Introduction

Childhood attention-deficit/hyperactivity disorder (ADHD) carries risk for elevated substance use and substance use disorder (SUD) by adulthood. Stimulant medications, a first-line treatment for ADHD, should decrease substance use given the prominence of impulsivity in models of addiction and the acute efficacy of stimulants for ADHD symptoms, including impulsivity. However, early exposure to stimulants may cause neurobiological and behavioral sensitization to other drugs and thus increase the risk for harmful substance use. A 2013 meta-analysis found no associations consistent with either protective or adverse effects of stimulants on substance use or SUD. After further individual studies with mixed results, a 2014 meta-analysis supported protective associations for cigarette smoking, and large epidemiologic studies demonstrated protective associations for SUD. In all observational studies, a vexing methodologic barrier to inferring causal effects of stimulants on substance use is the confounding influence of factors presumed to drive ADHD treatment and substance use. Recent studies have increased scientific rigor, yet none have comprehensively and prospectively assessed both substance use and the wide range of demographic, clinical, and psychosocial factors contributing to both substance use and treatment. The current study was designed to remedy this.

Age confounds the stimulant–substance use association. As substance use escalates through adolescence, well before most SUD diagnoses, adolescents with ADHD are increasingly unlikely to continue taking prescribed stimulants. Thus, without adjustment for age, associations between stimulant medication and substance use may be spurious. Groenman and colleagues found no differences in adolescent daily smoking across age-matched stimulant treatment profiles. A further complication is that some adolescents resume stimulant medication after months or years. A within-participant analysis of filled stimulant and atomoxetine prescriptions found that substance use–related emergency department visits were less frequent during medicated, vs unmedicated, periods for adolescents and adults with ADHD. However, data excluded frequent substance use, class of substance was not measured, data were limited to the commercially insured, and most substance use does not lead to hospitalization. Thus, additional research must address the effects of age through adolescence to the peak period of substance use (early adulthood), specificity of substance use class, and medication continuation, cessation, resumption, or stable termination.

Other factors driving both stimulant treatment and substance use are numerous, including sociodemographics, symptom severity, psychiatric comorbidities, functional impairment, psychosocial treatment and factors affecting it (eg, insurance), and parent characteristics (psychopathology, parenting practices/attitudes). Some confounders have been covaried in prior studies but only at single time points. Adjustment for changing influences over time is needed to test another popular treatment hypothesis: that younger age of treatment initiation and longer duration improve substance use outcome. For the first time, to our knowledge, we apply a causal analytic method, marginal structural models, to address this dynamic confounder problem.

In this study, we use prospective longitudinal data from a randomized multisite clinical trial of ADHD treatments in childhood to study the association between stimulant treatment and substance use through adolescence into early adulthood. We focus on substance use due to its critical role in SUD ontogeny. We implement a 2-part analytic series investigating the impact of age adjustment when examining contemporaneous associations between stimulants and substance use (part 1) and whether cumulative stimulant treatment from childhood through early adulthood predicts early adult substance use by comprehensively adjusting for time-varying covariates that may underlie stimulant–substance use associations (part 2).

Methods

Participants were children aged 7 to 9 years with DSM-IV combined-type ADHD recruited between 1994 and 1996 from 6 sites in the US and 1 in Canada. Children (95-99 per site) were randomly assigned to 1 of 4 treatment groups: medication management, multicomponent behavior therapy, their combination, or referral to usual community care. The Multimodal Treatment Study of ADHD (MTA) timeline, recruitment, diagnostic procedures, treatment, and sample demographics appear in prior publications. Demographics, including race and ethnicity, were reported by parents at baseline when participants were children. Informed consent was monitored by the 7 respective university institutional review boards. Treatment group assignment was not associated with later substance use.

Participants were assessed at baseline prior to randomization, at 3 and 9 months, at conclusion of the 14-month treatment, and at 2, 3, 6, 8, 10, 12, 14, and 16 years after baseline. Substance use data were provided at least once in adulthood (12, 14, and/or 16 years after baseline) by 81.3% of participants (mean [SD] age, 25.1 [1.07] years at 16-year assessment) and 95% were reassessed at least once between the 2- and 16-year assessments. Number of waves participated and substance use were nominally associated; these results and wave-specific retention are available elsewhere.

Measures

Substance Use

The self-report Substance Use Questionnaire, adapted for the MTA, included frequency of use for alcohol, marijuana, cigarettes, and several illicit and prescription drugs. Substance Use Questionnaire data were from the 2-year (mean [range] age, 10.5 [8.8-12.5] years) through 16-year (mean [range] age, 25.1 [22.5-28.6] years) assessments. Our harmonization procedure across development, and additional measure description, is detailed by Molina et al. A National Institute of Health Certificate of Confidentiality facilitated honest reporting.

Heavy drinking was the higher score of 2 items assessing frequency of binge drinking and drunkenness (How many times…[1] did you drink 5 or more drinks when you were drinking? …[2] have you gotten drunk or very, very high on alcohol?). Marijuana use was assessed with 1 frequency of use question. Past-year frequency ratings for each of these variables were coded to 1 of 4 levels: 0 (none), 1 (<once per month), 2 (at least monthly; less than weekly), and 3 (once per week or more). Daily smoking was a binary variable (0 = no, 1 = yes) reflecting 1 or more cigarettes per day. For other substance use, participants reported misuse of prescription medications including stimulants, sedatives, and opioids and use of heroin, inhalants, hallucinogens, cocaine, and “other substances to get high.” Maximum frequency of any illicit substance use (or prescription drug misused) was calculated; past-year frequency was coded to 1 of 3 levels: 0 (none), 1 (<once per month), and 2 (once per month or more often).

Stimulant Treatment

From the 14-month assessment to age 18 years, stimulant treatment information was collected via the Services for Children and Adolescents Parent Interview. After age 18 years, participants self-reported treatment. Stimulant treatment was analyzed as a binary variable reflecting a minimum of 10 mg/d d,l-methylphenidate–equivalent of any stimulant for 50% or more of the days in the past year across the longitudinal follow-up (see eMethods 1 in Supplement 1 for details and rationale). Medication use prior to entering the MTA, retrospectively reported by parents, was binary (presence = 1, absence = 0). Prior publications extensively describe medication assessments.

Analytic Plan

Parts 1 and 2 of this study used the substance use and stimulant medication measures described above. Part 1 models added age and socioeconomic disadvantage in a hierarchical fashion. These analyses tested the hypothesis that contemporaneous (past-year), interacting with recent (1 year prior) medication status would be associated with substance use before and after controlling for age. Part 2 tested the association of cumulative stimulant medication use with adult substance use using causal inference methods that adjusted for potential time-varying confounds. We repeated these analyses for alcohol, marijuana, and other SUD to check extension of our results to diagnosable SUD.

Part 1

We estimated generalized multilevel linear models for ordered categorical outcomes in 4 steps. First, effects of stimulant medication use at the current assessment (past-year), year prior, and their interaction were estimated to capture unadjusted effects of concurrent stimulant use and patterns of medication continuation and discontinuation. Second, we added index variables marking the passage of time over assessments and capturing differential rates of change through adolescence and early adulthood (as in Molina et al² ). Third, we added participants’ ages relative to others at the same assessment. Finally, we added baseline parents’ combined household income (1, <$10 000 to 9, ≥$75 000) and household (dis)advantage covariates reflecting family structure (single- vs 2-parent households) and parents’ education (see eMethods 2 in Supplement 1 for variable coding details). Models accommodated missing data with full information maximum likelihood estimation.

