Methods
Data
Data for this secondary analysis were extracted from the Drug and Alcohol Treatment Information System (DATIS;
www.datis.ca). DATIS is a centralized information system that collects data on client characteristics and service use from publicly funded addiction treatment agencies in Ontario [
39]. Participating agencies (approximately 170) provide a mix of outpatient, residential, and withdrawal management services, which are covered through the province’s universal health insurance program and accessed free of charge by residents. DATIS does not collect information from services provided in hospitals, private treatment clinics, by physicians (e.g., in primary care settings), or by mental health care providers. Self-help groups (e.g., Alcoholics Anonymous/Narcotics Anonymous) are also excluded.
DATIS utilizes a web-based platform for data entry and management. Service providers enter data at admission after the initial appointment, including sociodemographic information, substance use, and other admission details. The probabilistic matching algorithm within Oracle (UTL_MATCH) was used to identify and link records across clients in the database, based on provincial health card number, gender, first and last name, last name at birth, and date of birth [
40]. DATIS staff conduct systematic data quality checks and collaborate with the service providers to review data annually.
We extracted de-identified data from the 36 agencies that offered integrated programs. A number of agencies with integrated programs were excluded because they did not use the DATIS clinical-tracking module to record outpatient visit dates, needed to quantify treatment participation (
n = 7 of 36). Study data from 29 agencies cover the seven-year period from April 1, 2008 to March 31, 2015 (
N = 7352 treatment episodes for 5162 women). All participants are women who were pregnant or parenting children under 6 years old at admission. A treatment episode was defined as a series of one or more visits separated by intervals of less than 60 days, i.e., a 60-day service-free period indicated the start of a new episode [
18,
41]. We selected this seven-year study period for the broader evaluation of treatment processes and outcomes, as it corresponds to a period of stable funding for this suite of programs. In addition, by 2008, agencies had transitioned to using a web-based data entry and management for DATIS.
The institutional Research Ethics Boards of the Centre for Addiction and Mental Health and Ryerson University approved the study.
Measures
We calculated three episode-level measures of treatment participation: first visit interval (number of days between the first and second visit), retention (number of days between the first and last visit), and intensity (number of visits per episode). By definition, episodes that consisted of only one visit are excluded from the calculation of first visit interval. Single-visit episodes were assigned values of 0 for retention and 1 for intensity.
Client-level measures included age (continuous), education (less than high school vs. high school or more), employment status (employed full- or part-time vs. unemployed or not in the labour force), social assistance (received provincial income or disability support vs. no), marital status (single, widowed, divorced, or separated vs. married or partnered), and pregnancy status (yes vs. no or unsure, where unsure includes women who reported possibly or unknown). Involvement with the legal system was captured as a dichotomous variable (yes vs. no, where yes referred to awaiting trial or sentencing or being on probation or parole at the time of admission). Treatment mandates were based on whether a third party required the client to enter treatment, coded as: no mandate, legal mandate (a choice between treatment or jail, or treatment as a condition of probation or parole), child protection mandate, or mandate from an employer or school.
For each treatment episode, clients can identify up to 5 problem substances (i.e., substances that they want to address during treatment) from a list of 17 substance types. For each substance that they endorse as a problem, they are then asked to indicate how often they used it in the past month. We created dichotomous indicators to denote problems with alcohol, stimulants (powder cocaine, crack, other amphetamines including methamphetamine), cannabis, and opioids (prescription opioids, heroin, opium, and over-the-counter-codeine), which were the top 4 problem substances that women reported. Past-month frequency of use of problem substances was coded as: none, up to twice weekly, three or more times per week, and binge use (referring to periodic excessive use). For those who reported multiple problem substances, the highest frequency reported was used. Past-year non-medical injection drug use was coded as yes vs. no. All variables were self-reported by clients.
Analysis
We described the sociodemographic, substance use, and admission characteristics of women admitted to integrated programs, and calculated summary statistics for first visit interval, retention, and intensity. We used multilevel regression to model the predictors of the three participation variables, all of which are positively skewed count variables with no negative values. In a situation of positive skewness and variance much larger than the mean, using a statistical model that assumes the presence of normally distributed residuals will often be inappropriate because it can lead to incorrect confiden
ce intervals and
p values [
42]. Instead, regression models for count data are usually more appropriate, e.g., Poisson and negative binomial models. We examined the unconditional means and variances of the participation variables and decided that count regression models could be suitable. To determine the optimal model form, we calculated model-predicted probabilities for outcome values of 1–10 (0–10 for retention) from Poisson and negative binomial regressions, including zero-inflated versions [
43], and plotted them against observed probabilities. For all 3 outcomes, predicted probabilities from the negative binomial models best resembled observed probabilities.
