If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Corresponding author at: Peter Boris Centre for Addictions Research, St. Joseph's Healthcare Hamilton, West 5th Campus/McMaster University, 100 West 5th Street, Hamilton, ON L8P 3R2, Canada.
Peter Boris Centre for Addictions Research, McMaster University & St. Joseph's Healthcare Hamilton, Hamilton, ON, CanadaHomewood Research Institute, Guelph, ON, Canada
Investigated predictors of premature addiction treatment termination (dropout).
•
Regression revealed that drug severity and PTSD severity were significant predictors.
•
Latent profile analysis revealed four latent subgroups of patients.
•
A lower risk group had exclusively high alcohol severity.
•
A higher risk group had high illicit drug severity and comorbid psychiatric severity.
Abstract
Background
While inpatient programs are a common setting for addiction treatment, patients' premature termination is a major concern. Predicting premature treatment termination has the potential to substantially improve patient outcomes by identifying high-risk profiles and suggesting care paths that might reduce dropout. The current study examined the predictors of premature termination from an inpatient addiction medicine service.
Methods
In 1082 patients admitted to a large inpatient addiction medicine service, we used intake assessments of severity of alcohol use disorder, illicit drug use disorder, post-traumatic stress disorder (PTSD), anxiety disorders, and major depressive disorder to predict planned termination (n = 922) or premature termination (n = 160). We used two complementary analytic approaches—traditional binary logistic regression and a data-driven latent profile analysis (LPA).
Results
Binary logistic regression revealed that alcohol use severity, illicit drug use severity, and PTSD severity significantly predicted termination status, although alcohol use severity notably exhibited an inverse relationship. The LPA revealed four distinct profiles, with one profile exhibiting a significantly higher rate of premature termination and another exhibiting a significantly lower rate of premature termination. The high-risk profile was characterized by high drug severity, high comorbid psychopathology (PTSD, depression, and anxiety symptoms), but low alcohol severity. The low-risk profile was characterized by high alcohol severity, but low drug use and low comorbid psychopathology.
Conclusions
These results provide converging evidence that illicit drug severity and psychiatric severity, and particularly PTSD, were associated with premature termination. Moreover, the LPA revealed distinct latent subgroups of patients with meaningfully higher and lower risk of premature termination, suggesting that addiction services should develop strategies for identifying high-risk individuals or develop care paths for high-risk symptom clusters. Approaches that are trauma-informed or otherwise focus on the management of comorbid psychiatric conditions may be particularly appropriate for reducing premature termination.
The use of alcohol and other psychoactive substances is highly prevalent and is associated with considerable burden due to death, disability, and injury (
The global burden of disease attributable to alcohol and drug use in 195 countries and territories, 1990–2016: A systematic analysis for the global burden of disease study 2016.
); they were responsible for approximately 99.2 million (alcohol) and 31.8 million (drug use) disability adjusted life years (DALYs) globally in 2016 (
The global burden of disease attributable to alcohol and drug use in 195 countries and territories, 1990–2016: A systematic analysis for the global burden of disease study 2016.
