The Efficacy of Cognitive-Behavioral Therapy and Psychodynamic Therapy in the Outpatient Treatment of Major Depression: A Randomized Clinical Trial

This study was a randomized clinical trial with an allocation ratio of 1:1 for CBT and psychodynamic therapy. The study design was approved by the Dutch Union of Medical Ethics Trial Committees for mental health organizations.

Participants were referred by their general practitioner to one of three psychiatric outpatient clinics in Amsterdam. Inclusion criteria were presence of a major depressive episode according to DSM-IV criteria, as assessed with the Mini-International Neuropsychiatric Interview-Plus (15); a Hamilton Depression Rating Scale (HAM-D [16]) score ≥14; age 18–65 years; and written informed consent after receiving a complete description of the study.

Exclusion criteria included presence of psychotic symptoms or bipolar disorder, severe suicidality warranting immediate intensive treatment or hospitalization, substance misuse or abuse in the past 6 months, pregnancy, inability to meet trial demands, and use of psychotropic or other medications that might influence mental functions. Patients on an antidepressant regimen were included only if the medication they were currently taking was judged to be inefficacious by both the patient and the intake psychiatrist. If so, the medication was tapered off under medical supervision, and baseline assessment took place after a washout period of at least 1 week after the medication was completely stopped (16/25 of these patients subsequently restarted medication during the study).

Both psychotherapies comprised 16 individual sessions within 22 weeks and were conducted according to published treatment manuals (17, 18). Additionally, both were suitable for application in a broad group of patients, including those of a non-Northwest European (immigrant) cultural background. CBT was based on the principles described by Beck (19) and included behavioral activation and cognitive restructuring according to a session-by-session protocol with homework assignments. Short-term psychodynamic supportive psychotherapy (18, 20–24) was used to represent the psychodynamic intervention. This modality involved an open patient-therapist dialogue that used supportive and insight-facilitating techniques to address the emotional background of the depressive symptoms by discussing current relationships, internalized past relationships, and intrapersonal patterns.

Psychotherapists in both treatment conditions were psychiatrists or psychologists with at least a master’s degree who completed either a 3-day course in short-term psychodynamic supportive psychotherapy or a 100-hour basic CBT training course accredited by the Dutch Association for Behavioral and Cognitive Therapy. Moreover, all therapists adequately conducted at least one intensively supervised therapy case in accordance with the relevant treatment manual as judged by a study supervisor. Although no formal assessments were conducted, manual fidelity was checked by means of biweekly supervision sessions, chaired by a study supervisor, in which audiotaped material was discussed. All study supervisors were registered supervisors with either the Dutch Association of Psychoanalytic Psychotherapy (one supervisor was in training) or the Dutch Association for Behavioral and Cognitive Therapy. Differences in the mean number of years that the supervisors had been conducting their respective modalities were minimal (CBT, 10.9 years [SD=11.0]; psychodynamic therapy, 9.7 years [SD=2.9]) but somewhat larger with respect to mean years conducting supervision (CBT, 14.6 years [SD=10.3]; psychodynamic therapy, 6.3 years [SD=3.2]), although neither difference was significant.

Thirty-seven CBT therapists and 56 psychodynamic therapists treated, on average, 4.2 patients (range, 1–16) and 3.0 patients (range, 1–12), respectively. No differences between treatment conditions were found with regard to the average number of times a patient was discussed in supervision (CBT, 4.3; psychodynamic therapy, 4.6) or the mean number of therapy sessions patients received (CBT, 10.6; psychodynamic therapy, 10.9; mean numbers are lower than the maximum of 16 because of premature termination, dropout, and patients missing sessions). Regarding therapist protocol adherence, CBT therapists reported a mean score of 7.1 (scale range, 0–10) over 1,218 CBT sessions. Conditions did not differ regarding the mean number of years of clinical experience therapists had after completing their master’s degree or medical degree (CBT, 7.5 years [SD=7.3]; psychodynamic therapy, 7.4 years [SD=6.7]), but CBT was more often conducted by psychologists, and psychodynamic therapy was more often conducted by psychiatrists (χ²=109.80, df=1, p<0.001). Furthermore, CBT was conducted more often by a female therapist than was psychodynamic therapy (χ²=15.91, df=1, p<0.001). We therefore conducted a sensitivity analysis controlling for therapist gender and profession (psychologist, psychiatrist).

Patients with severe depression (HAM-D score >24) at baseline (N=129) and those with moderate depression at baseline who developed severe symptoms during psychotherapy monotherapy (N=21) were offered additional antidepressant medication administrated by a psychiatrist (who was not the patient’s psychotherapist) according to a protocol starting with extended-release venlafaxine at 75 mg/day, which could be increased to a maximum dosage of 225 mg/day. In case of intolerance or non-response, venlafaxine was switched to either citalopram or nortriptyline. Pharmacotherapy consultations addressed symptom evaluation, side effects, and adherence. Three research psychiatrists supervised pharmacotherapy.

