Comparative Benefits and Harms of Second-Generation Antidepressants: Background Paper for the American College of Physicians FREE

Major depressive disorder (MDD) is the most prevalent axis I disorder, affecting more than 16% of U.S. adults during their lifetime (1). In 2000, the economic burden of depressive disorders was an estimated $83.1 billion (2), more than 30% of which was attributable to direct medical expenses.

Pharmacotherapy dominates the medical management of MDD. Since the mid-1980s, second-generation antidepressants have gradually replaced tricyclic antidepressants and monoamine oxidase inhibitors as first-line medications, primarily because of their lower toxicity in overdose and similar general efficacy (3). These newer treatments include selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors, selective serotonin and norepinephrine reuptake inhibitors, and other second-generation drugs (Table 1).

To date, only 2 systematic reviews have assessed the comparative efficacy and harms of second-generation antidepressants (3 - 4). These studies reported no substantial differences in efficacy or harms among agents. However, because of a lack of direct head-to-head comparisons, assessments in both studies were primarily qualitative. Consequently, uncertainties persist about the differences among the drugs for which sufficient head-to-head evidence is lacking.

We systematically assessed evidence on the comparative benefits and harms of second-generation antidepressants for the acute, continuation, and maintenance phases of treatment of MDD; subsyndromal depression; and dysthymia and the comparative efficacy and effectiveness for such accompanying symptoms as anxiety, insomnia, or neurovegetative symptoms. We also sought to determine whether efficacy, effectiveness, and harms differed among subgroups of patients on the basis of age, sex, race or ethnicity, or comorbid conditions.

To our knowledge, this is the first meta-analysis of second-generation antidepressants to assess quantitatively all possible comparisons among drugs in this class. We update findings of an earlier report on these pharmaceuticals (5) for the Agency for Healthcare Research and Quality.

An open process (described at http://www.effectivehealthcare.ahrq.gov) involving the public, the Agency for Healthcare Research and Quality's Scientific Resource Center for Effective Health Care program, and various stakeholder groups produced key questions. We followed a standardized protocol for all review steps (5).

We searched MEDLINE, EMBASE, PsychLit, Cochrane Central Register of Controlled Trials, and International Pharmaceutical Abstracts from 1980 to April 2007. We used Medical Subject Heading terms when available and keywords when appropriate. We combined terms for depressive disorders with a list of 12 specific second-generation antidepressants—bupropion, citalopram, duloxetine, escitalopram, fluoxetine, fluvoxamine, mirtazapine, nefazodone, paroxetine, sertraline, trazodone, and venlafaxine—and their specific trade names. We limited electronic searches to “adult 19 + years,” “human,” and “English language.”

We manually searched reference lists of pertinent review articles and letters to the editor and used the Center for Drug Evaluation and Research database (up to April 2007) to identify unpublished research submitted to the U.S. Food and Drug Administration. The Scientific Resource Center invited pharmaceutical manufacturers to submit dossiers on completed research for each drug. We received dossiers from 3 pharmaceutical companies (Eli Lilly and Company, Indianapolis, Indiana; GlaxoSmithKline, Philadelphia, Pennsylvania; and Wyeth, Madison, New Jersey).

Two persons independently reviewed abstracts and relevant full-text articles. To assess efficacy or effectiveness regarding response, speed of onset, remission, maintenance of remission, and quality of life, we included head-to-head controlled trials of at least 6 weeks' duration that compared 1 drug with another. Because head-to-head evidence was lacking for many comparisons, we included placebo-controlled trials for indirect comparison models. To assess harms (specific adverse events, rates of adverse events, and discontinuations attributable to adverse events), we also examined data from observational studies with at least 100 participants and follow-up of at least 12 weeks. To assess differences of benefits and harms in subgroups and patients with accompanying symptoms, we reviewed both head-to-head and placebo-controlled trials. We included meta-analyses if we found them to be relevant for a key question and of good or fair methodological quality (6).

If both reviewers agreed that a study did not meet eligibility criteria, we excluded it. We also excluded studies that met eligibility criteria but were reported only as an abstract. Investigators resolved disagreements about inclusion or exclusion by consensus or by involving a third reviewer.

