Exploring Heterogeneity in Recovery from Substance Use Disorder Following Mindfulness-Based Relapse Prevention: A Latent Profile Analysis

Abstract

Introduction:

Substance use disorder (SUD) recovery is heterogeneous. Yet, over the last 50 years, substance use treatment providers and researchers have often defined success as sustained abstinence from substance use. An often overlooked but equally valid pathway to recovery for persons with SUD is non-abstinent recovery. However, most of the literature on non-abstinent recovery exists for individuals with alcohol use disorder (AUD) with few studies of non-abstinent recovery for other types of SUD. Literature exploring the mechanisms that lead to non-abstinent recovery is also lacking. As such, the current study aimed to examine recovery profiles for individuals (N =454) recruited in two randomized clinical trials comparing mindfulness-based relapse prevention with cognitive-behavioral relapse prevention and/or treatment as usual.

Methods:

Latent profile analysis empirically derived profiles of recovery following outpatient aftercare SUD treatment. Multinomial logistic regression examined associations between treatment assignment and recovery profile, including potential psychological mediators (e.g., mindfulness) and contextual moderators (e.g., annual household income).

Results:

Analyses supported four recovery profiles: (1) low-functioning frequent substance use; (2) low-functioning infrequent substance use; (3) high-functioning frequent substance use; (4) high-functioning infrequent substance use. There were no significant interaction effects of race or ethnicity by treatment type, or household income by treatment type, in predicting recovery profiles. Trait mindfulness, craving, and psychological flexibility failed to mediate the association between treatment assignment and recovery profile; however, there were statistically significant differences in trait mindfulness with individuals expected to be classified in the low-functioning infrequent substance use profile showing significantly lower levels of trait mindfulness compared to individuals in the two high-functioning profiles.

Conclusions:

Findings suggest that recovery from SUD is heterogeneous, and profiles of recovery based on dimensions of substance use and functioning can be identified across a variety of SUD, including among people with co-occurring SUD. Additionally, trait mindfulness appears to be a differentiating factor across recovery profiles. Further research is needed to explore how psychological and social factors may moderate and influence both abstinent and non-abstinent forms of recovery.

Introduction

Non-Abstinent Recovery

Despite abstinence from substances being the primary target for addiction treatment programs for decades (Mayock & Butler, 2021), a growing body of literature continues to elucidate the multidimensional and heterogenous pathways to recovery that exist for individuals with substance use disorders (SUD; Eddie et al., 2022; Hasin et al., 2017; Subbaraman & Witbrodt, 2014; Witkiewitz et al., 2021). Notably, not all individuals seeking treatment for SUD accept abstinence as a goal (Probst et al., 2015), and many do not consider abstinence necessary for success following treatment (Costello et al., 2020; Kaskutas et al., 2014; Laudet, 2007). Likewise, several empirical studies indicate that abstinence need not always be primary, as moderation and reduction in substance use are associated with improvements in health and wellbeing (Cheong et al., 2020; Fan et al., 2019; Henssler et al., 2021; Witkiewitz, Kranzler, et al., 2021). As such, it is essential for SUD treatment providers to meet those seeking their services where they are at in their recovery journey with whatever goals they deem to be personally meaningful – whether abstinent or otherwise (Körkel, 2021). Doing so requires an embrace of a broader definition of recovery that does not so much concern itself with the quantity or frequency of one’s substance use, but instead holds an individual’s overall quality of life as of the utmost importance (Witkiewitz, Montes, et al., 2020).

Non-abstinent recovery has been validated as a legitimate SUD treatment outcome by numerous empirical studies – particularly within the alcohol use disorder (AUD) literature. Witkiewitz and colleagues have conducted one such series of studies using data from Project MATCH (Witkiewitz et al., 2019; Witkiewitz, Wilson, et al., 2021) and the COMBINE study (Swan et al., 2021; Witkiewitz, Pearson, et al., 2020) to examine recovery profiles for individuals with AUD. In this literature, four profiles of recovery have been consistently identified across a range of time points: (1) low-functioning frequent heavy drinking, (2) low-functioning infrequent heavy drinking, (3) high-functioning occasional heavy drinking, and (4) high-functioning infrequent non-heavy drinking. Across these studies, it has been consistently shown that non-abstinent recovery is viable, evidenced by a subset of individuals who attain high psychosocial functioning despite continued heavy drinking.

Despite this evidence supporting the validity of non-abstinent recovery for individuals with AUD, there is a paucity of empirical literature on the validity of non-abstinent recovery for individuals with other SUD (Xin et al., 2022). There are only a handful of studies examining functional outcomes associated with non-abstinent reductions in substance use, specifically for cannabis (Borodovsky et al., 2022; Sherman et al., 2021) and stimulants (Amin-Esmaeili et al., 2021; Roos et al., 2019; Votaw et al 2024). Results from these studies support the notion that those who achieve non-abstinent reductions in substance use report functional outcomes that are comparable to the functional outcomes of those who abstain completely. However, little literature exists beyond this small collection of studies. Further, although there has been growing research on the underlying mechanisms of behavior change that drive abstinence-based recovery from SUD (e.g., Hagger et al., 2020; Mechanisms of Behavior Change Satellite Committee, 2018), there are no known investigations of specific mechanisms that might be particularly salient for non-abstinent forms of recovery.

Mindfulness-Based Relapse Prevention

Mindfulness-based relapse prevention (MBRP) is an evidence-based treatment for individuals with addictive behaviors (Bowen et al., 2021). MBRP draws upon cognitive-behavioral principles and incorporates these with mindfulness practices aimed at enhancing one’s ability to engage with the present moment in a flexible, accepting, and non-judgmental way (Bowen et al., 2021; Witkiewitz & Marlatt, 2004). To date, there have been nearly a dozen randomized controlled trials comprising over 1,000 participants that have examined the efficacy of MBRP (Grant et al., 2017). Across this literature, MBRP has been supported as an efficacious intervention across various populations and types of SUD, even when compared to other empirically-supported treatments (Bowen et al., 2009, 2014; Davis et al., 2019; Massaro et al., 2022; Somohano & Bowen, 2022; Witkiewitz et al., 2014). Further, evidence suggests that third-wave treatments, including MBRP, might be particularly effective for non-dominant racial and cultural groups (Dela Cruz et al., 2022; Félix-Junior et al., 2023; Greenfield et al., 2018; Killeen et al., 2023; Skrzynski et al., 2023; Witkiewitz et al., 2013). Despite a firm establishment of efficacy, further research on the mechanisms and processes of behavior change underlying MBRP is needed, particularly regarding non-abstinent recovery outcomes.

