Predictors of dropout from group therapy among patients with bipolar and substance use disorders

Abstract

Objective

Bipolar and substance use disorders frequently co-occur. Integrated treatment for these disorders has been shown to be effective at reducing substance use, but no study has examined attrition from dual diagnosis group therapy. The current study identified baseline demographic and clinical characteristics that predict treatment dropout among patients with co-occurring bipolar and substance use disorders.

Method

Using binary and multivariate analyses, baseline data were analyzed as part of a randomized controlled trial of integrated group therapy for bipolar and substance use disorders.

Results

Cigarette smoking, recent mood episode, and lack of a college education were strong predictors of dropout after controlling for demographic and substance use variables.

Conclusions

Given the strength of smoking as a predictor of dropout as well as the high rate of smoking among this population, a greater focus on the relationship between smoking and bipolar disorder is warranted.

1. Introduction

The high rate of substance use disorders among patients with bipolar disorder (Grant et al., 2004) and the poor prognosis associated with the co-occurrence of these disorders (Salloum and Thase, 2000) highlight the particular need for effective treatments for this population. We have developed a new treatment, Integrated Group Therapy (IGT), which has been shown to reduce substance use (Weiss et al., 2000; Weiss et al., 2007). However, no studies have examined treatment retention specifically.

In the current follow-up study to the previous report, we tested a shortened form of IGT, delivered by drug counselors. Baseline predictors of dropout from group therapy were examined among patients diagnosed with both bipolar and substance use disorders. Our aim was to identify baseline demographic and clinical predictors of attrition. Given the lack of research on treatment dropout among this dually-diagnosed population, we chose to compare a range of demographic, substance use, and mood characteristics as predictors of attrition.

2. Method

2.1 Participants and procedures

This study used data collected as part of a group therapy study for patients with co-occurring bipolar and substance use disorders randomized to either IGT or standard Group Drug Counseling. Study inclusion criteria were current diagnoses of bipolar disorder and substance dependence other than nicotine, substance use within the past 60 days, and a mood stabilizer regimen for ≥2 weeks. Exclusion criteria included concurrent group treatment, residential treatment restricting substance use, and acute danger to self or others.

Groups met weekly for 12 weeks in the evenings. Before the first session, subjects completed a pre-group interview with the group leader and signed a “contract” emphasizing the importance of consistent treatment attendance. Reminder calls were made the day before each scheduled group session. Patients missing a session without notification were contacted afterwards; transportation was arranged for patients citing travel difficulties. Dropout was defined as missing at least the last 4 groups (i.e., the last 4 weeks, following the method of Siqueland et al., 1998).

Study procedures were reviewed and written informed consent was obtained at the initial appointment. This protocol was approved by the McLean Hospital Institutional Review Board. (See Weiss et al., 2007, for more detail on methods.)

2.2 Measures

Background data were collected via a self-administered questionnaire; diagnostic information was derived from the Structured Clinical Interview for DSM-IV (SCID; First et al., 1996).

2.3 Data analysis

Stepwise logistic regression models identified predictors of treatment group dropout, with p≤.10 in an intermediate model required for retention. Data were analyzed using statistical packages (SPSS 15.0.1 and Stata/SE 8.2.).

3. Results

3.1 Demographic and diagnostic characteristics

Patients included 25 women and 36 men, 18–65 years old (mean=38.3±11.1), and primarily white (92%). They were generally well-educated (49% graduated from college), yet half (54%) were currently unemployed, and 40% had an annual household income under $35,000. Most were unmarried (54% never, 16% divorced, and 2% widowed.)

Most patients (79%) had bipolar I disorder; 21% had bipolar II or bipolar disorder not otherwise specified. Most patients (71%) met DSM-IV criteria for mania, depression, or a mixed episode during the 30 days prior to baseline, with a mean of 2.3±1.7 weeks ill. Most met lifetime dependence criteria for both drug and alcohol dependence (66%), 26% met for alcohol dependence alone, and 8% for drug dependence alone. Just over half (57%) were current smokers, consuming roughly one pack per day (mean=0.9±0.4).

3.2 Dropout characteristics

Twelve patients (19.7%) dropped out of group therapy: they were younger (31.5±11.5 vs. 40.0±10.4; t(59)=2.47, p=.02), less likely to be college educated (16.7% vs. 57.1%; χ²(1)=6.32, p<.02), and more likely to be current cigarette smokers (91.7% vs. 49.0%; χ²(1)=7.18, p<.01). Dropouts reported more weeks of mood episodes in the month prior to baseline (3.3±1.0 vs. 2.1±1.8; t(32)=−3.15, p<.01). Group dropout did not vary by race, gender, marital or employment status, household income, treatment group, type of substance or bipolar disorder, multiple substance dependence diagnoses, daily cigarette intake, history of previous individual psychotherapy, or likelihood of 12-step self-help group attendance.

3.2 Multivariate models to predict dropout

Demographic variables were entered on the first step, with less education predicting dropout (see Model 1, Table 1). Substance use variables were added next, with current smoking, as well as education, predicting dropout (see Model 2). Finally, mood was entered (i.e., the occurrence of a mood episode in the baseline month; see Model 3): smoking, lower education level, and a recent mood episode were each significant predictors of dropout in the final model. Backward elimination stepwise regression was run as a reliability check; the similarity of results for both models supported the validity of our model.

