A Social Cascades Model of Addiction: Examining Effects of Psychological Interventions on Social and Substance Use Outcomes via Meta-Analytic Review

Social Environments and Substance Use: A Chicken-and-Egg Conundrum

Substance problems rarely occur in isolation but are accompanied by a wide range of impairments, particularly in the social domain. Research in clinical samples suggests that individuals with SUDs tend to have small social networks but a higher percentage of drug-using associates. These individuals also receive less positive social support and are highly susceptible to the experience of loneliness (Ingram et al., 2020; Thomas et al., 2022; Tracy & Biegel, 2006). Notably, SUDs are among the most stigmatized health conditions, a factor that often drives individuals to conceal their substance problems and delay treatment, further exacerbating social alienation (Livingston et al., 2012). Together, the challenging social realities associated with SUDs serve as a critical link in perpetuating the vicious cycle that maintains substance problems, triggers relapses, and undermines individuals’ efforts to reduce use (Sun, 2000; Tracy & Biegel, 2006).

Researchers have attributed the association between substance problems and social environments to two primary pathways. In the broader addiction literature, it is often assumed that social stressors are the driving factors of substance problems (hereafter referred to as the “social environment → substance use” pathway). Longitudinal research has identified a wide range of environmental risk factors as precursors of subsequent substance problems. In the domain of peer relationships, peer rejection and peer substance use robustly predict later adolescent and young adult cigarette use, alcohol use, binge drinking, and marijuana use (Andrews et al., 2002; Hussong, 2002; Lynne-Landsman et al., 2010). Within the family context, family dysfunction characterized by low support and high conflict increases the subsequent risk of substance use during adolescence and young adulthood (Sánchez-Queija et al., 2016; Skeer et al., 2009; Van Ryzin et al., 2012). In the domain of committed relationships, one partner’s substance use predicts another partner’s use over time (Cornelius et al., 2016; Windle & Windle, 2014, 2018), while aggression and partner incarceration also contribute to elevated risk of the later development of SUDs (Ahmadabadi et al., 2019; Marziali et al., 2024). More broadly, decreased social contact and social support have also been found to predict later substance use (Peirce et al., 2000).

Conversely, the association between social problems and substance use could also reflect the harmful effects of substance use on social environments (hereafter referred to as the “substance use → social environment” pathway). In longitudinal research, substance use often precedes a variety of negative social consequences. For instance, teenage drug use prospectively predicts impaired romantic attachments, increased conflicts with parents, association with deviant peers, and greater loneliness over time (Bonar et al., 2022; Newcomb & Bentler, 1988; Poulin et al., 2011). Among couples, substance use, especially discordant use, has a prospective association with poorer relationship quality across age groups (Birditt et al., 2016; Foulstone et al., 2016; Testa et al., 2018). Substance use is also found to negatively impact social functioning over time, including reduced contact with family and friends and lower educational attainment (Peirce et al., 2000; Tucker et al., 2006).

In clinical contexts, it is common for individuals to enter treatment presenting with substance use problems bundled together with a cluster of social challenges. In more severe forms of addiction, individuals’ social resources become so depleted that substance use may turn into the last thread of hope for a moment of relief, despite knowing that using will only worsen the situation – a vicious cycle that can appear almost unbreakable. While both substance use and pathogenic social environment could contribute to this cycle, it is not always viable for patients to engage in extended treatments targeting each purported change process. This, plus high dropout rates among patients with SUDs, often leaves care providers a narrow window to deliver interventions most likely to have a meaningful impact to address these clustered issues before patients disengage. Yet, this presents a classic chicken-and-egg conundrum in terms of intervention sequencing. So, where should we begin? Axiomatically, the effectiveness of psychosocial treatments is driven by the degree of congruence between the causal relations relevant to the presenting symptoms and the causal mechanisms targeted by the treatment (Borkovec & Costonguay, 1998; Haynes et al., 1999; Persons, 2005; Rosen & Davison, 2003). Developing effective interventions for addiction, therefore, requires a precise understanding of the process driven by the interplay between social environments and substance use, as well as the relative strength of these pathways. If the clustering of substance and social problems were driven more by the “social environment → substance use” pathway than its reverse, treatments might benefit from integrating frameworks that address mediating processes linking stressors emerging from one’s social environment to substance use from the outset. In contrast, if the “substance use → social environment” pathway were to account for more of this clustering, treatments may prioritize targeting substance use behaviors, with improvements in the social domain expected to follow as substance use subsides.

Despite longstanding speculation in the addiction field regarding the coexistence of “substance use → social environment” and “social environment → substance use” pathways, empirical evidence has yet to allow researchers to confidently deduce causality or quantify the relative strength for either pathway. These two bodies of scientific literature have largely developed independently, with limited attempts to reconcile findings or bring theoretical constructs into conversation. Moving forward, the field of addiction science might benefit from embracing models that account for multiple, interacting pathways between social environment and substance problems.

A Cascades Model of Substance Use and Social Environments

Offering a more comprehensive framework for approaching the chicken-and-egg conundrum, cascades models represent among the most widely applied theories for describing processes by which functioning in one domain, level, or system influences another over time to shape the course of psychopathology (Gottman, 1993; Masten & Cicchetti, 2010; Masten & Coatsworth, 1998). Developmental cascades, in particular, refer to “the cumulative consequences for development of the many interactions and transactions occurring in developing systems that result in spreading effects across levels, among domains at the same level, and across different systems or generations” (Masten & Cicchetti, 2010). Here, we propose that the emergence of addictive behaviors, together with pathogenic social environments, represents a culmination of numerous cascading sequences that arise from, interact with, and are subsequently shaped by both the social fabric of one’s environment and patterns of substance use over time (see Figure 1). This process is analogous to how an avalanche is formed: a spontaneous event, whether it be rain or rockfall (a first taste of alcohol, a house party involving illicit drugs), sets off a small amount of snow moving along the slope initially (recreational use); as the snow accumulates (repeated use, associations with deviant peers), the growing momentum and force have the capacity to alter the terrain and carry the boulders and trees along the way (shapes social environment to be more conducive to substance use), propagating rapidly and ultimately resulting in varying degrees of impact (development of SUDs).

The usage of different substances is naturally associated with distinctive social features. Alcohol, (e-)cigarettes, and cannabis, for example, have been widely integrated into social life (Fairbairn & Kang, 2025; Fairbairn & Sayette, 2014; Lipperman-Kreda et al., 2018; Marron, 2017), where these substances are readily accessible, and their use in social contexts is largely normalized. In contrast, illicit drugs such as opioids and stimulants are historically associated with social marginalization and stigma (Room, 2005). The social networks of individuals who heavily use or are dependent on illicit drugs are often perceived as more “deviant,” with shared social activities oriented toward obtaining, using, and sometimes distributing illicit drugs (Moore, 1993).