Eight patients (6.2%) did not start the recommended pharmacotherapy, and 14 (10.8%) switched medication during treatment. The number of patients not starting pharmacotherapy, the pharmacotherapy dosages used, and patient-reported medication adherence did not differ significantly between treatment conditions.

The primary outcome measure was posttreatment remission rate (HAM-D score ≤7). Secondary outcome measures included 1-year follow-up remission rates and posttreatment and follow-up observer-rated (HAM-D) and patient-rated mean depression scores (Inventory of Depressive Symptomatology–Self-Report [IDS-SR] [25]).

Trained research assistants (master’s-level graduate students in clinical psychology) assessed the HAM-D according to the Dutch scoring manual (26). Assessors participated in biweekly 1-hour peer supervision sessions, in which audiotaped interviews were discussed. The average intraclass correlation coefficient over 46 audiotaped assessments scored by multiple assessors was 0.97. Both observer- and patient-rated depression measures showed good reliability at baseline assessment (Cronbach’s alpha: observer-rated, 0.75; patient-rated, 0.78). HAM-D assessors were not blind to treatment condition. We therefore requested the assessors’ hypotheses regarding treatment effects at the posttreatment and follow-up assessments and conducted a sensitivity analysis controlling for these variables.

Separate random allocation sequences were generated for each of the three clinics by one of the authors (J.P.) using the SPSS random number generator (SPSS, Chicago). Randomization was stratified by gender and age (<32.5 years and >32.5 years). Research assistants, aware of the allocation sequence, enrolled participants and assigned them to interventions.

Analyses were based on an intent-to-treat sample, including all patients randomly assigned. Patients were considered drop-outs if they completed less than eight psychotherapy sessions. Response was defined as a reduction in HAM-D scores ≥50% at posttreatment.

Given the hierarchical data structure, linear mixed-model analyses were used for continuous outcomes, and logistic generalized estimating equation analyses were used for dichotomous outcomes. These analyses were preferred over logistic mixed-model analyses because of the instability of the latter (27). In examining pre- to posttreatment outcomes, we excluded follow-up data (for which additional help-seeking could not be controlled) from the analyses. Mixed-model analyses were conducted according to a three-level structure (therapist, patient, and repeated measures). Location (clinic) was included as a covariate in a sensitivity analysis, rather than as a level because of the small number of categories (N=3). Mixed-model analyses were performed with MLwiN, version 2.22 (http://www.bristol.ac.uk/cmm/software/mlwin). All other analyses were performed with SPSS, version 16.0.

We started with a basic model including main effects for treatment and time and a time-by-treatment interaction. Time was treated as a categorical variable to assess treatment effects at the different time points. To control for possible confounders, we next added the following sets of covariates: 1) clinic and number of patients with a baseline HAM-D score >24; 2) demographic characteristics (as listed in Table 1); 3) depression characteristics (as listed in Table 1); 4) therapist gender and profession; 5) the HAM-D assessors’ treatment outcome hypotheses; and 6) help-seeking during the follow-up period (as reported in Table S1 in the data supplement that accompanies the online edition of this article). The estimated main effects for treatment at different assessment points under these different models are reported as odds ratios with 95% confidence intervals for remission rates and differences in means for continuous outcomes. For the latter, we also calculated effect sizes (Cohen’s d) and 95% confidence intervals using Comprehensive Meta-Analysis, version 2.2.046 (Biostat, Englewood, N.J.).

We then used a two-step strategy for the interpretation of outcomes. First, we examined whether treatment outcomes differed significantly. We considered treatment differences to be nonsignificant if the 95% confidence interval included 1.00 or 0.00 for odds ratios and effect sizes, respectively. This constituted our primary research question. Second, when differences were nonsignificant, we next examined whether psychodynamic therapy was noninferior to CBT by comparing outcomes to pre-specified noninferiority margins. These margins were determined based on the expert opinion of 10 experienced depression research clinicians (unaware of preliminary trial results) who rated a difference of 10% for remission rates and 2.6 HAM-D points (equivalent to a Cohen’s d value of 0.30) as the maximum difference allowable to conclude noninferiority. Based on a maximum difference of 10% (remission rates of 12% and 22%), noninferiority margins for remission were set at an odds ratio of 0.49. For all continuous outcome measures, noninferiority margins were prespecified at a Cohen’s d value of 0.30. We compared the 95% confidence intervals of the effect sizes and odds ratios with the above-mentioned prespecified noninferiority margins and considered psychodynamic therapy noninferior to CBT on a given outcome measure if the 95% confidence interval limit did not exceed the prespecified noninferiority margin for that measure. This constituted our secondary research question. We repeated the main analyses in the subgroup of moderately depressed patients (HAM-D score ≤24) receiving psychotherapy only and in the subgroup of severely depressed patients (HAM-D score >24) receiving combined psychotherapy and pharmacotherapy in order to investigate whether pooling these subgroups may have obscured differential patterns of results.

