Risky decision-making as an antecedent or consequence of adolescent cannabis use: findings from a 2-year longitudinal study

Abstract

Background and aims

Although poor decision-making (DM) has been correlated with problematic cannabis use (CU), cross-sectional designs make it difficult to determine whether poor DM represents an antecedent and/or consequence of CU. The current study measured bidirectional associations between CU and DM among adolescents over 2 years and compared these findings to those observed with episodic memory, which is consistently reported as a consequence of CU. We also measured the role of DM as a risk factor for cannabis use disorder (CUD) onset.

Design

Two-year longitudinal study with five bi-annual assessments.

Participants

Participants were 401 adolescents aged 14–17 years at baseline.

Setting

Miami, Florida, USA.

Measurements

CU frequency and CUDs were assessed at each time-point through the Drug Use History Questionnaire and Structured Clinical Interview for DSM-IV respectively. Neurocognition was assessed at odd time-points throughout the Iowa Gambling Task, Game of Dice Task and Cups Task [decision-making (DM)] and the Wechsler Memory Scale IV and California Verbal Learning Test II (episodic memory). We used latent growth curve modeling to examine bidirectional influences between CU and neurocognition over time. We applied discrete time survival analyses to determine whether baseline DM predicted CUD onset.

Findings

Greater lifetime CU frequency was associated with poorer episodic memory at baseline (bs = −14.84, −16.44, Ps = 0.038, 0.021). Greater CU escalation predicted lesser gains in immediate episodic memory (b = −0.05, P = 0.020). Baseline DM did not predict CU escalation (b = 0.07, P = 0.421), nor did escalation in CU predict changes in DM (b = 0.02, P = 0.352). Baseline DM also did not predict CUD onset (adjusted OR = 1.01, 95% confidence interval = 0.98–1.06).

Conclusions

This study replicates findings that poorer episodic memory in adolescents appears to be a consequence of cannabis use, even among adolescents at earlier stages of use. Poor decision-making does not appear to be either a consequence of or a risk factor for escalating cannabis use or onset of cannabis use disorder among adolescents.

INTRODUCTION

Cannabis use (CU) is prevalent among adolescents, with approximately 12.5% of Americans aged 12–17 years reporting past year use, and approximately 8% of these reporting near daily use [1]. Reports from nationally representative samples have consistently shown that the risk of developing a cannabis use disorder (CUD) is approximately 9% among lifetime cannabis users [2,3], although risk may be higher among adolescents, with reported CUD incidence rates of approximately 13.7% in smaller community samples [4]. Factors at multiple levels of influence, from molecular to environmental, have been shown to impact the risk of progressing into regular patterns of use or of developing CUDs. These include adolescent onset of use, male sex, peer influence, positive use expectancies, low self-esteem, drug availability, family stressors and family history of substance use, among others [5–11]. A growing body of work has examined whether poorer cognitive performance, particularly within the realm of executive functioning, may also place individuals at greater risk for CUDs [12,13].

Several aspects of executive functioning have been identified as potential risk factors for addiction to cannabis and other substances, including poor attention, working memory, cognitive flexibility, inhibitory control and decision-making (DM) [12,14–20]. Impairments in risky DM, or the ability to make optimal choices that maximize reward in the presence of risk, have been reported cross-sectionally among adult cannabis users [21–30], and may persist even after prolonged abstinence [23,29]. Others have even found cross-sectional dose–response relationships between CU history and DM among adults, suggesting that these impairments may appear as a result of CU [29,31]. Indeed, there is a plausible neurobiological basis for an association between CU and DM, as chronic CU results in the down-regulation of CB₁ receptors, which are densely located in brain regions involved in planning and DM, such as the ventral tegmental area, striatum and prefrontal cortex [32–35]. Concordantly, cross-sectional studies have shown associations between greater cannabis-related problems and lower gray matter volume in the orbitofrontal cortex [36], as well as increased regional cerebral blood flow in the ventromedial prefrontal cortex among adult chronic cannabis users during a DM task [37].

In contrast, several studies have failed to find associations between CU and DM [38–40], and others suggest that these relationships may be nuanced. For instance, a cross-sectional study by Gonzalez et al. [41] found that, although there were no differences in DM performance between young adult cannabis users and non-users, poorer DM was associated with a greater number of CUD symptoms. In addition, greater amounts of CU predicted a higher number of cannabis-related problems, but only among those with the poorest DM [42]. These findings suggest that DM was not affected by CU per se, but was more relevant to experiencing problems from use.

The majority of the studies examining associations between CU and DM have relied upon adult samples, while less is known about these relationships during adolescence. Poorer or riskier DM has been cross-sectionally documented among regular adolescent cannabis users [43,44], adolescents with CUD in full remission [45] and adults who initiated use during adolescence [46], with null results reported in only one cross-sectional [47] and one longitudinal study [48]. Similar to adult studies, findings from cross-sectional and prospective neuroimaging work suggest that adolescent chronic cannabis users show alterations in brain activation in regions such as the anterior cingulate cortex, ventromedial prefrontal cortex and orbitofrontal cortex while performing DM tasks [37,45,49,50]. These differences in brain activation have been documented even in the absence of between-group differences in behavioral task performance [37,49,50]. Recent studies on adolescents have shown that DM-related activation of frontal brain regions, rather than task performance, predicts adverse cannabis-related outcomes such as CUD relapse [45] and later escalation in CU [49]. It is important to note that adolescence is a sensitive developmental period characterized by large-scale neuromaturational changes that make adolescents simultaneously more likely to initiate and engage in substance use and more vulnerable to the neurotoxic and other adverse effects of substances [51,52]. Thus, further elucidating the role of DM in adolescent CU could have important implications for prevention and intervention efforts aimed at this population.

A major limitation of the aforementioned body of work is its overwhelming reliance upon cross-sectional designs, which makes it difficult to discern whether impairment in DM may be an antecedent or consequence of CU. For instance, impairments in DM that persist even after prolonged abstinence [23,29,45] could represent permanent adverse consequences from use, or evidence of a vulnerability that may predate CU and places individuals at risk for escalation in use and addiction. Alternatively, it is possible that poor DM may be both (a) a risk factor for becoming a regular user or developing problematic CU and (b) a consequence of heavy CU. Given ongoing neural changes associated with both increased risk-taking and greater vulnerability to adverse effects, this could be especially true for adolescents.

