Background

The combined use of multiple substances is common among young people in Europe (European Monitoring Centre for Drugs and Drug Addiction [EMCDDA, 1]. Alcohol, cannabis, and tobacco are the drugs most frequently misused by youngsters aged 18–25, and simultaneous use of several substances significantly predicts other risky behaviors (e.g. [2,3]). Polyconsumption exacerbates problems derived from substance use [4–7], and recent studies show gambling-related problems to be predicted, among several regressors, by being male and by dependence scores for alcohol, tobacco, and marijuana [8–11]. In other words, comorbidity of addictive behaviors comprises both substance and non-substance-related problems.

Except for gambling disorder, however, no consensus exists regarding the conceptualization of putative behavioral addictions as bona fide addictive disorders. In the last edition of DSM (DSM5), the category substance-related and addictive disorders includes a subcategory for substance use disorders and another one for behavioral addictions. However, the latter, so far, only includes gambling disorder, yet none of the other putative behavioral addictions proposed in the literature. Gambling disorder is thus currently present in the last versions of both the DSM [12] and the ICD [13] as a behavioral addiction, and so it is gaming disorder in the ICD-11 (but not in the DSM-5). Problematic internet use, compulsive buying/shopping, eating/food addiction, and hypersexuality are not currently classified as addictive disorders in any of the main psychiatric nosologies, in spite of which they are frequently conceptualized as addictive by part of the scientific community and the general public [14–16]. Still, and regardless of whether they qualify or not as addictive disorders, there is little doubt that some types of over-engagement in these activities can have a negative impact on wellbeing, health, and social and occupational functioning. In other words, although addiction is still a disputed term in these domains [15], there is some consensus that some of these behavior patterns can become problematic and thus deserve clinical attention [17–19].

The pertinence of exploring co-occurrence of drug abuse with other addictive/dysregulated behaviors not involving the use of drugs is reinforced by recent evidence. For instance, according to some studies (e.g. [20,21]), women with disordered eating and compulsive buying are more likely to have used and misused substances. And some level of co-occurrence of problematic video gaming and substance use also seems to exist, especially in young males [3,10].

For some authors, there are important neurocognitive similarities between excessive involvement in some of these activities and diagnosable addictive disorders (substance use and gambling disorders) [22–25]. These shared mechanisms–including weakened top-down control and executive mechanisms, altered reward processing and sensitivity, attentional biases and cue reactivity, and emotion dysregulation [15,26]–would justify the interest in exploring patterns of co-occurrence across behavioral domains. In other words, a path to better understand these conditions is to carefully quantify their degree of overlap with well-established addictive disorders. However, to date, research has tracked coincidences more closely than potential divergences between substance and non-substance related problems (and between different putative behavioral addictions). In this regard, previous studies have unveiled a distinctive link of problematic gaming and other sedentary leisure activities with cannabis use [3,10]. This link, in turn, seems to be underpinned by personality and individual differences factors that can be dissociated from the ones responsible for a more general and well-known overlap between addictive behaviors [10]. The corroboration of these associations while controlling for relevant confounders is, however, still pending.

Gender can also play a role in this pattern of associations, and gender differences regarding these problematic behaviors have been reported. For example, binge eating and other eating disorders, as well as compulsive buying/shopping are more prevalent in females, whereas gambling and gaming disorder, excessive internet use, hypersexuality, and substance use disorders are more prevalent in males [27–30]. However, the modulating role of gender in associations between different addictive/dysregulated behaviors remains underexplored. Different patterns of co-occurrence across genders would imply that gender-related traits and risk factors underlie such associations, and, consequently, that etiological paths to such behavioral problems (and interventions to prevent or treat them) should also be gender-informed.

The present study

Here, we focus on cannabis use problems (as measured by the CAST and SDS instruments [32,33]) and their independent associations with an array of non-substance-related behavioral problems (problematic gambling, excessive Internet and video games use, compulsive buying, dysregulated eating, and hypersexuality), measured with the MultiCAGE CAD-4 instrument [34].