PROC GLIMMIX in SAS v.9.3 was used to run multilevel negative binomial regression models, accounting for the clustering of clients within treatment agencies. The first treatment episode per person was selected for analysis. Predictor variables were entered as fixed effects, with a random effect for agency. Variables with statistically non-significant coefficients were removed from the models (alpha = 0.05). Changes in model fit after deleting fixed effects were examined using the − 2 loglikelihood (−2LL), Akaike information criterion (AIC), and Schwarz Bayesian information criterion (BIC), for which lower values indicate better fit.
Missing values ranged from 0 to 4.2% among the predictor variables (Table
1). Records with missing values were excluded from the regression models (final n for each model shown in the tables).
Table 1
Characteristics of women attending integrated programs in Ontario (2008–09 to 2014–15)a
Education: |
Less than high school | 2332 | 45.2 |
High school or more | 2677 | 51.9 |
Missing | 153 | 2.9 |
Receiving social assistance: |
No | 1645 | 31.9 |
Yes | 3296 | 63.9 |
Missing | 221 | 4.2 |
Employment status: |
Not employed | 4331 | 83.9 |
Employed full or part time | 732 | 14.2 |
Missing | 99 | 1.9 |
Marital status: |
Married or partnered | 1622 | 31.4 |
Not married | 3491 | 67.6 |
Missing | 50 | 1.0 |
Pregnant at admission: |
No, possibly, or unknown | 4168 | 80.7 |
Yes | 974 | 18.9 |
Missing | 20 | 0.4 |
Problem substance(s): b |
Alcohol | 2259 | 43.8 |
Stimulants c | 2124 | 41.2 |
Cannabis | 1980 | 38.4 |
Opioids d | 1617 | 31.3 |
Past-year injection drug use: |
No | 4591 | 88.9 |
Yes | 551 | 10.7 |
Missing | 20 | 0.4 |
Maximum frequency of substance use, past 30 days: |
Binge use | 148 | 2.9 |
3 times per week to daily | 2164 | 50.6 |
Up to 2 times per week | 857 | 16.6 |
No use | 1466 | 28.4 |
Missing | 77 | 1.5 |
Legal system involvement: |
No | 3798 | 73.6 |
Yes | 1247 | 24.2 |
Missing | 117 | 2.2 |
Treatment mandate: |
None | 3233 | 62.6 |
Legal system | 221 | 4.3 |
Child protection services | 1444 | 28.0 |
Employer or school | 143 | 2.8 |
Missing | 121 | 2.3 |
Discussion
The goals of this study were to describe the population of women attending integrated programs in Ontario, and to evaluate levels and predictors of participation in treatment. Overall, this population of women was facing numerous barriers to accessing the resources and opportunities needed for health: in addition to being pregnant or new mothers experiencing problematic substance use, most were unemployed, on social assistance, and single. Despite this, programs appeared to be able to successfully engage most women, once they were admitted to treatment. Although rates of treatment participation did vary across subgroups defined by sociodemographic and admission characteristics, effect sizes tended to be small on average, providing little evidence in general of sociodemographic inequities in participation.
The high rates of participation among these integrated programs are encouraging. Although we were unable to identify the reasons underlying the high participation rates with these administrative data, findings from other parts of our evaluation of these programs provide further context. Coordination across agencies and sectors at the levels of service delivery and policy was seen to form a key part of what constitutes effective integrated care for this population [
35], and may contribute to how these programs are able to maintain engagement among the women who access them. Qualitative investigation of women’s perspectives of these programs revealed the central role played by counsellor support for the emotion regulation and executive functioning features of the therapeutic relationship [
38] – components that have been found elsewhere to link with positive outcomes [
44,
45]. Factors such as multi-sectoral service coordination and therapeutic supports for emotion regulation and executive functioning may be particularly important for pregnant and parenting women who are accessing substance use services, given that they face numerous social and structural barriers to health (e.g., poverty, substance-related stigma, gender discrimination) [
46‐
49].
The programs in our study were open-ended, rather than having a designated length, and we found that women continuously attended for an average of 125 days, or 4 months. This is comparable to lengths of stay for integrated programs reported elsewhere [
1]. Further, we extend the literature by including counts of visits and the length of time between the first and second visit. In this suite of programs, women attended appointments about once a week for the 4 months that they were engaged. In the absence of data on client outcomes, it is difficult to speculate on the clinical significance of visit frequency for this population. There is some evidence from the mental health literature that weekly sessions of psychotherapy are associated with faster improvements in mental health (no data on specific diagnoses were available in this study) [
50]. Weekly sessions have also been recommended in clinical guidelines for psychotherapeutic interventions for anxiety disorders [
51,
52].
