Mechanisms of Behavior Change in Substance Use Disorder With and Without Formal Treatment

Abstract

This article provides a narrative review of studies that examined mechanisms of behavior change in substance use disorder. Several mechanisms have some support, including self-efficacy, craving, protective behavioral strategies, and increasing substance-free rewards, whereas others have minimal support (e.g., motivation, identity). The review provides recommendations for expanding the research agenda for studying mechanisms of change, including designs to manipulate putative change mechanisms, measurement approaches that expand the temporal units of analysis during change efforts, more studies of change outside of treatment, and analytic approaches that move beyond mediation tests. The dominant causal inference approach that focuses on treatment and individuals as change agents could be expanded to include a molar behavioral approach that focuses on patterns of behavior in temporally extended environmental contexts. Molar behavioral approaches may advance understanding of how recovery from substance use disorder is influenced by broader contextual features, community-level variables, and social determinants of health.

1. INTRODUCTION

Substance use and substance use disorder (SUD) are leading preventable causes of morbidity and mortality worldwide and are associated with excessive social and economic costs and human suffering. The majority of adults in many parts of the world engage in some substance use, and SUD is one of the most prevalent psychiatric disorders. Substance use and SUD are highly comorbid with psychiatric disorders. As such, clinical psychologists are likely to encounter clients who engage in substance use and may meet criteria for SUD.

The goal of this article is to review the scientific literature that examined how individuals make changes to their substance use behavior and recover from SUD. Most individuals recover from SUD in the absence of formal treatment (Tucker et al. 2020a). Thus, the current review focuses on the processes of behavior change with and without formal treatment and provides a narrative review of studies that have investigated mechanisms of behavior change. The review also provides recommendations for expanding the research agenda.

1.1. Defining Substance Use Disorder, Remission, and Recovery

The Diagnostic and Statistical Manual for Mental Disorders, 5th edition (DSM-5) (Am. Psychiatr. Assoc. 2013) and the International Statistical Classification of Diseases and Related Health Problems, 11th revision (ICD-11) (WHO 2022) currently guide the definition of SUD, and there is some variation in the symptoms necessary to meet criteria for a particular SUD between these two diagnostic manuals. The DSM-5 requires individuals to meet 2 of 11 symptoms in the past 12 months to meet criteria for SUD, whereas the ICD-11 requires 2 of 3 symptoms in the past 12 months. Both systems include symptoms related to loss of control over substance use, and substance use taking over other activities and resulting in life-health problems, tolerance, and physiological withdrawal effects.

The definitions of SUD based on the DSM-5 and ICD-11 represent disorders that are heterogeneous with respect to etiology and clinical presentation and for which all symptoms are not yet agreed upon. With 11 potential symptoms and only 2 symptoms required to meet criteria for SUD per the DSM-5, there are 2,048 symptom combinations that could result in the diagnosis of SUD based on DSM-5 (Witkiewitz et al. 2019). The ICD-11 definition of SUD has 3 potential symptoms and 2 symptoms required to meet criteria for SUD; thus, there are 8 symptom combinations that could result in the diagnosis of SUD based on ICD-11.

The DSM-5 and ICD also provide somewhat divergent definitions of remission from SUD. Per the DSM-5, remission is based solely on not meeting symptoms of the disorder, and abstinence or the amount of substance consumed is not considered in defining remission. The ICD also has definitions of remission that are based on not meeting symptoms of the disorder and further requires abstinence from substance use (e.g., sustained full remission is defined as meeting no criteria of disorder and abstinence lasting 12 months or longer).

Other recent consensus definitions of recovery from SUD include broader aspects of functioning than the narrowly defined criteria for remission outlined in the DSM-5 and ICD. For example, the Substance Abuse and Mental Health Services Administration defines recovery as “a process of change through which individuals improve their health and wellness, live a self-directed life, and strive to reach their full potential” (SAMHSA 2011, p. 3). Based on the latest empirical research examining recovery pathways, our research team has defined recovery as “a dynamic process of change characterized by improvements in health and social functioning, as well as increases in well-being and purpose in life” (Witkiewitz et al. 2020, p. 10). These definitions, which prioritize health and functioning, do not consider the amount of substance consumed by individuals, and do not require abstinence from substance use, represent a shift from the ICD definition of remission and from prior conceptualizations of recovery as requiring sustained abstinence.

1.2. Epidemiology of Substance Use Disorder and Recovery

According to the 2019 National Survey on Drug Use and Health (NSDUH; Lipari & Park-Lee 2019), the majority of individuals aged 12 or older in the United States (60.1%) used psychoactive substances in the past month, including 50.8% who used alcohol and 13.0% who used an illicit drug (including cannabis, which is legal in some US states). The 2012–2013 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) found that nearly one-third of the US population (29.1%) met lifetime criteria for alcohol use disorder (AUD) (Grant et al. 2015), and nearly one-tenth (9.9%) met lifetime criteria for another SUD (Grant et al. 2016).

Population data also indicate that remission or recovery from SUD is the most common outcome for individuals who use substances and previously met criteria for SUD (Fan et al. 2019, McCabe et al. 2018). Further, the majority of those who start using substances during adolescence and young adulthood will desist from risky substance use during young adulthood, a process sometimes termed maturing out (Dawson et al. 2006).

Epidemiological data, longitudinal studies, and cross-sectional retrospective surveys have found that many individuals recover from alcohol and other drug problems in the absence of any formal treatment (e.g., alcohol specialty treatment, counseling services conducted by a licensed treatment professional) or participation in mutual help groups [e.g., Alcoholics Anonymous (AA)]. Thus, “natural” recovery, or recovery in the absence of treatment, is the most common pathway, not the exception (for a review, see Tucker et al. 2020a). For example, among those surveyed in NESARC who met criteria for AUD, 65.8% of respondents experienced at least partial remission from AUD, and only 22.8% ever received formal treatment (Fan et al. 2019).

1.3. Treatment of Substance Use Disorder

Although most individuals with SUD never receive treatment and the majority recover without it, studies of mechanisms of behavior change among persons with SUD have largely focused on those who are seeking treatment, engaging in research on specific treatments for SUD, and/or participating in mutual help organizations (Magill et al. 2020). A much smaller number of studies examined mechanisms of behavior change outside of the treatment context. Because this review necessarily relies heavily on research with individuals who received treatment, the next section provides a brief overview of treatment-seeking and treatments for SUD.

1.3.1. Treatment-seeking and the treatment gap.

In the 2019 NSDUH, approximately 1.5% of US residents aged 12 and older (4.2 million people) received some type of substance use treatment in the past year, and 1.0% (2.6 million people) received substance use treatment at a specialty facility (Lipari & Park-Lee 2019). Of the 21.6 million people aged 12 or older in 2019 who needed substance use treatment, only 12.2% received treatment at an SUD treatment specialty facility. The majority of those who did not seek treatment (95.7%) reported that they did not feel they needed treatment.

Seeking treatment for SUD is often driven by family, court, or employer mandates (Edlund et al. 2012) rather than by self-motivation or self-selection into treatment. Individuals who meet criteria for SUD and who experience greater problem severity are more likely to seek treatment (Edlund et al. 2012), but other influences such as the role of the social network in promoting or discouraging substance use and treatment-seeking also need to be considered in understanding mechanisms of behavior change (Longabaugh et al. 2010). However, pathways into treatment do not necessarily influence outcomes, as those who are mandated to receive treatment show outcomes comparable to those of nonmandated individuals (Weisner et al. 2009).

