Pretreatment social network characteristics relate to increased risk of dropout and unfavorable outcomes among women in a residential treatment setting for substance use

Abstract

Increased retention in residential treatment for substance use disorder (SUD) has been associated with more favorable clinical outcomes for residents. Yet SUD treatment dropout remains high. It is essential to uncover factors contributing to these high rates. Little is known about whether features of an individual’s social network prior to treatment entry are related to number of days in treatment or to clinical status at treatment termination. To examine these relationships, we analyzed data from 241 women (58.5% Hispanic) entering an SUD residential treatment facility, who agreed to participate in a parent randomized control trial. We assessed characteristics of these women’s social networks prior to treatment entry at baseline. We extracted clinician-determined progress at treatment termination and days in treatment two months after treatment entry from clinical records. Data-driven analyses using purposeful selection of predictors showed that the overall size of the social network was associated with increased likelihood of being classified as having achieved good clinical progress in treatment at termination and that number of drug users in the pretreatment social network was related to staying fewer days in treatment. Contrary to our hypothesis, we found no significant associations between other pretreatment social support network characteristics (i.e., social support) and treatment retention or clinical discharge status. Future research should examine how features of social networks change through treatment and how these changes relate to treatment outcomes.

1. Introduction

Current rates of dropout from substance use disorder (SUD) treatment remain high (e.g., 30.4% in studies evaluating psychosocial treatments delivered in person; Lappan, Brown, & Hendricks, 2020). Identifying factors contributing to early treatment dropout is critical to improving treatments to more effectively and consistently engage those with SUD. Most extant studies on SUD treatment dropout have focused on demographic features (Brorson, Ajo Arnevik, Rand-Hendriksen, & Duckert, 2013), such as age (e.g., Brorson et al., 2013), sex (e.g., Mckowen et al., 2017; but see Stahler, Mennis, & DuCette, 2016), and minority status (e.g., Mennis & Stahler, 2016); or clinical characteristics, such as comorbid diagnoses (e.g., Krawczyk et al., 2017) and type of substance (e.g., Stahler et al., 2016).

Social factors or the characteristics of the social network of an individual with SUD might similarly affect treatment retention, particularly for women. While the relationships are complex when considering treatment type, social factors such as lack of support, more children, and more drug users in the social networks all might be specific risk factors for poorer retention and clinical outcomes in women with SUD (S. F. Greenfield et al., 2007). Research on the relationship between pretreatment social network composition and treatment retention is scarce. One recent study suggested that social identification with a therapeutic community eased the challenges of early treatment for individuals receiving SUD treatment within therapeutic communities (Haslam et al., 2019). More research is needed to evaluate how social network characteristics before treatment entry influence the likelihood of terminating residential treatment during the early stages of treatment among women with SUD.

The structure and complexity of social networks of individuals with SUD have so far been largely studied through self-report. The Important People Drugs and Alcohol scale (IPDA) and its predecessors (Important People and Activities Inventory, IPA; Important People Inventory, IPI) are empirically validated scales with good psychometric properties (Owens & Zywiak, 2016; Zywiak et al., 2009) and, by far, the most commonly used methods of social network evaluation in the SUD realm (Stone, Jason, Light, & Stevens, 2016). For example, IPA and IPI were included in the most prominent large-scale studies on the treatment of alcohol problems, projects MATCH (Zywiak, Longabaugh, & Wirtz, 2002) and COMBINE (Longabaugh, Philip, William, & Stephanie, 2010). These scales take inventory of the most important people in a participant’s network (social network alters) and the frequency and quality of contact between the participant and the alters. While the IPA and IPI focus on alcohol use only, the IPDA examines alcohol and other drugs (AOD) use. The rich social network information collected with the IPDA is then used to compose an evaluation of the participant’s social network in terms of its size, importance, substance use involvement, and level of support, with considerable variability across studies in what scores are calculated and how (Stone et al., 2016). While the IPDA does not capture all social network aspects that may be important (e.g., family conflict; Saunders et al., 2016), it provides valuable insight into the structural characteristics of the social network.

A large social network with or without daily contact, for example, can be indicative of less isolation and a stronger connection with the community. However, the qualities of the alters in the network might be important for substance use, because AOD use patterns among those alters can make AOD use more acceptable or available. Indeed, research on individuals with SUD has sometimes shown that larger (daily) network size is linked to better mental health and quality of life (Best & Lubman, 2017; Muller, Skurtveit, & Clausen, 2017), fewer days with drug use before treatment entry (Zywiak et al., 2009; but see Owens & Zywiak, 2016), and reduced AOD use severity and number of days with use following treatment (Zywiak et al., 2009). On the other hand, a higher number of AOD users in a network has been related to worse AOD treatment outcomes (Best & Lubman, 2017; Petrakis & Simpson, 2017; Tracy et al., 2016). Both the size of the network and its substance involvement have also been associated with alcohol use severity scores (Zywiak et al., 2009), possibly due to the highly social nature of alcohol consumption. Research has not consistently observed these relationships, thus more research is needed (Stone et al., 2016).

Nevertheless, the literature underscores the importance of examining both the size of the social network and the number of substance users in it, since these characteristics might separately contribute to SUD outcomes. We are not aware of any studies that have evaluated the relationship between these social network characteristics and early residential treatment retention or clinical progress. Social network characteristics and dropout might be associated since entering residential treatment can limit contact with social network alters, which can pose treatment continuation challenges.

The social support women receive might also influence the decision to leave residential treatment. One study among women in either inpatient or outpatient SUD treatment found that social support that alters from the treatment network (e.g., peers in treatment or service providers) provided was associated with increased likelihood to complete treatment, while social support from non-treatment-related alters was not (Jun, Tracy, & Min, 2017). This finding shows that specific aspects of social support might have divergent effects on treatment retention. Most studies using the IPDA have examined the effects of either general or treatment-specific social support, i.e., the attitudes of alters regarding participant’s decision to receive treatment. Some inconsistencies exist regarding the relationship between either general or treatment-specific social support and SUD indices, even though the general direction observed in research suggests that more support is linked to better outcomes (Stone et al., 2016). It is possible that both general and SUD treatment support would be related to the likelihood of dropping out.

