Divergent Outcomes in Cognitive-Behavioral Therapy and Pharmacotherapy for Adult Depression
Although the average depressed patient benefits moderately from cognitive-behavioral therapy (CBT) or pharmacotherapy, some experience divergent outcomes. The authors tested frequencies, predictors, and moderators of negative and unusually positive outcomes.
Sixteen randomized clinical trials comparing CBT and pharmacotherapy for unipolar depression in 1,700 patients provided individual pre- and posttreatment scores on the Hamilton Depression Rating Scale (HAM-D) and/or Beck Depression Inventory (BDI). The authors examined demographic and clinical predictors and treatment moderators of any deterioration (increase ≥1 HAM-D or BDI point), reliable deterioration (increase ≥8 HAM-D or ≥9 BDI points), extreme nonresponse (posttreatment HAM-D score ≥21 or BDI score ≥31), superior improvement (HAM-D or BDI decrease ≥95%), and superior response (posttreatment HAM-D or BDI score of 0) using multilevel models.
About 5%−7% of patients showed any deterioration, 1% reliable deterioration, 4%−5% extreme nonresponse, 6%−10% superior improvement, and 4%−5% superior response. Superior improvement on the HAM-D only (odds ratio=1.67) and attrition (odds ratio=1.67) were more frequent in pharmacotherapy than in CBT. Patients with deterioration or superior response had lower pretreatment symptom levels, whereas patients with extreme nonresponse or superior improvement had higher levels.
Deterioration and extreme nonresponse and, similarly, superior improvement and superior response, both occur infrequently in randomized clinical trials comparing CBT and pharmacotherapy for depression. Pretreatment symptom levels help forecast negative and unusually positive outcomes but do not guide selection of CBT versus pharmacotherapy. Pharmacotherapy may produce clinician-rated superior improvement and attrition more frequently than does CBT.
Cognitive-behavioral therapy (CBT) and pharmacotherapy are, on average, equally and moderately efficacious acute-phase treatments for unipolar depression (1). For example, 50%−60% of patients receiving CBT or pharmacotherapy for depression respond, compared with 40% receiving pill placebo (2, 3), and the average CBT or pharmacotherapy patient has posttreatment symptom levels roughly 0.3 SD below those of patients receiving pill placebo (4, 5). However, these favorable averages may obscure some patients’ negative outcomes (6). Moreover, beyond conventionally defined response (≥50% symptom reduction) and remission (posttreatment scores ≤7 on the Hamilton Depression Rating Scale [HAM-D] ), only patients with unusually positive outcomes have no residual symptoms with attendant dysfunction (8).
In this study we focused on patient-level treatment outcomes diverging from the average. We analyzed two specific negative outcomes: deterioration (symptom severity increases from pre- to posttreatment) (9) and extreme nonresponse (severe depressive symptoms posttreatment) (10), although many others are possible (11). Complementarily, we defined two unusually positive outcomes: superior improvement (≥95% symptom reduction from pre- to posttreatment) and superior response (no depressive symptoms posttreatment). We estimated the frequency of these outcomes and tested antecedents in a large patient-level database (N=1,700) pooled from 16 randomized clinical trials comparing CBT to pharmacotherapy. Such analyses may clarify which patients are more likely to have negative or unusually positive outcomes, help match treatments to patients, and through such matching increase treatments’ overall benefits (12).
The frequency and causes of deterioration, extreme nonresponse, superior improvement, and superior response in CBT and pharmacotherapy for depression are unclear. Perhaps 5%−10% of psychotherapy patients show reliable deterioration (increases exceeding the symptom measure’s reliable change threshold), with evidence of higher risk among children than adults, in routine practice compared with randomized clinical trials, and in psychotherapy considered more broadly than CBT specifically (9, 13). Reliable deterioration may occur with similar frequency in pharmacotherapy for depression (14). The concept of extreme nonresponse was introduced in CBT research (10), but observed rates have varied widely in CBT patients, 6% (15) in one sample and 22% in another (10). Finally, based on distributions of symptom scores, superior response and superior improvement possibly occur in <10% of treated depressed patients (8, 16). The current large-sample analyses were designed to clarify these outcome frequencies in pharmacotherapy versus CBT for depression.