Part 2

We fit a series of marginal structural models that adjusted for 75 covariates (eTable 1 in Supplement 1) that may confound the estimated effect of stimulant use on substance use, vis-à-vis their associations with both stimulant treatment and substance use. Marginal structural models yield unbiased estimates of causal effects assuming ignorability (ie, no unmeasured confounding), positivity (ie, individuals have a nonzero probability of treatment), and consistency (ie, an individual’s treatment status does not affect another individual’s outcome). Covariates included baseline variables and time-varying covariates measured repeatedly. The substance use outcomes for these analyses were the 16-year observations. We used multiple imputation for missing data from baseline to the 16-year assessment. We used a bayesian multiple imputation by chained equations approach for nested data in Mplus (version 8.3), saving 100 imputed data sets. We then estimated time-varying propensity scores reflecting individuals’ propensities to take stimulant medications at each assessment based on the covariates (propensity scores near 1 or 0 indicate that taking stimulant medication is extremely likely or unlikely, respectively). We used the super learner package in RStudio version 2022.07.0 (R Foundation) (a suite of machine learning tools) to derive a flexible propensity score model comprising the covariates in eTable 1 in Supplement 1 (eg, determine necessary interactions among covariates, specify covariates’ functional forms) without overfitting to the data. Finally, we estimated linear probability regression models for substance use outcomes at the 16-year assessment. Estimated effects were adjusted for confounders by weighting cumulative stimulant treatment, stimulant treatment prior to study entry, and their interaction by the propensity scores. Specifically, they were weighted by the product of the inverse of the time-varying propensity scores, which were stabilized to protect against potential bias due to observations with extreme probabilities. Thus, stimulant treatment became statistically disaggregated from the covariates, mirroring random treatment assignment, to ensure that any remaining differences in substance use among individuals with varying levels of stimulant treatment over time must be due to stimulant treatment itself; thus, a causal effect of cumulative stimulant treatment on substance use could be estimated under the assumptions described above). First, we estimated the effect of total cumulative years of stimulant medication exposure across the follow-up period. Second, we estimated the effect of the number of consecutive years of stimulant medication exposure. Analyses were conducted in RStudio version 2022.07.0 (RStudio PBC).

Two-sided P values were statistically significant at .05. All study questions and analysis plans are documented elsewhere.³² The part 2 strategy was preregistered before data analysis.⁴⁴ Analysis took place between April 2018 and February 2023.

Results

Rates of Substance Use and Stimulant Medication Use

A total of 579 children with DSM-IV combined-type ADHD were recruited with a mean (SD) age of 8.5 (0.8) years at baseline. A total of 115 participants (20%) were African American; 48 (8%), Hispanic; and 351 (61%), White. The Figure shows the rates of substance use and stimulant use across all assessments (and mean age by assessment). As reported elsewhere, substance use increased steadily through adolescence and remained stable through early adulthood. Mean percentages across the 12-, 14-, and 16-year follow-up assessments were 36.5% for daily smoking, 29.6% for marijuana use at least weekly, 21.1% for heavy drinking at least weekly, and 6.2% for other substance use at least monthly (eTable 2 in Supplement 1). The Figure shows that the share of adolescents using stimulant medication declined precipitously through adolescence from nearly 60% at the 2- and 3-year assessments to 7.2% on average in early adulthood (eTable 2 in Supplement 1).

Figure. Rates of Stimulant and Substance Use at the 2-Year Through 16-Year Follow-up Assessments of the Multimodal Treatment Study of Attention-Deficit/Hyperactivity Disorder

Part 1

Initial models predicting substance use from contemporaneous and prior year stimulant medication use and their interaction showed that stimulant use was associated with reduced odds of heavy drinking, using marijuana, daily smoking, and other substance use (Table 1, model 1). Consistent with stimulant use patterns described above, however, these associations were no longer present after accounting for developmental change in substance use through adolescence into early adulthood (model 2; eTable 3A-D in Supplement 1). Model 3 added relative age, and the same pattern emerged: neither contemporaneous nor prior year stimulant medication use was associated with any substance outcome, and the absence of interaction effects indicated no evidence of association between medication continuation or discontinuation with elevated or reduced substance use. In general, standard errors for stimulant use effects were large, indicating low precision in the estimates of stimulant-substance use associations.

Table 1. Selected Results of Generalized Multilevel Linear Models Testing Effect Sizes of Stimulant Use on the Log Odds of Heavy Drinking, Marijuana Use, Daily Smoking, and Other Substances Used, When Time and Age Trends are Absent (Model 1) vs Present (Model 3)

In addition to the substance use trends, the estimated effects of relative age showed that youth who were older than average within a given assessment were most likely to use substances. For example, the odds of daily smoking in a given year were 2.5 times higher for people a year older than average during that assessment (Table 1, model 3: B = 0.92; odds ratio, 2.51). The same pattern persisted after adding income and household (dis)advantage in model 4 (eTable 3A-D in Supplement 1). Given the absence of any statistically significant estimated effects of stimulant medication at this stage. we did not pursue additional sociodemographic, clinical, or familial covariates. These were comprehensively assessed in part 2.

Part 2

As displayed in Table 2, results of the marginal structural models provided no evidence of an effect of cumulative years of stimulants with daily smoking, marijuana use, or other substance use at a mean of age 25 years. Each additional year of cumulative stimulant medication was estimated to increase individuals’ likelihood of any binge drinking/drunkenness vs none in the past year by 4% and of monthly vs less frequent binge drinking/drunkenness by 3.5% (95% CI, 0.01-0.08; P = .03 and 95% CI, 0.0003-0.07; P = .048, respectively). However, using a familywise Bonferroni-corrected α level for multiple tests (P = .006), these 2 effects (of 18) became nonsignificant. Results showed no evidence that stimulant medication use prior to study entry or its interaction with cumulative years of stimulants increased the likelihood of any substance use at a mean age of 25. Similarly, in Table 3, results did not provide evidence that more years of continuous, uninterrupted stimulant medication use or its interaction with stimulant use prior to study entry had effects on any substance use at a mean age of 25 years. Results accounted for inconsistently medicated treatment patterns (see full models in eTable 4 in Supplement 1).

Table 2. Marginal Structural Model Results: Cumulative Stimulant Medication Use^a

Table 3. Marginal Structural Model Results: Continuous, Uninterrupted Stimulant Medication Use^a

Results from parts 1 and 2 were robust to substitution of substance use with alcohol, marijuana, and other SUD (eTables 5A-C, 6A-B in Supplement 1).

Discussion

Recent reviews concluded that stimulant medication protects against well-established substance use risk for children with ADHD, but critical methodologic limitations were also identified. Using prospective data, the current study addressed 2 key weaknesses of previous reports: (1) adjustment for age across the developmental phases of adolescence and early adulthood when substance use escalates and (2) adjustment for time-varying confounders of the association between stimulant treatment and substance use using causal inference methods.

Part 1 analyses, correcting for age, did not provide evidence that individuals currently or recently taking stimulants report lower rates of substance use than individuals not taking stimulants, although large standard errors prevent us from completely ruling out possible protective (or harmful) effects. Findings highlighted opposing developmental trends in substance use and stimulant medication through adolescence into early adulthood: as substance use escalates, stimulant medication use declines. Extensive research has documented these trends separately but never adjusted for them when studying association of stimulant use with substance use. Without adjusting for developmental trends, we might have mistakenly concluded a protective association in our longitudinal analyses. We also found no contemporaneous by prior year stimulant treatment interaction, indicating no evidence that patterns of stimulant continuation or discontinuation were related to substance use.

The results suggest that more and different treatment may be needed for individuals with ADHD to reduce escalating substance use known to precede SUD. Substance use prevention must begin early when reward seeking and social context drive initial substance use and precede habit formation. Preventive interventions may also need to directly address perceptions that some substances (eg, marijuana) are therapeutic. Our findings may differ from recent US commercial health care claims data because we examined more prevalent substance use behavior vs rare presentations to the emergency department. A Swedish registry study showing protective associations also relied on hospital visits and death and criminal records. However, our results extended to diagnosable SUD, which is more proximal clinically to emergency department visits. Similar to our findings, Groenman and colleagues also did not find protection by any stimulant treatment patterns for daily smoking. The large database studies provide reassurance that the most serious substance use–related end points (eg, hospitalization, death, adjudication) are not elevated as a function of recent stimulant treatment.

Another long-standing hypothesis is that early, continuous stimulant treatment should protect children with ADHD from harmful substance use. Part 2 analyses did not show that longer duration of stimulant treatment predicts less substance use in adulthood. In fact, cumulative stimulant treatment was associated with increased heavy drinking, although the effect size was small and did not survive experimentwise error correction. The marginal structural models adjusted for a comprehensive suite of time-varying confounders measured longitudinally from childhood to adulthood, and results persisted whether or not children were taking stimulant medication before study entry. Even if stimulants were initiated before the mean age of 8 years, our results were not consistent with the hypotheses of protection or harm in relation to substance use or SUDs. Mannuzza and colleagues found that early treatment (before age 9 years) predicted lower SUD in adulthood, and Groenman and colleagues found protection from SUD in 103 adolescents treated early and consistently at a high dose (53.4-mg methylphenidate). Both studies were limited by not comprehensively adjusting for the wide range of confounding variables that drive decreasing stimulant treatment and increasing substance use over time.

