Exploring the Link between ADHD and Cannabis Use in Swedish Ninth Graders: The Role of Conduct Problems and Sensation-Seeking

Abstract Background: Attention deficit hyperactivity disorder (ADHD) has in several studies been linked to substance use, including cannabis use. However, crucial gaps remain regarding how to understand this association. Analyzing the association between ADHD and substance use is complicated because of a pronounced overlap between ADHD, conduct problems, and traits such as sensation-seeking. Objectives: Using data from a large and nationally representative study among Swedish adolescents, this study explored the role of conduct problems, but also of sensation-seeking, in accounting for the association between ADHD and cannabis use. Results: There was a notable association between ADHD and cannabis use that was attenuated when conduct problems were controlled for. The association between cannabis use and conduct problems, in turn, was attenuated when sensation-seeking was controlled for. Individuals with both ADHD and conduct problems were more likely to have used cannabis than individuals with ADHD only, but not compared with individuals with conduct problems only. Conclusions: Whereas conduct problems largely explain the link between ADHD and cannabis use, sensation-seeking seems to account for the association between conduct problems and cannabis use.


Introduction
Attention deficit hyperactivity disorder (ADHD) is defined as a neurodevelopmental disorder characterized by inattention, hyperactivity, and impulsivity (Epstein & Loren, 2013). ADHD is associated with a large number of negative consequences. For example, an ADHD diagnosis in childhood predicts poorer outcomes more than 30 years later, in social, occupational, and economical domains, and a higher prevalence of substance use disorders has also been identified in this group (Klein et al., 2012). Higher levels of ADHD symptoms in childhood predict early initiation and more use of alcohol and marijuana in adolescence (Elkins et al., 2018) and adulthood (Howard et al., 2015).
This study explores the link between ADHD and cannabis use in adolescence. Though the relationship between ADHD and substance use is widely researched (e.g., Heradstveit et al., 2022;Lee et al., 2011;Yoshimasu et al., 2016), a number of questions remain unresolved . This study specifically addresses the role of conduct problems (CP) and sensation-seeking (SS) in accounting for the association between ADHD and cannabis use. Precisely how CP and ADHD are related to substance use remains unresolved (Heradstveit et al., 2022;Malone et al., 2010) and there are few studies on how ADHD, CP, and SS relate to substance use (Moggi et al., 2020;Van Eck et al., 2012). Though ADHD, CP, and SS have played central roles in substance use research for a long time, we need to know more about how each of these factors is associated with substance use, considering the concurrent contributions of the others.
We take advantage of a large and nationally representative Swedish population study on alcohol and other substance use among adolescents attending ninth grade (i.e., 15-16 years old). Nationally representative data are crucial for understanding the association between ADHD and substance use, as clinical data on this association are not generalizable to the entire youth population (Lynskey & Hall, 2001). Compared with the relationships of ADHD and alcohol, tobacco, or generic measures of illicit drug use, the relationship between ADHD and cannabis use remains somewhat underexplored (Sibley et al., 2014). Also, much of the available research on the ADHD-cannabis link is from the US, where the prevalence rates of youth cannabis use are much higher than those in Sweden (Carliner et al., 2017). The marginalization of cannabis use in Swedish society (Ekendahl et al., 2020) potentially means that the link between ADHD and conduct problems with cannabis use may be more robust than evidenced in other settings. In line with this, prior research shows that social and behavioral risk factors are more strongly related to cannabis use in lower prevalence countries than in higher prevalence countries (Sznitman et al., 2013). Here, we are particularly interested in how ADHD may be associated with cannabis use during a developmental period in adolescents' lives when the prevalence of cannabis use is significantly lower than a couple of years later (CAN, 2021). It is imperative to assess this relationship quite early during adolescence given the number of negative consequences that have been attributed to early-onset substance use (e.g., Poudel & Gautam, 2017).