), a sizable proportion of patients do not have successful outcomes. Therefore, understanding factors that predict treatment outcomes for patients is a high priority in clinical research (
). In addition, poor self-efficacy, relationship status, and gender have all been found to influence drug and alcohol use following addiction treatment (
). It is important to note that research has shown completion of SUD treatment to be significantly associated with decreased rates of relapse and future readmission (
In addition to helping us to understand post-treatment relapse, research should examine successful treatment completion as a critical indicator of patient outcomes. Treatment completion can be a critical first step in the process of recovery (
). Studies estimate that approximately 17–57% of individuals drop out from inpatient addictions treatment, while more general estimates suggest that dropout rates may be as high as 50% within the first month of treatment (
). The reason for this large range of outcomes may be due to the heterogeneity of inpatient treatment settings and the variability within treatment approaches, client motivation, and use of treatment modalities administered (e.g., psychotherapy, pharmacotherapy) (
). High rates of attrition are problematic for several reasons; they prevent administration of a full dose of treatment and can instill and influence treatment-related biases (i.e., expectations of symptom improvement, expectations of healthcare providers) (
As a result, an emerging body of literature has focused on factors that predict early drop-out from addictions treatment. For example, in a moderately sized sample (n = 122),
examined clusters of characteristics associated with dropout from outpatient SUD treatment. Their results suggested that individuals who were unemployed and had higher alcohol consumption were more likely to prematurely withdraw from treatment. Interestingly, they also found that this group of patients had more dependent, phobic, and schizotypal personality features than other groups of patients (
). Further, another study that examined outpatient treatment found higher retention (i.e., lower premature drop out) in individuals who were male, Caucasian, and had a high employment composite scores (as measured by length of their longest full time job, minimum monthly income, and employment prospects at the time of the study) (
). A larger study (n = 3649) that aimed to identify predictors of attrition found that younger age, greater incidence of cognitive dysfunction, more drug use, and lower alcohol use increased the probability of premature treatment termination (
). These results were echoed by a large systematic review that investigated risk factors for drop-out from addictions treatment. It highlighted lower education, younger age, cognitive deficits, low treatment alliance, and a comorbid personality disorder as the most consistently observed risk factors across the studies reviewed (
). Importantly, research has also identified factors that contribute to social stability (marital status, employment, and fewer prior arrests) as significant predictors of patient retention beyond 60 days of treatment (
With respect to program-related factors, literature suggests that higher staff ratios, greater per capita expenditure, and smaller, decentralized clinics have lower rates of attrition (
). A study investigating premature termination of inpatient addictions treatment in the UK (n = 187) found that patients with a weaker counselor-rated alliance dropped out of treatment significantly sooner than those with higher patient-counselor ratings (
noted that 91% of the literature was exclusively focused on demographic factors (i.e., age, sex, race). These factors are necessarily very coarse, not amenable to change, and likely to tell only a small fragment of a much larger story. Further, the majority of studies have relatively small sample sizes, meaning that they have relatively low statistical power. Possibly related to this, few studies have gone beyond traditional linear models in predicting premature termination. While traditional statistical methods are useful in identifying single predictors, they do not allow for the examination of symptom clusters or latent subgroups that are differentially related to program attrition.
The current study aimed to address a number of these limitations. Specifically, the study sought to identify correlates of premature termination from an inpatient addiction medicine service (AMS), using a large sample of more than a thousand patients. Beyond demographic characteristics, the study assessed common comorbid psychopathology (depression, anxiety, and posttraumatic stress disorder [PTSD]) and severity of SUD in predicting premature termination. We used two main analytical strategies—logistic regression and latent profile analysis (LPA)—to predict premature termination. These two strategies are complementary insofar as one is a traditional linear variable-centred approach (e.g., logistic regression) and the other is a person-centred approach that seeks to determine whether latent subgroups of patients are present (e.g., LPA). The former examines the linear relationships between predictors and outcomes, whereas the latter delineates unobserved configurations of correlations among key variables to ascertain underlying clusters of individuals, and then examines those clusters in relation to the outcome. As a heuristic, a variable-centred analysis can be thought of as a mean-level grouping strategy or “one-size-fits-all” approach, whereas a person-centred approach can be thought of as a pattern-based analyses or “which-size-fits-best” approach.
2. Methods
2.1 Participants
Participants were 1082 individuals admitted to a 105-bed inpatient AMS located in a larger mental health and addictions treatment centre in southwestern Ontario, Canada. The program offered group-based treatment that was 35 days in length to adults aged 19+ with alcohol and/or substance use disorders and specialized programming (56 days in length) for patients with concurrent post-traumatic stress disorder (PTSD). Overall, the AMS admits ~1000 patients each year and uses an abstinence-based approach to recovery, which is informed by 12-step facilitation therapy. Treatment is provided by a multidisciplinary team comprising physicians with certifications in addictions medicine and registered addictions counselors. Programming is paid for through public (e.g., Ontario Health Insurance Program) or semiprivate/private insurance, and direct payment.
Among the sample, 922 patients (age = 44.12 (11.34) years old, 66% male) completed the program as planned and 160 patients (44.57 (11.25) years old, 61% male) discharged early from the program (i.e., premature discharge; discharged home unplanned or signed out against medical advice). Participant characteristics can be found in Table 1.