The role of social networks might vary by gender and research has observed some gender differences in social network characteristics. In general, women’s networks tend to be larger and denser than those of their male counterparts (Szell & Thurner, 2013). A few studies have addressed gender differences in social networks among substance using samples. Mixed results have been reported on whether men or women have more substance using members in their social network (Grella & Joshi, 1999; vs. Mohr, Averna, Kenny, & Del Boca, 2001). Consistently, however, the literature suggests that women with SUD are more likely to have substance using partners than men with SUD (Bendtsen, Lejman Dahlström, & Bjurulf, 2002). The interactions between social networks and treatment outcomes might also differ by gender among individuals with SUD (Litt, Kadden, & Tennen, 2015). For example, females receiving medication-assisted treatment for opiate use disorder reported a higher number of AOD abstainers in their social network than males in one study (Kidorf, Latkin, & Brooner, 2016). Litt and colleagues (2015) found that women received less support to reduce drinking than men. Further, changes in social networks mediated stronger relationships between Alcoholics Anonymous meeting attendance and reduced substance use in men than in women (Kelly & Hoeppner, 2013). While most studies on social network in SUD have used mixed gender samples and have not tested for gender differences, a few have studied women exclusively (e.g., Min et al., 2013; Nargiso, Kuo, Zlotnick, & Johnson, 2014; Tracy et al., 2012), highlighting the importance of understanding the contribution of social networks to SUD treatment outcomes among women.

Another often overlooked moderating factor of the effects of social networks on treatment dropout might be ethnoracial group (S. F. Greenfield et al., 2007). Most studies have not tested ethnoracial differences, even though research has observed such differences in SUD treatment retention and response. Hispanic and non-Hispanic black individuals are less likely to complete SUD treatment (Mennis, Stahler, El Magd, & Baron, 2019), with some variability in which group is more at risk of leaving treatment early across the U.S. (Arndt, Acion, & White, 2013) or substances of choice (Mennis & Stahler, 2016; Stahler et al., 2016). Studying social networks could increase our understanding of the factors that contribute to racial/ethnic disparities in treatment retention and progress. Research has shown that ethnic groups vary in the interconnectedness of their networks and the social support that their networks provide (Flores et al., 2019). SUD treatment services utilization also varies by race with non-Hispanic blacks relying less on formal service providers than other racial groups (Perron et al., 2009).

Despite researchers’ increased interest in the role of social networks in SUD, little is known about the effects of social networks on treatment retention. We are not aware of any studies that have examined the relationship between social network characteristics and the likelihood of premature treatment discontinuation (e.g., staying fewer days in treatment before leaving treatment) in the early treatment stages. The early treatment stages are critical, as a high percentage of individuals drop out in the first 30 days (Loveland & Driscoll, 2014). Staying in treatment longer might facilitate receipt of more services and better outcomes (L. Greenfield et al., 2004). However, this is not a given. Individuals might remain longer in treatment without achieving the same progress as others who spent fewer days in treatment. Variability in problem severity, motivation, or adherence to treatment requirements might all be associated with a mismatch between treatment duration and progress even if treatment is identical. Therefore, treatment progress is just as important to study as is premature termination. Research has shown that the amount of social support a participant received from individuals involved in their treatment relates to the likelihood of treatment completion at discharge (Jun et al., 2017), but it is unclear whether other characteristics of the social network at treatment entry affect treatment progress at discharge.

To fill these significant gaps in research, this study used data from women admitted to a female-only residential SUD treatment facility who were taking part in a larger randomized clinical trial on a novel mindfulness-based program for substance use (Moment-by-Moment Women in Recovery, MBRP-W; Amaro & Black, 2017). Results from the parent study showed that women who participated in the mindfulness-based program were more likely to leave treatment with improved progress (Black & Amaro, 2019) after 5 months of treatment. In the current study, we evaluated the impact of social network structure characteristics on early treatment dropout while controlling for intervention participation, as pre-treatment networks are more likely to have greater effects on treatment retention measures early in treatment rather than later in treatment when changes in social networks could be expected as a result of treatment.

Using a data-driven approach, we first examined the impact of validated pretreatment social network characteristics from the IPDA on both the risk of staying fewer days in treatment in the first eight weeks after treatment entry and the likelihood of failing to achieve satisfactory treatment progress by the time of discharge within the same timeframe. Second, we evaluated whether racial/ethnic group moderated these effects. Our hypotheses were that both the likelihood of staying longer in treatment and treatment completion would be lower among women with social networks characterized by higher substance use and limited general and treatment support. We further expected a moderation of these effects by race/ethnicity. Due to the mixed findings on the role of ethnicity in treatment retention, we did not have a hypothesis about the specific pattern of moderation by ethnoracial groups.

2. Methods

2.1. Participants

We recruited participants (N = 245) from new enrollees in a female-only residential SUD treatment facility in a large metropolitan area in the United States. The facility provided comprehensive gender-specific residential SUD treatment. Services provided included relapse prevention, (psycho)education, individual and group therapies, case management, education and job training, child services, and referrals to external services (e.g., medical). Facility clinicians recommended duration of treatment on an individual basis; treatment could last up to one calendar year (Black & Amaro, 2019).

On a rolling basis and within approximately two weeks of treatment entry, facility clinicians initially screened participants for eligibility for the parent study (i.e., a randomized control trial of a 6-week novel behavioral intervention). If interested in learning more, they referred participants to the research interviewer who obtained informed consent (for detailed information about the complete study protocol, see Amaro & Black [2017]). All participants met the parent study inclusion criteria (i.e., current patient at the facility, female, age 18–65, SUD diagnosis, and English proficient) and none met exclusion criteria (i.e., unable to provide informed consent, cognitively impaired, untreated psychotic disorder, suicidal ideation in the past 30 days, enrolled in the site’s program for incarcerated women, beyond 6 months of pregnancy, refusing to have study interviews audio recorded, or declining to permit analysis of their data for future studies).

