Exercise, Decision-Making, and Cannabis-Related Outcomes among Adolescents

Abstract

Objective: Poor decision-making may represent a risk factor for adverse cannabis-related outcomes, whereas exercise has been linked to better executive functioning and substance use outcomes. This study examines the associations between self-reported exercise and cannabis use (CU) outcomes over 6 months among adolescents, and whether these are mediated by exercise-related effects on decision-making. Method: Participants were 387 adolescents aged 15–18 who completed two assessments 6 months apart. Self-reported past 6-month hours/week of exercise were assessed at baseline. At the 6-month follow-up, participants completed measures assessing past 6-month CU frequency, presence of CU disorder (CUD), and CU-related problems, as well as risky decision-making tasks (Iowa Gambling Task, Game of Dice Task, Cups Task), which were used to derive a latent construct of decision-making. We used prospective mediation to examine the role of decision-making in the relationship between exercise and CU outcomes. Results: More self-reported exercise at baseline predicted greater CU frequency at the 6-month follow-up, but did not predict the presence of a CUD, or cannabis-related problems. After controlling for confounds, baseline exercise did not predict better decision-making at follow-up. Decision-making did not predict CU outcomes, and indirect effects of decision-making were not significant. Conclusions: Contrary to hypotheses, adolescents reporting more exercise at baseline also reported higher CU frequency in our sample. This association may be explained by factors like sample characteristics or sports types, but more research is needed to explore this. Results did not support a mediating role for decision-making in the associations between exercise and CU outcomes.

Supplemental data for this article can be accessed online at https://doi.org/10.1080/10826084.2021.1906279.

Keywords:

• Cannabis use
• decision-making
• exercise
• adolescence
• mediation

Introduction

Participation in sports and exercise has been consistently linked to increased alcohol use and lower cigarette use (Kwan et al., 2014; Terry-McElrath et al., 2011). Associations with cannabis use (CU), on the other hand, are less clear, as studies have often lumped cannabis in with other illicit substances (Kwan et al., 2014; Lisha & Sussman, 2010; Terry-McElrath et al., 2011). Some studies have found that adolescents and young adults who report greater engagement in sports and exercise are less likely to use cannabis (Barber et al., 2001; Darling, 2005; Dawkins et al., 2006; Dever et al., 2012; Henchoz et al., 2014; King et al., 2016; Terry-McElrath et al., 2011), whereas others have found no relationship (Aaron et al., 1995; Eccles & Barber, 1999; Mahoney & Vest, 2012; Wichstrøm & Wichstrøm, 2009), or suggested moderation by factors such as age, sex, exercise intensity, and team involvement (Boyes et al., 2017; Kwan et al., 2014). There is some evidence that exercise and team involvement work synergistically to predict lower CU, although this association may also vary with specific sports (Ford, 2007; Lisha & Sussman, 2010; Terry-McElrath et al., 2011). Exercise interventions have also been linked to increased abstinence rates in the treatment of substance use disorders (Wang et al., 2014; Zschucke et al., 2012), including one specifically targeted to cannabis use disorders (CUDs; Buchowski et al., 2011). Several factors have been identified as potential mediators of the association between exercise and positive substance use outcomes, including decreases in internalizing symptoms, decreased stress reactivity, increased social support, development of adaptive coping strategies, and increased self-efficacy (Wang et al., 2014; Zschucke et al., 2012). However, the potential mediating role of exercise-related effects on cognition has not yet been examined.

