Disparities in Alcoholics Anonymous Participation from 2000 to 2020 Among U.S. Residents With an Alcohol Use Disorder in the National Alcohol Survey

Abstract

Objective:

Mutual-help groups (MHGs) like Alcoholics Anonymous (AA) are effective for resolving alcohol use disorders (AUDs), but few studies have examined disparities in MHG participation, particularly recently. We used five waves of National Alcohol Survey data to investigate whether prevalence of AA attendance among those with a lifetime AUD differed by race/ethnicity, age, and sex, directly testing whether these associations varied with time.

Method:

Analyses pooled weighted data from 2000 to 2020, including only participants with a lifetime AUD and identifying as non-Hispanic White, Latinx/Hispanic, or non-Hispanic Black/African American (N = 8,876). Logistic regression models examined associations between lifetime AA attendance and survey year, race/ethnicity, age, and sex; models also tested for differences in demographic effects across survey year using interaction terms.

Results:

In bivariate models, AA attendance was significantly less prevalent among participants identifying as Latinx/Hispanic (vs. White); ages 18–29 (vs. 30–64); and female (vs. male). Survey year was unrelated to AA attendance, and all interactions involving survey year were nonsignificant. In the final multivariate model (which controlled for severity and other help-seeking), disparities persisted for those identifying as Latinx/Hispanic (vs. White; adjusted odds ratio [aOR] = 0.63) and ages 18–29 (vs. 30–64; aOR = 0.35); AA attendance was also less prevalent among Black/African American (vs. White) participants (aOR = 0.59), but sex became nonsignificant.

Conclusions:

Results replicate and extend sparse findings regarding disparities in MHG attendance and suggest a stagnation in AA's growth and reach to underserved populations. Findings highlight the need to more effectively facilitate MHG attendance (and perhaps broader social network change) among racial/ethnic minorities and emerging adults.

Alcoholics anonymous (AA) and other mutualhelp groups (MHGs) form a ubiquitous component of the U.S. substance use treatment system and a core part of mandated treatment (Hai et al., 2022; Roman et al., 2022; Substance Abuse and Mental Health Services Administration [SAMHSA], 2014). Moreover, an extensive, rigorous evidence base reveals that MHG participation is effective for substance use disorder (SUD) resolution (Kelly, 2017; Kelly et al., 2020; Zemore et al., 2018b). Nevertheless, few studies have investigated potential disparities¹ in the use of MHGs. The current study uses national data to investigate disparities in the use of AA—the most prevalent MHG by far (Kelly & Yeterian, 2011)—from 2000 to 2020. We primarily focus on disparities for racial/ethnic minority (vs. White) and emerging adult² (vs. older) participants because existing evidence tentatively points to disparities in these subgroups, but we also examine sex exploratorily.

Disparities in MHG participation for racial/ethnic minority and emerging adults

Because they are free and widely available in the United States, MHGs are hypothetically accessible to nearly all Americans experiencing alcohol and other drug problems. Further, AA offers a wide diversity of meetings across the United States, including meetings for people of color, in different languages, for young people, and for women (Alcoholics Anonymous, 2023). Nonetheless, prior studies have illuminated specific barriers to MHG attendance for both racial/ethnic minority and emerging adults (Chartier & Caetano, 2011; Hoeppner et al., 2014; Mason & Luckey, 2003; Zemore et al., 2021). These populations are under-represented both in the founding and current membership of 12-step groups (Alcoholics Anonymous, 2014) and other MHG options (Zemore et al., 2017), which may contribute to challenging encounters and differences of opinion. Indeed, racial/ethnic minority participants attending 12-step groups have reported experiencing concern and mistrust of other participants; heightened scrutiny and discrimination; and conflict with 12-step philosophy, values, and practices (Durant, 2005; Harper, 1976; Hoffman, 1994; Jilek-Aall, 1978, 1981; Smith et al., 1993; Venner et al., 2012). Further, young adults ages 18–24 who are receiving SUD residential treatment have reported negative reactions to attendees’ opinions and personalities and age differences generally in connection with 12-step attendance (Labbe et al., 2014), and evidence suggests that lower religiosity among emerging (vs. older) adults may render 12-step attendance difficult (Jones, 2021; Kelly & Moos, 2003; Winzelberg & Humphreys, 1999).