Addiction scientists have proposed theories to help explain how social environments affect substance use and how substance problems reshape social environments. Positive social relations have been widely documented to reduce stress, provide support, model healthy behaviors, and enhance social functioning (Brent et al., 2014; DeVries et al., 2003; Smith & Christakis, 2008; Wrzus et al., 2013). Given the diverse benefits of positive social relations, some theories view the presence or absence of such social resources as protective or risk factors for substance problems (see also Moos, 2007). Social Control Theory posits that individuals’ behaviors are bound by strong ties to family, friends, community, and broader society. When these attachments become weak or disrupted, delinquent behaviors, such as alcohol and drug abuse, are more likely to occur, as individuals are less motivated to engage in responsible actions and adhere to existing social standards (Hirschi, 2017). Social Learning Theory emphasizes the role of modeling and imitation, suggesting that substance use can be learned through observing substance-specific behaviors of family members or peers and internalizing social norms regarding substance use. Other theories recognize the reinforcing effects of substance use within the broader social environment via affective means. Stress and Coping Theory suggests that substance use often functions as a maladaptive coping strategy in response to stress exposure, including stress arising from strained relationships. According to Behavioral Economics Theory, the availability of a rich reservoir of natural rewards in one’s social environment can not only prevent individuals from becoming addicted to a single, artificial reward source for pleasure (Grigson, 2008) but also serve as a healthier reward substitute during recovery (Fisher et al., 2016). While social isolation and distressing relationships have affected many with SUDs and potentially contributed to maintaining dysfunctional substance use behaviors, the assumption that individuals with substance problems are deprived of close social bonds – and that the cure is to increase social connections – is oversimplified. Social Reward Theories suggest that substance use may function as a leisure activity producing social-cohesive effects in group contexts; the combination of social rewards and drug reinforcement offers powerful incentives for continued use in social contexts, particularly affecting individuals embedded in substance-use-facilitating social networks (de Wit & Sayette, 2018; Gurrieri et al., 2021). Altogether, substance-supporting social networks and various social stressors are theorized to constitute a “pathogenic social environment” that is conducive to substance use and problems.

Conversely, substance problems are also theorized to shape the social environment. One broad mechanism involves the pharmacological effects of psychoactive drugs in the context of both acute intoxication and chronic use, which can alter social behavior and damage social relations. For example, Intoxication-Aggression Theories emphasize the role of acute intoxication in facilitating interpersonal conflicts and verbal/physical aggression via various mechanisms, such as perceptual distortions, disinhibition, or expectancy effects (Foran & O’Leary, 2008; Moore et al., 2008). The causal link between intoxication and aggression has been most consistently demonstrated for alcohol in placebo-controlled experiments, though certain stimulant drugs such as cocaine and amphetamine have also been implicated (Kuypers et al., 2020). Relatedly, Theories of Drug-Induced Impulsivity suggest that substance use may increase impulsivity by impairing inhibitory control, causing lapses in attention, and reducing sensitivity to negative consequences (De Wit, 2009; Perry & Carroll, 2008). Because strong social relations are built on reliability, trust, and mutual support, impulsiveness manifested in the social realm, such as saying things one later regrets, displaying emotional outbursts, and violating interpersonal boundaries, can naturally undermine relationship quality and contribute to social rejection and isolation. Neurocognitive Theories of Addiction highlight the role of neuroadaptations caused by chronic drug exposure in the disruption of social behaviors (Volkow et al., 2011). Psychoactive drugs are known to hijack the brain’s reward and cognitive systems, leading to impaired social cognition and impulse control as well as the devaluation of natural rewards. These changes help explain why addicted individuals may evince diminished motivation in pursuing natural rewards, such as hanging out with family and friends, and continue using substances even in the face of negative consequences for their social relationships, school, employment, and other areas of social functioning (Volkow et al., 2011). Other theories adopt a social-ecological perspective in understanding how individuals themselves can play an active role in modifying characteristics of their immediate social environments. Social Selection Theory posits that individuals are more likely to choose friends, romantic partners, or become involved with social groups that demonstrate more similar characteristics, including patterns of substance use, than with those who exhibit dissimilar features (McPherson et al., 2001). In other words, people may self-select to be involved in social environments that normalize or sustain their use. The Stress-Strain-Coping-Support Model proposes that living with or having a close relationship with someone who struggles with substance problems constitutes a form of stressful life circumstance, which may cause a significant strain on affected family members (Orford et al., 2010). Without adequate support, affected family members may distance themselves or develop maladaptive behaviors (e.g., substance use, controlling behaviors) that further erode their relationship with the user. Therefore, as addiction progresses, individuals’ interpersonal relationships also tend to suffer, further contributing to a social environment characterized by difficulties in close relationships (Adams, 2008).

Substance Use Treatments and Theorized Mechanisms of Change

The development of substance use interventions is often grounded in theoretical understandings of what causes and maintains addiction, and each treatment package typically comprises a combination of behavioral change techniques targeting mechanisms of action across intrapsychic, interpersonal, and environmental levels. Nonetheless, it is generally understood that change ultimately occurs through the patient, who is essentially the driver of their own life, bearing responsibility for modifying substance use patterns and social networks, repairing relationships with loved ones, and making day-to-day decisions that support recovery. By design, substance use interventions rely on patients’ agency to facilitate behavioral change, implementing techniques that enhance patients’ motivation and modify thoughts and beliefs about substance use, while also equipping them with behavioral and social skills to manage cravings and cope with various high-risk social situations that may trigger substance use or relapses.

In this regard, cognitive and behavioral approaches have been widely used in treating substance problems. Focusing on the interplay between individuals’ thoughts, emotions, and substance use behaviors, Cognitive-Behavioral Therapy (CBT) aims to facilitate behavioral change by modifying maladaptive cognitions surrounding substance use. Core components of CBT include conducting a functional analysis of drug use, identifying triggers, pro-drinking thoughts, and high-risk situations, as well as teaching skills to challenge maladaptive thoughts to promote behavioral change. Motivational Interviewing (MI) integrates person-centered principles with cognitive-behavioral strategies to help patients resolve ambivalence surrounding cutting down or quitting substance use and increase intrinsic motivations to make behavioral change. Key strategies of MI include eliciting patients’ change talk, minimizing sustain talk, as well as asking open questions and affirming patients to promote self-efficacy (Miller & Rollnick, 2013). Grounded in principles of operant conditioning, Contingency Management (CM) represents a classic behavioral treatment that aims to reinforce a target behavior (e.g., negative drug screen) through tangible reinforcers (e.g., providing cash, vouchers) and withhold the incentives when these behaviors do not occur (for review, see Petry, 2000). Compared to no treatment, each of these interventions has demonstrated moderate-to-large effects in improving substance-related outcomes, yet no one type of treatment is superior to another (Magill et al., 2019; Prendergast et al., 2006; Smedslund et al., 2011), suggesting that they all successfully manipulate potentially shared individual-level mechanisms underlying substance use. While, in theory, the aforementioned treatments primarily target individual-level mechanisms of change, it is important to note that they are not “asocial.” Recognizing the role of social factors in substance use, many CBT- and MI-based treatment programs incorporate social skills training modules – such as refusal skills, assertiveness training, and communication skills – to help patients manage and, when possible, influence their social environments.