We used a structured, Web-based data abstraction form (SRS 4.0, TrialStat, Ottawa, Ontario, Canada) onto which trained reviewers abstracted data from each study and assigned an initial quality rating. A senior reviewer read each abstracted article, evaluated completeness of data abstraction, and confirmed the quality rating. Investigators resolved disagreements by discussion and consensus or by consulting an independent party.

We assessed the internal validity (quality) of trials on the basis of predefined criteria and applied ratings of good, fair, or poor (5,7 - 8). Primary elements of quality assessment included randomization and allocation concealment, similarity of compared groups at baseline, blinding, use of intention-to-treat analysis, and overall and differential loss to follow-up. To assess observational studies, we used criteria involving selection of case patients or cohorts and control participants, adjustment for confounders, methods of outcomes assessment, length of follow-up, and statistical analysis (9). We rated studies with a fatal flaw in 1 or more categories as poor quality (Appendix Table 1) and did not include them in our analyses for this review unless no other head-to-head evidence was available. To identify effectiveness studies, we used a tool that distinguishes efficacy trials from effectiveness studies on the basis of certain elements of study design (10). Such studies have greater generalizability of results than efficacy trials because they enroll less selected study populations, use treatment modalities that mimic clinical practice, and assess health outcomes along with adverse events.

Lacking clear definitions about the equivalence of dosages among second-generation antidepressants in the published literature, we developed a roster of low, medium, and high dosages for each drug based on the interquartile dosing range (5). We used this roster, which does not indicate dosing equivalence, to detect gross inequalities in dosing that could affect comparative efficacy and effectiveness.

If data were sufficient, we conducted meta-analyses of head-to-head comparisons. Efficacy outcomes included the relative benefit of achieving response (more than 50% improvement from baseline), which reflects the ratio of benefits in one treatment group to benefits in another, and the weighted mean difference of changes on the Hamilton Depression Rating Scale or the Montgomery-Asberg Depression Rating Scale.

For each meta-analysis, we conducted a test of heterogeneity (I² index) and applied both random- and fixed-effects models. We report the random-effects results because the results from both models were very similar in all meta-analyses. We assessed publication bias by using funnel plots and the Begg adjusted rank correlation test (11) based on the Kendall τ coefficient.

Because no head-to-head evidence was available for the majority of drug comparisons, we conducted adjusted indirect comparisons (5). We employed meta-regressions of placebo-controlled trials by using individual drugs as covariates. When the number of trials was insufficient for meta-regressions, we used modified network meta-analysis (12). Evidence suggests that indirect comparisons agree with head-to-head trials if component studies are similar and treatment effects are expected to be consistent in patients included in different trials (13), although these assumptions are usually not verifiable.

All statistical analyses used StatsDirect Statistical Software program, version 2.3.8 (StatsDirect, Sale, United Kingdom); Stata, version 9.1 (StataCorp, College Station, Texas); and SAS, version 9.1 (SAS Institute, Cary, North Carolina).

We rated the strength of the available evidence for specific key questions and outcomes in a 3-part hierarchy (high, moderate, and low) (5) by using a modified GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) approach (14 - 15) that incorporates 4 key elements: study design, study quality, consistency of results, and directness (availability of data on outcomes or populations of interest).

The Agency for Healthcare Research and Quality participated in formulating the key questions and reviewed and commented on planned methods and data analysis. The Agency had no role in study selection, quality ratings, or interpretation and synthesis of the evidence, although staff reviewed interim and final evidence reports and distributed them for external peer review by outside experts.

We identified 2318 citations from searches and reviews of reference lists (Figure 1). Of the 203 included studies (Appendix Tables 2 to 11), 140 (69.0%) were financially supported by pharmaceutical companies and 19 (9.3%) by governmental agencies or independent funds. For 44 (21.7%) studies, we could not determine the funding source.

Overall, we found no substantial differences in comparative efficacy and effectiveness of second-generation antidepressants for treatment of MDD (Tables 2, 3, and 4 and Figures 2, 3, and 4). This finding pertains to the acute, continuation, and maintenance phases of treatment; to patients with accompanying symptom clusters; and to subgroups defined by age, race or ethnicity, sex, or comorbid conditions (we found only sparse evidence for subgroups). Nevertheless, second-generation antidepressants are not identical drugs. They differ somewhat with respect to onset of action and frequency of some adverse events. Generally, effectiveness studies with less stringent eligibility criteria provided results similar to those of efficacy trials, indicating good generalizability of our findings to primary care populations.