Recovery schools for improving behavioral and academic outcomes among students in recovery from substance use disorders: a systematic review

Substance use disorders (SUDs) among youth are a major public health problem. In the United States, for example, the incidence of SUDs increases steadily after age 12 and peaks among youth ages 18–23 (White et al., 2009). Although not every youth who experiments with alcohol or illicit drugs is diagnosed with an SUD, approximately 7–9% of 12–24 year olds in the United States were admitted for public SUD treatment in 2013 (SAMHSA, 2016). The true prevalence of SUDs among youth in the United States is likely higher than 7–9%, however, given that many youth receive treatment in private or in non‐specialty settings (Laudet, Harris, Kimball, Winters, & Moberg, 2014), and other youth may never receive treatment for their substance use problems (SAMHSA, 2015). Data available on substance use patterns and treatment availability in other nations also suggests that there are significant numbers of youth worldwide in need of some form of substance use treatment and/or aftercare, although this research is primarily limited to developed nations. For example, in Australia, adolescents aged 10–19 years old comprised approximately 12% of all treatment admissions from 2012–2013 (Australian Institute of Health and Welfare, 2014). Similarly, the European Council estimated that across surveyed countries, youth constituted anywhere from 16% (Italy) to 65% (Czech Republic) of the overall substance use treatment population (Council of Europe, Pompidou Group, 2006). And, in 2011, approximately 28,000 adolescents were newly admitted to outpatient treatment across Europe (European Monitoring Centre for Drugs and Drug Addiction, 2013).

Substance use problems can have numerous detrimental consequences on the academic, social, and general well‐being of youth. This might include negative effects on school or work performance, legal problems, and substance use tolerance and progression (American Psychiatric Association [APA], 2013). Prior research has documented the multiple negative effects from prolonged and heavy substance use, including diminished memory and cognitive abilities, reduced grades, a decreased likelihood of finishing high school or attending post‐secondary education, problems attaining or keeping employment, higher rates of more acute and chronic health conditions than those without a history of use, poverty, and family and social problems (Brown & Tapert, 2004; Homel, Thompson, & Leadbeater, 2014; Larm, Hodgins, Larsson, Samuelson, & Tengström, 2008; Leslie et al., 2016; Lisdahl, Wright, Medina‐Kirchner, Maple, & Shollenbarger, 2014; Menasco & Blair, 2014; Newcomb & Bentler, 1988; Patrick, Schulenberg, & O'Malley, 2016; Silins et al., 2014; Squeglia, Jacobus, & Tapert, 2009; Thoma et al., 2011). Given the numerous negative effects associated with heavy substance use among youth, it is important to understand what programs and interventions might be effective in assisting youth with SUDs.

The period of adolescence "is characterized by more biological, psychological, and social role changes than any other stage of life except infancy" (Holmbeck, Devine, Wasserman, Schellinger, & Tuminello, 2012, p. 431). These changes begin in adolescence and continue into emerging adulthood and involve pubertal and cognitive development, and identity, relationship, and achievement transitions (Schulenberg & Maggs, 2001). Generally, the period of adolescence lasts from ages 10–19, although the World Health Organization's term "young people" includes individuals between 10–24 years of age (World Health Organization, 2016).

Adolescent neurodevelopment has been described as a biologically critical period of heightened vulnerability to experimentation and habitual use of substances (Chambers, Taylor, & Potenza, 2003). For example, due to neurodevelopmental processes, adolescents are more prone to impulsive behaviors, risk taking, and drug and reward seeking compared to adults—all of which are behaviors that are linked to substance use (Chambers et al., 2003; Simon & Moghaddam, 2015). Indeed, animal studies of neural processing have demonstrated that adolescent and adult brains respond differently to rewards, with activity in adolescents' brain regions of learning and habit formation potentially predisposing them to form problematic substance use habits (Sturman & Moghaddam, 2012). This area of research has also demonstrated that adolescents consume larger quantities of alcohol than adults, a behavior that continues into adulthood, suggesting that drinking habits begun early in life, i.e., during adolescence, may worsen relative to habits begun during adulthood and extend for longer periods of time (Serlin & Torregrossa, 2015).

In addition to neurodevelopmental findings, research has demonstrated that adolescents develop expectancies about potential benefits or risks of substance use (Schulenberg & Maggs, 2001); these expectancies can vary depending on adolescents' social networks and settings, and affect when and how they decide to use substances. The social context is especially important as adolescents undergo relationship changes with parents by seeking to differentiate themselves as autonomous individuals and youth create relationships with peers that become more influential. Youth also tend to overestimate the prevalence of drinking in their social environment, with college students viewing drinking as highly normative; youth may also experience increased sociability and report better bonding with peers as a result of drinking (Schulenberg & Maggs, 2001). Social influence has also been consistently identified as a predictor of return to substance use among recovering adolescents (Ramo et al., 2012).

