Testing a social mechanism of recovery high schools: Peer affiliation as a mediator of substance use outcomes

Abstract

Purpose:

Limited options for developmentally appropriate continuing care represent a critical barrier for youth in recovery. As one of the few options designed specifically for adolescents, recovery high schools (RHSs) warrant investigation to better understand their effects and the mechanisms through which they operate. This study examined the long-term effects of RHS attendance and tested whether peer-level social change operates as a mechanism.

Methods:

Data were drawn from a longitudinal, group-design study of RHS effectiveness. Twelve-month follow-up substance use outcomes were compared between 146 adolescents who enrolled in RHSs after treatment and a propensity score-balanced comparison group of 117 adolescents who enrolled in traditional schools. Time-lagged mediation models tested whether recovery-positive peer affiliation at the 6-month follow-up mediated these effects.

Results:

Relative to the comparison group at the 12-month follow-up, adolescents who attended an RHS were less likely to have alcohol use disorder (OR = 0.31) or a drug use disorder (OR = 0.42), reported fewer cannabis use days (IRR = 0.62), and had higher odds of abstinence (OR = 2.53) after adjusting for baseline levels. Recovery-positive peer affiliation mediated the beneficial effects of RHS attendance on meeting criteria for a drug use disorder, cannabis use frequency, and abstinence.

Conclusions:

These findings provide robust evidence of the long-term beneficial effects of RHS attendance and indicate that some benefits result from improvements in recovery-positive peer affiliation. Improving access to youth-focused recovery supports that facilitate positive social network changes could help reduce relapse risk and sustain treatment effects.

1. Introduction

Over 50% of adolescents use drugs or alcohol within 90 days of an acute residential or outpatient treatment episode (Passetti et al., 2016), underscoring the importance of access to effective continuing care. Although contemporary approaches to treating substance use disorders (SUDs) should ideally be developmentally appropriate and grounded in the unique biopsychosocial contexts in which adolescent substance use occurs, youth-focused recovery supports are limited (Hennessy et al., 2019; Welsh et al., 2025). Recovery high schools (RHSs) are one of the few options designed specifically for adolescents, providing therapeutic and peer-based support in a learning environment grounded in recovery principles (Moberg and Finch, 2008). By facilitating relationships with other youth in an environment steeped in recovery-oriented norms and values, RHSs establish the social conditions necessary to promote and sustain positive behavior changes (Finch and Frieden, 2014; Karakos, 2014). There is a growing body of evidence supporting RHSs (Hennessy et al., 2025), but knowledge gaps remain. This study sought to address these gaps by evaluating the long-term effects of RHS attendance and testing whether a peer-level social change mechanism drives them.

Adolescents face the greatest risk of relapse immediately following treatment, with the sharpest decline in abstinence occurring between discharge and three months post-discharge in inpatient and outpatient samples (Brown and Ramo, 2006; Cornelius et al., 2003). Initial relapse among adolescents primarily occurs within a social context, placing those who return to pre-treatment social networks during this particularly vulnerable post-treatment window at elevated risk (Brown et al., 1989; Ciesla, 2010; Cornelius et al., 2003). Conversely, adolescents who affiliate with recovery-positive peers—peers who are supportive of recovery and model low-risk social norms by discouraging or avoiding substance use—are less likely to relapse (Anderson et al., 2007; Ciesla, 2010; Ciesla et al., 2008; Pivovar, 2014; White, 2006). Thus, depending on the attitudes and behaviors of its members, a social group can serve as either a risk or protective factor during recovery. This pattern extends to broader social environments in which youth are embedded, where exposure to substance use, criminal activity, and violence increases relapse risk (Gangi and Darling, 2012; Garner et al., 2014, 2007). Therefore, connecting youth to recovery-positive peers and alternative social environments is particularly beneficial as they transition out of treatment.