The current study examined longitudinal associations between CU and DM in a sample of 401 adolescents followed for 2 years, in an effort to determine whether poor DM is a risk factor for increasing CU and problems from use, a consequence of use or both. Preliminary findings with this cohort [53] showed a modestly sized, marginally significant cross-sectional association between DM and CU, such that greater CU was associated with riskier DM at baseline. We also previously found a modestly sized association between increasing CU and declines in episodic memory, consistent with a large body of literature documenting poorer episodic memory performance from CU [53–56]. In the current study, we first examine bidirectional influences between cumulative lifetime CU and DM over 2 years to determine whether (a) baseline DM predicts escalation in CU over time and (b) whether escalation in CU is associated with changes in DM. To ascertain whether findings are specific to DM, we also conduct the aforementioned analyses with episodic memory performance (instead of DM). We predict that poorer DM at baseline will be associated with greater escalation in cannabis use and that DM will show little to no change with escalating CU over time, consistent with its more prominent role as a risk factor, rather than consequence of cannabis use. Conversely, we have no theoretical basis to suspect that episodic memory would predict increases in CU and therefore hypothesize that it will show declines with escalation in use, consistent with poorer memory being a consequence of use. Secondly, the current study examines whether baseline DM predicts the onset of a CUD. To our knowledge, this is the first longitudinal study to concurrently examine bidirectional influences between CU and DM among adolescents. Our findings will contribute new evidence on neuropsychological predictors and sequelae of CU and addiction, thereby helping to develop more targeted intervention and prevention programs.

METHOD

Participants

Participants were youths recruited through distribution of flyers throughout Miami Dade County, including at middle and high schools, as well as word-of-mouth referrals, for a longitudinal study on DM and CU trajectories (R01 DA031176, Principal Investigator: R.G.). All participants underwent telephone screening that included questions on demographics, mental health, substance use exposure and medical history to determine study eligibility. Because our goal was to be able to identify neurocognitive effects and risk factors specific to CU, we aimed to recruit a sample of adolescents who were at risk for CU escalation and for whom cannabis was their primary drug of choice. Towards this end, the sample over-represented participants with recent use of cannabis. Specifically, exposure to alcohol, nicotine or cannabis—even if minimal (i.e. only a sip or a puff)—was an inclusion criterion. However, by design, 10% of the sample was allowed to have no substance use history, to avoid incidental identification of our participants as substance users and to allow a portion of the sample to be cannabis-naive at baseline (although these participants were retained in the study even if they initiated CU during the course of the study). Additional inclusion criteria were aged 14–17 years at baseline and ability to read and write English. Exclusionary criteria included self-reported developmental disorders, neurological conditions, significant birth complications, in-utero drug exposure, history of a traumatic brain injury or loss of consciousness for > 10 minutes, use of psychotropic medications with known neurocognitive effects [with the exception of stimulants for attention deficit hyperactivity disorder (ADHD)] and a formal diagnosis or history of treatment for a mental health disorder (except for ADHD and conduct disorders, given their high comorbidity with substance use). These criteria served to minimize the presence of potential confounds that sometimes present alongside CU and may impact cognitive functioning.

In addition, in order to avoid participants entering the study at ceiling levels of CU that may obfuscate the possibility of detecting escalation, or participants with significant other drug use to an extent greater than their CU, participants were excluded if their responses during screening indicated heavy use of alcohol (defined as > 13 drinks in a week or more than six to seven drinks in a day on more than three lifetime occasions) or cannabis (defined as using multiple times per day, every day, for > 12 weeks) or the presence of an alcohol or CU disorder based on responses to items from the Substance Dependence Severity Scale [57]. To ensure that our sample consisted of participants who used cannabis as their primary drug of choice, those reporting prior use of any non-prescription drug (other than alcohol, nicotine or cannabis) or recreational use of any prescription drug more than 10 times, any use in the 2 weeks prior to screening or use of any drug to a greater extent than cannabis were also excluded. Although participants were excluded during screening if their responses suggested prolonged more than daily CU or a CUD, we did not exclude participants if they developed a cannabis or other substance use disorder during the study or if their answers to a structured clinical interview suggested that they met criteria for a substance use disorder during any study assessment, including baseline. As with substance use, participants who developed any mental health disorders during the course of the study or whose responses to a structured interview suggested that they met criteria for a mental health disorder at any assessment (including baseline) were not excluded. Rates of internalizing and externalizing disorders are presented in Table 1.

Table 1.

Participant demographics, substance use, mental health, and neuropsychological performance by assessment wave.

	Assessment wave
	T1 (n = 401)	T2 (n = 391)	T3 (n = 383)	T4 (n = 380)	T5 (n = 387)
Demographics (mean, SD)
Age	15.40 (0.72)	15.96 (0.81)	16.38 (0.71)	16.92 (0.78)	17.39 (0.75)
Sex (% male)	54.1	53.7	54.0	53.9	54.0
Race (% white)	76.8	76.7	76.2	76.8	77.0
Ethnicity (% Hispanic/Latino)	89.8	89.5	89.3	89.7	89.7
Years of education	9.11 (0.84)	9.74 (.90)	10.08 (.85)	10.73 (0.88)	11.07 (0.84)
Years of maternal Education	14.23 (2.49)
WRAT-4 reading standard score	108.31 (14.73)
Substance use Characteristics (median, IQR)
Lifetime cannabis use (days)	21.00 (1.00, 144.50)	36.00 (2.00, 186.00)	62.00 (3.00, 266.00)	82.00 (4.00, 350.00)	108.00 (8.50, 483.00)
Lifetime alcohol use (days)	5.00 (1.00, 19.50)	8.00 (1.00, 31.00)	13.00 (3.00, 38.00)	17.00 (5.00, 56.50)	24.00 (7.50, 72.00)
Lifetime nicotine use (days)	.00 (0.00, 3.00)	.00 (0.00, 6.00)	1.00 (0.00, 9.25)	1.00 (0.00, 12.75)	2.00 (0.00, 21.00)
Current CUD (%)	13.2	11.5	19.6	17.6	23.3
Lifetime CUD (%)	22.2	29.4	37.8	42.4	47.8
THC + oral fluids Toxicology (%)	3.5	–	9.5	–	20.7
Mental health
Current internalizing Disorder (%)	5.5	–	4.7	–	4.9
Current externalizing disorder (%)	11.0	–	8.1	–	5.9
Lifetime internalizing disorder (%)	16.2	–	21.9	–	26.5
Lifetime externalizing disorder (%)	12.2	–	17.2	–	19.1
Neuropsychological performance (mean, SD)a
Cups task, risky choices gain domain*	50.00 (10.00)	–	50.04 (9.30)	–	50.31 (9.84)
Cups task, risky choices loss domain*	50.00 (10.01)	–	49.64 (9.88)	–	49.85 (9.79)
GDT, risky choices*	50.00 (9.99)	–	46.13 (9.16)	–	45.28 (9.13)
IGT net score (reverse-scored)*	50.00 (10.00)	–	46.91 (11.90)	–	44.90 (12.78)
CVLT-II total immediate recall	50.00 (10.00)	–	52.50 (10.01)	–	54.78 (10.54)
WMS-IV logical memory I	50.00 (10.00)	–	51.52 (9.98)	–	54.03 (9.89)
WMS-IV designs I	50.00 (10.00)	–	52.75 (10.29)	–	53.73 (10.60)
CVLT-II long delay free recall	49.99 (10.01)	–	52.70 (9.66)	–	55.18 (9.89)
WMS-IV logical memory II	50.00 (10.00)	–	52.11 (9.72)	–	56.12 (10.20)
WMS-IV designs II	50.00 (10.00)	–	51.38 (10.92)	–	53.01 (11.08)