Indeed, exercise has been linked to a variety of positive effects on cognitive function spanning multiple domains, including intellectual functioning, perceptual skills, memory, and executive functioning (Hillman et al., 2008; Sibley & Etnier, 2003), which can often be observed within two to three month periods (Best, 2010; Tomporowski et al., 2008). Within the domain of executive functioning, higher levels of exercise have been associated with improvements in attention, reasoning, set-shifting, processing speed, inhibitory control, error monitoring, cognitive flexibility, and working memory (Best, 2010; Guiney & Machado, 2013; Hillman et al., 2014; Kamijo & Takeda, 2010; Kelly et al., 2014). All of these aforementioned executive processes are involved in decision-making, i.e. the ability to make optimal choices that maximize reward in the presence of risk (Bechara & Damasio, 2002; Pacheco-Colón et al., 2019). Despite this, effects of exercise on decision-making have rarely been studied. One recent study found that a 3-month exercise program resulted in reduced effort-discounting, but had no impact on risky decision-making (Bernacer et al., 2019). In other words, participants were more willing to expend physical effort to obtain monetary rewards postintervention, whereas their valuing of monetary rewards based on explicit risk probabilities remained unchanged. Thus, there is some evidence that exercise may impact certain aspects of decision-making, which could potentially affect substance use and other risky behaviors.

Poor decision-making has been identified as a potential risk factor for problematic substance use, placing certain individuals at greater risk for escalation in use and subsequent addiction. It has been linked to earlier age of onset of substance use disorders (Tarter et al., 2003), faster transition to drug co-use (Lopez-Quintero et al., 2018), and relapse across various substances (Bowden-Jones et al., 2005; Nejtek et al., 2013; Passetti et al., 2008; Paulus et al., 2005; Verdejo-Garcia et al., 2014). Specifically with regards to CU, Gonzalez et al. (2012) found that, despite no differences in decision-making performance between young adult cannabis users and non-users, poor decision-making was associated with a higher number of CUD symptoms among cannabis users. Greater CU also predicted a higher number of cannabis-related problems, but only among those with poor decision-making (Gonzalez et al., 2015). Thus, there is some evidence that poor decision-making may represent a neurocognitive risk factor for adverse cannabis-related outcomes.

The current prospective study examines the associations between exercise, decision-making, and CU among adolescents over six months. Based on the aforementioned findings, we hypothesize that more exercise will be associated with positive CU outcomes, including lower CU frequency, lower odds of having a CUD, and lower severity of CU-related problems six months later. In addition, we predict that these effects will be mediated by exercise-related effects on decision-making, such that higher levels of exercise will be associated with better decision-making performance, which will in turn predict better CU outcomes. Findings from these analyses will begin to answer whether exercise-related cognitive gains can be leveraged in the prevention of cannabis addiction among adolescents.

Method

Participants

Participants were 387 adolescents aged 15–18 who were primarily cannabis users and were predominantly Hispanic/Latino (90%) and White (77%). Participants were recruited from Miami-Dade County middle and high schools, as well as through word-of-mouth referrals, as part of a larger study examining the effects of different adolescent CU trajectories on episodic memory and decision-making (R01 DA031176; N = 401). This parent study recruited a sample of adolescents who were at risk for escalation in CU; detailed inclusion and exclusion criteria for the parent study have been described in detail elsewhere (Duperrouzel et al., 2019), and are also included in the Supplementary Material.

Procedure

Parental consent and participant assent were obtained for all participants prior to the baseline assessment of the parent study. Participant consents were also obtained for youths who became of legal age during the course of the study. All study procedures were approved by the Institutional Review Board at Florida International University (IRB-13-0065 for parent study; IRB-19-0117 for current study).

The parent study involved five assessments conducted at 6-month intervals over a two-year period. The current study only includes data collected at the fourth and fifth assessments of the parent study. The fourth assessment of the parent study is referred to as “baseline” for this study, as it was the first measurement wave at which a large number of participants completed the exercise measure, whereas the fifth assessment of the parent study is here referred to as the “6-month follow-up.” Participants from the parent study were included in the current study if they completed at least one of these two measurement waves (N = 387).

Measures

Exercise

The Sports and Activity Involvement Questionnaire (SAIQ) is a questionnaire originally developed for use in the Adolescent Brain and Cognitive Development (ABCD) Study (Barch et al., 2018). It was adapted as a self-report questionnaire for use with participants in the current study. The SAIQ collects detailed information regarding participants’ involvement in sports, exercise, and other types of activities over the past 6 months and past 30 days. A full list of the activities included in this questionnaire can be found in the Supplementary Material. Participants indicated the number of months, weeks per month, days per week, and minutes per day that they spent on each endorsed activity. Using the total number of minutes spent on sports and exercise, we calculated the average number of hours per week spent on sports and exercise over the past 6 months, which was used as our measure of exercise. Participants also indicated whether they engaged in each activity as part of a team. Participants were coded as being involved in teams if they reported team involvement for most of the activities endorsed. Because the SAIQ was added to our protocol after parent study onset, it was completed by 138 participants (36% of the sample) at the current study’s baseline assessment.