In addition, MHG attendance among racial/ethnic minority and emerging adults may be affected by powerful barriers to seeking SUD services generally. Studies have documented diverse barriers to help-seeking among Latinx/Hispanic and Black/African American populations (e.g., cultural barriers, stigma, logistical barriers; Pinedo et al., 2022; Zemore et al., 2021) and emerging adults (e.g., low readiness for abstinence/change, lack of social support for help-seeking; Davis et al., 2017; Mason & Luckey, 2003; Melnick et al., 2014; Sinha et al., 2003). In addition to affecting MHG attendance directly, these barriers may have an impact on MHG attendance indirectly, via impacting utilization of specialty SUD treatment. Related, national data point to substantially lower rates of SUD treatment utilization among both Latinx/Hispanic (vs. White) individuals (Chartier & Caetano, 2011; Mancini et al., 2015; Mulia et al., 2014; Zemore et al., 2014, 2018a) and emerging (vs. older) adults (SAMHSA, 2022) with SUDs, along with lower SUD treatment retention among both Latinx/Hispanic and Black/African American (vs. White) adults (Arndt et al., 2013; Saloner & Cook, 2013). Specialty treatment constitutes a major referral route for MHGs, especially 12-step groups (Alcoholics Anonymous, 2014); therefore, this may imply lower frequency and intensity of MHG referral.

Although some studies have examined race/ethnicity in relation to MHG participation, evidence is limited and results inconclusive. A comprehensive review of the U.S. literature on racial/ethnic disparities in MHG participation among adults and adolescents with SUD treatment need (Zemore et al., 2021) identified 19 articles, including 8 national epidemiological studies and 11 treatment/community studies. Most articles compared Latinx/Hispanic and/or Black/African American adults to White adults. The evidence tentatively suggested disparities for Latinx/Hispanic populations (especially immigrants, women, and adolescents) and Black/African American women and adolescents. However, data were sparse and quite dated overall, and design limitations were noted. For example, most studies were under-powered for testing racial/ethnic disparities, and studies commonly conducted only bivariate analyses. Even where multivariate analyses were conducted, studies rarely controlled for SUD severity.

Regarding age, a handful of studies suggest age-related disparities in MHG participation among those with SUDs, but current epidemiological studies are lacking. Clinical studies comparing MHG attendance across age groups have generally (but not always) found that younger age is associated with lower rate and frequency of attendance (Bergman et al., 2018). For example, Project MATCH (Matching Alcoholism Treatments to Client Heterogeneity), a large trial comparing alcohol treatments, reported lower rates of past-90-day AA attendance among participants ages 18–29 (52%) versus those 30 and older (64%) at 3-month follow-up across conditions; further, among those attending AA, those under age 30 attended half as often on average as those 30 and older (Hoeppner et al., 2014). However, clinical studies adopting a broader definition of young adulthood, comparing those 18–39 to those 40 and older, have shown similar rates of 12-step attendance across age groups (Satre et al., 2004, 2012). Meanwhile, in an older epidemiological study that combined national data from 1991–1992 and 2001–2002 and examined those with a lifetime AUD, lifetime MHG attendance was significantly greater in all age groups versus those 18–29 (Chartier & Caetano, 2011).

Recent U.S. national data on disparities and the current study

Recently, Hai et al. (2022) examined correlates of and trends in MHG attendance in the 2002–2018 U.S. National Survey on Drug Use and Health (NSDUH), investigating racial/ethnic, age-based, and sex disparities. Multivariate analyses revealed that among those with a past-year SUD, prevalence of past-year MHG attendance was significantly lower among those identifying as Black/African American (vs. White) and reporting low (vs. high) English proficiency, and significantly higher among those ages 26–34 or 35–64 (vs. 18–25). Overall MHG attendance was stable over time. Reported disparities are consistent with the evidence described above. However, Hai et al.’s (2022) study examined only past-year MHG attendance among those with a past-year SUD, so results are relevant only to MHG participation initiated in the context of a very recent or ongoing SUD. Also, because Hai et al.’s (2022) study did not test whether disparities in MHG participation varied over time, results cannot speak to whether disparities have improved, remained stable, or worsened over time.