2.2. Overview of procedures

After obtaining informed consent, study interviewers administered a protocol that included a baseline interview, which collected information on demographics (e.g., age, race/ethnicity), substance use, mental health, and other psychological characteristics using REDCap (Harris et al., 2009). Subsequently, we randomized participants to one of two study conditions, they received the intervention and were contacted for two additional measurement points: at postintervention and 8 months thereafter (see Amaro & Black [2017] for full description of parent study methods). The University of Southern California Institutional Review Board approved all procedures of the parent study.

2.3. Measures

2.3.1. Treatment termination and progress

We derived study participants’ treatment entry dates, discharge dates, and clinician-determined progress at discharge from clinical records that the treatment facility provided. We calculated days until discharge based on treatment admission date and discharge date and they were censored at 8 weeks after the start of the treatment at the study site. Using a data-driven approach, we selected to examine dropout within the first 8 weeks after treatment entry. Specifically, in our sample, 20.3% of the sample had terminated treatment within this window, which allowed us to statistically test our research question.¹ Other studies have shown that dropout is common in this time frame. For example, participants with SUD dropped out after 8.4 weeks in residential treatment in Norway (Brorson, Arnevik, & Rand, 2019) and participants with methamphetamine SUD dropped out after 8.5 weeks in randomized control trials of psychopharmacological interventions (Cook, Quinn, Heinzerling, & Shoptaw, 2017). Clinicians discharged participants from the residential facility either voluntarily, by clinical recommendation, or upon incarceration. At discharge, a team of clinicians involved in the individual care of each participant determined clinical status. We used two types of clinical status at 8 weeks after treatment entry for this study: 1) still in treatment, completed treatment, or left treatment with satisfactory progress; or 2) left treatment without satisfactory progress. Facility clinicians determined that participants completed treatment when a participant had achieved therapeutic goals and they needed no further intensive treatment regardless of the duration of treatment, while satisfactory progress was recorded when participants achieved some therapeutic goals partially or fully but required continued intensive treatment. Clinicians determined that satisfactory progress was not achieved when the individual either left treatment too soon after admission, had returned to substance use (determined by participant disclosure or a positive drug test), or did not follow the rules within the residential facility (e.g., became violent).

We assessed pretreatment social network characteristics at baseline by using the Important People Drug and Alcohol scale (IPDA; Zywiak et al., 2009). We asked participants to focus on the 30 days prior to treatment entry and identify up to six important people, who were aged 12 or older, and to categorize the relationship with the listed social contact (e.g., mother, father, partner). For each person listed, participants reported the following: 1) frequency of contact, ranging from 0, “single contact”, to 7, “daily contact”; 2) rating of importance, ranging from 1, “not at all”, to 6, “extremely”; 3) degree of general support the person provided to the participant, ranging from 1, “not at all supportive”, to 6, “extremely supportive”; 4) alcohol use status (light, moderate or heavy drinker, abstainer or recovering alcoholic); 5) frequency of alcohol use, ranging from 0, “not at all”, to 7, “daily”; 6) ratings of support for participant’s alcohol use, ranging from 1, “left or made you leave when you were drinking” to 5, “encouraged it”; 7) drug use status (light, moderate, or heavy drug user; abstainer or recovering drug user); 8) frequency of drug use, ranging from 0, “not at all”, to 7, “daily”; 9) ratings of support for participant’s drug use, ranging from 1, “left or made you leave when you were using drugs” to 5, “encouraged it”; and 10) support for participant coming to SUD treatment, ranging from 1, “strongly opposes it” to 6, “strongly supports it”. Participants were also able to respond that they do not know the correct answer to some of these questions.

A number of social network indices can be derived from the IPDA and have been previously validated (Owens & Zywiak, 2016; Zywiak et al., 2009). We calculated nine social network characteristics: 1) number of people in the social network, 2) number of people in daily contact with the participant, 3) the average importance rating of the four social network members rated the highest, 4) number of alcohol users in social network, 5) number of drug users in social network, 6) network’s AOD involvement, 7) general social support, 8) treatment support, and 9) support for AOD use. We calculated network’s AOD involvement (index 6) based on drug status and frequency of use of social network members adjusted for frequency of contact. We calculated general (index 7) and treatment (index 8) support composite scores based on minimum, maximum, and average ratings of the respective questions among the top four most important social network members. The AOD use support composite scores were based on support given for participants’ drinking and drug use from the four most important members of the social network. Scoring algorithms for this study were based on a manual, obtained from the first author of the original IPDA validation paper (Zywiak, personal communication, July 31, 2017). Rank normalization as implemented in the R function “rz.transform” from the “cape” package was used when scoring the IPDA to normalize the data distribution (Tyler et al., 2013).

2.3.2. Demographics

Participants reported age, ethnoracial group, marital status, pregnancy, and number of children in their legal custody during the baseline interview. We determined ethnoracial group based on two questions: 1) whether the participant was Hispanic/Latina and 2) what races (black or African American, white, American Indian, Alaskan Native, Asian/Pacific or Other) they endorsed. The race question was considered only for non-Hispanic study participants. We coded individuals who were not Hispanic and endorsed multiple races in addition to black or African American as non-Hispanic black. We coded non-Hispanic individuals with other multiple races as Other.

2.3.3. Treatment mandated status

We obtained information about whether an external agency had mandated treatment for the individual from clinical records. Participants in this study received treatment voluntarily or criminal court or child protective services mandated them to treatment (Department of Children or Family Services, DCFS).

2.3.4. Mental health diagnoses

Facility clinicians diagnosed SUD and other comorbid mental health conditions based on criteria from either the 4-TR or 5th edition of the Diagnostic and Statistical Manual of Mental Health Disorders (DSM; American Psychiatric Association, 2000, 2013), using the Adult Initial Assessment (AIA Los Angeles County Department of Mental Health, 2011). We divided participants into three categories: 1) those with alcohol use disorder only (AUD), 2) those with drug use disorder only (DUD), and 3) those with both AUD and DUD (AUD+DUD). We also recorded comorbid disorders besides AUD and DUD (e.g., PTSD).