Because recovery is non-linear and often marked by cycles of relapse and treatment re-entry (Dennis and Scott, 2007), it is critical youth in recovery are connected to post-treatment recovery supports to address their long-term needs and help maintain improvements. Continuing care, which includes recovery supports ranging from less intensive clinical services to community-based mutual help groups, aims to do so by increasing recovery capital and sustaining motivation (McKay, 2021). Meta-analytic findings indicate that long-term treatment and recovery supports yield better outcomes than short-term treatment alone (Beaulieu et al., 2021), suggesting that suboptimal outcomes may reflect insufficient duration or lack of support rather than treatment ineffectiveness. A substantial barrier for recovering adolescents is the limited options for youth-focused, developmentally appropriate recovery supports. This gap is particularly important due to differences between recovering adolescents and adults (Finch et al., 2020). For example, social situations are the most common antecedent of relapse among adolescents, whereas negative emotional states more often precipitate relapse in adults (Chung, 2013; Ramo and Brown, 2008). This along with the salience of peer affiliation suggests that adolescents would likely benefit from approaches designed to facilitate recovery-positive social networks and reduce exposure to social-environmental risk factors. Moreover, because recovering youth often struggle relating to experiences of their adult counterparts, they may be more reluctant to attend and derive fewer benefits from mixed-age recovery supports (Kelly et al., 2005; Nash, 2020). Thus, it is important to link adolescents to options tailored to their developmental stage and that connect them to other youth in recovery.

One of the few continuing care options that address these considerations is RHSs, which provide students with an alternative social and academic environment based around recovery and associated norms. Although their structures vary across institutions, all RHSs operate under the principle that recovering adolescents can provide support to one another (Karakos, 2014; Moberg and Finch, 2008). In general, RHSs have smaller enrollment sizes than traditional high schools, cultivating a stronger sense of community among students. This is integral as students are urged to uphold the standards and expectations of RHSs both for their own health and for the benefit of the school community. As such, the environment promotes a unique sense of peer pressure centered around recovery, directly opposing the pressure to use substances typically experienced in traditional high schools (Lloyd, 2009). Despite the genetic underpinnings of peer affiliation and substance use (Kendler et al., 2014), biometrical genetic studies implicate peer groups as a meaningful target for intervention (e.g., Gillespie et al., 2009), supporting the potential value of shaping peer contexts to promote recovery.

There is a nascent but growing body of evidence supporting the beneficial effects of attending an RHS after treatment. Compared to a propensity score-balanced comparison group of adolescents who returned to traditional high schools following treatment, adolescents who attended an RHS for at least 28 days reported fewer days of cannabis use and higher abstinence rates at 6-month and 12-month follow-ups (Finch et al., 2018; Weimer et al., 2019) and fewer days of alcohol use, general substance use, and intoxicated illegal activity at the 12-month follow-up (Tanner-Smith et al., 2020; Weimer et al., 2019). However, the mechanisms underlying these effects have not been tested, representing a key limitation in understanding how RHSs support recovery. Moreover, because prior studies operationalized RHS attendance using a 28-day threshold at 6-month and 12-month follow-ups, it is unknown whether attending an RHS for any length of time during the particularly vulnerable 3-month post-treatment window has beneficial effects (Brown and Ramo, 2006; Tomlinson et al., 2004). Additionally, the effects of RHS attendance on SUD diagnoses have not yet been examined.

The social identity model of recovery conceptualizes recovery as a socially mediated process that unfolds through changes in social group affiliation (Best et al., 2016). As an individual distances themself from a social group whose norms and values are intertwined with substance use and affiliates with a social group defined by recovery, a recovery-based social identity emerges. The individual internalizes the group’s recovery-oriented norms and values, which shape their attitudes and behaviors. The more an individual sees themself as similar to its members, the more salient their recovery-based social identity becomes. However, the success of this transition may depend on the social resources and peer environments available to the individual. Adolescents may find it particularly challenging to cut ties with pre-treatment friends given the emphasis placed on peer acceptance and belongingness during this developmental stage. Without access to recovery-positive peers to whom they can relate, they may be more likely to reaffiliate with their pre-treatment social groups. RHSs facilitate recovery-positive peer affiliation by providing access to such a group of peers and fostering connection through a shared sense of community and belonging, helping fill the social void left from discontinuing relationships with substance using peers (Karakos, 2014; Lloyd, 2009). In this way, recovery-positive peer affiliation is a potential pathway through which RHS attendance influences recovery and therefore warrants examination as an underlying mechanism.