The current study used population-representative data from five waves of the National Alcohol Survey (NAS) spanning 2000–2020 to investigate whether lifetime AA attendance among those with a lifetime AUD was associated with race/ethnicity, age, and sex, directly testing whether these associations varied with time.³ We hypothesized, based on the above review, that in controlled analyses, lifetime AA attendance would be less prevalent for both Latinx/Hispanic and Black/African (vs. non-Hispanic White) participants and for emerging (vs. established) adults. We also expected no increases in AA attendance over time. Although we explored sex differences, we made no hypotheses for sex because of the mixed evidence base relating sex to AA attendance (Timko, 2008; Witbrodt & Delucchi, 2011). We also made no hypotheses about temporal effects on disparities because of lack of available evidence. Results should inform SUD service implementation and MHG referral practices specifically among underserved groups, which is valuable given the important role of MHGs in initiating and sustaining recovery.

Method

Data and sample characteristics

Data were from five waves (2000, 2005, 2010, 2015, and 2020) of the NAS, a cross-sectional probability survey of adults ages 18 or older in the United States (analytic N = 8,876). The 2000–2015 surveys applied random digit dial (RDD) sampling and computer-assisted telephone interviewing (CATI) to administer interviews (Drabble et al., 2021; Greenfield et al., 2019). The 2020 survey used RDD sampling and CATI in addition to address-based sampling via mail push-to-web for completion of online surveys; it also included online surveys from a nonprobability web panel (ICF, 2020; Kerr et al., 2022; Martinez et al., 2022). Surveys oversampled Latinx/Hispanic and Black/African American respondents, and data were weighted to be demographically representative of the U.S. population.

The analytic sample was limited to those endorsing 2 or more of 11 lifetime AUD criteria specified in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5, American Psychiatric Association, 2013). We included only those identifying as non-Hispanic White, Latinx/Hispanic, or non-Hispanic Black/African American because the study was underpowered for other racial/ethnic subgroups. We excluded 622 cases missing data on lifetime AA attendance because participants were balloted (n = 363), did not complete the full survey (n = 254), or provided no response (n = 14).

Measures

Survey year. Survey year was a categorical variable identifying whether respondents participated in the 2000, 2005, 2010, 2015, or 2020 NAS.

Lifetime AA attendance. All surveys from 2000 to 2020 assessed lifetime AA attendance among those with lifetime AUD. Participants were asked, “Have you ever gone to anyone—a physician, AA, a treatment agency, anyone at all—for a problem related in any way to your drinking?” Those responding affirmatively were asked to specify which services they had used among a list including AA (2000–2010 surveys) or were explicitly asked, “Did you go to a mutual-help group, such as Alcoholics Anonymous?” (2015–2020 surveys). Lifetime AA attendance was coded positive for those reporting any AA attendance.

Focal demographic variables. All surveys assessed ethnicity and race, inquiring, “Are you Hispanic, Latino/a/, or of Spanish origin?” (yes or no), and separately, “What is your race?” Response options for the latter question included Black, White, American Indian or Alaska Native, and other, with the Black option modified to Black or African American in the 2015–2020 surveys; options also included diverse Asian and Pacific Islander groups, with some changes in these across surveys. Responses were combined to create a single variable indicating identification as non-Hispanic White, Latinx/Hispanic (any race), non-Hispanic Black/African American, or other (the latter being excluded from analysis). Surveys also solicited age, categorized to contrast emerging adults (ages 18–29), established adults (ages 30–64), and older adults (ages 65 or older; Arnett et al., 2014).

In the 2000–2015 NASs, participants were asked whether they were male or female. Starting in 2020, participants were asked “What is your sex?”; options included “male,” “female,” and “do not identify as male or female.” Those selecting the latter option were then asked, “What sex were you assigned at birth, on your original birth certificate?” with options being “male” and “female.” Analyses were limited to those indicating male/female sex, excluding just one case who did not respond to the follow-up.

Other demographic characteristics. All surveys assessed current marital status (married/cohabiting, separated/divorced/widowed, or never married), educational attainment (high school or less, some college, or college/4-year degree or more), employment status (not employed or employed full/part-time), and health insurance (any or none).

Substance use treatment history. For the 2000–2015 NASs, lifetime substance use treatment was assessed with a single question: “Have you ever received treatment from a substance abuse treatment program (‘chemical dependency or substance abuse treatment program’ in earlier surveys) for either alcohol or drugs?” (yes or no). For the 2020 NAS, lifetime substance use treatment was coded positive based on a positive response to this first question and/or a positive response to a second question, “Have you ever gone to anyone—a physician, AA, a treatment agency, anyone at all—for a problem related in any way to your drinking?” and affirmation that the participant had attended an alcoholism or drug treatment program in follow-up questions. Multiple questions were used for 2020 because of a skip pattern whereby those indicating any alcohol services use in the second question, or no lifetime marijuana use and no past-year drug use of any kind, were skipped out of the first question.⁴

Lifetime AUD severity. All surveys assessed lifetime AUD criteria following the DSM-5 (American Psychiatric Association, 2013). This variable could be successfully harmonized across years because the 2000–2010 surveys included an item assessing craving.

