At pretreatment, t tests and chi-square tests showed that the two groups did not differ on gender, age, number of years of education, use of antidepressant medication, number of previous treatments for alcohol abuse, severity of alcohol dependence on the ADS, quantity and frequency of alcohol use, and measures of depressive symptoms and mood. Likewise, the two groups did not differ on baseline measures of dysfunctional attitudes, pleasant events, and self-efficacy. Ten participants in the CBT-D condition and 9 participants in the RTC condition met DSM–III–R criteria (per SCID-P) for past history of major depression when the organic rule-out (for alcohol involvement) was not applied. However, only 1 participant (in RTC condition) met criterion for past major depressive episode when the organic rule-out was applied (i.e., during a period while not drinking).
Of the 35 participants who completed baseline measures, one participant (in RTC condition) did not complete the posttreatment assessment but was included in all other analyses. Thirty-two participants (91%) completed all follow-assessments (1-, 3-. and 6-month); we were unable to locate 2 participants in the CBT-D condition and 1 participant in the RTC condition. Thus, our follow-up rate at 6 months was 91.4% (32 of 35). Chi-square analyses showed no significant differences between groups in attrition at either interval.
Participants were in partial hospital treatment for an average of 21.2 calendar days (SD = 4.5), and the mean number of days between pre- and posttreatment assessments was 18.9 (SD = 4.4). There were nonsignificant differences between treatment conditions on length of partial hospital treatment and length of time between assessments (ps > .10). Participants attended an average of 7.43 (SD = 1.0) treatment sessions, and there were no significant differences between CBT-D and RTC conditions (p > . 10). Twenty-one patients were prescribed antabuse on discharge, with no significant differences between treatment conditions (p > .10).
Validity of Self-Reported Alcohol Use
Self-report drinking data at follow-up were compared with the reports from the one SO who provided information as to each participant's drinking. Because of attrition and inability to contact several SOs at each follow-up point, 59 pairs of drinking reports were compared. The agreement between participant and SO reports was 80% (47 of 59) for the participant drinking at all during follow-up. Seven SOs claimed that participants were abstinent when the participants claimed they drank, whereas five SOs claimed that participants drank when the participants reported abstinence. If verification is performed only to confirm participants' self-reported abstinence (i.e., cases where participant reported drinking and SO reported abstinence are not included), agreement between SO and participant reports improves to 92% (54 of 59). The agreement between participant and SO reports was 68% (34 of 50) for the classification of the participants' drinking during the 6 months posttreatment as heavy drinking; that is, more than six drinks for men and more than five drinks for women. (An additional nine SOs were unable to report on the extent of patients' drinking, thus the agreement for this heavy drinking analysis was based on 50 pairs of drinking reports.) Tfen SOs claimed that the participant was not drinking heavily when the participant reported he or she was, whereas 6 SOs claimed the participant was drinking heavily when the participant reported they were not. If verification is performed on the participant's claim of not drinking heavily, agreement between SO and participant reports improves to 88% (44 of 50). SO reports significantly correlated with participant reports for number of drinking days during the 6 months post-treatment, r = .43, p < .03. When data from SOs who said they were not very confident about their knowledge of the participant's drinking were eliminated from these analyses (four people), the correlation improved, r = .56, p < .005. The participant reported more drinking days in 12 cases, the SO reported more drinking days in 19 cases, and the reports agreed in 21 cases. Overall, these data provide strong support for the validity of patients' self-reported alcohol use over the follow-up period.
Cohort or Seasonal Effects
Given that treatment was administered in five sequential, non-overlapping cohorts, we examined the possibility of a cohort effect by comparing the three CBT-D cohorts to each other and the two RTC cohorts to each other on both baseline variables (HAM-D, BDI, POMS Depression subscale, POMS Anxiety subscale, DAS, PES-MR, drinking self-efficacy, age, years of education, number of previous treatments for alcohol use, ADS, and age of onset of alcohol dependence) and drinking outcome variables. Nonparametrics, Mann–Whitney U and Kruskal–Wallis one-way analysis of variance (ANOVA) were used because of small sample sizes. No significant differences were found for any of the baseline or outcome variables between either the RTC cohorts or the CBT-D cohorts, all ps > .09.