2.3.5. Alcohol and drug use frequency

We collected data capturing alcohol and drug use frequency using the Alcohol and Drug Timeline Follow Back (TLFB) Calendar (Sobell & Sobell, 1992). The TLFB instrument assesses frequency of substance use through prompting individual retrospection by asking about personally relevant events (e.g., birthdays, anniversaries) and whether they were in any controlled environment (e.g., treatment facility or jail). For every day of the 8 months preceding treatment facility entry, we asked study participants to report whether they had used any alcohol, illicit drugs or prescription drugs to get high, as well as the number of cigarettes smoked. Based on these data, we established type of substances used within this 8-month window. We did not consider alcohol use (if not to intoxication) and cigarette use for these analyses. Most participants used more than one substance within this period.

2.3.6. SUD severity

We obtained information about SUD severity using a modified Addiction Severity Index – Lite (Cacciola, Alterman, McLellan, Lin, & Lynch, 2007). We asked participants about money spent on alcohol, as well as the number of days they experienced alcohol or drug problems in the 30 days prior to treatment entry. Further, they reported their subjective evaluation of how bothered they were by these drug and alcohol problems and how important they considered treatment to address these problems on a 5-point Likert scale (ranging from 0, “not at all”, to 5, “extremely”). We calculated alcohol and drug use severity scores based on these responses and TLFB alcohol and drug use data. We calculated family problems composite scores on the basis of self-reported relationship satisfaction, problems getting along with members of the social network, number of days during which social problems were experienced, and subjective evaluation of family problems. We also calculated legal problems composite scores on the basis of questions regarding days of illegal activities, amount of illegal income, and subjective evaluation of legal problems. We calculated all ASI composite scores according to the ASI scoring manual (McGahan, Griffith, Parente, & McLellan, 1986). While these composite scores cannot be easily interpreted, they can be used to statistically control for differences in SUD severity.

2.4. Statistical analyses

We used two separate analyses to examine the associations between pretreatment social network characteristics with two outcome variables: 1) days in treatment calculated from date of treatment entry to discharge censored at the eighth week after treatment entry; and 2) treatment discharge status as measured by a dichotomous outcome variable indicating a) whether the participant was still in treatment, or completed treatment or had left treatment with satisfactory progress; or b) whether the participant had left treatment without satisfactory progress by the eighth week after treatment entry. We used Cox hazard-proportional regression in the first analysis; and we used logistic regression in the second analysis.

All Cox and logistic regression models were initially fitted for each of the nine pretreatment social network characteristics. Then, social network characteristics with a p-value smaller than .25 were fitted into a final model to estimate their collective effects. We used purposeful selection (Bursac, Gauss, Williams, & Hosmer, 2008; Zhang, 2016) in both the initial and final models to select the appropriate set of control variables.

The purposeful selection procedure (Bursac et al., 2008; Zhang, 2016) included four steps. First, univariate models for each of the candidate control variables were fitted. We kept variables with p-values smaller than .25 for further multivariate selection. In the second step, multivariate models were fitted with all variables selected from the first step. Second, we iteratively checked variables, and we removed a variable from the multivariate models only if the variable’s influence was not significant at the .05 level, and the variable was not a confounder. We treated a control variable as a confounder if removing this variable from the multivariate model would cause at least a relative change of 10% in the coefficient estimate of the independent social network variable. In the third step, we checked the model goodness-of-fit of the multivariate model derived from step 2. We added control variables back to the model if the model lacked goodness-of-fit. Finally, we added theoretically informed interaction terms to the multivariate model. We kept an interaction term in the model if the term was significant and the model goodness-of-fit was significantly improved. Compared with other popular variable selection methods such as stepwise selection, the purposeful selection considers confounding and interaction effects, and, therefore, is likely to produce an accurate estimation of an association of interest.

We forced all the regression models to include the main and interaction effects of the social network characteristic and non-Hispanic white and non-Hispanic black as dummy coded variables. We used as a reference group individuals with Hispanic ethnicity, since this was the largest ethnoracial group in our sample. In the regression models we also controlled for variables selected from the baseline characteristics as shown in Table 1, using purposeful selection.

Table 1.

Descriptive of sample demographic characteristics (N=241).

Demographic characteristics	Mean (95% CI) or Frequency (%)
Age	32.11 (30.99~33.23)
Ethnoracial group
Non-Hispanic White	50 (20.7%)
Non-Hispanic Black	50 (20.7%)
Hispanic	141 (58.5%)
Marital status
Married	17 (7.1%)
Never married	179 (74.3%)
Divorced/Widowed/Separated	45 (18.7%)
Treatment mandated by
Criminal justice	112 (46.5%)
Child protective services (CPS)	81 (33.6%)
Not mandated	48 (19.9%)
Pregnant	18 (7.5%)
ASI composite scores
Alcohol severity in the 30 days before treatment entry	0.13 (0.11~0.16)
Drug score in the 30 days before treatment entry	0.15 (0.14~0.17)
Legal score in the 30 days before baseline	0.20 (0.17~0.24)
Family score in the 30 days before baseline	0.25 (0.22~0.28)
Number of underage children in participant’s legal custody
No child	18 (8.3%)
Have child(ren) but 0 in participant’s legal custody	117 (54.2%)
1–2 children in participant’s legal custody	62 (28.7%)
≥3 children in participant’s legal custody	19 (8.8%)
SUD diagnoses
Alcohol use disorder (AUD) only	23 (9.8%)
Drug use disorder (DUD) Only	180 (76.6%)
Both AUD and DUD	32 (13.6%)
Mental health diagnosis besides SUD	137 (56.8%)
Ever used during the 8 months before treatment entry
Alcohol to intoxication	120 (49.8%)
Marijuana	136 (56.4%)
Methamphetamines	187 (77.6%)
Cocaine/Crack	33 (13.7%)
Opiates	21 (8.7%)
Other Drugs	49 (20.3%)

We implemented the purposeful selection procedure in R, version 3.4.4 (R Core Team, 2018). We estimated Cox regression models by maximum likelihood as implemented in the “coxph” function of the R package “survival” (Therneau & Grambsch, 2000). Logistic regression models were fitted by maximum likelihood as implemented in the R base function “glm” (R Core Team, 2018).