In addition to empowering individuals to initiate and maintain behavioral change, substance use interventions may incorporate strategies to more directly engage patients’ immediate social networks throughout the treatment process. Some treatments focus on helping individuals repair relationships with significant others and strengthen existing social ties during treatment. For example, Behavioral Couples Therapy (BCT) represents a prototypical behavioral intervention for spousal substance abuse. BCT incorporates strategies to enhance social support for the patient and improve relationship functioning between patients and their partners, such as increasing positive couple behaviors, improving couples’ communication skills, and when relevant, addressing the partner’s own substance use (McCrady & Falanagan, 2021; O’Farrell & Fals-Stewart, 2000). For adolescent substance abuse, treatments often adopt a family system perspective that emphasizes reducing family conflict and enhancing family cohesion. Family therapy helps parents improve parenting skills, increase parental monitoring, and improve communication among family members (Austin et al., 2005; McCrady & Falanagan, 2021). Some interventions help individuals to reduce social isolation and create more natural rewards in their social environment. For example, the Community Reinforcement Approach (CRA) focuses on incentivizing the engagement of social and recreational activities that reinforce a substance-free lifestyle (Abbott et al., 1998; Meyers et al., 2011). Additionally, some interventions aim to facilitate changes in the composition of SUD patients’ social networks by increasing abstinence-supporting social ties. While there have been manualized treatment efforts to intentionally alter social network composition, such change may also take place naturally in the context of mutual-help groups, such as Alcoholics Anonymous (AA). Core practices in AA, such as storytelling in group meetings and the cultivation of sponsor-sponsee relationships, are essential for the fostering of pro-abstinence social networks (White & Kurtz, 2008). Over time, other forms of 12-step facilitation groups (e.g., Narcotics Anonymous) – largely modeled after AA – have also emerged to support recovery for drug addictions.

Over the past several decades, substance use treatments have evolved to encompass a diverse and flexible set of behavioral change techniques that are often combined in various ways across treatment programs, sometimes extending beyond the therapeutic orientation labels under which they are described. Therefore, distinctions among intervention types based on mechanisms of action are often blurred. Nonetheless, assumptions regarding the existence and strength of both “substance use → social environment” and “social environment → substance use” pathways are pervasive in the design of substance use interventions. These treatments distinguish themselves from one another based on several grounds, such as the extent to which they emphasize individual agency, what skills are considered necessary for change to happen, and the degree of scaffolding provided to not only patients but also to their surrounding social network members to facilitate recovery. Specifically, while many SUD treatments integrate consideration of social processes, they diverge in the extent to which: 1) social network members are directly involved in treatment; and 2) social factors represent a primary treatment focus vs one domain among many. These differences are often reflected in interventions’ theoretical framework, treatment format, as well as core intervention components involved, providing a reasonable, albeit heuristic, approximation for manipulating substance use levels and various features of the social environment in the context of RCTs. Based on these distinctions, in the context of this review, we categorize treatments as with or without a “significant social component” (see Methods for more detailed definitions) and hereafter refer to them as “socially oriented” and “individually oriented” treatments, respectively, for linguistic simplicity.

The Current Study

Substance problems have long been observed to emerge within social environments characterized by social isolation, strained relationships, decreased social support, and substance-use-facilitating social networks. Meaningful social relations are theorized to help combat addiction, and many substance use interventions have been developed based on this premise. Although various social processes are robustly associated with substance problems, the underlying mechanisms linking them remain unclear. We propose a cascades model that describes the bidirectional effects between social environment and substance use, where deterioration or improvement in the social domain can drive changes in substance use, and vice versa. Leveraging over two thousand effect sizes extracted from RCTs of substance use interventions, our study aims to test the cascades model to provide a comprehensive understanding of the relationship between pathogenic social environments and substance problems. Below, we note the incremental study aims pursued to achieve this goal. Our aims are derived such that evidence for the cascades model can be accumulated through progressive levels of analysis, beginning with tests of association, such as those reflected in prior research, and ultimately culminating in the test of cross-domain effects that leverage the design features of RCTs.

Transparency and Openness

The conduct and reporting of the current study follow the Preferred Reporting Items for Systematic reviews and Meta-Analyses statement (PRISMA2020; Page et al., 2021). The protocol for the review and meta-analysis was preregistered with the PROSPERO database (ID: CRD42024536527), an international systematic review registry. All study aims were preregistered, with the exception of Aim 2 (cross-lagged analysis), which was identified as a viable analysis after all effect sizes had been coded. We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. Data and code for replicating the primary results of this meta-analysis can be found at https://doi.org/10.17605/OSF.IO/W693F. There was no use of generative artificial intelligence in the preparation of this research.

Information Sources and Search Strategy

The search aimed to identify randomized controlled trials (RCTs) that manipulated and/or quantitatively evaluated social outcomes in treatment-seeking populations with substance use problems. A systematic search of PubMed, PsycINFO, and ProQuest Dissertations and Theses Global databases was conducted between May 8^th to 10^th, 2024. The key search terms used to identify articles focused on psychological interventions, social outcomes, and substance use are listed in Table 1. In addition, the reference lists of 18 published systematic reviews and/or meta-analyses – covering topics including psychological interventions to improve social outcomes (e.g., Zagic et al., 2022), social network/support interventions (e.g., Swinkels et al., 2023), and randomized controlled trials for treating SUD (e.g., Dutra et al., 2008) – were scanned for eligible studies. Reference lists of eligible studies were scanned to identify any additional studies that met the inclusion criteria. A total of 20,927 abstracts were scanned for potential inclusion in this review (see Figure 3 for PRISMA flow diagram). For the complete list of review articles as well as the search strategy, filters, and limits used, see Supplemental Material Table S1.

Eligibility Criteria

Selection Process

The study selection process is illustrated in Figure 3. The first author screened the titles and abstracts of study records and removed duplicate articles. Following the initial screening stage, the full text of each remaining article was independently reviewed by the first author and trained research assistants to determine its eligibility. Any conflicts of opinion regarding eligibility were resolved through discussion or by consulting a third reviewer (the second author of this study). In cases where studies did not include sufficient statistical information for effect size calculations, the corresponding and/or first authors were contacted directly. If there was no response to the initial email, two additional emails were sent for follow-up. Studies were ultimately excluded from the meta-analysis if study authors no longer had access to the requested data and/or if we did not hear back a month after the second follow-up email. Finally, we were able to obtain raw data from four large-scale RCTs through open-data sharing initiatives, including the MATCH study (Project MATCH Research Group, 1998) and studies from the NIDA Clinical Trials Network (Tai et al., 2021).

Study Coding and Effect Size Calculation

Development of the data extraction protocol for study characteristics was guided by the PICO framework (i.e., Patient/Population, Intervention, Comparator, Outcome) (Schardt et al., 2007). A structured table was used to guide the data extraction process. All study characteristics and effect size calculations were independently coded by both the first author and the third author to minimize errors. The coding process had two stages. First, study information (i.e., PICO characteristics) as well as raw effect size data (e.g., sample size, mean, SD) were blind coded and then cross-checked between two coders for each included study. Second, each coder then independently calculated effect sizes based on cross-checked data (inter-rater reliability ICC = .99). Disagreements for study characteristics coding were resolved via discussion.