Substance use

We assessed three cannabis-related outcomes at the 6-month follow-up through the following three questionnaires. The Drug Use History Questionnaire (DUHQ) is a semi-structured interview assessing frequency and amount of use of 16 different drug classes during a participants’ lifetime, the past 6 months, and the past 30 days (Duperrouzel et al., 2019; Rippeth et al., 2004) For follow-up visits, examiners queried participants’ typical frequency and amount of use for each month in the 6-month assessment interval. We used past 6-month frequency (in days) of CU at the 6-month follow-up as one of our CU outcome measures. To account for the influence of other substance use on cannabis outcomes, we covaried for past 6-month frequency of alcohol and nicotine use.

We also used the substance use modules of the Structured Clinical Interview for DSM-IV (SCID-IV) to diagnose the presence of alcohol and other substance use disorders at the 6-month follow-up. We used a dichotomous variable indicating the presence of a CUD (abuse or dependence) in the past 6 months as one of our cannabis outcomes.

Finally, participants who reported a history of CU also completed the Marijuana Problems Scale (MPS), a 35-item self-report questionnaire with adequate validity and reliability (Buckner et al., 2010; Stephens et al., 2000). This instrument assesses negative social, personal, occupational, and physical consequences experienced as a result of CU over the past 6 months. We used the total MPS score at the 6-month follow-up as one of our cannabis outcomes.

Decision-making

Decision-making was assessed through three computerized tasks at the 6-month follow-up. The Cups Task measures decision-making under conditions of specified risk for both gain and loss domains (Levin & Hart, 2003). In this task, participants were shown a display of 2, 3, or 5 cups on each side of the screen, and were instructed to choose a cup from one of the two sides for a total of 54 trials. One side always yielded a definite reward or a smaller loss, while the other side provided a chance for a greater reward for a total of 54 trials. We used the number of total risky choices from the gain and loss domains as our indices of decision-making for this task.

Participants also completed the Game of Dice Task (GDT), which assesses decision-making under explicit risk conditions (Brand et al., 2005). Participants were asked to guess the result of a die throw by selecting combinations of one, two, three, or four numbers for a total of 18 trials. Low-risk choices (i.e. combinations of three or four numbers) are associated with greater probability of smaller gains, whereas high-risk choices (i.e. combinations of one or two numbers) are associated with lower probability of higher gains. The total number of risky choices was used as our index of decision-making for the GDT.

Finally, participants completed the Iowa Gambling Task (IGT), which measures decision-making under conditions of ambiguous risk (Bechara et al., 1994). Across 100 trials, participants were instructed to select from four card decks, which included two “good” decks (Decks C and D), and two “bad” decks (Decks A and B), while trying to earn as much money as possible. We used the reverse-scored IGT Net Score (i.e. choices from good decks minus choices from bad decks) as our index of DM for this task (Bechara, 2007).

We then used the following four indices—total risky choices in the gain domain from the Cups and total risky choices in the loss domain from the Cups Task, total risky choices from the GDT, and reverse-scored Net Score from the IGT—to derive a latent construct of decision-making, the properties of which have been described in detail elsewhere (Pacheco-Colón et al., 2019), for use in our analyses. This variable served as our primary measure of decision-making across all analyses. Utilizing latent variables reduces measurement error and results in increased power and less biased estimates (Little et al., 2006). This is particularly important for mediation analyses, as measurement error associated with a mediator can severely impact parameter estimates (Muthén & Asparouhov, 2015). Of note, higher scores in our latent construct of decision-making indicate higher risk-taking, and thus, worse decision-making performance.