Patient characteristics may predict negative and unusually positive outcomes in CBT or pharmacotherapy for depression. Pretreatment markers of greater pathology (e.g., comorbidity, suicidality, unemployment, prior hospitalizations) have predicted less improvement in CBT (17, 18) and pharmacotherapy (19, 20). In addition, higher pretreatment levels of depressive symptoms, plus poorer pretreatment functioning and therapeutic alliance, have predicted extreme nonresponse in CBT (10, 15). Broadly, pretreatment symptom levels may relate directly to posttreatment symptom levels but inversely to the direction of symptom change (21). Thus, greater pretreatment severity may predict higher probabilities of extreme nonresponse and superior improvement but lower probabilities of superior response and deterioration. In contrast, demographic variables have been inconsistent predictors of CBT or pharmacotherapy outcomes (18, 22), perhaps because detecting modest effects without large individual-patient samples is difficult.
Whether treatments or illness trajectories cause particular outcomes is difficult to determine (6). Unipolar depression is often episodic, and mood reactivity within episodes is common (23). Thus, deterioration and extreme nonresponse could be adverse effects of treatment, unrelated to treatment, or even benefits of treatment (i.e., some patients would have been worse off untreated). Similarly, superior improvement and superior response may sometimes reflect “spontaneous” improvements due to biochemical, learning, or life events unrelated to treatment.
Pharmacotherapy and CBT likely share some active ingredients, including patients’ and clinicians’ expectancies of benefit and clinicians’ regular supportive interactions with patients. In addition, CBT may reduce negative emotion via top-down processes, including patients’ skill acquisition and use (24), whereas pharmacotherapy may act directly on neurochemical pathways to reduce negative emotion in bottom-up processes (25). Different hypothesized mechanisms highlight the possibility that CBT and pharmacotherapy may produce divergent outcomes for different patient subpopulations.
Data from randomized clinical trials support identification of potential causes of patient outcomes (6). Different frequencies of deterioration, extreme nonresponse, superior improvement, and superior response between patients randomly assigned to CBT or pharmacotherapy, for example, would suggest that these events may result from components of particular treatment protocols. Moreover, randomized clinical trials allow testing of moderators of outcomes. Moderators are interactions between treatments (e.g., CBT versus pharmacotherapy) and patient characteristics that provide information about individuals’ differential outcomes in one treatment versus another (18).
The goals of this study were to 1) estimate the proportions of patients with negative outcomes (deterioration, extreme nonresponse) and unusually positive outcomes (superior improvement, superior response) in CBT or pharmacotherapy for depression; 2) test predictors of these outcomes, including the type of treatment (CBT versus pharmacotherapy) and patient demographic and clinical characteristics; 3) test whether patient characteristics moderate the effects of treatment (CBT versus pharmacotherapy) on these outcomes; and 4) contrast pretreatment characteristics of patients with opposing outcomes (deterioration versus superior improvement; extreme nonresponse versus superior response).
Identification and Selection of Studies
We analyzed individual patient data pooled from randomized clinical trials comparing CBT to pharmacotherapy for adults (≥18 years) with unipolar depression, defined by major depressive disorder, dysthymic disorder, or elevated ratings of depressive symptoms on standard measures. Psychological interventions with cognitive restructuring as a core component defined CBT for study selection (1).
We accessed an existing database of trials to identify potential studies for this meta-analysis (www.evidencebasedpsychotherapies.org). The database was updated through January 2014 for the current search, and 24 studies met the inclusion criteria. We contacted study authors by e-mail to invite their participation. We requested patient-level data including pre- and posttreatment depressive symptom scores, demographic variables, and clinical characteristics. Authors provided patient-level data for 16 trials, which are described in Table S1 in the data supplement that accompanies the online version of this article.
