAS)-A.

Regarding self-reports of alcohol craving on the VAS-C, the Friedman test revealed a statistically significant difference across the VR environments in the AUD patients' group (χ<sup>2</sup> (4) = 22.83 *p* = 0.000), but not in the SD group (*p* > 0.05). In the AUD group, post-hoc analyses with Wilcoxon rank test showed significant differences in self-reported alcohol craving on the VAS-C between the neutral and at-home environments (*Z* = −3.041, *p* = 0.002), between neutral and bar environments (*Z* = −3.180, *p* = 0.001), between neutral and restaurant environments (*Z* = −3.110, *p* = 0.002), and between neutral and pub environments (*Z* = −3.110, *p* = 0.002). As can be seen in Figure 3, no statistically significant differences in alcohol craving reports on the VAS-C were found across the four VR alcohol-related environments. Both Figures 2 and 3 indicate significantly higher scores in the AUD group than in the SD group regarding self-reported anxiety and alcohol craving levels on the VAS-A and VAS-C, respectively. As there were more significant differences in terms of anxiety levels compared to craving levels, we consider that cue-induced anxiety responses can better differentiate between SD and AUD patients than self-reported craving responses.

The results of the Mann–Whitney U test indicated a statistically significant difference between AUD and SD groups when comparing the mean anxiety and alcohol craving responses in each VR alcohol-related environment (*p* < 0.05). As can be seen in Table 2, self-reported anxiety and alcohol craving levels were higher in the AUD group than in the SD group across all VR alcohol-related environments. There were no statistically significant differences between AUD and SD groups regarding their anxiety and alcohol craving responses in the neutral environment (*p* > 0.05).

**Figure 3.** Self-reported mean levels of alcohol craving (C) on VAS-C.


**Table 2.** Differences in anxiety and alcohol craving self-reports on VAS-A and VAS-C.

<sup>a</sup> IQR, interquartile range; <sup>b</sup> Mann–Whitney U test. Bonferroni adjustment for multiple testing; \*\*\* *p* < 0.001; \* *p* < 0.05; <sup>c</sup> env, environment.

In addition, in the AUD group, there was a strong, positive correlation between alcohol dependence severity as shown by AUDIT and total craving reported on the VAS-C (*rs* = 0.783, *p* = 0.002) and total anxiety reported on the VAS-A (*rs* = 0.650, *p* = 0.016). In addition, there was a significant relationship between trait-anxiety as assessed by the STAI (the trait part of the questionnaire) and the total alcohol craving score reported on the VAS-C (*rs* = 0.667, *p* = 0.013). A positive correlation was also found between cue-induced craving as explored by the MACS-VR and total alcohol craving responses reported on the VAS-C (*rs* = 0.581, *p* = 0.03). No other correlations were found in anxiety and craving levels assessed by different instruments (*p* > 0.05). Regarding the SD group, the only correlations were found between cue-induced craving as assessed by MACS-VR and mean total craving reported on the VAS-C (*rs* = 0.629, *p* = 0.016) and mean total anxiety reported on the VAS-A (*rs* = 0.707, *p* = 0.005).

Addressing the age and gender imbalance in our study, ANCOVA analyses were run to determine the effect regarding gender and age on total mean levels of anxiety and alcohol craving in SD and AUD patients. After converting the scores to ranks, we performed ANCOVA analyses on the rank scores. After adjustment for gender, there was a statistically significant group difference regarding their total mean anxiety levels reported on VAS-A *F* (1, 24) = 39.313, *p* < 0.001, partial η<sup>2</sup> = 0.621. Post-hoc analysis was performed with a Bonferroni adjustment. Anxiety levels were significantly greater in AUD patients (M = 54.8, SE = 6.33) than the SD group (M = 10.5, SE = 3.47), a mean difference of 13.7, 95% confidence interval (CI) (9.23, 18.3), *p* < 0.001. Similarly, after adjusting for gender, there was a statistically significant group difference regarding total mean craving levels reported on VAS-C *F* (1, 24) = 10.075, *p* < 0.005, partial η<sup>2</sup> = 0.296. Post-hoc Bonferroni analysis revealed that AUD patients (M = 58.2, SE = 6.05) reported significantly greater craving levels than the SD group (M = 30.79, SE = 5.4) (mean difference of 9.63, 95% CI (3.37, 15.90), *p* < 0.005).

In addition, after adjusting for age, there were statistically significant differences between SD and AUD groups in terms of their cue-induced responses. There was a statistically significant difference between the two groups regarding their total mean anxiety levels *F* (1, 24) = 12.317, *p* < 0.005, partial η<sup>2</sup> = 0.339. Anxiety levels were significantly greater in AUD patients (M = 48.23, SE = 6.33) compared to the SD group (M = 23, SE = 3.4) (mean difference of 17.65, 95% CI (7.27, 28.04), *p* < 0.005). Finally, there was no statistically significant difference between groups in terms of total mean craving regardless of age *F* (1, 24) = 3.764, *p* > 0.05, partial η<sup>2</sup> = 0.136. These results can be appreciated in Figures 4 and 5.

**Figure 4.** Self-reports of total mean anxiety in AUD and social drinkers (SD) groups.

**Figure 5.** Self-reports of total mean craving in AUD and SD groups.

Finally, the participants rated the ALCO-VR software according to the following variables: user-friendliness (MAUD = 9.15, SD = 0.89; MSD = 8.93, SD = 0.91), overall quality of the software (*MAUD* = 8.23, SD = 1.58; MSD = 8, SD = 1.3), realism of the environments (MAUD = 7.77, SD = 1.53; MSD = 7.8, SD = 1.5), and realism of the alcoholic beverages (MAUD = 7.38, SD = 1.93; MSD = 7.21, SD = 1.31).

#### **4. Discussion**

The current study was centered on testing the effectiveness of ALCO-VR software to induce anxiety and craving responses. Its secondary objective was to assess which craving or anxiety responses could best differentiate between the AUD and SD groups. The VR alcohol-related environments induced significantly greater anxiety levels than the neutral environment in our clinical group, but not in our control group. On the contrary, the VR alcohol-related environments induced greater craving levels

than the neutral environment in both clinical and control groups. More significant alcohol craving responses were found among AUD patients.

Our results are consistent with previous research [35] and confirm the effectiveness of the ALCO-VR platform for inducing anxiety and craving responses. Patients diagnosed with AUD displayed high levels of anxiety during exposure to alcohol-related contexts compared to our control group. Furthermore, craving responses in the AUD group were even higher than in the SD group during exposure to the VR alcohol-related environments than during exposure to the neutral environment. This response pattern is supported by previous research showing greater levels of craving within alcohol-related VR environments [36,50]. In substance use disorders, VR smoking-related environments elicited higher craving levels in nicotine-dependent undergraduate students than in the control condition [51]. In a virtual reality-methamphetamine (VR-METH) cue-exposure paradigm, methamphetamine users experienced higher craving levels than in the neutral condition [40]. Additionally, in behavioral addictions, a significant internet gaming VR environment induced higher craving levels in patients diagnosed with internet gaming disorder than in a control group [52]. Similarly, in gambling disorder, frequent gamblers experienced higher craving induced by the VR environment than non-frequent gamblers [53].