In addition, specific changes related to this developmental period have been identified as risk factors for more problematic substance use (e.g., early maturation, affiliation with deviant peers, and conflicts with parents; Weisz & Hawley, 2002). Adolescent development processes as well as the presence of these additional factors are both risk factors for problematic use, setting this population apart from adults and thus, treatments and aftercare supports should take developmental issues into account (Deas, Riggs, Langenbucher, Goldman, & Brown, 2000; Weisz & Hawley, 2002). In support of this contention, addiction literature has documented different substance use precipitants for adolescents than for adults, with adolescents more likely to return to use under social pressure and also exhibiting greater complexity in these patterns (Ramo & Brown, 2008). Adolescents often use multiple substances and may use a broader array of substances than adults (Deas et al., 2000).

Although not all experimentation with substances leads to problematic use, problematic use can transition to a substance use disorder (SUD), i.e., "a maladaptive pattern of substance use leading to clinically significant impairment or distress" (APA, 1994). Substance abuse and dependence were originally diagnosed separately with substance abuse considered an early phase and dependence the more severe manifestation; however, the Diagnostic and Statistical Manual of Mental Disorders‐V (DSM‐V) modified the definition of an SUD to combine abuse and dependence into one category with different levels (APA, 2013). An SUD is manifested in multiple ways, including substance use that repeatedly impacts school or work performance or contributes to legal problems. Individuals with an SUD demonstrate an increasing need for more of the substance to achieve the same effect (i.e., tolerance) and taking the substance for longer or in larger doses and experience unsuccessful attempts to quit using the substance. Thus, recovery from an SUD is a process involving many factors, the hallmark of which is reduction or complete abstinence of use. Recovery also refers to overall healthy functioning and has been defined as "voluntarily sustained control over substance use, which maximises health and wellbeing and participation in the rights, roles and responsibilities of society" (UK Drug Policy Commission, 2012).

Given the biological, psychological, and social developmental changes in adolescence, it is important to attend to their distinct developmental issues when focusing on adolescents in recovery (Weisz & Hawley, 2002). Thus, for youth diagnosed with an SUD there are a variety of adolescent‐specific treatment options available, which fall within a spectrum of varying intensity from early intervention, such as screening, brief intervention, and referral to treatment (SBIRT: Bien, Miller, & Tonigan, 1993; Madras et al., 2009) to intensive inpatient treatment (American Society of Addiction Medicine [ASAM], 2013). However, SUDs are often experienced as chronic conditions; thus, multiple treatment episodes and ongoing recovery supports after treatment are often necessary to assist with the recovery process (Brown et al., 2001; Ramo et al., 2012; White et al., 2004). Indeed, research has demonstrated that youth seeking SUD treatment do not always engage in or successfully complete treatment (Kaminer, Burleson, Burke, & Litt, 2014; Pugatch, Knight, McGuiness, Sherritt, & Levy, 2014; Winters, Stinchfield, Latimer, & Lee, 2007). Lack of engagement in treatment is due to a variety of factors including denial about the extent of the problem, motivation, emotional reasons, life stressors, financial or insurance barriers, peer influence and social norms, and access and availability of substances combined with triggers to use (Gonzales, Anglin, Beattie, Ong, & Glik, 2012; Wisdom, Cavaleri, Gogel, & Nacht, 2011), and has been linked to community factors such as median family income (Jones, Heflinger, & Saunders, 2007). Among youth who successfully complete substance use treatment programs, 45–70% return to substance use within months of treatment discharge (Anderson et al., 2007; Brown, et al., 2001; Ramo et al., 2012; White et al., 2004).

Perhaps the greatest risk of relapse for youth in recovery is during the late adolescent and emerging adulthood years (Anderson, Ramo, Cummins, & Brown, 2010); thus, youth with an SUD require developmentally appropriate, sustained, and multi‐pronged intervention and follow‐up support (Gonzales et al., 2012). To this end, research has demonstrated the importance of structured continuing care supports after treatment for youth in recovery from substance use. These continuing care supports can include, for instance, a dedicated case manager, home visits, meetings with caregivers, or other environmental supports for the youth and their family (Godley et al., 2014; Stanger, Ryan, Scherer, Norton, & Budney, 2015; Tanner‐Smith, Wilson, & Lipsey, 2013). Indeed, youth who engage in recovery supports posttreatment have the greatest likelihood of abstinence from substance use (Brown et al., 2001; Hennessy & Fisher, 2015). Engagement in substance‐free peer environments is one recovery support system that shows particular promise, and has been linked to reduced substance use and increased psychosocial functioning among youth (Anderson, Ramo, Schulte, Cummins, & Brown, 2008; Nelson, Henderson, & Lackey, 2015; Terrion, 2012).

Success and engagement at school and in postsecondary education are critical to healthy youth development. For youth in recovery from SUDs, school attendance, engagement, and achievement build human capital by motivating personal growth, creating new opportunities and social networks, and increasing life satisfaction and meaning (Keane, 2011; Terrion, 2012; 2014). Upon discharge from formal substance use treatment settings, schools become one of the most important social environments in the lives of youth with SUDs. Healthy school peer environments can enable youth to replace substance use behaviors and norms with healthy activities and prosocial, sober peers.