Because using a sequential, nonoverlapping design also raises the possibility of a seasonal effect, we divided cohorts into seasons on the basis of the date of the first day of treatment. One CBT-D cohort started in the winter, 1 CBT-D and 1 RTC cohort in the spring, 1 RTC cohort in the summer, and 1 CBTD cohort started in the fall. The four groups were compared on the following baseline variables: HAM-D, BDI, POMS Depression subscale, POMS Anxiety subscale, DAS, PES-MR, drinking self-efficacy, age, number of years of education, number of previous treatments for alcohol use, ADS, and age of onset of alcohol dependence. No significant differences were found using the Kruskal–Wallis one-way ANOVA, allps > .20. Thus, we feel confident that there were no biases that were due to either cohort effects or seasonal effects.
Relationship Between Baseline Variables and Drinking Outcomes
Before testing the effects of treatment condition on drinking outcomes (percent days abstinent and drinks per day at both baseline to 3-month follow-up and 3- to 6-month follow-ups), partial correlation coefficients were computed to examine whether the following baseline variables were significantly related to drinking outcomes: HAM-D, BDI, POMS Depression subscale, POMS Anxiety subscale, DAS, PES-MR, drinking self-efficacy, gender, age, number of years of education, number of previous treatments for alcohol use, ADS, age of onset of alcohol dependence, and past history of major depressive disorder (excluding organic rule out). Because of the number of correlations, we used a significance level of p < .01 to control for Type I error. Controlling for the corresponding baseline drinking variable, none of the baseline variables were significantly related to drinking outcomes at either follow-up paint.
In addition, we computed partial correlations to examine whether use of antidepressant medication at admission and at discharge from the partial hospital program was related to drinking outcomes. In the RTC condition, 6 participants were on antidepressant medication on admission and 8 participants were on antidepressant medication at discharge. In the CBT-D condition, 4 participants were on antidepressant medication at admission and 6 were on antidepressant medication when discharged. There were no between group differences at either admission or discharge and antidepressant use was not significantly related to any drinking outcome variables, all ps > .10.
Treatment Effects: Change in Depressive Symptoms and Depressed and Anxious Mood
We used repeated measures analyses of covariance (ANCOVA), with baseline percent days abstinent as the covariate, to investigate the effect of time (pretreatment vs. posttreatment) and treatment condition on measures of depressive symptoms and mood. Separate 2 × 2 (Treatment × Time) analyses were conducted for the three depression measures and for anxious mood.
Significant main effects for Time were found on the HAM-D, F(1, 32) = 54.93, p < .001; BDI, F(1, 32) = 42.36, p < .001; POMS Depression subscale, F(1, 30) = 17.04, p < .001; and POMS Anxiety subscale, F(1, 30) = 14.25, p < .01, with participants in both conditions reporting less depressive symptoms and negative mood from pre- to posttreatment. No significant main effects of treatment were found on the HAM-D, BDI, or either POMS subscale. However, a significant Treatment × Time interaction was found on the HAM-D, F(1, 32) = 4.15, p = .05; POMS Depression subscale, F(1, 30) = 8.5, p < .01; and POMS Anxiety subscale, F(1, 30) = 5.68, p < .03; with patients in the CBT-D condition showing a greater decrease in depressive symptoms and negative mood than those in the RTC condition between pre- and posttreatment (see ). No significant Treatment × Time interaction was found with the BDI.
Figure 1 Depressive symptoms and negative moods by treatment condition and time. CBT-D = cognitive-behavioral treatment for depression; RTC = relaxation training control; HAM-D = Modified Hamilton Rating Scale for Depression; POMS = Profile of Mood States; Depression (more ...)