Furthermore, although there were no statistically significant differences across the VR environments in our SD group, we observed a gradual increase in alcohol craving during exposure to the VR alcohol-related environments compared to the neutral scenario. We interpret this result as being representative of a group of young social drinkers, whose drinking patterns involve social interaction and peer pressure, especially in bars and pubs. The result also confirms the contextual specificity of craving for alcohol [49,54].

It is worth mentioning that both SD and AUD patients did not self-report high levels of anxiety and there were no statistically significant differences between groups at baseline measurement. For instance, 20 or 30 self-reported anxiety levels on a VAS from 0 to 100 during the VR neutral environment did not represent significant levels of anxiety. The fact that the SD group had slightly lower anxiety levels in VR alcohol-related environments indicated that these participants did not associate alcohol-related stimuli with the aversive consequences of drinking patterns. However, in our examination of the effectiveness of the ALCO-VR, we found significant differences between AUD and SD groups in anxiety and craving responses across all VR alcohol-related environments. Interestingly, in terms of cue-induced anxiety and craving responses, AUD patients reported similar levels of craving and anxiety during exposure to the VR alcohol-related environments. However, SD cue-induced anxiety responses were lower than their craving responses. SD had a different response tendency from the clinical group regarding anxiety levels. Cue-induced anxiety responses better differentiated the AUD patients and the SD group than craving. It should be noted that craving is elicited by stimuli in both groups, but it may not be associated with anxiety in SD, while, due to the ambivalence of patients in treatment (want to drink and want to remain abstinent) the same cravings may lead to high levels of anxiety.

Addressing age and gender imbalance in this study, our results confirm that AUD patients self-reported greater anxiety and alcohol craving levels compared to participants from the SD group. Similar results were found in a previous study in patients diagnosed with bulimia nervosa (BN), in which anxiety responses discriminated best between BN patients and healthy participants [42]. Although BN and AUD are apparently different disorders, we consider these similar results as an indicator that BN and AUD share common causal and maintenance mechanisms [55].

Apart from the craving and anxiety responses self-reported on the VAS-C and VAS-A respectively, our participants completed the following instruments: AUDIT, MACS, STAI, and MACS-VR. The results showed significantly higher scores in all the instruments in the clinical group than in the control group. The strong correlation between the results of these questionnaires and the scores reported on the VAS-C and the VAS-A confirms the validity of the ALCO-VR software for measuring cue-induced alcohol craving and anxiety, and so we consider that ALCO-VR software may be a useful EMA tool to assess craving and anxiety levels. Interestingly, the MACS-VR instrument exploring alcohol craving

immediately after ALCO-VR exposure indicated higher levels of cue-induced alcohol craving than the MACS instrument exploring alcohol craving in the week prior to the experiment. This confirms that the ALCO-VR software can elicit momentary alcohol craving not only in AUD patients, but in SD as well. Therefore, we argue that the ALCO-VR software is a valid instrument for exploring cue-induced anxiety and alcohol craving responses and can differentiate between groups of AUD patients and SD.

The current study is part of a larger project aiming to develop a VR therapy tool based on the cue-exposure paradigm to alcohol-related cues and contexts in patients with severe AUD and with several failed attempts to cease alcohol use. First, we conducted a pilot study testing a VR software based on alcohol-related cues and environments; the results indicated that the software distinguished between heavy drinkers (HD) and light drinkers (LD) in terms of behavioral parameters. LD displayed a preference for non-alcoholic drinks, whereas most heavy drinking individuals preferred alcoholic beverages [56]. In view of these results, we attempted to develop a new VR platform designed to treat patients diagnosed with AUD who were considered "resistant-to-treatment-as-usual" (TAU). A common procedure for identifying specific triggers for alcohol craving is to conduct exhaustive interviews with relevant populations, in our case, with patients with AUD. Therefore, we conducted a second study in which we developed an ad-hoc questionnaire to determine significant factors such as cues and contexts that elicit craving in AUD patients with the aim of creating naturalistic real-life environments [49]. We developed the ALCO-VR software based on the results of the previous study. In addition, the current study is the preceding step to test the efficacy of the cue-exposure therapy (CET) approach using the VR technology in AUD patients.

As substance use disorders and eating disorders resemble addictive behavior mechanisms [57], our study was based on a previous study centered on patients diagnosed with BN, which was conducted at VR-Psy Lab from the University of Barcelona. The aim of the study was to test the efficacy of a VR-cue exposure therapy (VR-CET) for patients diagnosed with BN. The study showed positive results in terms of abstinence rates, anxiety, and craving levels in patients with BN in the VR-CET group compared to the control group. The first step was to perform exhaustive interviews with BN patients to determine triggering factors for binge eating behaviors [58]. Based on the results of that study, VR environments were created and tested in BN patients to examine the elicitation of food craving [42]. These two studies were essential to develop a VR-CET for BN patients. The efficacy of the VR-based therapy software was tested in a multi-site clinical trial and it was found to achieve better outcomes than cognitive-behavioral therapy (CBT) both at the post-treatment assessment [59] and at six-month follow-up [55]. Several limitations should be noted with regard to our study. First, our control group consisted mostly of young women from the Faculty of Psychology, where the majority of students are female. Second, our clinical and control samples were limited. For future studies, we propose that larger samples should be used and that the effectiveness of the ALCO-VR software to differentiate between patients with long-term versus short-term abstinence period in terms of alcohol craving and anxiety should be tested. In addition, to address the gender and age imbalance between our groups, we will consider these limitations for upcoming studies of the same project.

#### **5. Conclusions**

The current study demonstrated the validity of the ALCO-VR software to induce alcohol craving and anxiety responses, particularly in patients diagnosed with AUD. The ALCO-VR can perform the ecological assessment of momentary cue-induced alcohol craving and anxiety levels, and can differentiate between individuals with AUD and those that are SD.

Considering the promising results of this study, the next step would be to test the therapeutic use of the software in patients diagnosed with AUD. The ALCO-VR is currently being implemented in a multi-site clinical trial testing its efficacy as a cue-exposure therapy tool for patients diagnosed with AUD, who are considered resistant to TAU. The ultimate goal of the project is to reduce relapse rates upon completion of treatment, since relapse remains one of the greatest challenges in AUD treatment and follow-up.