The analyzed sample included 1,700 adults from 16 randomized trials comparing CBT (N=794) with pharmacotherapy (N=906) for depression. As shown in Table 1, most patients were women, of majority ethnicity, educated beyond high school, and of middle age. Most patients (96.8%) had major depressive disorder, 1.4% had dysthymic but not major depressive disorder, and 1.8% had no depression diagnoses but elevated depressive symptom scores before treatment (HAM-D: mean=16.19, SD=4.78; BDI: mean=21.38, SD=9.36).
|Variable||Total Number of Studies||Total Number of Patients||Mean||SD||%|
|Education >12 years||12||1,439||65.5|
|Employed full time||11||1,289||52.0|
|Major depressive disorder||16||1,700||96.8|
|Depressive symptom scores|
|17-item Hamilton Depression Rating Scale|
|Pretreatment with missing values||14||1,466||19.18||4.58|
|Pretreatment with imputed values||14||1,471||19.18||4.58|
|Posttreatment with missing values||14||1,037||8.60||6.49|
|Posttreatment with imputed values||14||1,471||9.17||6.32|
|Beck Depression Inventory-II|
|Pretreatment with missing values||14||1,400||30.44||9.79|
|Pretreatment with imputed values||14||1,491||30.34||9.81|
|Posttreatment with missing values||14||1,074||11.16||9.84|
|Posttreatment with imputed values||14||1,491||12.04||9.55|
Descriptive Statistics for Individual Patient Variables in Meta-Analysis of Divergent Outcomes in CBT and Pharmacotherapy for Depression
Depressive symptom severity.
Fourteen studies used the clinician-report HAM-D, 14 studies used the patient-report BDI, and 12 used both measures. The 17-item HAM-D yields scores from 0 to 52. The 21-item BDI yields scores from 0 to 63. We converted first-edition BDI scores (26) to the slightly higher second-edition metric (27) before analysis. Higher HAM-D and BDI scores mark greater depressive symptoms.
“Any deterioration” was defined as an increase of ≥1 point on the HAM-D or BDI from pre- to posttreatment. “Reliable deterioration” was an increase meeting the p<0.05, two-tailed, change threshold of ≥8 points on the HAM-D (28) or ≥9 points on the BDI (29). The any-deterioration group contained all patients with deterioration, whether reliable or not.
“Superior improvement” was a pre- to posttreatment decrease in HAM-D or BDI score of ≥95%, marking improvement beyond traditional response thresholds (≥50%) and contrasting with deterioration.
“Superior response” was a posttreatment HAM-D or BDI score of 0, marking the absence of measured depressive symptoms and contrasting with extreme nonresponse.
We multiply imputed missing BDI and HAM-D scores among the studies using these measures to allow intent-to-treat analyses. We used the Markov chain Monte Carlo method in PROC MI to generate 10 complete data sets, computed outcomes for each patient in the complete data sets, conducted multilevel analyses in each data set with PROC GLIMMIX, and pooled the results via PROC MIANALYZE in SAS 9.3 software (SAS Institute, Cary, N.C.). Because the patient data were nested within studies, multilevel analyses controlled for the random effect of study. Logistic regression models predicted binary outcomes (e.g., presence versus absence of deterioration) from the fixed effects of treatment (CBT versus pharmacotherapy), a predictor (e.g., age, gender), and the interaction of treatment and the predictor (i.e., the moderator effect). We contrasted dichotomous and continuous pretreatment characteristics of patients with opposing outcomes in logistic and normal linear multilevel models, respectively.
Previous Analyses of the Individual Patient Database
Whereas the current analyses focused on outcomes diverging from the mean, past studies using the individual patient database focused on averages and conventional outcomes (31, 32). Average treatment effects did not differ significantly between the eight excluded versus 16 included studies, suggesting that the included studies are representative (32). Among the included studies, pharmacotherapy relative to CBT produced statistically significantly lower average posttreatment HAM-D (but not BDI) scores, although the difference (0.88 HAM-D points) was small (32). Conventionally defined response (HAM-D decrease of ≥50%) and remission (posttreatment HAM-D ≤7), respectively, did not differ significantly between pharmacotherapy (63.4%, 51.0%) and CBT (57.5%, 47.0%) (32). Finally, gender (31) and symptom severity (32) did not moderate CBT versus pharmacotherapy’s effects on posttreatment symptom averages.