^aT scores were calculated using the T1 mean and standard deviation.

^*Higher scores for these tests denote worse performance.

SD = standard deviation; IQR = interquartile range. Mental health was assessed through the Diagnostic Interview Schedule for Children using DSM-IV criteria [58]. SD = standard deviation; WRAT-4 = Wide Range Achievement Test 4; IQR = interquartile range; CUD = cannabis use disorder; CVLT = California Verbal Learning Test; WMS = Wechsler Memory Scale.

A total of 2837 participants underwent telephone screening. Based on the above-described inclusion/exclusion criteria, 1183 participants were determined to be ineligible for the study. The most commonly endorsed exclusionary criteria at time of screening were no reported substance use (20.3% of screened individuals), self-reported psychiatric disorders (12.2% of screened individuals) and self-reported use of psychotropic medications (5.4% of screened individuals). Of the eligible participants, 1253 youths were not enrolled for other reasons (e.g. no longer interested or parental consent for study enrollment was not obtained). Thus, a total of 401 participants were enrolled into the study. This sample of 401 participants is used throughout most of the analyses presented in the current study, the sole exception being the analyses examining whether baseline DM predicts the onset of a CUD, which utilize a subset of participants—those adolescents who were cannabis users but had a negative lifetime history of CUD at baseline (n = 226).

Procedure

All study procedures and protocols were approved by the Institutional Review Board at Florida International University. We obtained participant assent and parental consent for all participants prior to the baseline assessment. Participant consents were also obtained for those youths who turned 18 years old during the course of the study. Participants received monetary compensation for their participation.

The study involved five assessments conducted at 6-month intervals over 2 years, which consisted of in-person assessments at baseline (T1), 1-year follow-up (T3) and 2-year follow-up (T5) and telephone assessments at the 6-month follow-up (T2) and 18-month follow-up (T4). Substance use, mental health and other self-report data were collected at all measurement waves (T1–T5). Neuropsychological and toxicology testing were conducted during in-person assessments (T1, T3 and T5). Participants received monetary compensation for their time in the study, earning $75 for completing T1, $50 for T2, $100 for T3, $75 for T4 and $130 for T5. In addition, participants who completed all five assessments received a $50 bonus.

Measures

Demographics

We collected information on participant demographics, including age, sex, ethnicity, race and education, at each assessment (Table 1).

Substance use

Substance use history.

The Drug Use History Questionnaire (DUHQ) is a detailed semi-structured interview used to assess frequency and amount of use of 16 different drug classes [i.e. alcohol, nicotine, cannabis, K2/spice/synthetic cannabinoids, cocaine, methamphetamine, other stimulants, heroin, other opiates, benzodiazepine, barbiturates, ecstasy, hallucinogens, other club drugs, phenylcyclohexyl piperidine (PCP) and inhalants] during a participant’s lifetime (only at baseline), the past 6 months and the past 30 days [53,59]. For follow-up assessments, examiners queried participants’ typical frequency and amount of use for each month during the 6-month assessment interval. Cumulative lifetime frequency of cannabis use was calculated at each time-point by adding the lifetime frequency at baseline and the relevant 6-month follow-up frequencies. As in a prior study with this cohort [53], we used lifetime frequency (in days) of CU at each of the five time-points as our primary measure of cannabis use. Cumulative lifetime frequencies of alcohol and nicotine use were also included as covariates in our analyses.

Substance use disorders.

We 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 (abuse or dependence) at each assessment. We used CUD onset as the outcome variable for our second set of analyses. Onset of CUD occurred when cannabis abuse or dependence occurred at any time-point between T2 and T5 among participants without history of CUD at baseline.

Toxicology testing.

Participants were instructed to abstain from using any drugs for at least 24 hours prior to their in-person assessments. During these assessments, we collected saliva samples with the Intercept® Oral Fluid Drug Test (OraSure Technologies, Inc., Bethlehem, PA, USA), which were sent for laboratory testing to Forensic Fluids Laboratories (Kalamazoo, MI, USA) to determine recent use [limit of detection for tetrahydrocannabinol (THC) - 1 ng/ml]. Toxicology testing using oral fluids was chosen over urine because it is perceived as less intrusive and embarrassing by youths, minimizes risk of tampering and registers positive only for very recent use (up to approximately 24 hours for THC and up to 3 days for other drugs). Indeed, it was only very recent use that was of primary interest to us, as we wanted to avoid subacute effects of cannabis from influencing performance on neurocognitive measures. Urine toxicology can test positive for THC metabolites well after subacute effects on neurocognitive subside. Thus, positive THC toxicology test results were used as covariates in analyses to account for residual effects of very recent CU on neurocognition. Because the toxicology testing was analyzed by an external laboratory, we could not exclude participants based on positive results at the time of the assessment.

Neuropsychological testing

Neuropsychological performance in the domains of DM and episodic memory was assessed via performance-based measures administered at in-person assessments (T1, T3 and T5). These tests have been described in greater detail elsewhere [53].

DM.