*J. Clin. Med.* **2019**, *8*, 1153

**Author Contributions:** Conceptualization, J.G.-M., M.F.-G. and A.G. (Alexandra Ghi¸tă); methodology, J.G.-M., M.F.-G. and A.G. (Alexandra Ghi¸tă); software, J.G.-M., M.F.-G., A.G. (Alexandra Ghi¸tă) and B.P.-G.; validation, A.G. (Alexandra Ghi¸tă), O.H.-S., Y.F.-R., B.P.-G., M.M., L.O., S.M., L.T. and A.G. (Antoni Gual); formal analysis, A.G. (Alexandra Ghi¸tă); investigation, A.G. (Alexandra Ghi¸tă), O.H.-S., Y.F.-R., B.P.-G., M.M., L.O., S.M., L.T. and A.G. (Antoni Gual); resources, J.G.-M.; data curation, A.G. (Alexandra Ghi¸tă); writing—original draft preparation, A.G. (Alexandra Ghi¸tă); writing—review and editing, A.G. (Alexandra Ghi¸tă), J.G.-M., O.H.-S., M.F.-G., B.P.-G., M.M., A.G. (Antoni Gual); supervision, J.G.-M.; project administration, J.G.-M.; funding acquisition, J.G.-M.

**Funding:** This research was funded by the Spanish Ministry of Health, Social Services and Equality, Delegation of the Government for the National Plan on Drugs (FEDER/UE/Project 2016I078: "ALCO-VR: Virtual Reality-based protocol for the treatment of patients diagnosed with severe alcohol use disorder"). The study also received support from AGAUR, Generalitat de Catalunya, 2017SGR1693.

**Acknowledgments:** This study was supported by the Spanish Ministry of Health, Social Services and Equality, Delegation of the Government for the National Plan on Drugs (FEDER/UE/Project 2016I078: "ALCO-VR: Virtual Reality-based protocol for the treatment of patients diagnosed with severe alcohol use disorder"). The study also received support from AGAUR, Generalitat de Catalunya, 2017SGR1693.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Motor Adaptation Impairment in Chronic Cannabis Users Assessed by a Visuomotor Rotation Task**

**Ivan Herreros 1,**†**, Laia Miquel 2,3,\*,**†**, Chrysanthi Blithikioti 2,3, Laura Nuño 2, Belen Rubio Ballester 4, Klaudia Grechuta 4, Antoni Gual 2,3, Mercè Balcells-Oliveró 2,**‡ **and Paul Verschure 4,5,**‡


Received: 31 May 2019; Accepted: 16 July 2019; Published: 18 July 2019

**Abstract:** Background—The cerebellum has been recently suggested as an important player in the addiction brain circuit. Cannabis is one of the most used drugs worldwide, and its long-term effects on the central nervous system are not fully understood. No valid clinical evaluations of cannabis impact on the brain are available today. The cerebellum is expected to be one of the brain structures that are highly affected by prolonged exposure to cannabis, due to its high density in endocannabinoid receptors. We aim to use a motor adaptation paradigm to indirectly assess cerebellar function in chronic cannabis users (CCUs). Methods—We used a visuomotor rotation (VMR) task that probes a putatively-cerebellar implicit motor adaptation process together with the learning and execution of an explicit aiming rule. We conducted a case-control study, recruiting 18 CCUs and 18 age-matched healthy controls. Our main measure was the angular aiming error. Results—Our results show that CCUs have impaired implicit motor adaptation, as they showed a smaller rate of adaptation compared with healthy controls (drift rate: 19.3 +/− 6.8◦ vs. 27.4 +/− 11.6◦; t(26) = −2.1, *p* = 0.048, Cohen's *d* = −0.8, 95% CI = (−1.7, −0.15)). Conclusions—We suggest that a visuomotor rotation task might be the first step towards developing a useful tool for the detection of alterations in implicit learning among cannabis users.

**Keywords:** cerebellum; cannabis; implicit motor learning; motor adaptation; visuomotor rotation

#### **1. Introduction**

Cannabis is the most consumed illicit drug of abuse worldwide, ranking right after alcohol and tobacco [1,2]. The effects of the principal active component of cannabis, Delta9-tetrahydrocannabinol (THC), on the brain are still not well characterized. The cerebellum has several motor and cognitive functions [3] and has been suggested as a crucial structure for the addiction brain network [4]. It is well known that the cerebellum has a very high density of Cannabinoid receptor type 1 (CB1) receptors; however, little is known about its functional deficits due to chronic cannabis use. Motor adaptation is a cerebellum-dependent motor function that can be assessed with a visuomotor rotation (VMR) task and it might be altered in chronic cannabis users. Until now, there are no diagnostic tools that allow us to rate possible cerebellar impairments. We suggest that the VMR task could become a tool to assess cerebellar alterations due to chronic cannabis use (CCU).

Given the current trend towards the tolerance/legalization of its medical and recreational uses [5,6], it is crucial to clarify how long-term cannabis consumption affects brain function and performance. Numerous studies have evaluated the behavioral effects of acute and non-acute cannabis use [7], showing effects of cannabis not only on cognitive processes such as attention and memory [8–10], but also on psychomotor tasks, where reaction times and the ability to inhibit motor actions are impaired [11,12]. In addition, acute cannabis consumption increases the risk of driving accidents two- to three-fold [13]. However, there is still no correlation between the known molecular effects of cannabis and changes in overt behavior. This hampers our ability to assess cannabis-induced levels of dysfunction and the associated risks of chronic use and cannabis addiction.

Progress in understanding the consequences of cannabis requires linking its effects on the biological substrate with deficits in performance. THC [14] acts in the central nervous system through the CB1 receptor [15,16]. CB1 receptors are located in presynaptic terminals and have a regulatory effect on synaptic transmission [17,18]. The distribution and density of CB1 receptors in the brain may indicate which brain areas are more affected by the prolonged use of cannabis. Even though CB1 receptors are expressed all over the brain [19,20] (e.g., hippocampus, amygdala, striatum), their concentration is especially high in the molecular layer of the cerebellum [19]. In particular, cannabis consumption has been shown to down-regulate the CB1 receptors in the cerebellar molecular layer [21], affecting cerebellar plasticity mechanisms [22]. Indeed, chronic use of cannabis affects cerebellar-dependent behaviors both in animals and humans—specifically eye-blink conditioning, which is one of the basic paradigms for studying cerebellar associative learning [23]. CB1 knock-out mice and mice exposed to CB1 antagonists display a decreased acquisition of cerebellar-dependent eye-blink responses [24]. In humans, this result has been replicated in non-acute chronic cannabis users tested after 24 h of abstinence [25]. However, it is still not clear how deficits in eye-blink conditioning translate to easily administered diagnostic tests and instrumental activities of daily life.

Motor adaptation is a cerebellar-dependent function that has not been studied in chronic cannabis users (CCUs). Through adaptation, the motor control system readjusts to changes in the dynamics of their musculoskeletal system and body configuration [26,27]. Motor adaptation occurs in automatically when a difference between the predicted trajectory and the observed movement exists. Motor adaptation is frequently studied with visuomotor rotation paradigms [28], where a specific perturbation creates a mismatch between the motor action and the observed outcome. Healthy individuals will gradually compensate for this mismatch and adjust to sizeable perturbations, without awareness of the process (implicit motor learning). One way to speed up adaptation is by introducing a strategy that enables participants to explicitly adjust their aiming direction to hit a given target, as Mazzoni and Krakauer did (MK task) [29]. By inserting a rule, healthy participants manage to immediately cancel aiming errors (explicit motor learning) but at the expense of setting the predicted trajectory and the actual trajectory in conflict. In other words, with practice, participants make increasingly larger errors (implicit motor learning occurs). In general, cerebellar pathologies induce reduced adaptation [30,31]. Specifically, in the MK task, cerebellar-ataxic patients, despite using an explicit strategy [32], performed "better" (with less systematic error), as implicit adaptation is impaired.