Racial and Socioeconomic Disparities in Addiction Treatment Outcomes

When looking across the literature on non-abstinent recovery, MBRP, and SUD treatment more broadly, a clear pattern is found – treatment outcomes are not equivalent across racial and socioeconomic groups (Collins, 2016; Pouille et al., 2021; Saloner & Cook, 2013; Sheffer et al., 2012; Subbaraman & Witbrodt, 2014; Swan et al., 2021; Williams, 2016). More specifically, there is evidence that contextual features significantly predict SUD treatment outcomes; for example, race/ethnicity (serving as a proxy for contextual factors such as socioeconomic status, neighborhood-level factors, and access to quality SUD treatments) has been shown to impact access to medications for opioid use disorder and treatment retention and completion (Stahler et al., 2021), while proximity to substance use outlets is associated with return to substance use following SUD treatment (Slutske et al., 2019; Joshi et al., 2022). There is sparse literature that explicitly investigates disparities between racial and socioeconomic groups in non-abstinent recovery. However, recent findings suggest that differences in SUD treatment outcomes between racial groups are not exclusive to abstinent forms of recovery (Witkiewitz et al., 2019; Witkiewitz, Pearson, et al., 2020). In a more recent paper that examined the social determinants of health associated with the four previously mentioned recovery profiles consistently identified in the AUD literature, various socioeconomic factors were found to predict recovery profiles (Swan et al., 2021). Significant differences in recovery profiles were found based on racial identity and low-functioning profiles were differentiated from high-functioning profiles based on significant differences in education, poverty, income, and health insurance. Given that ethnoracial and socioeconomic disparities serve as proxies for systemic inequities, these findings suggest that individuals from less-resourced communities face additional barriers to achieving recovery, non-abstinent or otherwise. Though, given this small body of literature, further research is needed.

Aims and Hypotheses of the Current Study

To address the relative lack of research on non-abstinent recovery for SUD beyond AUD, the current study seeks to explore recovery profiles in a sample of individuals with SUD based on dimensions of psychosocial functioning and frequency of substance use. Additionally, it explores if the type of SUD treatment (i.e., MBRP, cognitive-behavioral relapse prevention, or treatment as usual) predicts non-abstinent forms of recovery. Further, it explores if treatment moderates the association of race and socioeconomic indicators on recovery, and if psychological factors mediate the relationship between treatment and recovery. Based on previous literature that supports the efficacy of MBRP over comparison treatments (Bowen et al., 2014), we hypothesized that assignment to MBRP will be associated with a greater likelihood of membership in high-functioning recovery profiles (Hypothesis 1). Because ethnoracial and socioeconomic disparities have been shown to differentiate AUD recovery profiles (Swan et al., 2021), we hypothesized that individuals belonging to ethnoracial minority groups, and those with lower socioeconomic opportunities, will be most likely to belong to low-functioning recovery profiles due to systemic inequities (Hypothesis 2). We further hypothesized that treatment assignment (specifically MBRP) would moderate this association such that MBRP would buffer the effect of systemic inequities on recovery (Hypothesis 3). Finally, we hypothesized that mechanisms of change relevant to MBRP (i.e., craving, mindfulness, and psychological flexibility) would mediate the association between treatment assignment and recovery type (Hypothesis 4).

Method

Data Sources

The current study utilized secondary data analysis of samples from two randomized controlled trials of MBRP: Bowen et al. (2009) [R21DA010562] and Bowen et al. (2014) [NCT01159535]. Both trials utilized an intent-to-treat design and tested the efficacy of eight weeks of group MBRP compared to treatment as usual (Bowen, 2009; 2014), as well as cognitive-behavioral relapse prevention in the Bowen et al. (2014) trial. These trials were identical in length, frequency, and format in delivering MBRP, treatment as usual, and cognitive-behavioral relapse prevention as aftercare SUD treatments. In both parent trials, the treatment as usual condition comprised weekly group-based 12-step facilitation focused on sustained abstinence from substances as an outpatient aftercare treatment. More information on each of the treatment conditions can be found in the parent trials of the current study. The Bowen et al. (2009) study included follow-up assessments at 2- and 4-months following treatment, and the Bowen et al. (2014) study included follow-up assessments at 3-, 6-, 9-, and 12-months following treatment.

Participants

The parent studies recruited participants in both samples (N = 454; see Table 1 and 2) following inpatient or intensive outpatient SUD treatment. Participants in both studies met the following inclusion criteria: 18 years or older, English fluency, ability to attend treatment sessions, medical clearance, and agreement to randomization to treatment. Exclusion criteria included current psychotic disorder, dementia, an imminent risk of harming self or others, or participation in previous MBRP trials. All participants provided written informed consent, and the University of Washington Institutional Review Board approved procedures. Procedures and intervention specifics can be found in the publications for the parent trials (Bowen et al., 2009, 2014).

Table 1.

Demographic Makeup of All Three MBRP Parent Trials

Trial	N	Sex (% Male)	Race/Ethnicity (% Non-Hispanic White), see Table 2 for detailed breakdown	Income (% <$15k)	Primary SUD
Bowen et al. (2009)	168	64%	55%	78%	Alcohol: 46.3% Cocaine: 26.3% Methamphetamine: 12.00%
Bowen et al. (2014)	286	71%	58%	75%	Alcohol: 49.3% Cocaine: 12.3% Methamphetamine: 12.01%
Total	454	68%	57%	76%	Alcohol: 48.08% Cocaine: 17.44% Methamphetamine: 12.00%

Note. N = Sample Size; Income = percent below $15,000 baseline annual household income; Primary SUD = top three most common primary drugs of choice by percent type.