Table 1.

Stepwise logistic regression models predicting group dropout during baseline month (N = 61)¹

	Model 1	Model 2	Model 3
	Demographics	+ Substance use	+ Mood
Baseline measures	(p=.062)	(p<.001)	(p<.001)
Demographic
Age	0.9 (0.9–1.0)
Female	1.0 (0.2–4.4)
Less than college degree	5.6 (0.96–33.1)†	9.2 (1.5–55.7)*	8.3 (1.4–48.6)*
Married	0.7 (0.1–5.2)
Working	1.4 (0.3–7.6)
Substance use past month
Cigarette smoker		15.9 (1.7–152.6)*	11.5 (1.2–106.8)*
Days used >1 substance		0.9 (0.8–1.1)
Mood
Episode in past month			8.5 (0.9–84.3)†
Nagelkerke R2	25.1%	37.5%	42.0%
C statistic	0.78	0.84	0.85

^†p<0.10

^*p<0.05

¹Coefficients are odds ratios, with 95% confidence intervals.

Adding substance use variables substantially increased the explained variance (R²Δ=12.4%). Adding mood also contributed to predicting dropout (R²Δ=4.5%), with education and smoking remaining strong predictors. The odds ratios for the three significant variables in the final model were relatively similar (OR=8.3–11.5). The final model appeared to be a remarkably strong predictor of group dropout, with an area under the ROC (Receiver Operating Characteristic) curve of 0.85; the model also produced a pseudo-R² of 42.0%.

4. Discussion

We developed a predictive model of dropout from group therapy using baseline demographic, substance use, and mood variables among patients with both bipolar and substance use disorders. Cigarette smoking, a recent mood episode, and having less than a college education were the strongest predictors of treatment dropout after adjusting for demographic, substance use, and mood variables.

Lower educational attainment has previously been found to be correlated with dropout from substance abuse treatment (Sayre et al., 2002; Siqueland et al., 1998). Education is correlated with socioeconomic status, which, in turn, may be associated with greater access to resources making treatment adherence easier (e.g., geographic and job flexibility, insurance coverage; Stark, 1992). Additionally, the educational components of our treatment might have been less appealing to these patients.

Our finding that mood episode at baseline was associated with dropout could reflect the symptoms of bipolar disorder that interfere with daily functioning: impaired judgment, diminished insight, and increased engagement in pleasurable and often risky behaviors; some patients continue to experience functional, social or occupational difficulties related to their disorder even between episodes (American Psychiatric Association, 1994). It is quite possible that current or even residual symptoms made treatment completion particularly difficult for some patients.

We were surprised to find that cigarette smoking was a very strong predictor of dropout in both bivariate and multivariate analyses. High rates of nicotine dependence have been well-documented among both mentally ill and substance-dependent patients (Grant et al., 2004), although nicotine cessation is rarely a major focus of treatment for either mood or other substance use disorders (Currie et al., 2003; Richter et al., 2004). Recent research has identified a potentially deleterious relationship between smoking and mood disorders, with bipolar smokers having an increased number of suicide attempts (Oquendo et al., 2004; Ostacher et al., 2006), worse illness course (Ostacher et al., 2006), and higher illness severity, including more psychotic symptoms (Waxmonsky et al., 2005), and co-occurring anxiety, attention deficit hyperactivity, and substance use disorders (Ostacher et al., 2006; Waxmonsky et al., 2005). Our finding adds to the growing literature suggesting that cigarette smoking might be associated with a particularly negative course among patients with mood and substance use disorders.

This analysis was limited by several factors, including small sample size. Additionally, our dropout rate was much lower than that seen in community psychiatric and substance abuse treatment programs (Stark, 1992) reflecting the high level of outreach by the study staff (e.g., weekly reminder calls). Further analyses in community settings would be helpful to confirm the generalizability of our results. Smoking data were collected without diagnostic criteria; thus, we were unable to analyze length or severity of nicotine dependence. We did ask subjects to report the number of packs smoked per day, but did not find a correlation with dropout. Extensive nicotine data might enable specification of the complexities of the relationship between smoking and dropout, including identifying moderators of the relationship. Despite the small sample size, however, the final model was a remarkably robust predictor of dropout. Further, given our low dropout rate, the patients who did leave treatment probably represent a particularly difficult to treat subgroup, suggesting a need for enhanced efforts to engage and motivate patients presenting with these baseline characteristics.

This study identifies lower educational attainment, recent mood episode, and smoking as strong baseline predictors of group therapy dropout among people with both bipolar and substance use disorders, frequently co-occurring disorders with a particularly poor prognosis. This study adds to a growing body of literature suggesting that smoking may be an indicator of particularly negative outcomes among patients with bipolar and co-occurring substance use disorders (Oquendo et al., 2004; Ostacher et al., 2006; Waxmonsky et al., 2005), suggesting a need for further research on nicotine use in this population. Clinicians might take note of patients presenting with this constellation of characteristics and enhance efforts at treatment engagement and retention.