Unfortunately, however, some of the most significant risk factors for substance use are embedded within school environments, including perceived peer use, association with substance‐using peers, alcohol or drug availability, and stressors such as academic challenges (Derzon, 2007; Mason, Mennis, Linker, Bares, & Zaharakis, 2014; Svensson, 2000; Wambeam et al., 2014). Indeed, in a nationwide survey of high school students in the United States, approximately 26% of respondents were offered, sold, or given an illicit drug on school property (CDC, 2012). This trend is especially problematic for youth in recovery from SUDs: for example, in a study of recovering youth, almost all adolescents who returned to their old school after treatment reported being offered drugs on their first day back in high school (Spear & Skala, 1995). College students suffer similar environmental risks, particularly given the high rates of, and social norms that approve of alcohol consumption on campus. For example, a study of seven universities in Great Britain demonstrated that approximately 70% of enrolled students reported heavy drinking at least once during the previous two weeks. Across five New Zealand universities, 37% of student respondents reported at least one binge drinking episode during the previous week and approximately 68% scored in the hazardous drinking range on the Alcohol Use Disorders Identification Test consumption scale (Kypri et al., 2009). In the United States, 35–39% of college students reported binge drinking (i.e., five/four or more drinks in one sitting for males/females, respectively) at least once in the past month (Monitoring the Future, 2013; SAMHSA, 2014) and research has documented a rapid escalation in alcohol consumption during the first two years of college in the United States (Schulenberg & Maggs, 2001). Indeed, approximately one‐half of SUD treatment admissions for college students in the United States were for alcohol use (Center for Behavioral Health Statistics and Quality, 2012). In addition, a recent review highlighted that approximately 20% of college students in the United States use drugs on a monthly basis and that 5% reported daily use (Dennhardt & Murphy, 2013). One study of recovering college students found that the majority of students were addicted to multiple substances while only a few students were solely addicted to alcohol (Cleveland, Harris, Baker, Herbert, & Dean, 2007). Given the prevalence of binge drinking and substance use among this population as well as the social acceptability of alcohol consumption during the college years, it is no surprise that the college environment has been described as "abstinence hostile" for youth in recovery (Cleveland et al., 2007).

Given the many social and environmental challenges faced by youth in recovery from substance use, recovery‐specific institutional supports are increasingly being linked to educational settings. The two primary types of education‐based continuing care supports for youth in recovery are recovery high schools (RHS) and collegiate recovery communities or programs (CRC): both settings will be referred to under the broad umbrella term of "recovery schools" for this review. Federal offices in the United States recognize these two educational programs as viable supports for youth after they complete formal substance use treatment programs (National Institute on Drug Abuse [NIDA], 2014; Office on National Drug Control and Policy [ONDCP], 2014).

RHSs and CRCs seek to build upon the social connectedness contained within the educational environment (Finch & Frieden, 2014). The recovery school model draws heavily on the notion of social capital, or the social support, connections, and access to resources available through social networks, as one of the strongest factors influencing educational and recovery outcomes (Coleman, 1988; Cloud & Granfield, 2004; Meier, 1999). Connectedness between youth, family, peers, school, and community generates social capital and can potentially translate into academic achievement for youth (Meier, 1999).

RHSs support the recovery and academic achievement of students by fostering connectedness and social capital in a setting that addresses issues related to addiction and recovery with dedicated attention to healthy adolescent growth and development. They provide a safe environment for youth in recovery to practice the skills learned in treatment, use structured activities so that youth learn how to engage with their own emotional state and physiological responses they previously avoided through substance use, and use the sober community to model positive behaviors of peers and mentors (Finch & Frieden, 2014). The primary elements through which this underlying theory of change is operationalized are: build a richer base of peer and family connection, social support, and accountability; minimize contact with negative peers to increase school engagement and reduce the risk of returning to use; provide students the opportunity to meet peers in recovery and to incorporate skills learned in treatment; quickly respond to problematic behaviors or symptoms of a co‐occurring disorder due to the small school environment and specialized school staff; promote contact with positive peers and adults outside school by requiring participation in support groups after school; and support graduation from high school by providing an accredited curriculum taught by licensed teachers (Tanner‐Smith et al., 2014).

Similar to RHSs, CRCs also support students' academic goals through fostering a positive, sober peer environment that provides companionship as well as emotional and recovery social support (Harris et al., 2014). Youth in recovery who choose to enroll in postsecondary education have additional tasks that can challenge their recovery. For example, unlike youth attending high school, youth in college are tasked with finding their own network of social support and developing healthy peer relationships, moving away from home and learning to live outside of their caregiver's supervision, and setting and attaining educational and career goals (Harris et al., 2008). For youth in recovery, these tasks must be undertaken with the added pressure of finding a peer environment that supports abstinence or sobriety, a difficult task in an abstinence‐hostile environment. Thus, through their available programming and dedicated on‐campus recovery support, CRCs seek to create an internal community of recovery support, advocate for sobriety among the larger community, and support students as they transition to young adulthood. CRCs are vital environmental supports for youth in recovery wishing to attend postsecondary education: indeed, in a recent survey of students in 29 CRCs across the United States, one‐third of respondents stated they would not have chosen to attend higher education without a campus recovery support system (Laudet et al., 2016).

