Non-Engagement in Substance Use Treatment among Women with an Unmet Need for Treatment: A Latent Class Analysis on Multidimensional Barriers

Abstract

Background:

Women with substance use disorders experience multifaceted barriers in accessing substance use treatment. Little is known about how these barriers may aggregate. Using a person-centered approach, this study evaluates patterns of treatment barriers and the factors associated with experiencing distinct sets of barriers among women.

Methods:

Data were from the NSDUH (2015-2019). 461 adult women with an unmet need for substance use treatment in the last year reported on 14 treatment barriers. Latent class analysis examined classes of barriers; multinomial logistic regression assessed predictors of class membership.

Results:

Three classes were identified: just not ready to abstain (71.2%), logistical barriers and stigmatization (18.2%), and barriers across all dimensions (10.6%). Higher education (aOR:1.94, p=0.03) and psychological distress (aOR:2.19, p=0.02) predicted higher odds—and identifying as African American predicted lower odds (aOR:0.17, p=0.02)—of belonging to the “Logistics and Stigma Barriers” class relative to the “Just Not Ready” class. Similarly, higher education (aOR: 2.57, p=0.02) and having children (aOR:2.28, p=0.03) predicted higher odds—and marriage predicted lower odds (aOR:0.22, p=0.02)—of belonging to the “High and Diverse Barriers” class relative to the “Just Not Ready” class. Having children predicted higher odds (aOR: 2.93, p=.02), and marriage predicted lower odds (aOR:0.19, p=0.02) of belonging to the “High and Diverse Barriers” class relative to the “Logistics and Stigma” class.

Conclusion:

A lack of readiness to abstain, socioeconomic circumstances, and family obligations are main barriers to SUD treatment among women. Interventions incorporating motivational interviewing, family systems, and social networks are needed.

1. Introduction

In 2019, only 10.8% of women aged 12 or older with a substance use disorder (SUD) in the U.S. received any treatment for substance use (Substance Abuse and Mental Health Services Administration, 2020). This extensive treatment gap between those having SUDs and those who receive SUD treatment may indicate both universal treatment barriers across sex and unique treatment barriers experienced by women (Stringer and Baker, 2018; Taylor, 2010). For instance, both men and women may face internal barriers linked with decreased treatment utilization, including a perceived need for treatment and a lack of readiness to stop using substances (Ali et al., 2015; Choi et al., 2019). Life circumstances that generally predict experiencing barriers for both men and women include living in rural areas without nearby access to treatment (Edmond et al., 2015), and unstable access to health insurance (Ali et al., 2015). Yet, gender-specific barriers to substance use treatment persist.

Gendered responsibilities, such as caregiving, result in women being more likely than men to experience logistical obstacles to receiving needed treatment (Agterberg et al., 2020; Jones et al., 2019). American women spend a disproportionate amount of time providing childcare compared to men, partly attributed to the inaccessibility of free or affordable childcare in the United States broadly and at SUD treatment centers (Zamarro and Prados, 2021; Gould and Cooke, 2015). In 2020, only 5.5% of American SUD treatment facilities provided childcare (Substance Abuse and Mental Health Services Administration, 2021a).

In addition to having responsibilities that give rise to logistical barriers to treatment, women who use substances are more likely than men to perceive stigma for seeking help (Agterberg et al., 2020; Stringer and Baker, 2018). Stigmatization may be especially relevant for pregnant women who use substances, as these women can perceive judgment by healthcare professionals and fear the possibility of being reported to Child Protective Services (Adams et al., 2021; Falletta et al., 2018; Frazer et al., 2019; Stringer and Baker, 2018).

Moreover, the lack of access to treatment programs that are specifically trauma-informed may be an additional barrier to women seeking treatment for SUD. Women with SUDs often have high lifetime and recent traumatic experiences often necessitating trauma-informed care for SUD, which helps improve both client retention and satisfaction (Hales et al., 2019; Jones et al., 2019; Keyser-Marcus et al., 2015). Though trauma-informed services may enhance SUD treatment for women in general, within-group variations exist among women with SUDs, particularly among those with or without children. Around 40% of women in treatment for SUDs have children, and women with children vary significantly in drugs of choice, overdose history, health, domestic violence experiences, and SUD treatment outcomes compared to those without children (Canfield et al., 2021). As such, within-group differences among women with SUDs necessitate women-specific research on barriers to treatment.