Frequency and Treatment Differences in Outcomes
Table 1 displays descriptive statistics for the variables analyzed. With the random effect of study controlled for, attrition (defined as posttreatment HAM-D and BDI scores both missing) was more frequent in pharmacotherapy (18.5%) than in CBT (12.0%) (odds ratio=1.67, p<0.01) but did not correlate significantly with pretreatment HAM-D or BDI score (p>0.25). Patients began treatment with moderate to severe depressive symptoms and ended with mild symptoms, on average, according to their scores on the HAM-D and BDI.
Figure 1 shows scatterplots of HAM-D and BDI scores pre- and posttreatment with regions marking the analyzed outcomes (see Figure S1 in the online data supplement for plots separating CBT and pharmacotherapy). Visual analysis of the scatterplots suggested that patients with deterioration versus superior improvement, and extreme nonresponse versus superior response, marked opposite tails of the symptom-change and posttreatment symptom-level continua, respectively.
The correlation between continuous HAM-D and BDI scores was moderately high (r=0.85) for pooled pre- and posttreatment observations. Similarly, the percentage agreement between the HAM-D and BDI scores was moderately high in identifying patients with versus without deterioration (92.3%), reliable deterioration (98.2%), extreme nonresponse (94.3%), superior improvement (88.8%), and superior response (92.3%). Nonetheless, clinician ratings (HAM-D) and patient ratings (BDI) sometimes differed regarding the presence of the analyzed outcomes (Figure 2).
Our first goal was to estimate the frequency of outcomes (see Table 2). Overall, 5%−7% of patients showed any deterioration (pre- to posttreatment increase ≥1 point on the HAM-D or BDI), 1% showed reliable deterioration (increase of ≥8 HAM-D or ≥9 BDI points), 4%−5% showed extreme nonresponse (posttreatment score on the HAM-D ≥21 or BDI ≥31), 6%−10% showed superior improvement (HAM-D or BDI decrease ≥95%), and 4%−5% showed superior response (posttreatment HAM-D or BDI=0). About 13% of patients had any negative outcome (deterioration or extreme nonresponse), and 15% had an unusually positive outcome (superior improvement or superior response) on the HAM-D or BDI.
|Estimated Proportion of Patients, Derived From Multilevel Logistic Regression Models Adjusting for Random Effect of Study|
|Outcomeb||%||95% CI||%||95% CI||%||95% CI||Treatment Difference (p)|
|HAM-D||7.1||4.7, 10.7||7.7||4.8, 12.0||6.6||4.1, 10.5||0.48|
|BDI||5.2||3.4, 8.0||5.4||3.3, 8.8||5.0||3.0, 8.2||0.78|
|HAM-D||0.9||0.4, 1.8||1.2||0.5, 2.6||0.6||0.2, 1.9||0.32|
|BDI||1.1||0.5, 2.5||0.8||0.3, 2.5||1.4||0.6, 3.4||0.37|
|HAM-D||5.4||3.7, 7.7||5.3||3.4, 8.1||5.5||3.6, 8.2||0.88|
|BDI||4.3||2.9, 6.4||4.6||2.9, 7.3||4.1||2.5, 6.5||0.62|
|Any negative outcome||13.3||9.9, 17.6||14.1||10.1, 19.2||12.7||9.1, 17.6||0.54|
|Unusually positive outcomes|
|HAM-D||6.4||4.4, 9.2||4.8||2.9, 7.7||7.7||5.2, 11.3||0.04|
|BDI||9.8||6.4, 14.7||8.9||5.6, 13.9||10.6||6.8, 16.1||0.32|
|HAM-D||3.9||2.6, 5.9||3.2||1.8, 5.5||4.5||2.9, 7.0||0.23|
|BDI||5.4||3.0, 9.4||4.5||2.4, 8.3||6.2||3.4, 10.9||0.17|
|Any unusually positive outcome||15.1||10.6, 21.0||12.8||8.5, 18.7||16.9||11.7, 23.7||0.06|
Frequencies of deterioration, extreme nonresponse, and superior response did not differ significantly between CBT and pharmacotherapy. Treatment with pharmacotherapy significantly increased patients’ chances, relative to CBT, of superior improvement on the HAM-D (odds ratio=1.67) but not on the BDI. (Among the five trials including pill placebo arms, deterioration, extreme nonresponse, superior improvement, and superior response did not differ significantly among CBT, pharmacotherapy, and placebo, perhaps owing to lower statistical power; see Table S2 in the online data supplement. However, the frequency of any negative outcome was lower with pharmacotherapy than with placebo [odds ratio=0.59].)