1.3.2. Modest efficacy of current treatments.

In the second half of the twentieth century, the general field of psychotherapy research focused on the question “Does treatment work?” The primary method used to answer this question is the randomized clinical trial, and the field searched intensively for the most efficacious treatments, whether behavioral, pharmacological, or a combination of the two. Miller & Wilbourne (2002) reviewed progress to date in alcohol treatment research at the turn of the twenty-first century and documented trends also found in other areas of psychology and psychiatry. Specifically, substantial progress had been made in the development of a number of efficacious treatments, treatments for SUD were moderately efficacious, and most treatments were similarly efficacious despite being guided by diverse underlying theories of change.

Given this state of evidence, in recent decades researchers have shifted their focus to why particular treatments are effective and for whom. The goal of this work is to better understand the mechanisms by which treatments are effective so that existing treatments can be optimized (Longabaugh & Magill 2011). Gaining a better understanding of mechanisms of behavior change among people who change in the absence of treatment could also help refine prevention and treatment approaches in order to target the most critical change mechanisms.

1.3.3. Pretreatment change and the placebo effect.

In addition to the prevalence of recovery in the absence of treatment, many individuals with SUD who seek treatment will change their behavior before starting treatment (Epstein et al. 2005). This pretreatment change phenomenon may be due to events and change processes that occur before treatment is initiated; for example, seeking treatment may help consolidate initial self-change efforts (Tucker 1995). Additionally, several studies indicate significant changes in substance use behavior among those individuals who receive placebo medications (Weiss et al. 2008). Selection effects into treatment (self versus other initiated), pretreatment change, and the placebo effect each bring additional complexity to the study of mechanisms of behavior change.

2. HOW DO WE IDENTIFY MECHANISMS OF BEHAVIOR CHANGE?

A mechanism of behavior change can be defined as the process(es) by which behavior change occurs; in the treatment setting, it can be defined as the mechanisms of a specific treatment or the process(es) through which a specific treatment effects change (Kazdin 2007). The question of how to identify a mechanism of behavior change is far more complex and requires consideration of many interacting variables. Hill (1965) outlined key considerations for causal inference in observational data, which have been adapted to recommendations for studying mechanisms of behavior change in psychotherapy studies (Kazdin 2007, Kazdin & Nock 2003). Although the causal inference approach described below dominates research examining mechanisms of behavior change in SUD, there is also some relevant research guided by a molar behavioral perspective, which focuses on changes in patterns of behavior in environmental contexts over time (Baum 2002, Rachlin 1992, Tucker et al. 1995, Vuchinich 1995). From this perspective, behavior change is investigated through the study of temporally extended environment–behavior associations, intertemporal choice, and allocation of behavior and resources to substance use and substance-free activities in context.

2.1. Causal Inference Approach to Studying Mechanisms of Behavior Change

The key requirements for establishing a mechanism of behavior change based on a causal inference approach are strong association, specificity, gradient, experiment, temporal relation, consistency, and plausibility and coherence (Hill 1965, Kazdin & Nock 2003). Statistical mediation has been the dominant methodological approach to testing mechanisms of behavior change from a causal inference perspective. Researchers have considered many different statistical mediators, defined as intervening variables that explain an association between the treatment (independent variable) and the outcome (dependent variables), to support specific mechanisms of change. Yet, even a statistically significant mediating (indirect) relationship might not explain the process(es) by which change occurs, and establishing mediation is not sufficient to establish a mechanism of behavior change.

A heuristic representation of the key requirements for testing mechanisms of behavior change from a causal inference framework (Hill 1965, Kazdin 2007) is shown in Figure 1 with the statistical mediation model (black outlined boxes) at the center of the figure. Most published studies that examined mechanisms of behavior change used statistical mediation tests designed to establish the significance of the mediated effect (e.g., the product of coefficients approach tests whether the product of the a-path and the b-path is significant), and the c′-path (the association between the independent variable and the outcome when the mediator is included in the model) is not significant (MacKinnon 2008), which may provide evidence for the strong association requirement. Plausibility and coherence are defined by testing mechanisms that are supported by the scientific knowledge base or clinical experience. Temporal precedence is established by measuring the mediators and outcomes across multiple time points and providing evidence that the independent variable precedes and causes the mediator and that the mediator precedes and causes the outcome (Krentzman et al. 2013, MacKinnon 2008, Witkiewitz et al. 2018). Gradient, which assesses whether greater changes in proposed mechanisms are related to subsequent greater changes in outcomes, is sometimes considered, particularly with respect to the dose of treatment received and the number of treatment sessions attended (Witkiewitz et al. 2011). Consistency of replication, which is the replication of findings across samples, is rare, although some examples exist (Hartzler et al. 2011, Maisto et al. 2015).

Figure 1.

The statistical mediation model (black outlined boxes) at the center of the figure shows the steps for a simple mediation model in which an independent variable (such as treatment) is strongly associated with the mediator (the a-path) and the mediator is strongly associated with the outcome (the b-path). The association between the independent variable and the outcome without the mediator in the model (c-path, solid line) should be reduced when the mediator is included in the model (c′-path, dashed line). As described by Hill (1965) and Kazdin (2007), other requirements for establishing a mechanism of behavior change are also displayed, including plausibility and coherence of the proposed mechanisms based on scientific literature or clinical practice, consistency of findings via replication, temporal precedence of the independent variable preceding the mediator and the mediator preceding the outcome, gradient determining whether greater changes in proposed mechanisms are related to subsequent greater changes in mediator and outcomes, specificity that other unmeasured variables do not cause the mediator or outcomes, and experimental control of the independent variable.

The experiment requirement is partially supported in studies where the independent variable of treatment is randomized, but this provides only partial support because even when the independent variable is randomized, the mediator typically cannot be randomized. Therefore, while the a-path is an unbiased estimate of the causal effect of the experiment on the mediator, both the b-path and the c′-path are not causal. Experimental approaches to targeting the mediator provide one potential strategy for meeting the experiment requirement for causal inference, yet it is often difficult to target processes of change (Shadel et al. 2017).

A primary reason why these paths may not be causal is the inability to rule out additional omitted variables that could provide an alternative explanation for the mechanism of change (Tofighi et al. 2019). Such an omitted variable, called an omitted confounder, could provide an alternative explanation for the mechanism through which behavior change occurs. The specificity assumption, which is also known as the no-omitted-confounder assumption (Tofighi et al. 2019), is only rarely addressed in SUD research (Hsiao et al. 2019). Specificity assumes that any changes in the outcome are uniquely explained by the hypothesized mediator and that other variables (i.e., omitted confounders) do not influence behavior change (MacKinnon 2008).

Hill (1965) also outlined several additional important considerations for causal inference that were not described as key requirements either by Hill or in subsequent work applied to psychotherapy studies per Kazdin & Nock (2003). First, statistical significance does not in and of itself provide evidence of strong association, and association does not indicate causation. Statistical significance can be driven by sample size, such that small mediation effects could be statistically significant even when, in practical terms, the effects are not large enough to support a strong association or a meaningful indirect effect. Standardized effect size measures are available to provide a better metric of the strength of the association in mediation analyses (Miočević et al. 2018). Further, even large associations might not be robust to potential confounding relationships (Hsiao et al. 2019, Tofighi et al. 2019), such that even in the presence of strong relationships, the evidence for a mechanism of behavior change may not be sufficient to conclude that all environmental, behavioral, and contextual variables were accounted for by the model. Limited research considers whether mediation effects are free from confounding or whether there is strong evidence to support a purported mechanism of behavior change using the Hill (1965) or Kazdin & Nock (2003) criteria for testing mechanisms of behavior change.

