Source: DeRubeis et al., Archives of General Psychiatry, Vol 62, Apr 2005
Tags: cognitive therapy, CT, antidepressant medications, ADM, treatment outcomes, response rates, remission rates, site-by-treatment interaction, therapist experience, placebo-controlled, HDRS, moderate to severe depression, MDD, effect size, clinical implications, APA guidelines
At 8 weeks, both cognitive therapy and medications outperformed placebo, with no significant difference between the two active treatments. At 16 weeks, response rates were identical (58% each). However, a significant site-by-treatment interaction emerged: medications outperformed CT at Vanderbilt (where CT therapists were less experienced) but not at Pennsylvania (where CT therapists were highly experienced). The study's headline conclusion is that CT can match medications for moderate to severe depression, provided the therapists are sufficiently skilled.
Response rate
The proportion of patients meeting a predefined threshold of improvement. At 8 weeks: HDRS score of 12 or lower. At 16 weeks: a multi-assessment criterion designed to prevent a transient symptom spike from disqualifying an otherwise improving patient.
Remission rate
A stricter criterion than response. Requires meeting all response criteria plus a final HDRS score of 7 or lower. Represents a return to near-normal functioning, not merely improvement.
Site-by-treatment interaction
A statistical finding indicating that the relative effectiveness of treatments differs across study locations. In this trial, the CT vs ADM comparison produced different results at Pennsylvania and Vanderbilt. When this interaction is significant, you cannot simply report a single overall treatment effect; you must look at each site separately.
Cochran Q test
A statistical test used to compare response classifications (responder vs non-responder) across treatment conditions while adjusting for site. An extension of the chi-square test to stratified contingency tables.
Augmentation
Adding a second medication when the primary one has not produced adequate response. In this trial, non-responders to paroxetine after 8 weeks could be augmented with lithium carbonate or desipramine hydrochloride. Sixty-four per cent of augmented patients received lithium, 56% desipramine, and some received both in combination or sequence.
Anxiolytic effects
Anti-anxiety properties of a medication. Paroxetine's anxiolytic effects became relevant in explaining why patients with comorbid anxiety disorders responded particularly well to ADM, contributing to the superior ADM outcomes at Vanderbilt (where anxiety comorbidity was more prevalent).
This is the only phase where all three conditions can be compared, since placebo ended at 8 weeks.
Response rates at 8 weeks:
ADM: 50% (95% CI, 41% to 59%)
CT: 43% (95% CI, 31% to 56%)
Pill placebo: 25% (95% CI, 16% to 38%)
Cochran Q test controlling for site confirmed a significant difference across the three treatments (P = .006). Both active treatments significantly outperformed placebo:
ADM vs placebo: significant (P = .001)
CT vs placebo: significant (P = .04)
ADM vs CT: not significant (P = .40)
Response rates did not differ between sites for any treatment condition.
Continuous HDRS scores at 8 weeks (HLM analysis):
All three groups improved over the first 8 weeks, but the rate of improvement differed by treatment:
ADM vs placebo: significant advantage (P = .006), effect size 0.60 (medium)
CT vs placebo: nonsignificant trend (P = .09), effect size 0.44 (small to medium)
ADM vs CT: not significant (P = .46), effect size 0.16 (trivial), favouring ADM
The site-by-treatment interaction was not significant at 8 weeks (P = .19), meaning the pattern of results was broadly similar at both sites during this phase.
The key takeaway from the 8-week data is that both active treatments produced clinically meaningful improvement over placebo, with medium effect sizes. The categorical analysis (response rates) showed both clearly beating placebo. The continuous analysis showed a slightly weaker signal for CT, but the CT vs ADM difference was not significant by either method.
With placebo ended, the 16-week comparison is ADM vs CT head-to-head.
Response rates at 16 weeks:
ADM: 58% (95% CI, 48% to 66%)
CT: 58% (95% CI, 45% to 70%)
Difference: not significant (P = .92)
Response rates were identical overall. The test of a site-by-treatment interaction was a nonsignificant trend (P = .07), and within-site comparisons were not significant at either location.