A key MTA (and other longitudinal studies of children with ADHD) finding is that most patients no longer take stimulants in adulthood. Yet, if stimulant treatment creates long-term protection, that benefit should extend beyond treatment cessation. (In previous reports, we did find long-term effects in another domain, growth impairment, even though only 53 participants were consistently taking medication to the 10-year assessment, and fewer were taking medication after this point.) Stimulants may not permanently change some important substance use risk processes, including impulsivity and poor delay of gratification. Tamminga et al recently reported that methylphenidate-driven improvements in cognitive performance, including response inhibition and delay aversion while receiving medication, disappeared once treatment ceased. There still remain possible benefits to combined pharmacotherapy (including stimulants), psychoeducation, and psychotherapy for individuals with current ADHD and SUD.

Limitations

Limitations include that we lacked medical records to verify medication history, although confidence in our data is supported by studies showing concordance between parent report and medical records. Nonmarijuana illicit substance use and prescription misuse rates were very low, precluding testing associations within classes of other substance use. Finally, although our multisite sample included female individuals (20%) and children identified as African American (20%) and Hispanic (8%), it was insufficiently powered for a strong test of moderation by sex, race, or ethnicity.

Conclusions

Taken together, our 16-year prospective, comprehensive, and developmentally sensitive analyses of stimulant medication associations with substance use failed to support any hypotheses of substance use protection or harm from stimulant treatment for ADHD. Although these results contrast with recent conclusions of protection found in other data sets, across all studies the findings lend a measure of comfort in the consistent lack of evidence that stimulant treatment predisposes children with ADHD to later substance use.

Introduction

Individuals diagnosed with attention-deficit/hyperactivity disorder (ADHD) during childhood face an increased likelihood of developing substance use issues and substance use disorder (SUD) in adulthood. Stimulant medications, a primary treatment for ADHD, are theorized to reduce substance use, as they address impulsivity, a known factor in addiction. However, an alternative view suggests that early stimulant exposure could heighten sensitivity to other drugs, thereby increasing the risk of harmful substance use.

Past research has yielded mixed results regarding the link between stimulant treatment and substance use. A 2013 review of multiple studies found no clear protective or harmful association. While a 2014 review suggested a protective link for cigarette smoking, and large population studies indicated protection against SUD, these observational studies face a significant challenge: other factors influencing both ADHD treatment choices and substance use make it difficult to determine a direct cause-and-effect relationship.

Age also complicates understanding the stimulant-substance use relationship. As substance use increases during adolescence, individuals with ADHD often stop taking prescribed stimulants. Without accounting for age, any observed links between stimulant medication and substance use might be misleading. For instance, one study found no difference in adolescent daily smoking among age-matched groups based on stimulant treatment. Additionally, some individuals restart medication years later, and while medication periods were linked to fewer emergency visits for substance use in one study, that data excluded frequent use and specific substance types, limiting its broad applicability.

Many other factors can influence both stimulant treatment and substance use. These include social and economic background, symptom severity, co-occurring mental health conditions, functional challenges, access to psychosocial treatments, and parental factors like mental health and parenting styles. While some studies have attempted to adjust for these influences, they have often done so at a single point in time. A comprehensive adjustment for how these factors change over time is necessary to properly assess whether starting treatment earlier or continuing it longer affects substance use outcomes. This study applies a specific statistical method called marginal structural models for the first time in this context to address this complex problem.

This study uses long-term data from a large clinical trial on ADHD treatments in childhood. The aim is to examine the relationship between stimulant treatment and substance use through adolescence into early adulthood, focusing on how substance use contributes to the development of SUD. The research is divided into two parts: the first investigates the impact of age adjustment on current links between stimulants and substance use, and the second explores whether cumulative stimulant treatment from childhood predicts early adult substance use, thoroughly adjusting for various changing factors that might influence these associations.

Methods

The study included children aged 7 to 9 years who had DSM-IV combined-type ADHD. These children were recruited between 1994 and 1996 from six sites in the US and one in Canada. Participants, with 95-99 children per site, were randomly placed into one of four treatment groups: medication management, multicomponent behavior therapy, a combination of both, or referral to standard community care. More details on the Multimodal Treatment Study of ADHD (MTA) timeline, recruitment, diagnostic methods, treatments, and participant demographics are available in previous reports. Parents provided demographic information, including race and ethnicity, at the start of the study. The institutional review boards at each of the seven universities oversaw the informed consent process. It was noted that the initial treatment group assignment did not predict later substance use.

Participants underwent assessments at various points: at the study's start before randomization, at 3 and 9 months, after the 14-month treatment concluded, and then at 2, 3, 6, 8, 10, 12, 14, and 16 years following the baseline assessment. Substance use data from adulthood (at 12, 14, and/or 16 years post-baseline) were collected from 81.3% of participants, who were, on average, 25.1 years old at the 16-year assessment. Additionally, 95% of participants were re-evaluated at least once between the 2-year and 16-year assessments. The number of assessments a participant completed had a minor link to their substance use; further details on this and participant retention at each assessment point are published elsewhere.

Measures

The study utilized a self-report Substance Use Questionnaire, specifically adapted for the MTA. This questionnaire gathered information on the frequency of use for alcohol, marijuana, cigarettes, and various illegal and prescription drugs. Data from this questionnaire were collected during assessments ranging from 2 years (average age 10.5 years) to 16 years (average age 25.1 years) after baseline. The methods for standardizing these measures across different developmental stages, along with further descriptions, are detailed in a publication by Molina et al. A Certificate of Confidentiality from the National Institute of Health helped ensure participants felt comfortable providing honest information.

Substance Use

Heavy drinking was determined by the higher score from two questions about binge drinking and getting drunk. Marijuana use was measured using a single question about frequency. For both heavy drinking and marijuana use, past-year frequency was categorized into four levels: none, less than once per month, at least monthly but less than weekly, or once per week or more. Daily smoking was recorded as a simple yes/no, indicating one or more cigarettes per day. For other substances, participants reported misuse of prescription medications (stimulants, sedatives, opioids) and use of heroin, inhalants, hallucinogens, cocaine, and other drugs taken to get high. The highest frequency of any illegal substance use or prescription drug misuse was calculated, with past-year frequency categorized into three levels: none, less than once per month, or once per month or more often.

Stimulant Treatment

Information on stimulant treatment was gathered through interviews with parents from the 14-month assessment until participants reached 18 years of age. After age 18, participants self-reported their treatment. Stimulant treatment was considered active if a participant took at least 10 mg/day of d,l-methylphenidate equivalent of any stimulant for 50% or more of the days in the past year during the study's follow-up period. (More detailed information and reasons for this definition are provided in an online supplement). Parents also provided retrospective reports on medication use before participants entered the MTA study, recorded as either present or absent. Earlier publications contain extensive descriptions of how medication use was assessed.

Analytic Plan

This study's two parts utilized the substance use and stimulant medication measurements previously described. In Part 1, statistical models progressively incorporated age and socioeconomic disadvantages. These analyses examined whether stimulant medication use in the past year, in conjunction with use from the prior year, was linked to substance use, both before and after accounting for age. Part 2 investigated the relationship between the total amount of stimulant medication used over time and adult substance use. This part employed causal inference methods, which adjusted for potential confounding factors that changed over time. The analyses were repeated for alcohol, marijuana, and other substance use disorders (SUD) to determine if the findings extended to clinically diagnosable conditions.

Part 1 Generalized multilevel linear models were used to analyze the ordered categorical outcomes, following a four-step process. First, the effects of stimulant medication use in the past year, in the year before, and their combined effect were estimated. This captured both the immediate impact of current use and patterns of continuing or stopping medication. Second, variables were included to mark the passage of time across assessments, accounting for different rates of change throughout adolescence and early adulthood, consistent with previous research. Third, the analysis incorporated participants’ ages relative to other participants at the same assessment point. Fourth, baseline information on parents’ combined household income (ranging from less than $10,000 to $75,000 or more) and factors indicating household advantage or disadvantage, such as family structure (single- or two-parent households) and parents' education, were added. (Detailed variable coding information is available in an online supplement). The models were designed to handle missing data using a statistical method called full information maximum likelihood estimation.

Part 2 A series of marginal structural models were employed to adjust for 75 potential confounding variables that could influence the relationship between stimulant use and substance use, given their links to both treatment and substance use. These models aim to provide unbiased estimates of causal effects, assuming that all relevant confounding factors have been measured, that individuals have a real chance of receiving treatment, and that one person's treatment status does not affect another's outcome. The confounding variables included both baseline data and factors that changed over time, measured repeatedly. The substance use outcomes for these analyses were based on observations from the 16-year assessment.