ADHD, conduct problems, sensation-seeking, and substance use
There is a pronounced overlap between ADHD and related conditions such as conduct disorder and oppositional defiant disorder on the one hand (see , and between CP and substance use on the other (Heron et al., 2013;Roy, 2008). This makes it uncertain whether ADHD is actually associated with substance use, and studies including CP have yielded inconclusive findings (see Malone et al., 2010). The term CP refers to a heterogenous collection of behaviors that violate norms in social and legal domains, including, for example, destroying property, stealing, and fighting (Hinshaw, 1987). The heterogeneity also includes different developmental trajectories (Bevilacqua et al., 2018). ADHD and CP are considered conceptually distinct (Hinshaw, 1987;Waschbusch, 2002) and ADHD was, in the DSM-5, redefined as a neurodevelopmental disorder rather than a disruptive one (Doernberg & Hollander, 2016), further underscoring this distinction.
Nonetheless, the borders between ADHD and CP are fuzzy and create analytical challenges. For instance, the association between ADHD and substance use may be completely or partly due to conduct disorder (Lee et al., 2011). Several studies suggest that ADHD is not related to substance use outcomes (see, e.g., Fergusson et al., 2007;Flory & Lynam, 2003) or only weakly related thereto (Yoshimasu et al., 2016) when adjustment is made for CP. Further, it has been asserted that ADHD plays only a minor role in explaining substance use (Lynskey & Hall, 2001). A meta-analysis focusing specifically on illicit drug use found no significant association between ADHD and use when controlling for CP (Serra-Pinheiro et al., 2013). However, in another meta-analysis (Groenman et al., 2017), ADHD was associated with substance use disorder even when cooccurring CP were considered. Thus, there are conflicting findings even at the level of meta-analyses. The first research question addressed in this study is therefore to what extent the association between ADHD and cannabis use is accounted for by CP.
Moreover, available research suggests that CP may be best conceptualized as a mediator of the association between ADHD and substance use (e.g., Brook et al., 2010;Fergusson et al., 2007;Klein et al., 2012;Sibley et al., 2014; see also Waschbusch, 2002). If the association between ADHD and cannabis use is attenuated when CP are controlled for, we have evidence of mediation rather than of confounding (MacKinnon et al., 2000). The interpretation of this is not that ADHD is an unimportant correlate of cannabis and substance use, but rather that CP may at least in part "explain" the ADHD-cannabis link.
To the extent that ADHD and CP have independent associations with cannabis use, they should predict cannabis use over and above what ADHD and CP do separately. Individuals with combined ADHD and CP would also have poorer substance use outcomes than those with only one of these conditions (see Flory & Lynam, 2003;Roy, 2008 for reviews). For instance, August et al. (2006) found that individuals with both ADHD and externalizing problems had more severe substance use outcomes than those with ADHD only and controls, and that the ADHD-only group did not differ from controls. Molina et al. (1999) found that the combined group had higher levels of substance use than the group with CP only, but also that the ADHD-only group did not differ from controls in substance use outcomes. In another study, the authors found that the prevalence of substance use (including past six-month marijuana use) was highest in the combined group, as compared with ADHD only, CP only, and neither of these (Molina & Pelham, 2003). Moreover, an early meta-analysis found that the combined group may have more CP than CP-only or ADHD-only groups and that the same pattern held for hyperactivity and inattention problems, though the evidence was less strong in that case (Waschbusch, 2002).
The reason for the elevated risk of substance use in the combined group may be due to more severe conduct and ADHD symptoms than in other groups. However, other, more recent research has failed to find any real differences in substance use problems between those with ADHD and CP and those with CP only (Yoshimasu et al., 2016). Contrary to the studies above, Yoshimasu et al. (2016) found substantially elevated odds of substance use problems among individuals with ADHD only compared with those with neither ADHD nor CP. Given these conflicting findings, a second research question to be addressed in this study is whether the presence of both ADHD and CP in adolescents is more strongly related to cannabis use than the presence of either of these conditions in isolation.
A third research question is whether SS may account for the relationship between ADHD, CP, and cannabis use. SS has been defined as "the seeking of varied, novel, complex, and intense sensations and experiences, and the willingness to take physical, social, legal, and financial risks for the sake of such experience" (Zuckerman, 1994, p. 27). It is considered a facet of the personality trait impulsivity (see e.g., Stautz & Cooper, 2013) and as such has close links to ADHD. However, it is conceptually distinguished from both ADHD and CP, as the latter two are considered disorders, not personality traits (Moggi et al., 2020). SS has been linked to cannabis use net of ADHD, CP, other personality factors, and background factors such as sex and age (Walther et al., 2012). "Novelty-seeking" is higher in individuals with ADHD than in controls (Purper-Ouakil et al., 2010), and a study of college students found that SS mediated the association between ADHD and risky behaviors (Graziano et al., 2015). Other research has established a link between SS and CP (Martin et al., 2004). Thus, SS appears to be intertwined with both ADHD and CP, and understanding its role in accounting for the links between ADHD, CP, and cannabis use is of importance. To the best of our knowledge, only a few studies have explored how these three constructs are related to substance use, of which one was carried out among males only (Moggi et al., 2020) and one addressed stimulant use only (Van Eck et al., 2012). Our study contributes to the scarce research in this field.