Given the above observations, we sought to assess the impact of chronic cannabis use (CCU) in motor adaptation using a VMR paradigm. We hypothesize that THC, acting as an exogenous agonist of the cannabinoid receptors, will diminish cerebellar plasticity. For this reason, we look for a measurable impact on motor adaptation—a cerebellar-dependent process that facilitates maintaining accurate control of motor behavior [27]—that can be studied with the VMR task [29,32]. We expect that, alike to cerebellar ataxic patients, chronic cannabis consumers will have reduced implicit motor adaptation and thus will perform more accurately than healthy controls on the VMR task.

#### **2. Experimental Section**

#### *2.1. Design*

We conducted a case-control study. The study protocol is registered in clinicaltrials.gov, ID: NCT02816034. The study was approved by the Ethics Committee of Hospital Clínic de Barcelona (decision number HCB/2016/0018).

#### *2.2. Participants*

Study participants were between 18 and 50 years old. We included in the experimental group individuals following Diagnostic and Statistical Manual of Mental Disorders (DSM–V) criteria for Cannabis Use Disorder that screened positive for cannabis in the urine analysis performed the same day of their assessment. For the experimental and control groups, we recruited right-handed participants, without another active substance-use disorder (except of tobacco use disorder), with normal or corrected-to-normal vision. Participants with a cognitive impairment, such as mental retardation, or with psychotic disorders were excluded. Cannabis users were recruited among the patients of the Addiction Unit of a tertiary Hospital and their acquaintances. The sample included 17 CCUs and 18 healthy age-matched controls. All the participants signed an informed consent form before the initiation of the experiment. Participants were compensated for their effort with a voucher equivalent to a lunch ticket.

#### *2.3. Procedures*

A clinical interview was performed by a psychiatrist and a psychologist with mental health and addiction expertise. Socio-demographic data, including gender and age, were collectedduring the clinical interviews. DSM-V criteria were used to diagnose patients with Cannabis Use Disorder. Furthermore, we gathered information about the frequency and quantity of cannabis consumed during the last week and the last 6 months using an ad hoc questionnaire. We used the first question of the Cannabis Use Disorder Identification Test (CUDIT) [33] to determine the frequency of cannabis use. In order to quantify cannabis use we used the Standard Joint Unit (1 SJU = 1 joint = 7 mg THC) [34]. The Scale for the Assessment and Rating of Ataxia (SARA) [35] was administered to assess ataxia. The SARA is a semi-quantitative scale, which has 8 items and assesses motor changes in gait, stance, sitting, speech disturbances, finger chase, nose-finger chase, fast alternating hand movements and heel–shin slide.

Participants sat in front of a desk (Figure 1A,B) and were asked to control a cursor that was displayed on a screen, using an optical pen. Its position was recorded using a high-resolution digitizing tablet (Intuos3, Wacom, Saitama, Japan) at a sampling rate of 100 Hz. During the task, every trial required moving the stylus with the right hand over the digitization tablet from an initial central position. Visual feedback was projected on a surface placed in front of the subject. This surface occluded the subject's view of the movements of their arm. The projection displayed eight circles over the surface, arranged in a circle (Figure 1C,D). For each trial, a target marked with a bulls-eye appeared inside one of these eight circles. Participants were instructed to perform a center–out reaching movement from the initial position towards the target with the pen. Participants moved as fast as possible to reach the target, which in turn turned red Then they then returned the pen to the start position. Depending on the phase of the experiment, the correct target was either the bulls-eye or the clockwise adjacent marker.

**Figure 1.** Perspective (**A**) and top view (**B**) of the experimental setup. Subjects were seated approximately 25 cm in front of a table with both hands occluded to vision. Movement trajectories were recorded through a digitizing tablet (Intuos Pro, Wacom, Saitama) located at waist height at the lower part of the table. We used a projector (BenQ MP512ST, Texas Instruments, Taoyuan, Taiwan), mounted 55 cm above the table, to display the behavioral tasks. The size of the projection was 46 × 61 cm; data were sampled and stimuli displayed at 100 Hz. Prior to the experiment, we adjusted the seat in order to ensure that the distance between the center of the projection and the eyes of each participant was approximately 40 cm. (**C**) Visuomotor task. A centered home position (gray square) and eight circles (red) are projected over the surface, arranged in a circle. For each trial, a target appeared inside one of these eight circles (green circle). (**D**) During the Rotation and Rotation Strategy phases, the visual feedback (black arrow) of actual center–out reaching movements (green arrow) deviate by 45 degrees in a counter-clockwise direction.

The experimental protocol constisted of 7 phases and a total amount of 242 movements. These movements were grouped in 30 blocks of 8 trials and 1 of only 2trials. We replicated the protocol performed in cerebellar ataxic patients [32]. The protocol started with two baseline adaptation phases (B1 and B2) separated by a cognitive control phase (practice strategy, PS). Each baseline phase consisted of 24 movements (three blocks) where the participants moved towards the indicated target, and the visually projected cursor reproduced the participants' movement trajectory. In the PS phase, participants were asked to aim to the marker adjacent to the target in the clock-wise direction (that is, the marker appearing at 45 degrees to the clock-wise direction). After these three phases, a visual rotation was introduced without warning during two trials. In these trials, the mapping of the movement-to-visual coordinates was altered such that the position of the displayed cursor was rotated 45 degrees counter-clockwise relative to the starting position. In order to force fast reaching movements,

trials were repeated when movements were not completed within a time limit of 1.5 seconds after target appearance.

After these two trials, participants started the main experimental phase, i.e., the rotation plus strategy (RS) phase, where they were instructed to overcome the rotation by applying the strategy learned in the PS phase. The RS phase lasted 80 trials (10 blocks), that is, ten movements towards each target in random order. The last phase of the experiment started with a no-feedback washout (NF) phase, where participants were instructed not to apply the rotation strategy anymore, while they performed a series of 8 reaching movements, in the absence of visual feedback. Finally, in an 80 trials long washout phase (W), visual feedback was reinstalled in the absence of any rotation. The whole task lasted approximately 20 min.

#### *2.4. Analysis*

We measured performance in terms of the angular error—the angular difference between the executed movement and target directionsThis measure can be very noisy on a trial by trial basis, mostly because participants sometimes lapse and they apply the aiming rule incorrectly. For this reason we used the median angular error per block of 8 trials, noting that within a block all eight target positions were presented once.

One patient and one control were removed from the final sample because of laterality problems and were not included in the analyses. An additional patient was discarded because her behavior was a clear outlier (3 SD) from the mean performance score and was not included in the analyses. Additionally, to avoid acute cannabis use effects on performance, we removed from the analysis all participants that had consumed cannabis within the 6 hours prior to performing the task (*n* = 4), as this interval has been used as a criterion for cannabis' acute effects [36].

