Major depressive disorder, which has a point prevalence of 5% in adolescents, is associated with significant morbidity and family burden (1). It also is a major contributor to suicidal behavior and completed suicide (2). Hence, improvements in the treatment of major depressive disorder in adolescents should be of significant public health value. To this end, the National Institute of Mental Health in 1999 funded the Treatment for Adolescents With Depression Study (TADS) (3). TADS is a randomized, controlled trial that is intended to evaluate the effectiveness of three active treatments for adolescents with major depression: clinical management with fluoxetine, cognitive-behavioral therapy (CBT), their combination (fluoxetine plus CBT), and, acutely only, clinical management with pill placebo as a control condition.

Previous publications from the TADS described the study aims, rationale, and design (3), characteristics of the study group (4), and the results of acute (5) and long-term (6) randomized treatment. To briefly summarize, CBT alone was less effective than combined treatment or fluoxetine and not significantly more effective than placebo at week 12 (5). Secondary analyses at 12 weeks (7) also showed superiority for combined treatment with respect to speed of response (8), quality of life and functioning (9), remission (10), and overall safety (11). CBT “caught up” with fluoxetine at week 18, and combined treatment reached maximum medical benefit at that point, several months earlier than fluoxetine or CBT, with all treatments converging on an approximately 80% response rate by week 36 (6). In a recent paper from the TADS, Rohde and colleagues found that the majority of adolescents who had not achieved sustained response during acute treatment did achieve that level of improvement during continuation and maintenance therapies, with CBT and combined treatment showing greater persistence of sustained response than fluoxetine (12).

In epidemiological studies, most children and adolescents who experience an episode of depression typically recover within 1 to 2 years from baseline (13, 14), and approximately one-half of those who recover subsequently relapse (15). In the TADS study group, the mean episode duration at intake was greater than 1 year, and more than one-half had received previous treatment within the current episode, suggesting that in the absence of evidence-based treatment, depression persists (4). Likewise, in follow-ups of psychotherapy (16) or pharmacotherapy (17) randomized, controlled trials, 25% to 50% of treated subjects experience a relapse within 6 months to 2 years after treatment. In one study showing an advantage for CBT over other psychosocial treatments (18), there were no differences in clinical outcomes, including relapse, at 2-year follow-up (16). Thirty percent of the subjects had a recurrence of depression, and 21% were depressed during at least 80% of the follow-up period. In the only trial of maintenance psychotherapy in depressed teens of which we are aware, CBT booster sessions did not reduce the rate of recurrence but did appear to accelerate recovery among participants who were still depressed at the end of the acute phase (19). In a recent study, Emslie and colleagues, using a placebo-controlled discontinuation design, found a higher rate of relapse (69%) in those who discontinued fluoxetine after 12 weeks of treatment compared to those who continued to take fluoxetine for an additional 6 months (42%) (20). Taken together, these findings strongly suggest that in the absence of evidence-based treatment or with only short-term treatment, depression is a waxing and waning illness and that new intervention strategies are required to promote long-term reductions in depressive symptoms (21).

To this end, TADS was designed to test the hypothesis that longer-term treatment would improve outcomes and promote stability of remission. We now report effectiveness outcomes for subjects assigned initially to combined treatment, fluoxetine only, and CBT only across 1 year (TADS stage IV) of naturalistic follow-up in which TADS treatments were stopped and treatment was available openly in the community. For ethical and feasibility reasons, the subjects assigned to placebo were treated openly after week 12, and the placebo group was not included in these analyses by design.

The randomized, controlled trial was conducted in 13 academic and community sites in the United States. Stages I, II, and III consisted of 12, 6, and 18 weeks of acute, consolidation, and continuation treatment, respectively. Following discontinuation of TADS treatments at the end of stage III, stage IV consisted of 1 year of naturalistic follow-up. The methods for the TADS have been extensively documented in previous publications (3–6, 22–29), and only the aspects of the study that are directly relevant to the stage IV analyses will be presented here.

End of Stage III

Immediately before the week 36 visit, which constituted the end of active treatment and the beginning of stage IV, the site team reviewed each subject’s clinical status and assigned an end-of-treatment score on the severity measure of the Clinical Global Impression (CGI) (30) and recommended additional treatment as follows.