Materials and methods
The data used in this study are from the baseline of Futura01, a recently initiated Swedish cohort study of people born in 2001 (see e.g., Sjödin et al., 2021). This nationally representative sample covers approximately 5% of the entire Swedish birth cohort that year (n = 5549). The data were collected during school hours in 2017, when participants were in ninth grade (15-16 years old). Students filled out a paper and pen questionnaire during school hours under exam-like conditions. In a first step, a random sample of 500 schools from all of Sweden was randomly selected by Statistics Sweden. Schools were sampled with a probability proportional-to-size (PPS) sampling design, where a school's inclusion probability was proportional to its number of ninth-grade students. The participation rate for schools was 68.6% (n = 343). One class per school was then selected using a PPS sampling procedure and all students in the selected class who were present on the day of the survey were asked to participate in the study. The study was approved by the Ethical Review Board of Stockholm (number 2017/5.2) and by the Swedish Ethical Review Authority (number 2020-01272). The primary focus in Futura01 is to track the long-term consequences for young people growing up in a period of historically low levels of alcohol consumption, but it also includes a large number of variables on, for example, illicit drug use, behavioral and emotional problems, family relationships, and school satisfaction. The present study included a total of 4666 individuals who had non-missing data on all included variables (84% of all participants). They came from 334 different school classes, with an average of 14 students per class. Respondents with valid answers regarding all included variables (see further below) differed significantly (p < 0.05) from those who had missing data on any variable except cannabis use. The analytical sample had a lower prevalence of ADHD medicine and CP than those with missing data. A larger share were females and lived with both their parents and the group had a higher grade point total (GPT), but a lower level of SS.

Dependent variable
The dependent variable was lifetime prevalence of cannabis use, thus distinguishing those who had used cannabis from those who had not. The measure was derived from a question on how many times a respondent had used cannabis (marijuana or hashish), ranging from 0 to 40 or more times. The prevalence of cannabis use in the sample was low (5.27%, see Table 1) and we chose not to use a more detailed measure, to reduce problems with sparse data.

Independent variables
The primary independent variables were ADHD, CP, and SS. While Futura01 does not measure ADHD or CP using diagnostic criteria, we employed two measures that could serve as proxies. ADHD was measured by a question asking whether participants had ever received medicine for ADHD (with examples given by the brand names Ritalin, Concerta, and Strattera). Individuals who responded "don't know" (n = 148) were coded as missing.
We note that the use of ADHD medicine as indicative of ADHD can be problematic. Pharmacological treatment has been shown to reduce ADHD symptoms (see Sayal et al., 2018) and to improve subsequent substance use outcomes (Boland et al., 2020). Having ADHD medicine as a proxy creates a risk that the strength of the relationship between ADHD and cannabis use is underestimated. On the other hand, this potential underestimation will be counteracted if those who are prescribed medicine present more severe symptoms than those who are not. A Swedish register study has shown that individuals with ADHD who medicate are more likely to have a drug use disorder than those who do not, and that this difference is explained by the severity of the condition (Sundquist et al., 2015). Also, the non-ADHD group in this study may include some individuals with ADHD who are not medicating. About 75% of Swedish children and youth aged 10-17 years who are diagnosed with ADHD are prescribed medicine (Socialstyrelsen, 2014). Therefore, the prevalence of ADHD in this study is most likely underestimated.
The measure of CP assessed whether participants had committed legal violations. An additional item embedded in this block of items covered being involved in a fight, but we excluded this as it was impossible to interpret as it could reflect self-defense. The relevant items assessed whether participants had shoplifted, stolen a bike, destroyed property, engaged in burglary, lit something on fire (e.g., a car), or carried a weapon (e.g., a knife). Each item was rated on a four-level scale, ranging from 1 = never to 4 = more than five times. The internal consistency of the legal violations scale was good (alpha = 0.74). Those who had a score of 8 or lower on the legal violations scale (91.73%) were distinguished from those who scored 9 or higher (8.27%). Based on these variables, we created four subgroups of individuals for the second set of analyses (see below): (1) both ADHD and CP, (2) ADHD only, (3) CP only, and (4) neither ADHD nor CP. As the measure of CP covered only legal violations, it did not assess all features of CP. For instance, criteria for conduct disorder also include violations of rules and aggression (see e.g. Fairchild et al., 2019). SS was measured using BSSS-4 (Brief Sensation-Seeking Scale-4) (Stephenson et al., 2003). BSSS-4 is a four-item measure of SS, capturing specific dimensions ("experience seeking," "disinhibition," "thrill and adventure seeking," "boredom susceptibility") hypothesized to underlie SS (Stephenson et al., 2003). BSSS-4 correlates strongly with the original Brief Sensation Scale and is associated with cannabis, alcohol, and tobacco use (Stephenson et al., 2003). The internal consistency of BSSS-4 was high (alpha = 0.82).