If not stated otherwise, sample estimates are reported as mean +/− standard deviation, and qualitative data are described as percentages. Chi-square test and t-tests were done to compare groups depending on the type of data. Bonferroni correction was performed for multiple comparisons and Cohen coefficient for effect size.

#### **3. Results**

CCUs (*n* = 11, 4 women) had a mean age of 28.7 +/− 8.6 years old, which was age-matched by the control group (CG) (*n* = 17, 8 women; 30.7 +/− 7.1 years old). CCUs were diagnosed with a severe cannabis use disorder (≥ 6 criteria) and 75% were marijuana smokers. One subject was diagnosed with an attention deficit hyperactivity disorder in childhood but did not fulfill DSM-V criteria at the time of inclusion in the study. Experimental subjects showed significant differences in educational level compared to controls. For more details on clinical and sociodemographic characteristics, see Table 1.


**Table 1.** Sociodemographic and clinical characteristics of the experimental (chronic cannabis users) and control groups.


**Table 1.** *Cont.*

SD: Standard Deviation; SJU: Standard Joint Unit (1 SJU = 7 mg THC); SDU: Standard Drink Unit (1 SDU = 10 gr of pure alcohol). SARA: Scale for the Assessment and Rating of Ataxia. Bonferroni correction = *p*-value = 0.006.

The evolution of the angular error during the VMR task was our main behavioral outcome (Figure 2A). Cannabis users as well as controls demonstrated an average evolution of the angular error consistent with the expected interference between the use of an explicit strategy and an implicit adaptation process, as reported for healthy subjects in relevant studies [28,37]. Both groups experienced an increase in error (a drift towards positive angular errors) as trials accumulated in the RS phase. To assess implicit motor adaptation, we used both the maximum aiming error (degrees in cursor space) during the RS phase, or the *peak drift* [31], together with the *drift rate*, measured as the slope of the adaptation curve in the RS phase (in degrees per trial). To reduce the noise that was introduced by outlying trials, both measures were computed based on block scores, obtained as the median error for the eight trials in each block. Drift rate was significantly smaller for CCUs compared to controls (0.1 +/− 0.1◦/trial vs. 0.2 +/− 0.1◦/trial; t(26) = −2.2, *p* = 0.037, Cohen's *d* = −0.9, 95% CI = (−1.8, −0.1)) and so was peak drift (19.3 +/− 6.7◦ vs. 27.4 +/− 11.6◦; t(26) = −2.1, *p* = 0.048, Cohen's *d* = −0.8, 95% CI = −1.7, −0.1), confirming the experimental hypothesis that chronic cannabis consumption can reduce motor adaptation. However, even though CCUs tended to display larger after-effects than controls, which were measured as the remnant error during the no-feedback washout phase (9.1 +/− 8.7◦ vs. 13.5 +/− 4.5◦; t(26) = −1.8; *p* = 0.09), the difference did not reach significance. Notably, adaptation occurred very rapidly at the onset of the RS phase in comparison to the literature [38], which suggests that the use of continuous feedback might have provoked a fast explicit adaptation process in addition to implicit adaptation.

The difference in performance was not due to an intrinsic difference between CCUs and controls in unperturbed aiming, as both groups scored similarly in both baseline periods (−2.3 +/− 3.4◦ vs. −2.1 +/− 2.5◦; t(26) = −0.1, *p* = 0.89 and −1.4 +/− 1.8◦ vs. −1.4 +/− 2.1◦; t(26) = −0.1, *p* = 0.95 in B1 and B2, respectively). Differences were also minor in execution at either the onset or the offset of the PS phase (block 4: −6.4 +/− 5.8◦ vs. −5.0 +/− 5.2◦; t(26) = −0.7, *p* = 0.51 and block 6: −4.0 +/− 5.7◦ vs. −2.1 +/− 3.1◦; t(26) = −1.1, *p* = 0.26). No significant differences were found between CCUs and the CG in VMR task performance in terms of timings.

Finally, even though the experimental and control groups did not differ significantly in terms of alcohol and tobacco consumption, one could not a-priori discard their involvement in the reduced adaptation observed in this study, as they both are psycho-active substances. To clarify this, we performed an ordinary least-squares analysis, where we added the levels of tobacco and alcohol consumption as possible confounds in addition to the group variable. Given the reduced sample size, the contrast between CCUs vs. controls only at the trend level (*p* = 0.10), but this was in striking contrast with the almost null influence observed both for tobacco (*p* = 0.75) and alcohol (*p* = 0.99).

**Figure 2.** Motor adaptation in chronic cannabis users (CCUs) and control participants (Conts). (**A**): Evolution of the average error. Vertical lines separate different phases of the experiment, indicated by the labels: B1, first baseline period; PS: practice strategy; B2: second baseline period; R: rotation; RS: rotation plus strategy; NF: no feedback washout; W: regular washout. (**B**): Rate of error increase in the RS phase. Shaded areas and error bars indicate the 95% confidence interval of the mean.

#### **4. Discussion**

This is the first study that analyzes the effects of chronic cannabis consumption on a visuomotor rotation task to assess the processes involved in motor adaptation. CCUs displayed impaired adaptation, manifested as lower aiming errors in the VMR task than controls.

Our results are consistent with our hypothesis that chronic cannabis use has an impact on cerebellar-dependent functions and, more specifically, on motor adaptation. The visuomotor adaptation paradigm has been used as a marker of cerebellar damage in ataxic patients, who show impaired implicit learning, putatively, due to disrupted forward internal models [32]. More specifically, cerebellar ataxic patients accumulate fewer aiming errors during the rotation and strategy (RS) phase, indicating a decreased motor adaptation [32]. It is interesting to note that the effect size of the original study with ataxic patients is very sizeable (Cohen's d = −2.627), while the effect size on CCUs is moderate (Cohen's d= −0.8). This is not surprising, as the clinical manifestation of cerebellar dysfunction of ataxia is severe, while cannabis-induced cerebellar alterations do not lead to a clinical phenotype (all the participants assessed scored zero on the SARA scale). However, we suggest that our results are clinically meaningful for two reasons: First of all, in several brain disorders, early alterations in the patient's brain are present even decades before the first clinical symptoms appear [39,40]. Our paper shows an altered performance of CCUs in a visuomotor adaptation task, indicating subtle cerebellar alterations in the users´ brains. These results were obtained after excluding experimental participants under the acute effects of cannabis (on the suggestion of an anonymous reviewer). Interestingly, even though this reduced the sample size (from *n* = 33 to *n* = 28), statistical significance, and therefore the underlying effect size, increased (from d = −0.7 to −0.8). Our finding suggests that CCUs adapt less when they are not under the acute effects of cannabis. This deficit might be explained by the hypothesis of CB1 receptor down-regulation due to cannabis use [21]. More specifically, it has been shown that repeated cannabis exposure causes endocytosis of the endocannabinoid receptors, resulting in reduced activity of those receptors in the absence of their exogenous agonist. If this hypothesis explains our results, it would be plausible that acute cannabis use in CCUs may indeed hide an underlying impairment in cerebellar-dependent motor adaptation. Although our results show very subtle alterations compared to ataxic patients, these might be the first signs of a future clinically significant impairment that will only manifest with continuous heavy use [41]. Future longitudinal studies will clarify whether these

early cerebellar alterations prelude long-term clinical symptoms. Secondly, it is noteworthy that even though, CCUs show no measurable motor deficits clinically, they share a certain degree of alterations in motor adaptation with cerebellar ataxic patients. Future research will show the reliability of the task as a potential tool for assessing brain alterations, as well as whether the latter are present exclusively in cannabis users or they are part of addiction-induced brain alterations.