2.2. Molar Behavioral Approach to Studying Mechanisms of Behavior Change

In contrast to the causal inference approach, the molar behavioral approach seeks to identify environment–behavior associations over time in the service of a final cause analysis (Baum 2002, Rachlin 1992). Specifically, there is recognition that discrete behavioral acts are emitted as one part of a broader pattern of behavior that occurs over time in an environmental context. Thus, mechanisms of behavior change can be understood only by studying the pattern of behavior over time. Further, contexts change over time and effect changes in behavioral patterning. This account takes a broader temporal perspective than a causal inference approach to studying antecedents and consequences at a particular point in time or over a delayed time course (as in longitudinal panel data) and seeks to measure meaningful associations between contextual variables that reliably covary with behavior patterns over time. It does not require that a given environmental stimulus precede a discrete behavioral act in a temporally contiguous relationship, as required in the causal inference approach, but instead focuses on establishing correlations over time between environmental conditions and patterns of behavior. Five decades of basic and applied research on behavioral choice support this molar approach to understanding behavioral allocation patterns, including allocation patterns that involve the use of substances (Bickel et al. 2014, Vuchinich 1995).

2.3. Summary of Approaches to Studying Mechanisms of Behavior Change

The causal inference and molar behavioral approaches have distinct assumptions about behavior change that guide the experimental and other empirical approaches to studying mechanisms of change. To date, the majority of research has been conducted within a causal inference framework and has not examined behavior–environment associations that may support behavior change. With these inherent limitations considered, the next section of this review provides a summary of available research that has examined mechanisms of behavior change in SUD. We begin with an overview of mechanisms of behavior change that have been examined within the context of formal treatments. The subsequent section covers the wide range of mechanisms of behavior change that may occur within and outside of the treatment context (“common” mechanisms). A summary of the review findings is provided in Table 1.

Table 1.

Mechanisms of behavior change in substance use disorder

Mechanism	Evidence for mechanism of behavior change	Targeted by which treatments?	Relevant citation(s)
Coping skills	Mixed: some evidence for quality and breadth of coping skills rather than utilization	CBT, network support therapy, CM	Magill et al. (2020)
Contingent incentives	Some evidence for incentives contingent upon negative urine samples, but no evidence for incentives contingent upon other behaviors (attending treatment, taking medication)	CM	Litt et al. (2008), Petry & Carroll (2013)
In-session client language	Both change talk and sustain talk during sessions may mediate outcomes; more robust evidence for sustain talk	MI, CBI	Magill et al. (2018)
Therapist language and behavior	Robust evidence for working alliance as a predictor of outcomes, but less evidence for clinician relational skills and outcomes	MI, CBI	Magill et al. (2018)
Craving	Evidence for reductions in self-reported craving and neural reactivity to craving as mediators of substance use treatment outcomes	Mindfulness-based interventions, CBI, CBT, naltrexone, topiramate, methadone	Hölzel et al. (2011), Schacht et al. (2013), Witkiewitz et al. (2014)
Mutual help involvement	AA social network support and greater involvement with mutual help groups, and not attendance alone, associated with better outcomes	AA, TSF	Kelly (2017), Kelly et al. (2011)
Motivation to change	Some evidence for in-session client language, particularly sustain talk (i.e., lack of motivation to change); however, robust investigations are lacking	MI, CBI, CBT	Apodaca & Longabaugh (2009), Magill et al. (2018)
Self-efficacy	Self-efficacy is a robust mediator of substance use outcomes, but it does not appear to be selective to a specific treatment, and directionality of self-efficacy to outcomes is unclear	MI, CBI, CBT, CM, network support therapy	Kadden & Litt (2011)
Social support	Positive social support, particularly positive social support that focuses on substance use, is a more robust mediator of outcomes	MI, CBI, CBT, TSF, network support therapy	Longabaugh et al. (2010), McCrady (2004)
Affective states	Negative affect is associated with substance use outcomes, less support for positive affect; generally, association is not selective to a specific treatment, and directionality of affect to outcomes is unclear	Integrated treatments that target affective symptoms and substance use, antidepressants	Baker et al. (2012), Martin et al. (2011), Witkiewitz & Villarroel (2009)
Motives and expectancies	Motives and expectancies predict substance use behaviors, but few studies have evaluated temporal precedence of motives and expectancies as mechanisms of behavior change	CBI, CBT, expectancy challenge intervention	Gesualdo & Pinquart (2021), Votaw & Witkiewitz (2021)
Substance use identity	Some evidence of substance use identity being associated with substance use outcomes, but no studies evaluated identity change as a mechanism	Cognitive training	Montes & Pearson (2021)
Protective behavioral strategies (PBS)	Good evidence of PBS as a mechanism of change, particularly for alcohol harm reduction; less evidence for PBS direct actions to limit or change manner of drinking or for other substance use	Brief interventions, SFAS, PBS interventions	Leeman et al. (2016), Martens et al. (2011), Murphy et al. (2012), Pearson (2013)
Behavioral economic indicators	Reducing demand for substances by increasing engagement in future-oriented positive alternatives to substance use, changing behavioral allocation away from substances to substance-free delayed rewards, and measures of substance demand and discounting predict substance use outcomes	SFAS, community reinforcement approach, CM, network support therapy, TSF, CBT, experimental manipulations	Acuff et al. (2020), Murphy et al. (2022), Tucker et al. (2021)
Neurobiological, neurocognitive, and physiological factors	Some evidence that brain structure, connectivity, behavioral control, reward system processing, neural cue reactivity, stress physiology, and neural activation during discounting tasks predict substance use outcomes	CBT, CM, MI, mindfulness-based interventions, heart rate variability biofeedback, naltrexone, goal management training, cognitive training	Eddie et al. (2022), Ekhtiari et al. (2016), Owens et al. (2019), Verdejo-García et al. (2008)

Abbreviations: AA, Alcoholics Anonymous; CBI, combined behavioral intervention; CBT, cognitive behavioral therapy; CM, contingency management; MI, motivational interviewing; PBS, protective behavioral strategies; SFAS, substance-free activity session; TSF, 12-step facilitation.

3. TREATMENT-SPECIFIC MECHANISMS

Several forms of behavioral and pharmacological treatments are available for the treatment of SUD. As described next, there are unique therapeutic elements of these treatments that are hypothesized to affect specific mechanisms of behavior change.

3.1. Mechanisms of Behavior Change in Cognitive Behavioral Therapy

Cognitive behavioral therapy (CBT) is a behavioral treatment for SUD that focuses on modifying an individual’s thoughts and feelings as they relate to behavior (Carroll 1999). CBT targets antecedents to substance use, such as cognitions (e.g., “I am a failure”), emotions (e.g., sadness or anxiety), and situational factors (e.g., a social gathering where alcohol is freely available). Once these antecedents to substance use are identified, individuals learn specific coping skills, such as reappraisal of cognitions, affect regulation, development of values-based alternatives to substance use, and substance use refusal skills. CBT exerts causal effects on substance use outcomes, and these outcomes are superior to the effects of minimally treated comparison conditions (Magill et al. 2019).

The hypothesized mechanism of behavior change in CBT is coping skill utilization and enactment, whereby CBT enhances individuals’ coping skills and increased coping skill utilization is thought to support reductions in substance use or abstinence. Based on the results of systematic reviews, the evidence on coping skills as a mediator of the association between CBT and substance use outcomes is mixed (e.g., Magill et al. 2020). Research has indicated that the quality and breadth of coping skills (e.g., using a wide array of coping responses consistently that are tailored to specific substance use situations), not the quantity, likely robustly mediate the effects of CBT on substance use outcomes (Kiluk et al. 2010, Witkiewitz et al. 2018). Despite this promising evidence for coping as a mediator of CBT, it remains unclear whether coping is unique to CBT because evidence indicates that other behavior treatments, including network support therapy and contingency management (CM), also enhance coping skills (Litt et al. 2008).