Table 2.

Descriptive Statistics of Final Analyzed Sample

	% of Overall Sample	Mean (S.E.)	SD	Median
MBRP	43.3	–	–	–
RP	19.3	–	–	–
TAU	37.4	–	–	–
Male	67.8	–	–	–
Female	30.9	–	–	–
Non-Binary	0.2	–	–	–
Hispanic	8.8	–	–	–
Non-Hispanic	90.9	–	–	–
Black	24.7	–	–	–
American Indian/Alaskan Native	7.5	–	–	–
Asian	0.4	–	–	–
Hawaiian/Pacific Islander	0.7	–	–	–
Other/Mixed Race	14.3	–	–	–
a $0 – $4999	60.0	–	–	–
a $5000 – $9999	9.7	–	–	–
a $10000 – $14999	5.1	–	–	–
a $15000 – $19999	4.0	–	–	–
a $20000 – $34999	4.6	–	–	–
a $35000 – $49999	2.4	–	–	–
a $50000+	0.9	–	–	–
Age	–	39.24 (0.51)	10.73	40.00
b Drinking Days	–	1.25 (0.29)	5.56	0.00
b Drug Use Days	–	1.68 (0.37)	7.19	0.00
c SIP	–	2.41 (0.27)	4.58	0.00
c SDS	–	3.61 (0.29)	4.46	2.00
c BAI	–	10.27 (0.77)	12.83	5.00
c BDI	–	11.70 (0.70)	11.56	8.10
c AAQ	–	4.56 (0.37)	6.45	4.00
c FFMQ	–	3.30 (0.11)	1.94	3.13
c PACS	–	1.94 (0.65)	11.41	1.00

Note. Descriptive statistics for the final analyzed sample (N = 454; Bowen et al., [2009] and [2014]). S.E. = Standard Error; SD = Standard Deviation; MBRP = Mindfulness-Based Relapse Prevention; RP = Cognitive-Behavioral Relapse Prevention; TAU = Treatment As Usual; SIP = Short Inventory of Problems; SDS = Severity of Dependence Scale; BDI = Beck Depression Inventory; BAI = Beck Anxiety Inventory; AAQ = Acceptance and Action Questionnaire; FFMQ = Five Facet Mindfulness Questionnaire; PACS = Penn Alcohol Craving Scale.

^aAnnual household income reported at baseline.

^bNumber of substance use days reported over the last 2–3 months at follow-up.

^cReported at follow-up.

Measures

The study assessed psychosocial functioning and substance use variables used to derive recovery profiles at follow-up. However, due to differences in follow-up time points between the trials, we used the 2-month follow-up data from Bowen et al. (2009) and 3-month follow-up data from Bowen et al. (2014), to chronologically align the follow-up measures as closely as possible and to represent the first follow-up assessment after the end of treatment. Both studies assessed mediators at post-treatment (8 weeks following baseline assessment).

Substance Use

For the current study, substance use was measured via the Timeline Follow Back (TLFB). The TLFB is a time-based tool that allows self-reported quantitative estimates of the quantity and frequency of use for various substances over the previous 90-day period. Previous literature has indicated support for the TLFB as a robust indicator of recent engagement in substance use (Hjorthøj et al., 2012).

Psychosocial Functioning

To identify the levels of psychosocial functioning, the following self-report measures were used as indicators:

Short Inventory of Problems (SIP).

The SIP is a self-report assessment tool that assesses the consequences of substance use across five specific domains: physical, inter-personal, intra-personal, impulse control, and social responsibility (Blanchard et al., 2003). Via a 4-point Likert-type scale, individuals endorse the frequency at which they experience a range of consequences. A strong body of literature supports SIP as a robust measure of substance use consequences across various populations and SUD (Kiluk et al., 2013). The SIP indicates strong reliability within the parent trials utilized here (MacDonald’s ω = 0.983).

Severity of Dependence Scale (SDS).

The SDS is a brief five-item self-report measure used to assess psychological dependence on substances (Gossop et al., 1995). Via a 4-point Likert-type scale, individuals indicate the frequency by which they experience dependence. The SDS indicates strong reliability within the parent trials utilized here (MacDonald’s ω = 0.911).

Beck Depression Inventory (BDI).

The BDI is a popular self-report assessment tool used to assess the severity of depressive symptoms (Beck et al., 1961). Via a 4-point Likert-type scale, individuals indicate the frequency by which they experienced varied symptoms associated with depressive disorders. We found the BDI to have strong reliability across the parent trials utilized here (MacDonald’s ω = 0.939).

Beck Anxiety Inventory (BAI).

The BAI is a popular self-report assessment tool used to assess the severity of anxiety disorder symptoms (Beck et al., 1988). Via a 4-point Likert-type scale, individuals indicate the frequency by which they experienced varied symptoms associated with generalized anxiety disorder. We found the BAI to have strong reliability across parent trials (MacDonald’s ω = 0.967).

Mediating Mechanisms

The current study examined the following variables, assessed at post-treatment, as potential mechanisms of behavior change mediating the association between treatment assignment and recovery profile at follow-up:

Mindfulness.

The Five-Facet Mindfulness Questionnaire (FFMQ; Baer et al., 2006) assessed trait mindfulness. The FFMQ is a commonly used self-report measure of participant mindfulness which has been found to be reliable and valid across various populations (Baer et al., 2008; de Bruin et al., 2012; Karatepe & Yavuz, 2019; Sauer et al., 2013). The FFMQ consists of 39 self-report items with a Likert-type scale ranging from 1 (totally disagree) to 5 (totally agree). The FFMQ further assesses five sub-facets of mindfulness: observing, describing, acting with awareness, non-judgment, and non-reactivity. The FFMQ indicates strong reliability within the parent trials (MacDonald’s ω = 0.911).

Acceptance.

The Acceptance and Action Questionnaire (AAQ), a 7-item self-report measure, was used to assess participants’ general psychological flexibility via a 7-point Likert-type scale (Hayes, 1996). The AAQ indicates moderate reliability within the parent trials (MacDonald’s ω = 0.777).