Globally, these results indicate that VMR tasks might be a simple, non-invasive and novel way to start exploring a basic cerebellar function in CCUs. The cerebellum is crucial for motor adaptation, among other functions. As implicit learning is an essentialskill for everyday functioning, our findings are the first step towards understanding more complex cerebellar deficits. Our results are in line with the increasingly recognized role of the cerebellum in higher cognitive functions and emotional processes, and a recent review that points out a cerebellar involvement in cannabis addiction [42]. Future research should examine whether there are changes in the performance of users after cessation of use, as well as if this task can capture cerebellar alterations in individuals dependent on other substances, such as tobacco, alcohol, amphetamines, etc.

This study has limitations that have to be taken into consideration when interpreting the results. First of all, the small sample size might have compromised the power of the study, so further studies should increase the sample size and also explicitly indicate in the protocol that participants are required to refrain from cannabis use for at least 6 hours prior to the study. Regardless of the small sample size, the moderate effect size might be due to the fact that CCUs do not show any clinically significant motor deficits (SARA score equals to 0), unlike ataxic patients. Secondly, significant differences were found between groups in terms of educational level. While we propose that educational level should not be an important confounder, due to the implicit nature of the task, tobacco and alcohol use might have had an effect on performance. Indeed, even though an ordinary least-squares analysis showed inexistent correlation with performance in the present study, one should take into account for future studies that reduced adaptationmight arise due to the synergistic effects of all substances, and not exclusively of cannabis. Thirdly, even if the small differences that we found in motor adaptation are representative of cerebellar alterations in CCUs, no causation can be inferred, as the brain differences might have been preceded cannabis use. Future longitudinal studies should clarify this matter. Finally, in previous studies where participants only received end-point position feedback, the systematic aiming error accumulated slowly during the 80 trials of the RS phase. However, we opted for providing participants with continuous feedback, evaluating the angular orientation using the initial part of the trajectory. The increased level of feedback has introduced discrepancies with the literature. First, as intended, it may have accelerated (implicit) adaptation, consistently with Taylor et al. [38], where a different VMR task was used. Indeed, the drift rate in our control group (0.2◦/trial) is twice as fast as in Taylor et al. [32] (circa 0.1◦/trial). Secondly, it has produced a shift in the maximum peak drift, which here reached 27.2◦ for the control group whereas in previous experiments it remained near 10◦ [29,32]. This increased peak drift might be due to fast explicit processes [38], as suggested by the high level of adaptation already manifested in the first block of the RS phase. However, we expect to have minimized the influence of the explicit learning in our main result by reporting the drift rate, which was computed taking into account the median performance for each one of the ten blocks of eight trials of the RS phase. In other words, our measure of drift rate is not affected by the high level of adaptation which had already been reached at the first block of the RS phase. Besides, even though we report a tendency in the after-effects that is consistent with our hypothesis of decreased implicit adaptation in CCUs, the difference only reached trend-level significance (*p* < 0.09). Hence, the replication of these results with a bigger sample size may be needed to confirm that the difference in performance reported here is due to putatively-cerebellar implicit motor learning. Indeed, a possibility to avoid involving explicit processes would be using newer VMR paradigms more precisely targeted at measuring implicit learning [43].

#### **5. Conclusions**

Taken altogether, CCUs showed a moderately decreased motor adaptation compared to controls, assessed with the VMR task. This tool is easy to implement and might be able to detect subtle changes in brain function before clinical symptoms of damage occur. Future research will clarify whether the VMR task can emerge as a potential assessment tool of brain dysfunction secondary to chronic cannabis consumption.

**Author Contributions:** Conceptualization, I.H., L.M., M.B.-O., P.V.; Methodology, I.H., L.M.; Data analysis, I.H.; L.M. Investigation, I.H., L.M., C.B., L.N., K.G., B.R.B.; Writing—Original Draft Preparation, I.H., L.M.; Writing—Review & Editing, All authors; Supervision, M.B.-O., P.V., A.G.

**Funding:** This work was supported by MINECO "Retos Investigacion I+d+i", Plan Nacional project, SANAR (Gobierno de España)—under agreement TIN2013-44200-REC, FPI grant nr. BES-2014-068791, and by the European Research Council under grant agreement 341196 (CDAC).

**Acknowledgments:** The authors wish to thank all subjects participating in this study. CERCA Programme/ Generalitat de Catalunya.

**Conflicts of Interest:** LM has received honoraria from Lundbeck, outside of the work for this project. AG has received honoraria and travel grants from Lundbeck, Janssen, D&A Pharma and Servier, all outside the work for this project. The other authors declare that they have no competing interests.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **What Do Real Alcohol Outpatients Expect about Alcohol Transdermal Sensors?**

**Pablo Barrio 1,2,3,\*, Lidia Teixidor 1, Magalí Andreu <sup>1</sup> and Antoni Gual 1,2,3**


Received: 3 April 2019; Accepted: 3 June 2019; Published: 5 June 2019

**Abstract:** Objective: Little is known about the potential acceptability of alcohol transdermal sensors among alcohol-dependent outpatients in routine clinical settings. The aim of the present study was to investigate patients' attitudes towards alcohol transdermal sensors, as well as features associated with enhanced acceptability and usability. Methods: A cross-sectional survey among routine alcohol outpatients was conducted. The Drug Attitude Inventory (DAI-10) was adapted to the field of alcohol transdermal sensors for attitudes assessment. Likert-type and multiple-choice questions were used for acceptability and usability evaluation. Results: 68 patients completed the questionnaire, and the DAI-10 mean score was 3 (standard deviation (SD) = 6.5). Internal consistency revealed a Cronbach alpha of 0.613. The score of a single The score of a single Likert-type question about overall perceived value was 7.4 (SD = 2.6). Its correlation with mean DAI-10 scores was *r* = 0.633, with *p* < 0.001. Relapse prevention and a stricter treatment control from therapists were the main reported advantages. Perceived stigma was the main disadvantage. Features increasing device discretion would enhance its acceptability. Conclusions: The data suggest that transdermal sensors could play a role in the clinical treatment of alcohol outpatients and concerns regarding stigma should be taken into account. Future designs should try to minimize size and visibility and stigma concerns should be discussed with patients.