3.2. Mechanisms of Behavior Change in Contingency Management

CM is a behavioral treatment that involves providing individuals with tangible incentives upon emission of a target behavior. Tangible incentives are provided in the form of prizes or vouchers with discrete monetary values that can be exchanged for retail goods or services. Target behaviors include objectively verified abstinence or engagement in a variety of desirable therapeutic behaviors (e.g., attending treatment, taking prescribed medication). Research on CM indicates that, relative to no-incentive comparison conditions, CM significantly increases abstinence and attendance among individuals with SUD (Pfund et al. 2022), and greater magnitude and more frequent incentives are significantly associated with abstinence (Petry et al. 2004), supporting a dose–response relationship.

The hypothesized mechanism of behavior change in CM is the contingent-incentive system, particularly its effect on increasing engagement in target behaviors. A study by Litt and colleagues (2008) indicated that increases in marijuana abstinence at posttreatment mediated the association between CM and marijuana abstinence at 5–8 months posttreatment. Studies by Petry and colleagues (e.g., Petry & Carroll 2013) found that increases in abstinence mediated the association between CM and positive treatment outcomes. Specifically, CM was positively associated with cocaine abstinence at posttreatment, and cocaine abstinence at posttreatment in turn was positively associated with improved quality of life and lower depressive symptoms at 9 months following treatment. These studies provide emerging support that, in abstinence-incentive systems that require participants to submit urine samples that are negative for drugs, increases in abstinence may serve as a mechanism of behavior change in CM. However, no studies have been conducted on increases in positive target behaviors as mediators of the association between CM and outcomes.

3.3. Mechanisms of Behavior Change in Motivational Interviewing

Motivational interviewing (MI) is a “collaborative conversation style for strengthening a person’s own motivation and commitment to change” (Miller & Rollnick 2013, p. 12). MI is geared toward strengthening individuals’ motivation using a combination of both relational skills (i.e., empathy, collaboration) and technical (e.g., open-ended questions, reflections) skills. Motivation and commitment are reflected in individuals’ in-session statements. Statements in favor of changing substance use (e.g., “I need to stop using”) are referred to as “change talk,” and statements against changing substance use (e.g., “I feel better when I use”) are referred to as “sustain talk.”

MI yields modest effects on substance use outcomes compared with standard care control conditions (Miller & Wilbourne 2002). At present, three hypotheses about how MI works have been proposed: the technical hypothesis, the relational hypothesis, and the conflict resolution hypothesis (Magill & Hallgren 2019). Most research centers on testing the first hypothesis that technical skills affect language uttered by the client during treatment sessions (i.e., change talk and sustain talk) and that these statements (particularly greater change talk and less sustain talk) subsequently affect outcomes. Several studies support that individuals’ statements within MI sessions mediate the association between MI consistent skills and substance use outcomes (e.g., Barnett et al. 2014). However, examinations of the individual paths of the statistical mediation model suggest that sustain talk is a more consistent predictor of unfavorable outcomes than change talk is of favorable outcomes (Magill et al. 2018).

The relational hypothesis of MI suggests that clinicians’ use of relational skills facilitates individuals’ engagement in the change process (e.g., expressing interest in change, seeking information about change) and that engagement subsequently affects substance use outcomes. Some studies of clinicians’ use of relational skills generally indicated that empathy during an MI session was positively associated with engagement and negatively associated with substance use outcomes (e.g., Moyers et al. 2016). Several studies found that client-reported strength of the working alliance with the therapist was associated with better outcomes (Hartzler et al. 2011, Maisto et al. 2015). However, meta-analyses have not demonstrated a robust association between relational skills and substance use outcomes (Magill et al. 2018). One possible explanation for this discrepant finding is that studies evaluating the efficacy of MI generally comprise clinicians with high relational skills, and there is too little variability to establish an association between relational skills and outcomes (Magill & Hallgren 2019).

The conflict resolution hypothesis, which theorizes that exploration and resolution of ambivalence mediate the association between MI and substance use outcomes, is the least researched of the three mechanistic hypotheses. Perhaps the primary reason for the lack of research is that measuring and statistically examining a dynamic process like ambivalence has been challenging (Magill & Hallgren 2019). Presently, there is no agreed-upon measurement of ambivalence, and research on the conflict resolution hypothesis may not progress until there is consistent support for measurement of ambivalence.

3.4. Mechanisms of Behavior Change in Mindfulness-Based Treatments

Over the past two decades, SUD treatments have increasingly incorporated mindfulness-based practices. Contemporary mindfulness-based treatments are conceptualized and “practiced as a type of focused, purposeful, nonjudgmental attention” (Witkiewitz et al. 2014, p. 514). For example, mindfulness-based relapse prevention was developed as a manualized aftercare treatment that incorporates mindfulness practices and CBT skills (Bowen et al. 2011), and mindfulness-oriented recovery enhancement was developed as a stand-alone treatment that combines mindfulness practices, CBT skills, and positive psychology approaches (Garland et al. 2019).

Mindfulness-based interventions are increasingly supported as an efficacious SUD treatment (Korecki et al. 2020), but there has been less work on the mechanisms of behavior change in mindfulness-based interventions. Multiple mechanisms have been proposed as potential mediators of treatment effects including attentional control, present-moment awareness, craving, emotion- and self-regulation, mindfulness, changes in brain function and structure, and acceptance (Garland et al. 2019, Hölzel et al. 2011, Witkiewitz et al. 2014). A few such proposed mechanisms, including craving, self-compassion, self-regulation, mindfulness, awareness, and acceptance, have been tested and supported in mediation models (Garland et al. 2019, Roos et al. 2019, Witkiewitz et al. 2013). Existing research, however, on mechanisms of behavior change in mindfulness-based interventions has evaluated few of the requirements for causal inference.

3.5. Mechanisms of Behavior Change in Mutual Help/12-Step Facilitation

Mutual help groups and 12-step facilitation (TSF) are grounded in the principles of AA (AA 1953, Nowinski et al. 1995). Peers typically lead mutual help groups, whereas treatment professionals lead TSF. Both treatments involve unique practices, including acceptance of powerlessness, belief in a higher power, commitment to AA, and disease attribution (Kelly 2017). Research generally indicates that greater mutual help group involvement is associated with better outcomes, and TSF significantly increases mutual help group attendance and produces moderately better outcomes than standard care control conditions and may be as efficacious as other active treatment conditions (e.g., CBT) and treatment-as-usual conditions such as medication management (Kelly et al. 2020).

The proposed central mechanisms of behavior change in mutual help groups/TSF are spirituality and mutually shared experiences. It is theorized that participation in these treatments increases spirituality and mutually shared experiences, which in turn promote better substance use outcomes. However, reviews have found little support for spirituality and mutually shared experiences as mechanisms of behavior change (Kelly 2017). Presently, research suggests that people may use mutual help groups/TSF in different ways and that mutual help/TSF may mobilize other social, cognitive, and affective mechanisms (Kelly 2017) and provide social support for abstinence (Kelly et al. 2011).