Analyses

All analyses were conducted in Stata (Version 16; Stata-Corp LP, College Station, TX) using sample weights and variance estimation techniques adjusting for the complex survey design. We first conducted descriptive analyses of the sample and bivariate logistic regressions to examine associations between lifetime AA attendance and survey year, focal demographic variables (i.e., race/ethnicity, age, and sex), and other covariates (i.e., other demographic variables, substance use treatment history, and lifetime AUD severity). For variables with multiple categories, p values for specific category odds ratios (ORs) were interpreted only when the overall Wald test was significant.

We then ran multivariate models including survey year, focal demographic variables, and interactions between survey year and these demographics to test whether any demographic differences varied across survey year. Models were run with and without all covariates to examine whether interaction effects changed with covariate adjustment. To further explore interactions between focal demographics and time, we plotted adjusted prevalence rates of any lifetime AA attendance by race/ethnicity, age, and sex groups across survey years from fully adjusted models. Nonsignificant interaction terms were dropped from the final multivariate model to facilitate interpretation of the main effects.

Results

Bivariate analyses

Table 1 displays sample characteristics and findings from bivariate models testing associations of all study variables with lifetime AA attendance. Considering race/ethnicity, age, and sex in turn, lifetime AA attendance was least prevalent among those identifying as Latinx/Hispanic (7.6%), followed by Black/African American (9.9%) and White (10.8%); least prevalent among those 18–29 (4.8%), followed by 65 plus (11.7%) and 30–64 (12.2%); and less prevalent among women (8.1%) than men (11.7%). Similarly, in unadjusted analyses, odds of lifetime AA attendance were significantly lower among those identifying as Latinx/Hispanic (vs. non-Hispanic White; OR = 0.68), ages 18–29 (vs. 30–64; OR = 0.36), and female (vs. male; OR = 0.66). Odds of AA attendance were also lower among those who were married/cohabiting and never married (vs. separated/divorced/widowed; ORs = 0.59 and 0.55, respectively), reported a college degree (vs. high school or less; OR = 0.53), reported part-/full-time employment (vs. no employment; OR = 0.60), and had any health insurance (vs. none; OR = 0.70). AA attendance was more prevalent among those reporting any substance use treatment (vs. none; OR = 24.63) and at higher AUD criteria counts (OR = 1.79) but was not associated with survey year.

Table 1.

Sample characteristics by lifetime AA attendance among those with a lifetime AUD (unweighted N = 8,876)

Variable	Total sample		Any (vs. no) lifetime AA attendance		Unadjusted ORs for AA attendance (weighted)
	Unweighted n	Weighted %	Unweighted n	Weighted %	OR	[95% CI]	p
Race/ethnicity
White (ref.)	5,662	77.0	657	10.8	–	–
Latinx/Hispanic	1,717	13.2	144	7.6	0.68	[0.52, 0.88]	***
Black/African American	1,497	9.8	174	9.9	0.90	[0.70, 1.64]
Age, in years
18–29	1,901	24.7	87	4.8	0.36	[0.27, 0.49]	***
30–64 (ref.)	5,977	67.5	759	12.2	–	–
≥65	939	7.9	125	11.7	0.95	[0.73, 1.25]
Gender
Male (ref.)	5,086	60.8	635	11.7	–	–
Female	3,790	39.2	340	8.1	0.66	[0.55, 0.79]	***
Marital status
Married/cohabiting	4,856	61.6	476	9.6	0.59	[0.47, 0.73]	***
Separated/divorced/widowed (ref.)	1,640	15.0	269	15.3	–	–
Never married	2,178	23.4	210	9.1	0.55	[0.42, 0.73]	***
Educational attainment
High school grad or less (ref.)	3,060	36.9	387	11.9	–	–
Some college	2,602	32.1	340	11.9	1.00	[0.81, 1.22]
College grad or more	3,192	31.0	246	6.7	0.53	[0.42, 0.66]	***
Employment status
Not employed (ref.)	2,959	31.0	444	13.7	–	–
Employed full-/part-time	5,863	69.0	521	8.7	0.60	[0.50, 0.71]	***
Insurance status
Not insured (ref.)	1,229	13.7	163	13.5	–	–
Insured	7,531	86.4	801	9.8	0.70	[0.55, 0.89]	**
Substance use treatment history
None (ref.)	7,659	88.7	378	4.6	–	–
Any treatment	989	11.4	562	54.2	24.63	[19.79, 30.65]	***
Survey year
2000 (ref.)	1,579	17.0	150	9.8	–	–
2005	1,703	19.3	206	11.3	1.16	[0.90, 1.51]
2010	1,616	16.5	183	8.8	0.88	[0.66, 1.18]
2015	1,714	22.2	200	10.4	1.07	[0.79, 1.43]
2020	2,264	25.0	236	10.8	1.11	[0.86, 1.44]
Lifetime AUD criteria count	Unweighted M (SE)	Weighted M (SE)	Unweighted M (SE)	Weighted M (SE)	OR	[95% CI]	p
	3.89 (0.03)	3.86 (0.03)	7.62 (0.09)	7.55 (0.12)	1.80	[1.73, 1.88]	***