Previous studies have used latent class analysis (LCA) on both male and female samples to understand the heterogeneity of treatment barriers experienced by various subgroups (Qi et al., 2013; Schuler et al., 2015; Wang et al., 2016). Of note, studies all identified a class (or subgroup) characterized by low or few barriers to accessing treatment and found demographic and substance use variables predictive of class membership. Yet, to the best of the authors’ knowledge, no prior studies have assessed the latent classes of barriers that might be experienced by women specifically. To understand the heterogeneity of ‘women’s experiences and to assess how different types of barriers may be experienced in tandem, we use LCA to evaluate barriers to treatment experienced by various subgroups of women. We also assess how demographic and other circumstantial variables and specific substance dependencies in the last year may predict class membership.

2. Material and Methods

2.1. Study Overview

We use data from the National Survey on Drug Use and Health (NSDUH), a cross-sectional, self-reported, publicly available, and nationally representative dataset collected annually in the United States (Substance Abuse and Mental Health Services Administration., 2020) and approved by the Institutional Review Board at Research Triangle International. We use five waves of data from 2015-2019 to generate a sample size of adult women needing substance use treatment sufficient for analyses. These particular waves were chosen because a partial questionnaire redesign in 2015, and the COVID-19 pandemic in 2020, rendered data incomparable with prior years (Substance Abuse and Mental Health Services Administration, 2021b, 2016).

2.2. Sample and eligibility

Participants were eligible for this sub-study if they were female, were 18 years of age or older, indicated that they needed treatment or counseling for the use of alcohol or drugs in the last 12 months, and did not receive that treatment. Two individuals who did not indicate whether they had received treatment or counseling during the previous 12 months were removed from the analytical sample. In total, the analytical sample included 461 people.

2.3. Measures

2.3.1. Barriers to receiving treatment

Among all surveyed adult women (N = 114,736) 1.4% received treatment for substance use in the past year, and only 0.4% (N = 461) reported that they needed treatment in the last 12 months and that they did not receive this treatment. Only this latter group reported their reasons for treatment non-reception and were included in this study. Participants responded to the question, “Which of these statements explain why you did not get the treatment or counseling you needed for your use of [alcohol/any drug/alcohol or any other drug”]?” Participants could endorse up to nine items, concluding with a tenth item, “some other reason or reasons.” If participants affirmed this last item, they were presented with five additional items, ending with a sixth item, “some other reason or reasons.” In total, a participant could endorse 14 specific barriers to treatment. These items were treated as indicator variables for the LCA. Two items referring to “some other reason or reasons” were not included as indicator variables due to lack of specificity. These 14 items are listed in Table 1. These barriers fell within four categories: logistical concerns, stigma, perception of treatment necessity, and personal readiness. 19 (4.1%) respondents did not endorse any 14 pre-specified barriers.

Table 1.

Barriers to Attaining Treatment or Counseling for Substance Use

Item	Category	N (%)
No healthcare coverage and ‘couldn’t afford the cost	Logistical Concerns	110 (23.9%)
Treatment not covered by healthcare coverage	Logistical Concerns	42 (9.1%)
No transportation/too far away/too inconvenient	Logistical Concerns	56 (12.1%)
‘Didn’t find a program that offered the type of treatment or counseling you wanted	Logistical Concerns	51 (11.1%)
Not ready to stop using	Personal Readiness	168 (36.4%)
There were no openings in the programs	Logistical Concerns	25 (5.4%)
Did not know where to go to get treatment	Logistical Concerns	83 (18.0%)
Concerned that getting treatment or counseling might cause your neighbors or community to have a negative opinion of you	Stigma	80 (17.4%)
Concerned that getting treatment or counseling might have a negative effect on your job	Stigma	80 (17.4%)
‘Didn’t think you needed treatment at the time*	Perception of Need	40 (8.7%)
Thought you could handle the problem without treatment*	Perception of Need	64 (13.9%)
‘Didn’t think treatment would help*	Perception of Need	19 (4.1%)
‘Didn’t have time (because of job, childcare, or other commitments)*	Logistical Concerns	43 (9.3%)
‘Didn’t want others to find out that you needed treatment*	Stigma	42 (9.1%)

Note.