Level 1

Participants with a summary CGI severity score of 1 or 2 (normal or borderline ill) and who were otherwise well, i.e., did not need treatment for other disorders or problems, received a recommendation to discontinue all treatment unless 1) in the opinion of the site team, continued treatment was indicated, for example, because of a history of relapse, or 2) there was a strong subject or family preference for continuing treatment.

Level 2

Those with a CGI severity rating of 3 (mildly ill) or who required other treatments were given a recommendation (as appropriate clinically) to continue TADS-like treatment(s) and to add other treatments if necessary.

Level 3

Those with a CGI severity score of 4 or worse (moderately ill or worse) received whatever recommendation the treating site deemed clinically and ethically appropriate.

With respect to clinical referrals, each site used a standardized resource list of referrals that included psychologists, social workers, and psychiatrists who in the opinion of the referring site were committed to evidence-based mental health care.

Stage IV Procedure

In stage IV, subjects were assessed by their independent evaluator, blind to treatment condition, with an abbreviated battery at 3 and 9 months and with a complete battery at 6 months and at the end of 1 year. A debriefing session was held at the end of each assessment and up to two “adjunctive services and attrition prevention” sessions (31) were provided. In addition, subjects were told that if their depressive symptoms worsened between assessment visits, they should call their primary clinician to schedule an additional visit (within 3 days) for assessment. If the subject appeared to the clinician to be relapsing, a visit with the independent evaluator was scheduled and the subject was referred for open community treatment according to previously established referral procedures.

TADS employed two primary outcome measures: 1) the scalar Children’s Depression Rating Scale—Revised total score (32) and 2) responder status, which was defined as being rated much improved or very much improved on the CGI improvement scale (30). In addition to responder status, we examined remission status by using a cutoff score of 28 or less on the Children’s Depression Rating Scale (10, 33). Data from the adolescent self-report, the Reynolds Adolescent Depression Scale (34), were included because of the prominent place accorded adolescent self-report in the CBT literature. To ascertain suicidal ideation, we employed a “flag” score of 31 or greater on the Suicidal Ideation Questionnaire—Junior High School Version (35), which denotes suicidal ideation of sufficient severity to warrant prompt clinical evaluation. Psychometric properties and intercorrelations for all measures were acceptable (4). Since the subjects were not receiving study treatment and were not seen except for scheduled assessment visits, we did not assess (and do not report) the incidence of new suicidal events or other serious adverse events during stage IV follow-up.

The participants and at least one parent provided informed consent or assent. Institutional review boards at each site approved and monitored the protocol. Safety monitoring was done quarterly by a National Institute of Mental Health data safety and monitoring board.

Stage IV Statistical Analysis Plan

The primary effectiveness and safety analyses were conducted by using an intention-to-treat principle, by which the analysis included all subjects randomly assigned to the three active treatments (combined treatment, fluoxetine, and CBT) regardless of adherence to study treatment or procedures. Longitudinal analyses of the primary and secondary scalar outcome measures were conducted by using quadratic random coefficients regression models designed to assess individual subject trajectories from baseline to the end of stage IV. Specifically, the impact of treatment on outcome was modeled as a function of fixed effects for treatment, natural log of time, and clinical site (and their two-way and three-way interaction terms) as well as the random effects for subject and subject-by-time interactions. Site was retained, but its interactions were omitted from the final model owing to statistical nonsignificance.

To test for treatment differences in remission rates over time and to generate estimated probabilities of remission in the three treatment arms from week 6 to the end of stage IV, we used a generalized linear mixed model for binary outcomes, applying the generalized estimating equation method. This model included treatment, time, treatment-by-time interaction, and site. Baseline response scores were not included because all subjects met the criteria for DSM-IV major depressive disorder before treatment initiation. For the dichotomized outcome of response based on CGI-rated improvement, a logistic regression model applying a last-observation-carried-forward imputation method and controlling for site was conducted to examine the effects of randomized treatment on response at the end of stage IV. Although a longitudinal analysis employing a generalized estimating equation approach was preferred, such a model was not possible owing to the high treatment response rate in all three treatment arms by month 6 of the stage IV follow-up period.

For hypotheses stipulated in the statistical plan, the nominal significance level was set a priori at a two-tailed type I error rate of 0.05 for the omnibus tests designed to compare all three treatment arms. If the treatment or a treatment-by-time (linear or quadratic) interaction term was significant, then pairwise comparisons were conducted by means of a closed test procedure with an alpha of 0.05 for each test. In the event of a nonsignificant omnibus result, a sequential rejective approach was planned to safeguard the type II error rate. Analyses were conducted by using SAS 8.2 software (SAS Institute, Cary, N.C.), with PROC MIXED employed for the random regressions on the scalar outcomes, PROC GENMOD for applying the generalized estimating equation method to the remission outcome, and PROC LOGISTIC for the CGI improvement outcome.