**Keywords:** alcohol dependence; transdermal sensor; attitudes; stigma

#### **1. Introduction**

Alcohol remains a first-order global health problem, with 15 million people affected in the European Union [1]. Recent publications in the United States also warn about an increase in the prevalence of alcohol use disorders during the last decade [2]. The impact it has on both individuals and society is of an enormous dimension, both medically and economically [3,4]. A significant share of these consequences is attributable to the severest form of alcohol use, i.e., alcohol dependence, which is currently considered a chronic disease, with a relapsing–remitting nature. Despite some controversies in the field, abstinence has been the prevailing therapeutic goal in most of the existing settings, being considered the safest and most efficient pathway to early recovery [5,6]. Therefore, for a great majority of professionals dealing with alcohol dependence, monitoring abstinence becomes an indispensable task. Traditional tools for abstinence assessment have consisted of patients' self-reports and alcohol biomarkers. Despite recent improvements, especially in alcohol biomarkers [7,8], there remain some relevant limitations that could be overcome with the use of alcohol transdermal sensors.

Interestingly, in harm reduction or controlled drinking paradigms, alcohol transdermal sensors could also be of special utility, given the importance of quantifying the amount of alcohol ingested when patients chose to reduce their intake instead of abstaining. In quantifying the amount of alcohol consumed, transdermal sensors seem to be the most sensitive tool. In that sense, they could potentially be the ideal partner to the paradigm of heavy use over time [9], where the emphasis is put quantitatively on the amount of ethanol ingested rather than discretely qualifying patients as abstinent or not. This may suggest that transdermal sensors could be seen as a valuable tool to facilitate and improve digital phenotyping in addiction patients [10,11].

Transdermal sensors have been available for some years now. Their main use has been within the justice system [12]. There are also several studies in the literature assessing the devices' validity and functioning, and there have already been randomized trials to evaluate their efficacy and patients' experiences [13–17]. Together, they suggest that transdermal sensors could be effective in helping alcohol patients. They also point out towards good acceptability and feasibility. However, no study has been conducted in real practice, or everyday settings, so little is known about patients' acceptability and attitudes towards such devices in the routine, daily clinical setting of patients undergoing routine treatment with no monetary compensation or legal imperatives. Since they offer an ongoing, 24 h measurement of alcohol through sweat, it could be hypothesized that patients could see them under an autonomy-restricting perspective. However, this has not been evaluated in a formal study. Therefore, it seems important, in order to better understand and anticipate patients' acceptance toward these devices, to assess their attitudes towards alcohol transdermal sensors. Ideally, gathering end-users' perspectives and preferences should facilitate device implementation in routine settings, and therefore enable maximum reach among real patients.

The aim of the present study, conducted among patients who have never used a transdermal device, was therefore two-fold: firstly, to investigate patients' attitudes towards alcohol transdermal sensors, as well as their perceived utility; and secondly, to investigate what features of transdermal sensors would enhance patients' acceptance and usability.

#### **2. Methods**

#### *2.1. Study Design and Subjects*

We performed a cross-sectional survey among alcohol-dependent patients attending the outpatient service of the Addictive Behaviors Unit of the Clinic Hospital of Barcelona. Subjects were specifically recruited among those attending the routine urine screening program. Ethics consent was granted from the Hospital Clinic Ethics Committee. Informed written consent was obtained from all participants prior to participation in the study.

#### *2.2. Instrument*

For this study, a specific questionnaire was designed and all questions were in Spanish. The questionnaire consisted of 3 main parts. The first was aimed at evaluating patients' perceived utility and preferences regarding transdermal sensors. It consisted of 3 multiple choice and 7 Likert-type questions. The second part was designed to obtain patients' beliefs and attitudes towards alcohol transdermal sensors as part of their treatment. Given the lack of similar research for this specific subject, we took advantage of the extensive literature regarding the Drug Attitude Inventory-10 [18], which was initially designed to test attitudes of patients with schizophrenia towards medication in order to correlate it to medication adherence. It consists of ten items; each one is scored with either 1 or −1, depending on whether the response signals a positive attitude towards medication or not. To avoid response bias, half of the items are worded positively, and half negatively. We therefore kept the same wording and the same order of the questions. Its score ranges from −10 to 10. We adapted its ten items to the present study, replacing the concept of medication by that of alcohol transdermal sensors. While such an adaptation of the Drug Attitude Inventory (DAI-10) has not been previously validated in the literature, two previous studies with alcohol patients (one cohort attending Alcoholics Anonymous groups and another cohort obtained from the same outpatient population of this study) performed a

similar adaptation of the DAI-10, showing good psychometric properties [19,20]. In addition, we still considered it a good approach to capture patients' attitudes, given the lack of more specific, validated instruments for our study aims. The third was devised to gather basic sociodemographic characteristics. At the beginning of the questionnaire, pictures of the existing transdermal devices were printed (the WristTAS and the SCRAM). Patients received no monetary compensation for participation in the study.

#### *2.3. Procedure*

The professional responsible for receiving patients and their urine specimens was an experienced nurse in the addiction field. Within a four-week period at the beginning of 2018, while the nurse received patients, patients were offered the option of participating in the questionnaire. To be included, patients had to be adults diagnosed with an alcohol use disorder and be attending the urine screening program. The main exclusion criteria were the presence of cognitive decline (either based on the nurse's clinical judgment or patient history) and being in a state of intoxication. Patients with any other condition that, in the opinion of the investigators, could have compromised the validity of responses were not offered participation (e.g., an altered psychopathological state). Patients were reassured that all data provided would be kept totally anonymous. Once completed, questionnaires were kept safe until the end of the study, at which point they were analyzed.

#### *2.4. Statistical Analysis*

A descriptive analysis of sociodemographic data was conducted. The mean and the standard deviation were used for continuous variables; percentages were used for qualitative variables. Regarding the adapted version of the DAI-10, an internal consistency analysis was carried out with Cronbach's alpha. Concurrent validity was assessed via bivariate correlations between the DAI-10 total score and a 0 to 10 Likert scale measure assessing the potential perceived value of alcohol transdermal sensors as part of patients' treatment. Finally, in order to evaluate if attitudes were influenced by any specific variable, we conducted a lineal regression model with DAI-10 scores as the dependent variable and age, sex, level of instruction, therapeutic objective, and length of urine testing as predictors. Analyses were conducted with SPSS (IBM Corp. Released 2015. IBM SPSS Statistics for Windows, Version 23.0., IBM Corp, Armonk, NY, USA).

#### **3. Results**

During the study period, a total of 110 patients attended the urine screening program. Of that, 80 subjects were offered participation in the study (reasons for exclusion: intoxication of 11 patients, cognitive decline of eight patients, other reasons were not specified for 11 patients). A total of eight patients declined participation and four subjects accepted but did not return a questionnaire. That left a total of 68 patients completing the questionnaire. The mean age of the sample was 52.9 years (standard deviation (SD) 12.3). The 66% majority were men, and 39% of the sample were currently employed. Regarding education status, 14% had primary education, 28% had secondary education, 28% had a technical qualification, and 30% had a university qualification. Most of the patients (72%) underwent screening twice a week, 20% once a week, and 8.2% less than once a week. Most of the patients (78%) reported abstinence to be their therapeutic objective. Only a small minority (18%) reported drinking reduction as their aim. Patients had been attending the screening program for an average of 11 months (SD 10.5).