Notes: For categorical variables, column percentages are listed for the total sample and row percentages are listed for lifetime Alcoholics Anonymous (AA) attendance. Associations between sample characteristics and AA attendance were tested in weighted bivariate (logistic) regression models. To protect against over-interpreting differences between categories for variables with multiple categories (e.g., race/ethnicity, age, marital status, etc.), p values for specific category ORs are shown and interpreted only when the overall Wald test for the group of dummy variables representing these categories was significant. AUD = alcohol use disorder; OR = odds ratio; CI = confidence interval; ref. = reference.

^**p < .01;

^***p < .001.

Interaction models

Both unadjusted and adjusted models tested interactions between survey year and focal demographic variables, but no significant interactions emerged for race/ethnicity, age, or sex (not shown; all ps > .25). Supplemental Figures 1–3 display the prevalence rates of lifetime AA attendance for each demographic subgroup across survey year, based on adjusted models. (Supplemental material appears as an online-only addendum to this article on the journal's website.) Despite some variation from survey to survey, the figures show overlapping confidence intervals, suggesting no clear differences over time in associations between AA attendance and race/ethnicity, age, and sex.

Multivariate model excluding nonsignificant interactions

Table 2 displays findings from a single multivariate model including survey year, key demographics, and all covariates; interactions are omitted as all interactions for survey year and focal demographics were nonsignificant. Paralleling the bivariate models, results show significantly lower odds of AA attendance among those identifying as Latinx/Hispanic (vs. White; adjusted OR [aOR] = 0.63) and ages 18–29 (vs. 30–64; aOR = 0.35). However, AA attendance was also lower among those identifying as Black/African American (vs. White; OR = 0.59), and the sex difference was no longer significant. Other significant covariates in this model included any substance use treatment history (vs. none; aOR = 10.45) and lifetime AUD criteria count (aOR = 1.62).

Table 2.

Final multivariate model examining predictors of lifetime Alcoholics Anonymous attendance among those with a lifetime AUD (unweighted N = 8,876)

Variable	aOR	[95% CI]	p
Race/ethnicity
White (ref.)	–	–
Latinx/Hispanic	0.63	[0.43, 0.94]	*
Black/African American	0.59	[0.38, 0.92]	*
Age, in years
18–29	0.35	[0.22, 0.55]	***
30–64 (ref.)	–	–
≥65	1.09	[0.74, 1.62]
Gender
Male (ref.)	–	–
Female	0.92	[0.69, 1.21]
Marital status
Married/cohabiting	0.90	[0.63, 1.27]
Separated/divorced/widowed (ref.)	–	–
Never married	0.94	[0.60, 1.48]
Educational attainment
High school grad or less (ref.)	–	–
Some college	1.05	[0.75, 1.48]
College grad or more	0.96	[0.68, 1.35]
Employment status
Not employed (ref.)	–	–
Employed full-/part-time	0.80	[0.58, 1.08]
Insurance status
Not insured (ref.)	–	–
Insured	0.87	[0.58, 1.31]
Substance use treatment history
None (ref.)	–	–
Any treatment	10.45	[7.79, 14.01]	***
Lifetime AUD criteria count	1.62	[1.55, 1.70]	***
Survey year
2000 (ref.)	–	–
2005	1.21	[0.84, 1.74]
2010	0.70	[0.47, 1.06]
2015	1.04	[0.67, 1.60]
2020	0.72	[0.48, 1.08]

Notes: To protect against over-interpreting differences between categories for variables with multiple categories (e.g., race/ethnicity, age, marital status, etc.), p values for specific category ORs are shown and interpreted only when the overall Wald test for the group of dummy variables representing these categories was significant. AUD = alcohol use disorder; aOR = adjusted odds ratio; CI = confidence interval; ref. = reference.