Missing data from baseline to the 16-year assessment were handled using multiple imputation. A Bayesian multiple imputation approach was used for nested data within Mplus software, creating 100 complete datasets. Following this, time-varying propensity scores were calculated. These scores indicated how likely individuals were to take stimulant medication at each assessment based on the confounding variables, with scores close to 1 meaning highly likely and scores close to 0 meaning highly unlikely.

The "super learner" package in RStudio, a set of machine learning tools, was used to develop a flexible propensity score model. This allowed for identifying necessary interactions among variables and defining their functional forms without making the model too specific to the current data.

Finally, linear probability regression models were used to estimate substance use outcomes at the 16-year assessment. The estimated effects were adjusted for confounding factors by weighting cumulative stimulant treatment, stimulant treatment before study entry, and their interaction using the calculated propensity scores. Specifically, these variables were weighted by the inverse product of the time-varying propensity scores, which were adjusted to prevent bias from observations with very high or low probabilities.

This weighting process effectively separated stimulant treatment from the confounding variables, simulating a random assignment to treatment. This ensured that any remaining differences in substance use among individuals with varying levels of stimulant treatment over time could be attributed directly to the stimulant treatment itself, allowing for an estimation of its causal effect under the stated assumptions. The analysis proceeded in two steps: first, estimating the effect of the total number of years an individual was exposed to stimulant medication throughout the study period, and second, estimating the effect of the number of consecutive years of stimulant medication exposure. All analyses were performed using RStudio software.

A statistical significance level of .05 was used for two-sided P values. All research questions and analysis plans for this study are documented in other publications. Specifically, the strategy for Part 2 was officially registered before any data analysis began. The analysis period for this study spanned from April 2018 to February 2023.

Results

The study included 579 children diagnosed with DSM-IV combined-type ADHD, with an average age of 8.5 years at the start of the study. The participant group included 115 (20%) African American, 48 (8%) Hispanic, and 351 (61%) White individuals. A visual representation illustrates the rates of substance use and stimulant medication use across all assessment periods, alongside the average age at each assessment. As noted in other reports, substance use consistently increased throughout adolescence and then remained stable into early adulthood. Across the 12-, 14-, and 16-year follow-up assessments, the average percentages for specific substance uses were: 36.5% for daily smoking, 29.6% for at least weekly marijuana use, 21.1% for at least weekly heavy drinking, and 6.2% for at least monthly use of other substances. The visual representation also indicates a sharp decline in stimulant medication use among adolescents, dropping from nearly 60% at the 2- and 3-year assessments to an average of 7.2% in early adulthood.

Part 1 Initial statistical models examined the relationship between substance use and stimulant medication use in the current and previous year, including their combined effect. These models initially suggested that stimulant use was linked to a reduced likelihood of heavy drinking, marijuana use, daily smoking, and other substance use. However, these links disappeared once developmental changes in substance use through adolescence and early adulthood were accounted for. When relative age was added to the models, there was still no association found between current or previous year stimulant medication use and any substance use outcome. Furthermore, no interactions were observed, meaning that patterns of continuing or discontinuing medication did not show a clear link to increased or decreased substance use. Overall, the statistical uncertainty for stimulant use effects was high, indicating that the estimates of the stimulant-substance use associations were not very precise.

Beyond substance use trends, the analyses revealed that individuals who were older than the average age within a particular assessment period were more likely to use substances. For example, the likelihood of daily smoking in a given year was 2.5 times higher for individuals who were one year older than the average during that assessment. This pattern remained consistent even after adjusting for income and household advantages or disadvantages. Since no statistically significant effects of stimulant medication were found at this stage of the analysis, further adjustment for additional social, clinical, or family-related factors was not pursued in Part 1. These factors were thoroughly examined in Part 2 of the study.

Part 2 The results from the marginal structural models showed no evidence that cumulative years of stimulant medication had an effect on daily smoking, marijuana use, or other substance use when participants were, on average, 25 years old. Although each additional year of cumulative stimulant medication initially appeared to increase the likelihood of any binge drinking or drunkenness by 4% and monthly binge drinking or drunkenness by 3.5%, these small effects became statistically insignificant when adjusted for the large number of tests performed. Additionally, there was no indication that stimulant medication use before the study began, or its interaction with cumulative stimulant use, increased the likelihood of any substance use by age 25. Similarly, there was no evidence that more years of continuous, uninterrupted stimulant medication use or its interaction with pre-study stimulant use affected any substance use at age 25. The analysis considered various patterns of medication use, including inconsistent treatment.

The findings from both Part 1 and Part 2 remained consistent even when the analysis focused specifically on alcohol, marijuana, and other substance use disorders (SUDs).

Discussion

Recent comprehensive reviews have suggested that stimulant medication may protect children with ADHD from developing substance use problems, a known risk. However, these reviews also highlighted significant methodological weaknesses in the previous research. This current study, utilizing prospective data, aimed to address two major limitations: first, appropriately adjusting for age across adolescence and early adulthood, a period when substance use typically increases; and second, accounting for confounding factors that change over time in the link between stimulant treatment and substance use, using advanced causal inference methods.

The analyses in Part 1, after correcting for age, did not provide clear evidence that individuals currently or recently using stimulants reported lower rates of substance use compared to those not using stimulants. However, the wide range of possible estimates means that small protective or harmful effects cannot be entirely dismissed. The study revealed contrasting developmental patterns: substance use increases through adolescence into early adulthood, while stimulant medication use decreases during the same period. While previous research has documented these trends separately, they had not been jointly considered when examining the association between stimulant use and substance use. Failure to adjust for these developmental trends could lead to incorrect conclusions about a protective association. Furthermore, no interaction was found between current and prior year stimulant treatment, suggesting that patterns of continuing or discontinuing stimulant use were not linked to changes in substance use.

The study's findings suggest that individuals with ADHD may require additional or different types of treatment to mitigate the rising substance use that often precedes SUD. Prevention efforts should start early, during the period when factors like seeking rewards and social influences primarily drive initial substance use and the formation of habits. These interventions might also need to challenge the belief that certain substances, such as marijuana, have therapeutic benefits. The results of this study may differ from those based on recent US health care claims data because this study focused on more common substance use behaviors rather than less frequent events like emergency department visits. Similarly, a Swedish study that reported protective associations relied on data from hospital visits, deaths, and criminal records. However, the current study's findings did extend to diagnosable SUD, which is clinically more closely related to emergency department visits. Consistent with these findings, another study by Groenman and colleagues also found no protective effect of any stimulant treatment patterns on daily smoking. Large database studies do, however, offer reassurance that severe substance use outcomes, such as hospitalization, death, or legal issues, are not increased by recent stimulant treatment.

A long-held belief is that early and continuous stimulant treatment in children with ADHD should protect them from harmful substance use. However, the Part 2 analyses did not demonstrate that longer stimulant treatment duration predicted less substance use in adulthood. While cumulative stimulant treatment showed a small association with increased heavy drinking, this effect was no longer statistically significant after rigorous correction for multiple comparisons. The marginal structural models in this study accounted for a wide array of confounding factors that changed over time, measured from childhood to adulthood. The results remained consistent regardless of whether children had taken stimulant medication before entering the study. Even when considering stimulant initiation before the average age of 8 years, the findings did not support hypotheses of either protection or harm regarding substance use or SUDs. Although some studies, such as those by Mannuzza and Groenman, found protective effects from early or consistent high-dose treatment, these studies were limited because they did not fully adjust for the numerous confounding variables that influence both the decline in stimulant treatment and the rise in substance use over time.

A significant finding from the MTA and other long-term studies of children with ADHD is that most individuals discontinue stimulant use by adulthood. If stimulant treatment offered lasting protection against substance use, this benefit would ideally continue even after medication cessation. It is possible that stimulants do not permanently alter key processes that increase substance use risk, such as impulsivity and difficulty delaying gratification. Recent research by Tamminga et al. indicated that improvements in cognitive functions like impulse control and delay aversion, observed while on methylphenidate, vanished once treatment stopped. Despite these findings, potential benefits may still exist for individuals with co-occurring ADHD and SUD through a combination of medication (including stimulants), education, and psychotherapy.