3.6. Mechanisms of Behavior Change in Pharmacological Treatments

During the twentieth century, SUD became increasingly conceptualized as a medical condition amenable to pharmacological treatments rather than as a moral problem that can be controlled with rational decision making (Dakwar & Kleber 2017). This medical disease conceptualization aims to understand what factors contribute to the transition from casual to uncontrollable substance use. Thus far, research on the medical disease conceptualization identifies withdrawal, craving and cue reactivity, and aberrant reward processing as the neurobiological vulnerabilities to the onset of SUD and the target of pharmacological treatments (Dakwar & Kleber 2017).

For withdrawal, pharmacological treatments are used either to produce similar effects on receptors as reinforcing substances or to address the specific symptoms that are associated with withdrawal (e.g., sleep and mood disturbances, restlessness) (Dakwar & Kleber 2017). For example, agonist treatments like benzodiazepines, which affect gamma-aminobutyric acid similarly to alcohol, are used during acute withdrawal to reduce aversive withdrawal symptoms and are then gradually tapered to facilitate comfortable discontinuation (Witkiewitz et al. 2019).

Research suggests that both agonist and antagonist treatments reduce craving and cue reactivity. Methadone for opioid use disorder, for example, may protect against withdrawal symptoms and also reduce craving and cue reactivity (Ling et al. 1996). By comparison, antagonist treatments block substances from activating receptors to inhibit the reinforcing effects of substance use and subsequently reduce craving (Dakwar & Kleber 2017). Topiramate, a partial glutamate antagonist, and naltrexone, a mu-opioid receptor antagonist, may be effective by reducing craving among individuals who have AUD (Kranzler & McKay 2012).

Various pharmacological treatments have shown promise for addressing aberrant reward processing, and these treatments aim to normalize reward salience by targeting specific neurotransmitters. Treatments like amphetamine and modafinil are intended to target presynaptic and synaptic levels of dopamine and to increase reward salience to non–substance use activities, but there is little research support for their efficacy (Dakwar & Kleber 2017). Antidepressants like selective serotonin reuptake inhibitors are another treatment intended to restore reward processing through the resolution of negative affective states, and there is some preliminary support that, when coupled with CM, they promote abstinence among individuals who use cocaine (Torrens et al. 2005). Other nascent research suggests that pharmacological treatments like ketamine may normalize the balance between the valuation of immediate substance use and delayed non–substance use rewards (Dakwar & Kleber 2017).

4. COMMON MECHANISMS OF BEHAVIOR CHANGE

All the previously described SUD treatments comprise unique therapeutic elements, and these treatments confer significantly more benefit on outcomes than standard care conditions (typically community treatment programs that rely on 12-step programming and psychoeducation). Yet, head-to-head comparisons of these treatments generally indicate no significant differences in outcomes (Kelly et al. 2020). Thus, the SUD treatment literature is no exception to the general psychotherapy outcome literature finding about the equivalence of psychotherapy outcomes established several decades ago (Luborsky et al. 2002). The lack of significant differences in outcomes suggests that the unique therapeutic elements (i.e., elements that are common to all or most treatments) as well as factors outside of treatment may serve as mechanisms of behavior change. This section summarizes findings from treatment research on common mechanisms of behavior change (see also Table 1).

4.1. Motivation to Change

Motivation or readiness to change has long been considered an important mechanism of behavior change. Various measurements have been used to capture motivation, including analysis of insession language, movement through various stages of change, and several self-report measures of motivation. Research has indicated that higher motivation based on these measures generally predicts more favorable substance use outcomes (Bertholet et al. 2009). Research on motivation as a mediator of outcomes in SUD treatment has provided some support for in-session language, but there has been a lack of experimental studies conducted with stringent criteria for mediation (Apodaca & Longabaugh 2009). Further, few studies have examined motivation to change over time as a mechanism of change.

4.2. Coping and Coping Skill Utilization

Coping is a broad construct that encompasses skills such as emotion regulation, flexibility, distress tolerance, and mindfulness. Coping skill utilization has been commonly examined as a mechanism of behavior change in CBT for SUD (Magill et al. 2020), and this research has suggested that coping enhancements during treatment are associated with more favorable substance use outcomes at long-term follow-up (Kiluk et al. 2010; Roos et al. 2017, 2020), providing evidence of coping as a mechanism of behavior change. To date, most studies have examined changes in coping behavior in the context of treatment, and studies on coping behavior without treatment have focused on coping styles (e.g., problem solving, expressing emotion, avoiding the situation). These studies indicated that individuals who naturally recovered from AUD relied less on avoidant coping than did individuals with current heavy drinking (Russell et al. 2001).

4.3. Self-Efficacy

Self-efficacy is defined as an individual’s belief in their ability to achieve a desired outcome in prospective substance use situations (Bandura 1977). In theory, individuals with greater self-efficacy are more likely to resist substance use in high-risk situations, whereas individuals with lower self-efficacy are more likely to engage in substance use. Research supports this prediction and indicates that greater self-efficacy is associated with lower quantity and frequency of substance use (Kadden & Litt 2011). Treatment appears important in enhancing self-efficacy (Hartzler et al. 2011, Maisto et al. 2015, Witkiewitz et al. 2012), but it is unclear what treatment components enhance self-efficacy and how self-efficacy may change outside of treatment. Further, the direction of the association between self-efficacy and outcomes may be such that improvements in substance use predict greater self-efficacy (Kadden & Litt 2011).

4.4. Social Support Networks and Social Support

Social environments are integral to facilitating (or hindering) salutary substance use outcomes (McCrady 2004). Changes in social networks, like reducing contact with people who use substances, increasing contact with people who do not use substances, and engaging with people in mutual help groups, are important predictors of recovery (Kelly et al. 2011). Furthermore, individuals who build social networks that offer positive social support (e.g., family or friends who encourage help-seeking), as opposed to negative social support (e.g., family or friends who discourage help-seeking), show better recovery outcomes (McCrady 2004). Importantly, the research on social support as a predictor of substance use has found that social support specific to substance use may be more important in predicting substance use outcomes compared with general social support (Longabaugh et al. 2010).

4.5. Affective States

Negative affective states like anger, anxiety, and depression commonly co-occur with SUD. Studies of treatments have generally found that negative affective states are associated with more substance use and more substance use–related consequences (Sliedrecht et al. 2019, Witkiewitz et al. 2015). Conversely, positive affective states like greater quality of life and purpose in life have been associated with more favorable substance use outcomes (Martin et al. 2011). Integrated treatments targeting both substance use and affective states, primarily anxiety and depression, have greater effects on substance use and affective states than single focused treatments (Baker et al. 2012).

Outside of the treatment context, there is mixed support on the association between affective states and recovery outcomes. Some studies found that positive affective states are associated with more favorable recovery outcomes (Elms et al. 2018, Russell et al. 2001), and other studies found no association (Bischof et al. 2005). However, this research has multiple limitations, including the reliance on self-reported affect and recovery outcomes (Elms et al. 2018) and variation in how affective states are defined and assessed (Bischof et al. 2005). The direction of the association between affect and outcomes is unclear, and improvements in substance use may predict changes in affect (Witkiewitz & Villarroel 2009).

4.6. Motives, Expectancies, and Identity

According to the motivational model of substance use (Cox & Klinger 1988), substance use is influenced by complex relationships among expectancies for the desired effects of a substance, contextual features, and incentives. Specifically, individual learning history and past reinforcement of substance use and current context are hypothesized to be associated with the expected effects of a substance, and that association in turn is hypothesized to predict an individual’s decision to use substances in the moment (Cox & Klinger 1988). The motivational model of substance use has been widely tested in numerous studies using a causal inference approach, yet most studies have relied on trait measures of motives and have not examined contextual features, expectancies, or momentary motives for substance use (Votaw & Witkiewitz 2021). Further, the few studies that have tested the temporal precedence of motives as a mechanism of change have produced equivocal results (Votaw & Witkiewitz 2021). Relatedly, expectancy theory proposes that substance use is influenced by expectations of what substance use will provide in context, such that substance use behavior depends on prior learning and memories of reward or punishment from a given behavior in a given context. Motives and expectancies have been shown to predict substance use behavior, and there is some evidence that specific treatments, such as expectancy challenge interventions, may be particularly effective in targeting motives and expectancies, particularly for alcohol use (Gesualdo & Pinquart 2021).