The main advantages and disadvantages attributed to transdermal sensors by patients and their frequencies can be seen in Table 1. Three patients spontaneously reported a decrease in human contact as a potential downside of transdermal sensors. Factors associated with enhanced acceptability and its rating from patients in a scale from 0 to 5 can be seen in Table 2. For both tables, a sex-specific analysis was also conducted. The main difference observed was for stigma, women presenting almost a 2-fold percentage of response compared to men.


**Table 1.** Advantages and disadvantages attributed to transdermal sensors.

**Table 2.** Features associated with enhanced acceptability (rated from 0 (lowest) to 5 (highest)).


In a 0 to 10 scale, patients reported a mean score of 6.6 (SD 3.4) regarding the acceptance of a transdermal sensor if their therapist prescribed it. Regarding the adapted DAI-10 questionnaire, the mean score was 3.0 (SD 6.5). Internal consistency, measured with Cronbach's alpha, was 0.613, indicating fair reliability. The question about the overall perceived value of alcohol transdermal sensors showed a mean of 7.4 (SD 2.6). The correlation between this and the DAI-10 score, as a means to investigate concurrent validity, was *r* = 0.633, with *p* < 0.001. The regression model revealed no significant predictor of DAI-10 scores. The differences according to sex regarding the percentage of positive responses to each DAI-10 item can be seen in Figure 1.

**Figure 1.** Percentage of positive responses to the Drug Attitude Inventory (DAI-10) questionnaire for each item and according to sex.

Table 3 reports preferences regarding different kinds of devices for transdermal sensors. As can be seen, almost all patients preferred a device similar to a nicotine patch (63%), while the rest of the options were far less frequently preferred.


**Table 3.** Preferred devices for transdermal sensor implementation.

#### **4. Discussion**

In this study, we aimed to investigate patients' attitudes towards alcohol transdermal sensors. Globally, we believe the results suggest that patients perceive alcohol transdermal sensors as a potentially useful and valuable treatment device.

The adapted version of the DAI-10 scored above average among our sample. Similarly, the overall perceived value as rated in a 0 to 10 scale was high (7.4), and the correlation between both was rather high and clearly significant, suggesting a good concurrent validity. Additionally, the internal consistency of the adapted questionnaire, measured with Cronbach's alpha, was fair. These results suggest the tentative conclusion that the results of the adapted DAI-10 questionnaire could be considered to be reasonably valid.

Looking at patients' responses, the majority stated that preventing relapse and accomplishing their therapeutic goals regarding alcohol were the main functions of transdermal sensors. What is also interesting to note is the significant proportion of respondents that reported "showing professionals or their family that they do not drink" as a main function of transdermal sensors. It suggests that transdermal sensors could play an important role in patients' interaction with both professionals and their social network.

In line with the most frequently reported inconvenience of transdermal sensors (i.e., stigma associated with wearing it), the features that could most enhance the device's acceptability were precisely those that seemed to be designed to reduce the device's visibility (small size and discreteness). Further supporting these findings, the most preferred physical device was a patch similar to the ones used for nicotine replacement therapy. In line with our findings, previous reports about transdermal sensors have also emphasized stigma and embarrassment as major concerns reported by patients [21,22]. What is also worth mentioning is that patients' responses suggest that providing continuous feedback about the information gathered by the device is also essential in order to enhance patients' acceptability and usability.

The results from the previous survey by Alessi and colleagues [17], which was conducted among patients who wore a real sensor, suggest that stigma and embarrassment were not major concerns. This is in sharp contrast to our results. The difference could be explained by the different sample selection and study design procedures, since all patients in the aforementioned survey were gathered from studies conducted with contingency management procedures, and the majority of subjects belonged to non-clinical samples, suggesting a potential selection bias. For example, in one of the studies patients were heavy drinkers not seeking treatment. That means that social desirability and the wish to abstain might have been quite different in both samples. Another potential explanation could be the fact that patients wore the SCRAM as a foot bracelet. This could be considered a discrete device and therefore stigma concerns would have been reduced.

Looking at patients' responses, it seems that physical discomfort is also a main preoccupation among patients. It is probable that similar recommendations to the ones for dealing with stigma and embarrassment could be given. These were reduced size and visibility, which should also increase comfort with the device.

Overall, and in line with the increasing importance that patient-centered care is gaining in medical settings [23], we expect this study to offer relevant insights into patients' perspectives, motives, and attitudes towards alcohol transdermal sensors. We expect knowledge about patients' needs and worries to be increased, so that when, in the near future, they are offered to incorporate transdermal sensors into their treatment program, they can be offered this in a more realistic, friendly, and efficient manner.

Turning to the limitations of the present study, a key issue in interpreting all studies using questionnaires is social desirability [24]. Addiction itself is especially prone to such bias [25,26]. Therefore, although the questionnaires were completely anonymous, the presence of a social desirability bias cannot be totally ruled out.

Other relevant limitations should also be taken into account when interpreting our findings. First, we developed a new questionnaire. Although it was based on an extensively validated one (the DAI-10), it must be acknowledged that our study was not focused on questionnaire validation; therefore, more measurements could have been obtained in order to better validate it. That being said, the internal reliability and the concurrent validity were fair. Another important limitation stems from the fact that all patients belonged to a single outpatient center, a fact that might diminish external validity. Similarly, all patients were recruited from the urine screening program, so the question remains as to whether similar or different results would be observed in other clinical settings. It is also important to mention that the nature of this study was mainly descriptive. However, we tried to find some predictors of patients' attitudes towards transdermal sensors, but found no significant predictors, a fact that could partly be due to insufficient statistical power. Also relevant is the fact that no systematic screening instrument for the detection of cognitive decline or other excluding conditions was used. The clinical nature of the study also precluded an extensive evaluation of patients' clinical characteristics. Finally, four questionnaires were not fully completed and, therefore, we had a minor proportion of missing data, which we excluded from analysis.

#### **5. Conclusions**

Though our data suggest that concerns about stigma and embarrassment should be taken into account when planning the implementation of transdermal sensors in routine settings (more notably for female patients), our findings, together with those of previous studies, encourage further efforts to bring transdermal sensors to routine, clinical samples of alcohol-dependent patients.

Future designs of transdermal sensors should aim at reduced size and visibility, thus lowering stigma concerns. Patients should also be allowed full access to the information provided by the sensor.

**Author Contributions:** Conceptualization, P.B. and A.G.; Data curation, M.A.; Formal analysis, P.B.; Investigation, L.T.; Supervision, A.G.; Writing—review & editing, M.A. and A.G.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Article*