Although RHSs and CRCs operate along a similar theory of change in relation to sober peer support and community, they have different operating models and structures based on how they seek to intervene with youth and due to their placement in particular educational systems. For example, RHSs operate as both the formal school and recovery support system and there may be less of a clearer distinction between the classroom as an academic versus recovery space. As part of the educational system, RHSs are also able to award their students secondary diplomas when they graduate. Alternatively, CRCs are operated on a college campus but are primarily separate from academic classroom spaces, offering additional supports outside of the academic structure. Youth in CRCs may choose to discontinue their participation in the CRC activities but still remain enrolled at the college, while youth at RHSs need to remain sober for continued enrollment at the school. When RHS youth return to use substances, these situations are typically treated on a case‐by‐case basis, but if the youth continues to use substances without indicating motivation to change, they are eventually referred to another school setting to preserve the sober environment of the RHS. Given the close proximity of students and teachers and small school size of RHSs, there is also closer monitoring of youth behavior at the RHS, compared to CRCs, which are housed on larger campuses.

Multiple, interacting factors may affect whether participants enroll in or join a recovery school and once participating, will determine how the recovery school setting affects subsequent substance use and other behavioral outcomes. Most often studied are individual‐level factors such as demographic and socioeconomic characteristics.

Sex/gender and age have distinguished different pathways of youth recovery involvement and outcomes (Stevens, Estrada, Murphy, McKnight, & Tims, 2004; Wellman, Contreras, Dugas, O'Loughlin, & O'Loughlin, 2014). For example, male college students have higher annual prevalence rates of use of marijuana and most other drugs than female college students (Dennhardt & Murphy, 2013). There are also differences between males and females on mental health comorbidity and level of juvenile justice involvement (Stevens et al., 2004), which may interact with level of substance use and/or level of care received for that use (Oser, Karakos, & Hennessy, 2016). Mental health, specifically diagnoses of mental illness comorbidities, has also been studied in this population and found important to outcomes (Hersh, Curry, & Yaminer, 2014; Yu, Buka, Fitzmaurice, & McCormick, 2006). For example, one review of comorbid depression and adolescent treatment outcomes indicated mixed results depending on study characteristics and concluded that there is not a simple way to categorize the relationship between depression and substance use outcomes (Hersh et al., 2014).

Finally, evidence suggests that tangible resources affect recovery, possibly through either restricting or enhancing access to supportive social networks and other established recovery supports (Granfield & Cloud, 1999; Pesetski, 2015; Terrion, 2012); thus, measures of socioeconomic status are important indicators of potential inequity among this population.

This review included studies that used an experimental or quasi‐experimental design. For the primary analysis, eligible studies were those that compared outcomes for students enrolled in recovery school programs with students enrolled in one or more comparison conditions that did not involve a recovery school program. To be considered eligible, study designs were required to meet one of the following criteria:

i. Randomized controlled trial (RCT): Participants were randomly assigned to intervention and comparison conditions. Individual and cluster level randomization was acceptable. ii. Quasi‐randomized controlled trial (QRCT): Participants were assigned to intervention and comparison conditions via a quasi‐random procedure, such as birth date or student record number. iii. Quasi‐experimental controlled trial (QED) with individual level matching: Participants in the intervention and comparison conditions were allocated to conditions via a non‐random process, but participants were individually matched on at least one measure of substance use and on three student demographic characteristics (age, race/ethnicity, gender). iv. Quasi‐experimental controlled trial (QED) with pretest‐adjusted outcomes: Participants in the intervention and comparison condition were formed via a non‐random process, but the study authors adjusted for pretest differences between groups (e.g., as pretest‐adjusted posttest means, regression coefficients from models that adjusted for pretest). For those outcomes on which pretest data were not applicable (e.g., high school graduation), adjustment must have been done for a close proxy of a pretest. v. Quasi‐experimental controlled trial (QED) with pretest data: Participants in the intervention and comparison condition were formed via a non‐random process, but pretest data were available for each outcome. Pretest data must have been reported in a form that permitted assessment of the initial equivalence of the intervention and control groups on those variables via calculation of an effect size. For those outcomes on which pretest data were not applicable (e.g., high school graduation), data for a close proxy of a pretest must have been available.

Eligible student populations included students enrolled part‐time or full‐time in secondary (high school) or postsecondary (college or university) educational institutions. All ages of students were eligible for inclusion provided they were enrolled in an educational institution at the time of the study, but most secondary and postsecondary students were expected to be ages 14–25. Studies that included students who were not enrolled in educational institutions at the time of the intervention were not eligible for inclusion.

Eligible student populations were those comprised of students in recovery from substance use. The definition for "in recovery" is broad, and encompasses any student with a history of using substances who is motivated or interested in reducing their substance use or maintaining abstinence from substance use. Students who were mandated to a program because of behaviours related to substance use or actual problematic use were also considered eligible populations. Due to inconsistent reporting in primary studies, students in recovery were not required to have a formal SUD diagnosis (e.g., substance abuse or substance dependence diagnosis based on DSM criteria).

To be as inclusive as possible, no other eligibility restrictions were placed on the eligible participant populations. Students in recovery who were enrolled in educational institutions in any country were eligible for inclusion.