Substance use identity, defined as identifying as an individual who uses substances, has been associated with substance use, craving, and substance-related consequences (Montes & Pearson 2021). Recovery identities are associated with well-being and recovery from SUD (Dingle et al. 2019), and identity has been proposed as a potential mechanism of behavior change. However, attempts to change drinking identity via cognitive training have not yielded positive findings (Lindgren et al. 2015). The mechanisms through which identity is associated with substance use are thus unclear, and whether identity can be explicitly targeted via intervention is an understudied but important area for future research.

4.7. Protective Behavioral Strategies

Promoting the use of protective behavioral strategies (PBS) is an increasingly common component of brief motivational interventions particularly for young adult risky drinkers (Leeman et al. 2016, Murphy et al. 2012). PBS involve direct actions to stop, limit, or change the manner of drinking to reduce risks (e.g., alternating alcoholic and nonalcoholic drinks) or indirect actions to minimize or avoid risks of drinking (e.g., having a designated driver). Use of indirect strategies appears relatively more appealing to college students (Leeman et al. 2016), and their use is more predictive of reduced alcohol-related problems than use of direct strategies (Martens et al. 2011). Furthermore, PBS use is more consistently associated with reductions in alcohol-related negative consequences than with reductions in alcohol consumption (Leeman et al. 2016; Murphy et al. 2012, 2019; Pearson 2013; Voss et al. 2018). Collectively, findings to date support targeting PBS use as a mechanism of change at least among young adult risky drinkers who wish to limit drinking-related harms but continue to consume alcohol. Future studies should investigate the utility of PBS to reduce harms associated with use of other substances and in other populations.

4.8. Behavioral Economic Indicators of Strength of Preference for Substance Use

Behavioral economic models are guided by molar behaviorism and posit that recovery from SUD requires a temporal shift from a shorter to a longer view of the future and shifts in resource allocation (e.g., time and money) away from drug use toward valuable delayed non-substance-related rewards, which reinforce and stabilize the shifted recovery behavior patterns (Tucker et al. 2021). Such environmental enrichment reduces substance use and related harms (Acuff et al. 2020), and promoting such shifts in behavioral allocation has emerged as a key intervention target and outcome in behavioral economic interventions for harmful substance use (e.g., Murphy et al. 2012, 2019). Further, measures of relative behavioral allocation to substances add new information about problem severity that has predictive utility beyond substance use practices and problems (e.g., Tucker et al. 2009, 2020b, 2021).

4.8.1. Promoting substance-free activity engagement.

Several related lines of research have supported reducing demand for substances by increasing engagement in future-oriented positive alternatives to substance use. First, behavioral economic interventions like the substance-free activity session (SFAS) that is offered with a brief motivational intervention explicitly target increasing access to drug-free rewards and have been shown to significantly decrease alcohol use and problems out to 16 months postintervention in college students (Murphy et al. 2012, 2019). In addition, several efficacious SUD treatments include elements aimed at promoting engagement in beneficial drug-free activities, including the community reinforcement approach, CM, behavioral activation, CBT, network support therapy, and TSF, among others (Murphy et al. 2022), and numerous studies of individuals who sought help from treatment and/or AA have similarly found that stable long-term recovery is associated with postresolution improvements in life-health functioning (Eddie et al. 2021, Tucker et al. 2002).

4.8.2. Behavioral and monetary allocation.

Second, prospective studies of natural recovery attempts among untreated persons with AUD have shown that successful recovery, particularly moderation drinking, is accompanied by significant shifts in how participants allocate their money during the year before and after initiation of a recovery attempt. An integrative data analysis of five prospective natural recovery studies (Tucker et al. 2020b, 2021) yielded two major findings relevant to understanding the behavioral regulation processes involved in maintaining stable abstinence or moderation drinking during the year after recovery initiation. Those attempting natural recovery who had more balanced monetary allocation patterns (i.e., a more balanced index of proportional monetary allocation to alcohol versus savings), indicative of greater sensitivity to longer-term contingencies, were more successful in maintaining moderation drinking than those with less balanced patterns. Furthermore, those who successfully moderated their drinking had higher-value postresolution consumption bundles that involved receipt of large, theretofore delayed rewards (e.g., housing). These findings generally revealed multidimensional contexts defined in terms of economic behavior that were favorable and unfavorable to achieving stable recovery.

4.8.3. Measuring and manipulating economic demand and delay discounting.

Third, a large experimental literature has manipulated and measured behavioral economic demand for addictive commodities, typically by assessing the relative reinforcing value of substances using purchase tasks or choice procedures (Acuff et al. 2020). In a meta-analysis of 34 relevant experiments, cue exposure and reinforcer magnitude manipulations resulted in significant increases in substance demand metrics across studies, and stress/negative affect manipulations resulted in a significant increase in a single demand metric (O_max reflecting the maximum expenditure on substances). Pharmacotherapy, behavioral intervention, and external contingency manipulations produced a significant decrease in a single demand metric (intensity or demand when substances are free). Substance type explained some of the heterogeneity in findings. The authors concluded that the results suggested different mechanistic properties across demand manipulations and that various contextual factors can contribute to dynamic within-person fluctuations in demand. These mixed findings raise multiple future research questions concerning behavioral economic mediators, and this body of work illustrates how experimental analysis of demand indices can be used to understand the effects and mechanisms of change in SUD.

Longitudinal studies have been inconsistent in showing predicted associations between longer temporal discounting rates, reductions in substance use, and better substance-related outcomes, particularly among younger adults (e.g., Murphy et al. 2012, 2019). Although interventions like episodic future thinking (EFT) (Snider et al. 2016) have been developed to reduce discounting rates and lengthen the time frames over which individuals with SUD make choices between smaller sooner and larger later rewards, findings to date are mixed, and additional studies are needed.

4.9. Neurobiological, Neurocognitive, and Physiological Factors

Although less well investigated than the treatment-specific and common change mechanisms reviewed thus far, recent research investigating the role of neurobiological, neurocognitive, and physiological factors in SUD and behavior change has yielded some intriguing preliminary findings that are summarized here.

4.9.1. Brain structure and connectivity.

Models of neuroadaptation in SUD have been well established in the basic science literature (Koob & Volkow 2016). A small but growing literature has examined the neurobiological and physiological mechanisms of behavior change in SUD. Studies have provided compelling evidence of alterations in brain volume in individuals with SUD (Pando-Naude et al. 2021), including consistent cross-sectional evidence of lower gray and white matter volume in areas of the brain that support behavioral control, reward responsivity, interoceptive awareness, and attention among those with SUD compared with those without SUD. Substance use and recovery from substance use might also affect intrinsic functional connectivity (Nixon & Lewis 2019). Studies have generally found that connectivity integrity improves with abstinence, such that longer-term abstinence from substances may be associated with stronger executive functioning connectivity, weaker reward system connectivity, and decreased functional connectivity in interoception, inhibitory control, and reward circuits over time among those who persist in substance use.