Craving.

The Penn-Alcohol Craving Scale (PACS; Flannery et al., 1999) measured participants’ craving to use substances. The PACS is a 5-item self-report questionnaire that is a robust and widely used measure across the alcohol and substance use literature (Hartwell et al., 2019). For the purposes of the parent trial, the PACS was adapted to include craving for all substances, not exclusively alcohol. The PACS indicates strong reliability within the parent trials (MacDonald’s ω = 0.911).

Statistical Analyses

Latent Profile Analysis

Latent profile analysis (LPA) empirically derived distinct recovery profiles following outpatient aftercare treatment. LPA is an exploratory mixed modeling approach that examined heterogenous phenotypes/subgroups within observed continuous data (Goodman, 1974). It is commonly used to quantitatively define latent subpopulations within a larger population (Berlin et al., 2014). Latent profiles are determined based on the relationship between continuous indicator variables (e.g., variables that operationally define substance use and variables that operationally define functionality – as in the case of the current study). LPA assumes that unobserved subgroups (i.e., latent profiles) within the observed data generate patterns of responses to indicator variables. In LPA, two sets of parameters are estimated: (1) the probability of endorsing specific indicators given membership to a latent profile (i.e., item-response probabilities), and (2) the item-response means given the latent profile categorization (Peugh & Fan, 2013). Item-response means are then used to define the latent profiles/subgroups based on the patterns of item responses. Here we defined latent profiles based on two sets of indicator variables: psychosocial functioning and substance use frequency. As described above, psychosocial functioning variables included the SDS, SIP, BAI, and BDI. The study derived substance use frequency via the number of drinking days and the number of drug use days over the last 90 days as measured via the TLFB.

Model fit was assessed via the Bayesian Information Criteria (BIC) where a lower BIC score indicates a better fitting model and via the likelihood ratio test (LRT) which measures how much more likely the data would be under an alternative model with one fewer profile. A significant LRT indicates a rejection of the model with one fewer profile (Vuong, 1989). Additionally, we evaluated latent profile classification precision using model entropy. A model entropy value of 0.80 or greater was considered good classification (Nylund et al., 2007). The current LPA utilized robust maximum likelihood estimation to include all available data across the two clinical trials under the assumption that any missing data were missing at random. Covariates include gender (male, female, other) and age.

Beyond LPA, the current study utilized multinomial logistic regression to test Hypotheses 1 – 4. For each hypothesis, we ran multiple multinomial logistic regression models whereby we compared the association of each set of variables (e.g., race/ethnicity predicting latent profile) by comparing differing reference profiles (e.g., Profile 1 vs. Profile 2; see Table 5). We tested whether assignment to MBRP, and race/ethnicity, were significantly associated with latent profile classification. Additionally, we examined whether there was a treatment by income and treatment by race/ethnicity interaction on recovery profiles. Due to the low number of racial minorities per racial category (see Table 1), the current study defined race/ethnicity across two levels: (1) White and non-Hispanic; (2) person of color and/or of Hispanic ethnicity. Likewise, due to a minority of participants reporting greater than $5,000 annually (see Table 2), socioeconomic status (operationalized via annual household income) was defined across two levels: (1) individuals reporting an annual household income between $0 – $5000; (2) individuals reporting an annual household income of $5000 or more.

Table 5.

Results of Multinomial Logistic Regression

	Profile 1 vs. Profile 4 (reference = Profile 1)	Profile 2 vs. Profile 4 (reference = Profile 4)	Profile 3 vs. Profile 4 (reference = Profile 3)	Profile 1 vs. Profile 3 (reference = Profile 3)	Profile 2 vs. Profile 3 (reference = Profile 3)	Profile 1 vs. Profile 2 (reference = Profile 1)
	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)
	p	p	p	p	p	p
Hypothesis 1;a(MBRP = 0)
TAU	−0.30 (0.60) p=0.62	0.08 (0.60) p=0.90	−0.47 (0.69) p=0.50	−0.17 (0.86) p=0.85	−0.39 (0.86) p=0.65	−0.22 (0.80) p=0.78
RP	−0.59 (0.48) p=0.22	−0.08 (0.52) p=0.87	−0.31 (0.63) p=0.62	0.28 (0.75) p=0.70	−0.39 (0.77) p=0.61	−0.68 (0.66) p=0.31
Hypothesis 2;b(Non-Hispanic White = 1; Person of Color = 0);c(Baseline Income of $0 – $4999 = 1)
Race/Ethnicity Direct Effect	−0.55 (0.44) p=0.21	0.63 (0.49) p=0.20	0.51 (0.55) p=0.36	1.06 (0.66) p=0.11	1.06 (0.67) p=0.11	0.08 (0.62) p=0.90
Income Direct Effect	0.25 (0.45) p=0.58	0.98 (0.66) p=0.14	0.04 (0.61) p=0.95	−0.21 (0.71) p=0.77	1.02 (0.87) p=0.24	1.23 (0.78) p=0.11
Hypothesis 3;b(Non-Hispanic White = 1; Person of Color = 0);c(Baseline Income of $0 – $4999 = 1)
Race/Ethnicity Moderation	1.02 (0.78) p=0.19	−0.21(0.59) p=0.72	0.76 (0.62) p=0.22	1.02 (0.78) p=0.19	0.55 (0.79) p=0.49	−0.47 (0.77) p=0.54
Income Direct Effect	−0.21 (0.71) p=0.77	0.98 (0.66) p=0.14	0.04 (0.61) p=0.95	−0.21 (0.71) p=0.77	1.02 (0.87) p=0.24	1.23 (0.78) p=0.11

Note. This table presents all results for each multinomial logistic regression model assessed. Age and gender are included as covariates for all models. Hypothesis 1 tests the association of treatment assignment with latent profile. Hypothesis 2 tests the association of race and ethnicity with latent profile, and further tests the association of income with latent profile. Hypothesis 3 tests the moderating effect of treatment type on the association of race and ethnicity on latent recovery profile, and further tests the moderating effect of treatment type on the association of income on latent recovery profile. Profile 1”low-functioning frequent substance use” profile (9.56% of the total sample). Profile 2: “low-functioning infrequent substance use” profile (7.99% of the total sample). Profile 3: “high-functioning frequent substance use” profile (6.79% of the total sample). Profile 4”high-functioning infrequent substance use” profile (75.66% of the total sample). B = unstandardized beta coefficient; S.E. = standard error; p = p-value; MBRP = Mindfulness-Based Relapse Prevention; RP = Cognitive-Behavioral Relapse Prevention; TAU = Treatment as Usual.