Prediction and Moderation of Outcomes
Our second and third goals were to test predictors and moderators of outcomes. We tested predictors (main effects) and moderators (interaction effects) in a series of multilevel logistic regression analyses (see Table S3 in the data supplement for regression coefficients). The patient characteristics listed in Table 1 did not significantly moderate differential treatment effects (CBT versus pharmacotherapy), but several statistically significant predictors were evident.
First, a lower pretreatment level of depressive symptoms predicted deterioration, but only within measures (HAM-D or BDI). Patients with lower pretreatment HAM-D scores had a higher chance of any (odds ratio=1.92) and reliable (odds ratio=2.46) deterioration on the HAM-D. (Odds ratios for continuous predictors reflect changes of 1 SD on the predictor. Table 1 shows the SDs.) Similarly, patients with lower pretreatment BDI scores had a higher chance of any (odds ratio=2.48) and reliable (odds ratio=2.19) deterioration on the BDI.
Second, higher pretreatment symptom severity predicted extreme nonresponse both within and between measures. Patients with higher pretreatment HAM-D scores had a higher chance of extreme nonresponse on the HAM-D (odds ratio=1.89) and BDI (odds ratio=1.48). Similarly, patients with higher pretreatment BDI scores had a higher chance of extreme nonresponse on the BDI (odds ratio=2.25) and HAM-D (odds ratio=1.65).
Third, younger age (odds ratio=1.36) and higher pretreatment HAM-D scores (odds ratio=1.33) predicted superior improvement on the HAM-D (odds ratios for continuous predictors, e.g., age, reflect changes of 1 SD on the predictor; Table 1 shows SDs). Finally, full-time employment (odds ratio=1.98) and lower pretreatment BDI scores (odds ratio=1.46) predicted superior response on the BDI.
We followed up the pretreatment symptom scores’ significant prediction of outcomes within the HAM-D by testing specific depressive symptoms captured by individual scale item scores (eight studies provided item-level HAM-D data). Lower pretreatment depressed mood (odds ratio=1.60), less guilt (odds ratio=1.37), and fewer general somatic symptoms (odds ratio=1.56) predicted any deterioration. (Odds ratios for HAM-D items reflect changes of 1 point on the item.) Individual HAM-D items did not predict reliable deterioration significantly. Initial insomnia (odds ratio=1.58), middle insomnia (odds ratio=1.57), and weight loss (odds ratio=1.70) predicted extreme nonresponse. Finally, gastrosomatic symptoms predicted superior improvement (odds ratio=1.51).