Estimated IQ

The Wide Range Achievement Test—4th Edition (WRAT-4) Word Reading subtest was used to estimate participants’ IQs at the baseline assessment of the parent study (Wilkinson & Robertson, 2006). We used participants’ standard scores on this test to covary for effects of global cognitive function on decision-making performance.

Demographics

Participants’ sex and age at the baseline assessment were also used as covariates in our analyses.

Analytic plan

First, we examined the independent direct effects of the predictor on later outcomes. Specifically, we conducted three separate regression models examining the effect of baseline exercise on each of our three CU outcomes at the 6-month follow-up (Path c).

Figure 1. Covariate-adjusted mediation model examining decision-making as mediator of the association between past 6-month exercise at baseline and past 6-month cannabis use frequency at the 6-month follow-up. All estimates represent unstandardized partial regression coefficients. Note: GDT = Game of Dice Task; IGT = Iowa Gambling Task; CU = cannabis use; 6MFU = 6-month follow-up; **Significant at p < 0.001; *Significant at p < 0.05.

Figure 2. Covariate-adjusted mediation model examining decision-making as mediator of the association between past 6-month exercise at baseline and past 6-month presence of a CUD assessed at the 6-month follow-up (CFI = 0.94; RMSEA = 0.03). All estimates represent unstandardized partial regression coefficients. Note: GDT = Game of Dice Task; IGT = Iowa Gambling Task; CUD = cannabis use disorder; 6MFU = 6-month follow-up; **Significant at p < 0.001; *Significance at p < 0.05.

Figure 3. Covariate-adjusted mediation model examining decision-making as mediator of the association between past 6-month exercise at baseline and total MPS score at the 6-month follow-up (CFI = 1.00, RMSEA = 0.00). All estimates represent unstandardized partial regression coefficients. Note: GDT = Game of Dice Task; IGT = Iowa Gambling Task; MPS = Marijuana Problems Scale; 6MFU = 6-month follow-up; **Significance at p < 0.001; *Significance at p < 0.05.

To evaluate decision-making as a mediator of the relationship between exercise and CU outcomes, we then ran prospective mediation models which included three paths: the effect of past 6-month exercise at baseline on decision-making performance at the 6-month follow-up (Path a), the effect of decision-making performance at the 6-month follow-up on cannabis outcomes at the 6-month follow-up (Path b), and the direct effect of past 6-month exercise at baseline on cannabis outcomes at the 6-month follow-up after controlling for the mediating influence of decision-making (Path c′). We ran a total of three such models, using past 6-month CU frequency, past 6-month presence of a CUD, and total MPS score at the follow-up as outcomes, respectively. The first two models (CU frequency and disorder as outcomes) included the full sample (N = 387). However, participants with no history of CU (n = 66) were excluded from the analyses involving the MPS score, as this questionnaire examines problems experienced as a result of CU and was therefore not administered to non-users.

Finally, analyses were repeated, controlling for theoretically relevant covariates. Covariate-adjusted models controlled for the influence of sex, baseline age, and IQ on decision-making, as well as for the influence of sex, baseline age, and concurrent use of alcohol and nicotine on CU outcomes.

All analyses were conducted using Mplus 8. We used the Mplus INDIRECT command to assess the significance of the indirect effect (Path a × b). To account for nonnormality in our data and avoid assumptions regarding the distribution of the indirect effect, we estimated standard errors and confidence intervals of model path coefficients using nonparametric bootstrap sampling (20,000 samples).

Missing data

Because the parent study had a 99% retention rate, there were low rates of missingness in the CU and decision-making variables. However, because collection of exercise data began after parent study onset, the SAIQ was completed by 138 participants at the current study’s baseline assessment (∼36% of the sample). Missingness in the exercise questionnaire was related to CU, such that users with missing exercise data also reported more CU. To ensure that data were missing at random, we examined the effect of exercise on cannabis-related outcomes using only those participants with complete exercise data; results were unchanged. Thus, we handled missing data in this and all other study variables using full information maximum likelihood (FIML). This method can be applied to an incomplete dataset to produce parameter estimates that more accurately describe the entire sample. FIML uses information from all available data points to construct parameter estimates under the assumption that the data are missing at random, as in the current study. FIML has been shown to outperform other methods for handling missing data even with large proportions of missing data (Xiao & Bulut, 2020).