The operational definition of non-substance-related problematic behaviors is thus tightly linked here to the instrument used to measure them. The MultiCAGE tool extends the CAGE screening (for risk of alcohol use disorder in primary care settings) [35] to other putative addictive disorders. In all subscales, items refer to subjective perception of the problem, perception by relevant others, feelings of guilt, and lack of control or abstinence symptoms. Criterion validity has been established for alcohol and substance use subscales. For the other subscales, criterion validity has not been established, and a higher score is interpreted as indicating a higher risk of suffering from a condition requiring clinical attention. In terms of delineation of problematic behaviors, the alcohol, illegal drugs, and gambling subscales measure the risk of suffering from substance use and gambling disorders as defined in main psychiatric classifications. Definitions of problematic Internet use, problematic video gaming, compulsive shopping/buying, and hypersexuality mostly overlap with the definition of behavioral addiction according to the components model. Eating problems are defined in a more heterogeneous manner, as they can equally capture restrictive eating behavior, binging, and purging, i.e. they measure eating problems in a way that cannot be equated to eating/food addiction.

Regarding substance use, cannabis-related problems were singled out for the present study for practical and theoretical reasons. Beyond the evidence that cannabis is addictive and thus cannabis abuse presents many of the characteristics observed in other addictive disorders [36,37], cannabis is the most frequently used illegal drug in western countries [1], and has been previously linked to sedentary leisure activities, and, particularly, to digital media use [3,10,38,39]. Therefore, there are at least two reasons why cannabis-related problems are expected to be linked to non-substance-related dysregulated/addictive behaviors. On the one hand, substance-related problems are known to be located at the externalizing end of the externalized-internalized behavior continuum, along with gambling disorder, hypersexuality, and at least some patterns of dysregulated eating [40–45]. And, on the other, the link between cannabis use and sedentary leisure activities could reveal potential shared reward, motivation, and personality-related mechanisms that could be relevant to understand problematic Internet use and video gaming. As noted earlier, a privileged association between excessive video gaming and cannabis use, and the role of personality dimensions therein, have been previously reported [3,10].

Validated scales were used to assess cannabis abuse (CAST; [46]) and severity of dependence (SDS; [47]), and these were separately used as predictors of risk scores of dysregulated behaviors as measured by the MultiCAGE CAD-4 screening (excessive gambling, excessive video gaming, excessive internet use, compulsive buying, dysregulated eating, and hypersexuality), while controlling for relevant confounders. Importantly, the large sample size allowed to take gender into consideration for analyses, and thus to test differences between male and female participants in all measures of interests, but also for different association patterns between cannabis CAST/SDS and MultiCAGE scores across genders. To our knowledge, this is the first attempt to test these associations using comparable instruments for all potentially dysregulated non-substance-related behaviors, while considering the role of gender. Although, in view of the antecedents reviewed earlier, positive associations are expected, especially for tech-related sedentary behaviors, our hypotheses regarding the specific pattern of elevated risks remain open (and thus analyses exploratory).

In summary, to our knowledge, the associations between cannabis use and a sufficiently wide-ranging array of non-substance-related dysregulated/addictive behaviors have never been explored, either by themselves or in relation to gender. The main aim of this study was thus to explore such associations, using comparable measurements for different problematic behaviors, and the role of gender therein. The results yield both theoretical and practical relevance. On the one hand, a precise characterization of the differential associations between non-substance addictive or dysregulated behaviors and cannabis-related problems (including a potential replication of the association between cannabis-related problems and problematic use of the Internet and video games) can help identify separable etiological mechanisms for conditions that are frequently conceptualized and treated as comparable. On the other hand, and beyond theoretical considerations, given the large social acceptance and high rate of use of cannabis among youngsters, a better depiction of its relationships with other mental health hazards is valuable to understand its potential risks, and to design better prevention tools.