Limitations

The study had several limitations. One limitation was the absence of medical records to confirm medication history, though other studies have shown that parent reports often align well with medical records, which increases confidence in the data collected. Another limitation was the very low rates of illegal substance use other than marijuana and misuse of prescription drugs. This prevented a detailed analysis of associations within specific classes of these substances. Finally, despite including female participants (20%) and individuals identifying as African American (20%) and Hispanic (8%) across multiple sites, the sample size was not large enough to rigorously test whether the study's findings differed significantly based on sex, race, or ethnicity.

Conclusions

In summary, this 16-year prospective study, which comprehensively and developmentally examined the links between stimulant medication and substance use, found no evidence to support hypotheses that stimulant treatment for ADHD either protects against or causes harm related to substance use. While these results differ from some recent studies that suggested a protective effect, the overall body of research consistently shows no indication that stimulant treatment in childhood increases the risk of later substance use in individuals with ADHD. This provides some reassurance regarding this common treatment.

Introduction

Childhood attention-deficit/hyperactivity disorder (ADHD) increases the risk for higher substance use and substance use disorder (SUD) in adulthood. Stimulant medications are a primary treatment for ADHD. It has been hypothesized that these medications should reduce substance use because addiction often involves impulsive behavior, and stimulants effectively treat ADHD symptoms like impulsivity. However, some theories suggest that early exposure to stimulants might make individuals more sensitive to other drugs, potentially increasing harmful substance use.

A 2013 broad review of studies found no clear evidence that stimulants either protected against or worsened substance use or SUD. Following more individual studies with mixed findings, a 2014 review suggested a protective effect for cigarette smoking. Large population studies later showed protective effects against SUD. In all observational studies, a challenge in determining the true impact of stimulants on substance use has been that other factors, which influence both ADHD treatment and substance use, can confuse the results.

Recent studies have become more rigorous. However, none had comprehensively and consistently evaluated both substance use and the wide range of demographic, clinical, and social factors that contribute to both substance use and treatment. This study was designed to address these gaps.

Age is a complicating factor in understanding the link between stimulant use and substance use. As substance use increases during adolescence—long before most SUD diagnoses—adolescents with ADHD are less likely to continue taking prescribed stimulants. Therefore, without accounting for age, any apparent links between stimulant medication and substance use might be misleading. For instance, one study found no differences in daily smoking among adolescents with different stimulant treatment patterns when age was considered. Another complication is that some adolescents restart stimulant medication after months or years. A study that looked at individual participants found that emergency room visits related to substance use were less common during periods when adolescents and adults with ADHD were taking medication compared to when they were not. However, that study did not include frequent substance use, did not specify the type of substance, only included commercially insured individuals, and most substance use does not lead to hospitalization. Therefore, more research is needed to understand the effects of age across adolescence into early adulthood (when substance use peaks), the specific types of substances used, and patterns of medication use, such as continuation, stopping, restarting, or ending treatment permanently.

Many other factors influence both stimulant treatment and substance use. These include social and demographic characteristics, the severity of ADHD symptoms, other mental health conditions, daily functional difficulties, access to and type of psychosocial treatments (like therapy, influenced by factors such as insurance), and parental characteristics (like mental health or parenting styles). Some of these factors have been adjusted for in previous studies, but only at single points in time. Adjusting for how these influences change over time is necessary to test another common theory: that starting treatment at a younger age and continuing it longer leads to better substance use outcomes. This study used a specific causal analysis method, called marginal structural models, to address this problem of factors changing over time.

In this study, researchers used information gathered over time from a randomized clinical trial of ADHD treatments in childhood. The goal was to examine the link between stimulant treatment and substance use from adolescence into early adulthood, focusing on substance use because of its critical role in the development of SUD. The study had two main parts: the first investigated how accounting for age affects the immediate links between stimulants and substance use, and the second examined whether total stimulant treatment from childhood through early adulthood predicted substance use in early adulthood, by thoroughly adjusting for factors that changed over time and could influence these associations.

Methods

Participants were children aged 7 to 9 years who had combined-type ADHD, based on DSM-IV criteria. They were recruited between 1994 and 1996 from six sites in the US and one in Canada. The children were randomly assigned to one of four treatment groups: medication management, behavior therapy, a combination of both, or referral to usual community care. Details about the Multimodal Treatment Study of ADHD (MTA) timeline, recruitment, diagnostic methods, treatments, and participant demographics have been published previously. Parent reports provided demographic information, including race and ethnicity, at the study's start. The group a child was assigned to did not affect later substance use.

Participants were assessed at the study's start, at 3 and 9 months, after the 14-month treatment period ended, and then at 2, 3, 6, 8, 10, 12, 14, and 16 years after the study began. Substance use data were collected at least once in adulthood (at the 12-, 14-, or 16-year assessments) from 81.3% of participants, who were, on average, 25.1 years old at the 16-year assessment. Ninety-five percent of participants were assessed at least once between the 2-year and 16-year assessments. The number of assessment waves a participant completed was slightly related to substance use, and more details on this and retention rates are available elsewhere.

Substance Use Measures

Substance use was measured using a self-report Substance Use Questionnaire, adapted for the MTA study. This questionnaire asked about the frequency of using alcohol, marijuana, cigarettes, and various illegal and prescription drugs. Data from this questionnaire were collected from the 2-year assessment (when participants were, on average, 10.5 years old) through the 16-year assessment (when participants were, on average, 25.1 years old). A National Institute of Health Certificate of Confidentiality was used to encourage honest reporting of substance use.

Heavy drinking was defined by the higher score from two questions asking about the frequency of binge drinking and drunkenness. Marijuana use was assessed with one frequency question. For these substances, past-year use frequency was categorized into four levels: none, less than once per month, at least monthly but less than weekly, or once per week or more. Daily smoking was a binary variable, indicating one or more cigarettes per day. For other substances, participants reported misuse of prescription medications (stimulants, sedatives, opioids) and use of heroin, inhalants, hallucinogens, cocaine, and other drugs used to get high. The highest frequency of any illicit substance use or prescription drug misuse in the past year was calculated and coded into three levels: none, less than once per month, or once per month or more often.

Stimulant Treatment Measures

Information on stimulant treatment from the 14-month assessment until age 18 was gathered through parent interviews. After age 18, participants self-reported their treatment. Stimulant treatment was considered active if a participant took a minimum daily dose equivalent to 10 mg of d,l-methylphenidate for at least 50% of the days in the past year during the follow-up period. Information on medication use before entering the MTA study was reported retrospectively by parents and was recorded as a binary variable (present or absent). Previous publications provide extensive details on how medication assessments were conducted.

Analytic Plan

This study used two main analytical approaches. Part 1 examined immediate and recent stimulant medication use in relation to substance use, adjusting for age and socioeconomic factors over time. This analysis tested whether concurrent stimulant use and patterns of continuing or stopping medication were linked to substance use, both before and after accounting for age. Part 2 used advanced causal inference methods, specifically marginal structural models, to evaluate the long-term impact of total stimulant treatment from childhood on adult substance use. This involved thoroughly adjusting for a wide range of factors that changed over time and could influence both stimulant use and substance use. These methods aimed to isolate the potential causal effect of stimulants by statistically accounting for complex confounding influences, assuming no unmeasured confounding, that individuals had a non-zero probability of treatment, and that one individual's treatment status did not affect another's outcome. The analyses for Part 2 also considered the total number of years of stimulant exposure and the number of consecutive years of use. For both parts, analyses were also performed for diagnosable SUDs related to alcohol, marijuana, and other substances to see if the findings extended to these more serious outcomes.

Results

Rates of Substance Use and Stimulant Medication Use

A total of 579 children with combined-type ADHD were recruited, with an average age of 8.5 years at the start of the study. Of these participants, 20% were African American, 8% were Hispanic, and 61% were White. Rates of substance use and stimulant use across all assessments were observed, with mean age increasing with each assessment. Substance use steadily increased throughout adolescence and remained stable into early adulthood. Across the 12-, 14-, and 16-year follow-up assessments, average rates were 36.5% for daily smoking, 29.6% for weekly marijuana use, 21.1% for weekly heavy drinking, and 6.2% for monthly use of other substances. The percentage of adolescents using stimulant medication dropped sharply through adolescence, from almost 60% at the 2- and 3-year assessments to an average of 7.2% in early adulthood.

Part 1 Findings

Initial statistical models that predicted substance use based on current and prior year stimulant medication use showed that stimulant use was associated with lower odds of heavy drinking, marijuana use, daily smoking, and other substance use. However, these associations disappeared after accounting for the typical increase in substance use during adolescence into early adulthood. When relative age was added to the models, neither current nor prior year stimulant medication use was associated with any substance use outcome. There was also no evidence that continuing or stopping medication was linked to increased or decreased substance use. Generally, the estimates for stimulant use effects had large uncertainties, suggesting low precision in determining the links between stimulant use and substance use.