Studies have shown that a shorter interval between visits early on in a treatment episode is associated with a higher rates of engagement or retention (Acevedo et al., 2015b; Lee et al., 2012). In the present study, the rate of drop out after a single visit was 14%, which is considerably lower than the 23–50% reported elsewhere in standard treatments [
53,
54]. Maintaining participation in treatment is a perennial challenge in substance use treatment generally [
55,
56], and high rates of drop-out in the initial weeks of treatment is one signal that programs are not meeting the needs and/or expectations of clients. To the extent that longer retention is associated with positive outcomes in the longer term (e.g., reduced substance use, arrests, and incarceration) [
15‐
17,
57], then such measures offer up important information on program performance. Our findings suggest that these integrated programs have achieved a fair degree of success in at least engaging women in services after admission. Further work is planned evaluating the link between these indicators of engagement and maternal and child health outcomes.
Examining the predictors of participation provides insight into variation across the treatment population, and whether additional efforts are required to meet the needs of specific subgroups. Adjusting for substance use and other admission characteristics, older age consistently predicted better participation. Women also stayed in the program longer and attended more visits if they had a high school diploma or were pregnant. These findings differ from what has been reported previously in that high income, being married and unemployed were associated with longer retention in women [
31]. In the present study, the magnitude of the association between these characteristics and participation outcomes was small (i.e., incident rate ratio < 2.0) [
58]. Nonetheless, as systems and services continue to evolve to support women, results suggested that younger women and women with lower education may need additional supports for participation. Further, pregnancy at admission predicted longer retention and greater intensity, yet only a small minority of women was pregnant when they entered treatment. Given the potential for better maternal and child outcomes associated with earlier engagement in integrated programs (including prenatal care and support for the social determinants of health, as well as substance use treatment), additional outreach efforts may be warranted to engage women while they are pregnant [
11,
46]. There is a need for qualitative studies of the ways in which barriers to care are experienced across the population, attending to the intersections between identities, and how these impact on participation on integrated treatment (e.g., [
59,
60]). Further, systematic quantitative exploration and statistical modeling of the impacts of structural violence at multiple levels (individual, community, and population) is also needed to inform the development of policies and practices that both promote health and reduce health disparities.
Our findings join a growing body of research in supporting the use of integrated (comprehensive and holistic) service models for women who have problematic substance use, and suggests that efforts to scale up women-focused programs are warranted [
1,
7,
9‐
13]. As noted earlier, with only 36 such programs operating across the province, most Ontario communities do not offer integrated substance use services for pregnant and parenting women. National studies in Canada have identified gaps in the capacity of substance use treatment agencies to offer comprehensive services addressing maternal and child health [
61]. With a narrowing gender gap in rates of substance use and related problems in many countries [
62‐
66], effective programming for women has myriad public health implications for women’s health and well-being, as well as fetal and child development. There is a need for dedicated attention and resources to support the evolution of substance use service systems as they work to ensure that they are able to address the unique contexts of substance use among women.
In this study, mandates from the legal system or child protection services tended to be associated with lower participation. Specifically, both types of mandates were associated with a prolonged interval between the first and second visits, and legal mandates were associated with a lower overall number of visits. Previous findings on the association between mandates and retention are equivocal: some studies have reported that legal and employer mandates are associated with prolonged retention [
26,
67,
68], while others have found that mandates are associated with higher dropout rates [
69,
70]. Research into the effectiveness of mandated treatment has emphasized mandates from the legal system; however, only 4.3% of the women in this treatment population were mandated to treatment by the legal system. There is limited work examining mandates from child protection services, and their impact on treatment processes and outcomes. Given the key role that such mandates play in promoting treatment entry in this population (i.e., over one in four women in this population was mandated through child protection services), these issues deserve further attention.
Strengths of this study are its population focus and the inclusion of a suite of integrated programs embedded within a broader system of psychosocial treatment for substance use. That said, the data source excluded private treatment and substance use services received outside of the publicly funded system of specialized psychosocial services. The measures of sociodemographic characteristics and substance use were self-reported and are, therefore, subject to potential reporting and recall biases. As with any secondary analysis of administrative data, not all potentially important predictor variables were available; specifically, we lacked measures of rural versus urban location, mental health problems, previous treatment experiences, race and ethnicity, and poly-substance use (over and above the use of substances self-reported to be causing problems). For the subset who were pregnant at admission, we have no information on how far along the women are in their pregnancies. There was also no information on family structure or custody. Previous work reported that having more than two children predicted earlier drop out from treatment [
71]. Finally, because this study is based on data from a treatment system, we are unable to address issues of access to integrated programs. Although we found only weak associations between sociodemographic variables and participation once women had entered these programs, it is nonetheless possible that inequities exist across population subgroups in access to services in the first place.