4.9.2. Behavioral and cognitive control.

Lower behavioral and cognitive control, commonly measured by response inhibition during behavioral tasks and self-reported impulsivity, tend to be observed among individuals with SUD, and a limited number of studies suggest that lower control predicts poorer treatment outcomes in SUD treatment (Verdejo-García et al. 2008). Behavioral and self-report measures often modestly correlate with abnormalities in brain structure and function (Balodis & Potenza 2020), and these deficits can be targeted via cognitive interventions (e.g., goal management training; Anderson et al. 2021).

4.9.3. Neural activation to cues and craving.

A large research literature on craving and cue reactivity generally indicates that individuals with SUD have greater craving and reactivity to drug cues (Schacht et al. 2013) and that greater activation in areas of the brain involved in reward processing during the presentation of substance use cues is associated with poorer treatment outcomes (Seo et al. 2013). Several studies have targeted craving and cue reactivity using a variety of treatment approaches, and some evidence suggests that reductions in craving are a mechanism of behavior change following specific interventions (i.e., mindfulness-based interventions, coping interventions) (Westbrook et al. 2011, Witkiewitz et al. 2011).

4.9.4. Stress physiology.

Stress, negative emotional states, and physiological reactivity have also been widely studied, and neurocognitive and physiological differences have been observed in individuals with and without SUD and among those who relapse to substance use. Engagement of stress circuitry that includes the hypothalamus–pituitary–adrenal (HPA) axis and the extended amygdala is known to play a key role in the development of cue and stress reactivity responses to alcohol (Blaine & Sinha 2017). In addition to neural measures of stress and negative affect, heart rate variability and other cardiovascular processes have been proposed to serve as mechanisms that may influence substance use behaviors and potentially could be targeted directly with biofeedback interventions (Eddie et al. 2022).

4.9.5. Neuroeconomics.

Neuroeconomics attempts to bring together a behavioral economic perspective on substance use with cognitive neuroscience in an effort to better understand decision-making processes and neural correlates of behavioral choices (Stanger et al. 2013). Several studies have examined neural correlates of reward-based decision making, temporal discounting, substance demand, and substance-related reinforcement and have generally found involvement of frontostriatal circuitry and connectivity to be associated with impulsive and impaired choices (Owens et al. 2019). Neural activation during temporal discounting tasks was found to be associated with substance use treatment outcomes in an intervention that incorporated CBT, CM, and MI among adolescents with alcohol and/or cannabis use disorder, such that greater activation of reward-motivation networks during impulsive choices predicted a greater percentage of alcohol and/or cannabis use days during and following treatment (Elton et al. 2019).

4.9.6. Summary of work examining neurocognitive factors.

Although the aforementioned studies yielded some intriguing preliminary findings about the role of neurobiological, neurocognitive, and physiological factors in SUD and behavior change, few studies used within-subjects designs that followed an individual across time. Thus, how the brain may change and the neuroadaptations that result from changes in substance use are not well understood. The neurocognitive mechanisms by which behavioral change occurs for specific treatments are even less understood. Preliminary evidence from multiple studies points to a variety of mechanisms, depending on the specific treatment (Schacht et al. 2017, Westbrook et al. 2011). For example, mindfulness-based interventions may be effective in reducing anterior cingulate activation (Westbrook et al. 2011), and naltrexone may be effective via reductions in striatal activation (Schacht et al. 2017). Moreover, most research examining neurocognitive and physiological mechanisms of SUD has been conducted in laboratory settings that have little ecological validity, has used small samples and potentially unreliable experimental tasks, and has been limited by neuroimaging protocols that may not translate to real-world behavior change (Ekhtiari et al. 2016).

5. FUTURE DIRECTIONS: NEXT STEPS AND IMPORTANT QUESTIONS TO BE ADDRESSED

5.1. Broadening Definitions of Recovery and Successful Outcomes

Most of the research reviewed is based on studies that have explicitly focused on substance use outcomes and levels of substance use. Many studies used binary abstinence outcomes as indicating success in treatment, and thus a great deal of information may have been lost about the reductions in use that might occur and the mechanisms of behavior change in supporting substance use reductions and recovery outcomes that are not based on substance use (e.g., well-being, purpose in life). Recent work has moved toward expanding definitions of SUD recovery to consider broader life-health functioning, well-being, quality of life, and purpose in life as outcomes that are more meaningful to persons with lived experience of SUD and that may encourage more individuals with SUD to engage in treatment and community support services (Witkiewitz et al. 2020). It is unclear whether the same mechanisms of behavior change involved in reducing substance use will be supported with respect to these broader functional outcomes. For example, behavioral treatments or pharmacotherapies that help reduce craving may be effective in reducing substance use via reductions in craving, but such treatments might not be effective in changing patterns of behavior other than substance use. Although reducing substance use allows an opportunity to establish new behavior patterns and make other behavior changes (e.g., engaging in substance-free activities), it could also be the case that reducing craving likely does little to support other lifestyle changes.

More generally, research to date has largely focused on individual change in the person receiving treatment and viewed the individual as the primary change agent, and the broader socioecological context and social determinants of health that might support or hinder recovery efforts have been neglected. The molar behavioral model has much to offer the field for thinking about mechanisms of behavior change from a broader socioecological framework, given that the study of behavior–environment patterns over time requires careful attention to the broader contextual features of the environment that will facilitate or hinder specific behavioral allocation patterns. The SFAS intervention research of Murphy and colleagues (2012, 2019) and the natural recovery research of Tucker and colleagues (2021) epitomize this molar approach and illustrate how meaningful measures of behavioral allocation can be collected in the natural environments in which drinking is maintained and changed over long periods of time.

5.2. Future Directions for Research Design

Most of the research reviewed is based on longitudinal panel designs that test statistical mediation using a causal inference approach. Thus, a recommended direction for future research is to improve upon several dimensions of research design, measurement, and analysis, as described here.

5.2.1. Design issues: manipulating the mechanism and measuring temporal patterning.

From a causal inference perspective, research is needed that goes beyond longitudinal panel designs and assumptions of temporality from longitudinal panel data. Experimentation is a central feature of the causal inference approach, and recent work has considered the possibility of direct manipulation of the proposed mechanism of change (Acuff et al. 2020). For example, Shadel and colleagues (2017) attempted to directly manipulate self-efficacy in a study examining changes in self-efficacy as the mechanism of change in smoking cessation treatment. The self-efficacy manipulation increased smoking cessation rates, but the mediating relationship among the treatment, self-reported self-efficacy, and smoking outcomes was not supported. In other words, directly manipulating self-efficacy resulted in better outcomes, but change in self-reported self-efficacy was not the mechanism by which change occurred. Other work attempted to manipulate therapist behaviors via a dismantling randomized clinical trial design (Morgenstern et al. 2012) and also did not find support for the proposed mechanism of change.

From a molar behavioral perspective, research designs are needed that can investigate the extended temporal patterning of behavior–environment associations that involve substance-related and substance-free behaviors. Many aspects of the environmental context that matter to human behavior cannot be experimentally manipulated, but behavior can be measured longitudinally to assess and represent shifts in behavior allocation patterns over time. Carefully considering the broader behavior–environment contexts, relations between rates of drug-related and drug-free reinforcement, and rates of substance use and engagement in drug-free alternatives could advance understanding of how individuals change behavior and what processes are associated with positive change. For example, Tucker and colleagues (e.g., 2009, 2021) conducted several studies of monetary expenditure and drinking behavior patterns at the daily level that were aggregated over varying intervals with the goal of understanding how people make changes in both processes over time.