Subject Characteristics

The full TADS study group (N=439) has been extensively described and compared to both clinical and epidemiological samples (4). The participants ranged in age from 12 to 17 years and had a primary DSM-IV diagnosis of major depressive disorder. The baseline clinical and demographic characteristics of the 327 adolescents assigned to an active treatment arm (fluoxetine, CBT, or combined treatment) are as follows. The mean age was 14.6 years (SD=1.5), 45.0% were male, 74.0% were Caucasian, 11.3% were African American, and 9.8% were Hispanic. Depression ranged from mild to severe depression. The total raw scores on the Children’s Depression Rating Scale ranged from 45 to 98, and the mean score at entry was 60.8 (SD=10.3), which translates to a T score of 75.7 (SD=6.5), which is above the 98th percentile on the Children’s Depression Rating Scale. Among the 234 adolescents who completed a week 36 assessment (the entry point into stage IV), there were no statistically significant between-treatment differences in demographic variables, family income, functioning, or pattern of comorbidity, with three exceptions. Before treatment, more subjects receiving fluoxetine (four of 73, or 5.5%) than combined treatment (one of 78, or 1.3%) or CBT (none of 76, or 0.0%) had a score on the Suicidal Ideation Questionnaire above 31, indicating clinically significant suicidality (Fisher’s exact test, overall p=0.05). There was also a higher rate of current disruptive behavior disorder for those receiving combined treatment (12 of 81, or 14.8%) than for the fluoxetine group (four of 73, or 5.5%) or the CBT group (three of 76, or 3.9%) (Fisher’s exact test, overall p=0.04). In addition, there was a higher rate of current anxiety disorder for CBT (eight of 76, or 10.5%) relative to combined treatment (two of 81, or 2.5%) or fluoxetine (one of 73, or 1.4%) (Fisher’s exact test, overall p=0.03).

Subject Disposition

A total of 2,804 subjects were screened by phone (gate A). Of these, 1,088 signed consent for evaluation of inclusion and exclusion criteria (gate B) and 439 completed the baseline assessment and were subsequently randomly assigned to treatment (gate C). This report concerns only the 327 subjects randomly assigned to active treatment conditions: combined treatment (N=107), fluoxetine (N=109), and CBT (N=111). Of these 327 subjects, 234 (71.6%) remained in the study at week 36. Of these, 178 (54.4%) completed stage III in the treatment condition to which they had been randomly assigned. Sixty-six percent of the eligible subjects (N=215) participated in at least one stage IV assessment. During stage IV, 49 (45.8%) of the 107 in combined treatment, 46 (42.2%) of the 109 receiving fluoxetine, and 53 (47.7%) of the 111 in the CBT group completed all four stage IV assessments. The proportions were somewhat higher for those who completed an assessment at month 12: 65 (60.7%) of the 107 in combined treatment, 56 (51.4%) of the 109 receiving fluoxetine, and 67 (60.4%) of the 111 CBT subjects. During stage IV, 215 (65.7%) of the 327 families provided information on mental health services. Data collapsed across stage IV indicated that 2.8% of the subjects received inpatient mental health care, 34.0% received outpatient mental health care, 34.0% received services from a nonphysician mental health provider, and 17.2% received services from a physician mental health provider. There were no differences between treatment groups in the type of services received in stage IV.

Benefits

Table 1 presents the adjusted mean scores on the Children’s Depression Rating Scale, Reynolds Adolescent Depression Scale, and Suicidal Ideation Questionnaire, as well as the response and remission rates by the stage IV assessment point for each treatment and the total group. Figure 1 illustrates the trajectory of scores on the Children’s Depression Rating Scale for the three active treatments across stages I–IV. Figure 2 illustrates comparable trajectories for the Reynolds Adolescent Depression Scale and Suicidal Ideation Questionnaire, as well as the response and remission measures for the three active treatments across stage IV. When presented in either tabled or graphical format, in all cases the data show that improvement in the group average scores continued on a positive trajectory during stage IV follow-up.