DM was assessed through the Game of Dice Task (GDT [60]) the Cups Task [61,62] and the Iowa Gambling Task (IGT [63]). As described in prior publications [53,64], we used the following four indices as our measures of DM for each task: total risky choices from the GDT, total risky choices in the gain domain and total risky choices in the loss domain from the Cups Task and the reverse-scored net score from the IGT. For our primary measure of DM, we transformed these raw scores into T scores [mean = 50, standard deviation (SD) = 10] using the mean and SD of each measure using data from the entire cohort at the baseline assessment, and computed their average. This was performed to convert raw scores to a common metric and facilitate interpretation of change throughout study visits. Of note, higher scores on our index of DM indicated greater risk-taking (i.e. worse DM).

Episodic memory.

Episodic memory was assessed through the California Verbal Learning Test, 2nd edition (CVLT-II [65]) and the Logical Memory and Designs subtests of the Wechsler Memory Scale, 4th edition (WMS-IV [66]). For our measure of immediate memory, we used the total immediate recall across all trials of the CVLT-II, and the Logical Memory I and Designs I immediate recall trials. For our measure of delayed memory, we used the long delay free recall trial of the CVLT-II and the logical memory II and designs II delayed recall trials. As was performed with the DM measures, we transformed raw scores from these trials into T scores using the scores on each measure at the baseline assessment and computed their average. This yielded indices of immediate memory and delayed memory for use in our analyses.

Estimated IQ.

The Wide Range Achievement Test 4—word reading subtest (WRAT-4 reading) requires participants to correctly read words aloud and is often used as a proxy of general intelligence [67]. We used this index to estimate participants’ IQs at baseline, including it as a covariate in our analyses.

Analytical plan

Bidirectional associations between CU and neurocognition.

To address our aim of examining bidirectional influences between lifetime CU frequency and neurocognition throughout the 2-year study period, we used latent growth curve modeling (LGCM). We first ran four separate unconditional linear growth models to examine growth in cannabis use, DM, immediate memory and delayed memory. Subsequently, we specified a total of three separate multivariate LGCM, also known as parallel process models [68]. Each of these models simultaneously estimates the CU growth curve and one of the neurocognition growth curves, with separate models run for DM, immediate memory and delayed memory.

Within each parallel process model (see Fig. 1), we specified the following parameters to address study hypotheses: (a) to examine associations at baseline, the CU intercept was correlated with the intercept of the relevant neurocognitive variable; (b) to examine whether baseline neurocognitive performance predicted change in cannabis use, we regressed the CU slope on the intercept of the relevant neurocognitive variable; (c) to examine whether baseline CU predicted change in neurocognitive performance we regressed the slope of the relevant neurocognitive variable on the CU intercept; and (d) to examine whether changes in CU predicted changes in neurocognition we regressed the slope of the relevant neurocognitive variable on the CU slope.

Figure 1.

Covariate-adjusted parallel process LGCM of cannabis use and decision-making (${\chi }_{(133)}^2=606.48$, P < 0.05; CFI = 0.90; RMSEA = 0.10, SRMR = 0.13; AIC = 17618.23; SABIC = 17 645.08). Paths are presented in unstandardized metric, and T1–T5 referto the assessment time-points. CFI = comparative fit index; RMSEA = root mean square error of approximation; AIC = Akaike information criterion; SABIC = sample-size adjusted Bayesian information criterion; LGCM = latent growth curve modeling; SRMR = standardized root mean square residual

Finally, we identified theoretically relevant covariates which may influence associations between CU and neurocognition, which included sex [69], estimated IQ [70], baseline age [51], use of other substances such as alcohol and nicotine [71,72] and toxicology results, suggesting acute THC intoxication at each available time-point [73]. We then re-tested all parallel process models and compared findings after accounting for these covariates. Results from unadjusted models can be found in the Supporting information. The current study was not pre-registered on a publicly available platform; thus, results should be considered exploratory.

All LGCMs were conducted using Mplus version 8 [74]. Models were specified using maximum likelihood estimation with standard errors and a χ² statistic that are robust to non-normality (MLR). We assessed model fit using absolute fit indices, as well as relative fit indices such as the sample-size adjusted Bayesian information criterion (SABIC) and the Akaike information criterion (AIC). Comparative fit index (CFI) values of 0.95 or greater were used to indicate excellent fit and values between 0.90 and 0.94 were used to indicate acceptable fit [75]; root mean square error of approximation (RMSEA) values of less than 0.05 were used to indicate excellent fit [75]; and standardized root mean square residual (SRMR) values below 0.08 were considered to indicate good fit [76].

Missing data.

All 401 participants provided complete substance use and mental health data at their baseline assessment, and there were low rates of missingness for follow-up assessments (see Table 1). We used full information maximum likelihood (FIML) estimation to handle missing data. FIML uses all available data to construct parameter estimates under the assumption that the data are missing at random.

DM as a predictor of CUD onset.

To examine the role of DM as a risk-factor for developing a CUD, we then implemented discrete time survival (DTS) analyses examining whether DM performance at baseline predicted the onset of a CUD during the 2-year study period and to estimate the predictive hazard of developing a CUD across the last four assessments. Importantly, these analyses utilized a subset of 226 participants who were cannabis users but had a negative lifetime history of CUD at baseline. Data were organized in a person—time format, where each individual was at risk at each wave observed from baseline until the onset of a CUD or the last assessment (for censored cases). Estimates of DTS models are presented as odds ratios (ORs), which are analogous to hazard ratios estimates from survival models. All analyses were re-run after controlling for baseline age, sex, estimated IQ, lifetime frequency of alcohol and nicotine use and positive THC toxicology results. DTS models were fitted using Stata version 14.2.

RESULTS

Change in CU over time

The unconditional linear growth model of CU showed acceptable fit (Table 2). On average, lifetime frequency of CU increased significantly (β = 0.792; i.e. large effect) during the 2-year period (P < 0.001). However, there was significant variability in the slope of cannabis use, suggesting that individual participants varied substantially in their patterns of change over time. The correlation between the CU intercept and slope was also large and significant (P < 0.001), indicating that participants with more frequent CU at baseline demonstrated greater escalation in CU during the course of the study.

Table 2.

Fit indices and estimates for unconditional linear growth models of cannabis use and neurocognition.

	χ2	d.f.	CFI	RMSEA	SRMR	AIC	SABIC	Intercept $\overline{x}$	Slope $\overline{x}$	Intercept σ2	Slope σ2	Cov (I/S)
Cannabis use	166.37**	10	0.95	0.20	0.02	12 048.28	12 056.48	10.72**	4.22**	397.67**	28.41**	56.91**
Decision-making	4.06*	1	0.98	0.09	0.02	7285.52	7292.08	49.84**	−0.59**	19.37**	1.07*	−0.92
Immediate memory	< 0.001	1	1.00	0.00	0.00	7418.52	7425.09	50.00**	1.01**	38.43**	0.30	0.86
Delayed memory	3.37	1	1.00	.08	0.02	7409.18	7415.75	49.83**	1.16**	39.56**	0.20	0.74

^**P < 0.001;

^*P < 0.05.