Results

Participant characteristics

Demographics and substance use characteristics of our sample are presented in Table 1. Use of drugs other than alcohol, nicotine, and cannabis was low. Most commonly endorsed drugs were hallucinogens (n = 51), benzodiazepines (n = 38), and cocaine (n = 37), with most of these participants endorsing use ≤2 days over the past 6 months.

Table 1. Participant demographic, substance use, and neurocognitive characteristics (N = 387).

	M ± SD or %
Demographics
Baseline age	16.92 ± 0.78
Sex (% Male)	54.0%
Race (% White)	77.0%
Ethnicity (% Hispanic/Latino)	89.7%
WRAT-4 Reading Standard Score	108.35 ± 14.79
Neurocognition 6-month follow-up
Cups Task, risky choices gain domain	17.15 ± 5.56
Cups Task, risky choices loss domain	17.11 ± 6.37
Game of Dice Task, risky choices	5.37 ± 4.70
Iowa Gambling Task, Net Score	10.03 ± 28.60
Substance use 6-month follow-up	Md [IQR] or %
Past 6-month cannabis use (days)	14.00 [0.00, 105.00]
Past 6-month alcohol use (days)	4.00 [1.00, 15.00]
Past 6-month nicotine use (days)	0.00 [0.00, 2.00]
Past 6-month hallucinogen use (days)	0.00 [0.00, 0.00]
Past 6-month benzodiazepine use (days)	0.00 [0.00, 0.00]
Past 6-month cocaine use (days)	0.00 [0.00, 0.00]
Lifetime cannabis use (days)	108.00 [8.50, 483.00]
Past 6-month cannabis use disorder (%)	25.4%
Marijuana problems Scale Score	2.00 [0.00, 4.00]

Note: M = mean; SD = standard deviation; Md = median; IQR = interquartile range.

CU frequency

The direct effect of past 6-month exercise at baseline on past 6-month CU frequency at the 6-month follow-up was significant (Path c: b = 3.02., SE = 0.89, p = 0.001, 95% CI [1.28, 4.78]). This association was contrary to our hypotheses, such that for every additional hour/week of exercise reported at baseline, there was a 3.02-day increase in past 6-month CU frequency at the follow-up.

Our hypothesized mediation model (Supplementary Figure 1) revealed a marginally significant path from exercise at baseline to decision-making at the follow-up (Path a: b = –0.04, SE = 0.02, p = 0.059, 95% CI [–0.08, –0.004]), such that more exercise predicted less risk-raking or better decision-making. The path from decision-making to past 6-month CU frequency at the 6-month follow-up was not significant (Path b: b = 1.26 1, SE = 5.46, p = 0.818, 95% CI [–9.41, 9.74]). After controlling for the role of decision-making, the direct effect of exercise on CU frequency was largely unchanged (Path c′: b = 2.99, SE = 0.94, p = 0.002, 95% CI [1.05, 4.78]). However, the indirect effect of exercise on CU frequency via decision-making was not significant (Path a × b: b = –0.04, SE = 0.22, p = 0.841, 95% CI [–0.61, 0.30]).

As shown in Figure 1, after controlling for effects of sex, age, and IQ, the effect of exercise on decision-making (Path a) was not significant. On the other hand, the direct effect of exercise on CU frequency (Path c′) remained significant even after controlling for covariates.

CUD

The direct effect of past 6-month exercise at baseline on past 6-month CUD assessed at the 6-month follow-up was not significant (Path c: b = 0.02, SE = 0.02, p = 0.197, 95% CI [–0.02, 0.06]).