^*denotes the items which were presented only after endorsing having “some other reason or reasons”, rather than those items which were presented to sample on the whole

Participants that endorsed the latter “some other reason or reasons” were asked to type the most important other reason they did not receive treatment. A research staff member coded these items into categories; these categories were then verified independently by two analysts, and disagreements were resolved via discussion. Open-ended responses that coincided with the 14 pre-specified barriers were included in the counts for the corresponding barrier. Fifty-three open-ended responses were coded as affirmative responses to pre-specified barriers.

2.3.2. Predictors

Participants reported their age using five levels: 18-25, 26-34, 35-49, 50-64, and 65 years or older.

Race/ethnicity was truncated into a four-level variable, Non-Hispanic White, Non-Hispanic Black/African American, Hispanic, and Other. “Other” includes Non-Hispanic Native American/Alaska Native, Non-Hispanic Native Hawaiian/Other Pacific Islander, Non-Hispanic Asian, and Non-Hispanic in more than one race. This variable was truncated for two reasons: 1) to avoid over-saturating the polytomous logistic regression model and 2) to accommodate subgroups with small sample sizes, as the groups included as “Other” accounted for only 9.8% of the sample in total. “Non-Hispanic White” was treated as the reference group.

Women reported on their educational attainment using an 11-point scale ranging from Fifth grade or less grade completed to College graduate or higher. This variable was truncated into a binary variable, 0 (High school diploma/GED or less) and 1 (greater than High school diploma/GED), for the reasons described above.

Participants reported their employment in the past week using nine possible options (see Table 2). This variable was truncated into a binary variable, 0 (had a job or volunteer work in the last week) and 1 (was not employed in the past week).

Table 2.

Counts of Study Variables

	N (%)
Year
2015	93 (20.2%)
2016	85 (18.4%)
2017	92 (20.2%)
2018	88 (19.1%)
2019	103 (22.3%)
Age
18-25	166 (36.0%)
26-34	114 (24.7%)
35-49	134 (29.1%)
50-64	37 (8.0%)
65+	10 (2.2%)
Race
Non-Hispanic White	292 (63.3%)
Non-Hispanic Black/African American	58 (12.6%)
Non-Hispanic Native American/Native Alaskan	17 (3.7%)
Non-Hispanic Native Hawaiian/Pacific Islander	2 (0.4%)
Non-Hispanic Asian	5 (1.1%)
Non-Hispanic More than One Race	21 (4.6%)
Hispanic	66 (14.3%)
Educational Attainment
Less than High School Diploma/GED	79 (17.1%)
High School Diploma/GED	138 (29.9%)
Some College Credit, No Degree	136 (29.5%)
Associates Degree	44 (9.5%)
College Graduate or Higher	64 (13.9%)
Employment in Past Week
Worked at full-time job	155 (33.6%)
Worked at part-time job	61 (13.2%)
Has job or volunteer work, but did not work	31 (6.7%)
Unemployed and looking for work	53 (11.5%)
Disabled	44 (10.0%)
Keeping house full-time	29 (6.3%)
In school/training	12 (2.6%)
Retired	4 (0.9%)
Does not have a job for some other reason	68 (14.8%)
Marital Status
Married	99 (21.5%)
Widowed	11 (2.4%)
Divorced/Separated	97 (21.0%)
Never Married	254 (55.1%)
Health insurance all last year	314 (68.1%)
Rurality	104 (22.6%)
Last-Year Serious Psychological Distress	295 (64.%)
Has at least one child in household	160 (34.7%)
Las Year Substance Dependence *
Alcohol	199 (43.2%)
Marijuana	51 (11.1%)
Cocaine	35 (7.6%)
Heroine	38 (8.2%)
Hallucinogens	7 (1.5%)
Inhalants	2 (0.4%)
Methamphetamine	63 (13.7%)
Pain Relivers	82 (17.8%)
Tranquilizers	26 (5.6%)
Stimulants	29 (6.3%)
Sedatives	6 (1.3%)

Note.

^*Categories are not exclusive from one another

Women reported on whether there was any time in the last 12 months that they “did not have any kind of health insurance or coverage” as either 0 (yes) or 1 (no). The higher value indicates stable access to health insurance for the last 12 months.