The following Patient characteristics were coded: (a) sample size; (b) attrition rate, defined as proportion of participants who discontinued the treatment or were lost to follow-up across the entire duration of the study; (c) average age of participants at study initiation; (d) sex (% males); (e) race/ethnicity (% non-White); (f) type of primary substance used (alcohol, nicotine, cannabis, opioids, stimulants, and mixed); (g) substance use-related inclusion criteria applied in the study; (h) substance use severity at baseline (high: the majority of the sample exhibited symptoms of dependency; moderate: the sample met diagnostic criteria for a SUD without signs of dependency; low: treatment-seeking individuals not meeting formal diagnoses).

The following Intervention characteristics were coded: (a) theoretical orientation guiding the intervention (12-step-oriented; acceptance and commitment therapy; behavioral treatment; contingency management approach; family system orientation; motivational interviewing; unspecified social models, etc.); (b) whether there is a significant “social component” associated with the intervention (yes/no); (c) for interventions with a significant social component (i.e., “socially oriented interventions”), type of social component (community reinforcement; family involvement; group therapy; social network intervention; or social support); (d) for interventions without a significant social component (i.e., “individually oriented interventions”), whether the treatment includes any “social module,” defined as behavioral change techniques that operate primarily through “bodily process” as mechanism of action but may still indirectly influence individuals’ social context (e.g., social skills training, refusal skills, assertiveness training) (yes/no); (e) the number of “social modules” included in individually oriented interventions; (f) planned treatment duration (in months); (g) planned number of total sessions; (h) planned session duration. Based on information gathered for (g) and (h), treatment intensity was estimated by calculating the total number of intervention hours participants assigned to each treatment were expected to receive.

The following Comparator characteristics were coded: (a) nature of the control group (alternative treatment; attention placebo control; treatment element placebo; treatment-as-usual; wait-list control); (b) comparison type between intervention and control group (Social vs Individual treatment; Individual treatment vs inactive/minimally active control; Social treatment vs inactive/minimally active control); (c) planned treatment duration (in months); (e) number of total sessions; (f) session duration. Based on information gathered for (e) and (f), treatment intensity was estimated with the total number of hours participants assigned to the control group were expected to receive.

The following Outcome characteristics were coded: (a) type of social outcome measure assessed (relationship quality; social network characteristics; social support; social functioning); (b) type of substance measured (alcohol; barbiturates; cannabis; nicotine; opioids; stimulant; sedatives; or unspecified); (c) type of substance outcome measure assessed (abstinence; frequency; quantity; heavy/binge drinking; substance-use related problems; severity; or other); (d) for each coded effect size, the study stage at the time of assessment (pre-treatment; treatment; post-treatment; or follow-up) and (e) the assessment timing, defined in months post-intake, were recorded.

Other information coded included publication status of the study and year of publication.

Risk of Bias Assessment

We used the revised Cochrane risk of bias tool for randomized trials (Sterne et al., 2019) to assess the quality of our included studies. Each article was evaluated by two research assistants in five domains: bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. An overall level of risk of bias (high risk, some concerns, low risk) for each study was subsequently determined based on the ratings across the five domains. Differences in ratings were resolved by discussion or by consulting a third reviewer.

Data Analytic Plan

All analyses were conducted using R Statistical Software (R Core Team, 2022). Publication bias was assessed using multiple methods: funnel plot symmetry, Duval and Tweedie’s Trim and Fill Test, and the examination of publication status as a moderator in the meta-regression models.

Descriptive Statistics

Our search of the literature produced 79 independent randomized controlled trials meeting eligibility criteria, including 13,946 individual participants. A total of 2,131 effect sizes were derived from these samples – 942 effect sizes for social outcomes (k = 72), 1,032 for substance use outcomes (k = 70), and 157 cross-sectional correlations between social and substance use outcome measures (k = 8). Note that, although correlation values were reported in fewer studies, most were derived from participant-level data in the context of large-scale RCTs (also see Selection Process). On average, across all studies, participants were 64.11% male, with an average age of 34.52 (SD = 10.35, range = 15.13 ~ 55.36) at study initiation. 89.87% of studies (k = 71) were published, with the earliest article appearing in 1986. Included studies were predominantly conducted in western cultures: of our 79 samples, 86.08% (k = 68) were collected in the US, 5.06% (k = 4) in Europe, 7.59% (k = 6) in Asia, and 1.27% (k = 1) in the Middle East. The majority of studies (k = 51; 64.56%) recruited clinical samples characterized by high levels of substance use symptom severity. Across all studies, nearly one-third of patients (k = 21; N = 4,150) primarily presented with problems associated with alcohol use, followed by nicotine use (k = 11; N = 3,647), opioid use (k = 12; N = 984), cannabis use (k = 4; N = 694), and stimulant (cocaine/methamphetamine) use (k = 7; N = 615); the remaining studies (k = 24; N = 3,856) recruited patients with various substance use issues without targeting a specific type of substance problem for treatment. A detailed summary of study characteristics is provided in Supplemental Material Table S4.

The most commonly reported effect sizes for social outcome measures assessed close relationship quality, accounting for 38.39% of effect sizes. Social network characteristics – including the function and composition of patients’ social networks – represented the second most common social outcome (31.95%), followed by social support (19.77%), and social functioning (9.89%). Table 2 presents definitions, representative measures, and sample items used to assess social outcomes considered in this study. For substance use outcomes, the most commonly reported effect sizes were frequency of use (42.87%), followed by abstinence (16.51%), severity of SUD symptoms (13.18%), quantity of use (12.35%), alcohol- or drug-related problems (8.43%), heavy/binge drinking (5.11%), and other substance use outcomes (1.54%). In terms of study quality, 74% of studies received an overall high risk of bias rating, 21.9% were rated as some concerns, and 4.1% were rated as low risk (see Supplemental Material Table S3 for a complete list of ratings by domain).

Publication Bias and Manipulation Checks

Publication bias was examined separately for social and substance use outcomes. Publication status did not emerge as a significant moderator for either social outcome effect sizes, b = 0.16, se = 0.11, p = .160, or substance use outcome effect sizes, b = 0.22, se = 0.18, p = .227, though published studies tended to yield larger effects than did unpublished studies. Additionally, we visually inspected funnel plots (Figure 4) generated using both effect sizes aggregated at the study level following the Borenstein et al. (2009) procedure as well as all individual effect sizes. In each case, effect sizes were plotted against their corresponding standard errors. Across both types of outcomes and plotting procedures, while outliers were present, effects were reasonably evenly distributed around the mean, approximating the shape of a funnel and suggesting no substantial evidence of bias. This observation was further supported by the Trim and Fill test, which indicated no missing studies due to publication bias.