Understanding the effects of RHS attendance and the mechanisms through which they operate can inform the refinement and development of youth-focused continuing care. The present study sought to extend the evidence base of RHSs through the following aims: (1) evaluate long-term substance use recovery outcomes for adolescents who attended RHSs for any length of time within the 3-month post-treatment window; and (2) identify if and to what extent recovery-positive peer affiliation mediates beneficial effects.

2. Methods

2.1. Parent study procedures

Data were drawn from a prospective, quasi-experimental evaluation of RHS effectiveness (Finch et al., 2018). High school-aged youth who recently completed formal SUD treatment (N = 294) were recruited from treatment facilities and RHSs in Minnesota, Texas, and Wisconsin. Randomly assigning participants to schools was neither feasible nor ethical (Tanner-Smith and Lipsey, 2014); therefore, study conditions were formed naturally through participants choosing to enroll in an RHS (intervention group, n = 171) or traditional high school (comparison group, n = 123). Trained data collectors conducted in-person interviews at baseline (study enrollment) and 3-month, 6-month, and 12-month follow-ups from December 2011-February 2017. Participants were not restricted from seeking treatment or continuing care during the study. University of Minnesota’s Institutional Review Board approved primary data collection procedures, and University of Oregon’s Institutional Review Board approved this secondary analysis. As described below, the propensity score-balanced analytic sample for the present study included 263 participants.

2.2. Measures

2.2.1. SUDs

Two M.I.N.I. modules (Sheehan et al., 1998) screened for SUDs based on DSM-IV criteria: (1) the alcohol module captured symptoms of alcohol abuse and dependence; and (2) the other drug module captured symptoms of abuse and dependence for cannabis, cocaine, stimulants, narcotics, hallucinogens, inhalants, tranquilizers, or multiple substances. At the 12-month follow-up, participants reported symptoms from the past six months. To align with DSM-5 criteria (released after data collection began), abuse and dependence were collapsed into a single level to create two binary variables for alcohol use disorder (AUD) and general drug use disorder (DUD; 1 = yes, 0 = no).

2.2.2. Substance use frequency and abstinence

Alcohol, cannabis, and other drug use frequency were assessed using the Timeline Followback method (Sobell and Sobell, 1992). At the 12-month follow-up, participants reported the number of days they used alcohol, cannabis, and other illicit drugs during the past 90 days. Complete abstinence from alcohol, cannabis, and other drugs was also derived from these responses (1 = yes, 0 = no).

2.2.3. RHS attendance

Participants reported the school(s) they attended during the study period at each timepoint. Participants who enrolled in an RHS at any point during the period between baseline and the 3-month follow-up were considered the intervention group, and all other participants were considered the comparison group.

2.2.4. Recovery-positive peer affiliation

Recovery-positive peer affiliation was measured using items from the Personal Experiences Inventory (Winters and Henley, 1989), Peer Substance Use Test (Chassin et al., 1993), and a researcher-developed item (α = .91). At the baseline and 6-month follow-up, participants rated their agreement with 13 items describing their friends’ substance use behaviors and attitudes on a 4-point scale. At the 6-month follow-up, they also rated the extent to which they agreed their current peer group supports their recovery on a 10-point scale. Confirmatory factor analyses were conducted for both timepoints, modeling the items as ordinal indicators and estimating parameters with diagonally weighted least squares. Factor scores were extracted and used as the mediator (6-month follow-up) and a covariate (baseline) in the mediation models. Factor loadings and fit indices are reported in Supplementary Material 1.