Table 1Treatment success rate and participant characteristics across planned vs. prematurely discharged patients. Values reflect mean (standard deviation), mode, or percentage response, and contrasts reflect t-tests or χ2 tests.
All patients (N = 1082)
Planned termination (n = 922)
Premature termination (n = 160)
p
Age
44.12 (11.34)
44.57 (11.26)
41.52 (11.50)
0.001
Gender
65% male
66% male
61% male
0.29
Marital status
33% never married
33% never married
42% never married
0.03
Education
56% university/college
57% university/college
50% university/college
0.09
Employment prospects
67% employed/seeking
79% employed/seeking
57% employed/seeking
<0.001
Length of stay
39.10 (8.77)
23.62 (13.67)
<0.001
Alcohol use severity (ICD-9)
6.85 (3.92)
7.04
5.75
<0.001
Drug use severity (DUDIT)
16.83 (16.78)
15.8
22.76
<0.001
Depression severity (PHQ-9)
14.03 (7.41)
13.65
16.19
<0.001
Anxiety severity (GAD-7)
11.38 (6.30)
11.03
13.36
<0.001
Trauma severity (PCL-5)
36.26 (22.70)
34.85
44.39
<0.001
Notes: ICD-9: International Classification of Diseases, 9th Edition; DUDIT: Drug Use Disorders Identification Test; PHQ-9: Patient Health Questionnaire-9; GAD-7; Generalized Anxiety Disorder 7 Item Scale; PCL-5; Posttraumatic Stress Disorder Checklist (DSM-V).
Patients completed an intake assessment battery to obtain information regarding symptomatology associated with mood, anxiety, and substance use. Patients completed the paper-based assessment battery as part of the standard clinical practice within the first seven days of admission and served a primary goal of informing patient care. Assessments were self-report and participants completed them between October 19, 2015, and April 18, 2017. We later transcribed ata for research purposes. We obtained discharge status via program administrative data, which indicated whether the patient was discharged as planned (i.e., standard completion) or unplanned (i.e., premature termination). We requested both clinical and administrative data for analysis via an approved research protocol from the Regional Centre for Excellence in Ethics, Research Ethics Board in Guelph, Ontario, Canada.
2.3 Intake assessment measures
We assessed symptoms of major depressive disorder (MDD) using the Patient Health Questionnaire (PHQ-9;
), which is a brief, self-report, nine-question measure. Each question is scored from 0 (not at all) to 3 (nearly every day), yielding a maximum score of 27, with the following clinical ranges: 5 = mild; 10 = moderate; 15 = moderately severe; 20 = severe depression. We assessed symptoms of anxiety disorders using the Generalized Anxiety Disorder-7 (GAD-7;
), which is a brief, self-report, 7-item measure. Each item is rated from 0 (not at all) to 3 (nearly every day) and a composite score out of 21 is generated, with scores of 5, 10, and 15 reflecting mild, moderate, and severe anxiety, respectively. We assessed symptoms of PTSD using the Posttraumatic Stress Disorders Checklist for DSM-5 (PCL-5;
) to characterize traumatic exposures descriptively. Very high rates of traumatic exposure were present and only 5% reported no exposure to any events on the LEC-5. Consistent with this, the mean score on the PCL-5 (Table 1) exceeded the recommended cut-off of 33. We assessed self-report alcohol use disorder using the alcohol items from the Psychoactive Substance Use Module from the International Classification for Diagnosis (ICD)–10 Symptom Checklist for Mental Disorders (
). This measure consists of 11, yes/no questions to determine the presence of symptomology that is consistent with alcohol use disorder. We assessed other psychoactive drug severity using the Drug Use Disorders Identification Test (DUDIT), which is an 11-item measure of drug use frequency and severity (
Journal of substance abuse treatment the psychometric properties of the drug use disorders Identi fi cation test (DUDIT): A review of recent research ☆.