An a priori power analysis (14) indicated that 300 participants were required (alpha=0.05, 1 − β=0.80) to answer our primary research question. To detect the 10% difference in remission rates between conditions that constituted the noninferiority margin (alpha=0.05, 1 − β=0.80), 344 participants were needed (using SPSS SamplePower for equivalence studies, one-tailed). Power to detect an outcome difference of a Cohen’s d value of 0.30 for continuous outcome measures was 0.87.

The CONSORT diagram for the study is presented in Figure 1. From April 2006 to December 2009, 4,866 patients were assessed for eligibility during a standard intake procedure; 570 (11.7%) were found to be potentially eligible and invited for baseline assessment. Of these patients, 229 (40.2%) did not meet inclusion criteria or were not willing to participate. Therefore, 341 patients were randomly assigned to CBT (N=164) or psychodynamic therapy (N=177). Demographic and clinical characteristics of the sample are summarized in Table 1. No significant differences were found between the two treatment conditions.

No significant differences were found between treatment conditions regarding the proportion of patients who did not complete treatment (CBT, 31.1%; psychodynamic therapy, 25.9%). The majority of patients who dropped out were those who missed treatment appointments without specifying a reason (53.9%). Recruitment was discontinued when the intended number of participants was reached. One-year follow-up assessments were conducted from April 2007 to January 2011.

Based on observed data, 24.3% (N=27/111) of the patients in the CBT condition and 21.3% (N=26/122) in the psychodynamic therapy condition met the remission criterion at the posttreatment assessment. Observed response rates were 38.7% (N=43/111) for CBT and 36.9% (N=45/122) for psychodynamic therapy. Estimated odds ratios for remission at different assessment points are listed in Table 2. At the posttreatment assessment, the odds ratio was 0.82 (95% CI=0.45–1.50), indicating that remission rates did not differ significantly. The lower limit of the odds ratio’s 95% confidence interval exceeded the prespecified noninferiority margin of 0.49. This pattern of results did not change when different sets of covariates were added (Table 2).

Mean observer- and patient-rated depression scores during treatment for both groups are presented in Figure 2; depressive symptom scores in both conditions improved over time. Estimated observer- and patient-rated mean differences at different assessment points are presented in Table 2, along with effect sizes of the posttreatment differences between conditions. At week 22, the estimated observer-rated difference between treatment conditions was 0.24 points (SE=0.90) on the HAM-D, corresponding with an effect size (Cohen’s d) of 0.02 (95% CI=−0.24 to 0.27), indicating that treatment differences were nonsignificant. The estimated patient-rated difference between treatment conditions was 1.94 points (SE=1.92) on the IDS-SR, corresponding with an effect size of −0.08 (95% CI=−0.38 to 0.22), also indicating that differences were not significant. The upper limits of the 95% confidence intervals for both effect sizes did not exceed the prespecified noninferiority margin of 0.30. This pattern of results did not change when controlling for different covariates, although the noninferiority margin was slightly exceeded when controlling for clinic and number of patients with baseline HAM-D scores >24 (HAM-D estimated mean difference=0.64 [SE=0.81]; Cohen’s d=0.05, 95% CI=−0.21 to 0.31).

In the moderately depressed subgroup, observed remission rates for CBT and psychodynamic therapy were 26.5% (N=18/68) and 27.7% (N=23/83), respectively, with estimated odds ratios (1.02, 95% CI=0.50–2.06), observer ratings (Cohen’s d=−0.05, 95% CI=−0.37 to 0.27), and patient ratings (Cohen’s d=−0.24, 95% CI=−0.61 to 0.13) all indicating that psychodynamic therapy was noninferior to CBT. In the severely depressed subgroup receiving additional pharmacotherapy, observed remission rates for CBT and psychodynamic therapy were 20.9% (N=9/43) and 7.7% (N=3/39), respectively, with estimated odds ratios (0.31, 95% CI=0.08–1.26), observer ratings (Cohen’s d=0.21, 95% CI=−0.23 to 0.64), and patient ratings (Cohen’s d=0.17, 95% CI=−0.35 to 0.69) all indicating no significant differences, without demonstrating noninferiority of psychodynamic therapy relative to CBT.