2.2.5. Model covariates

Sex, race/ethnicity, age, and baseline measures of the outcomes were included as covariates in the statistical models (independent of the propensity score weights, described below). Due to multicollinearity between baseline and follow-up abstinence, models predicting abstinence controlled for baseline measures of alcohol, cannabis, and other drug use frequency. Race/ethnicity was dichotomized due to limited variability in the sample (1 = White, non-Hispanic, 0 = other race/ethnicity). The mediation models also controlled for baseline recovery-positive peer affiliation in all paths.

2.3. Analytic strategy

2.3.1. Propensity score estimation and weighting

In the absence of random assignment, baseline differences between intervention and comparison groups can threaten internal validity and limit causal inference (Imbens and Rubin, 2015). To address this, propensity score methods were used to adjust for differences in participant characteristics that might otherwise be confounded with RHS attendance. Over 50 baseline characteristics drawn from the following domains were included as covariates in the propensity score model: demographics, psychosocial factors, mental and physical health, treatment history, substance use and related consequences, continuing care interest/involvement, and school attendance/performance.

To address some missing baseline data in the propensity score estimation (ranging from 0.3%−7.1%, see Supplementary Material 2), 20 datasets of the identified covariates were multiply imputed using chained equations. The imputation model was specified in accordance with best practices for addressing missing baseline data in propensity score models (Coffman et al., 2020). The datasets were pooled into a single baseline covariate dataset, and ridge regression was used to estimate the propensity scores with the WeightIt R package (Greifer, 2025). The propensity score was defined as the probability of attending an RHS for any length of time between baseline and the 3-month follow-up. To ensure comparability between the intervention and comparison groups, cases outside the region of common support were excluded (Guo and Fraser, 2014). Inverse probability of treatment weights (IPTW) were calculated with the average treatment effect on the treated estimand and used to balance the groups on the estimated propensity scores (Benedetto et al., 2018). The intervention and comparison groups were considered balanced when the weighted baseline covariates had standardized mean differences below |0.25| and no significant differences (Ho et al., 2007). Supplementary Material 2 compares the baseline covariates with and without the IPTW applied.

2.3.2. Aim 1

Generalized path models were estimated with cluster robust standard errors to correct for correlated errors associated with participant nesting within schools. Missing data were handled using full information maximum likelihood, and patterns of missingness in baseline predictors were examined to confirm the missing at random assumption was tenable. Cases were weighted with the IPTW. Regression paths were drawn from the group indicator (1 = intervention, 0 = comparison) to the 12-month follow-up outcomes. Binary logistic regression models were fitted to estimate the effects of RHS attendance on AUD, a DUD, and abstinence. Negative binomial regression models were fitted to estimate the effects of RHS attendance on substance use frequency counts. Analyses were performed in Mplus (Muthén and Muthén, 2024).

2.3.3. Aim 2

Using the propensity score weighting and generalized path modeling approach described above, time-lagged mediation models were estimated to test whether recovery-positive peer affiliation (6-month follow-up) mediated the effects of RHS attendance. Three paths were simultaneously estimated: a path (effect of RHS attendance on recovery-positive peer affiliation), modeled linearly given the approximate normality of the factor score; b path (effect of recovery-positive peer affiliation on the outcome); and c’ path (direct effect of RHS attendance not operating through the mediator). The indirect effect was the product of the a and b paths (ab), and the total effect was the sum of the direct and indirect effects (c = c’ + ab). Confidence intervals for indirect effects were obtained using bias-corrected bootstrap resampling with 5,000 iterations (MacKinnon et al., 2004).