3. Results

3.1. Study sample

Two hundred and forty-one women made up the analytic sample for the current study. From the total study sample, we excluded four participants, who identified as other race/ethnicity besides the three primary groups: Hispanic, non-Hispanic black, and non-Hispanic white. The average age of participants in this study was 32.11 years and the majority were Hispanic (58.5%), never married (74.3%), and mandated to SUD treatment by the criminal justice system (46.5%) or child protective services (33.6%). Most were diagnosed with drug use disorder (DUD, 76.6%) and had a co-occurring mental health disorder other than SUD (56.8%). The most common substances used in the 8 months prior to treatment entry were methamphetamines (77.6%), marijuana (56.4%), and other drugs (alcohol to intoxication 49.8%, cocaine/crack 13.7%, and opiates 8.7%). Additional descriptive statistics about the study sample can be found in Table 1. At 8 weeks post-treatment entry, 183 (75.93%) participants were still in treatment, 12 (4.98%) had not completed treatment, but left treatment with satisfactory progress, and 50 (20.75%) had left without satisfactory progress. No participants had completed treatment within the timeframe.

3.2. Description of pretreatment social network characteristics

Women in this sample reported an average of 4.28 important people (95% CI: 4.06~4.49) in their pretreatment social networks and, on average, participants were in daily contact with 3.62 (95% CI: 2.76~4.48) of these alters prior to treatment entry. Among pretreatment social network members rated by participants as important, very important, or extremely important, the average importance rating was 5.82 (95% CI: 5.78~5.85), which is in between very important (rating of 5) and extremely important (rating of 6). Among these important alters in women’s pretreatment network, the average number of alcohol and drug users was 1.59 (95% CI: 1.42~1.77) and 1.02 (95% CI: 0.86~1.73), respectively. In their network of important people, more than one fourth (23%, n = 56) reported having at least one parent who was an alcohol user and a few reported having at least one partner (21.2%, n = 52) or ex-partner (11.8%; n = 29) who used alcohol. In terms of drug use, a few (7.5%, n =18) reported having at least one parent who used drugs and a few reported having at least one partner (21.2%, n = 52) or ex-partner (11%, n = 27) who was a drug user. Participants also listed people who did not fall into either of these two relationship categories (e.g., sibling, friend, other family member) and were drug or alcohol users. Descriptive statistics for the rest of the social network characteristics examined were as follows: substance involvement of social network members M = 26.48, 95% CI = 23.10~29.85, Range = –12.91~105.43; general support M = 24.11, 95% CI = 23.24~24.99, Range = 3.43~32.31; treatment support M = 18.71, 95% CI = 18.20~19.22, Range = 6.26~22.57; and support for substance use M = 6.23, 95% CI = 5.44~7.01, Range = –0.40~25.97.

3.3. Pretreatment social network characteristics and days in treatment

This section presents results of three models (see Table 2). The Cox regression analysis showed the associations between two pretreatment social network characteristics and days to treatment termination were significant in the initial models (Models I & II in Table 2). We found that a larger number of people in the client’s social network was significantly associated with a smaller risk of staying fewer days in treatment (HR = .53, 95% CI = .28~1.00, p = .05); and a larger number of drug users in a social network was significantly associated with greater risk of fewer days in treatment (HR = 1.58, 95% CI = 1.02~2.44, p = .04). The Appendix Table I shows the complete results for the initial models for each of the nine social network characteristics. A third characteristic, number of social network members in daily contact, had a p-value smaller than .25 (HR = .80, 95% CI = .61~1.04, p = .09) and based on a priori criteria, we included it in the final model. Model III of Table 2 includes the final model with the three social network characteristics mentioned above.

Table 2.

Cox hazard-proportional regression analyses of the association between pre-treatment social network characteristics and the days to treatment termination, right censored at 8 weeks after treatment starts.¹

IPDA variables and their interaction with race/ethnicity	Outcome: Days to treatment termination
	Model I1		Model II1		Model III1
	Hazard ratio (95% CI)	p	Hazard ratio (95% CI)	p	Hazard ratio (95% CI)	p
Number of people in social network 2	.53 (.28~1.00)	.05	NA		.49 (.19~1.30)	.15
× Non-Hispanic white	1.22 (.41~3.58)	.72			.93 (.12~6.92)	.94
× Non-Hispanic black	1.99 (.53~7.44)	.31			1.71 (.31~9.48)	.54
Number of drug users2	NA		1.58 (1.02~2.44)	.04	1.85 (1.13~3.02)	.01
× Non-Hispanic white			.65 (.23~1.86)	.42	.71 (.23~2.20)	.55
× Non-Hispanic black			.62 (.26~1.49)	.29	.56 (.22~1.39)	.21
Number of people in daily contact2			NA		.79 (.59~1.05)	.10
× Non-Hispanic white					1.37 (.77~2.44)	.28
× Non-Hispanic black					1.29 (.77~2.16)	.22
Control Variables
Non-Hispanic white	.45 (.05~3.84)	.47	.99 (.28~3.52)	.99	.64 (.01~28.32)	.82
Non-Hispanic black	.29 (.02~4.37)	.37	2.17 (.76~6.21)	.15	.49 (.02~10.36)	.65
Mental health diagnosis besides SUD	.36 (.21~.63)	<.001	.45 (.26~0.78)	.004	.43 (.24~0.77)	.004
DCSF mandated treatment	1.36 (.71~2.63)	.36	1.36 (.71~2.60)	.35	1.67 (.83~3.37)	.15
Not mandated treatment	2.38 (1.22~4.61)	.01	2.36 (1.18~4.70)	.01	2.41 (1.17~4.98)	.02
Age	.97 (.93~1.01)	.13	.97 (.93~1.01)	.10	NA
Married	.49 (.12~2.04)	.32	.46 (.10~2.00)	.30
Divorced/ Widowed/Separated	1.49 (.74~2.98)	.26	1.54 (.75~3.15)	.24
Used opiates during 8 months before treatment entry	NA		2.25 (1.04~4.86)	.04	2.17 (1.02~4.65)	.05

¹Cox hazard-proportional models were fitted for each of the nine IPDA variables separately. Only models with significant IPDA variables are presented (Model I & II). See Appendix Table I for full results. Model III was fitted for all IPDA variables with p-value smaller than .25 in Appendix Table I. Control variables were selected separately for each of the model fitted using purposeful selection method from age, race/ethnicity, marital status, mandating agency, pregnancy, ASI alcohol, drug, legal and family problems ASI scores, number of children in legal custody, mental health diagnosis other than SUD, AUD/SUD diagnosis, intervention randomization, and use of alcohol, marijuana, methamphetamines, cocaine/crack, opiates, and other drugs 8 months before treatment entry.