^aTreatment assignment was dummy-coded such that MBRP is the reference group.

^bRace and ethnicity were modeled as a binary categorical predictor such that persons of color are compared to non-Hispanic White individuals.

^cIncome was coded as a binary categorical predictor such that individuals who reported a baseline annual income of $0 – $4999 were compared to individuals who reported an annual income of $5000 or greater.

Multinomial logistic regression also examined potential mechanisms of change that promote recovery from SUD following aftercare treatment. To this end, we tested whether there was a significant association between treatment condition (i.e., MBRP versus cognitive-behavioral relapse prevention or treatment as usual) and craving, mindfulness, and psychological flexibility (a-path), and if in turn these variables were associated with latent recovery profile membership (b-path). The products of coefficients approach in a latent class mediation modeling framework, which multiples the a*b paths, was used to test the significance of the mediated (i.e., indirect) effect of these variables (Hsiao et al., 2021; MacKinnon, 2008; Witkiewitz et al., 2018).

The study cleaned and aggregated all data sets in IBM SPSS Statistics version 28.0.0.0 (190; Field, 2013) and conducted all analyses in Mplus version 8.8 (1; Muthén & Muthén, 2017).

Results

Descriptive Statistics

The full sample (N = 454) was 32% female and 57% identified as non-Hispanic White, with a mean age of 39.24 years (SD = 10.73), see Table 2 for full demographics. Notably, the sample comprised individuals from generally low socioeconomic backgrounds with most of the sample reporting an annual household income well below the federal poverty line. At baseline, 60% of the sample reported an annual income between $0 – $4,999 with 3.8% of the sample reporting an annual income of greater than $20,000. While the two parent trials did not specifically target individuals from low socioeconomic backgrounds, the parent trials did indeed recruit individuals who were just exiting community-funded intensive inpatient or intensive outpatient SUD treatment, and as such, this context may have contributed to many individuals not having the opportunity to establish stable employment prior to and at the time of recruitment. Regarding treatment assignment, 196 received MBRP, 170 received treatment as usual, and 88 were randomized to relapse prevention. Of the total sample, 180 participants had no missing data on any included measures at baseline and follow-up. When controlling for age and gender, there were no significant differences in measures of psychosocial functioning (i.e., BAI, BDI, SDS, and SIP) or hypothesized mechanisms (i.e., AAQ, FFMQ, and PACS) based on treatment assignment.

Latent Profile Analysis

Profile Enumeration

To identify recovery profiles, we tested 2- through 6-profile solutions based on indicators of substance use frequency, and psychosocial functioning. Superior model fit as defined by the LRT, BIC, and model entropy, as well as parsimony, and overall interpretability of the profile solution supported a 4-profile solution over the 2-, 3-, 5-, and 6-profile solutions. LRT supported the 4-profile solution (LRT = −3947.89, p = 0.004), over the 5- and 6-profile solutions, and BIC supported the 4-profile solution (BIC = 7937.519) over the 2- and 3-profile solutions. BIC, LRT, and entropy for all tested models are presented in Table 3.

Table 3.

Model Fit Statistics by Number of Profiles

Number of Profiles	BIC	LRT (p-value)	Entropy
2	8169.63	−4261.95 (p < 0.00)	0.825
3	8062.34	−4022.36 (p < 0.00)	0.848
4a	7937.52	−3947.89 (p < 0.01)	0.848
5	7870.10	−3864.67 (p = 0.71)	0.796
6	7790.54	−3806.68 (p = 0.70)	0.806

Note. This table presents all statistics used to assess latent profile solutions for models by which substance use frequency was derived via the number of drug use and drinking days, and by which psychosocial functioning was defined via the total scores of the Beck Anxiety Inventory, Beck Depression Inventory, Severity of Dependence Scale, and the Shorty Inventory of Problems; BIC = Bayesian Information Criteria; LRT = Likelihood Ratio Test.

^aProfile solution selected based on model fit, parsimony, and interpretability of profiles.

Profile Interpretation

Table 4 presents the standardized scores for the psychosocial and substance use indicators across each of the four profiles. The classification precision of the 4-profile solution was acceptable (entropy = 0.848). Individuals with expected classification in Profile 1 (9.56% of the sample), which we named the “low-functioning frequent substance use” profile, reported relatively more frequent days of drug and alcohol use and relatively higher levels of depression and anxiety, substance use consequences, and dependence severity. Individuals with expected classification in Profile 2 (7.99% of the sample), which we named the “low-functioning infrequent substance use” profile, reported less frequent days of drug and alcohol use, and higher levels of depression and anxiety, substance use consequences, and dependence severity. Individuals with expected classification in Profile 3 (6.79% of the sample), which we named the “high-functioning frequent substance use” profile, reported relatively more frequent days of drug and alcohol use and relatively lower levels of depression and anxiety, substance use consequences, and dependence severity, while those with expected classification in Profile 4 (75.66% of the sample), named the “high-functioning infrequent substance use” profile, reported relatively less frequent days of drug and alcohol use and relatively lower levels of depression and anxiety, substance use consequences, and dependence severity. Figure 1 depicts the four latent profiles by indicator type.

Table 4.