The random regression analyses on the Children’s Depression Rating Scale total scores across the entire study identified a statistically significant linear time effect (F=202.07, df=1, 292, p<0.001), time-by-treatment interaction (F=12.13, df=2, 292, p<0.001), and quadratic time-by-treatment interaction (F=11.72, df=1, 264, p<0.001). The effect of site was significant (F=2.46, df=12, 311, p=0.005). Planned contrasts at week 36 and at stage IV months 3, 6, 9, and 12 identified superiority for combined treatment relative to CBT at week 36 (F=5.08, df=1, 267, p=0.03); all of the stage IV contrasts were nonsignificant.

With a positive response defined as a CGI improvement score of 1 (very much improved) or 2 (much improved), logistic regression using the last observation carried forward indicated that the effect of treatment on the response rate at stage IV month 12 was not statistically significant (χ²=4.20, df=2, p=0.13), nor was site (χ²=9.70, df=9, p=0.38). The adjusted rates of response at month 12 were 82.2% for combined treatment, 75.2% for fluoxetine, and 70.3% for CBT. Planned contrasts at month 12 identified no significant differences among the treatment groups.

When a cutoff score of 28 or less on the Children’s Depression Rating Scale was used to define remission, generalized estimating equation analyses indicated that the effect of time (χ²=123.25, df=8, p<0.001) and the treatment-by-time interaction (χ²=30.67, df=16, p=0.02) were statistically significant; the effect of site also was significant (χ²=25.09, df=9, p=0.003). All but one of the planned stage IV contrasts were nonsignificant; combined treatment was superior to fluoxetine at month 6 (χ²=4.15, df=1, p=0.05).

Random regression analyses of the Reynolds Adolescent Depression Scale total score across time identified a statistically significant linear time effect (F=75.87, df=1, 286, p<0.001), time-by-treatment interaction (F=13.34, df=2, 286, p<0.001), and quadratic time-by-treatment interaction (F=11.60, df=2, 261, p<0.001). The effect of site was significant (F=2.62, df=12, 314, p=0.002). Planned contrasts at weeks 36 and stage IV months 3, 6, 9, and 12 identified superiority for combined treatment relative to CBT at week 36 (F=5.22, df=1, 282, p=0.03); all other stage IV contrasts were nonsignificant.

Random regression analyses of the Suicidal Ideation Questionnaire total score across time identified a statistically significant linear time effect (F=49.53, df=1, 273, p<0.001), quadratic time effect (F=4.53, df=1, 228, p<0.04), and time-by-treatment interaction (F=3.93, df=2, 273, p<0.03). The effect of site was nonsignificant. Planned contrasts at weeks 36 and stage IV months 3, 6, 9, and 12 were nonsignificant.

Loss of Benefits

To evaluate the extent to which subjects deteriorated during stage IV, we examined worsening on the CGI severity scale, Suicidal Ideation Questionnaire, and remission measure. According to the CGI severity rating at week 36, 81.7% of the study group were minimally impaired (CGI score ≤3). During stage IV, 13.1% deteriorated to mildly to extremely impaired (score=4–7): 15.1% in combined treatment, 9.9% in the fluoxetine group, and 14.5% in the CBT group. There were no statistically significant differences across the three treatment groups. According to the suicidality flag, defined as a total score of 31 or greater on the Suicidal Ideation Questionnaire, 91.1% of the sample did not report clinically significant suicidality at week 36. During stage IV, 6.4% developed clinically significant suicidality: 5.9% of those in combined treatment, 7.6% in the fluoxetine group, and 6.4% in the CBT group. There were no statistically significant differences across the three treatment groups. Using a score of 28 or less on the Children’s Depression Rating Scale as the remission threshold, we found that 59.9% of the subjects were in remission at week 36. During stage IV, 32.8% of those in combined treatment, 28.6% of the fluoxetine group, and 30.4% of the CBT group lost remission status at some point. There were no statistically significant differences across the three treatment groups.