Participants

The target population for this study were the college students at the University of Granada, Spain. 856 participants from a population of 47096 were recruited using probabilistic stratified sampling from 23 Degree programs [Psychology, Speech therapy, Tourism, English, History, Literature, Business management, Economics, Biology, Physics, Optics, Teacher training (primary education), Teacher training (early childhood education), Education sciences, Law, Medicine, Pharmacy, Social work, Political sciences, Sociology, Computer science, Civil Engineering, and Telecommunications engineering]. Mean age of the sample was 21.12 years (SD = 7.23). 37.62% (322) reported their gender was male, and the rest, female. No participants reported to have a gender identity other than male or female. Participants voluntarily filled all questionnaires, as well as a form with their sociodemographic information in a pen-and-paper format during a break between lectures. Instructions were provided by the second and fourth authors (both of whom are experienced in psychological assessment, and stayed in the room during the whole session and watched it to ensure participants behaved as instructed). All participants were informed about the aims and procedure, and about the possibility of withdrawing from the study at any time. All participants provided a written informed consent. The study was approved by the Human Research Board of the University of Granada. The instruments described below were applied to all participants.

MultiCAGE CAD-4 [34]

This questionnaire is an extension of the CAGE risk screening for alcohol abuse [48], developed to estimate the risk of putative addictive disorders in several behavioral domains (gambling, buying, alcohol use, illegal drugs use, hypersexuality, internet use, video gaming, and eating behavior). According to the original validation criteria, meeting 0 criteria is interpreted as absence of risk, 1 criterion indicates detectable but low risk of problems, and meeting 2 or more criteria is interpreted as indicating a high or very high risk of suffering a clinically significant condition.

The original tool was validated in primary care settings with members of the general population, that is, with people attending public healthcare centers for any reason, not necessarily related to the problematic behaviors assessed. A version of the scale developed specifically for putative technological addictions (including the 8 items for Internet and video games of the version used in this study) was also validated in the general population, outside the primary care setting, and using a snowball sampling method. The Internet and video games subscales of this version yielded psychometric properties very similar to the ones of the original scale [49].

The MultiCAGE CAD-4 has also yielded good reliability for all the subscales, as well as very similar within-subscale factor loadings for all items. Item interchangeability is strongly suggestive of the existence of a unique continuous construct underlying each of the subscales. Reported Cronbach α values were 0.84, 0.73, 0.88, 0.70, 0.82, 0.79, 0.79, and 0.73, and 0.76, for alcohol, gambling, illegal drugs, eating, Internet, video games, buying, and sex subscales, respectively [34]. To our knowledge, criterion validity has been established only for the substance use scales, not the behavioral scales. A cut-off score of 2 was observed to have 92.4% diagnostic sensitivity for alcohol use disorder, 100% for heroin and cannabis use disorder, and 94.1% for cocaine use disorder [34]. Among the non-substance related scales, Internet and video games related problems have been observed to weakly but significantly correlate with executive functioning, social behavior, and emotional control problems, as well as with general mental health [49]. Only non-substance subscales will be used as variables of interest in the present study.

Fagerström test for nicotine dependence [50]

This instrument measures the intensity of physical addiction to nicotine, and consists of 6 items evaluating the amount of cigarettes smoked, compulsion, and other signs of dependence to nicotine. 4 of these items have two response options and these are recorded as 0/1, the other 2 items have 4 response options, scored 0–3. Item-by-item scores are summed, and the total score is interpreted as a measure of nicotine dependence (< 4: minimal, 4–7: moderate, and > 7: high). In a validation study, Fagerström scores were significantly correlated (0.33 and 0.42) with plasma cotinine levels. Cotinine was used in this validation study instead of nicotine, as it is relatively insensitive to the immediate effects of smoking and constitutes a more stable measure of chronic intake [51]. The Spanish version used here yielded an acceptable Cronbach α value of 0.66 [52].

Severity of Cannabis Dependence Scale (SDS; [53])

Severity of cannabis dependence was measured using the Spanish version of the SDS [47]. This questionnaire consists of 5 Likert-type items, each in a 0–3 scale. Total score ranges between 0 and 15. Psychometric properties of both the English and the Spanish version are good [47,54], with a Cronbach α of 0.82, and an ICC coefficient of 0.83 for the Spanish version. A cut-off of 3 is normally used to define moderate cannabis dependence, and 7 for severe dependence [46].