The categorization of socially vs individually oriented interventions represents an important distinction in the manipulation of substance use levels and social processes in the current review, but the inherently multifaceted nature of substance use interventions means few are truly “asocial.” Nonetheless, of 17 studies that incorporated individually oriented intervention arms, which were ultimately compared against a minimally active control group, 14 incorporated some social modules (Maximum = 1-2 units), and 3 included no social modules. When estimated as a proportion of total therapy hours, social units accounted for an average of 14.2% of treatment content within these interventions. An additional 34 studies incorporated an individually oriented treatment as a comparison group against a socially oriented treatment; of these, 1 included no social units, 17 included 1-3 social units (8.0% treatment content), and 16 could not be precisely quantified due to incomplete reporting. Taken together, findings of this coding effort indicate that, while no approach to categorization is perfect, our individual-level intervention category nonetheless represents a reasonable heuristic for identifying interventions with low to negligible social content.

Overall, substance use treatments produced a small but significant effect after aggregating all substance outcomes across post-baseline time points, based on 842 effect sizes, d [95% CIs] = 0.26 [0.15, 0.37], g = 0.26 [0.15, 0.36], k = 70. This confirms that, on average, substance use treatments were delivered effectively and produced an expected effect of reducing substance use/problems among treatment-seeking individuals with substance problems. Further, when comparing substance use intervention with a significant social component to an active, individually oriented substance use treatment (“Social vs Individual Comparison”), multilevel models based on 320 effect sizes yielded a small but significant treatment effect on social outcomes, d = 0.15 [0.05, 0.25], g = 0.15 [0.05, 0.25], k = 26. This finding indicates that incorporating a significant social component into substance use interventions provides added social benefits above and beyond those offered through individually oriented substance use treatments.

Aim 1: Correlation between Social and Substance Use Outcomes

Aim 2: Cross-Lagged Analyses

To this point, we have established correlations between substance use and social outcomes not only at the level of individual participants, but also in the magnitude of randomized intervention effects. However, while establishing core associations, these initial analyses fail to address the question of temporal precedence – which comes first? Note that studies included in our meta-analysis not only contained intervention effect sizes at the level of the study, but also frequently contained repeated measurements of both social and substance use outcomes as these emerged in response to treatment over time. As such, leveraging this data to further understand this relationship and temporal dimensions thereof, we next conducted lagged analyses aimed at exploring treatment effects over time within studies, examining whether changes in the social domain preceded later changes in the substance use domain, and vice versa. So here, we look at whether treatment-related changes in social outcomes observed across earlier periods of time predicted later changes in substance use in the subsequent period, after controlling for autocorrelation, and vice versa.

To ensure that effect sizes binned together across studies reflect comparable trial stages, here we control for “time since treatment cessation,” such that time 0 corresponds to the end of treatment, negative values indicate months before treatment ends, and positive values indicate months since treatment cessation. Statistical power for the “lagged drug → social” model peaked at a 6-month lag (82.8%; see Figure 6). Cross-lagged analysis at this interval indicated that reductions in substance problems over the preceding 6 months were significantly associated with subsequent improvements in social outcomes, b = 0.39, se = 0.10, t = 3.81, p < 0.001. Similarly, statistical power for the “lagged social → drug” model peaked at a 5-month lag (70.2%; see Figure 6). Cross-lagged analysis at this interval indicated that improvements in social outcomes over the preceding 5 months were significantly associated with subsequent reductions in substance problems, b = 0.33, se = 0.12, t = 2.74, p = .007. Adding treatment duration as a covariate to the cross-lagged models did not change the significance level or the direction of cross-lagged effects. Taken together, these results indicate a bidirectional temporal association between social and substance use outcomes, such that intervention-related improvements in the social domain temporally preceded reduction in substance use problems, and vice versa.

Aim 3: Randomized Treatment Effects of SUD Interventions on Social Outcomes

While our prior analysis established temporal precedence in intervention effects, our findings still reflect correlative relationships between social and substance use outcomes. As such, they fail to rule out third variable confounds. To move beyond correlative findings, we begin by examining the overall effects of substance use interventions on social outcomes relative to active, minimally active, or inactive control groups. Analyses included a total of 695 effect sizes, yielding a small but significant treatment effect of improved social outcomes among treatment-seeking individuals with substance problems, d = 0.15 [0.08, 0.22], g = 0.15 [0.08, 0.22], k = 72 (see Supplemental Material Figure S1).² Results from likelihood ratio tests indicated significant heterogeneity at both the within-study level, $I_{level2}^2=22.21$, ${\sigma }_{level2}^2=0.03$, p < .001, and the between-study level, $I_{level3}^2=49.47$, ${\sigma }_{level3}^2=0.07$, p < .001. Moderator analyses revealed a significant moderating effect of social component type on social outcome effect sizes, with interventions integrating social support strategies or involving family members producing the largest improvements in participants’ social outcomes. In addition, while theoretical orientation was also a significant moderator, post hoc analyses revealed that this was largely driven by interventions using behavioral contingency management, which showed significantly smaller effects than any other theoretical orientation type. Otherwise, treatment effects on social outcomes were largely generalizable across types of social outcome, methodological factors, treatment phases, and population characteristics. Of particular interest, primary abused substance type did not significantly moderate treatment effects on social outcomes, F(5, 692) = 1.05, p = .388. While overall moderation findings were non-significant, exploratory subgroup analyses indicated effect sizes comparable in magnitude among participants primarily using nicotine, g = 0.21 [0.14, 0.28], alcohol, g = 0.19 [0.05, 0.32], and mixed substances, g = 0.17 [0.05, 0.30], with negligible effects observed for opioids, stimulants, and cannabis. Results of all moderator analyses are presented in Table 3.

Aim 4a: Targeted Comparison for the “Substance Use → Social Environment” Pathway

Our prior analyses indicate that substance use interventions had overall positive effects on social outcomes relative to active, minimally active, and inactive controls. However, the observed effects may stem from a combination of two sources: first, the improved social outcomes may result from a direct effect elicited by elements of the psychological interventions, particularly those involving a significant social component; second, it may reflect a downstream effect in response to reduced substance use achieved through substance use treatment.