Means and Standard Errors for Indicators of the 4-Profile Solution

	Profile 1	Profile 2	Profile 3	Profile 4
	Low Functioning, Frequent Substance Use	Low Functioning, Infrequent Substance Use	High Functioning, Frequent Substance Use	High Functioning, Infrequent Substance Use
	Mean (S.E.)	Mean (S.E.)	Mean (S.E.)	Mean (S.E.)
a Substance Use Frequency
b Drinking Days	2.51 (0.64)	−0.93 (1.11)	2.04 (0.60)	1.35 (0.68)
c Drug Use Days	3.06 (0.33)	2.37 (0.30)	2.76 (0.29)	2.26 (0.44)
Psychosocial Functioning
SIP	21.17 (1.63)	13.61 (1.02)	6.10 (0.52)	0.16 (0.03)
SDS	2.77 (0.31)	2.01 (0.29)	1.34 (0.20)	0.60 (0.05)
BDI	1.81 (0.26)	1.77 (0.25)	1.39 (0.22)	0.88 (0.05)
BAI	1.32 (0.25)	1.19 (0.20)	0.78 (0.21)	0.72 (0.05)

Note. This table presents the standardized means and standard errors for each indicator of the 4-profile solution. S.E. = Standard Error; SIP = Short Inventory of Problems; SDS = Severity of Dependence Scale; BDI = Beck Depression Inventory; BAI = Beck Anxiety Inventory.

^aMeans and standard errors of drinking and drug use days are unstandardized. Drinking and drug use days are modeled as count indicators via a negative binomial hurdle model.

^bUnstandardized threshold mean = 1.95; threshold standard error = 0.16.

^cUnstandardized threshold mean = 2.19; threshold standard error = 0.17.

^dStandardized means and standard errors of all psychosocial functioning variables are reported here.

Figure 1. Plot of Standardized Means of 4-Profile Solution by Indicator.

Note. This figure presents the standardized means of each of the latent profile indicators of the 4-profile solution. Profile 1”low-functioning frequent substance use” profile (9.56% of the total sample). Profile 2: “low-functioning infrequent substance use” profile (7.99% of the total sample). Profile 3: “high-functioning frequent substance use” profile (6.79% of the total sample). Profile 4”high-functioning infrequent substance use” profile (75.66% of the total sample). SIP = Short Inventory of Problems; SDS = Severity of Dependence Scale; BDI = Beck Depression Inventory; BAI = Beck Anxiety Inventory.

Multinomial Logistic Regression

Multinomial logistic regression was used to test the association of each latent profile (Profiles 1 – 4 described above) with treatment assignment. In these models, we dummy-coded treatment via two variables where MBRP was the reference group for both treatment as usual and cognitive-behavioral relapse prevention. Despite our hypothesis that individuals assigned to MBRP would be more likely to belong to the high-functioning profiles (Hypothesis 1), we found no evidence for treatment assignment in predicting profile membership when controlling for age and gender (p > 0.05; see Table 5 for full statistics).

Additionally, based on our hypothesis that persons of color would be more likely to belong to the low-functioning profiles (Hypothesis 2), we coded race and ethnicity as a binary predictor of profile membership in which we compared non-Hispanic white individuals to individuals of color and/or Hispanic ethnicity. We then tested the association of race and ethnicity on recovery profile and found no significant association between ethnoracial identity and recovery profile when controlling for age and gender (p > 0.05; see Table 5). We further examined whether treatment assignment partially moderated the effect of race/ethnicity and profile membership (Hypothesis 3). As shown in Table 5, there was no significant partial moderating effect of treatment assignment on race/ethnicity and profile membership when controlling for age and gender (p > 0.05).

Given our hypothesis that individuals with lower socioeconomic status (operationalized as self-reported baseline annual income) would be more likely to belong to the low-functioning profiles (Hypothesis 2), we coded income as a binary predictor of latent class membership by which individuals who reported a baseline annual income between $0 – $4999 were compared to individuals who reported annual income of $5000 or greater. We then examined income as a covariate predictor of profile membership. Results demonstrated no significant association between income and profile membership when controlling for age and gender (p > 0.05; see Table 5). We then examined if treatment assignment partially moderated the effect of income level and profile membership (Hypothesis 3). As shown in Table 5, there was no significant partial moderating effect of treatment assignment on socioeconomic status and profile membership when controlling for age and gender (p > 0.05).

Finally, we examined the indirect effects of craving, trait mindfulness, and psychological flexibility on the direct effect between treatment assignment and profile membership in three separate models controlling for age and gender. As shown in Table 6, we found no evidence of partial mediation for either craving, mindfulness, or psychological flexibility on the association of treatment assignment and profile membership. Failure to identify mediation disconfirmed Hypothesis 4. However, we did find a significant direct effect of trait mindfulness on latent profile membership for Profile 1 compared to Profiles 3 and 4, such that individuals in the “low-functioning frequent substance use” profile (Profile 1), had significantly lower trait mindfulness than individuals in the “high-functioning frequent substance use” (Profile 3) and “high-functioning infrequent substance use” profiles (Profile 4; p < 0.05; see Table 5 for full statistics).

Table 6.

Mediating Effects of Trait Mindfulness, Psychological Flexibility, and Craving

	Profile 3 vs. Profile 2 (reference = Profile 3)	Profile 1 vs. Profile 2 (reference = Profile 1)	Profile 4 vs. Profile 2 (reference = Profile 4)	Profile 3 vs. Profile 4 (reference = Profile 3)	Profile 1 vs. Profile 4 (reference = Profile 1)	Profile 3 vs. Profile 1 (reference = Profile 3)
	(B) (S.E.)	(B) (S.E.)	(B) (S.E.)	(B) (S.E.)	(B) (S.E.)	(B) (S.E.)
	p	p	p	p	p	p
Direct Effects
AAQ	0.10 (0.55) p=0.85	−0.61 (0.39) p=0.12	−0.63 (0.39) p=0.11	0.73 (0.42) p=0.08	0.02 (0.30) p=0.50	0.71 (0.43) p=0.10
PACS	>−0.01 (0.01) p=0.23	<0.01(<0.01) p=0.28	<0.01 (0.01) p=0.96	>−0.01 (0.01) p=0.52	<0.01 (0.01) p=0.62	<−0.01 (0.01) p=0.79
FFMQ	−0.62 (0.51) p=0.23	0.37 (0.64) p=0.57	−0.60 (0.51) p=0.24	−0.02 (0.06) p=0.72	−0.97 (0.48) p=0.04*	0.99 (0.48) p=0.04*
Indirect Effects
AAQ	<0.01 (<0.01) p=0.84	<0.01 (0.01) p=0.46	<0.01 (0.01) p=0.47	–	–	–
PACS	<0.01 (<0.01) p=0.96	<0.01 (<0.01) p=0.56	<0.01 (<0.01) p=0.82	–	–	–
FFMQ	<0.01 (<0.01) p=0.50	<0.01 (<0.01) p=0.53	<0.01 (<0.01) p=0.50	–	–	–