Remission rates at 16 weeks:
ADM: 46% (95% CI, 37% to 55%)
CT: 40% (95% CI, 28% to 53%)
Here the site-by-treatment interaction was significant (P = .04):
At Pennsylvania: no significant difference in remission rates (P = .37)
At Vanderbilt: ADM remission higher than CT remission at the level of a nonsignificant trend (P = .05)
Continuous HDRS scores at 16 weeks (HLM analysis):
No significant main effect for treatment overall, but a significant site-by-treatment interaction (P = .02, effect size 0.36):
At Pennsylvania: no significant difference between ADM and CT (P = .20, effect size 0.37)
At Vanderbilt: ADM produced significantly greater change than CT (P = .04, effect size 0.57, a medium effect)
This is the most important nuance in the study. The overall finding (CT = ADM) masks a real difference between sites, and the authors traced that difference to two contributing factors.
Factor 1: Patient population differences at Vanderbilt
Axis I comorbidity was more prevalent in the ADM condition at Vanderbilt (87%) than at Pennsylvania (52%). Patients with Axis I comorbidity, especially anxiety disorders, responded particularly well to ADM. This makes pharmacological sense: paroxetine has established anxiolytic properties.
Critically, patients with Axis I comorbidity responded equally well to ADM at both sites. The difference was not that Vanderbilt's ADM was better, but that Vanderbilt's ADM sample had more of the patients who tend to do well on paroxetine.
Factor 2: Therapist experience differences
All three CT therapists at Pennsylvania had 7 to 21 years of CT experience. Two of three CT therapists at Vanderbilt had only 2 years of experience.
The more experienced therapists at Pennsylvania produced CT outcomes at least comparable to ADM. The less experienced therapists at Vanderbilt produced CT outcomes inferior to ADM. This mirrors the TDCRP pattern: in that earlier trial, sites with less CT experience showed large advantages for ADM, while the site with greater CT experience showed negligible differences.
Within ADM:
Better response: patients with Axis I comorbidity (especially generalised anxiety disorder)
Worse response: patients with chronic depression; unemployed patients
Within CT:
Better response: patients meeting criteria for the melancholic subtype of MDD
Worse response: patients comorbid for social phobia
Mean daily paroxetine dose climbed from about 14 mg in week 1 to approximately 39 mg by week 8. During weeks 9 to 16, the mean was about 37 mg/day. Of the 101 ADM patients continuing past week 8, 47 (47%) received augmentation. A difference emerged in prescribing practices:
Vanderbilt psychiatrists followed a more aggressive augmentation strategy (maintaining higher paroxetine doses alongside augmenting agents)
Pennsylvania psychiatrists followed more conventional practice (reducing paroxetine when augmenting)
The authors examined whether this prescribing difference explained the ADM site differences and concluded it did not.
⚠️ The headline result, that CT and ADM produced identical 16-week response rates (58% each), directly challenged the prevailing APA guideline that moderate to severe depression should be treated primarily with medication.
⚠️ The site-by-treatment interaction is the study's most clinically significant finding. It means the answer to "Is CT as good as medication?" is conditional: yes, when delivered by experienced therapists; possibly not, when delivered by less experienced ones.
⚠️ The effect sizes tell the real story at 8 weeks. ADM vs placebo was 0.60 (medium); CT vs placebo was 0.44 (small to medium). ADM vs CT was only 0.16 (trivial). Both treatments worked. The difference between them was clinically unimportant.
⚠️ Do not confuse "response" with "remission." At 16 weeks, response rates were equal, but remission rates showed a site-dependent pattern (ADM favoured at Vanderbilt, no difference at Pennsylvania). This distinction matters for clinical decision-making and for exam questions.
⚠️ The finding that melancholic depression predicted better CT response is counterintuitive, since melancholia is often assumed to be more "biological" and thus more medication-responsive. Worth flagging for exam revision.
⚠️ The TDCRP comparison is central context. This study was designed as a direct response to the TDCRP's finding that ADM was superior to CT in moderate-to-severe depression. The fact that DeRubeis et al. found no such superiority, while identifying therapist experience as a likely explanation for the TDCRP's results, is a major contribution.