3. Results

3.1. Propensity score-balanced analytic sample

Thirty-one cases were outside the region of common support, yielding a propensity score-balanced analytic sample of 263 participants aged 13–19 (M = 16.3 years). Between the baseline and 3-month follow-up, 55.5% enrolled in an RHS and 45.5% enrolled in a non-RHS. Approximately 73% were White, non-Hispanic and 57.8% were male. At baseline, 63.9% met criteria for AUD, and 95.1% met criteria for a DUD. At 6-month and 12-month follow-ups, 80.6% and 71.5% of participants were retained, respectively; retention rates were not significantly different between groups. Table 1 displays the unweighted demographics and study variables for each group and total analytic sample.

Table 1.

Analytic Sample Characteristics by Study Group

		Non-RHS (n = 117)		RHS (n = 146)		Total (n = 263)
Characteristics	Range	M / %	SD	M / %	SD	M / %	SD
Demographics
Male	0.0 – 1.0	59.8	%	56.2	%	57.8	%
Female	0.0 – 1.0	40.2	%	43.8	%	42.2	%
Age	13.0 – 19.0	16.1	1.1	16.4	1.1	16.3	1.1
White	0.0 – 1.0	70.1	%	74.7	%	72.6	%
Multiple	0.0 – 1.0	10.3	%	10.3	%	10.3	%
Hispanic	0.0 – 1.0	11.1	%	5.5	%	8.0	%
Other race/ethnicity	0.0 – 1.0	8.5	%	9.6	%	9.1	%
Treatment History
Inpatient	0.0 – 1.0	44.4	%	61.0	%	53.6	%
Intensive outpatient	0.0 – 1.0	60.7	%	54.8	%	57.4	%
Outpatient	0.0 – 1.0	30.8	%	37.7	%	34.6	%
Detoxification	0.0 – 1.0	12.8	%	11.6	%	12.2	%
Baseline
Alcohol use disorder	0.0 – 1.0	60.7	%	66.4	%	63.9	%
Drug use disorder	0.0 – 1.0	94.9	%	95.2	%	95.1	%
Alcohol and drug use disorder	0.0 – 1.0	58.1	%	63.7	%	61.2	%
Cannabis use days	0.0 – 91.0	57.5	33.7	54.9	34.8	56.1	34.3
Alcohol use days	0.0 – 91.0	16.3	23.6	18.8	24.7	17.7	24.2
Other drug use days	0.0 – 91.0	19.0	28.3	29.2	34.4	24.6	32.2
Abstinence	0.0 – 1.0	0.9	%	3.4	%	2.3	%
RPPA	1.0 – 4.0	2.0	0.4	1.9	0.6	1.9	0.5
RPPA factor score	−0.7 – 1.1	0.0	0.3	0.0	0.3	0.0	0.3
6-Month Follow-Up
RPPA	−2.5 – 2.2	−0.4	0.9	0.3	1.0	0.0	1.0
RPPA factor score	−1.5 – 1.3	−0.2	0.5	0.1	0.5	0.0	0.5
12-Month Follow-Up +
Alcohol use disorder	0.0 – 1.0	28.0	%	13.2	%	19.7	%
Drug use disorder	0.0 – 1.0	59.8	%	36.8	%	46.8	%
Alcohol and drug use disorder	0.0 – 1.0	23.2	%	12.3	%	17.0	%
Cannabis use days	0.0 – 90.0	26.4	35.0	12.5	22.1	18.6	29.2
Alcohol use days	0.0 – 87.0	5.8	13.2	2.6	5.4	4.0	9.7
Other drug use days	0.0 – 90.0	5.0	15.0	6.0	14.7	5.6	14.8
Abstinence	0.0 – 1.0	25.6	%	46.2	%	37.2	%

Note. Based on unweighted cases;

⁺Four participants newly met criteria for a drug use disorder and seven for alcohol use disorder at the 12-month follow-up (no differences between groups; p = .320 and p = .701, respectively). RHS = recovery high school; M = mean; SD = standard deviation; RPPA = recovery-positive peer affiliation.