Note. This table presents all direct and indirect effects of trait mindfulness, craving, and psychological flexibility on treatment and latent profile. Age and gender are included as covariates for all models.. Profile 1”low-functioning frequent substance use” profile (9.56% of the total sample). Profile 2: “low-functioning infrequent substance use” profile (7.99% of the total sample). Profile 3: “high-functioning frequent substance use” profile (6.79% of the total sample). Profile 4”high-functioning infrequent substance use” profile (75.66% of the total sample). B = unstandardized beta coefficient; S.E. = standard error; p = p-value. AAQ = Acceptance and Action Questionnaire; PACS = Penn Alcohol Craving Scale; FFMQ = Five Facet Mindfulness Questionnaire.

^*Indicates a significant p-value at alpha = 0.05 significance level.

Discussion

Despite decades of research perpetuating the myth that abstinence is the only viable treatment outcome for individuals with SUD (Mayock & Butler, 2021), there is growing recognition of a more holistic definition of recovery that supports a plurality of pathways to wellbeing, where improvements in quality of life, meaning, and fulfillment are seen as the key indicators of success following SUD treatment (Hagman et al., 2022; Tucker & Witkiewitz, 2022). To this end, the current study quantitatively examined heterogenous recovery profiles in a sample of individuals with SUD. We sought to replicate profiles of recovery identified in the AUD literature that differed on the bases of substance use frequency and psychosocial functioning (Swan et al., 2021; Witkiewitz et al., 2019; Witkiewitz, Pearson, et al., 2020; Witkiewitz, Wilson, et al., 2021). We further explored the association of treatment type on these recovery profiles, as well as the moderating effects of treatment type on the association of race/ethnicity and socioeconomic status on recovery profiles. Finally, we sought to investigate whether hypothesized mechanisms of behavior change in MBRP, specifically craving, psychological flexibility, and trait mindfulness, mediated the association between treatment type and recovery profiles.

We replicated the four recovery profiles previously identified in the AUD literature, though notably, we did so in an outpatient sample of individuals with various and sometimes co-occurring SUD (see Table 4 and Figure 1). To our knowledge, this is the first time these profiles have been enumerated in a sample that did not primarily consist of individuals with AUD. Of note, when operationalizing high relative psychosocial functioning, rather than abstinence, as the key indicator of recovery, a majority of the sample (82.45%) achieved recovery (Profiles 3 & 4), with a subset of these individuals (6.79%) achieving recovery despite relatively high levels of continued substance use (Profile 3). We also found evidence for a profile of individuals who abstained or used substances infrequently but did not experience comparatively high psychosocial functioning (Profile 2; 7.99%). These findings offer two key takeaways that align with existing literature: (1) recovery is possible whether or not one abstains (Hasin et al., 2017; Tucker & Witkiewitz, 2022), and (2) abstinence does not necessarily result in recovery (Eddie et al., 2022; Swan et al., 2021; Witkiewitz, Pearson, et al., 2020). These conclusions suggest that SUD treatment programs that are based on an abstinence-only approach may not be beneficial for individuals who do not require or desire abstinence (Probst et al., 2015), and for those who do not experience better mental health outcomes even after abstaining, these programs are potentially limited and ineffective (Tucker & Simpson, 2011; Wilson et al., 2016). Taken together, by replicating these four profiles in a sample of individuals with SUD, we add to the growing literature that recognizes non-abstinent recovery as a viable SUD treatment outcome and reiterates that abstinence-only frameworks are not the only answer (Tucker & Witkiewitz, 2022).

Despite evidence that MBRP is more effective than cognitive-behavioral relapse prevention in targeting co-occurring mood disorders (Roos et al., 2017), we did not find any evidence that treatment type was a predictor of recovery profile. Failure to support this hypothesis suggests that these treatments are equivalent regardless of the type of recovery an individual attains (i.e., whether abstinent or non-abstinent). This finding aligns with previous literature which presents recovery as an individual process informed by one’s goals, values, and greater life context (Witkiewitz, Montes, et al., 2020). These findings further suggest that the processes by which recovery is achieved are applicable irrespective of recovery goals and may be found across differing treatments. In other words, it is not the treatment that determines the form of recovery one attains, rather individuals can achieve recovery via a multitude of treatment approaches.

This finding is particularly notable when considered within the broader literature on treatment equivalency (Budd & Hughes, 2009; Mechanisms of Behavior Change Satellite Committee, 2018). It is findings like this that fuel an interest in mechanisms of behavior change more broadly (Kazdin, 2007). Given that evidence-based treatments for SUD have demonstrated relatively equivalent efficacy despite heterogeneity among treatment targets (Magill et al., 2015, 2020), there is a movement away from the application of evidence-based manuals and toward the application of evidence-based process of change (Hayes et al., 2022). The failure of the present study to identify differences across treatments supports the trend towards recognizing shared processes in successful interventions, rather than emphasizing distinct treatment modalities. While it is informative to delineate what facilitates recovery (i.e., processes of change), these findings suggest that it matters more where one is going (i.e., towards a life of greater meaning and fulfillment) than how exactly one gets there (i.e., whether through MBRP, cognitive-behavioral relapse prevention, etc.).

Indeed, the current study did seek to identify mechanisms of behavior change that may facilitate recovery regardless of substance use frequency. However, we failed to find evidence for partial mediation of craving, trait mindfulness, or psychological flexibility in predicting the association between treatment type and recovery profile. This finding, which contradicts our hypotheses, may be explained by the failure of treatment type to predict recovery profile in the first place (as explained above). It is also possible that these mechanisms of change may not be the most salient processes for this population or for these specific treatments (Magill et al., 2020; Witkiewitz et al., 2013; Moniz-Lewis, in prep). Further, we are limited in our ability to examine the extent to which these potential processes of change may be relevant or irrelevant beyond retrospective self-report in research settings. Without appropriate temporally granular methods that are sensitive to interactions among multi-level factors (e.g., individual, community, etc.), such questions remain difficult to assess (Chang et al., 2021; Lamont et al., 2018).