We conducted prospective chart analysis of 1143 individuals, of whom we identified 1082 as having completed the assessment battery (95%) and included in the final sample. We used independent sample t-tests to compare continuous variables across patients in the premature discharge and planned completion groups. We compared categorical variables using the Pearson Chi-squared tests. We completed bivariate correlations among all measures to examine zero-order associations and test for potential multicollinearity. First, for the traditional linear analysis, we included demographic and clinical variables that demonstrated statistically significant differences between the two groups via bivariate analyses as covariates and clinical predictors (respectively) in a logistic regression model. We excluded statistically nonsignificant variables to create a final, more parsimonious, model. Second, we completed an LPA to investigate the association between demographic and clinical variables and program completion status, as well as to identify subgroups that may be differentially related to premature discharge. We used the following variables to classify patients: (1) depressive symptomatology; (2) anxious symptomatology; (3) trauma-related symptomatology; (4) severity of alcohol use; and (5) severity of drug use. We used an LPA over a latent class analysis because the variables of interest were continuous, rather than categorical. Furthermore, although cut-off scores are available for the measures, using dichotomized variables would artificially truncate variability and the associated conditions are dimensional in nature (e.g.,
). We used the following indices to examine goodness of fit: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample size adjusted BIC, Lo-Mendell-Rubin test (LMR), and entropy. In this approach, a smaller AIC and BIC represent a parsimounous solution and thereby a better model fit. We used the LMR to compare whether a k profile solution results in a better fit compared to a model with k-1 profiles. Finally, entropy represents the model's overall classification quality with values closer to 1, suggesting less entropy and thus better model classification. We examined profile assignment probability of the optimal class solution to investigate precision of group classification. We completed data analyses using SPSS v.24, Mplus v.7.0, and R v. 3.4.4 statistical software.
3. Results
3.1 Bivariate analyses
Among demographic variables assessed, we found age, marital status, employment prospects, and program upon discharge to be significantly different between groups (Table 1). We found no significant differences in sex between groups (p > 0.05). We found significant differences between groups for all investigated clinical variables, including alcohol use severity, drug use severity, depressive symptomatology (PHQ-9), anxious symptomatology (GAD-7), and trauma-related symptomatology (PCL-5) (p < 0.001). A bivariate correlation matrix (Fig. 1) highlighted strong positive associations between the GAD-7 and PHQ-9 (r = 0.82, p < 0.05), and PCL-5 and both the PHQ-9 (r = 0.67, p < 0.05) and GAD-7 (r = 0.68, p < 0.05). Interestingly, alcohol use severity was negatively correlated with drug use severity (r = −0.41, p < 0.05). Associations among other variables differentiating program completion status varied from negligible to moderate magnitude.
Fig. 1Heatmap of associations (zero order correlations) among the measures used. [rs > |0.09, p < 0.05].
Notes: Drug Use Severity was measured by the Drug Use Disorders Identification Test; Alcohol Use Severity was measured by the: International Classification of Diseases, 9th Edition; Trauma Symptom Severity was assessed with the Posttraumatic Stress Disorder Checklist for DSM-5; Anxiety Symptom Severity was assessed with the Generalized Anxiety Disorder 7 Item Scale; Depression Symptom Severity was assessed with the Patient Health Questionnaire-9.
We entered clinical and demographic variables that differed significantly between groups (as demonstrated in the bivariate analyses) into a single logistical regression model. Alcohol use severity (ß = −0.069, p = 0.004), drug use severity (ß = 0.011 p = 0.062), trauma symptomatology (ß = 0.015, p < 0.001), and employment prospects (ß = −0.875, p < 0.001) all emerged as significant predictors of group termination status. These variables predicted 85.06% of group status correctly (Cox and Snell R2 = 0.059, Nagelkerke R2 = 0.105) (Table 5).
3.3 Latent profile analysis
While we evaluated five latent profiles, we deemed a four-profile solution to be the optimal profile solution based on the following indicators: (1) superior AIC and BIC relative to two, three, and five profile solutions; (2) the highest entropy value of all five solutions; and (3) Lo-Mendell-Rubin tests demonstrating superior model fit compared to the two- and three-profile solution (Table 2). In other words, the four-profile solution was superior to the two- and three-profile solutions on multiple measures, and better than the five-profile solution, as well as best overall in terms of entropy. Furthermore, four distinct and theoretically interpretable profiles emerged, further supporting the interpretation (Fig. 2).
Table 2Model fit statistics across latent class solutions.