All estimates represent unstandardized values.

CFI = comparative fit index; d.f. = degrees of freedom; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; AIC = Akaike information criterion; SABIC = sample-size adjusted Bayesian information criterion; Cov (I/S) = covariance between intercept and slope.

Change in DM over time

The unconditional linear growth model of DM showed excellent fit. Results indicated that, on average, DM performance improved over time (P < 0.001), as evidenced by moderate decreases in risk-taking (β = −0.571), which could be consistent with brain maturation and/or practice effects. There was also a significant amount of variability in the slope of the DM variable, indicating that participants improved at different rates. The correlation between the DM intercept and slope was not significant (P = 0.371), indicating that initial levels of DM were not related to changes in DM performance.

Change in episodic memory over time

For both immediate and delayed memory, the unconditional linear growth models showed excellent fit (Table 2). On average, participants’ immediate and delayed memory performance increased significantly over time (Ps < 0.001); these represented large effects (βs = 1.860 and 2.619). Conversely, the slope variances were not significant, indicating that participants’ memory performance improved uniformly among participants among study visits. The correlations between intercepts and slopes were not significant for either immediate (P = 0.436) or delayed memory (P = 0.494).

CU and DM

As illustrated in Fig. 1, the association between intercepts was not significant (P = 0.188), indicating that lifetime days of CU reported at baseline were not related to baseline DM performance. Consistent with our hypothesis, the CU slope did not predict the DM slope (P = 0.352), suggesting that escalation in CU did not result in decrements in DM performance over time. However, contrary to our hypotheses, neither did the DM intercept predict the CU slope (P = 0.421), indicating that baseline DM performance did not predict escalation in CU during the course of the study. The association between CU intercept and DM slope was also not significant (P = 0.544), indicating that lifetime CU at baseline did not predict change in DM performance over time. Detailed results from this model are listed in Table 3.

Table 3.

Detailed estimates for covariate-adjusted parallel process model of cannabis use and decision-making.

Parameter	Standardized estimate (β)	Unstandardized estimate (standard error)	P-value	Lower 95% CI	Upper 95% CI
DM intercept ←→ CU intercept	0.093	7.232 (5.498)	0.188	−3.545	18.008
CU intercept → DM slope	−0.078	−0.003 (0.006)	0.544	−0.014	0.008
DM Intercept → CU slope	0.059	0.065 (0.081)	0.421	−0.094	0.224
CU slope → DM slope	0.131	0.022 (0.023)	0.352	−0.024	0.067
CU intercept ←→ CU slope	0.539	45.752 (6.609)	< 0.001	32.799	58.705
DM intercept ←→ slope	−0.087	−0.299 (0.794)	0.706	−1.856	1.257
Sex→ CU intercept	−0.147	−5.414 (1.855)	0.004	−9.050	−1.778
Sex→ CU slope	−0.221	−2.158 (0.528)	< 0.001	−3.192	−1.124
Sex→ DM intercept	0.187	1.659 (0.637)	0.009	0.409	2.908
Sex→ DM slope	−0.012	−0.020 (0.194)	0.920	−0.401	0.361
Baseline age→ CU intercept	0.129	3.286 (1.093)	0.003	1.144	5.429
Baseline age→ CU slope	0.102	0.688 (0.362)	0.058	−0.022	1.398
Baseline age→ DM intercept	0.090	0.554 (0.455)	0.224	−0.338	1.447
Baseline age→ DM slope	−0.008	−0.008 (0.142)	0.952	−0.286	0.269
Estimated IQ → DM intercept	−0.016	−0.005 (0.023)	0.833	−0.049	0.040
Estimated IQ → DM slope	−0.148	−0.008 (0.006)	0.208	−0.020	0.004
Lifetime alcohol T1 → lifetime CU T1	0.198	0.730 (0.213)	0.001	0.313	1.147
Lifetime alcohol T2 → lifetime CU T2	0.176	0.672 (0.184)	< 0.001	0.311	1.033
Lifetime alcohol T3 → lifetime CU T3	0.171	0.685 (0.159)	< 0.001	0.373	0.998
Lifetime alcohol T4 → lifetime CU T4	0.171	0.716 (0.165)	< 0.001	0.393	1.038
Lifetime alcohol T5 → lifetime CU T5	0.183	0.782 (0.164)	< 0.001	0.461	1.103
Lifetime nicotine T1 → lifetime CU T1	0.066	0.202 (0.164)	0.217	−0.119	0.523
Lifetime nicotine T2 → lifetime CU T2	0.061	0.174 (0.130)	0.178	−0.080	0.429
Lifetime nicotine T3 → lifetime CU T3	0.068	0.182 (0.096)	0.059	−0.007	0.371
Lifetime nicotine T4 → lifetime CU T4	0.076	0	0.016	0.037	0.361
Lifetime nicotine T5 → lifetime CU T5	0.083	0.221 (0.082)	0.007	0.061	0.381
THC toxicology T1 → DM T1	0.046	1.445 (1.688)	0.392	−1.864	4.754
THC toxicology T3 → DM T3	−0.022	−0.473 (0.998)	0.635	−2.430	1.483
THC toxicology T5 → DM T5	−0.004	−0.074 (0.818)	0.928	−1.677	1.529

DM = decision-making; CU = cannabis use; CI = confidence interval; THC = tetrahydrocannabinol; bidirectional arrows represent correlations and unidirectional arrows represent regression paths.

Bold-type estimates indicate significance at P <0.05.

CU and immediate memory

Table 4 lists estimates for the covariate-adjusted parallel process model of CU and immediate memory. The intercepts were significantly and negatively correlated (P = 0.038), as shown in Fig. 2. This was a modestly sized effect, and indicated that a greater number of lifetime days of CU reported at baseline was associated with worse immediate recall at baseline. In addition, contrary to what was observed with DM, the CU slope predicted the immediate memory slope (P = 0.020). This effect was moderate in size and indicated that participants with greater escalation in CU showed lesser gains in memory performance during the 2-year study period, consistent with our hypotheses. No significant associations were observed between the immediate memory intercept and the CU slope (P = 0.088), as well as between the CU intercept and immediate memory slopes (P = 0.834), such that baseline values of one variable did not predict change in the other.