The urbanicity of the county in which each participant lived was classified as a small metro, large metro, or non-metro, using the 2013 Rural-Urban Continuum Codes by the USDA (2013). This variable was truncated into a binary variable of 0 (metro) and 1 (non-metro or rural).

Serious psychological distress in the last year was assessed using the Kessler-6 scale (Kessler et al., 2003). Participants reported nervousness, hopelessness, restlessness, sadness, worthlessness, and feeling that everything was an effort, which they experienced during their worst month in the last year. These items were rated on a scale of 1 (none of the time) to 4 (all of the time). Participants with a cumulative score of 13 or higher were rated as having serious psychological distress (1), while those with scores less than 13 were not (0). While the NSDUH does not assess specific mental health diagnoses that might be particularly relevant for women, such as PTSD, this variable acted as a broad proxy for mental health concerns.

Participants reported their marital status as married, widowed, divorced, separated, or never been married. This variable was truncated into a binary variable, 0 (currently single) and 1 (presently married).

Women reported the number of children under 18 that reside in their household. Because the majority of respondents (65%) reported that they did not have children in their household, this variable was also truncated into a binary variable 0 (no children in the household) and 1 (at least one child in the household).

Substance dependency was determined using the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) criteria. Dependence was evaluated for eleven substances: alcohol, marijuana, cocaine, heroin, hallucinogens, inhalants, methamphetamine, prescription pain relievers, tranquilizers, stimulants, and sedatives. For alcohol and marijuana, dependence items were presented if participants endorsed that they had used alcohol and marijuana in the last year for more than five days or if they did not endorse the specific number of days that they had used alcohol or marijuana in the previous year. The dependency items were presented for the other substances if the participant had previously endorsed any use in the last year.

Six items were presented regarding the ‘participant’s use of each substance. These items included “Spent a great deal of time over a month or more getting, using, or getting over the effects of [the substance]” and “Inability to cut down or stop using [the substance] every time tried or wanted to.” Participants were asked a seventh question regarding their withdrawal symptoms from alcohol, pain relievers, cocaine, heroin, sedatives, stimulants, and methamphetamine.

Substance dependence was not exclusive to one substance. Three or more problematic-use items needed to be endorsed to be classified as dependent on a substance. The mean number of substances participants reported being dependent on during the last year was 1.17 (Range = 0-7). Notably, 113 participants (24.5% of the sample) did not report being dependent on any of the eleven substances in the last year, which may be due to two main reasons. First, women may be experiencing dependence on a substance that was not included in those above eight, such as cigarettes or other nicotine products. Second, women may have endorsed fewer than three diagnostic criteria but still personally felt that they needed treatment for their substance use.

Because few participants reported dependence on hallucinogens (1.5%), inhalants (0.4%), or sedatives (1.3%), these variables were excluded as predictors of class membership; ultimately, eight binary variables regarding substance dependency were included in polytomous logistic regression models as predictors of class membership.

2.4. Analytical strategy

Using Latent Class Analysis, our study followed a three-step method—during which the model with the ideal number of classes is determined using unconditional models—to avoid over-extracting classes during the class enumeration process (Nylund-Gibson and Masyn, 2016; Weller et al., 2020). The 14 barriers described above in 2.3.1 were treated as indicator variables.

A class solution was selected by comparing models with 1-6 classes (Nylund-Gibson and Choi, 2018). The criteria for comparison of the models were the Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), sample size adjusted BIC (saBIC), and consistent AIC (cAIC). For each criterion, lower values were an indicator of improved model fit (Nylund-Gibson and Choi, 2018). For each model, 1000 starting values were tested to find the stable and global log-likelihood values rather than the local log-likelihood values.

Additionally, entropy—an assessment of how well a model defines its identified classes—and the lowest average latent class posterior probability were considered. For entropy, higher values indicated greater accuracy in defining classes (Clark and Muthén, 2009). The lowest average latent class posterior probability value among the classes for each model should be at least .70 to indicate acceptable class separation (Nagin, 2005). Interpretability of the item response probabilities was also considered in choosing the appropriate number of classes.

After selecting an LCA model, each participant was assigned a class membership, which was assigned using the largest modal posterior probabilities from the final LCA model. Class membership was then treated as a manifest nominal outcome variable predicted in a polytomous logistic regression model. The log odds were converted into odds ratios for ease of interpretability. Additionally, to aid in interpreting the predictors of class membership, the demographics of the predictors were also recalculated and stratified by class membership.