The average latent profile probabilities for most likely profile membership is shown in Table 3. Probabilities were approaching 1.0; therefore, we considered them to be very high. We characterized Profile 1 (27.7%) by comparably higher levels of alcohol use, the lowest levels of drug use, and the lowest levels of psychopathology (depressive, anxious, and trauma-related symptomatology) among profiles. As a result, we designated Profile 1 as High Alcohol/Low Psychiatric Severity and exhibited the highest program completion rate, encompassing 92.8% of individuals with planned discharge, and 7.2% premature termination. We characterized Profile 2 (27.6%) by the lowest levels of alcohol use, high levels of drug use, and highest levels of psychopathology. Consequently, we designated this profile as High Drug Use/High Psychiatric Severity. Individuals in this profile exhibited the highest rates of premature termination (23.1%), and lowest levels of planned discharge (76.9%) among profiles examined. We characterized Profile 3 (28.6%) by the highest levels of alcohol use, lowest levels of drug use, and comparably high levels of psychopathology. Therefore, we designated this profile as High Alcohol/High Psychiatric Severity; accounting for 14% of patients who prematurely terminated treatment and 86% in the planned discharge group. Finally, Profile 4 (16.1%) demonstrated the lowest levels of alcohol use among all patients, comparably high levels of drug use, and low levels of psychopathology. We designated this class as High Drug Use/Low Psychiatric Severity. Approximately 15% of patients in this profile prematurely terminated treatment, whereas 85% had a planned discharge.
Table 3Average latent profile posterior probabilities for most likely latent class membership N (Row) by latent class C (column).
Finally, results of χ2 tests reve3aled that the four distinct profiles had significant differences overall in terms of discharge rates (Table 4). Specifically, Profile 1 and Profile 2 were significantly different from the other profiles (Table 4), with a significantly lower rate of discharge for Profile 1 and a significantly higher rate of discharge for Profile 2.
Table 4Latent profile analysis class comparisons based on rates of premature termination.
Table 5Clinical predictors of premature discharge using a binary logistic regression.
B
SE
Odds ratio
Wald χ2
P-value
Constant
−1.530
0.298
0.216
26.386
0.000
Employment prospects
−0.875
0.186
0.417
22.135
0.000
Alcohol severity (ICD-9)
−0.069
0.024
0.934
8.192
0.004
Illicit drug severity (DUDIT)
0.011
0.006
1.011
3.492
0.062
PTSD severity (PCL-5)
0.015
0.004
1.015
12.809
0.000
Notes: ICD-9: International Classification of Diseases, 9th Edition; DUDIT: Drug Use Disorders Identification Test; PCL-5; Posttraumatic Stress Disorder Checklist (DSM-V).
The current study examined the predictors and profiles of premature (unplanned) discharge from an inpatient SUD treatment service, using two parallel analytic methods—a traditional binary logistic regression and data-driven latent profile analysis. The binary logistic regression highlighted that alcohol use severity, drug use severity, trauma symptomatology, and employment status all emerged as significant predictors of program completion status; predicting 85.06% of discharge/termination status. A parallel data-driven approach conducted to investigate further associations between demographic and clinical variables and program completion status found four distinct profiles of patients with differing levels of alcohol and drug use and psychopathology. Of these profiles, the best predictor of premature discharge was a profile of patients that endorsed high drug use, high psychopathology, and low alcohol use (High Drug Use/High Psychiatric Severity)—this group encompased 23.1% of individuals who met criteria for premature discharge. Alternatively, the profile of patients endorsed the highest planned discharge and lowest premature termination experienced high alcohol use, low drug use, and low psychopathology (High Alcohol Use/Low Psychiatric Severity). To our knowledge, there are no other studies that have used latent profile analysis to investigate the predictors of premature termination among patients seeking inpatient addictions treatment.