Follow-up assessments were conducted with 192 (56.3%) participants (Figure 1). More patients reported having received additional treatment during the follow-up period in the CBT condition (N=41 [44.6%]) than in the psychodynamic therapy condition (N=32 [33.0%]), but this difference did not reach significance (x²=2.67, df=1, p=0.10) (see Table S1 in the online data supplement).

Based on observed data, 34.7% (N=33/95) of patients in the CBT condition and 26.8% (N=26/97) of those in the psychodynamic therapy condition met the remission criterion. The estimated odds ratio of remission rates at follow-up was 0.74 (95% CI=0.41–1.34) (Table 3), indicating that remission rates did not differ significantly. The lower limit of the odds ratio’s 95% confidence interval exceeded the prespecified noninferiority margin of 0.49. This pattern of results did not change when different sets of covariates were added.

Estimated observer- and patient-rated mean differences at follow-up and corresponding effect sizes are presented in Table 3. The estimated observer-rated difference between treatment conditions was 1.94 points (SE=1.01) on the HAM-D (Cohen’s d=0.14, 95% CI=−0.14 to 0.42), and the estimated patient-rated difference was 2.99 points (SE=2.22) on the IDS-SR (Cohen’s d=0.12, 95% CI=−0.23 to 0.48), both indicating that treatment differences were not significant. The upper limits of the 95% confidence intervals for both effect sizes exceeded the prespecified noninferiority margin of 0.30. This pattern of results did not change when controlling for different covariates.

In the moderately depressed subgroup, observed remission rates were 40.0% (N=22/55) in the CBT condition and 35.4% (N=23/65) in the psychodynamic therapy condition (odds ratio=0.84, 95% CI=0.41–1.73; observer rated: Cohen’s d=0.08, 95% CI=−0.28 to 0.44; patient rated: Cohen’s d=0.02, 95% CI=−0.40 to 0.45). In the severely depressed subgroup receiving additional pharmacotherapy, observed remission rates were 27.5% (N=11/40) for CBT and 9.4% (N=3/32) for psychodynamic therapy (odds ratio=0.27, 95% CI=0.07–1.08; observer rated: Cohen’s d=0.32, 95% CI=−0.15 to 0.79; patient rated: Cohen’s d=0.40, 95% CI=−0.27 to 1.07). All these differences were nonsignificant, but noninferiority could not be demonstrated for any of them.

Serious adverse events during treatment and follow-up were mostly increases in depressive symptoms or suicidality for which additional treatment was needed (see Table S2 in the online data supplement). No differences were found between the treatment conditions with regard to the proportion of patients reporting serious adverse events during treatment (CBT, 6.1%; psychodynamic therapy, 6.2%) or follow-up (CBT, 1.8%; psychodynamic therapy, 2.3%).

This study has a number of strengths. First, several elements contribute to the generalizability of the study’s findings. Treatment was provided in regular psychiatric outpatient clinics by a large number of therapists with different experience levels. Patients were not recruited by advertisement but instead were referred by general practitioners, and no selection criteria with regard to previous treatment or suitability for psychotherapy were applied. Patients with relatively low socioeconomic status were included, and almost one-half of the patient sample reported a non-Northwest European (immigrant) cultural background. Second, we included severely depressed patients who were additionally treated with pharmacotherapy prescribed according to a protocol. Third, this is, to our knowledge, the largest randomized controlled trial to date comparing CBT and psychodynamic therapy in the treatment of depression (N=341). By comparison, a meta-analysis of psychodynamic therapy for depression included a total of 421 patients across six CBT-psychodynamic randomized controlled trials (9). Finally, this was the first study to test whether psychodynamic therapy can be demonstrated to be noninferior to CBT in the treatment of depression.

This study also has a number of limitations. First, a substantial number of patients did not complete treatment or were lost to assessment. Second, treatment fidelity was not systematically assessed but was instead checked by means of intensive supervision by experienced supervisors and subjective therapist ratings. These ratings suggested adequate adherence to the CBT manual, but no such measure was available for psychodynamic treatment. Third, HAM-D assessors were not blind to treatment condition, and therefore we cannot rule out observer bias. However, controlling for assessor-rated treatment expectations did not alter the pattern of results, and results were similar for observer- and patient-rated outcomes. Fourth, research assistants enrolling participants were aware of the allocation sequence, which may have introduced selection bias. However, no significant differences were found with regard to any of the sample baseline characteristics. Fifth, although noninferiority margins were carefully thought through and based on clinical expert opinion, they were still set in an arbitrary fashion. Sixth, we could not prevent patients from seeking additional treatment during the follow-up period, and a nonsignificant finding suggested that patients in the CBT group may have returned to treatment more than those in the psychodynamic therapy group. However, controlling for additional treatment in the follow-up period did not change the general pattern of results. Finally, the study did not include a control condition.