The subsequent mediation model (Supplementary Figure 2) showed a significant association between past 6-month exercise at baseline and decision-making performance at the 6-month follow-up (Path a: b = –0.04, SE = 0.02, p = 0.035, 95% CI [–0.07, –0.01]), such that more exercise predicted less risk-taking. However, decision-making did not predict past 6-month CUD at the 6-month follow-up (Path b: b = –0.003, SE = 0.09, p = 0.979, 95% CI [–0.18, 0.19]). After controlling for the mediating role of decision-making, the direct effect of baseline exercise on CUD at follow-up remained unchanged (Path c′: b = 0.02, SE = 0.02, p = 0.224, 95% CI [–0.02, 0.06]). The indirect path from exercise to CUD via decision-making was also not significant (Path a × b: b = 0.00, SE = 0.00, p = 0.982, 95% CI [–0.01, 0.01]).

As shown in Figure 2, the effect of exercise on decision-making (Path a) became nonsignificant after controlling for covariates. All other findings remained unchanged.

Cannabis-related problems

Among cannabis users in our sample, the effect of past 6-month exercise at baseline on CU-related problems reported at the 6-month follow-up was not significant (Path c: b = 0.03, SE = 0.08, p = 0.704, 95% CI [–0.12, 0.21]). Similar to previous models, our hypothesized mediation model (Supplementary Figure 3) revealed a significant association between baseline exercise and decision-making at the 6-month follow-up (Path a: b = –0.05, SE = 0.02, p = 0.026, 95% CI [–0.10, –0.01]), as well as a nonsignificant association between decision-making and CU-related problems at the follow-up (Path b: b = 0.06, SE = 0.52, p = 0.906, 95% CI [–0.87, 1.16]). The direct effect of exercise on CU-related problems remained nonsignificant after accounting for the role of decision-making (Path c′: b = 0.55, SE = 0.34, p = 0.106, 95% CI [–0.10, 1.22]). The indirect effect via decision-making was also not significant (Path a × b: b = –0.03, SE = 0.02 p = 0.276, 95% CI [–0.11, 0.001]).

As illustrated in Figure 3, after controlling for other covariates, the direct effect of exercise on decision-making became nonsignificant, whereas all other findings remained unchanged.

Post hoc exploratory moderation analyses

In an effort to better understand our unexpected finding, we conducted a series of post hoc analyses to further explore the association between baseline exercise and CU frequency at the 6-month follow-up. Because these variables have previously been identified as moderators, we conducted separate regression models to determine whether (a) sex and (b) involvement in team sports moderated this relationship.

Results revealed a significant interaction between sex and exercise, (b = –3.02, SE = 1.47, p = 0.039, 95% CI [–5.61, .23]). Probing of this interaction revealed that the association between exercise and CU was primarily driven by males (b = 3.07, SE = 1.11, p = 0.006, 95% CI [0.88, 5.25]), rather than females (b = 0.05, SE = 1.17, p = 0.435, 95% CI [–2.02, 5.90]). The interactive effect of exercise and involvement in team sports was not significant (b = 2.55, SE = 2.11, p = 0.226, 95% CI [–1.86, 6.55]).

Discussion

The current study examined the associations between engagement in exercise and various CU-related outcomes among adolescents, and whether decision-making performance mediated these relationships. Our results suggest that, although participation in exercise did not predict later presence of a CUD or CU-related problems, there was a significant association between self-reported exercise at baseline and greater CU frequency at the 6-month follow-up, even after controlling for covariates. Contrary to our hypotheses, none of these association were mediated by exercise-related effects on decision-making. Across models, more exercise at baseline predicted better decision-making performance at the follow-up (although this effect was marginally significant in one of our models), but this association became nonsignificant after controlling for covariates. Decision-making, on the other hand, did not predict any of the CU-related outcomes explored in this study.