Control variables
Control variables included sex (male/female), grades, and family structure. Grades capture the GPT in mathematics, Swedish, and English in the previous semester, and ranges from 0 to 60. Higher values imply higher grades. The family structure variable distinguished individuals living with both their parents from those in other family structures. The original question asked respondents whether they lived with both their parents, with their mother, with their father, with parents having joint custody, or that none of these arrangements applied to them. The question did not specify whether "parents, " "mother, " or "father" referred to biological parents only, so there is a risk that respondents interpreted it differently. Although a number of additional control variables could have been included, we were careful not to "overcontrol" the models by including variables (e.g., emotional problems) that might mediate rather than confound the associations (Li, 2021).

Statistical analyses
The associations between ADHD, CP, and other variables were estimated using multilevel logistic regressions. Because the participants were nested in school classes, the regressions included a random intercept for school class. To answer the three research questions, two sets of analyses were carried out. The first set, addressing research questions 1 and 3, estimated the association between ADHD and cannabis use in a total of four models, beginning with an unadjusted model and with consecutive models controlling for CP and the other variables. The next set of regression models included groups with different combinations of ADHD and CP as the exposure of primary interest, instead of separate measures of ADHD and CP. These sets of analyses corresponded to the second research question and included three models. The first model assessed the crude association between ADHD-CP groups and cannabis use; the second model controlled for family structure, GPT, and sex, and the final model also adjusted for SS.
A shortcoming of logistic regression is that the magnitude of the odds ratios (ORs) cannot be directly compared across models because the metric changes when additional variables are introduced into the model (Breen et al., 2018;Mood, 2010). Because of this, average marginal effects (AMEs) derived from the regression models are also presented for both sets of analyses. These describe the associations in terms of percentage point differences in the probability of having used cannabis and can be directly compared across models (Mood, 2010). The basic principle of AMEs, in the case of this study, is to predict each respondent's probability of cannabis use derived from the logistic regression estimates, comparing the predicted value if having used ADHD medicine with the predicted value if not having used ADHD medicine, keeping the value of the other variables as observed under both predictions (Williams, 2012). The sum total of the differences is then averaged over respondents. The same rationale applies for the second set of analyses, where the probability of cannabis use was compared for each of the ADHD-CP groups, in relation to the combined group.
Statistical analyses were performed in Stata 16 (StataCorp, 2019). Multilevel logistic regressions were run using the melogit command and AMEs were computed using the margins command. Table 1 presents descriptive statistics for the included variables. The lifetime prevalence of cannabis use was about 5% in the sample, and 4% had at some point been prescribed ADHD medicine. Our categorization of CP meant that around 8% were defined as having CP. The vast majority (89%) belonged to the group with neither ADHD nor CP and the smallest group comprised those who had both conditions. A majority lived with both their parents, and there was a slightly smaller share of males than females.