The review included any recovery school program that was designed to support the academic success of students in recovery from substance use. Recovery schools are broadly defined as educational institutions, or programs at educational institutions, that are developed specifically for students in recovery and address recovery needs in addition to academic development. Eligible recovery school programs could be at the secondary (Recovery High Schools [RHS]) or postsecondary level (College Recovery Communities [CRC]) and should meet the standards detailed by the Association of Recovery Schools (2013) or by the Association of Recovery in Higher Education (2015) as outlined below. Both sets of guidelines provided by the two organizations are broad enough to include schools and programs of varying designs, yet stringent enough to exclude treatment‐based programs to ensure the review's focus on academic recovery supports. Although RHSs were required to meet the broad standards detailed by the ARS to be considered eligible for the review, these programs did not need to be officially accredited by ARS to be included in the review (because the accreditation process is fairly new, may be costly for some locations, has not been widely implemented at this time, and is currently only available in the United States) (ARS, 2016). Given the different models of RHSs and CRCs, eligibility criteria differed by program type.

Recovery High Schools (RHSs) were required to meet the inclusion criteria described below to be eligible for inclusion in the review. Eligible RHS programs must have:

i. had as their primary mission to provide education to youth in recovery from substance use or co‐occurring disorders; ii. had an explicit goal of providing academic or educational instruction to high school students; iii. had an explicit goal of providing a high school environment oriented around recovery from substance use or co‐occurring disorders and require previous treatment and/or commitment to sobriety; iv. met state requirements for awarding a secondary diploma; v. had direct contact with one or more students, and provided educational instruction via face‐to‐face, online, telephone, or video communication; vi. been available to any student in recovery from substance use or co‐occurring disorders who met state or district eligibility requirements for attendance.

Collegiate Recovery Communities (CRCs) must have met the inclusion criteria described below to be eligible for inclusion in the review. Eligible CRCs must have:

i. had an explicit goal of providing recovery support services and focused on encouraging abstinence from use of substances; ii. had an explicit goal of providing a collegiate environment oriented around recovery from substance use, which could be met through any of the following or a combination of the following activities for youth in recovery: (a) had a dedicated space for students in recovery to meet and support each other, (b) offered group mutual aid meetings, and/or (c) offered individual counseling services with trained, specialized staff; iii. been located in a postsecondary educational setting, had direct contact with one or more students, and provided supports via face‐to‐face, online, telephone, or video communication.

Comparison Condition. Eligible comparison conditions were those that included traditional educational programs or services that did not explicitly have a substance use recovery focus. A school specifically designed for students with mental health diagnoses who did not have a substance use disorder was also considered to be an eligible comparison condition.

There were no eligibility criteria based on duration of follow‐up. All follow‐up periods of eligible studies were considered.

Eligible studies could be conducted in any country. Eligible settings for intervention delivery were educational settings for youth in recovery and are described in detail in the section discussing intervention eligibility criteria (above). If the location of the educational program was in a substance use treatment center, then the program was not eligible for inclusion in this review because educational programs delivered in formal substance use treatment settings prioritize treatment services over academics and may involve more intensive treatment services. Thus, programs in these settings did not qualify as a posttreatment or continuing care recovery support environment.

Eligible studies could have been published in any language and reported in any form or type of publication, including but not limited to journal articles, books, book chapters, theses and dissertations, technical reports, conference papers, and other unpublished but disseminated formats. Studies were required to have been reported in 1978 or later, given that the first collegiate recovery program was developed in 1977 (White & Finch, 2006) and the first recovery high school was developed in 1979 (A. Finch, personal communication, June, 15, 2015; Ruben, 2000).

As addiction and recovery supports such as recovery schools span multiple disciplines, education, social science, and public health electronic databases were searched. Search terms varied by database, but generally included three blocks of terms and appropriate Boolean or proximity operators, if allowed: blocks included terms that addressed (1) intervention; (2) population; (3) outcomes. We searched the following 11 electronic databases (hosts): Cochrane Central Register of Controlled Trials (CENTRAL), Cochrane Database of Abstracts of Reviews of Effects (DARE), Education Resources Information Center (ERIC, via ProQuest), Education Database (via ProQuest), International Bibliography of the Social Sciences (IBSS, via ProQuest), PsycINFO (via ProQuest), PsycARTICLES (via ProQuest), PubMed, Social Services Abstracts (via ProQuest), Sociological Abstracts (via ProQuest), and Web of Science. See Table 1 for the full search strings used in the ProQuest, PubMed, and Web of Science hosts. All electronic searches were originally completed in November of 2016 and an update was completed July 29, 2018.

As suggested in the Campbell Systematic Review literature search guide (Kugley et al., 2016), we originally planned to conduct journal hand searches in the most current issues of journals where a large number of potentially eligible studies were found. However, given that only one unpublished manuscript was ultimately eligible for inclusion in the review, we instead conducted hand searches of journals in which at least five retrieved references were located. Thus, we conducted hand searches for the following 17 journals: Addiction, Addictive Behaviors, Alcoholism Treatment Quarterly, Archives of Disease in Childhood, Child Welfare, Drug and Alcohol Dependence, Health Psychology, Journal of Cancer Education, Journal of Child & Adolescent Substance Abuse, Journal of Drug Education, Journal of Groups in Addiction & Recovery, Journal of Psychoactive Drugs, Journal of School Health, Journal of Social Work Practice in the Addictions, Journal of Substance Abuse Treatment, Peabody Journal of Education, and The American Journal on Drug & Alcohol Abuse. All hand searches were originally completed in February of 2017 and were updated on September 12, 2018.