We did find that individuals with expected classification in the low-functioning frequent substance use profile (Profile 1) reported significantly lower levels of trait mindfulness than individuals with expected classification in the high-functioning profiles (Profiles 3 & 4). This finding aligns with existing research that has found mindfulness to be inversely correlated with symptoms of anxiety and depression (two key features of our psychosocial functioning indicator) and to improve psychosocial functioning overall (Bowlin & Baer, 2012; Hofmann et al., 2010). Mindfulness may also be a protective factor against the development of severe consequences resulting from substance use, meaning that individuals with higher levels of mindfulness are less likely to use substances in ways that cause harm (Chiesa & Malinowski, 2011; Vinci et al., 2016, 2020). For example, a recent study of young adults found that certain facets of mindfulness such as acting with awareness and non-judgment, were linked to lower levels of alcohol-related consequences, but not the amount of alcohol consumed (Single & Keough, 2021). Similar evidence for mindfulness as a protective factor has been found in both clinical and community samples who engage in substance use (Brett et al., 2018; Fernandez et al., 2010; Frohe et al., 2020; Wisener & Khoury, 2019). These findings bear important clinical implications, suggesting that SUD treatment providers should incorporate mindfulness training into their treatment regimens as it appears that mindfulness enhances psychosocial functioning regardless of one’s recovery goal (i.e., abstinent or non-abstinent). It is plausible that mindfulness supports those with abstinence goals in skillfully responding to triggers and cravings that would otherwise lead to an undesired return to substance use, while further supporting those with non-abstinent goals to make conscious choices about how to engage in substance use such that the positive consequences of substance use are maximized while the negative consequences of use are minimized.

Our study did not find a correlation between race, ethnicity, or socioeconomic status and recovery profile classifications, contrasting with findings in non-abstinent AUD recovery research (Swan et al., 2021). This discrepancy may stem from our limited representation of socioeconomic status, which might not fully represent factors influencing SUD treatment disparities (Guerrero et al., 2017; Jackson et al., 2022; Reskin, 2012). Socioeconomic status was oversimplified into two income groups due to a skewed sample distribution. Specifically, the majority of the analytic sample (60%) reported an annual household income of less than $5,000 (see Table 2). As such, to have sufficient statistical power to explore the potential predictive and moderating effects of socioeconomic status (operationalized as annual household income), we were required to distill the sample into two groups: those reporting above $5,000 annually, and those reporting below $5,000 annually. Importantly, the high number of individuals reporting an annual income below the federal poverty line is a strength of the current study as it situates the findings of the current study in an underserved population that has historically been underrepresented in research on mindfulness-based treatments more broadly (Waldron et al., 2018). For example, the finding that mindfulness enhances psychosocial functioning regardless of recovery type (i.e., abstinent vs. non-abstinent) is especially notable in a sample comprised of individuals largely below the federal poverty line as many individuals from low socioeconomic backgrounds face unique stressors impacting recovery as well as unique barriers to accessing traditional forms of treatment (Collins, 2016; Lewis et al., 2018; Saloner & Cook, 2013; Sheffer et al., 2012). This finding highlights the potential effectiveness of mindfulness training in supporting recovery among individuals facing significant socioeconomic challenges. However, the low-income level of the current sample is also a limitation in that it reduces the generalizability of these findings to individuals with SUD who come from more affluent backgrounds. As such, future studies should employ larger samples with more robust socioeconomic measures to uncover potential disparities in SUD treatment. Similarly, the binary racial and ethnic categorization, which was also necessary given the sample distribution, may conceal nuanced differences linked to broader socio-cultural factors, such as economic inequality and systemic racism, which are crucial in understanding SUD treatment outcomes (Farahmand et al., 2020; Marsh et al., 2015; Probst et al., 2020) and also associated with non-abstinent AUD recovery (Swan et al., 2021).

Notwithstanding those already mentioned, the present study is not without additional limitations. Recognizing the constraints of our exploratory secondary data analysis, future research should implement longitudinal studies focusing on individuals pursuing non-abstinent recovery from the outset. Such studies would more effectively reveal the effectiveness and mechanisms of this recovery pathway. Additionally, comparing individuals’ substance use and psychosocial functioning before and after treatment across different recovery profiles may provide deeper insights into their evolution over time. The current study is further limited by the measures used to define the indicators of the recovery profiles. We relied on self-report measures to define our profile indicators, which may not have accurately captured the constructs of interest as intended. Additionally, the measures used to define psychosocial functioning here (BAI, BDI, SIP, and SDS) differed from those in previous literature (Swan et al., 2021; Witkiewitz, Kranzler, et al., 2021; Witkiewitz, Pearson, et al., 2020). Nonetheless, we still replicated the same four recovery profiles. This replication suggests that a superordinate construct of psychosocial functioning may adequately be approximated by, or at least correlate with, the measures used in this study and those prior. However, without future research that tests the factor structure of psychosocial functioning via the variables included in this study, and others, this possibility remains unknown.

Conclusion

While the literature on non-abstinent recovery is in its infancy, the lived experiences and stories of resilience from individuals who have achieved recovery despite non-abstinence are not (Tucker & Witkiewitz, 2022; Witkiewitz, Montes, et al., 2020). The present study replicated four profiles of recovery from SUD that have been identified in AUD literature. In doing so, it supported a profile of non-abstinence recovery in which individuals evidenced high psychosocial functioning despite frequent substance use. We further found support for trait mindfulness as an important factor in differentiating high-functioning and low-functioning recovery groups.

Highlights:

• The current study identified four heterogeneous recovery profiles.

• Non-abstinence was validated as a recovery profile.

• Mindfulness was a significant differentiator among the recovery profiles.