Cannabis Abuse Screening test (CAST; [55])

The CAST is a screening tool for cannabis abuse in the general population that has also been proven valid and reliable for adolescents and young adults [32,47,56]. It consists of 6 items with 5 response option (from 0: never, to 4: very often), referred to the last 12 months. The total score is computed as the sum of the item-by-item responses. The CAST has been validated in adolescent and adult samples [47,57,58] and shows good reliability (AUC = 0.82). Cut-offs of 3 and 7 are used to detect moderate and severe cannabis addiction.

Statistical analyses

First, gender differences in dependence and abuse of cannabis and in signs of non-substance addictive/dysregulated behaviors were explored. SDS and CAST scores were treated as continuous non-normally distributed variables, and nonparametric Mann-Whitney’s U tests were used to compare male and female participants.

MultiCAGE subscale scores range between 0 and 4, depending on the number of items endorsed. Separate scores were thus obtained for the different behavioral domains under scrutiny. The small range of values precludes using MultiCAGE subscales as continuous variables, so their ordinal nature was retained when testing gender effects. Moreover, given that our sample consisted of college students (not exclusively of patients or high-risk individuals), some subscales yielded very low observation frequencies for high scores. In view of that, and in order to ensure a sufficient number of observations per level, high scores were collapsed in a single level. More precisely, individuals in each subscale were classified as without risk of problems (MultiCAGE = 0), with low but detectable risk (1), and with high risk of problems (2–4). As noted above, the high-risk cutoff has only been validated for substance-related MultiCAGE subscales, not for purely behavioral ones. Labels for the latter have been established by mere analogy and must be interpreted with caution. This, however, does not affect to the ordinal nature of the 0–2 scale resulting after collapsing high values. The association between gender and risk for each potentially problematic behavior was estimated using Goodman and Kruskal’s γ ordinal correlation index. This set of analyses was performed in JASP statistical software [59]. Please note that this analysis in particular could have been affected by a reduction of sensitivity of the scale to differences in the high end, so it is important it is confirmed by further analyses (in which all values were retained, and modelled using a Poisson distribution).

To test the associations of CAST and SDS scores with the several MultiCAGE non-substance scales, two sets of regressions were run. As noted above, MultiCAGE subscale scores are discrete (0–4), with most observations being 0, and very few 3 and 4-score observations. In view of this, the R package lme4 [60] was used to model MultiCAGE scores as Poisson-distributed (i.e. as a count of discrete items endorsed; no collapsing was required in this case). For each MultiCAGE score, a mixed-effects model was built, with the Degree program the participant was enrolled at as the only random-effects factor, age, gender, Fagerström score, and MultiCAGE alcohol score as fixed-effects covariates, and cannabis dependence/abuse (either SDS or CAST total scores) as main fixed-effects input variable. The effects of covariates, including nicotine dependence (Fagerström score) and risk of alcohol use disorder (MultiCAGE–alcohol, i.e. CAGE) will be estimated and reported, and are included in the models to ensure the effects of cannabis problems/dependence, which are the main predictors of interest, are not explained away by concurrent use of other substances. Although comparing the association patterns of non-substance-related problem with different substance-related problems is not the main aim of the present study, differences in such patterns (and what they could imply for further research) will be mentioned and briefly discussed.

To test the role of gender, initial saturated models also included the interactions of gender with SDS/CAST, Fagerström score, and MultiCAGE alcohol score. Each of these interactions was retained in the final regression model only if substantially contributed to model fit, according to a hierarchical comparison (using the Akaike Information Criterion and a χ² test). For the sake of readability, only statistics for predictors included in the final models will be reported here. Incidence rate ratios (IRR) were computed for SDS/CAST effects, and significance was determined by family-wise Bonferroni-corrected p < 0.05 (p_corrected = 0.05/x, with x representing the number of contrasts in the hypothesis-relevant family). For gender–MultiCAGE correlations, the number of tests to be performed was 6 (one gender effect per MultiCAGE subscale), and the significance threshold was established at p_corrected = 0.0083. For SDS/CAST effects, the number of tests considered for family-wise correction will be those of theoretical interest (CAST and SDS direct effects and their interactions with gender in the best-fitting models). As detailed later, family-wise correction yielded in this case a threshold p_corrected = 0.0035.

Data and code for these analyses can be accessed at the Open Science Framework website (https://osf.io/2jqnh/).