The following 11 websites were also searched for grey literature sources: Alcohol Studies Database (http://www2.scc.rutgers.edu/alcohol_studies/alcohol/↗), Association of Recovery in Higher Education (http://collegiaterecovery.org↗), Association of Recovery Schools (https://recoveryschools.org↗), Drug & Alcohol Findings Project (http://findings.org.uk/index.php?s=eb↗), International Clinical Trials Registry (http://apps.who.int/trialsearch/↗), National Institute on Alcohol Abuse and Alcoholism (https://www.niaaa.nih.gov↗), NIH RePORTER (https://projectreporter.nih.gov/reporter.cfm↗), ProQuest Dissertations and Theses Global (https://search.proquest.com/pqdtglobal/↗), SafetyLit (http://www.safetylit.org/index.htm↗), SAMHSA (http://www.samhsa.gov↗), and Theses Canada (http://www.bac‐lac.gc.ca/eng/services/theses/Pages/theses‐canada.aspx↗). All website searches were originally completed in October of 2016 and were updated on July 29, 2018.

Leading authors and experts in the field of youth addiction and recovery were also contacted for additional studies via email in February of 2017. The bibliographies of relevant reviews and included studies were also searched to identify additional references for review. Finally, in February of 2017 and then again in September 2018 we conducted forward citation searching using the website Google Scholar. We chose Google Scholar for forward searching since this database produces similar results to other search engines such as Web of Science (Tanner‐Smith & Polanin, 2015).

Two reviewers independently screened studies' titles and abstracts to assess for potential eligibility; any disagreements were resolved via discussion and consensus. Potentially eligible studies were then retrieved in full text and these full texts were reviewed for eligibility, again using two independent reviewers. Disagreements between reviewers were resolved via discussion and consensus. If eligibility could not be determined due to missing information in a report, we contacted study authors for this information. We used FileMaker Pro for the entire literature screening and data extraction process.

Two graduate‐level student reviewers were trained on the coding manual by the review's primary author (EAH). All studies were independently double‐coded by these two reviewers, using a predetermined codebook (see Hennessy et al., 2017 for the codebook). Coding disagreements between the two reviewers were resolved via discussion and consensus. Because the sole study eligible for the review was conducted by some of the authors of this review (AJF, EAH, ETS), this study was coded by the two graduate student reviewers who were not connected to the study. This process, including discrepancy resolution, was supervised by a senior reviewer not connected to the study (NS).

The primary categories for coding were as follows: participant demographics (e.g., age, DSM diagnoses for mental health and substance use, gender, grade level, race/ethnicity, socioeconomic status); intervention setting (e.g., country, academic level of focus, urbanicity); intervention components and curriculum delivery (e.g., community service, parent involvement, online coursework, in‐person lessons); study characteristics (e.g., attrition, duration of follow‐up, study design, participant dose, sample N); outcome construct (e.g., type, description of measure); outcome results (e.g., timing at measurement, baseline and follow‐up mean and standard deviation). Additionally, external peer reviewers requested that additional data be extracted from the reports connected to the single eligible study. These additional data included study aims, manuscript focus, timing of report, sample size at the time of the report, type of data collected/presented, and outcomes reported and were collected from those ancillary reports by EAH and checked by NS.

The risk of bias in included studies was assessed using the Cochrane risk of bias tool for

non‐randomized studies (Sterne et al., 2016), the ROBINS‐I. This tool requires that at the protocol stage of the review, two sets of items are determined: (1) confounding factors and (2) co‐interventions that could be different between groups with the potential to differentially impact outcomes. For this review, we specified the following confounding factors at the protocol stage: prior substance use, prior academic achievement, mental health comorbidities, and readiness to change. We also specified the following co‐interventions as potentially different between groups: self‐help attendance (e.g., 12‐Step programs) and other substance use counseling services (e.g., outpatient treatment).

We used the ROBINS‐I tool to assess risk of bias for all seven risk of bias subdomains and an assessment of the overall risk of bias for the study. Because the sole study eligible for the review was conducted by some of the authors of this review (AJF, EAH, ETS), two graduate‐level reviewers extracted data using the ROBINS‐I tool, and a senior reviewer (NS) resolved any discrepancies.

If we had identified any eligible randomized studies, we would have used the Cochrane risk of bias tool for randomized studies (Higgins & Green, 2011); however, no eligible randomized studies were identified in our search.

We anticipated that outcomes in included studies would likely be reported as continuous measures using a variety of scales (e.g., numeric grades or test scores, frequency or quantity of use); thus, the standardized mean difference (SMD; d) was the primary effect size metric used to quantify study findings. All standardized mean difference effect sizes were corrected with the small sample adjustment (Hedges' g). Effect sizes were coded such that positive values (> 0) indicated a beneficial effects of recovery schools (i.e., higher academic achievement, less substance use).