Unexpectedly, we found that adolescents who reported more hours/week of exercise at baseline also reported greater past 6-month frequency of CU at the follow-up assessment. This effect remained significant even after controlling for the effects of age, sex, and concurrent use of alcohol and nicotine on CU. Post hoc exploratory analyses also revealed that this effect was driven by males, as males in our sample reported higher levels of both exercise and CU. These findings are inconsistent with those of several studies documenting protective effects of exercise and sports participation against CU (Barber et al., 2001; Darling, 2005; Dawkins et al., 2006; Dever et al., 2012; King et al., 2016; Terry-McElrath et al., 2011). These discrepancies may be explained, at least in part, by the characteristics of our sample. First, in contrast to previous studies, our study examined these effects in a predominantly Hispanic/Latino sample (90%). In addition, our participants were part of a larger study, the inclusion/exclusion criteria of which were successfully applied to recruit a sample of adolescents at risk for escalation in CU (Duperrouzel et al., 2019; Hawes et al., 2019). Thus, by design, our participants may have had certain characteristics (e.g., personality traits, lower perceived risk of substance use) that made them more likely to use cannabis and other substances, and may not be representative of all adolescents. In addition, associations between exercise and CU may have been influenced by other variables not assessed by the current study, including but not limited to parental monitoring and peer deviance (Dever et al., 2012; King et al., 2016). It is possible that participants in our sample may have had lower parental monitoring and thus more unsupervised time with peers, which may have placed them at greater risk for experimentation with substances and subsequent escalation. Finally, many of the studies examining the associations between sports participation and later CU utilize data that were collected over a decade ago. The recent proliferation of recreational and medical cannabis laws has been accompanied by increased acceptance of CU and decreases in perceptions of risk (Hughes et al., 2016; Pew Research Center, 2016). Thus, it is possible that associations between exercise and CU may be different now than they previously were.

It should also be noted that there is significant cross-study variability in the assessment and operationalization of participation in sports and exercise. For instance, some studies have employed binary variables to measure participation in sports (Darling, 2005), with others using Likert scales to indicate the extent of involvement in sports and exercise (Darling, 2005; Dever et al., 2012; King et al., 2016; Terry-McElrath et al., 2011), or continuous variables to represent time or years spent in these activities (Dawkins et al., 2006). Because we were interested in the cognitive benefits of exercise, we opted for the latter, as minutes per week has been described as the most predictive index of total health benefits (Piercy et al., 2018). Furthermore, although the terms “sports and exercise,” and “physical activity” are sometimes used interchangeably, they are not equivalent (Khan et al., 2012). Sports and exercise contribute to physical activity, but physical activity can also include activities performed through work, chores, at home, and while traveling. Effects on health-related behaviors, such as substance use, may be different for sports and exercise versus physical activity. Indeed, after adjusting for mutual influences, Henchoz et al. (2014) found that participation in sports and exercise was protective against later CU, but higher levels of physical activity were positively associated with later at-risk CU. Of note, the measure used in the current study primarily assessed sports and exercise, although it may also have captured some non-exercise components of physical activity (e.g. active transport through bike riding or skateboarding). Future studies should assess participation in sports and exercise, as well as overall levels of physical activity to determine which of these can be protective in the context of substance use, and for whom.

Nonetheless, several previous studies have found positive associations between sports participation and CU. For instance, Ewing (1998) found that sports participation predicted greater CU among high school males, whereas Peretti-Watel (2002) described a “U-curve” such that CU was highest for males at both the lowest and highest levels of physical activity. Furthermore, Ford (2007) found that this association varied with both sex and specific sport, such that female soccer players and male hockey players reported the highest levels of CU, with runners reporting the lowest. It is thus possible that engagement in sports and exercise may be a protective factor for some and a risk factor for others. More recently, a cross-sectional study found that adults who endorsed using cannabis concurrently with exercise reported more minutes of exercise (both aerobic and anaerobic) per week, as well as greater enjoyment of and motivation to exercise, and were more likely to be male (YorkWilliams et al., 2019). This is in line with our findings, as well as with athletes’ subjective reports that CU enhances their athletic performance (Nguyen, 2019), and evidence of lower body-mass index among cannabis users (Ross et al., 2020). It is therefore possible that increased enjoyment of exercise while under acute cannabis intoxication may have contributed to greater CU frequency, as observed in the current study. Large-scale longitudinal studies would allow for more fine-grained analyses to develop a nuanced understanding of the association between exercise and CU, and its directionality.