3.2. Aim 1 results

Table 2 presents coefficients, confidence intervals, and standardized effect sizes from models estimating associations between RHS attendance and 12-month follow-up outcomes. Relative to the comparison group at follow-up, RHS participants had lower rates of AUD (13.2% vs. 28.0%) and a DUD (36.8% vs. 59.8%). In adjusted models, RHS attendance was associated with 69% lower odds of AUD (Est. = −1.17, 95% CI = [−2.18, −0.17]) and 57% lower odds of a DUD (Est. = −0.86, 95% CI = [−1.71, −0.01]). RHS participants reported less frequent cannabis use (M = 12.5 vs. 26.4 days), with adjusted models showing a 38% lower incidence rate of cannabis use days (Est. = −0.47, 95% CI = [−0.89, −0.06]). Additionally, RHS participants had higher rates of complete abstinence (25.6% vs. 46.2%), with adjusted models revealing their odds of abstaining were 2.53 times higher than comparison group participants (Est. = 0.93, 95% CI = [0.00+, 1.85]). Alcohol and other drug use frequency did not significantly differ between groups.

Table 2.

Adjusted Associations Between Recovery High School Attendance and Recovery Outcomes at 12-Month Follow-Up

Outcome	Est.	95% CI	IRR	OR	p
Alcohol use disorder	−1.17	[−2.18, −0.17]	—	0.31	.022*
Drug use disorder	−0.86	[−1.71, −0.01]	—	0.43	.049*
Cannabis use days	−0.47	[−0.89, −0.06]	0.62	—	.027*
Alcohol use days	−0.75	[−1.54, 0.05]	0.47	—	.066
Other drug use days	0.15	[−0.67, 0.97]	1.16	—	.723
Abstinence	0.93	[0.00+, 1.85]	—	2.53	.049*

Note. Est. = unstandardized regression coefficient; 95% CI = 95% confidence interval around coefficient; IRR = incidence rate ratio; OR = odds ratio;

^*p < .05.

3.3. Aim 2 results

Table 3 presents the coefficients and bias-corrected bootstrapped confidence intervals for the path estimates, indirect effects, and total effects from time-lagged mediation models estimating the effects of RHS attendance on 12-month follow-up outcomes via 6-month follow-up recovery-positive peer affiliation. Recovery-positive peer affiliation mediated the beneficial effects of RHS attendance on a DUD (ab = −0.26, 95% CI = [−0.70, −0.02], 30% of the total effect), cannabis use days (ab = −0.22, 95% CI = [−0.65, −0.02], 56% of the total effect), and abstinence (ab = −0.52, 95% CI = [−0.13, −1.19]), 54% of total effect). Although RHS attendance reduced the likelihood of AUD (c = −1.19, 95% CI = [−2.41, −0.03]), there was no evidence of mediation by recovery-positive peer affiliation (ab = −0.10, 95% CI = [−0.55, 0.21]). There was also no evidence of indirect effects in models predicting alcohol and other drug use frequency.

Table 3.

Time-Lagged Mediation Models of Recovery High School Attendance (0–3 Months) Effects on Recovery-Positive Peer Affiliation (6 Months) and Recovery Outcomes (12 Months)

	RHS Attendance → RPPA		RPPA → Outcome		RHS Attendance → Outcome
Outcome	a	95% CI	b	95% CI	c’	95% CI
Alcohol use disorder	0.31*	[0.05, 0.57]	−0.32	[−1.30, 0.70]	−1.09	[−2.45, 0.15]
Drug use disorder	0.30*	[0.03, 0.55]	−0.86**	[−1.64, −0.30]	−0.62	[−1.55, 0.40]
Cannabis use days	0.31*	[0.04, 0.55]	−0.71**	[−1.83, −0.16]	−0.17	[−0.71, 0.57]
Alcohol use days	0.34*	[0.07, 0.59]	−0.10	[−0.57, 0.74]	−0.77	[−1.62, 0.24]
Other drug use days	0.31*	[0.05, 0.56]	0.11	[−1.39, 1.32]	0.07	[−1.79, 1.48]
Abstinence	0.34**	[0.08, 0.60]	1.52**	[0.81, 2.46]	0.45	[−0.65, 1.54]