With the aim of moving towards analyses capable of parsing treatment effects attributable to changes in substance use, we conducted analyses comparing substance use interventions to inactive or minimally active control conditions. Across a total of 378 effect sizes, this analysis revealed a small but significant effect for substance use interventions on social outcomes compared to inactive/minimally active controls, d = 0.15 [0.06, 0.24], g = 0.15 [0.06, 0.23], k = 49. After excluding effect sizes from interventions involving a significant social component, the remaining 78 effect sizes revealed an effect similar in magnitude, albeit now non-significant in this smaller subset of studies, d = 0.13 [−0.04, 0.30], g = 0.13 [−0.04, 0.29], k = 17. Follow-up analyses indicated substantial heterogeneity at the between-study level, $I_{level3}^2=70.11$, ${\sigma }_{level3}^2=0.10$, p < .001. Results of subgroup analyses parsing social outcome types indicated that positive effects selectively emerged for perceived social support, d = 0.16 [0.06, 0.25], g = 0.16 [0.06, 0.25], k = 5, $I_{level3}^2=0$, ${\sigma }_{level3}^2=0.00$, whereas effects were more variable across other types of social outcomes.³ As an exploratory step, we next conducted between-trial comparisons examining the effect of intervention type, with results indicating that trials of individual interventions that integrated some social modules (k=14) outperformed trials of interventions with no explicit social content (k = 3), F(1, 72) = 9.75, p = .003 (63.7% variance explained). While interpretation of individual-only interventions was constrained by representation (k = 3), subgroup analyses indicated that participants who received predominantly individually-oriented intervention integrating minimal social content (1-2 social modules; 17.2% total content) showed significantly improved social outcomes compared with controls, g = 0.22 [0.06, 0.37], k = 14, $I_{level3}^2=62.62$, ${\sigma }_{level3}^2=0.06$. Taken together, while acknowledging the inherently complex nature of individual intervention types currently represented in the addiction literature, results indicate modest support for the premise that improvements in social outcomes were partially driven by reduced substance use, though these downstream effects varied across studies and the type of social outcome measured.

Aim 4b: Targeted Comparison for the “Social Environment → Substance Use” Pathway

Next, we conducted a subgroup analysis of studies employing a “Social vs Individual Comparison” design on substance use outcomes. That is, we aimed to examine whether socially oriented substance use treatments further enhanced substance use outcomes compared to individually oriented treatments. The pooled effect across all studies examined together was small and statistically significant, d = 0.12 [0.02, 0.22], g = 0.12 [0.02, 0.22], k = 32. Substantial heterogeneity was observed at the between-study level, $I_{level-3}^2=84.41$, ${\sigma }_{level3}^2=0.06$, p < .001. Exploratory analysis parsing this effect by social intervention type indicated that the most pronounced effects emerged in treatments involving family members (e.g., couples therapy, family therapy), with a pooled effect size of d = 0.23 [0.05, 0.40], g = 0.22 [0.05, 0.40], k = 17, $I_{level-3}^2=67.46$, ${\sigma }_{level3}^2=0.11$. That is, involving a family member in the treatment provided greater benefits in improving substance use outcomes compared to benefits provided by individually oriented treatments. In contrast, minimal effects were observed for social network interventions, CRA, group therapy, and social support interventions when compared with individually oriented substance use interventions. Together, evidence for the added benefits of socially oriented interventions in enhancing substance use outcomes – above and beyond what is achieved through individually oriented substance use programs – was observed primarily in treatments involving family members, with comparable effects observed in other types of socially oriented interventions.

As an exploratory analysis, we also compared effect sizes for individually oriented interventions that contained some social modules to interventions that integrated no social modules. Note that these individually oriented interventions with social modules tended to represent more comprehensive treatment packages, whereas two studies that integrated no social elements were constrained to contingency management trials. As such, like all meta-analytic cross-study comparisons, these contrasts are likely to include substantial noise both related and unrelated to the integration of social modules. Nonetheless, it is notable that individually oriented treatments with social modules, g = 0.68 [0.06, 1.30], k = 13, yielded larger effects on substance use outcomes than treatments integrating no explicit social components, g = 0.19 [−0.09, 0.47], k = 2. Taken together, these findings tentatively suggest the potential usefulness of including discrete social modules in enhancing treatment effects.

Substance Use Interventions Produced Small but Significant Effects on Social Outcomes

Individuals with SUD often find themselves facing a recovery process characterized by a daunting social reality. Addiction is frequently accompanied by impoverished social networks, featuring high levels of conflict and an absence of meaningful social connections. In tandem with this, individuals may rely on drug use as the primary mechanism for entry and gaining acceptance into social spaces, so much so that substances become the only way they know to maintain the relationships that matter to them. Many individuals with SUDs enter treatment with either no resources or a lack of motivation for forming relationships independent of substance use. As such, while the necessity of altering these pathogenic social environments has been apparent, in light of the severity of the social issues, it is unclear whether interventions would have any effect on the social life of individuals with substance problems, and, even if an effect were to be found, it remains unclear the extent to which the effect would be helpful or harmful in nature. This is because substances can serve as a coping strategy in social settings, providing some social buffering and a fragile sense of belonging, so taking away this coping mechanism through substance use interventions may have an unpredictable impact on individuals’ social lives, at least in the short term. This is the first meta-analysis to address this question.

Results of Aim 3 analyses indicated that psychological interventions produced a small but significant improvement in social outcomes in treatment-seeking populations with substance problems. Although small, these effects can be considered meaningful, as the overwhelming majority of included studies featured control groups that received varying forms of clinical intervention, ranging from minimally active treatments to unspecified standard care to evidence-based substance use treatments. Notably, several included studies employed TAU as a control condition in a parallel-group design, which reflects standard practice in the field of intervention research and an ethical obligation to those in need of treatment. However, both the content and composition of TAU in control conditions are known to impact estimated treatment effects (Löfholm et al., 2013; Santa Ana et al., 2008), resulting in smaller effect size estimates (Götz et al., 2022; Kraft, 2020). Nonetheless, the small effects observed are also consistent with previous meta-analyses that reported small-to-moderate effect sizes when estimating effects of mental health treatments on social outcomes among various psychiatric populations beyond addiction (Hickin et al., 2021; Swinkels et al., 2023; Zagic et al., 2022).

An important theoretical question for the cascades model pertains to its generalizability – does it reflect a “universal” phenomenon, or does it manifest only under specific conditions and populations? In the current review, we observed moderate to high heterogeneity across study aims and levels of analyses—heterogeneity that was comparable to, or even lower than, levels typically reported in psychological research (Stanley et al., 2018). In some cases, this variability could be partially explained by specific moderators, while in others it was reduced through subgroup analyses. Nonetheless, there remains unexplained heterogeneity across levels of analyses, likely reflecting the influence of unmeasured moderators or mediators (e.g., genetic disposition, trait-level characteristics, variability in therapists) on pathways linking social environments and substance use. Here, it may be useful to consider the principle of multifinality in helping explain the variability observed across levels of analysis. That is, pathogenic social environments and substance problems represent two broad “endpoints” among which numerous developmental trajectories may exist – a reflection that various components within systems of social environments and addictive behaviors are not always characterized by one-to-one correspondence. Moreover, these components continue to serve as key players embedded within a web of factors that have the potential to influence the manifestation of specific pathways. In sum, findings generally support the idea that the cascades model reflects a governing principle underlying addiction recovery, though, as with most psychological phenomena, such cascading influence is not uniform.