Table 4.

Detailed estimates for covariate-adjusted parallel process model of cannabis use and immediate memory.

Parameter	Standardized estimate	Unstandardized estimate (standard error)	P-value	Lower 95% CI	Upper 95% CI
ImmMem intercept ←→ CU intercept	−0.146	−14.838 (7.156)	0.038	−28.864	−0.812
CU intercept → ImmMem slope	0.028	0.001 (0.005)	0.834	−0.010	0.012
ImmMem Intercept → CU slope	−0.095	−0.075 (0.044)	0.088	−0.161	0.011
CU slope → ImmMem slope	−0.333	−0.051 (0.022)	0.020	−0.094	−0.008
CU intercept ←→ CU slope	0.532	44.947 (6.430)	< 0.001	32.343	57.551
ImmMem intercept ←→ ImmMem slope	−0.003	−0.011 (0.659)	0.987	−1.302	1.280
Sex→ CU intercept	−0.147	−5.410 (1.856)	0.004	−9.047	−1.773
Sex→ CU slope	−0.212	−2.066 (0.516)	< 0.001	−3.077	−1.055
Sex→ ImmMem intercept	0.049	0.608 (0.741)	0.412	−0.845	2.061
Sex→ ImmMem slope	0.086	0.129 (0.173)	0.458	−0.211	0.469
Baseline age→ CU intercept	0.129	3.278 (1.094)	0.003	1.134	5.422
Baseline age→ CU slope	0.099	0.672 (0.356)	0.059	−0.025	1.369
Baseline age→ImmMem intercept	−0.009	−0.074 (0.507)	0.885	−10.067	0.920
Baseline age→ ImmMem slope	0.141	0.147 (0.117)	0.211	−0.083	0.376
Estimated IQ → ImmMem intercept	0.394	0.163 (0.024)	< 0.001	0.116	0.210
Estimated IQ → ImmMem slope	0.105	0.005 (0.006)	0.362	−0.006	0.017
Lifetime alcohol T1 → lifetime CU T1	0.200	0.737 (0.215)	0.001	0.317	1.158
Lifetime alcohol T2 → lifetime CU T2	0.179	0.680 (0.184)	< 0.001	0.319	1.040
Lifetime alcohol T3 → lifetime CU T3	0.173	0.694 (0.157)	< 0.001	0.386	1.002
Lifetime alcohol T4 → lifetime CU T4	0.174	0.725 (0.161)	< 0.001	0.408	1.041
Lifetime alcohol T5 → lifetime CU T5	0.185	0.791 (0.161)	< 0.001	0.476	1.107
Lifetime nicotine T1 → lifetime CU T1	0.066	0.203 (0.164)	0.215	−0.118	0.525
Lifetime nicotine T2 → lifetime CU T2	0.060	0.172 (0.129)	0.182	−0.081	0.426
Lifetime nicotine T3 → lifetime CU T3	0.066	0.178 (0.096)	0.063	−0.010	0.366
Lifetime nicotine T4 → lifetime CU T4	0.074	0.193 (0.082)	0.019	0.032	0.354
Lifetime nicotine T5 → lifetime CU T5	0.081	0.214 (0.081)	0.008	0.055	0.373
THC toxicology T1 → ImmMem T1	−0.032	−1.183 (1.064)	0.266	−3.268	0.902
THC toxicology T3 → ImmMem T3	−0.008	−0.210 (10.067)	0.844	−2.301	1.881
THC toxicology T5 → ImmMem T5	0.028	0.548 (0.749)	0.464	−0.920	2.016

ImmMem = immediate memory; CU = cannabis use; CI = confidence interval; THC = tetrahydrocannabinol.

Bidirectional arrows represent correlations and unidirectional arrows represent regression paths.

Bold-type estimates indicate significance at P < 0.05.

Figure 2.

Covariate-adjusted parallel process LGCM of cannabis use and immediate memory (${\chi }_{(133)}^2=561.47$, P < 0.05; CFI = 0.91; RMSEA = 0.10, SRMR = 0.13; AIC = 17 650.94; SABIC = 17 677.79). Paths are presented in unstandardized metric, and T1–T5 refer to the assessment time-points. CFI = comparative fit index; RMSEA = root mean square error of approximation; AIC = Akaike information criterion; SABIC = sample-size adjusted Bayesian information criterion; LGCM = latent growth curve modeling; SRMR = standardized root mean square residual

CU and delayed memory

As shown in Fig. 3, the correlation between the CU intercept and the delayed memory intercept was negative and significant (P = 0.021), consistent with hypotheses that participants with greater CU at baseline also demonstrated worse performance on delayed recall trials at baseline. Surprisingly, the delayed memory intercept was negatively and significantly associated with the CU slope (P = 0.006), such that worse performance on delayed recall trials predicted greater escalation in CU during the course of the study, even after controlling for estimated IQ. Associations between the CU intercept and delayed memory slope and between the CU and delayed memory slopes were not significant (Ps = 0.529 and 0.213, respectively), suggesting that neither baseline nor change in CU predicted changes in delayed memory performance. Detailed estimates for this model are shown in Table 5.

Figure 3.

Covariate-adjusted parallel process latent growth curve modeling (LGCM) of cannabis use and delayed memory (${\chi }_{(133)}^2=574.33$, P < 0.05; CFI = 0.91; RMSEA = 0.10, SRMR = 0.13; AIC = 17 659.63; SABIC = 17 686.47). Paths are presented in unstandardized metric, and T1–T5 refer to the assessment time-points. CFI = comparative fit index; RMSEA = root mean square error of approximation; AIC = Akaike information criterion; SABIC = sample-size adjusted Bayesian information criterion; SRMR = standardized root mean square residual

Table 5.

Detailed estimates for covariate-adjusted parallel process model of cannabis use and delayed memory.

