Pathways and mechanisms for the development of substance use disorders (SUD) in adolescents are complex and even after years of investigation remain poorly understood. Research efforts have generally been advancing in two main directions: (a) investigating purported biological and familial underpinnings of SUDs, and (b) examining environmental factors that influence the onset of SUDs. However, the interface of both constitutional [e.g., family history of substance use (FHSU)] and environmental (e.g., parenting, peer relations, neighborhood safety) aspects of adolescent substance experimentation and use remains unclear. For instance, prior research shows that FHSU positive youth (FHSU-P) are at increased risk for developing problem drug use (1, 2), and it is speculated that this elevated risk might be related to alterations in reward processing and behavioral inhibition (3, 4), or other processes generally associated with the functions of the prefrontal cortex (5). More generally, increased risk-taking, including substance use, occurs during adolescence, regardless of FHSU status, and might be linked to the imbalance in the developmental course of a rapidly developing motivation-reward system, which contributes to the pursuit of rewarding, exciting, and novel experiences and a more gradually developing cognitive control system, exercising restraint on potentially harmful impulses (6–8). Functional imbalance in the development of these systems may present clinically as high level of impulsivity, which can be measured with the appropriate psychometric assessments (9).

While FHSU is known to be a risk factor for SUD development, it is important to note that not all FHSU-P youth go on to use substances problematically or develop SUDs. Among FHSU-P youth, there are many mediating factors that may either heighten or diminish risk for SUD (10). Such protective effects can be attributed to both biological features like increased recruitment of the dorsolateral prefrontal cortex during inhibitory task (11) or increased resting state connectivity between brain regions implicated in cognitive control (12) but may also be linked to environmental influences (10). A review by Petraitis et al. (13) examined theories of adolescent substance use that focused on substance-specific conditions, social learning processes and attachment to family, and classified environmental influences into social, attitudinal and interpersonal factors. Subsequent literature has pointed out the complex nature of the interactions between personal and environmental factors in relation to both vulnerability and resilience. One review discussed the definition of resilience as “positive adaption inspite of adversity” and identified “protective” factors from three levels (e.g., family, school and community) (14) These included parental supervision, family bonding, positive peer connections and school and community engagement among others. Others have independently found that factors like “prosocial peers”, “home and school support”, and “meaningful community participation” appear to confer protective value in adolescents in relation to tobacco, alcohol and illicit drug use, however, they also noted that the beneficial effects are not universal but might be limited to a small set of factors specific to different populations (15). Indeed, exploring the family history of SUD requires examining the multidimensional and interacting pathways of personal and environmental influences, which may simultaneously contribute to risk and promote resilience, highlighting the importance of an integrative framework over isolated analyses.

Indeed, several models have been proposed to capture the interplay among various factors that contribute to SUD risk. The social development model theory presents a framework which stipulates that children learn patterns of behavior from the socializing agents of family, school, peers and various community institutions, which can affect future behaviors (16, 17), and can be used to reliably predict substance use in late adolescence (18). The Bronfenbrenner's Bioecological Model (19) examines the interactions between the person's unique characteristics, their immediate and broader societal environment and how these elements change over time. These multi-system interactions affect the development of adolescent health behavior, including substance use. In turn, the differential susceptibility hypothesis suggests that some youth are characterized by increased sensitivity to environmental contexts possibly linked to specific genetic makeup (20). Moreover, neurobiological changes that occur during adolescence as part of normative maturation of the central nervous system, may also contribute to the etiology of adolescent substance use (21–23). These models and theories suggest that a large group of young individuals who are at risk of developing SUDs (e.g., FHSU-P youth) will most likely be comprised of subgroups of individuals who share sets of environmental exposures that may influence different trajectories of engagement with drug use. This further suggests that FHSU alone may have a limited predictive value for SUD development and that more detailed stratification based on environmental and psychosocial risk factors is warranted to identify different levels of SUD risk (24–26). Taken together, examining subgroups based on environmental and psychosocial risk exposures within FHSU-P may identify individual- as well as group-level targets of intervention to decelerate the rate of substance use experimentation and use.

The Adolescent Brain and Cognition Development (ABCD) Study offers the largest cohort of young children with information about both constitutional predisposition for SUD (e.g., FHSU) as well as many indexes of environmental exposures. For example, a recent report from the ABCD cohort identified clusters of environmental exposure that showed consistent and replicable associations with multiple measures of brain organization (27). Similarly, another group found evidence that greater environmental risk and stress were associated with higher endorsement of distressing psychosis like experiences particularly among minority children from the ABCD study (28). A similar study, albeit in a small cohort, used clustering technique to identify groups based on parental substance use and examined its influence on stress and substance use vulnerability in children (29). Thus, availability of these data allows for the examination of a considerable number of interactions between FHSU and environmental exposures, and their association with key cognitive and clinical outcomes such as impulsivity, reward processing and psychopathological symptoms.