In addition to the trends in substance use, results showed that older youth within a given assessment period were more likely to use substances. For example, the odds of daily smoking in a given year were 2.5 times higher for individuals who were one year older than the average during that assessment. This pattern remained consistent even after accounting for income and household factors. Given the absence of any statistically significant effects from stimulant medication at this stage, additional social, demographic, clinical, or family factors were not pursued in this part of the analysis, as they would be comprehensively addressed in Part 2.

Part 2 Findings

Results from the causal models showed no evidence that the total number of years an individual took stimulants affected daily smoking, marijuana use, or other substance use by an average age of 25 years. Each additional year of stimulant medication was estimated to increase an individual's likelihood of any binge drinking or drunkenness by 4% and monthly binge drinking or drunkenness by 3.5%. However, when a stricter statistical threshold was applied to account for multiple tests, these two effects became statistically insignificant. The results also showed no evidence that stimulant medication use before the study began, or its interaction with total years of stimulant use, increased the likelihood of any substance use by age 25. Similarly, there was no evidence that more years of continuous, uninterrupted stimulant medication use, or its interaction with pre-study stimulant use, had an effect on any substance use at age 25. These findings accounted for inconsistent medication patterns.

The results from both Part 1 and Part 2 remained consistent even when diagnosable alcohol, marijuana, and other SUDs were analyzed instead of general substance use.

Discussion

Recent reviews have suggested that stimulant medication protects children with ADHD from substance use risks, but these reviews also highlighted significant study limitations. This study, using data collected over time, addressed two key weaknesses of previous research: first, adjusting for age across adolescence and early adulthood when substance use increases, and second, adjusting for factors that change over time and could confuse the link between stimulant treatment and substance use, using advanced causal inference methods.

The Part 1 analyses, which accounted for age, found no evidence that individuals currently or recently taking stimulants reported lower rates of substance use compared to those not taking stimulants. However, the large uncertainties in the estimates mean that small protective (or harmful) effects cannot be entirely ruled out. The findings revealed opposite developmental trends: as substance use increases through adolescence into early adulthood, stimulant medication use decreases. While research has separately documented these trends, they have not been adjusted for when studying the link between stimulant use and substance use. Without such adjustments, a protective link might have been wrongly concluded in this study's longitudinal analyses. No interaction was found between current and prior year stimulant treatment, indicating no evidence that patterns of continuing or stopping stimulants were related to substance use.

These findings suggest that individuals with ADHD might need different or more intensive treatment to reduce the increasing substance use known to precede SUD. Prevention efforts for substance use should start early, when reward-seeking and social influences drive initial substance use and before habits form. Preventive interventions may also need to directly address beliefs that some substances, like marijuana, have therapeutic benefits. This study's findings may differ from those based on recent US commercial health care claims data because this study examined more common substance use behaviors rather than rare emergency department visits. A Swedish registry study showing protective associations also relied on hospital visits, deaths, and criminal records. However, this study's results also extended to diagnosable SUDs, which are clinically more similar to emergency department visits. Consistent with these findings, another study also found no protection from stimulant treatment patterns for daily smoking. The large database studies provide reassurance that the most serious substance use–related outcomes (e.g., hospitalization, death, legal issues) are not increased by recent stimulant treatment.

Another long-standing theory is that early, continuous stimulant treatment should protect children with ADHD from harmful substance use. Part 2 analyses did not show that longer stimulant treatment predicted less substance use in adulthood. In fact, cumulative stimulant treatment was linked to increased heavy drinking, though this effect was small and did not remain statistically significant after correcting for multiple comparisons. The advanced models adjusted for a comprehensive set of factors that changed over time, measured from childhood to adulthood, and the results held true whether or not children were taking stimulant medication before the study began. Even if stimulants were started before the average age of 8 years, the results did not support the idea of protection or harm regarding substance use or SUDs. While some studies have suggested early treatment predicts lower SUD, they were limited by not thoroughly adjusting for the wide range of confounding variables that lead to decreasing stimulant treatment and increasing substance use over time.

A key finding from this study (and other long-term studies of children with ADHD) is that most patients no longer take stimulants in adulthood. However, if stimulant treatment provides long-term protection, that benefit should continue even after treatment stops. (Previous reports from this study did find long-term effects in another area, growth impairment, even though few participants continued medication consistently long-term.) Stimulants may not permanently change some important processes related to substance use risk, such as impulsivity and difficulty delaying gratification. Research has shown that improvements in cognitive performance, including impulse control and delay aversion, while on methylphenidate disappear once treatment stops. Still, there may be benefits to combined treatment approaches for individuals with current ADHD and SUD, including medication (like stimulants), education, and therapy.

Limitations

Limitations include the lack of medical records to verify medication history, although other studies show that parent reports often match medical records. Rates of non-marijuana illicit substance use and prescription misuse were very low, making it impossible to test specific associations within these substance classes. Finally, while the study sample included female individuals (20%), African American children (20%), and Hispanic children (8%), it was not large enough to strongly test if sex, race, or ethnicity influenced the results.

Conclusions

Overall, this 16-year, comprehensive, and developmentally sensitive analysis of stimulant medication's link with substance use did not support any hypotheses of protection or harm from ADHD stimulant treatment regarding substance use. Although these findings differ from recent conclusions of protection found in other datasets, across all studies, the consistent lack of evidence that stimulant treatment predisposes children with ADHD to later substance use provides a degree of reassurance.

Introduction

When children have attention-deficit/hyperactivity disorder (ADHD), they face an increased risk of using substances and developing substance use disorder (SUD) as adults. Stimulant medications are a primary treatment for ADHD. Given that impulsivity is a key factor in addiction, and stimulants effectively manage ADHD symptoms, including impulsivity, these medications should theoretically reduce substance use. However, some theories suggest that early exposure to stimulants might cause changes in the brain and behavior that make a person more sensitive to other drugs, thereby increasing the risk of harmful substance use.

A broad review of studies in 2013 found no clear evidence that stimulants either protect against or increase the risk of substance use or SUD. After more individual studies showed mixed results, another review in 2014 suggested a protective link for cigarette smoking. Large population studies have also indicated protective associations for SUD. In all these studies, a major challenge in determining if stimulants directly cause changes in substance use is that many other factors influence both ADHD treatment and substance use.

Recent studies have become more rigorous, but none have comprehensively and proactively evaluated both substance use and the wide range of personal, medical, and social factors that contribute to both substance use and treatment. The current study aimed to address this gap.

Age makes it harder to understand the relationship between stimulant use and substance use. As substance use rises during adolescence, well before most SUD diagnoses, teenagers with ADHD are less likely to continue taking their prescribed stimulants. Therefore, without accounting for age, any links found between stimulant medication and substance use could be misleading. One study found no differences in daily smoking among adolescents when comparing those with similar stimulant treatment profiles at the same age. A further complication is that some adolescents restart stimulant medication after months or years. A study that looked at the same individuals over time found that emergency room visits related to substance use were less common for adolescents and adults with ADHD when they were taking medication compared to when they were not. However, that data did not include frequent substance use, did not specify the type of substance, was limited to those with commercial insurance, and most substance use does not lead to hospitalization. Thus, more research is needed to look at how age affects substance use from adolescence through early adulthood (when substance use peaks), and to specifically examine different types of substances and how medication use changes over time (continuation, stopping, restarting).

Many other factors influence both stimulant treatment and substance use. These include social and personal details, the severity of ADHD symptoms, other mental health conditions, difficulties with daily functioning, and social support factors like insurance. Some previous studies accounted for these factors, but only at single points in time. Adjusting for influences that change over time is necessary to test another common treatment idea: that starting treatment at a younger age and continuing it longer improves substance use outcomes. For this study, researchers used a special analytical method called marginal structural models to address the problem of factors that change over time.

Methods

Participants were children aged 7 to 9 years old who had ADHD of the combined type, as defined by DSM-IV. They were recruited between 1994 and 1996 from six sites in the US and one in Canada. Children (95-99 per site) were randomly assigned to one of four treatment groups: medication management, multi-component behavior therapy, a combination of both, or referral to typical community care. Details about the Multimodal Treatment Study of ADHD (MTA) timeline, recruitment, diagnostic procedures, treatment, and participant demographics have been published previously. Parents provided demographic information, including race and ethnicity, at the start of the study when the participants were children. Ethics committees at the seven universities oversaw the informed consent process. The treatment group a child was assigned to did not affect their later substance use.