Our results highlighted four distinct subgroups characterized by differences in symptom severity of psychopathology and drug and alcohol use. The High Drug Use/High Psychiatric Severity profile was associated with the greatest premature termination of treatment. We hypothesize that this may occur for several reasons. First, the cognitive dysfunction and neurological consequences associated with more severe drug use may make treatment engagement more challenging and decrease patient motivation. Furthermore, patients with high levels of psychopathology may experience greater levels of negative emotionality (
), both of which have been identified as predictors of substance use relapse and may influence treatment completion. As a result, this profile may also warrant additional treatment resources that go beyond those that have historically been allocated for the treatment of AUD, specifically. Interestingly, individuals who endorsed high alcohol use and low drug use and psychopathology (High Alcohol Use/Low Psychiatric Severity) experienced the highest rate of planned termination (lowest premature termination). This suggests that individuals who solely endorsed high alcohol consumption and low consumption of other substances demonstrated the most favorable treatment outcome out of the four profiles of patients studied. Therefore, highlighting that AUD, without co-existing SUDs or other psychopathology, may be well suited toward this particular inpatient addiction treatment program/model; or AUD may present as a less complex disorder for which to provide treatment, compared to the treatment of other SUDs. This is consistent with the literature, which suggests that individuals with co-occurring severe mental illness and SUD will have adversely affected treatment outcomes and treatment course due to the additional psychological burden that their mental illness poses (
). In addition to this, the addictions program may not have adapted itself to support the increased complexity of clients over time. This highlights a need for ongoing program adaptation to meet the needs of clients with multiple and complex substance use challenges.
Studies using traditional statistical methods (i.e., binary logistic regression, general linear models) to investigate predictors of premature program discharge or treatment retention find similar results to those discussed in our study. Studies using binary logistic regressions have identified “labor problems” and unemployment as a predictor of treatment drop out; both of which can be conceptualized under the broader domain of psychosocial stability, which has been found to positively contribute to treatment completion (
). Individuals who are employed may be less likely to suffer from homelessness and have increased motivation to complete treatment (because of their motivation to remain employed) (
). In the context of this study, employment and employment insurance may have enabled many individuals to have the cost of their treatment covered in this facility. Therefore, our results confirm an important link between employment status and treatment completion, which may also highlight the predictive value of factors that contribute to psychosocial stability.
Our results also highlighted the role of trauma, and drug and alcohol use severity as independent predictors of premature program discharge. Several studies have highlighted the role of substance use as a maladaptive coping strategy to reduce symptoms associated with trauma and PTSD (
). Trauma has also been associated with an increased burden of disease that may warrant additional resources beyond those that drug and alcohol use treatment facilities offer (
). High drug use and lower levels of alcohol use also emerged as independent predictors of treatment termination. Drug use severity has been associated with higher rates of psychopathology and increased neurological sequelae, both of which may pose challenges relate to treatment engagement (
). Further, withdrawal from illicit substances often results in significant psychological and physical symptoms, which can create further challenges for SUD treatment (
The current study should be considered in the context of its strengths and limitations. Among its strengths are its large sample size; inclusion of predictors beyond demographic risk factors; and that it is the first study to use LPA to identify unobserved patient subgroups associated with premature discharge from an inpatient SUD treatment program. Specifically, using LPA allowed for the characterization of profiles of patients that appear to differentially complete or not complete treatment. Such patterns are useful for highlighting groups of patients who may be more vulnerable for premature discharge and may require additional resources and support to complete treatment. Further, to our knowledge, this is the largest study to date to examine premature discharge in this patient population. This large sample size allowed for the use of novel statistical methods, such as the LPA, in addition to traditional statistical approaches, such as logistic regression. The parallel analytic strategies allowed us to elucidate different profiles of individuals that leave treatment early; binary logistic regression allowed for the identification of independent predictors that contribute to premature discharge/termination, thereby providing a complementary comprehensive analysis of factors and group-based characteristics/profiles that contribute to premature discharge or termination.