Contrasts of Outcome Groups
Our fourth goal was to contrast the pretreatment characteristics of patients reaching opposing outcomes, deterioration versus superior improvement and extreme nonresponse versus superior response. As shown in Table 3, patients with any deterioration on the HAM-D had lower pretreatment HAM-D scores than patients with superior improvement. Similarly, patients with any deterioration on the BDI had lower pretreatment BDI scores than did patients with superior improvement. Finally, as shown in Table 4, patients with extreme nonresponse on the HAM-D or BDI had higher pretreatment scores on both symptom measures than did patients with superior response.
|Group With Any Deterioration Versus Group With Superior Improvement on HAM-D||Group With Any Deterioration Versus Group With Superior Improvement on BDI|
|Pretreatment Variable||Any Deterioration||Superior Improvement||Any Deterioration||Superior Improvement|
|Education >12 years||78.2||8.4||70.6||9.2||71.4||10.1||79.2||7.3|
|Group With Extreme Nonresponse Versus Group With Superior Response on HAM-D||Group With Extreme Nonresponse Versus Group With Superior Response on BDI|
|Pretreatment Variable||Extreme Nonresponse||Superior Response||Extreme Nonresponse||Superior Response|
|Education (>12 years)||68.2||10.1||70.2||10.1||68.7||10.1||80.5||7.7|
Among 1,700 depressed patients from 16 randomized clinical trials comparing CBT to pharmacotherapy, we found that 13% experienced negative and 15% experienced unusually positive symptomatic outcomes. The analyzed negative outcomes were any deterioration (5%−7% had increases of ≥1 HAM-D or BDI points), reliable deterioration (1% had increases ≥8 HAM-D or ≥9 BDI points), and extreme nonresponse (4%−5% had posttreatment HAM-D scores of ≥21 or BDI scores of ≥31). Negative outcomes may be more common in practice settings outside of research (12, 14) because frequent assessment of patients, feedback to clinicians about patients’ progress, and monitoring of treatment fidelity (e.g., clinicians’ competence and protocol adherence) in many randomized clinical trials parallel interventions shown to reduce deterioration (13). Patients’ unusually positive outcomes included superior improvement (6%−10% had HAM-D or BDI score decreases ≥95%) and superior response (4%−5% had posttreatment HAM-D or BDI scores of 0). Because few patients receiving CBT or pharmacotherapy reached absent (superior response) or minimal (superior improvement) residual depressive symptoms, our results highlight the potential value of continuing, sequencing, or augmenting treatments (19, 33).
Treatment with pharmacotherapy versus CBT increased patients’ odds of superior improvement from the clinician’s perspective (HAM-D: 7.7% versus 4.8%) but not from the patient’s (BDI). Pharmacotherapy also predicted greater attrition relative to CBT (18.5% versus 12.0%), and among the subset of studies with placebo arms, fewer negative outcomes overall relative to pill placebo (16.2% versus 24.8%). If replicated, more frequent superior improvement during pharmacotherapy than in CBT may signal operation of an unknown moderator (e.g., to be revealed by genotyping) among patients who tolerate treatment (e.g., do not drop out). In addition, younger age (among adults) predicted superior improvement on the HAM-D, whereas employment predicted superior response on the BDI. These latter findings fit broader patterns of better outcomes for patients with less (versus more) severe pathology (17–20). Treatment modality did not change patients’ odds of deterioration, extreme nonresponse, or superior response on the HAM-D or BDI.
Pretreatment symptom levels predicted outcomes and varied significantly between patients with negative versus unusually positive outcomes. In general, pretreatment symptom levels related directly to posttreatment symptom levels and inversely to the direction of symptom change. More specifically, lower pretreatment symptom severity increased risk for deterioration, and patients who deteriorated had lower pretreatment symptom levels than did patients with superior improvement, within measures (HAM-D or BDI). In addition, higher pretreatment symptom levels increased the odds of superior improvement within the HAM-D only. Specific depressive symptoms (HAM-D items) most relevant to these predictions included less depressed mood, less guilt, and fewer general somatic symptoms (deterioration) and more gastrosomatic symptoms (superior improvement).