For binary outcomes (e.g., abstinence from drugs), we used the odds ratio effect size metric to quantify study findings. We assumed that binary outcomes reflected different underlying constructs than continuous measures (i.e., abstinence from substance use represents a different underlying construct than frequency or quantity of substance use); therefore, all analyses presented the Hedges' g and odds ratio effect sizes separately.

To minimize any potential bias in the meta‐analysis results introduced by effect size outliers, we planned to Winsorize any effect sizes that were three or more standard deviations from the mean (Lipsey & Wilson, 2001); however, because only one study met eligibility criteria and was included in the review, we did not Winsorize any effect sizes as planned.

All reports of unique studies were reviewed to ensure that articles reporting on the same study were appropriately linked so that only unique study samples were included in each analysis. In our protocol we had originally planned to deal with within‐study effect size dependency based on how many studies were eligible for the review, because some meta‐analytic methods for handling statistical dependencies (e.g., robust variance estimation) require large numbers of included studies (e.g., see Tanner‐Smith & Tipton, 2014 for a summary). However, there was only one eligible study so no meta‐analyses were conducted.

We planned to contact primary study authors if data needed to calculate an effect size were missing from a report. In the one eligible study included in the review, no additional data were needed or requested from the authors.

We originally planned to assess and report heterogeneity using the x² statistic and its corresponding p value, I², and τ², but because only one eligible study was identified, a meta‐analysis was not possible and the assessment of effect size heterogeneity was not applicable.

We planned to use assessments appropriate for reporting and publication biases, such as the contour enhanced funnel plot (Palmer, Peters, Sutton, & Moreno, 2008), regression test for funnel plot asymmetry (Egger, Smith, Schneider, & Minder, 1997), and the trim and fill method (Duval & Tweedie, 2000). However, because only one eligible study was identified and a meta‐analysis was not performed, we did not conduct any assessments of reporting biases.

Given the expected diversity of populations and settings, we originally planned to conduct meta‐analyses using random‐effects inverse variance weights and to report 95% confidence intervals, with mean effect sizes reported separately by study design (RCTs and non‐RCTs), intervention type (RHSs and CRCs), and outcome domain (academics, substance use, and substance use problems). Because only one eligible study was identified and a meta‐analysis was not performed, we instead narratively present effect size estimates and their respective 95% confidence intervals for each outcome domain.

We planned to explore the following moderators using mixed‐effect meta‐regression models: (1) proportion of males; (2) mean age of sample; (3) proportion of students with mental health comorbidities; (4) proportion of students with juvenile justice involvement; (5) student socioeconomic status; (6) and student race/ethnicity. These effect size moderators have been established in prior research as potentially important factors in the role of youth recovery from substance use and were addressed in the background section of this review. Again, however, given that only one eligible study was identified and a meta‐analysis was not possible, we were unable to conduct the planned meta‐regression models. We instead narratively describe these characteristics for the one included study.

We originally planned sensitivity analyses to assess the robustness of review findings which included: (1) level of within‐study attrition; (2) whether or not a standardized assessment tool was used to measure outcomes; and (3) whether removal of Winsorized effect sizes to handle potential outliers substantively altered the review findings. Again, however, we were unable to conduct these sensitivity analyses given that only one study was eligible for inclusion in the review.

Of the 147 reports screened for eligibility at the full‐text level, 138 did not meet the inclusion criteria: 7 were review articles, 23 did not include an eligible RHS or CRC intervention, 3 were conducted outside of school settings, 6 did not include eligible populations of students with substance use problems, and 99 did not use eligible research designs to evaluate the effects of a recovery school program (see Table 4). Indeed, most identified reports that discussed RHSs or CRCs did not use quantitative comparison group designs to examine impacts; most articles simply described recovery schools in operation but did not include research or evaluation components. Among the few ineligible reports that did conduct research on recovery schools, all included pre‐experimental or observational research designs that lacked a comparison group.

We used the ROBINS‐I tool (Sterne et al., 2016) to assess risk of bias in the one included study (Finch et al., 2018), which used a non‐randomized QED to study the effects of recovery high schools. As specified in the protocol, risk of bias was assessed for four potential confounding factors (prior substance use, prior academic achievement, mental health comorbidities, and readiness to change) and two co‐interventions (self‐help attendance, other substance use counseling services). Risk of bias was assessed separately for the academic outcomes and substance use outcomes; however, given the similarity in measurements and methods across these two outcome domains, the risk of bias ratings were ultimately the same for both of these outcome domains (see Table 5).

The one included study was deemed to have serious risk of bias due to confounding. There was the potential for confounding of the effect of intervention in the study (given that students self‐selected into conditions). Although the authors used an appropriate analysis method to control for confounding factors in the domains of prior substance use, prior academic achievement, and mental health comorbidities, the authors did not control for any measures of students' readiness to change. Per the review protocol, risk of bias due to confounding could occur if the authors did not control for confounders in all four domains.

The one included study was deemed to have low risk of bias in the selection of participants into the study. There was no indication that any post‐intervention characteristics influenced selection into the study or analysis, and no indication that follow‐up or start of intervention times varied for most participants.

The one included study was deemed to have moderate risk of bias in the classification of interventions. The intervention status of participants was well‐defined, but was determined retrospectively at the end of the study period (i.e., intervention students were those who attended an RHS for at least 28 days). However, there was no reason to assume that classification of intervention status could have been affected by the knowledge of the outcome or the risk of the outcome.