	Indirect Effect		Total Effect
Outcome	ab	95% CI	c	95% CI
Alcohol use disorder	−0.10	[−0.55, 0.21]	−1.19*	[−2.41, −0.03]
Drug use disorder	−0.26*	[−0.70, −0.02]	−0.88	[−1.87, 0.11]
Cannabis use days	−0.22*	[−0.65, −0.02]	−0.39	[−0.92, 0.30]
Alcohol use days	−0.04	[−0.22, 0.30]	−0.80	[−1.63, 0.11]
Other drug use days	0.04	[−0.36, 0.57]	0.11	[−1.32, 1.44]
Abstinence	0.53**	[0.13, 1.19]	0.97	[−0.21, 2.12]

Note. RHS = recovery high school; RPPA = recovery-positive peer affiliation; a = unstandardized a path estimate; 95% CI = 95% bias-corrected bootstrapped confidence interval around estimate; b = unstandardized b path estimate; c’ = unstandardized c’ path (direct) estimate; ab = indirect effect estimate; c = total effect estimate;

^*p <.05,

^**p <.01.

4. Discussion

Drawing upon data from the only longitudinal, between-group evaluation of RHS effectiveness to date, the present study examined the long-term effects and underlying mechanisms of RHS attendance. Relative to the comparison group at the 12-month follow-up, youth who attended an RHS during the vulnerable 3-month post-treatment window were less likely to meet criteria for AUD or a DUD, more likely to be abstinent, and report fewer days of cannabis use. Recovery-positive peer affiliation mediated the beneficial effects of RHS attendance on a DUD, cannabis use frequency, and abstinence. These results contribute robust evidence of long-term beneficial effects of RHS attendance and indicate some benefits are due, in part, to associated increases in recovery-positive peer affiliation.

Overall, these results offer additional evidence that RHSs are an effective youth-focused continuing care option. In light of the limited number of RHSs currently operating in the United States (The Association of Recovery Schools, 2025), these results bolster support for policy initiatives to increase access to RHSs. A notable contribution of this study is the examination of RHS effects on SUDs, which is important because diagnostic criteria account for impairment, physical dependence, and high-risk behaviors not otherwise captured by only assessing frequency of use. Indeed, although RHS attendance was not associated with fewer alcohol use days, it significantly reduced the odds of AUD. This suggests that, even in the absence of a significant reduction in alcohol use frequency relative to the comparison group, RHS attendance still had clinically meaningful effects on alcohol-related functioning. Nonetheless, considering a previous finding that RHS attendance for at least 28 days over a 12-month period was associated with fewer days of alcohol use (Weimer et al., 2019), it is possible a higher dosage of RHS attendance over a longer period is necessary for observing beneficial effects on alcohol use frequency. Future research should therefore continue to evaluate how different RHS attendance patterns may influence different recovery outcomes.

To the best of our knowledge, this is the first study to test the underlying mechanisms of RHSs. The elucidation of recovery-positive peer affiliation as a pathway through which RHSs can yield beneficial effects indicates adolescents’ recovery journeys are not only shaped by receiving treatment and continuing care but also by their social networks. These indirect effects are consistent with findings from other studies that identified social-environmental factors as mediators of structured, clinically-focused continuing care interventions (Garner et al., 2014, 2007; Godley et al., 2019). If recovery is indeed a socially mediated process arising from changes in group membership, these findings highlight the critical role of recovery-positive peer affiliation in driving it. Whereas many recovery supports can only encourage such social network changes (Kelly et al., 2014), our a path estimates consistently showed RHS attendance can indeed facilitate recovery-positive peer affiliation, which in turn produced long-term beneficial effects on a DUD, abstinence, and cannabis use. These results suggest adolescent-focused continuing care should not simply focus on extending services beyond formal treatment but should also create and link youth to supportive environments that are well-positioned to facilitate positive social network change. This could include, for instance, an emphasis on peer-centered activities that can promote positive social connections and a sense of belonging, such as support groups, peer mentoring, collaborative recovery projects, and sober social events.