Psychotherapy may produce negative outcomes, or fail to evoke unusually positive outcomes, when psychotherapists are confrontational and over-interpret common experiences and patient characteristics as pathological (34). However, lack of generalization across measures in our analyses of deterioration and superior improvement also suggests methodological artifacts, such as regression to the mean within measures. The patient (BDI) and clinician (HAM-D) measures may also capture somewhat different information (35) and have finite reliability and validity (36), as do all measures. Finally, patients with lower pretreatment symptom scores may be entering treatment as a depressive episode is waxing, and their deterioration and lack of superior improvement during treatment might reflect the natural course of illness.
Higher pretreatment depressive symptom levels also predicted extreme nonresponse, and patients with extreme nonresponse had higher pretreatment symptom levels than did patients with superior response, both within and between the HAM-D and BDI. In addition, lower pretreatment symptom levels increased patients’ odds of superior response on the BDI. Specific depressive symptoms (HAM-D items) that predicted extreme nonresponse on the HAM-D included initial insomnia, middle insomnia, and weight loss. Although CBT and pharmacotherapy benefit many severely depressed patients (37), our analyses suggest that severe depression at the beginning of treatment adds risk for severe depression and reduces the likelihood of being symptom-free at the end of pharmacotherapy or CBT.
The pretreatment symptom levels and patient characteristics analyzed here (age, gender, education, ethnicity, pretreatment symptom severity, double depression, comorbidity) did not moderate treatment effects on deterioration, extreme nonresponse, superior improvement, or superior response. Consequently, the current results do not support differential treatment selection (CBT versus pharmacotherapy) from symptom levels and patient characteristics to prevent negative or potentiate unusually positive outcomes. Other patient characteristics or events outside of treatment may account for these outcomes and are potential targets for future research. Even so, empirical guidance for selecting CBT or pharmacotherapy for a given patient based on average posttreatment symptom scores is available (16, 38).
On the basis of the current findings, we recommend assessing symptom levels frequently and longitudinally (especially among patients with high pretreatment severity) and pursuing rapid corrective action (e.g., increased session frequency, switching or augmenting treatment) if patients do not progress adequately (12). However, other negative outcomes and adverse events, including discrete behaviors (e.g., suicide), psychosocial outcomes (e.g., divorce; domestic violence), dropping out of treatment, and medication side effects (e.g., sexual dysfunction) arguably are distinct from depressive symptom levels (11) and may require different preventive efforts. Similarly, positive outcomes distinct from depressive symptoms are also important therapeutic targets (e.g., social-interpersonal functioning) (39).
The current study has additional limitations that temper our conclusions. For example, although the sample was large, low base rates of divergent outcomes and attrition may have limited detection of predictors and moderators. In addition, many important patient characteristics (e.g., personality profiles, depressive cognitive content, therapeutic alliance, social support) were not analyzed and could be targets for future research on predictors and moderators of divergent outcomes (10, 40). Moreover, our results from randomized clinical trials may not generalize fully to routine clinical practice (12, 14). Similarly, generalization of our findings to other negative outcomes (e.g., adverse events, side effects) or positive outcomes (e.g., improvement in work or social functioning) not studied here is unknown. Finally, our analyses focused on trials of CBT versus pharmacotherapy and do not address combinations, sequences, dissemination, or the quality of implementation of treatments.
Preventing negative and potentiating unusually positive outcomes may improve the overall efficacy of CBT and pharmacotherapy for depression. In this effort, future research might profitably test replication and mechanisms of the current findings. For example, greater odds of clinician-rated superior improvement among younger (versus older) adults and those treated with pharmacotherapy (versus CBT) and greater odds of patient-rated superior response for patients with full-time (versus less) employment possibly reflect measurable intrapatient, environmental, or interacting mechanisms. Similarly, specific depressive symptoms linked with clinician-rated extreme nonresponse (insomnia, weight loss), deterioration (less depressed mood, less guilt, fewer general somatic symptoms), and superior improvement (gastrosomatic symptoms) may reflect sampling or measurement error but may also reveal patient subpopulations with different treatment-response trajectories.