5.2.2. Measurement issues: expanding the temporal units of analysis.

In an effort to move away from simple mediation models tested using longitudinal panel data, in the last two decades research on mechanisms of behavior change has become far more focused on smaller units of analysis, such as neural activation during a range of behavioral tasks, targeting of specific brain regions with noninvasive brain stimulation techniques (e.g., transcranial direct current and transcranial magnetic stimulation), and measurement of client utterances during treatment sessions (Feldstein Ewing et al. 2015). Generally, this work has found only modest correlations between neuroimaging parameters or client language and substance use outcomes, and many trials have failed to show an effect of targeting neural deficits with noninvasive forms of brain stimulation (Ekhtiari et al. 2019). Much of this work has been unable to measure causal relationships in a causal inference framework and has been far removed from considering environment–behavior covariation in context.

To address limitations of prior work and to gain a better understanding of mechanisms of behavior change in SUD, we propose that the field expand the temporal units of analysis in the measurement of potential mechanisms. Prior work has focused largely on the individual as change agent and measured individual differences in behavior at discrete temporally contiguous points in time, but behavior is continuously evolving and occurs in context. The field needs to move toward measurement of broader contextual features, community- and neighborhood-level variables, and social determinants of health. For example, individuals within communities with greater alcohol outlet density versus those within communities with more substance-free recreational opportunities may experience disparate substance use trajectories as well as pathways to recovery (Babor et al. 2011). Ignoring community-level effects may obscure important differences in how people initiate and maintain changes in substance use behavior.

We also propose that the field needs to focus on improving methods and measurement to assess individual mechanisms of behavior change in near real time, ideally via ambulatory assessment methods that can dynamically adjust to individual context and with minimal burden to participants. Momentary assessment and mobile health intervention tools can be used to deliver just-in-time adaptive interventions informed by microrandomized trials, which allow for the measurement of the effects of an intervention on purported mechanisms and behavioral outcomes in context (Klasnja et al. 2015). These tools will also benefit from ambulatory assessment and just-in-time adaptive assessment methods that take a more idiographic approach to measuring behaviors that are most important in understanding individualized behavior–environment associations with excellent temporal resolution (Burgess-Hull & Epstein 2021).

5.2.3. Analytic issues: move beyond simple mediation tests.

From an analytic perspective, the field should move beyond simple tests of statistical mediation and consider the broader array of quantitative methods that could be used to explore mechanisms of behavior change (Hallgren et al. 2018). Advances in methods for testing causal inference, including tests of the no-omitted-confounder assumption, should be routinely implemented in studies examining mediation using a causal inference framework (Hsiao et al. 2019, Tofighi et al. 2019). Time-varying effects models can be used to measure covariation in environment–behavioral outcomes over time and may allow greater understanding of the extended temporal environmental patterns that support or hinder behavior change (Meisel et al. 2021). Mixture modeling approaches may be useful in reducing heterogeneity in studies of environmental factors and substance use behavior (Vest et al. 2020, Witkiewitz et al. 2018). Mathematical modeling, dynamical systems theory, and machine learning methods may also be helpful in capturing nonlinear and person-specific associations between contextual and behavioral associations (Bekele-Maxwell et al. 2018, Chow et al. 2015, Soyster et al. 2021).

5.2.4. Analytic issues: considering individualized prediction.

Extending person-centered and person-specific analyses one step further, it is important to consider the likelihood that many individuals may change in different ways and at different times. Causal inference models assume that all individuals will be changing in the same way at the same time points, whereas a molar behavioral framework assumes that change can be understood by studying temporally extended patterns of behavioral and environmental events in context. It is likely that some subgroups of individuals are changing in similar ways and experience similar temporally extended patterns of behavioral and environmental events, whereas other subgroups of individuals may be changing in unique ways and experience unique patterns of behavioral and environmental contexts. Person-specific machine learning models (Soyster et al. 2021), predicted individual treatment effects models (Kuhlemeier et al. 2021), and likely responder analyses (Laska et al. 2020) are all recently developed analytic approaches that attempt to tease out individual differences in behavior patterns and outcomes, and these approaches hold particular promise for studying individual differences in mechanisms of behavior change.

5.3. Future Directions for Studying Mechanisms of Behavior Change Outside of Treatment and in Diverse Populations

One of the most important steps for advancing understanding of SUD mechanisms of behavior change is to focus more research on behavior change outside of treatment. Most individuals with SUD never receive treatment and still experience recovery from substance use problems (Tucker et al. 2020a). Examining behavior change outside of the treatment context could guide future prevention and intervention approaches and may provide additional information about environment–behavior associations and contexts that are most supportive of behavior change. From a causal inference perspective, it is critical to reconsider whether experimentation is essential and whether other tools might be useful for studying mechanisms of behavior change in the absence of intervention and experimentation. Even in the absence of experimentation, behavior and contexts can be measured and incorporated into longitudinal models of behavior change over time among persons with SUD (Tucker & Roth 2006). Further, studying treatment-by-context interactions is critical to understand behavior change. Treatment interventions do not occur independent of environmental context, and treatment may be more or less effective in mobilizing particular mechanisms of behavior change depending on contextual and socioecological factors (Latkin et al. 2013).

It is also imperative to expand research to historically disadvantaged populations who are underrepresented in research that examines mechanisms of behavior change in SUD and SUD treatment (Eghaneyan et al. 2020, Montgomery et al. 2020). Experiences of socioeconomic discrimination, colonization, and historical trauma are a few of the many risk factors for substance use that might play a particularly important role in understanding behavior change but are not measured in most studies (Haeny et al. 2019).

5.4. Limitations and Conclusions

There is a tremendously vast literature on mechanisms of behavior change in SUD, and the current review provides only a glimpse of research on the topic. The current review was not conducted using a systematic search strategy and did not follow published guidelines for systematic reviews. When possible, we include citations to systematic reviews and meta-analyses, and we encourage further systematic reviews in each of the areas discussed in the current review. Furthermore, most of the reviewed studies were based on a causal inference framework in treatment populations, and there was a dearth of studies examining mechanisms of behavior change outside of the treatment context. In our conclusions, we acknowledge differences between the causal inference and molar behavioral frameworks in understanding mechanisms of behavior change, and this review has inspired greater consideration of the differences and potential for consilience between these two approaches. Greater attention to temporally extended patterns of behavioral and environmental events in broader socioecological contexts may help increase understanding of the nuances of mechanisms of behavior change in SUD that occur with and without the experience of formal treatment.

SUMMARY POINTS.

1. Most individuals recover from alcohol and other drug problems in the absence of formal treatment or participation in mutual help groups, and many of the mechanisms by which people change alcohol and other drug use are not necessarily specific to receiving formal treatment or participating in mutual help.

2. Mechanisms of behavior change can be understood from a causal inference perspective as the process(es) by which a stimulus precedes a discrete behavioral act in a temporally contiguous relationship, or they can be understood from a molar behavioral perspective as meaningful associations between contextual variables that reliably covary with behavior patterns over time.

3. Unique therapeutic elements in behavioral and pharmacological treatments (i.e., elements that are common to all or most treatments) as well as factors outside of treatment may serve as mechanisms of behavior change.

4. Self-efficacy for changing behavior, the ability to cope with experiences of craving and withdrawal without engaging in substance use behavior, availability of substance-free rewards, building a life worth living without substance use, and use of protective behavioral strategies are all mechanisms by which substance use behavior change is likely to occur for many individuals.

5. Future research should examine temporally extended patterns of behavioral and environmental events in broader socioecological contexts to gain more insight into the nuances of mechanisms of behavior change in substance use that occur with and without the experience of formal treatment.