Given the lack of evidence that recovery-positive peer affiliation mediated the relationship between RHS attendance and AUD, future research should investigate how this beneficial effect operates. It is possible the b path between recovery-positive peer affiliation and AUD is moderated by individual and contextual factors, and as such, the mechanism is only present for specific subgroups of youth or under certain conditions (e.g., students with more socially motivated alcohol use). Thus, future analyses should test whether moderated mediation may be present. Moreover, the specific types of therapeutic services and supports offered by RHSs as well as the extent to which students engage in them may mediate recovery outcomes or link to other intermediary effects that can influence them, such as improved self-confidence and recovery motivation (Hennessy et al., 2019). In this regard, serial mediation models may be useful for testing sequential pathways through which the beneficial effects of RHS attendance operate. School climate (e.g., perceived staff support, norms/values) and peer support constructs (e.g., emotional/informational support) are other potential mediators that warrant further investigation (Karakos, 2014). Lastly, given all the possible mechanisms through which RHSs may influence recovery outcomes, future research would benefit from testing parallel mediation models to examine the unique and combined contributions of multiple mediators.

4.1. Strengths and limitations

This study addressed methodological weaknesses of prior mechanism studies by estimating time-lagged models to establish temporal precedence and formally testing mediation using a robust statistical approach (Meisel et al., 2024). Propensity scores were estimated to balance the groups on a comprehensive set of baseline covariates, mitigating the potential selection bias inherent in quasi-experimental designs. Nonetheless, while the methodological rigor enhances confidence in our findings, limitations remain. First, the analytic sample size is modest, which limits statistical power for detecting direct and total effects. Indeed, as our results showed, it is common for indirect effects to reach statistical significance in absentia of significant total effects because the statistical power to detect indirect effects is considerably greater (Kenny and Judd, 2014). Although a significant total effect is not required for testing mediation (Agler and De Boeck, 2017), the increased likelihood of a Type II error for the parameter remains a limitation. Second, the sample was comprised of U.S. adolescents who received formal substance use treatment, the majority of whom were White and middle-class, limiting generalizability to other populations. Replication studies should be conducted with more diverse samples, including youth who have not received formal treatment. Third, these findings relied exclusively on self-reported data, which can be subject to recall and social desirability biases. Finally, the illicit drug use frequency and general DUD outcomes combined a wide range of drugs. Heterogeneity within those outcomes may limit interpretability due to differences between drug types and may obscure drug-specific effects of RHS attendance. Thus, future studies should integrate data from other sources and use more refined measures of drug use and diagnoses.

5. Conclusion

This novel study evaluated the long-term effects of RHS attendance and tested to what extent they operate through a peer-level social change mechanism. Our findings contribute robust evidence of the beneficial effects of RHS attendance and indicate some benefits are due, in part, to associated improvements in recovery-positive peer affiliation. These results offer additional support for policy initiatives to increase access to RHSs and underscore the need for adolescent-focused recovery supports that can facilitate positive social network changes. Clinical implications include incorporating peer-centered components to foster positive social connections and a sense of belonging among youth. Future research directions include investigating other potential underlying mechanisms of RHSs as well as individual and contextual factors upon which they may be conditioned. Taken together, this study offers insights for refining and developing adolescent-focused continuing care, crucial steps needed to advance the science of recovery support.

• First study to test underlying mechanism of recovery high school (RHS) attendance.

• RHSs improved substance use disorder statuses, cannabis use, and abstinence.

• Recovery-positive peer affiliation mediated some observed beneficial effects of RHSs.

• Highlights value of youth-focused recovery supports that can facilitate positive social network change.