Relative to our previous reports for short-term treatment (week 12) (5) and longer-term treatment (week 36) (6), the new data from stage IV on benefits and harms over 1 year of naturalistic follow-up provide important new information. First, the stage IV analyses replicate previous findings showing that combined treatment reaches maximum medical benefit earlier (week 18) than fluoxetine and CBT (weeks 30 and 36, respectively) and that 9 months of treatment is superior to 12 weeks irrespective of treatment modality. Second, in contrast to previous reports on epidemiological and treatment samples showing high rates of clinical deterioration after short-term treatment, longer-term treatment results in sustained, clinically meaningful improvement even when active treatment is discontinued. Granting that loss of benefit as we define it represents very mild deterioration, 6% to 33% of the TADS patients, depending on the measure chosen, worsened following the cessation of TADS treatment. While this is smaller than the 69% relapse rate identified in a recently completed study using placebo during withdrawal of fluoxetine (20), it remains substantial and indicates a need for further improvements in long-term treatment of major depressive disorder in youth. Third, although rarely statistically significant, combined treatment generally maintained numerical superiority relative to CBT and fluoxetine on many if not all stage IV endpoints. Thus, the stage IV results are in line with earlier reports from the TADS showing a meaningful advantage for combined treatment over both monotherapies.

Limitations

Despite the inevitable methodological concerns that arise in effectiveness research (for a thorough discussion, see reference 7), TADS is generally thought to represent the state of the art in comparative treatment trials for this patient population (36–39). Nevertheless, interpretation of the stage IV results is compromised by three primary limitations: 1) substantial missing data in stage IV, 2) the lack of untreated and actively treated comparison groups, and 3) lack of detail regarding service utilization in stage IV.

While the analyses using only the subjects who actually entered stage IV and had at least one stage IV data point (unadjusted scores; data not shown) provided results identical to those in the intent-to-treat analyses (adjusted predicted scores), indicating that the findings are not driven by the analytic model, the main limitation of the present report is the extent of missing data in stage IV, since it is possible that greater subject retention would have contributed to lower stability in maintenance of response.

Without an untreated (or a placebo-treated) comparison group, it is impossible to know for certain that the TADS treatments benefited subjects beyond the simple passage of time (6, 7). However, we know from a recently completed trial of acute treatment discontinuation (20) that patients relapse more commonly if medication is discontinued early. In the TADS, subjects treated acutely with placebo were treated openly after 12 weeks, depending on clinical status, and by the end of 36 weeks had reached a clinical status equivalent to that of patients assigned to active treatment (40). We also know (data not shown) that placebo-treated subjects continued to do well in stage IV. Taken together, these findings imply that for the average patient, 6 to 9 months of active treatment is sufficient to maintain benefits once treatment is discontinued. To definitively establish the necessary duration of treatment, a randomized maintenance therapy clinical trial will be necessary.

Likewise, without an actively treated comparison group (e.g., a randomized discontinuation design in which subjects were randomly assigned to continue or to discontinue treatment), it is impossible to state with confidence that persistent benefit is identical to maximum benefit. Given that relatively few TADS subjects received mental health treatment during stage IV, continued maintenance treatment might have improved outcomes for the subjects who deteriorated, especially the 3% who needed hospitalization.

Finally, collection of data on services utilization in stage IV suffered from a much higher percentage of missing data than in stages I–III. For example, while we obtained data of reasonable quality on provider type, data on the type of treatment were mostly missing at some sites, making it impossible to predict treatment course by treatment type or exposure without the application of extensive and potentially questionable data imputation methods.

Conclusion and Clinical Implications

With publication of the findings from acute treatment (stage I), long-term treatment (stages I–III), and now the naturalistic 1-year follow-up (stage IV), the central findings from the TADS are as follows: 1) combined treatment meaningfully accelerates recovery from depression relative to CBT and fluoxetine, 2) longer-term treatment results in improved outcomes relative to short-term treatment, 3) longer-term treatment may decrease the chances of relapse or recurrence when treatment is discontinued, and 4) adding CBT to fluoxetine minimizes persistent suicidal ideation and treatment-emergent suicidal events associated with medication monotherapy. While not definitive in all respects, the TADS adds important new information to the existing literature and, along with the Adolescent Depression and Psychotherapy Trial (ADAPT) (41) and Treatment of Resistant Depression in Adolescents (TORDIA) study (42), begins to provide an evidence-based platform for establishing treatment guidelines for adolescent depression.

From the outset, TADS treatments were designed according to “best practice” standards and were manualized to allow ready dissemination (if warranted) into clinical practice at the conclusion of the trial (3). Given the fact that combined treatment is more cost-effective than fluoxetine alone, in part because of the higher suicidality-driven health care costs associated with medication monotherapy (43), there are substantial public health and health economic benefits that would accrue from providing evidence-based combined treatment to moderately to severely ill depressed teens. As the TADS comes to an end, we believe that it is now time for the TADS to inform programs of research focused on disseminating evidence-based care and on quality improvement initiatives in practice settings.