Treatment Effects: Change in Cognitive–Behavioral Process Variables
Repeated measures ANCOVAs, with baseline percent days abstinent as the covariate, were used to examine the effects of Time (pretreatment vs. posttreatment) and treatment on cognitive–behavioral process variables for the measurement of dys-functional attitudes, pleasant events, and drinking self-efficacy. The effect of Time was significant for dysfunctional attitudes, F
(1, 32) = 10.19, p
< .01, and approached significance for drinking self-efficacy, F
(1, 24) = 4.06, p
< .06, with both treatment conditions showing a decrease in dysfunctional attitudes and an increase in self-efficacy over time. There was a nonsignificant effect of Time for pleasant events (p
> .10). No significant effects of Treatment or Treatment × Time interactions were found for any of the three process variables. Given diese results, we did not proceed to test for mediation effects, as significant treatment effects on these variables were a prerequisite condition for mediation (Baron & Kenny, 1986
Treatment Effects: Changes in Drinking Status, Frequency and Quantity
For both the 0- to 3-month and 3- to 6-month intervals, participants were coded as having drank or not and as having drank heavily or not (i.e., consumed more than six standard alcoholic drinks for men, or five for women on any one day) during that 3-month interval. These results are presented in . For the first 3 months, both chi-square analyses (Treatment × Any Drinking and Treatment × Heavy Drinking) were nonsignificant. During the second follow-up interval (3 to 6 months posttreatment), the chi-square analysis was significant for any drinking, χ2(1, N = 32) = 4.22, p < .04, with 53% of CBT-D participants drinking (at least once) during this 3-month period compared with 87% of RTC participants. The chi-square analysis for heavy drinking during the second 3-month follow-up interval was nonsignificant.
Means and Standard Deviations of Dependent Variables by Treatment Condition
Drinking frequency and quantity
Because of skewness, all variables involving percent abstinent days were normalized using log transformations and mean number of drinks per day using inverse transformations, and the transformed variables were used in all analyses. However, raw scores were used in all figures for ease of interpretation. Separate 2 × 2 (Treatment × Time) repeated measures ANCOVAs were conducted for percent days abstinent and mean number of drinks per day. We entered both the 0- to 3-month and 3- to 6-month follow-up values as the repeated measure, with the corresponding variable at base-line (for past 180 days) serving as the covariate.
The repeated measures ANCOVA for percent days abstinent revealed a significant main effect for Time, F(1, 30) = 8.64, p < .01, as percent abstinent days at follow-up declined over time in both conditions. More relevant to the study hypotheses, the analysis revealed a significant effect of Treatment, F(1, 29) = 8.21, p < .01, and a Treatment × Time interaction that approached significance, F(1, 30) = 2.85, p = .10. Examination of reveals that the overall percentage of days abstinent is higher in the CBT-D condition than in the RTC and that the decrease in percent days abstinent from the first to the second follow-up interval is greater in the RTC condition than in the CBT-D. A simple effects test across groups at each time point revealed a significantly greater percent days abstinent at 0- to 3-month follow-up, t(30) = 2.25, p = .03, and at 3- to 6-month follow-up, t(30) = 2.95, p < .01, in the CBT-D condition relative to the RTC. In addition, a simple effects test within groups across time showed that, whereas the RTC condition had a significant decrease in percent days abstinent, t(14) = 2.48, p < .03, from the first to the second 3 months of follow-up, the CBT-D condition did not change significantly (p > .10).
Drinking frequency and quantity at follow-up by treatment condition and time. CBT-D cognitive–behavioral treatment for depression; RTC = relaxation training control.
On the mean number of drinks per day, the repeated measures ANCOVA revealed a significant main effect of Time, F(1, 30) = 8.64, p < .01; a nonsignificant effect of Treatment, F( 1, 29) = 2.67, p > . 10; and a significant Treatment × Time interaction, F(1, 30) = 4.10, p = .05. Examination of shows that mean number of drinks per day is lower in the CBT-D group between 3- and 6-month follow-up and that RTC patients had a greater increase in mean number of drinks per day than patients receiving CBT-D. A simple effects test between groups at each time point indicated that the two groups did not differ significantly at 0- to 3-month follow-up but that CBT-D patients drank significantly fewer drinks per day than RTC patients at 3- to 6-month follow-up, t(30) = 2.12, p < .05. A simple effects test within groups across time confirmed that, although the CBT-D condition did not change significantly from the first to the second 3 months (p > .10), the RTC condition's mean number of drinks per day increased significantly, t(14) = 2.62, p < .03.