Participants were assessed at the beginning of the study before randomization, at 3 and 9 months, at the end of the 14-month treatment, and at 2, 3, 6, 8, 10, 12, 14, and 16 years after the start of the study. Substance use data was provided at least once in adulthood (at 12, 14, and/or 16 years after the study began) by 81.3% of participants. Their average age at the 16-year assessment was 25.1 years. Also, 95% of participants were re-evaluated at least once between the 2-year and 16-year assessments. The number of assessments participated in and substance use were slightly related; these results and retention rates for each assessment period are available elsewhere.

Measures

The Substance Use Questionnaire, adapted for this study, asked about how often participants used alcohol, marijuana, cigarettes, and various illicit and prescription drugs. Data from this questionnaire were collected from the 2-year assessment (average age 10.5 years) through the 16-year assessment (average age 25.1 years). The method for making this data consistent across different ages, along with other measure descriptions, is detailed in a related publication. A special certificate from the National Institute of Health helped ensure honest reporting.

Heavy drinking was measured by asking two questions about the frequency of binge drinking (consuming 5 or more drinks) and getting drunk. Marijuana use was assessed with one question about frequency. Past-year frequency for these variables was categorized into four levels: none, less than once per month, at least monthly but less than weekly, and once per week or more. Daily smoking was recorded as a simple yes/no, meaning one or more cigarettes per day. For other substance use, participants reported misuse of prescription medications (stimulants, sedatives, opioids) and use of heroin, inhalants, hallucinogens, cocaine, and "other substances to get high." The highest frequency of any illicit substance use or prescription drug misuse was calculated; past-year frequency was coded into three levels: none, less than once per month, and once per month or more often.

Stimulant treatment information from the 14-month assessment up to age 18 years was gathered through interviews with parents. After age 18, participants self-reported their treatment. Stimulant treatment was recorded as a yes/no variable, meaning participants took at least 10 mg per day of any stimulant for 50% or more of the days in the past year during the study's follow-up. Medication use before entering the study, reported retrospectively by parents, was also a yes/no variable. Prior publications have extensively described how medication assessments were conducted.

Analytic Plan

Both Part 1 and Part 2 of this study used the substance use and stimulant medication measures described above. Part 1 models added age and socioeconomic disadvantage in a step-by-step way. These analyses aimed to test the hypothesis that current (past-year) medication status, interacting with recent (one year prior) medication status, would be linked to substance use before and after accounting for age. Part 2 examined the link between total stimulant medication use over time and adult substance use. It used statistical methods that adjust for potential influencing factors that change over time. These analyses were repeated for alcohol, marijuana, and other substance use disorders (SUD) to see if the results applied to diagnosable conditions.

In Part 1, generalized multilevel linear models were used for outcomes that fell into ordered categories. This involved four steps. First, the effects of stimulant medication use in the current year, the previous year, and their interaction were estimated to capture the effects of concurrent stimulant use and patterns of continuation or stopping medication, without initial adjustment for other factors. Second, variables were added to mark the passage of time over assessments and capture different rates of change through adolescence and early adulthood. Third, participants’ ages relative to others at the same assessment were included. Finally, the analysis added baseline parental income and household disadvantage factors, such as family structure (single-parent vs. two-parent households) and parents' education. The models used a statistical method that could handle missing data.

In Part 2, a series of marginal structural models were used. These models adjusted for 75 factors that might influence the estimated effect of stimulant use on substance use, given their links to both stimulant treatment and substance use. These models aim to provide accurate estimates of direct cause-and-effect relationships, assuming there are no hidden factors influencing the results, that treatment is possible for everyone, and that one person's treatment does not affect another's outcome. The influencing factors included those measured at the beginning of the study and those that changed and were measured repeatedly over time. Substance use outcomes for these analyses were taken from the 16-year observations. A statistical method was used to fill in any missing data from the start of the study to the 16-year assessment. Then, time-varying scores were calculated, showing how likely individuals were to take stimulant medications at each assessment based on the influencing factors. A set of machine learning tools was used to develop a flexible model for these scores, without making it too specific to this one study's data. Finally, statistical models were used to predict the likelihood of substance use outcomes at the 16-year assessment. The estimated effects were adjusted for influencing factors by weighting the total stimulant treatment, stimulant treatment before the study, and their combination based on how likely individuals were to take medication. This made stimulant treatment statistically independent from other factors, similar to how random treatment assignment works. This allowed for an estimation of a direct cause-and-effect relationship between total stimulant treatment and substance use, based on the assumptions mentioned earlier. First, the effect of the total cumulative years of stimulant medication exposure over the study period was estimated. Second, the effect of the number of consecutive years of stimulant medication exposure was estimated. A p-value of less than 0.05 was considered statistically significant. The research questions and analysis plans were documented elsewhere, and the Part 2 strategy was registered before data analysis began. The analysis took place between April 2018 and February 2023.

Results

A total of 579 children with ADHD of the combined type were recruited. Their average age at the start of the study was 8.5 years. Among them, 115 participants (20%) were African American, 48 (8%) were Hispanic, and 351 (61%) were White. Substance use steadily increased throughout adolescence and remained stable through early adulthood. The average percentages across the 12-, 14-, and 16-year follow-up assessments were 36.5% for daily smoking, 29.6% for marijuana use at least weekly, 21.1% for heavy drinking at least weekly, and 6.2% for other substance use at least monthly. The proportion of adolescents using stimulant medication dropped sharply through adolescence, from nearly 60% at the 2- and 3-year assessments to an average of 7.2% in early adulthood.

Initial models that predicted substance use based on current and prior year stimulant medication use, and their interaction, showed that stimulant use was linked to lower odds of heavy drinking, marijuana use, daily smoking, and other substance use. However, these associations disappeared after accounting for the typical changes in substance use from adolescence into early adulthood. When relative age was added to the models, neither current nor prior year stimulant medication use was linked to any substance outcome, and there was no evidence that patterns of medication continuation or discontinuation were related to increased or decreased substance use. Generally, the estimates for stimulant use effects had large margins of error, meaning the precision of the links between stimulant and substance use was low. Additionally, the estimated effects of relative age showed that youth who were older than average within a given assessment were most likely to use substances. For example, the odds of daily smoking in a given year were 2.5 times higher for individuals a year older than the average during that assessment. This pattern remained even after considering income and household disadvantage. Given the absence of any statistically significant effects of stimulant medication at this stage, no additional social, demographic, clinical, or family factors were pursued in this part of the analysis, as these were comprehensively assessed in Part 2.

The results from the marginal structural models in Part 2 provided no evidence that the total number of years an individual took stimulants affected daily smoking, marijuana use, or other substance use by an average age of 25 years. Each additional year of stimulant medication was estimated to increase an individual's likelihood of any binge drinking or drunkenness by 4%, and monthly versus less frequent binge drinking or drunkenness by 3.5%. However, when a statistical correction was applied for multiple tests, these two effects became non-significant. The results also showed no evidence that stimulant medication use before the study began, or its interaction with total years of stimulant use, increased the likelihood of any substance use by an average age of 25. Similarly, results did not show that more years of continuous, uninterrupted stimulant medication use or its interaction with stimulant use before the study began had any effects on substance use at an average age of 25 years. These results accounted for inconsistent medication patterns.

The findings from both Part 1 and Part 2 remained consistent even when substance use was replaced with alcohol, marijuana, and other substance use disorders (SUD).

Discussion

Recent reviews concluded that stimulant medication protects children with ADHD from the known risk of substance use. However, these reviews also noted important limitations in their research methods. This current study used data collected over time to address two main weaknesses of earlier reports: first, it adjusted for age across the developmental stages of adolescence and early adulthood when substance use typically increases; and second, it adjusted for factors that change over time and influence both stimulant treatment and substance use, using methods that help determine cause and effect.

Part 1 analyses, which accounted for age, did not provide evidence that individuals currently or recently taking stimulants reported lower rates of substance use than those not taking stimulants. However, large margins of error mean researchers cannot completely rule out potential protective or harmful effects. The findings highlighted opposite trends over time: as substance use increases during adolescence and early adulthood, stimulant medication use decreases. While extensive research has separately documented these trends, they had never been adjusted for when studying the link between stimulant use and substance use. Without adjusting for these developmental trends, researchers might have wrongly concluded a protective association in longitudinal analyses. Also, no interaction was found between current and prior year stimulant treatment, suggesting that patterns of starting or stopping stimulants were not related to substance use.