Regression models: research questions 1 and 3
Table 2 presents results from the multilevel logistic regressions. Model 1 shows the unadjusted OR for ADHD and the other variables. There was a significant association between ADHD and cannabis use across all models. When controlling for CP (model 2), the OR for ADHD was substantially smaller than in the bivariate case. This was confirmed when looking at the AMEs, where the association was reduced by 58% (see Table 3). However, when adjusting for family structure, gender, GPT, and SS in the final model, individuals with ADHD still had significantly higher odds of having used cannabis. CP were substantially and significantly associated with cannabis use. When controlling for SS, the OR for CP was greatly reduced (from 14.28 to 8.00), indicating that the relationship between CP and cannabis use to a considerable extent was accounted for by SS. This was also confirmed by the drop in the AMEs, as shown in Table 3. The AME for CP was reduced by 48% in model 4 compared with model 2, and this change was mostly due to SS (see Table 3). SS was positively and significantly associated with cannabis use, and significant associations were also observed for family structure and GPT. Students living with both their parents had lower odds of cannabis use, as did those with higher GPT. Males had significantly higher unadjusted odds than females, but this association was not significant in the adjusted models. Table 4 presents results from the multilevel logistic regressions predicting lifetime cannabis use in the ADHD-CP group and covariates. The AMEs are shown in Table 5. The ORs in Table 4 show differences in the odds of having used cannabis between individuals with ADHD only, those with CP only, and those with neither condition, respectively, compared with the base category of individuals with both ADHD and CP. Model 1 shows the unadjusted association between ADHD-CP groups, whereas model 2 is adjusted for family structure, GPT, and sex. In the third model, SS was added as a potential mediator of the link between ADHD-CP group and cannabis use (research question 3).

Regression models: research questions 2 and 3
The unadjusted model (model 1) shows a clear differential in the odds of cannabis use between the combined ADHD-CP group on the one hand, and the groups with either ADHD only or without ADHD or CP on the other. The combined group had odds of cannabis use that were about 14 times higher than those in the group with only ADHD (1/0.07 = 14.29, p < 0.001), about 2.5 times higher than those in the CP-only group (1/0.39 = 2.56, p < 0.01) and 50 times higher than those in the group with neither ADHD nor CP (1/0.02 = 50, p < 0.001). It should be noted that the confidence intervals are quite wide and that these estimates a contains six separate logistic regressions, i.e., one for each independent variable. *p < 0.05, **p < 0.01, ***p < 0.001. OR = odds ratio, ci = confidence intervals, aDHD = attention deficit hyperactivity disorder, cP = conduct problems, GPt = grade point total, SS = sensation-seeking.  *p < 0.05, **p < 0.01, ***p < 0.001. OR = odds ratio, ci = confidence intervals, aDHD = attention deficit hyperactivity disorder, cP = conduct problems, GPt = grade point total, SS = sensation-seeking.
thus are far from precise. When controlling for family structure, GPT, and sex (model 2), large and significant differences remained between the combined group, and the groups with only ADHD or neither ADHD or CP. The odds of cannabis use in model 2 were 12.5 times higher for the combined ADHD and CP group compared with the ADHD-only group (1/0.08 = 12.50, p < 0.001) and 25 times higher compared with the group without ADHD or CP (1/0.04 = 25, p < 0.001) (model 3). Yet while the CP-only group had lower odds of cannabis use than the combined group, this difference was not statistically significant. Model 3 added SS, which was substantially and significantly associated with cannabis use. A one-unit increase in SS increased the odds by about 23%. With SS included, the differences between the combined group and the group with ADHD or the group without ADHD or CP remained large and highly significant (p < 0.001 for both). However, there was a large reduction in the AMEs between the first and the third model. The AMEs were reduced by more than 50% for all three groups in model 3 compared with model 1, with the combined group as reference (see Table 5). While the change from model 2 to model 3 was substantial for the ADHD-only group and for the group with neither condition, this was not the case for the CP-only group. In this group, the major reduction occurred between model 1 and 2, suggesting that SS plays a minor role in explaining differences between individuals with both ADHD and CP and individuals with CP only.