The one included study was deemed to have inconclusive risk of bias (no information) due to deviations from intended interventions, assuming the effect of interest is assignment to the intervention. Not enough information was provided to determine whether any deviations from the intended intervention reflected what would be expected in usual educational practice.

The one included study was deemed to have inconclusive risk of bias (no information) due to missing data. Participants who were lost to follow‐up during the study period were excluded from the analysis, and missing data on other variables needed for the analysis were handled using multiple imputation. Attrition rates were not reported separately for the two groups, nor were reasons for missing data.

The one included study was deemed to have moderate risk of bias in the measurement of outcomes. The methods of outcome assessment were comparable across groups, and any error in outcome measurement was unexpected to be related to intervention status. However, the outcome assessors were not blinded so it is possible that measures might have been influenced by knowledge of the intervention condition.

The one included study was deemed to have low risk of bias in the selection of the reported results. There was no registered protocol or analysis plan; however, the outcome measurement and analyses were consistent and clearly defined, and there was no obvious indication of selected analyses or selected subgroup analyses.

Overall, the one included study was deemed to have serious risk of bias. This overall risk of bias disposition resulted from the serious risk of bias assessment in the confounding subdomain (due to a lack of measurement of readiness to change), and no other subdomains with critical risks of bias.

The authors reported findings for three outcomes in the academic performance domain: grade point average, truancy, and school absenteeism. All three academic outcomes were collected via self‐report at the study's six‐month follow‐up period. Based on the raw unadjusted outcome data, students in the RHS condition reported higher grade point averages (= 0.65, 95% CI [0.34, 0.97]), lower truancy (= 0.16, 95% CI [‐0.15, 0.46]), and lower absenteeism (= 0.35, 95% CI [0.05, 0.66]) than participants in the comparison condition. However, most of these effect sizes were attenuated when based on the adjusted outcome data from the regression models that controlled for pretest scores and the propensity score: grade point averages (= 0.26, 95% CI [‐0.04, 0.56]), truancy (= 0.01, 95% CI [‐0.29, 0.31]), and absenteeism (= 0.56, 95% CI [0.25, 0.87]). The authors did not report any subgroup findings for participant race/ethnicity, gender, socioeconomic status, mental health comorbidity status, or juvenile justice involvement.

The authors reported findings for four outcomes in the substance use domain: complete abstinence from alcohol/drugs, frequency of alcohol use, frequency of marijuana use, and frequency of drugs other than alcohol or marijuana. All four substance use outcomes were collected via self‐report at the study's six‐month follow‐up period. Based on the raw unadjusted outcome data, students in the recovery high school condition reported higher rates of abstinence (OR = 4.17, 95% CI [0.22, 78.74]), lower alcohol use (= 0.38, 95% CI [0.08, 0.69]), lower marijuana use (= 0.62, 95% CI [0.31, 0.93]), and lower other drug use (= 0.26, 95% CI [‐0.05, 0.56]) than participants in the comparison condition. Again, however, most of these effect sizes were attenuated when based on the adjusted outcome data from the regression models that controlled for pretest scores and the propensity score: abstinence (OR = 4.36, 95% CI [1.19, 15.98]), alcohol use (= 0.23, 95% CI [‐0.07, 0.53]), marijuana use (= 0.51, 95% CI [0.20, 0.82]), and other drug use (= 0.45, 95% CI [0.14, 0.76]). The authors did not report any subgroup findings for participant race/ethnicity, gender, socioeconomic status, mental health comorbidity status, or juvenile justice involvement.

The eligible RHS study did not report any secondary outcomes measuring students' substance use related problems.

The one study included in the review examined the effects of RHSs on high school students' outcomes. This study was conducted in three states (Minnesota, Texas, and Wisconsin) in the United States, and thus the findings may not be generalizable to students in other U.S. states or other countries (particularly lower‐ or middle‐income settings). Participants in this study were primarily White high school students, so the findings may not be generalizable to students from different racial/ethnic backgrounds or different ages. The study did not report findings for outcomes related to high school completion, college enrollment, cocaine use, or substance‐related problems; thus there is also no evidence regarding RHS effects for these outcomes. Finally, the study did not report subgroup findings for different types of students, so there is no evidence available regarding variability in effects across types of students.

No studies examining the effects of CRCs were identified for inclusion in the review. Thus, there is currently insufficient evidence regarding the effects of CRCs on college students' academic and substance use outcomes.

The one included study examining RHSs (n = 194 students) used a controlled quasi‐experimental design to compare outcomes for students who self‐selected to attend RHSs (or not) after being discharged from substance use treatment. The study was rated as having an overall serious risk of bias due to potential confounding, because participants' readiness to change was not measured or controlled for in the analyses (and readiness to change was identified at the protocol stage as an important confounding factor). Given the lack of evidence from any RCTs, and the serious risk of bias due to potential confounding in the one included QED study, the overall quality of evidence for RHSs is very low.

There is insufficient evidence regarding the effects of CRCs. No studies examining the effects of CRCs were identified for inclusion in the review.