In the current study patients completed a relatively limited number of assessments; individuals completed one assessment within the first week of admission to the program. Further, this assessment consisted of self-reported measures that allowed clinicians to obtain data from patients quickly and efficiently, but they did not complete objective diagnostic interviews or specific behavioral tasks. This approach may be affected by a patient's experiential state or introspective ability during the first 7 days of treatment, which may be a particularly difficult period. Relatedly, the self-report measures cannot disambiguate overlap in symptoms, such as negative affectivity attributable to depressive symptoms versus PTSD symptoms or sleep disturbance resulting from withdrawal versus nightmares. Another factor that may have influenced the results of the self-reported assessments include social desirability, which we did not assess. Our assessment also did not account for the presence of notable personality features or the presence of personality disorders, which are highly prevalent in individuals with SUDs (
). Also, the term premature termination encompasses substantial clinical heterogeneity, including both patients who may have chosen to leave against medical advice and patients who the program staff discharged for reasons such as violating the treatment's substance use policies (e.g., bringing drugs on the unit) or exhibiting aggressive or disruptive behavior. Others no doubt leave treatment for very personal reasons, such as work or childcare responsibilities, or the program simply does not “fit” their perceived needs. Thus, heterogeneity within the definition of premature termination itself creates ambiguity and made the dependent variable necessarily imprecise. Finally, this study was conducted in a semiprivate hospital and thus may reflect a subset of the population that has the means to cover the cost of treatment or was able to have the cost of treatment covered through their employer or provincial healthcare funds.
A final limitation of our study is the generalizability of our findings to other clinical settings. In this case, the SUD treatment was built upon a 12-step facilitation approach—the traditional “Minnesota Model”—and has a principal focus on abstinence. This is similar to many other treatment settings, but by no means all. For example, a growing body of literature has highlighted the benefits of harm reduction approaches (
). Furthermore, as this model of treatment is variably suited toward different patients, the heterogeneity in discharge and treatment outcomes may reflect the efficacy of this type of treatment among different patient profiles. In other words, premature treatment termination is not a monolithic construct, but reflects the interaction of the treatment program and patient characteristics. Predictors of premature termination may differ for treatment programs with different orientations. Moreover, the fit between patient perspectives and program orientation is a critical feature worth examining as a predictor in future studies.
5. Conclusion
Our results indicate that high illicit drug severity and high comorbid psychopathology are associated with the least favorable treatment outcomes (highest rate of premature termination). As such, our results highlight the need for SUD treatment programs to call on greater resources and implement greater management of comorbid psychopathology. Notably, two of the four profiles identified had high psychopathology, emphasizing the high rates of concurrent disorders in this population. This further underscores the need for resourcing treatment paths that focus on management of concurrent disorders and high levels of comorbidity. In this capacity, mindfulness-based relapse prevention may be useful to provide additional support and resources for individuals with concurrent disorders. Research shows that this therapeutic approach is efficacious in reducing addictive behaviors and symptoms of craving and may be a useful care path for individuals with concurrent disorders (
). Within comorbid psychopathology, trauma severity was significantly implicated, underscoring the need for a trauma-informed approach in addiction treatment. Employment status emerged as a strong predictor of premature treatment termination; this association is well-established in the literature on treatment completion and highlights the importance of addressing factors pertaining to psychosocial stability. In sum, our results implicate a number of significant predictors of premature termination that may inform treatment targets to optimize the likelihood of successful recovery.
Author statement
All authors have substantively contributed to the manuscript. None of the original material contained in the manuscript has been submitted for consideration nor will any of it be published elsewhere, except in abstract form in connection with scientific meetings. With regard to conflicts of interest, JM is a principal in BEAM Diagnostics, Inc., but no BEAM products were used in the data collected. No other authors have declarations. All appropriate institutional ethical approvals were in place and all necessary disclosures have been made.
Acknowledgments
We thank Dr. Harry Vedelago and Ms. Wendy Woo for their invaluable contributions to data collection, to the other clincians at the Homewood Health Addiction Medicine Service, and to Dr. Ben Goodman for his contributions to early data analysis. We are very grateful to the patients for their participation in this research.
Funding
This work was supported by charitable donations to Homewood Research Institute, a registered charity in Canada, and by the Peter Boris Chair in Addictions Research.
References
American Psychiatric Association
Diagnostic and statistical manual of mental disorders.
The global burden of disease attributable to alcohol and drug use in 195 countries and territories, 1990–2016: A systematic analysis for the global burden of disease study 2016.
Journal of substance abuse treatment the psychometric properties of the drug use disorders Identi fi cation test (DUDIT): A review of recent research ☆.
30333-0&id=ga1.jpg){kind=link}
30333-0&id=gr1.jpg){kind=link}
30333-0&id=gr2.jpg){kind=link}