Discussion
Using a large and nationally representative sample of Swedish adolescents in ninth grade, this study explored the association between ADHD and lifetime use of cannabis, and the extent to which CP and SS may account for this link. The structure of the relationships between ADHD, CP, and substance use has proved highly complex, and disentangling these relationships is crucial for both theoretical developments and preventive initiatives.
The first set of analyses confirmed that ADHD is related to cannabis use, but the association was profoundly reduced when adjusting for CP. This finding lends additional support to the view that CP are a mediator of the ADHD-substance use link (e.g., Klein et al., 2012;Sibley et al., 2014), although our cross-sectional data cannot exclude the possibility that they are rather a confounder. In fact, more than half of the unadjusted association between ADHD and cannabis use was accounted for by CP. SS, on the other hand, did not explain any remaining part of the ADHD-cannabis use link once CP, family structure, sex, and school grades were adjusted for. In contrast, the link between CP and cannabis use seemed to be explained by SS to a considerable degree.
Thus, SS appears to play different roles for adolescents with ADHD and those with CP in relation to cannabis use.
The second set of analyses explored whether individuals with combined ADHD and CP were more likely to have used cannabis than individuals with only one of these conditions. The unadjusted difference between the combined group and the group with only ADHD was substantial, confirmed by both the ORs and the AMEs. For instance, the probability of cannabis use was close to 40 percentage points higher in the combined group compared with in the ADHD-only group. While the association weakened when adjustments were made for the control variables and SS, it was still quite profound (19 percentage points according to the AME). The difference between the combined group and the CP-only group was much smaller, and was only statistically significant in the unadjusted model. Thus, these analyses lend support to the view that the combined group may be more "severe" than the ADHD-only group (Waschbusch, 2002), but not compared with the CP-only group. However, the latter finding should be treated with caution and not judged solely based on p-values, because there were few individuals in the combined group. Nonetheless, the results point at a much larger difference between the combined group and the ADHD-only and CP-only group, respectively, suggesting that the CP-only group is more likely to use cannabis than the ADHD-only group.
An interesting finding is that SS seemed to play a more central role in explaining the difference between the combined group and the ADHD-only group and the group with neither condition, than in explaining the difference between the combined group and the CP-only group. In the latter case, the changes in the relationship between the models were mostly due to the control variables (i.e., family structure, sex, and GPT). Thus, the combined and the CP-only group may be characterized by similar and higher levels of SS compared with the other groups, suggesting that SS is more closely linked to CP than to ADHD. The cross-sectional data used precludes us from teasing out the temporal relationships between the main variables in the two sets of analyses, so the findings are consistent with different causal structures. It is imperative that future work continues to address the temporal relationships between ADHD, CP, and SS, using large longitudinal data sets from the general population of adolescents. We suggest that future research should compare at least two partly opposing structures that may be derived from this and prior studies: a "cascade" model and a "partly mediation, partly confounding" model.
The cascade model postulates simple mediated paths, where ADHD predicts CP, CP predicts SS, and SS predicts substance use. Still, given that SS is considered to be a personality trait (Zuckerman, 1994), it may temporally precede CP, and as such partly confound the association between CP and substance use. In this "partly mediation, partly confounding model," ADHD may influence substance use through CP, but CP may also in part be determined by SS. We emphasize that a completely different conceptualization may be tenable as well, where at least some of the main variables explored here may be manifestations of an underlying latent variable, similar in spirit to the claim that a general p-factor underlies manifestations of different mental disorders (Caspi et al., 2014). In this view, cannabis use and CP may not be best treated as separate entities, but rather as expressions of a general rejection of societal norms. At least in Sweden, with its prohibitionist stance toward all illicit substance use regardless of age (Tham, 2021), this parsimonious conceptualization may have merit.
In addition to the fact that the study used cross-sectional data, there are also other limitations that should be noted. First, the measure of cannabis use was quite crude, only distinguishing between those who had ever used cannabis and those who had never done so. A more fine-grained measure would enable more detailed conclusions as to how ADHD, CP, and SS are related to cannabis use in adolescents. Also, only participants with data on all variables considered were included in the study. While they constituted about 85% of the entire sample, the differences between included and excluded participants on the considered variables mean that the prevalence estimates as regards variables such as ADHD and CP are overly conservative.
In conclusion, this study showed that the association between ADHD and cannabis use was largely accounted for by CP. The association between CP and cannabis was reduced when controlling for SS, suggesting that higher levels of SS in those with CP may explain their higher likelihood of cannabis use. Also, individuals with both ADHD and CP were more likely to have used cannabis compared with those with only ADHD, but not compared with those with only CP. Further research should continue exploring how ADHD, CP, and SS are linked to cannabis use. Knowing the causal structures between these variables would facilitate the development of appropriate cannabis preventive interventions.