All analyses were conducted using the statistical program R. The LCA models were calculated using the poLCA function in the poLCA package (Linzer and Lewis, 2011). The multinom function in the nnet package(Venables and Ripley, 2002) was used to conduct the polytomous logistic regressions. Sample size adjusted BIC (saBIC) and consistent AIC (cAIC) were calculated manually using equations described by Tein and colleagues (Tein et al., 2013) and Anderson and colleagues (Anderson et al., 1998), respectively. R² entropy was calculated manually, and the poLCA.posterior function was used to calculate posterior probabilities.

3. Results

3.1. Sample demographics

Descriptive statistics of the study variables are listed in Table 2. Among the sample, the majority (60.7%) belonged to age brackets of 35 years or younger, were Non-Hispanic White (63.3%), and had never been married (55%). Most of the sample (about 77%) did not have a college degree. Around 34.7% of women had a child younger than 18 in their household, and alcohol was the most frequently reported substance of dependence (43.2%), followed by pain relievers (17.8%) and methamphetamine (13.7%).

3.2. Class solution

The information criterion values and additional diagnostics criteria are presented in both Table 3 and Figure 1. BIC, saBIC, and cAIC indicated that the three-class solution demonstrated the best fit for the data. AIC indicated that the six-class solution demonstrated the best fit. However, the most significant improvements in AIC were from a one-class to a two-class solution and then from the two-class solution to a three-class solution; additional classes resulted in lesser improvement in AIC. The three-class solution was chosen for interpretation and further analysis. It should be noted that the entropy value for the three-class model (0.67) indicates a “medium” rather than “high” ability to classify participants into latent classes (Clark and Muthén, 2009). However, the lowest average latent class posterior probability for this model (0.87) met the cutoff for acceptability, indicating good separation among the classes.

Table 3.

Model Fit Information Criterion and Additional Diagnostic Criteria for Various Class Solutions

	LL	AIC	BIC	saBIC	cAIC	Smallest Class Count (%)	Entropy	Lowest ALCPP
1 Class	−2421.50	4870.99	4928.86	4970.30	4942.86	461 (100%)	-	-
2 Classes	−2285.63	4629.25	4749.12	4834.95	4778.12	52 (11.3%)	.86	.95
3 Classes	−2211.79	4511.58	4693.45	4823.68	4737.45	49 (10.6%)	.67	.87
4 Classes	−2175.97	4469.93	4713.80	4888.42	4772.80	47 (10.2%)	.72	.85
5 Classes	−2148.46	4444.91	4750.78	4969.80	4824.78	46 (10.0%)	.80	.88
6 Classes	−2126.52	4431.04	4798.92	5062.33	4887.92	34 (7.4%)	.81	.89

Note. L.L.-Log-Likelihood; AIC- Akaike Information Criterion; BIC- Bayesian Information Criterion; saBIC- Sample Size Adjusted BIC; cAIC- Consistent AIC; Entropy- R² entropy; ALCPP- Average latent class posterior probability; Bold text indicates the lowest value for each criterion, as relevant.

Figure 1.

Information Criterion Values Across Classes

This is an elbow plot of four information criterion, from classes from 1-6. The solid line depicts the Akaike Information Criterion (AIC), the dashed-dotted line depicts the Bayesian Information Criterion (BIC), the dotted line depicts the same-size adjusted Bayesian Information Criterion (saBIC) and the dashed line depicts the consistent Akaike Information Criterion (cAIC).

3.3. Conditional item response probabilities

Conditional item response probabilities for each of the three classes are detailed in Table 4 and illustrated in Figure 2. The three classes were similar in that they all featured a moderate probability of endorsing “You were not ready to stop using.” Otherwise, the three classes differed in the magnitude of the response probabilities and the domains of items endorsed.

Table 4.

Item response probabilities for three-class solution.