One source of variability that has been of interest in prior research is drug class. While understanding drug-specific mechanisms of action (e.g., receptor binding) is essential for developing pharmacological treatments (e.g., methadone treatment for opioid use disorder, naltrexone for alcohol dependence, and nicotine replacement therapy for smoking cessation), psychological interventions have been considered as comparatively less drug-specific, with similar techniques or treatment packages routinely applied across drug classes (e.g., refusal skills can be taught in treatments for various types of SUDs). Consistent with such suppositions, results of our moderator analysis (Aim 3) suggested that significant positive effects extended across various participant characteristics, including individuals with mild as well as severe substance use patterns and those using substances ranging from alcohol to opioids. While statistical power could potentially explain the difficulty in detecting moderator effects, subgroup analyses nonetheless indicated that intervention effects on social outcomes were consistently small in magnitude across drug groups. From a theoretical point of view, it is possible for cascades effects to operate in ways that invoke equifinality in social recovery at the drug-domain level, such that individuals across the spectrum of substance-related disorders may find themselves experiencing comparable improvements in social outcomes following treatments.

Social and Substance Use Outcomes Following Treatment are Correlated at Multiple Levels

Our examination of the cascades model linking social outcomes and substance use began with an investigation of the associations between the two domains (Aim 1). As expected, a significant participant-level correlation between social and substance use outcomes emerged, comparable in magnitude to effect sizes reported in previous meta-analyses examining associations between substance use outcomes and indicators of interpersonal attachment style and function/quality of social relationships (Beattie, 2001; Fairbairn et al., 2018). Participant-level correlations are known to be susceptible to individual-level biases, such as stable traits, reporting style, and common method bias; thus, we next conducted additional correlational analyses at the intervention level. Results indicated that interventions producing greater reductions in substance problems also tended to be more effective at improving patients’ social outcomes. Notably, the magnitude of the intervention-level correlation (r = .64) was considerably larger than the participant-level correlation (r = .18). Such discrepancies may reflect differences in this variability of data existing at the participant- versus intervention-level. In other words, it is possible that the participant inclusion criteria set for each study limited the variability in substance use that would be observed across individuals within the same sample, whereas the variability would be less constrained across studies recruiting a diverse range of populations, including those with mild and severe forms of substance use, and implementing various study designs. Another possibility is that individual-level data contain more “noise,” which was reduced when effects were aggregated at the level of the study, thereby yielding stronger intervention-level associations. Together, we note the usefulness of leveraging meta-analytic approaches to examine associations at multiple levels, with intervention-level correlations in particular offering stronger evidence for a potential causal link than individual-level correlations.

In Aim 3, results of cross-lagged analyses suggest that temporal associations between social and substance use outcomes following treatment appear to be bidirectional, with intervention-related improvements in the social domain preceding subsequent declines in substance-related symptoms, and vice versa. The question of temporality might be particularly relevant with processes of slow development (Hill, 1965), such as shifts in social environmental qualities and addictive behaviors, where the temporal ordering of changes is less obvious and can be observed as occurring concurrently. Of note, cross-lagged analyses do not provide enough information to warrant causal claims, as temporal precedence represents a “necessary but not sufficient” condition for causal inference. Further, cross-lagged analyses were conducted on intervention-level aggregated effect sizes in this meta-analysis, so potentially failing to capture individual-level cascades. Nonetheless, these analyses indicate the usefulness of cross-lagged findings as a proxy for examining the temporal ordering between treatment-related changes in social and substance use outcomes.

Evidence for the Substance Use → Social Outcome Pathway Remains Equivocal

Using targeted comparisons (Aim 4a) to further probe the “substance use → social outcome” pathway, we found that treatment effects specifically emerged in the social cognitive domain, with patients receiving individually oriented interventions reporting significantly greater perceived social support than controls. In contrast, individually oriented interventions produced limited effects on more “structural” elements of social life (e.g., social network characteristics, social relationship quality). These findings should be interpreted with caution, as null effects for structural social changes observed in the current review may reflect a combination of factors, including limited statistical power, the nature of the available literature assessing these outcomes, and the use of potentially less sensitive measures. As such, the observed differential effects across social outcome types should not be taken as direct evidence that cascading effects operate selectively through the social cognitive domain. That said, it remains plausible from a theoretical standpoint that different facets of social environments may participate differently in the cascading process. Perhaps compared to repairing social relationships or revamping social networks, which can often take time and may depend on broader environmental contingencies, individually-based treatments may more readily influence how patients perceive and interpret their social worlds. As such, another possibility is that structural changes unfold asynchronously with the intervention effects and may not be fully captured within the assessment windows of most included studies.

Individually oriented interventions with social modules produced greater improvements in social outcomes compared to those without, yet it was unclear the extent to which such changes are driven specifically by reduced substance use vs the inclusion of social modules. As such, causal evidence for the substance use → social outcome pathway remains inconclusive. This is due in part to the limited representation of “purely” individual interventions included for the current review, potentially reflecting a broader pattern of the addiction literature in which there is a lack of asocial interventions, except for atypical trials. Specifically, in the current review, individually oriented interventions without social modules were confined to three contingency management trials characterized by a narrow set of behavioral change techniques. In contrast, individually oriented interventions with discrete social units often represent more comprehensive treatment packages with standard social modules embedded within, such as refusal skills and assertiveness training. Interestingly, in addition to social outcomes, these more comprehensive treatment packages involving discrete social content were also more effective at changing substance use outcomes than those employing limited behavioral change techniques. These findings might also suggest to us that there exist potential boundary conditions for the manifestation of the “substance use → social environment” pathway, including that it may require reaching a certain threshold of improvement in substance use outcomes for downstream effects to occur in the social domain. However, future research is needed to establish causal relations for this specific pathway.

Evidence Provides Support for the Social Environment → Substance Use Pathway

Targeted comparisons (Aim 4b) examining substance use outcomes across trials employing a “Social vs Individual” design suggested that socially oriented interventions significantly outperformed individually oriented interventions, particularly in treatments involving close social relations in the context of couples/family therapy. Given that couples and family therapies represent a well-established area of SUD intervention research, a sizable sample of high-quality RCTs was included in this specific subgroup analysis, providing adequate power to detect such effects. In line with our current findings, previous meta-analytic reviews consistently reported that family involvement produces (added) benefits in terms of reducing substance use, increasing treatment initiation, as well as decreasing other psychiatric symptoms, with small-to-medium effect sizes (Ariss & Fairbairn, 2020; Meis et al., 2013; Merkouris et al., 2022). Results from Aim 4b further complement findings from existing longitudinal studies synthesized by Fairbairn et al. (2018), in which attachment style – the disposition to form social bonds – was found to temporally precede substance use outcomes, but not vice versa. Moreover, our analyses moved beyond associative evidence reported in Fairbairn et al. (2018) by examining the relationship between close social bonds and substance use in the context of RCTs. Overall, converging evidence points to a unique role of close social bonds in addiction recovery, a finding that is somewhat remarkable given longstanding challenges in demonstrating differential effects across various psychotherapy models (Luborsky et al., 2002). Future research is needed to explore what it is about these close social bonds that might render them potentially beneficial in this process.