Parameter	Standardized estimate	Unstandardized estimate (standard error)	P-value	Lower 95% CI	Upper 95% CI
DelMem intercept ←→ CU intercept	−0.159	−16.437 (7.104)	0.021	−30.362	−2.513
CU intercept → DelMem slope	0.109	0.004 (0.006)	0.529	−0.008	0.016
DelMem Intercept → CU slope	−0.156	−0.122 (0.045)	0.006	−0.210	−0.034
CU slope → DelMem slope	−0.216	−0.029 (0.023)	0.213	−0.075	0.017
CU intercept ←→ CU slope	0.522	43.652 (6.308)	< 0.001	31.287	56.016
DelMem intercept → DelMem slope	0.004	0.014 (0.613)	0.981	−1.187	1.215
Sex→ CU intercept	−0.147	−5.406 (1.856)	0.004	−9.044	−1.768
Sex→ CU slope	−0.212	−2.071 (0.512)	< 0.001	−3.074	−1.068
Sex→ DelMem intercept	0.049	0.619 (0.739)	0.402	−0.830	2.068
Sex→ DelMem slope	0.021	0.027 (0.168)	0.871	−0.302	0.356
Baseline age→ CU intercept	0.129	3.276 (1.095)	0.003	1.130	5.422
Baseline age→ CU slope	0.095	0.642 (0.354)	0.070	−0.052	1.336
Baseline age→ DelMem intercept	−0.004	−0.039 (0.505)	0.939	−10.029	0.952
Baseline age→ DelMem slope	0.191	0.172 (0.109)	0.115	−0.042	0.387
Estimated IQ → DelMem intercept	0.387	0.163 (0.024)	< 0.001	0.115	0.210
Estimated IQ → DelMem slope	0.033	0.001 (0.005)	0.791	−0.009	0.012
Lifetime alcohol T1 → lifetime CU T1	0.200	0.737 (0.214)	0.001	0.318	1.155
Lifetime alcohol T2 → lifetime CU T2	0.178	0.678 (0.184)	< 0.001	0.318	1.039
Lifetime alcohol T3 → lifetime CU T3	0.173	0.692 (0.158)	< 0.001	0.383	1.001
Lifetime alcohol T4 → lifetime CU T4	0.174	0.723 (0.162)	< 0.001	0.404	1.041
Lifetime alcohol T5 → lifetime CU T5	0.185	0.789 (0.162)	< 0.001	0.472	1.107
Lifetime nicotine T1 → lifetime CU T1	0.067	0.207 (0.164)	0.208	−0.115	0.528
Lifetime nicotine T2 → lifetime CU T2	0.061	0.175 (0.130)	0.178	−0.080	0.430
Lifetime nicotine T3 → lifetime CU T3	0.067	0.181 (0.096)	0.062	−0.009	0.369
Lifetime nicotine T4 → lifetime CU T4	0.074	0.195 (0.082)	0.018	0.034	0.356
Lifetime nicotine T5 → lifetime CU T5	0.082	0.216 (0.081)	0.008	0.056	0.375
THC toxicology T1 → DelMem T1	−0.018	−0.654 (0.926)	0.480	−2.469	1.161
THC toxicology T3 → DelMem T3	−0.017	−0.440 (0.932)	0.637	−2.266	1.387
THC toxicology T5 → DelMem T5	0.008	0.144 (0.825)	0.862	−1.473	1.760

DelMem = delayed memory; CU = cannabis use; THC = tetrahydrocannabinol; CI = confidence interval.

Bidirectional arrows represent correlations and unidirectional arrows represent regression paths.

Bold-type estimates indicate significance at P < 0.05.

DM and CUD onset

Of the 226 cannabis users without a history of CUD included in these analyses, a total of 101 developed a CUD during the course of the study, representing 964 person–time units of observation [incidence rate = 10.5, 95% confidence interval (CI) = 8.6, 12.7]. Predicted hazards of CUD onset at each wave of data collection are presented in the Supporting information. Adjusted discrete time survival models of the association between baseline DM and risk of CUD occurrence showed that DM did not predict CUD onset (adjusted OR = 1.01, 95% CI = 0.98 1.06). Adjusted predicted hazards of CUD onset across scores of DM at baseline are depicted in Fig. 4.

Figure 4.

Results of covariate-adjusted discrete time survival (DTS) models showing predicted hazards of cannabis use disorder (CUD) onset among adolescent cannabis users across scores of decision-making at baseline. The light grey area represents 95% confidence intervals

DISCUSSION

The current study examined longitudinal bidirectional associations between CU and DM among adolescents to determine whether DM impairments may be a risk factor for problematic cannabis use, a consequence of increased use or both. To ascertain whether findings were specific to DM, these aims were also examined in relation to episodic memory, which has consistently been reported to be a consequence of (rather than risk factor for) cannabis use. Consistent with our hypotheses, our results revealed that greater CU was associated with poorer episodic memory at baseline and greater escalation in use predicted lesser gains in immediate episodic memory performance over time. Conversely, baseline DM performance did not predict escalation in cannabis use, nor did escalation in CU predict changes in DM performance. In addition, baseline DM performance failed to predict the development of a CUD during the course of the study. These findings remained unchanged after controlling for important confounds, such as age, sex, estimated IQ, concurrent use of alcohol and nicotine and acute THC intoxication. Thus, in our cohort of adolescent cannabis users, no evidence was observed to support DM as a risk factor or consequence of cannabis use.

Even after controlling for effects of subacute intoxication via oral fluids testing for THC, adolescents reporting more lifetime days of CU showed worse immediate memory at baseline, and those with greater escalation in their use showed lesser improvements in immediate memory performance over time. Conversely, delayed memory was impacted cross-sectionally at baseline, but not longitudinally, consistent with other studies finding adverse effects of cannabis on the learning of new information, rather than with forgetting previously learned information [77,78]. Nonetheless, this does not rule out the possibility that delayed recall may also be affected with continued escalation. Overall, our results are consistent with evidence that CU results in the down-regulation of cannabinoid type 1 (CB1) receptors, which are found in high density in hippocampal regions [34]. We also replicate findings from a large body of work suggesting that CU results in small to moderate adverse effects on episodic memory [50–52] even in a young adolescent sample, with a more limited history of cannabis use.

Contrary to our hypotheses, our results did not support a role for DM as either a risk factor for, or a consequence of, increased CU among adolescents. Our findings are therefore not consistent with those of other longitudinal studies documenting associations between CU and poorer DM among adults [21,29]. These discrepancies could be explained, at least in part, by the young age of our sample, as both these studies involved cannabis users who were older and thus had heavier, more extensive histories of use than the current adolescent sample. It is possible that results may differ in an older sample with more extensive history of CU and a larger proportion of individuals with severe CUDs. That said, we note that the proportion of users in our sample who developed a CUD within the 2-year follow-up period was significantly higher than that typically reported for adults in national samples [2,3]. Another possibility is that individual differences in ongoing neuromaturational processes [79,80], particularly in frontal brain regions important for DM, may obfuscate results due to resulting variability in DM both cross-sectionally and longitudinally. In addition, prior work suggests that measures of impulsive DM may not be as consistently associated with substance use among humans as other aspects of impulsivity, such as behavioral disinhibition [81], which casts greater uncertainty over the contribution of DM to adolescent cannabis use. However, we note that our study focused upon escalation in CU and transition to CUD rather than initiation of use. Studies that include measures of DM and begin with a cohort of participants prior to substance use initiation could test this hypothesis.