Category	Item	Class 1: Just Not Ready (71.2%)	Class 2: Moderate Logistics and Stigma (18.2%)	Class 3: High and Diverse (10.6%)
Logistical Concerns	No healthcare coverage/’ couldn’t afford	0.19	0.40	0.27
	Treatment not covered	0.05	0.20	0.12
	No transportation/distance/inconvenience	0.09	0.21	0.16
	‘Didn’t find a program	0.06	0.21	0.22
	‘Didn’t have time	0.03	0.00	0.64
	No program openings	0.03	0.14	0.02
	‘Didn’t know where to go	0.07	0.45	0.34
Stigma	Might cause others to have a negative opinion	0.03	0.48	0.47
	Might have a negative effect on job	0.03	0.57	0.32
	‘Didn’t want others to find out	0.02	0.00	0.72
Need Perception	‘Didn’t think you needed treatment	0.08	0.00	0.32
	Thought you could handle the problem	0.08	0.00	0.82
	‘Didn’t think treatment would help	0.01	0.00	0.29
Personal Readiness	Not ready to stop using	0.35	0.41	0.46

Note. Items in this table are ordered by category, rather than by order of presentation to participants

Figure 2.

Item Response Probabilities for the three-class model

This figure depicts the item response probabilities for the three-class model. Gray polka dots indicate class 1, Just Not Ready; Stripes indicate class 2, Moderate Logistics and Stigma Barriers; Solid black indicates class 3, High and Diverse Barriers.

The majority of the sample (71.2%) had the highest probability of belonging to class 1. This class was labeled the Just Not Ready class. This class was characterized by low conditional item response probabilities across all items except for the item “You were not ready to stop using” (probability = 0.35).

18.2% of the sample had the highest probability of belonging to class 2. This class was labeled the Moderate Logistics and Stigma Barriers class. Aside from the item above regarding readiness to stop substance use, the response probabilities were focused on barriers related to logistics and stigma and were all moderate. For example, women in this class had a probability of 0.45 of endorsing the logistical barrier, “You did not know where to go to get treatment,” and a probability of 0.57 of endorsing the stigma-related barrier, “You were concerned that getting treatment or counseling might have a negative effect on your job.”

Finally, 10.6% of the sample had the highest probability of belonging to class 3. This class was labeled the High and Diverse Barriers class. This class featured high or moderately high item response probabilities across all three categories of barriers, aside from the previously mentioned item assessing readiness. Examples include a moderately high likelihood (0.64) of endorsing the logistical barrier, “You didn’t have time (because of job, childcare, or other commitments),” high likelihood (0.74) of endorsing the stigma barrier, “You didn’t want others to find out that you needed treatment.” High likelihood (0.82) of endorsing the needs-perception-related barrier, “You thought you could handle the problem without treatment.”

Demographics of each class, about the specific predictors of the polytomous regression, are detailed in Table 5.

Table 5.

Percentage of Endorsement of Demographic and Dependency Variables by Class Membership

	Just Not Ready (N=328)	Logistics and Stigma (N=84)	High and Diverse (N=49)
	N(%)	N(%)	N(%)
Race
Non-Hispanic White	195(59%)	66(79%)	31(63%)
Black/African American	51(16%)	2(2%)	5(10%)
Hispanic	48(15%)	12(14%)	6(12%)
Other	34(10%)	4(5%)	7(14%)
More than H.S. Education	154(47%)	56(67%)	34(69%)
Currently Employed	156(48%)	60(71%	31(63%)
Last Year Health Insurance	224(68%)	56(67%)	34(69%)
Rural	75(23%)	18(21%)	11(22%)
Last Year SPD	192(59%)	67(80%)	36(73%)
Marital Status	78(24%)	18(21%)	3(06%)
Any Children in Household	109(33%)	27(32%)	24(49%)
Last Year Dependency
Alcohol	134(41%)	41(49%)	24(49%)
Marijuana	36(11%)	10(12%)	5(10%)
Cocaine	24(7%)	7(8%)	4(08%)
Heroine	26(8%)	8(10%)	4(08%)
Methamphetamines	39(12%)	13(15%)	11(22%)
Pain Relievers	54(16%)	21(25%)	7(14%)
Tranquilizers	20(6%)	5(6%)	1(02%)
Stimulants	19(6%)	7(8%)	3(06%)

3.4. Polytomous logistic regression results

The results from all regression models are presented in Table 6. Class 1 (Just Not Ready) is the reference group in the first polytomous logistic regression model. This model was repeated while treating class 2 (Moderate Logistics and Stigma Barriers) as the reference group, allowing for the comparison of belonging to class 3 over class 2 (see Table 6).

Table 6.