²Square root was taken before entering the regression analysis to increase normality

The result of the final model (Model III in Table 2) shows that a larger number of drug users in the client’s social network was significantly associated with greater risk of staying fewer days in treatment (HR = 1.85, 95% CI = 1.13~3.02, p = .01). Neither larger number of people in the social network (HR = .49, 95% CI = .19~1.30, p = .15), nor larger number of people in daily contact (HR = 0.79, 95% CI = .59~1.05, p = .10) reached significance, when considering the three social network characteristics collectively.

The interaction terms of the two race/ethnicity dummy variables with social network characteristics did not reach significance, indicating no moderation. These dummy variables were also nonsignificant as control variables. Finally, two control variables were consistently significant in the models: first, having a mental health diagnosis other than SUD and/or AUD was significantly associated with a smaller risk of fewer days in treatment (all ps ≤ .004); and second, nonmandated treatment (i.e., entering treatment voluntarily) was significantly associated with a larger risk of fewer days in treatment (all ps ≤ .02). Having ever used opiates during 8 months before treatment entry was significantly associated with greater risk of fewer days in treatment in the initial model for number of drug users (p = .04) and the final model (p = 0.05).

3.4. Pretreatment social network characteristics and leaving treatment without satisfactory progress

In this section we present results of three models (see Table 3) testing the associations between pretreatment social network characteristics and leaving treatment without satisfactory progress. The initial logistic regression models identified two social network characteristics significantly associated with having left treatment without satisfactory progress at 8 weeks after treatment entry (Models I & II in Table 3). We found a larger number of people in the client’s social network to be significantly associated with smaller risk of leaving treatment without satisfactory progress (OR = .35, 95% CI = .13~.85, p = .03); and a larger number of people in daily contact was significantly associated with smaller risk of leaving treatment without satisfactory progress (OR = .68, 95% CI = .44~.96, p = .05). Appendix Table II shows the complete results for each initial model with the nine social network characteristics. A third characteristic, support for AOD use, had a significance level smaller than .25 (OR = 1.05, 95% CI = 0.98~1.14, p = .15) and we therefore included it in the final model. The final model (Model III in Table 3) shows results for analyses that included all three social network characteristics mentioned and their relationships to status at treatment termination. The final model results revealed that the only significant characteristic associated with status at treatment termination at 8 weeks after treatment entry was the number of people in the client’s social network (OR = .34, 95% CI = .12~0.86, p = .03).

Table 3.

Logistic regression analyses of the association between pre-treatment social network characteristics and treatment progress at discharge in the 8 weeks after treatment entry (Still in treatment/Completed treatment/Left Treatment with satisfactory progress vs. Left treatment without satisfactory progress).¹

IPDA variables and their interaction with race/ethnicity	Outcome: Leaving treatment without satisfactory progress
	Model I1		Model II1		Model III1
	Odds ratio (95% CI)	p	Odds ratio (95% CI)	p	Odds ratio (95% CI)	p
Number of people in social network2	.35 (.13~0.85)	.03	NA		.34 (.12~0.86)	.03
× Non-Hispanic white	2.08 (.56~8.51)	.28			1.54 (.36~6.95)	.56
× Non-Hispanic black	3.48 (.62~29.93)	.19			2.74 (.37~27.82)	.35
Number of people in daily contact2	NA		.68 (.44~0.96)	.05	0.74 (.49~1.05)	.12
× Non-Hispanic white			1.74 (.93~3.27)	.08	1.79 (.92~3.55)	.09
× Non-Hispanic black			1.66 (.89~3.16)	.11	1.61 (.82~3.28)	.17
Support for AOD use			NA		1.06 (.98~1.14)	.16
× Non-Hispanic white					0.98 (.85~1.11)	.73
× Non-Hispanic black					0.92 (.77~1.07)	.28
Control Variables
Non-Hispanic white	.20 (.01~2.61)	.23	.45 (.13~1.33)	.17	.21 (.01~2.89)	.26
Non-Hispanic black	.06 (.01~1.89)	.15	.41 (.09~1.48)	.21	.10 (.01~3.19)	.24
Mental health diagnosis besides SUD	0.34 (.17~0.68)	.003	.37 (.18~0.75)	.006	0.36 (.17~.74)	.006
Ever use alcohol during 8 months before treatment entry	0.67 (.33~1.34)	.26	NA		NA
DCFS mandated treatment	NA		1.29 (.54~3.04)	.56
Not mandated treatment			1.82 (.73~4.44)	.19

¹Logistic regression models were fitted for each of the nine IPDA variables separately. Only models with significant IPDA variables are presented (Model I & II). See Appendix Table II for full results. Model III was fitted for all IPDA variables with p-value smaller than 0.25 in Appendix Table II. Control variables were selected separately for each of the model fitted using purposeful selection method from age, race/ethnicity, marital status, mandating agency, pregnancy, ASI alcohol, drug, legal and family problems ASI scores, number of children in legal custody, mental health diagnosis other than SUD and AUD, AUD/SUD diagnosis, intervention randomization, and use of alcohol, marijuana, methamphetamines, cocaine/crack, opiates, and other drugs 8 months before treatment entry.