^*p < .05;

^***p < .001.

Discussion

Results summary

The current study used aggregate NAS data to investigate disparities across race/ethnicity, age, and sex in lifetime AA attendance among those with a lifetime AUD, directly testing whether disparities have changed over time. Supporting the study's three a priori hypotheses and paralleling recent NSDUH findings (Hai et al., 2022), in multivariate analyses lifetime AA attendance was less prevalent for both Latinx/Hispanic and Black/African (vs. non-Hispanic White) participants and for emerging (vs. established) adults.

Multivariate results closely parallel the current study's bivariate results, except that the Black/African disparity emerged only in the multivariate analysis. This pattern appears to suggest that Black/African American participants with lifetime AUD attended AA less than would be expected given their elevated lifetime AUD severity (M = 4.03 criteria) versus White participants (M = 3.82 criteria; p < .05) and greater prevalence of lifetime SUD treatment utilization (14.0%) versus White participants (11.4%; p < .10), with both higher AUD severity and SUD treatment utilization strongly predicting higher odds of AA attendance overall (Table 1). Further, although prevalence of lifetime AA attendance was significantly lower among women than men in bivariate analysis, sex was unrelated to AA attendance in the multivariate analysis. Loss of the gender effect in multivariate analysis is likely attributable to controlling for AUD severity: Women in the sample reported lower lifetime AUD severity (M = 3.53 criteria) than men (M = 4.07 criteria; p < .001), and a logistic regression of AA attendance including only gender and lifetime AUD severity as predictors produced a nonsignificant gender effect (p > .39).

Multivariate findings are particularly notable because our model incorporated comprehensive controls including demographics, substance use treatment history, and lifetime AUD severity. Importantly, multivariate findings confirmed study hypotheses among both the overall sample and only those who received specialty treatment (N = 989; see supplemental table). To support interpretation, we computed effect sizes for findings in the total sample (Table 2) using adjusted odds ratios (aORs; which were converted to aORs > 1 using the reciprocal;⁵ Norton et al., 2018) and guided by Chen et al. (2009). Given that prevalence of lifetime AA attendance in the current study was 10.3%, effect sizes were small to medium for both the White versus Latinx/Hispanic and White versus Black/African American effects (aORs = 1.59 and 1.69 respectively, with ORs = 1.46–2.49 denoting small-to-medium effects) and medium to large for the age 30–64 versus age 18–29 effect (aOR = 2.86, with ORs = 2.50–4.13 denoting medium-to-large effects).

We found no changes in AA attendance over time, and time (indexed by survey year) had no effect on relationships between AA attendance and race/ethnicity, age, or sex. The only other significant (positive) contributors to AA attendance in the multivariate model were having a lifetime history of substance use treatment and higher lifetime AUD severity. Thus, despite increased awareness of the benefits of MHG participation, AA's reach among individuals experiencing AUDs has remained stable across these decades, as have disparities among racial/ethnic minorities and emerging adults.

Clinical implications and directions for future research

Results point to robust disparities in AA utilization among those with lifetime AUDs for Black/African American, Latinx/Hispanic, and emerging adults, disparities that have persisted over the past two decades (as shown by the lack of interactions with time). Findings highlight a pressing need for service providers to more actively and responsively facilitate MHG participation for these populations. Ideally, MHG facilitation should be an iterative process that includes consideration of the barriers to participation anticipated and encountered by Black/African American, Latinx/Hispanic, and emerging adults. Providers are encouraged to become familiar with the range of effective MHG options (including 12-step alternatives; Zemore et al., 2017, 2018b), modes of participation, meeting formats, and specialized meetings tailored to racial/ethnic minority and young people, both locally and online. Individuals presented with a range of options may be more likely to identify an appealing way to participate; further, evidence suggests that AA participation among people of color and young people is enhanced when representation of their racial/ethnic or age group is greater (Humphreys & Woods, 1993; Labbe et al., 2013). In addition, providers might consider supporting the construction of more abstinence-supportive social networks outside of MHGs. Social network change appears to be the most powerful known factor driving the benefits of 12-step involvement for substance use outcomes (Kelly, 2017; Kelly et al., 2012), and clinical trial data show that people with AUD who add nondrinkers to their networks have significantly better drinking outcomes, with equal benefits accruing regardless of whether or not nondrinkers came into the network via AA (Litt et al., 2016).