Polytomous logistic regression model predicting class membership

	Just Not Ready < Logistics and Stigma		Just Not Ready < High and Diverse		Logistics and Stigma < High and Diverse
	OR(Beta)	SE	OR(Beta)	SE	OR(Beta)	SE
Intercept	0.06(−2.86)***	0.70	0.02(−3.75)***	0.90	0.41(−0.89)	1.05
Age	1.00(0.00)	0.14	1.06(0.05)	0.18	1.05(0.05)	0.21
Race (Non-Hispanic White)
Black/African American	0.17(−1.75)*	0.79	1.00(0.00)	0.58	5.76(1.75)	0.93
Hispanic	0.74(−0.3)	0.45	0.49(−0.72)	0.67	0.65(−0.43)	0.75
Other	0.33(−1.11)	0.61	1.24(0.21)	0.54	3.75(1.32)	0.74
Education	1.70(0.53)	0.32	2.36(0.86)*	0.41	1.38(0.33)	0.48
Current Employment	4.94(1.60)***	0.36	1.61(0.48)	0.39	0.33(−1.12)*	0.49
Last Year Health Insurance	0.59(−0.52)	0.38	1.39(0.33)	0.51	2.34(0.85)	0.57
Rural	1.01(0.01)	0.37	0.76(−0.28)	0.48	0.75(−0.29)	0.55
Last Year SPD	2.74(1.01)**	0.36	1.88(0.63)	0.41	0.69(−0.37)	0.51
Marital Status	1.12(0.11)	0.36	0.21(−1.58)*	0.65	0.18(−1.69)*	0.7
Children In Household	0.80(−0.22)	0.34	2.30(0.83)*	0.38	2.87(1.05)*	0.46
Last Year Dependency
Alcohol	1.06(0.06)	0.32	1.43(0.36)	0.40	1.35(0.3)	0.47
Marijuana	0.72(−0.32)	0.50	0.52(−0.65)	0.69	0.72(−0.33)	0.78
Cocaine	1.09(0.09)	0.63	1.88(0.63)	0.74	1.72(0.54)	0.86
Heroine	1.24(0.22)	0.61	1.34(0.29)	0.75	1.07(0.07)	0.87
Methamphetamines	1.26(0.23)	0.47	1.83(0.61)	0.54	1.46(0.38)	0.64
Pain Relievers	3.06(1.12)**	0.40	1.33(0.29)	0.54	0.43(−0.83)	0.6
Tranquilizers	0.63(−0.46)	0.68	0.47(−0.76)	1.14	0.74(−0.3)	1.24
Stimulants	0.94(−0.06)	0.56	1.00(0.00)	0.70	1.07(0.07)	0.79

Note.

^*p<.05,

^**p<.01,

^***p<.001;

Betas are log odds

3.5. Supplemental sensitivity analysis

Because participants had to opt-in to being presented with the last five barriers, a sensitivity analysis was run to assess which latent classes might be identified using only the first nine barriers. Tables and figures regarding this LCA are provided as supplemental material.

4. Discussion

The LCA model identified three classes of barriers women face in accessing treatment for SUDs. The most frequently endorsed barrier across the sample was a lack of personal readiness to stop using drugs. All three classes featured a moderate likelihood of endorsing this item. This finding suggests a strong need for interventions to encourage readiness for treatment. Motivational interviewing may be an effective way to prepare women for change and to help them understand the need for treatment, though not all findings support the efficacy of this approach (Frost et al., 2018). Aside from the similarity of the barrier to personal readiness, the three classes differed in the strength and type of the item response probabilities.

4.1. Just Not Ready

The majority of women (71.2%) in the sample belonged to the Just Not Ready class, which was characterized by a low likelihood of endorsing any barriers aside from personal readiness. This finding corroborates another nationally representative study, which found that most individuals in treatment belonged to a class that was likely to experience obstacles related to personal perceptions (Schuler et al., 2015).

Demographics of this class indicate that the women belonging to this class were particularly under-resourced, as less than half of those in this class were employed or had more than a high school education. The level of social disadvantage of this subgroup of women denotes the need to address basic needs in interventions for SUDs.

4.2. Moderate Logistics and Stigma Barriers

The next largest group (18.2%) belonged to a class with a moderate likelihood of endorsing logistical and stigma-related concerns. These concerns included being unable to afford the cost of care and concern about others’ negative opinions of their treatment-seeking. Most participants in this class were educated, employed, and experiencing psychological distress, suggesting that many of these women were working class and particularly stressed or experiencing mental illness.