²Square root was taken before entering the regression analysis to increase normality

We did not observe any significant moderation of race/ethnicity in the effects of social network characteristics on the likelihood of leaving treatment without satisfactory progress by 8 weeks after treatment entry. Finally, the control variable, having a mental health diagnosis besides SUD or AUD, was consistently significant in all the three models of Table 3 (all ps ≤ .006). No other control variables reached significance in the logistic regression models.

4. Discussion

Ours is the first study to examine the associations of pretreatment social network features with risk of staying fewer days in treatment and treatment progress in the early stages of residential treatment for women. Our results contribute to the literature on social networks in SUD. Results showed that the number of drug users in the pretreatment social network increased the risk of staying fewer days in treatment, which supported our hypothesis. However, hypotheses regarding the relationship of general and treatment support (i.e., the role of specified pretreatment network features) to the likelihood of staying longer during the first 8 weeks of treatment and to favorable treatment status at 8 weeks post-treatment entry were not supported, as we observed no significant relationships. Network size was associated with better treatment status (i.e., not leaving treatment without significant progress). Further, contrary to our hypothesis, we found no moderation of race/ethnicity in these analyses.

The current study highlights the importance of the size of individual networks and the prevalence of drug users in it just before entering treatment. While both of these indices showed significant associations with the risk of staying fewer days in treatment in initial models, only the number of drug users remained significant in the complete model. This suggests that among women in residential treatment for SUD, the social network composition is crucial. Contact with other drug users can help to maintain drug use through increasing access to AOD and behavioral modeling, and this contact usually decreases over time with residential treatment (Min et al., 2013; Nargiso et al., 2014).

The current study also found that women who identified a higher number of important people were more likely to remain in treatment or to have left after achieving significant progress toward completing their treatment goals than those with a fewer important people. This was true even when considering the number of important people with whom women had daily contact, which was significantly associated with treatment outcome at discharge in initial models. It is important to understand what drives this relationship. It is possible that succeeding in treatment is perceived as more important for women with larger social networks, because of expectations placed on them by multiple people. Also, having more people deemed as important might mean that women receive more instrumental help with tasks they cannot complete while in treatment, thus women can focus on their progress in treatment rather than on what needs to happen in the outside world. A larger social network (of people who women deem to be important) before entering treatment might additionally increase the perceived sacrifice of choosing residential treatment if reduced contact with more people is perceived as negative. Further, a large social network might be indicative of less functional impairment before treatment entry. Last, but not least, women who are able to develop meaningful relationships might establish stronger connections with peers in treatment, which might be of benefit for the recovery process (Jun et al., 2017; Mandell, Edelen, Wenzel, Dahl, & Ebener, 2008; Warren et al., 2020). All of these ideas are speculations at the present time and future research can parse these mechanisms.

Interestingly, various social network characteristics emerged as significant predictors of unfavorable outcomes in terms of retention and clinical progress. While overall network size reduced the likelihood of dissatisfactory clinical progress, number of drug users in woman’s network increased the risk of staying in treatment for fewer days. The two outcome variables that we examined could be unrelated to each other. While treatment dropout is considered an unfavorable treatment outcome, research has shown that most treatment responders rapidly improve in the beginning of treatment and then the rate of improvement can decline, so that additional treatment does not contribute beyond the point these responders have reached “good enough” clinical progress (e.g., Barkham et al., 2006). Upon treatment discontinuation, some individuals might also plan to pursue more treatment in the future (Aslan, 2015), which we have not assessed in the current study. Further, previous research has shown that network size and number of users in network relate to various mental health and substance use indices (Stone et al., 2016; Zywiak et al., 2009); our study adds to the literature that documents the discriminant validity of these two social network characteristics.

The current study did not find any indication that increased pretreatment general or treatment social support from people women deemed to be important was related to longer treatment retention or better clinician-evaluated outcome at discharge, despite being predicted. Previous research has linked social support to a variety of favorable clinical outcomes for individuals with SUD (Stone et al., 2016). However, we do not believe that the failure to find any evidence in this study invalidates previous results. Here, we examined the association of social support provided by alters with whom the participants associated in the month before they entered the residential facility. It is possible that the social support from these social network alters is more important for treatment initiation than treatment continuation. Support received by other individuals (e.g., facility clinicians and fellow residents) might be more influential during treatment (Jun et al., 2017). Future research should examine this differential effect of pretreatment and in-treatment social support during early stages of treatment and their associations with treatment retention, clinical status at discharge, and other treatment outcomes.

Contrary to our hypotheses, we failed to observe a moderation of race/ethnicity on the relationship between social network characteristics and the likelihood of early dropout or treatment progress at discharge. The participants in this study representing various race/ethnicity groups were similar in many aspects; therefore, previous differences found might have been related to other disparities between race/ethnicity groups. The only group differences that we found in this study were: Women who identified as Hispanic were younger overall and women who identified as non-Hispanic black had more comorbid diagnoses overall. A smaller percentage of the non-Hispanic black group reported having used methamphetamines, while a larger percentage of this group reported having used cocaine in the 8 months prior to treatment entry. Last, the mean drug use composite score for non-Hispanic blacks was lower than that of the Hispanic group of women.

In this study, we controlled statistically for differences in age, comorbid diagnoses, and drug use severity. Previous research has suggested that drug of choice might be an important variable predicting dropout and treatment progress, but we did not control for drug of choice, as we did not have these data. Eighty-four percent of the sample reported some use of stimulants and post hoc analyses showed no difference in treatment progress between women who used stimulants on more days than other substances (n = 167) and the rest of the sample (p = .38). Future studies should examine if drug of choice moderates the relationship between social network variables and treatment retention.

Throughout the models, having an additional mental health diagnoses (i.e., anxiety, depression) to AUD or SUD was associated with both reduced risk of staying fewer days in treatment and leaving treatment without satisfactory progress above and beyond social network factors. While this is not the usual direction of results (e.g., Krawczyk et al., 2017), the current finding replicates another study, which showed increased treatment retention for women with mental health comorbidity compared to their male counterparts (Choi, Adams, Morse, & Macmaster, 2015). Women with mental health comorbidity might have increased motivation for treatment and readiness to seek help, especially if the criminal justice system is involved (Webster et al., 2006). Therefore, they might remain invested in their treatment. In addition, clinicians might pay more attention to women with more complex mental health issues within a residential treatment setting and support them more throughout the treatment process than those without mental health comorbidity. Future research should examine what drives the relationship between psychiatric comorbidity and longer retention and better outcomes among women with SUD.