Contrary to some of our prior cross-sectional work with adult participants, we did not find an association between DM and CUD in the current study. Gonzalez et al. [41] reported that poorer DM predicted a higher number of CUD symptoms among young adult cannabis users despite no between-group differences in DM performance. In addition, for those with poor DM, greater CU frequency predicted more cannabis-related problems [42]. However, when we explored this with our current adolescent sample, we found no associations between baseline DM and the development of a CUD during the course of the study. Again, it is possible that the young age and relatively shorter histories of use in our sample may have contributed to these null findings. Further, although a number of our participants developed a CUD during the course of the study, it is important to note that most of the CUDs reported were of lower severity, with only a small subset of participants developing a severe CUD (i.e. ‘dependence’ based on DSM-IV nomenclature). Whether these associations are detectable in samples with more severe use patterns remains to be determined. Nevertheless, we note that use patterns in our cohort are more representative of those typically seen among adolescents [1]. Future studies may also expand to test associations with other CUD severity indices, such as symptom counts, as outcomes. Potential moderators of these associations, including levels of CU (e.g. frequency or amount) and presence of a mental health disorder, should also be explored [48]. Finally, given prior findings that DM task-related brain activation can predict later adverse outcomes among cannabis users [37,45,49,50], future studies should incorporate neuroimaging methods to assess DM in these populations.

Our study has several notable strengths, including the relatively large sample of young adolescents, the use of longitudinal LGCMs to examine bidirectional associations over time and the inclusion of several theoretically important confounding variables, including use of other substances, as covariates. In addition, the substance use characteristics of our sample over time suggest that our specific inclusion/exclusion criteria were successful in recruiting a cohort of adolescents for whom cannabis was the primary drug of choice, and who escalated in their CU over time. Nevertheless, our findings must be interpreted in light of several limitations. First, our cohort was a convenience sample screened to over-represent cannabis users and to minimize the presence of significant mental health and other substance use confounds that could affect neurocognition. Participants in our sample self-selected into the study and were recruited primarily from surrounding Miami-Dade County schools. We did not apply a sampling strategy to ensure representation of national demographics in our sample, which was predominantly white (76.8%) and of Hispanic/Latino ethnicity (89.8%). Although this is consistent with the demographic makeup of the greater Miami metropolitan area, it may limit generalizability to other racial, ethnic and cultural groups. However, we have no evidence to hypothesize that associations between CU and neurocognition may vary across common US ethnic groups. Secondly, to minimize the impact of mental health factors on both substance use and neurocognitive performance, participants were excluded from the study if they self-reported a formal diagnosis of or history of treatment for a psychiatric disorder during screening. National surveys suggest that internalizing disorders are relatively common among adolescents, with lifetime prevalence estimates of 14.3% for mood disorders and 31.9% for anxiety disorders [82]. Given our exclusion criteria, these rates were lower in our sample (5% met criteria for lifetime mood disorders and 14% for lifetime anxiety disorder at baseline). It is possible that associations between cannabis use and neurocognition may differ in samples with higher levels of internalizing psychopathology, considering the known effects of these factors on both substance use and neurocognition [83–86]. Nevertheless, it should be noted that enrolled participants were not excluded if they developed a psychiatric disorder during the course of the study, and approximately one-quarter of the sample met criteria for a lifetime internalizing disorder at the final assessment. Thirdly, despite observed escalation in cannabis use, the majority of participants in the sample had limited histories of CU due to their young age and did not report heavy, daily CU (e.g. average monthly use ranged of CU ranged from 4.6 days at baseline to 8.7 days at the final assessment). More work is needed to determine whether these results would differ in older samples with more chronic histories of heavy cannabis use. In addition, our primary measures of DM and episodic memory across analyses consisted of averages of several different indices. This method may obscure important task-specific information, but is less likely to result in Type 1 error. Future studies may explore relationships between CU and performance in individual tasks assessing more finely grained aspects of DM. Lastly, although participants were asked to abstain from drugs for 24 hours prior to each assessment and assessments were not completed if participants appeared intoxicated, a subset of the sample tested positive for THC in oral fluids at in-person assessments (see Table 1). Although we covaried for positive toxicology results for THC, given limits of detection for oral fluids testing it is possible that some participants may have experienced subacute effects at the time of testing and therefore exacerbated the magnitude of observed effects on memory performance in the current study.

In conclusion, the current study replicates prior findings that poorer immediate episodic memory is a consequence of escalating CU and extends this to a young adolescent sample at relatively early stages of use using a longitudinal parallel process design. Poorer memory among adolescent cannabis users may explain associations between CU and poorer academic outcomes [87,88] and further supports the contention that CU is not benign among adolescent users. However, our results do not support a role for DM as either a consequence of, or a risk factor for, escalating use or the development of a CUD among cannabis-using adolescents. Examination of potential moderators of the associations between CU and DM may reveal for whom and under what circumstances may DM relate to escalating CU and adverse cannabis-related outcomes.

Supplementary Material

Supplementary files

Table S1 Raw scores on neurocognitive measures at each assessment wave.

Table S2 Characteristics of the subsample used in DTS analyses at baseline (n = 226).

Table S3 Predictive hazards of CUD onset at each wave of observation (results from adjusted DTS models).

Figure S1 Unadjusted parallel process LGCM of cannabis use and decision-making (χ²(28) = 345.06, P < .05; CFI = .93; RMSEA = .17, SRMR = .04; AIC = 19511.87; SABIC = 19525.00).

Figure S2 Unadjusted parallel process LGCM of cannabis use and immediate memory (χ²(28) = 317.40, P < .05; CFI = .94; RMSEA = .16, SRMR = .02; AIC = 19625.09; SABIC = 19638.22).

Figure S3 Unadjusted parallel process LGCM of cannabis use and delayed memory (χ²(28) = 322.13, P < .05; CFI = .94; RMSEA = .16, SRMR = .02; AIC = 19619.82; SABIC = 19632.95).