The current results clarify expectations for outcomes in acute-phase CBT or pharmacotherapy for adult depression. First, whereas the majority of patients responded by conventional standards (32), we found that a few patients (13%) had negative outcomes and some (15%) had very positive outcomes. Second, pretreatment symptom levels help forecast negative and unusually positive outcomes: Patients with severe pretreatment symptoms are more likely to have large drops in symptoms (superior improvement) and unlikely to get worse (deteriorate), but they are also more likely to end treatment with severe symptoms (extreme nonresponse) and are unlikely to end treatment symptom-free (superior response). Conversely, patients with milder pretreatment symptoms are more likely to end treatment symptom-free (superior response) and unlikely to end treatment with severe symptoms (extreme nonresponse), but they are also more likely to get worse (deteriorate) and less likely to have large drops in symptoms (superior improvement). Finally, choosing pharmacotherapy versus CBT may increase patients’ odds of both discontinuing treatment and clinician-rated superior response.
2 : Antidepressants for major depressive disorder in patients with a co-morbid axis-III disorder: a meta-analysis of patient characteristics and placebo response rates in randomized controlled trials. Int Clin Psychopharmacol 2011; 26:69–74Crossref, Medline, Google Scholar
5 : A meta-analysis of factors impacting detection of antidepressant efficacy in clinical trials: the importance of academic sites. Neuropsychopharmacology 2012; 37:2830–2836Crossref, Medline, Google Scholar
9 : Clinical significance of the National Institute of Mental Health Treatment of Depression Collaborative Research Program data. J Consult Clin Psychol 1995; 63:321–326Crossref, Medline, Google Scholar
12 : Predicting negative treatment outcome: methods and estimates of accuracy, in Prevention of Treatment Failure: The Use of Measuring, Monitoring, and Feedback in Clinical practice. Washington, DC, American Psychological Association, 2010, pp 83–105Crossref, Google Scholar
13 : The efficacy and effectiveness of psychotherapy, in Bergin and Garfield’s Handbook of Psychotherapy and Behavior Change, 6th ed. Edited by Lambert MJ. New York, Wiley, 2013, pp 169–218Google Scholar
14 : Treatment as usual (TAU) for depression: a comparison of psychotherapy, pharmacotherapy, and combined treatment at a large academic medical center. Psychotherapy (Chic) 2013; 50:110–118Crossref, Medline, Google Scholar
17 : Acute phase cognitive therapy for recurrent major depressive disorder: who drops out and how much do patient skills influence response? Behav Res Ther 2013; 51:221–230Crossref, Medline, Google Scholar
19 : Selecting among second-step antidepressant medication monotherapies: predictive value of clinical, demographic, or first-step treatment features. Arch Gen Psychiatry 2008; 65:870–880Crossref, Medline, Google Scholar
20 : Clinical factors associated with treatment resistance in major depressive disorder: results from a European multicenter study. J Clin Psychiatry 2007; 68:1062–1070Crossref, Medline, Google Scholar
27 : Beck Depression Inventory, Second Edition Manual. San Antonio, Tex, Psychological Corp, 1996Google Scholar
31 : Gender as predictor and moderator of outcome in cognitive behavior therapy and pharmacotherapy for adult depression: an “individual patient data” meta-analysis. Depress Anxiety 2014; 31:941–951Crossref, Medline, Google Scholar
32 : Baseline depression severity as moderator of depression outcomes between cognitive behavioral therapy vs pharmacotherapy: an individual patient data meta-analysis. JAMA Psychiatry 2015; 72:1102–1109Crossref, Medline, Google Scholar
33 : Preventing depressive relapse and recurrence in higher-risk cognitive therapy responders: a randomized trial of continuation phase cognitive therapy, fluoxetine, or matched pill placebo. JAMA Psychiatry 2013; 70:1152–1160Crossref, Medline, Google Scholar
36 : Rating scales for depression, in Handbook of Clinical Rating Scales and Assessment in Psychiatry and Mental Health. Edited by Baer L, Blais MA. Totowa, NJ, Humana, 2010, pp 7–35Google Scholar