Women who entered residential treatment voluntarily stayed in treatment fewer days than women who were mandated to treatment by either the criminal justice system or child protective services. This relationship was significant beyond the association of social network factors on treatment duration. Lack of external pressures to stay in treatment is likely to drive this finding (Perron & Bright, 2008). As seen in our study, those mandated to treatment do not differ in clinical outcomes from those who enter treatment voluntarily (Kelly, Finney, & Moos, 2005); therefore, it is not clear that longer mandates to treatment is clinically beneficial for women mandated to treatment. More research is needed on the similarities and differences of treatment experiences, outcomes, and retention among women who enter treatment voluntarily or are mandated, since some research has reported negative consequences of compulsory treatment (Werb et al., 2016).

A few limitations of the current study are worth nothing. First, the IPDA does not capture some characteristics of the individual’s relationships with her social contacts, which have previously been examined in the broader literature on social network analyses (Campbell, Cranmer, Doogan, & Warren, 2019; i.e., reciprocity, Feeney & Collins, 2015; or conflict, Fish, Maier, & Priest, 2015; or clustering, Psylla, Sapiezynski, Mones, & Lehmann, 2017; Szell & Thurner, 2013). In the current study, we focused more on the quantity and substance use characteristics of social network alters, rather than the quality of the relationship with the individual (even though we assessed social support, which might be a proxy of the relationship quality). We also limited the number of social alters to six, which might have further affected the results. Last, treatment site clinicians evaluated treatment progress at termination without using a standardized research measure. While the clinical team used clear and well-defined criteria for the discharge status determination, we did not have a way to verify internal consistency in use of these criteria across clinical teams. Future research should use a more robust measure of treatment progress to validate the results from our study.

4.1. Conclusions

The current findings suggest that more drug users in one’s network is related to increased risk of staying fewer days in treatment and that overall social network size is associated with higher likelihood of better clinical outcome at 8 weeks post-treatment entry among women with SUD who are receiving residential treatment and consented to research participation. These results highlight the importance of studying both days in treatment and clinical status, since they might be related to different aspects of the experiences for women. Further, this study emphasizes the importance of structural characteristics of women’s social network above social support, which did not relate to any treatment outcome measures. Possible clinical implications might include incorporating specific strategies for engaging women with a large number of drug users in their networks to prevent early treatment dropout and focusing on increasing social network size among women with smaller social networks at treatment entry to improve outcomes, but research needs to examine these suggestions before they can be widely implementation. This study is a first and important step toward understanding how social networks before treatment entry relate to treatment retention and clinical outcomes in early residential treatment for women with SUD.

Highlights.

1. Structural characteristics of the social networks before treatment entry might be important for treatment retention and clinical outcomes

2. Pretreatment social network characteristics have been rarely studied in relation to dropout in residential treatment for substance use disorders

3. Number of drug users in the pre-treatment social network is related to increased risk for staying fewer days in substance use disorder residential treatment among women

4. Larger social networks prior to entering SUD treatment were found to be related to reduced likelihood of leaving treatment without satisfactory progress

5. General and treatment specific social support from important people prior to entering SUD treatment was not related to women’s risk of early treatment termination or clinical progress status at discharge at 8 weeks post-treatment entry

Table Appendices

Appendix Table I.

Cox hazard-proportional regression analyses of the association between pre-treatment social network characteristics and the days to leaving treatment, right censored at 8 weeks after treatment start.¹

Independent social network variables	Hazard ratio	95% CI	p
Number of people in social network2	.53	.29~1.00	.05
Number of people in daily contact2	.80	.61~1.04	.09
Average importance rating of most important people	.67	.23~1.93	.46
Number of alcohol users in social network2	1.01	.64~1.59	.96
Number of drug users in social network2	1.58	1.02~2.44	.04
Substance involvement of social network members	1.00	.99~1.01	.65
General support	.99	.94~1.04	.66
Treatment support	.95	.87~1.04	.29
Support for AOD use	1.03	.97~1.08	.33

¹Cox hazard-proportional models were fitted for each IPDA variable separately and with control variables selected using purposeful selection procedure from age, race/ethnicity, marital status, mandating agency, pregnancy, ASI alcohol, drug, legal and family problems ASI scores, number of children in legal custody, mental health diagnosis other than SUD, AUD/SUD diagnosis, intervention randomization, and use of alcohol, marijuana, methamphetamines, cocaine/crack, opiates, and other drugs 8 months before treatment entry.

²Square root was taken before entering the regression analysis to increase normality

Appendix Table II.

Logistic regression Analyses of the association between pre-treatment social network characteristics and treatment progress at 8 weeks after treatment start (Still in treatment/Completed treatment/Left treatment with satisfactory progress vs. Left treatment without satisfactory progress).¹

Independent social network variable	Odds ratio	95% CI	p
Number of people in social network2	.35	.13~.85	.03
Number of people in daily contact2	.68	.44~.96	.05
Average importance rating of most important people	.70	.21~2.63	.58
Number of alcohol users in social network2	1.19	.66~2.20	.56
Number of drug users in social network2	1.30	.73~2.34	.38
Substance involvement of support system members	1.00	.99~1.02	.69
General support	.98	.91~1.05	.52
Treatment support	.95	.85~1.07	.41
Support for AOD use	1.05	.98~1.14	.15

¹Logistic regression models were fitted for each IPDA variable separately and with control variables selected using purposeful selection procedure from age, race/ethnicity, marital status, mandating agency, pregnancy, ASI alcohol, drug, legal and family problems ASI scores, number of children in legal custody, mental health diagnosis other than SUD, AUD/SUD diagnosis, intervention randomization, and use of alcohol, marijuana, methamphetamines, cocaine/crack, opiates, and other drugs 8 months before treatment entry.

²Square root was taken before entering the regression analysis to increase normality