Relationship Between Depressive Symptoms and Drinking Outcome
Given our findings that CBT-D intervention resulted in greater decreases in depressive symptoms during treatment and in improved drinking frequency and quantity at longer term followup, we examined whether change in depressive symptoms mediated the relationship between treatment condition and these drinking outcomes from 3- to 6- months after treatment; that is, did patients in the CBT-D condition drink less at follow-up because of improved levels of depressive symptoms? For change in depressive symptoms to be a mediator, the following conditions must be met: (a) treatment condition is significantly related to change in depressive symptoms; (b) change in depressive symptoms is significantly related to drinking outcome; (c) treatment is significantly related to drinking outcome; and (d) after controlling for change in depressive symptoms, either treatment is no longer significantly related to drinking outcome or the relationship between treatment and outcome is significantly reduced (Baron & Kenny, 1986
To test these conditions, we performed a series of regression analyses. We controlled for the appropriate baseline drinking variable by entering it first in all regression analyses. First, change in depressive symptoms was regressed onto treatment condition (, Line A). Second, drinking outcome was regressed onto change in depressive symptoms (, Line B). Third, drinking outcome was regressed onto treatment condition (, Line C). Fourth, drinking outcome was regressed onto treatment condition after controlling for change in depressive symptoms (, Line D). According to the aforementioned criteria, evidence suggests that change in HAM-D is a mediator of the relationship between treatment condition and 3- to 6-month drinking outcomes. The only possible exception was that the relationship between treatment condition and percent days abstinent was still statistically significant after controlling for change in depressive symptoms and the decrease in the strength of the relationship between treatment and drinking required to show mediation only approached significance (p
<.08; Meng, Rosenthal, & Rubin, 1992
Figure 3 The mediating effect of change in depressive symptoms on the relationship between treatment and drinking outcome. A series of regression analyses were performed with the appropriate baseline drinking variable controlled for by entering it first in all (more ...)
Posttreatment Alcoholics Anonymous (AA) Participation
The standard partial hospitalization treatment for alcohol-dependent patients includes strong encouragement for participation in AA programs. Therefore, patients were asked at 1-, 3-, and 6-month follow-ups whether they had attended any AA meetings and if so, how many meetings they had attended at each followup interval. At 1 month, 94% of CBT-D and 64% of RTC patients had attended one or more AA meetings since posttreatment, χ2 (1 ,N = 27) = 3.92, p < .05; at 3 months, 87% of CBT-D and 42% of RTC patients had attended AA meeting (or meetings) since the 1-month follow-up, χ2 (1, N = 27) = 6.08, p < .02; and at 6 months, 81% of CBT-D and 27% of RTC patients had attended AA meetings since the 3-month follow-up, χ2 (1, N = 31) = 9.31, p < .01. There were also significant differences in the number of AA meetings attended by the CBT-D versus those attended by the RTC patients between posttreatment and 1 month (Mdns = 11 and 0, respectively), between 1 and 3 months (Mdns = 27 and 0, respectively), and between 3 and 6 months (Mdns = 31 and 0, respectively); Mann–Whitney tests, all ps < .005. Although patients were not asked directly if they had attended AA meetings before treatment, they were asked how many days they had received treatment in an outpatient setting, including AA, in the 30 days before starting treatment. No significant pretreatment difference was found between treatment conditions; CBT-D, M = 1.05, RTC, M = 2.38 (p > .10).
AA Participation and Drinking Outcome
Because there were treatment group differences on AA attendance during follow-up, we reexamined the relationship between treatment group and drinking outcomes after covarying the number of AA sessions attended. Specifically, using regression analyses, we examined the relationship between treatment group and drinking outcomes between 3- and 6-month follow-ups, with and without covarying the number of AA sessions attended from 0- to 3-months. As expected, without covarying AA sessions, treatment group significantly predicted 3- to 6-month percent days abstinent (r = .47, p = .006), number of drinks per day (r = .36, p = .04), and drinking status (i.e., totally abstinent vs. any drinking), model χ2 (1, N = 32) = 4.46, p = .03. After covarying me number of AA sessions attended from 0- to 3-months, we found that treatment group continued to significantly predict 3- to 6-month percent days abstinent (r2 change = .18, p = .01) and number of drinks per day (r2 change = .18, p = .01), and the prediction of 3- to 6-month drinking status approached significance, model χ2(1, N = 27) improvement = 2.69, p = .10.