Findings also highlight directions for future research. Research is needed to ensure that MHG facilitation is maximally effective among underserved groups and to explore how to optimize the effectiveness of facilitating social network change outside of MHGs among these populations. Both strategies are encouraged because the reach of MHGs may be inherently limited due to lack of representation of people of color and young people in both 12-step groups and 12-step alternatives (Zemore et al., 2017). Notably, although 12-step facilitation (TSF) interventions have been widely studied, little is known about their effectiveness among racial/ethnic minority and emerging adults (cf. Project MATCH analyses; Arroyo et al., 2003; Davis, et al., 2017; Tonigan et al., 2002; Villanueva et al., 2007). TSF interventions adapted to address challenges common among racial/ethnic minority and emerging adults may be especially effective, particularly when paired with referral to specialized meetings serving these populations. Studies in these areas should be supported by systematic documentation of barriers to MHG participation in underserved groups, as research on barriers to MHG participation is very sparse, limited to 12-step groups, and generally dated. We also encourage research examining which MHG types, meeting formats, meeting characteristics, and participation modes are most congenial to underserved groups. Findings can inform both selection of the optimal intervention target (e.g., facilitation of 12-step participation, 12-step alternative participation, and/or general social network change) and intervention content. Eventually, this line may lead to cost-effectiveness studies testing whether new and adapted approaches offer greater benefits at lower costs than existing approaches to establishing longterm social/peer recovery support.

Limitations

Several limitations stem from the limited availability of measures in the NAS series. First, the current study necessarily focused on AA to the exclusion of other MHGs. However, AA is both the most commonly used and most widely studied source of support for AUDs (Kelly & Yeterian, 2011; Zemore et al., 2017); therefore, data speaking to disparities in attendance are valuable. Second, because the current study was limited to a yes/no measure of AA attendance, results cannot speak to length or intensity of involvement. Attending AA once is unlikely to provide benefits. However, to our knowledge no national epidemiological studies assess length or intensity of MHG attendance, and power would likely be insufficient for such outcomes in these studies. Fortunately, results are complemented by clinical studies, which have often adopted nuanced participation measures. Third, the study examined only any/no lifetime specialty treatment utilization; we were not able to capture quantity of treatment received, type of treatment received, or receipt of other (non-specialty) alcohol and drug services, all of which may affect the likelihood and impact of AA referral. Thus, we cannot fully dismiss a role for differences in SUD services receipt in the obtained disparities. Still, we underline that racial/ethnic and age-based differences remained robust both in a multivariate regression including lifetime specialty substance use treatment as a covariate (Table 2) and in a sensitivity analysis examining only those reporting lifetime specialty treatment (Supplemental Table). This pattern of results is inconsistent with speculations that disparities in SUD services receipt substantially contribute to racial/ethnic and age-based disparities.

In other limitations, changes across time in NAS survey mode and response rate mean that temporal effects should be interpreted with caution. Related, the sample size for the most recent NAS was relatively small and figures may be seen to imply some temporal variation in disparities. Thus, additional analysis of very recent, large data sets to confirm disparities would be valuable. Finally, we were unable to explore disparities in AA attendance for racial/ethnic minority individuals not identifying as Latinx/Hispanic or Black/African American and could not examine whether sex qualified the observed race/ethnicity and age effects. There is reason to expect disparities in MHG participation for Asian and Native American populations (Zemore et al., 2018a), and sex may qualify racial/ethnic disparities in SUD services use including MHG attendance (Zemore et al., 2014).

Still, strengths of this study include the large national data sets spanning two decades; excellent measurement of AUD symptom history and lifetime AA attendance; and comprehensive incorporation of covariates including AUD severity and specialty SUD treatment. Overall, the present study contributes to the literature as one of only a few large epidemiological studies to examine racial/ethnic and age disparities in MHG attendance.