To test these interactions, in this proof-of-concept study, we used unsupervised machine learning to categorize FHSU positive (FHUS-P) youth into subgroups based on diverse sociodemographic and environmental exposures, which have previously been identified as key risk factors (30–32). Importantly, we operationalized FHSU to specifically capture only those relatives whose substance use was associated with significant functional or psychosocial impairment. Unlike broader definitions based solely on substance exposure, or formal diagnostic categories such as the family history of SUD that rely on clinical confirmation, our definition balances specificity and feasibility within a large-scale and non-clinical ABCD sample. Based on available evidence we hypothesized that a cohort of FHSU-P youth ages 9–10 from the ABCD study can be stratified into subgroups based on known sociodemographic and environmental factors using a machine learning clustering method. To compare the relative extent of risk between these subgroups, we planned to conduct group comparisons between FHSU Negative (FHSU-N) and each of the FHSU-P subgroups, as well as between FHSU-P subgroups on clinical factors such as parental behavioral reports, trait impulsivity, and reward-related behaviors. We hypothesized that the scores of these behavioral indexes will be significantly different among the subgroups and will help in distinguishing subgroups with greater and lesser risk for initiating problem substance use.

In this study, we identified distinct subgroups of adolescents with familial risk for substance use disorder (SUD) based on sociodemographic and psychosocial factors. These subgroups fell along a spectrum of environmental advantage (i.e., higher family income, supportive parents, positive school and peer engagement) to disadvantage (e.g., lower income, unmarried parents, poor school and peer engagement) and differed markedly in both psychopathology symptoms and impulsivity levels. In particular, the more advantaged subgroups (Subgroups 1 and 2) showed lower psychopathology and impulsivity, whereas the more disadvantaged subgroups (Subgroups 3, 4, and 5) exhibited heightened psychopathology symptoms and elevated impulsivity patterns across multiple domains. Surprisingly, reward sensitivity did not differ between the subgroups, although this may reflect methodological constraints of the task used to derive prediction error estimates. Interestingly, within the subgroups with more favorable factors, Subgroup 2 was predominantly defined by the presence of a non-working, possibly stay-home parent, yet high family income, whereas Subgroup 1 was defined by working parents and a combination of additional favorable factors. Indeed, the impact of the role of stay-home vs. working parent on behavioral development is an active area of research, with some studies showing that children with stay-home mothers may exhibit less aggressive behaviors (68–70), while others showing no difference in behaviors for children with either working or stay-home parents (71). These findings highlight the importance of psychosocial context, individual differences in impulsivity, and potential methodological limitations in measuring reward processing within at-risk youth populations.

We also found Subgroups 1 and 2 to have the lowest levels of psychopathology symptoms as well as impulsivity in several domains within the FHSU-P subgroups, suggesting potential protective factors of intact parenting and parental relationships that may mitigate the impact of familial SUD risk (72). Conversely, Subgroups 3, 4, and 5 exhibited more pronounced psychopathology symptoms, suggesting possibly heightened vulnerability in those adolescents with disadvantageous psychosocial exposures and emphasizing the need for personalized approaches to assessment and intervention that account for individual differences in symptom presentation and risk profiles. However, note that unlike Subgroups 3 and 4, Subgroup 5 had greater proportion of females compared to males. These findings, especially of the aspects of impulsivity, are consistent with reports that implicate impulsivity in the etiology and maintenance of SUD (73, 74). Interestingly, however, such heightened vulnerability did not appear to be associated with the density of SUD problems in the family, since only Subgroup 3 showed the highest FHD scores compared to both more protective Subgroups 1 and 2, and more vulnerable Subgroups 4 and 5. Moreover, FHD scores were not significantly different between Subgroups 1 (most protective) and 5 (most vulnerable). From treatment planning standpoint, these findings underscore the importance of considering individual differences in impulsivity when assessing individuals with FHSU-P and designing targeted interventions aimed at reducing impulsive behaviors.

In contrast to our hypothesis, FHSU-P subgroups did not differ in terms of reward sensitivity, specifically quantified as computationally derived behavioral measures of positive and negative reward prediction error. Although it is plausible that these subgroups may indeed be comparable on positive and negative reward prediction error, we must also note that the MID task used in ABCD Study is not an optimal task to derive prediction error measures and for examining learning (75, 76). Perhaps the lack of differences in RPE between FHSU-P subgroups reflects this methodological limitation, and therefore, needs further validation. Nevertheless, recent reports have assessed the reliability of RPE, derived from the MID task, in encoding in young populations (48, 77), supporting the notion that RPE variable computed from the MID task may have utility in identifying specific patterns of reward-related brain activation, which in turn may also correlated with impulsivity measures in young populations. In case of the current study, comparing the groups on the RPE variable is an initial assessment for any possible reward sensitivity differences among the subgroups. Based on the established practices of utilizing computational variables in brain neuroimaging, RPE will need to be incorporated in an independent set of whole brain analyses of the sample to investigate for differences in brain activation and connectivity among the subgroups, which in turn may be linked to future drug initiation. An alternative consideration is that although impaired reward processing is associated with drug use initiation, it may not be related to FHSU-P or the underlying biology (78, 79) and in some cases overshadowed by complex interplay of other behavioral traits (80).