Employment, psychological distress, and dependence on pain relievers predicted being more likely to belong to the Logistics and Stigma class than to the Just Not Ready class. These findings suggest that working women may be particularly apt to experience logistical barriers—such as the inability to afford to take time off work—and stigma-related barriers—such as worry that colleagues will find out about their need for treatment services. Findings regarding psychological distress and dependence on pain relievers underscore the need to address stigma and the emotional toll associated with having a SUD.

4.3. High and Diverse Barriers

Finally, the third and smallest class (10.6%) was characterized by a high likelihood of endorsing barriers across multiple domains, namely logistical concerns, stigma, and not perceiving the need for treatment. Examples include not having time to pursue treatment, not wanting others to learn about ‘one’s need for treatment, and thinking that one could handle their problem without treatment. Very few women in this class were married, while nearly half had children in their homes, suggesting that many were single mothers.

Gendered responsibilities (Zamarro and Prados, 2021), lack of access to child care (Gould and Cooke, 2015), and a perception of stigma for substance use by care providers, friends and family (Barnett et al., 2021) act as barriers to mothers who need SUD treatment, and this may be especially true of single mothers. Expectantly, these familial circumstances predicted class membership. Women with children in their homes were likelier to belong to the High class than the Moderate and Low classes.

Conversely, married women had higher odds of belonging to both classes with fewer barriers than the High class. There is some evidence that happy (McCrady et al., 2004) or close marriages (Heinz et al., 2009) predict improved treatment outcomes (Moos et al., 2002). Marriage may be a source of social support for women needing treatment. Alternatively, the same barriers that preclude participants from accessing treatment may also be those that would decrease one’s likelihood of currently being married (e.g., financial instability or limited time).

Concerning SES, participants with current employment had higher odds of belonging to the Logistics and Stigma class than to High and Diverse, supporting the interpretation that women in the former group may be working class. Participants with more education were likelier to belong to the High and Diverse class than to Just Not Ready, supporting the interpretation that women in the former group were particularly under-resourced.

4.4. Limitations

The main limitations of this study are five-fold: First, the barrier items were only presented to participants who did not receive treatment in the last year, and the patterns of barriers presented herein may differ from those experienced by women who accessed needed treatment. Secondly, the barriers assessed in the NSDUH dataset are not comprehensive; for example, barriers related to cultural concerns were not included. Third, the last five barriers were only presented to participants who endorsed that they had “other concerns” not included in the first nine items. However, it should be noted that these participants chose not to endorse “some other reason or reasons” and likely felt that the first nine items sufficiently described their experiences. Fourth, study protocols were such that the same individuals might have been assessed over multiple years. Finally, many women with SUDs have experienced past trauma and have symptoms of PTSD (Keyser-Marcus et al., 2015). However, these experiences and symptoms were not assessed in the NSDUH and were instead measured using the broader proxy variable of “serious psychological distress.”

4.5. Conclusion

This study highlights the heterogeneity of treatment barriers experienced by women with SUDs and an unmet need for treatment. The heterogeneity illustrates how these barriers may coalesce and be experienced in tandem. Clinicians and practitioners should remember that many women may face multiple barriers at once. For example, mothers in this sample had higher odds of belonging to a class of women who faced barriers related to stigma, logistics, lack of perceived need, and lack of readiness. In this instance, treating the logistical concerns alone—by providing childcare or contingency management financial incentives—may not be sufficient to help these women initiate needed treatment. Additional interventions—such as educational programs to reduce stigma in the community (see Livingston et al., 2012) and perhaps motivational interviewing designed to help initiate change (Frost et al., 2018)—may also be useful to help women initiate treatment.

Future work should continue to assess the heterogeneity of women’s barriers to initiating treatment for SUDs and explore how interventions can address the multiple types of barriers likely to be experienced by single mothers, working-class women, and those with psychological distress.

Highlights.

• Women face heterogenous barriers to accessing needed substance use treatment.

• Three classes of substance use treatment barriers were identified for adult women.

• Classes differed both in types of barriers, and likelihood of barrier endorsement.

• Familial, socioeconomic, and substance use experiences predicted class membership.