In line with our hypotheses, we found a significant association between baseline exercise and risky decision-making at the 6-month follow-up, such that more hours/week of self-reported exercise predicted less risk-taking, or better decision-making. However, this effect appeared to be better explained by the effects of age, sex, and IQ on decision-making. Decision-making encompasses many higher-order cognitive functions (e.g. working memory, response activation and inhibition, performance monitoring, reward learning). Previous work suggests that exercise effects on executive aspects of cognition are most often observed in the domains of inhibitory control and cognitive flexibility among pediatric populations. It is likely that the tasks employed in the current study do not sufficiently tap into these domains, which are more readily assessed through cognitive tasks, such as stop signal and flanker tasks (Erickson et al., 2015; Hillman et al., 2014; Westfall et al., 2018). Future work should examine whether exercise effects on these cognitive domains may mediate associations between exercise and substance use.

In addition, we found no association between decision-making and CU outcomes, including frequency of CU, presence of a CUD, and problems resulting from CU. These findings contribute to a mixed body of work examining the associations between CU and decision-making (Broyd et al., 2016; Crean et al., 2011; Gonzalez et al., 2012, 2015). Importantly, despite the wide range of CU in our sample, most of our adolescents reported relatively low levels of CU, with only about a third reporting chronic, near daily use, and most reporting low severity of CU-related problems (as illustrated by the narrow interquartile range of MPS scores shown in Table 1). It is thus possible that participants may not have been using cannabis at high levels long enough for more adverse cannabis-related outcomes to manifest. More longitudinal work is needed to explore the associations between decision-making and CU over longer periods of time.

Findings from the current study must be interpreted in light of several limitations. First, our measure of exercise relied on participant self-report, which has been shown to have low to moderate correlations with objectively measured physical activity (Prince et al., 2008), and is likely subject to social desirability and memory biases. Future studies should examine associations between more objective measures of exercise or fitness, such as pedometer metrics or measures of aerobic fitness, and substance use outcomes. Second, CU is very complex, with increasingly varied methods of use and potencies. Our study relied on days of use as our index of CU, as frequency indices have been shown to be slightly more reliable than reported amounts of use for adolescents completing timeline follow-back interviews (Levy et al., 2014). However, we were not able to examine the impact of other factors, such as route of administration, product type, and potency. Third, because the exercise measure was added after parent study onset, there was a large proportion of missing data in this variable. However, we used FIML to handle missing data in all study variables. FIML has been shown to produce more accurate parameter estimates than other methods, even when data are sparse (Xiao & Bulut, 2020). Also due to limitations regarding the exercise measure and parent study design, our study covers a 6-month time window, with decision-making and CU outcomes assessed at the same time-point. Although exercise effects can be observed over short periods of time (Best, 2010; Tomporowski et al., 2008), future studies should explore the impact of both exercise and decision-making on later CU outcomes over longer periods of time, as this would also help to clarify the temporality of these associations. Finally, as previously mentioned, our sample was predominantly Hispanic/Latino, which may limit generalizability to more diverse samples.

Conclusions

Higher levels of exercise at baseline predicted greater CU frequency at the 6-month follow-up, but did not predict the presence of a CUD, or CU-related problems. Baseline exercise predicted better decision-making at the follow-up, although this path was marginally significant in one of our models. However, this effect was better accounted for by the effects of sex, age, and IQ on decision-making. Decision-making did not predict CU-related outcomes. The indirect effects of decision-making were also not significant, and thus did not support a mediating role of decision-making in associations between exercise and CU outcomes among adolescents. To the best of our knowledge, ours is the first study to examine the associations between exercise, CU, and risky decision-making. Future studies should continue to examine the effects of exercise on cognition using objective measures of exercise and/or fitness to determine whether exercise-related cognitive gains can be utilized in prevention and treatment efforts aimed at substance-using adolescents.

Declaration of interest

The authors report no conflicts of interest.

Additional information

Funding

This work was supported by R01 DA031176 & U01 DA041156 (to R. Gonzalez), and F31 DA047750-01A1 (to I. Pacheco-Colón) from the National Institute on Drug Abuse.