A major novelty of this study is that it attempts to subcategorize a large cohort of young FHSU-P youth in relation to environmental exposures in order to investigate two main objectives: first, that a set of relevant exposures will be able to differentiate independent subgroups with either favorable (possibly protective) and unfavorable environmental exposures, and second, that these groups may show significant differences in some behavioral (e.g., impulsivity) and clinical measures. As such we view the reported results as preliminary but strongly suggest that investigations in similar directions may benefit from using clustering techniques (e.g., k-means, as used in this study) or latent class analyses [latent profile analysis (LPA) or latent class growth analysis (LCGA)], to stratify individuals with FHSU into unique subgroups for more personalized assessments. Indeed, the reported significant differences in impulsivity scores appears to be the most interesting findings especially as a starting point of more detailed investigation for possible brain based differences among the subgroups as it is well established that there are correlations between high impulsivity and indexes of brain morphology and physiology (81, 82). In addition, as this relationship will need to be replicated, it offers the possibility to identify individuals with particular exposure profiles that may benefit from interventions that specifically target impulsivity subtypes (e.g., sensation seeking vs. negative urgency).

Indeed, these types of investigation may have a far-reaching implication for both early identification for high risk individuals as well as provide guidance for prevention and treatment of substance use in adolescence. Conceptually, findings of robust moderating effects from environmental exposures can inform multisectoral prevention and early intervention strategies that improve outcomes for at-risk youth. For example, information on the moderating effect of environmental factors may allow for promoting initiative that, on one hand, may support environmental factors with positive contribution to preventing substance use initiation in youth and, on the other, to modify environmental factors that may facilitate or predispose youth to early drug experimentation. Such initiatives may need to be based in family, school or neighborhood settings and include home based family therapy and group activities promoting problem solving skills (e.g., resisting peers' pressure for drug use, socializing without drug use etc.). Additionally, characterizing subgroups of youth who may have high levels of impulsive behaviors may allow clinicians to select most appropriate therapeutic strategies for addressing specific behaviors or symptoms. As impulsive behaviors may vary in respect to their inherent features (e.g., sensation seeking vs. positive or negative urgency) therapeutic goals may differ in reducing urges to seek novel experiences vs. managing urges to respond impulsively to different types of outcomes. Further, identifying impulsive behaviors as part of underlying disorders like ADHD may require considerations for pharmacological interventions in addition to behavioral therapies. In sum, more detailed information about stratifying large groups of youth who may have familial loading for SUD into “at risk” subgroups characterized by unique profiles of interacting environmental exposures and psychological features may allow for more effective “precision” approaches of preventing and treating adolescent substance use and related disorders.

Taken together, the next steps for this project are to conduct analyses of the longitudinal data that are available from ABCD and potentially determine which of the subgroups may exhibit high levels of substance initiation, at what ages these behaviors may occur and what types of biological characteristics (i.e., brain activation and connectivity) may be predictive of these behaviors. These types of analyses will be potentially able to produce results that can be used as evidence to support any recommendations for preventive, psychotherapeutic and psychopharmacological treatments for substance using adolescents as we have outlined in the introduction section. Additionally, such results may be applicable to recommendations for policy initiatives like age limits for legal substance use, parental consent for substance use treatment for minors, and legal consequences drug related offences for minors.

We want to also acknowledge relevant limitations that are important to consider when interpreting the findings. Firstly, the selection of demographic variables was guided by existing literature, but the findings here reinforce previous reports that advantageous and disadvantageous psychosocial exposures tend to cluster together along a positive-negative axis (27, 83, 84). Next, as discussed earlier, while the MID task is commonly used in research, its ability to fully capture prediction error or elucidate reinforcement learning mechanisms may be limited. Third, while we have done analysis of FHSU density, it was not included as factor for any analysis, that may have varying degree of resilience factors. As the next step, enhancing the study by incorporating additional measures, such as socio-demographic variables, social determinants of health, FHSU density or complementary tasks assessing reinforcement learning, would provide a more comprehensive understanding of the phenomena under investigation. Moreover, conducting similar clustering analyses in the FHSU-N group could help determine whether the identified subgroups reflect patterns specific to familial risk or instead capture broader socio-demographic or behavioral structures present across the general adolescent population.

The findings of this study highlight the importance of psychosocial context in shaping the clinical presentation of adolescents with positive family history for substance use. Understanding these relationships is critical for informing targeted interventions aimed at mitigating the impact of familial risk factors on the development and maintenance of substance use disorders. Future research should further elucidate the mechanisms underlying these associations and explore novel intervention strategies tailored to individual risk profiles within the FHSU-P population.