The finding that family- and couples-based treatments outperformed individual therapy, while other “socially oriented” interventions did not, has important implications for refining the cascades model. This divergence might suggest that not all facets of social environments are created as equal in the context of addiction recovery. Instead of suggesting that all social connections would uniformly influence substance use behaviors, these results point toward a potential hierarchy of social influences, where the unique leverage of intimate, close-proximity relationships outweigh more general social network or peer-based interventions. This perspective also mirrors extensive evidence from ethnological, sociological, and psychological research showing that human social networks are found to be organized in layered structures, characterized by increasing network sizes and decreasing social closeness as social distance increases (Dunbar & Spoors, 1995; Hamilton et al., 2007; Zhou et al., 2005). Emerging evidence further suggests that these different layers of social closeness may even be processed differently at the brain level (Roseman-Shalem et al., 2024). The cascades model may benefit from incorporating this “layered” perspective in conceptualizing the strength of cascading sequences generated from various facets of social environments (e.g., family vs school/workplace vs community), such that closer relationships are more capable of invoking stronger cascades, both positively and negatively. Further research on this hypothesis is needed, as it may help provide a more precise clinical roadmap to help understand specific social mechanisms are more likely than others to trigger the downstream reductions in substance use.

The current review has important implications across several domains. First, for basic research and theory, our work represents the first large-scale effort to build theory that would bridge across two independent bodies of work in the field of addiction, applying a cascades model to the understanding of links between social environments and substance use. Second, for prevention research, the cascades model highlights the need for future research to both identify and develop strategies that effectively disrupt negative cascade sequences early on before they progress into full-blown addiction. At the same time, future research should not overlook any meaningful clinical encounter that may increase the likelihood of initiating a positive cascade (e.g., single-session interventions) that produces positive outcomes down the line. Third, for intervention research, we emphasize the importance of considering both patients’ substance problems and their social environments when refining existing and developing new interventions. By the time individuals decide to seek professional help, particularly when SUDs have progressed to their more severe forms, substance use and impairments in the social domain often become a “package deal,” so considering either substance problems or patients’ social challenges in a vacuum risks overlooking half of the equation. Future research could explore innovative strategies to address the social challenges commonly experienced by this population, with the aim of producing synergistic effects to enhance treatment outcomes.

Limitations and Future Directions

Limitations of the present meta-analysis should be noted. First, our analyses are not capable of identifying the mechanism that explains the cascade effects between social and substance use outcomes during recovery. While it is useful to consider the type of social component, or a lack thereof, involved in substance use interventions as a proxy to infer potential mechanisms targeted by the treatment, it is important not to overstate the specificity of inferences that might be derived from this approach. In reality, many substance use treatments target multiple change processes beyond the “orientation label” they claim, along with non-treatment-specific elements such as therapeutic rapport, all of which may contribute to a positive treatment outcome. Nonetheless, results from these RCTs move toward offering converging evidence for reciprocal effects between social and substance use outcomes, highlighting the importance for future intervention research to examine the change mechanisms associated with substance use interventions.

Second, as with most meta-analytic reviews, despite our extensive efforts to contact study authors, we were unable to include a number of studies in our analyses due to missing data on social outcomes. This limitation may, to some extent, reflect a broader tendency to overlook the systematic assessment and reporting of social outcomes in substance use treatment research. As such, while we found no direct evidence of publication bias, selective reporting of social outcomes may still contribute to a “file drawer” problem, potentially inflating pooled effect sizes observed in the current review. Related to this, several of our subset analyses included relatively few studies, and thus, statistical power for these analyses was limited. Thus, it is important for future studies to consider incorporating sensitive, valid, and diverse social outcome measures into the treatment development and evaluation process.

Third, social network interventions featuring 12-step oriented treatment programs and group therapy were sometimes implemented in the control arm and compared against researcher-designated treatments of interest (e.g., CBT, DBT) delivered with close supervision and high fidelity. This trial context may have potentially attenuated the observed effectiveness of social network approaches and group therapy. It is important to note that our findings should not be interpreted as indicating potential inferiority for group therapy or 12-step programs; instead, both demonstrated comparable effectiveness in treating substance problems when compared to individually oriented approaches, and even may be considered as more cost-effective, offering lower-cost and more accessible treatment options. Thus, factors such as researcher allegiance and treatment fidelity may confound our results, potentially attenuating the observed effectiveness of these interventions. Future research may consider analyses that account for both the dosage and quality of the interventions to provide a fairer evaluation of these treatments.

Fourth, similar to most meta-analyses, our moderator analyses were conducted at the study level for demographic variables, which can be underpowered and easily confounded by study-level factors. With the growing interest in transparency and open data initiatives, future reviews may consider analyzing participant-level data and examining whether effects observed in our current meta-analysis replicate. Fifth, our categorization of the control group was broad, encompassing waitlist control, attention placebo, and TAU. The inclusion of TAU in parallel-design studies warrants particular consideration, as TAU may involve active treatment components that mask the effects of the interventions under investigation, contributing to a conservative bias. Nonetheless, due to ethical concerns, studies with truly inactive controls are uncommon in this field, and it is notable that we were still able to detect meaningful effects across analyses in this review. Sixth, our search was restricted only to scholarly work written in English, introducing a mono-language bias that may limit the generalizability of the findings. Future research should include studies written in languages aside from English to evaluate whether the cascades model replicates across diverse linguistic and cultural contexts.

Finally, the focus of our review on substance use interventions and social outcomes does not represent the only approach for investigating the reciprocal effects between social factors and substance use within a meta-analytic framework. For example, pharmacological trials aimed at targeting craving, anhedonia, or other neurobiological processes could also potentially elicit cascading effects to modify patients’ social outcomes, warranting examination in future work. In the present meta-analysis, we had limited the scope of our reviews to psychological interventions only. While some included studies incorporated components that provide patients with access to or information about pharmacotherapy (for a list of promoting pharmacological support BCT, see Supplemental Material Table S2), these elements were implemented across intervention arms, which limited our ability to isolate their unique effects on social outcomes. Beyond pharmacological interventions, future research might also consider synthesizing evidence using different population eligibility criteria to further elucidate diverse mechanisms linking social variables and substance use. Here, we considered social recovery specifically within SUD populations, whereas another approach would be to examine substance use among individuals with social deficits. Social cognitive deficits in substance use populations may, in and of themselves, constitute a chicken-and-egg dilemma – these deficits may precede drug exposure, develop over time as a consequence of chronic substance use, or reflect a combination of both processes – and thus occupy an ambiguous position within the substance use ← →social environments cascades model, though they can certainly participate independently or interact with other cascading sequences that reinforce substance use and deterioration of social functioning. However, our included studies did not measure social cognitive factors (empathy, perspective taking, theory of mind) and lacked specificity in isolating intervention components that directly target these constructs. While our current review may not be best positioned to answer some of these questions, investigating the effects of mental health treatments on substance use outcomes among populations with pronounced social impairments (e.g., social anxiety, autism), for instance, might help shed light on them.

Conclusion

This meta-analysis explored the link between social and substance use outcomes as individuals begin substance use treatments. Evidence partially supports a reciprocal relationship described by the cascades model: benefits gained from either domain – social or substance use – may potentially propagate into the other, driving spillover effects over time. Results carry theoretical implications surrounding the co-occurring substance problems and social impairments frequently observed in populations with SUDs.