These results suggest that different or more extensive treatment may be needed for individuals with ADHD to reduce the increasing substance use that often leads to SUD. Preventing substance use should begin early, when the desire for reward and social context drive initial substance use, before habits form. Prevention efforts may also need to directly address beliefs that some substances, such as marijuana, have therapeutic benefits. The findings of this study may differ from recent US commercial health care claims data because this study examined more common substance use behaviors rather than rare emergency department visits. A study from Sweden that showed protective links also relied on hospital visits, deaths, and criminal records. However, the current study's results also extended to diagnosable SUD, which is more directly related to clinical issues like emergency department visits. Similar to these findings, another study also did not find protection from any stimulant treatment patterns for daily smoking. The large database studies provide reassurance that the most serious substance-related outcomes (e.g., hospitalization, death, legal issues) do not increase due to recent stimulant treatment.

Another long-standing idea is that early, continuous stimulant treatment should protect children with ADHD from harmful substance use. Part 2 analyses did not show that longer stimulant treatment predicts less substance use in adulthood. In fact, cumulative stimulant treatment was linked to increased heavy drinking, though this effect was small and did not remain statistically significant after correcting for multiple comparisons. The marginal structural models adjusted for a comprehensive set of factors that change over time, measured longitudinally from childhood to adulthood. The results held true regardless of whether children were taking stimulant medication before the study began. Even if stimulants were started before the average age of 8 years, the results were not consistent with the ideas of protection or harm regarding substance use or SUDs. One study found that early treatment (before age 9 years) predicted lower SUD in adulthood, and another found protection from SUD in adolescents treated early and consistently at a high dose. Both studies were limited because they did not comprehensively adjust for the wide range of factors that influence both decreasing stimulant treatment and increasing substance use over time.

A key finding from this study and other long-term studies of children with ADHD is that most patients no longer take stimulants in adulthood. Yet, if stimulant treatment offers long-term protection, that benefit should continue even after treatment stops. Stimulants may not permanently change some important processes related to substance use risk, such as impulsivity and difficulty delaying gratification. One recent study reported that improvements in thinking skills, including impulse control and delay aversion, while taking methylphenidate, disappeared once treatment stopped. There may still be benefits to combined drug therapy (including stimulants), education, and counseling for individuals who currently have both ADHD and SUD.

Limitations

Limitations of this study include not having medical records to confirm medication history, though confidence in the data is supported by studies showing agreement between parent reports and medical records. Rates of non-marijuana illicit substance use and prescription misuse were very low, making it impossible to test specific links within different classes of other substance use. Finally, while the study's multi-site sample included female participants (20%) and children identified as African American (20%) and Hispanic (8%), there was not enough data for a strong study on how sex, race, or ethnicity might change the results.

Conclusions

Overall, the 16-year, comprehensive, and developmentally sensitive analyses of stimulant medication links with substance use did not support any ideas of protection or harm from stimulant treatment for ADHD. Although these results differ from recent conclusions of protection found in other data sets, across all studies, the findings consistently show no evidence that stimulant treatment makes children with ADHD more likely to use substances later.

Introduction

Children with attention-deficit/hyperactivity disorder, known as ADHD, have a higher chance of using drugs or alcohol and developing substance use problems later in life. Medicines called stimulants are often the first choice for treating ADHD. It was thought these medicines might lower the chance of substance use because they help with problems like acting without thinking. However, some worried that taking stimulants early in life might make people more open to trying other drugs, leading to more substance use.

Studies in the past about stimulants and substance use have shown mixed results. Some found no clear link, while others suggested stimulants might help protect against smoking or other substance problems. A big problem with many of these studies is that other things can affect both ADHD treatment and substance use. It is hard to know if the medicine itself causes the change or if other factors are involved.

Age is a significant factor. As teenagers get older, they often use more substances. At the same time, teenagers with ADHD are less likely to keep taking their prescribed stimulants. This means that if studies do not consider age, it might look like stimulants are linked to substance use when they are not. Many other things can also affect both stimulant treatment and substance use, like family income, how serious ADHD symptoms are, other mental health problems, and how parents act. Earlier studies did not always look at how these things changed over time.

This study was designed to look at the link between stimulant medicine and substance use from childhood into early adulthood. It used careful methods to see if stimulant use at a certain time was linked to substance use, and if taking stimulants for a long time made a difference. It also looked at how age and many other changing factors might affect these links.

Methods

Children aged 7 to 9 years with a type of ADHD were chosen for this study between 1994 and 1996. They came from seven places in the US and Canada. The children were put into different groups to get different ADHD treatments or regular community care. Many details about how the study was set up have been shared in earlier reports. The type of treatment group a child was in did not affect their later substance use.

The children were checked many times over the years, from when they started the study until they were young adults. Most of the young adults provided information about their substance use, like alcohol, marijuana, cigarettes, and other drugs. They also shared how often they used them. A special rule was in place to help people feel safe telling the truth about their substance use.

Information about stimulant medicine use was also collected. This included how much medicine was taken and for how long. The study looked at stimulant use as taking a certain amount of medicine for at least half the days in a year. The study also checked if a child had taken stimulant medicine before joining the study.

The study looked at the information in two main ways. The first way checked if taking stimulants at a certain time was linked to substance use, after looking at how age changed things. The second way checked if taking stimulants for many years, from childhood into adulthood, was linked to substance use in early adulthood. This part of the study used special methods to deal with all the other factors that might confuse the results. It aimed to show if stimulant use had a direct cause-and-effect link with substance use.

Results

The study included 579 children with ADHD. About 20% were African American, 8% Hispanic, and 61% White. The study found that substance use steadily went up as people got older, from teenage years into early adulthood. For example, about 36.5% were daily smokers in early adulthood, and about 29.6% used marijuana at least once a week. At the same time, the number of young people taking stimulant medicine went down a lot, from almost 60% when they were younger to only about 7% in early adulthood.

The first part of the study looked at current stimulant use. At first, it seemed like taking stimulants was linked to less heavy drinking, marijuana use, daily smoking, and other substance use. However, when the study considered how substance use naturally goes up with age, this link disappeared. This means that simply being on stimulant medicine at a certain time did not show a clear connection to less substance use when age was taken into account. Being older than others at the same assessment time was linked to a higher chance of using substances.

The second part of the study looked at taking stimulants for many years over time. These results showed no clear proof that taking stimulants for more years stopped people from smoking daily, using marijuana, or using other substances by age 25. It did seem that each extra year of stimulant medicine use slightly increased the chance of heavy drinking, but this finding was very small and not strong enough to be certain after careful checks. The results also showed that taking stimulants before the study or taking them without breaks did not clearly protect against or cause more substance use.

Discussion

Recent reviews of studies had suggested that stimulant medicine might protect children with ADHD from substance use problems. However, those studies had some problems. This study aimed to fix two main problems: looking at age changes across growing up and handling many changing factors that affect both stimulant treatment and substance use.

The first part of the study, which looked at age, did not find clear proof that people taking stimulants at the time reported less substance use than those not taking them. The study showed that as people got older, substance use went up, but stimulant use went down. If the study had not looked at these trends carefully, it might have wrongly decided that stimulants offered protection. This suggests that more or different kinds of help might be needed for people with ADHD to lower their substance use risk.

The second part of the study looked at taking stimulants for many years. It found no clear proof that taking stimulants for a longer time led to less substance use in adulthood. While one small link showed a slight increase in heavy drinking with more years of stimulant use, it was not strong. This study used very careful methods to account for many different factors that could confuse the results. Even when stimulants were started at a young age, the study did not find that they either protected against or caused substance use problems later on.

Most people stop taking stimulants by the time they are adults. If stimulants truly offered long-term protection, that benefit should last even after stopping the medicine. This study did not find evidence for such lasting protection for substance use. It might be that stimulants do not permanently change things like acting without thinking, which can be linked to substance use.

This study had some limitations. It did not have medical records to confirm medicine use, though parent reports are usually accurate. There were also too few people who used illegal drugs or misused prescription drugs (other than marijuana) to study those groups well. Finally, even though the study included girls and different racial and ethnic groups, it did not have enough people to see if the effects of stimulants were different for these groups.

Conclusions

Overall, after looking at 16 years of information very carefully, this study found no clear proof that stimulant medicine for ADHD either protects against or causes harm related to substance use. While these findings are different from some other recent studies, it is comforting that all studies consistently show no evidence that stimulant treatment makes children with ADHD more likely to use substances later on.