Effect sizes (Cohen's d conventions):
Small: 0.2 to 0.5
Medium: 0.5 to 0.8
Large: above 0.8
Key effect sizes from this study:
ADM vs placebo at 8 weeks: 0.60 (medium)
CT vs placebo at 8 weeks: 0.44 (small to medium)
ADM vs CT at 8 weeks: 0.16 (trivial)
Site-by-treatment interaction at 16 weeks: 0.36 (small)
ADM vs CT at Vanderbilt, 16 weeks: 0.57 (medium)
Power calculation:
Based on an estimated drug-placebo effect size of 0.5 SD on the HDRS (mean difference 3.5, SD 7), a cell size of 60 or more was required for 80% power to detect a difference at the .05 level, two-tailed.
Q: At 8 weeks, did CT significantly outperform placebo?
A: It depends on the analysis. In the categorical (response rate) analysis, yes: CT response (43%) was significantly higher than placebo (25%), P = .04. In the continuous HLM analysis, the advantage was only a nonsignificant trend (P = .09), with a small-to-medium effect size of 0.44.
Q: What were the overall 16-week response and remission rates for ADM and CT?
A: Response rates were identical at 58% for both treatments. Remission rates were 46% for ADM and 40% for CT, but this difference was driven by a site-by-treatment interaction, with no significant difference at Pennsylvania and a borderline advantage for ADM at Vanderbilt.
Q: What two factors explained the site-by-treatment interaction?
A: First, the Vanderbilt ADM sample had a higher prevalence of Axis I comorbidity (especially anxiety disorders), and these patients responded particularly well to paroxetine because of its anxiolytic properties. Second, the CT therapists at Vanderbilt were less experienced (2 years vs 7 to 21 years at Pennsylvania), leading to weaker CT outcomes at that site.
Q: How did this study's findings differ from the TDCRP's conclusions?
A: The TDCRP found ADM superior to both CT and placebo among more severely depressed patients, leading to APA guidelines recommending medications as first-line treatment. This study found no overall superiority of ADM over CT, with identical 16-week response rates. The authors argued that the TDCRP's results likely reflected the same pattern seen at Vanderbilt: less experienced CT therapists produced weaker outcomes, making ADM appear superior.
Q: What was the study's main conclusion about the APA guideline recommending medications over CT for moderate to severe depression?
A: The findings did not support that guideline. CT appeared to be as effective as medications for moderate to severe depression, provided it was delivered by experienced cognitive therapists. The authors suggested the evidence base for the medications-first recommendation was weaker than previously assumed.
Q: Why might paroxetine have been especially effective in patients with comorbid anxiety?
A: Paroxetine is an SSRI with established anxiolytic (anti-anxiety) properties. Patients with comorbid anxiety disorders, which were the most common Axis I comorbidity in this sample, would have benefited from this dual action, receiving treatment for both their depression and their anxiety within the same medication.
Q: What does the finding about melancholic depression and CT response suggest?
A: Patients with the melancholic subtype of MDD responded better to CT, which is counterintuitive since melancholia is traditionally viewed as more "biological" and therefore more amenable to pharmacological treatment. This finding challenges the assumption that the more "endogenous" presentations of depression are best treated with medication.
cognitive therapy, cognitive behavioural therapy, CBT, antidepressant medication, pharmacotherapy, paroxetine, SSRI, moderate to severe depression, major depressive disorder, MDD, treatment response, remission, HDRS, Hamilton Depression Rating Scale, placebo-controlled trial, site-by-treatment interaction, therapist experience, therapist expertise, therapist competence, TDCRP, APA Practice Guidelines, treatment guidelines, effect size, Cohen's d, hierarchical linear model, HLM, augmentation, lithium, desipramine, anxiolytic, Axis I comorbidity, anxiety disorders, melancholic depression, atypical depression, DeRubeis, Hollon, Beck, treatment efficacy, psychotherapy outcome research, evidence-based treatment, clinical psychology, treatment matching