Nonpharmacological Versus Pharmacological Treatments for Adult Patients With Major Depressive Disorder

We searched MEDLINE® (via PubMed®), Embase®, the Cochrane Library, AMED (Allied and Complementary Medicine Database), PsycINFO®, and CINAHL (Cumulative Index to Nursing and Allied Health Literature) from January 1, 1990, through January 13, 2015. We used a combination of medical subject headings and title and abstract keywords, focusing on terms to describe the relevant population and interventions of interest. We limited the electronic searches to English-, German-, and Italian-language and human-only studies.

In addition, we manually searched reference lists of pertinent reviews, included trials, and background articles, and searched for gray literature relevant to this review following guidance from the Methods Guide for Effectiveness and Comparative Effectiveness Reviews for these steps.⁹

Inclusion and exclusion criteria are presented in Table 1.

Two trained research team members independently reviewed all titles, abstracts, and eligible full-text articles. We designed, pilot tested, and used a structured data abstraction form to ensure consistency of data abstraction. Trained reviewers initially abstracted data from each study. A senior reviewer then read each abstracted article and evaluated the completeness and accuracy of the data abstraction. We resolved discrepancies by consensus or by involving a third, senior reviewer.

To assess the risk of bias of studies, we used definitions based on AHRQ guidance.¹⁰ We rated the risk of bias for each relevant outcome of a study as low, moderate, or high. To determine risk of bias in a standardized way, we used the Cochrane Risk of Bias tool to appraise randomized controlled trials (RCTs).¹¹ Two independent reviewers assigned risk-of-bias ratings. They resolved any disagreements by discussion and consensus or by consulting a third, independent party.

Throughout this review we synthesized the literature qualitatively. When data were sufficient, we augmented findings with quantitative analyses.

For meta-analyses, we used random-effects (DerSimonian-Laird) and fixed-effects models to estimate comparative effects. We assessed statistical heterogeneity in effects between studies by calculating the chi-squared statistic and Cochran’s q. We used the I² statistic to estimate the magnitude of heterogeneity. We examined potential sources of heterogeneity using sensitivity analysis or analysis of subgroups. We assessed publication bias by checking study registries and using funnel plots and Kendall’s tests. However, given the small number of component studies in our meta-analyses, these tests have low sensitivity to detect publication bias.

Because of the dearth of studies directly comparing interventions of interest, we planned network meta-analyses a priori. Our outcome measure of choice was the rate of response on the Hamilton Depression Rating Scale (HAM-D), defined as at least a 50-percent improvement of scores from baseline. We included all placebo- and active-controlled RCTs detected through our searches that were homogeneous in study populations and outcome assessments and were part of a connected network. We employed a hierarchical frequentist approach using random-effects models.^12,13

We graded the strength of evidence (SOE) based on AHRQ guidance established for the Evidence-based Practice Centers.¹⁴ This approach incorporates five key domains: risk of bias, consistency, directness, precision, and reporting bias. Grades (high, moderate, low, insufficient) reflect the strength of the body of evidence for a specific outcome on the comparative benefits and harms of the interventions in this review. During the protocol development, we asked the Technical Expert Panel and the Key Informants to rank the relative importance of outcomes following a process proposed by the GRADE (Grading of Recommendations Assessment, Development and Evaluation) Working Group.¹⁵ We graded only those outcomes that Technical Expert Panel members and Key Informants deemed as important or critical for decisionmaking.

We assessed applicability of the evidence following guidance from the Methods Guide for Effectiveness and Comparative Effectiveness Reviews.¹⁶ We used the PICOTS (populations, interventions, comparators, outcomes, timing, settings) framework to explore factors that may affect applicability.

Our search strategies identified 7,813 possible articles. We excluded 7,368 references following independent dual title and abstract review, and another 390 references at the full-text review stage. Reasons for exclusion were based on eligibility criteria. Overall, we included 44 trials reported in 55 published articles. Of these, 42 trials pertained to KQ 1a and 5 to KQ 1b. Two trials pertained to KQ 2a, and no trials were identified for KQ 2b. In addition, of the 44 trials, 43 trials pertained to KQ 3a and 1 to KQ 3b; 3 pertained to KQ 4.

For network meta-analyses, we included data from 85 additional published trials and 27 unpublished trials. These trials addressed comparisons of interventions of interest that did not meet eligibility criteria for this report; they did, however, provide common comparators that we could use for network meta-analyses.

In all, 42 trials comparing SGAs with nonpharmacological treatment options for MDD provided direct evidence on acute-phase outcomes. Study durations ranged from 4 to 96 weeks. Most patients suffered from moderate to severe major depression. Many of the available trials had serious methodological limitations. Additionally, few trials adequately assessed harms or reported information on quality of life or functional capacity. The figures provide graphical overviews of response rates (Figure A) and discontinuation rates because of adverse events (Figure B) of SGAs compared with psychological interventions, CAM therapies, and exercise.

No trials were specifically designed to assess differences in our specified subgroups. Overall, only three trials addressing a subgroup of interest met the criteria for inclusion: one of subgroups defined by common accompanying psychiatric symptoms and two of subgroups defined by demographic characteristics. For common accompanying psychiatric symptoms, SGAs produced slightly higher remission rates than interpersonal psychotherapy in patients with a comorbid anxiety disorder but not in those without co-occurring anxiety (insufficient SOE). We had no evidence for any other common accompanying symptoms (insomnia, low energy, or somatization).

For subgroups defined by demographic characteristics, we included two trials. In one trial conducted in older adults, SGAs and St. John’s wort led to similar response rates and discontinuation rates because of adverse events (low SOE). The other trial included only minority (predominantly black and Latina) women and showed similar reduction in depressive symptoms between SGAs and CBT (insufficient SOE). We did not identify any trials assessing differences between men and women in effectiveness or harms (insufficient SOE).

No trials at all addressed effectiveness or harms in selected subgroups of patients who did not achieve remission following an initial adequate trial with one SGA (insufficient SOE).

Across all interventions, we graded the strength of evidence for benefits as moderate for only one comparison—namely, SGAs compared with CBT. Results from trials of this comparison indicate that SGAs and CBT have similar effectiveness regarding symptomatic relief in patients with mild to severe MDD. For risk of harms, we graded the strength of evidence as moderate for some outcomes of three comparisons—namely, SGAs compared with CBT, acupuncture, and St. John’s wort. Patients treated with SGAs had a higher risk of experiencing adverse events or discontinuing treatment because of adverse events than patients treated with CBT, acupuncture, or St. John’s wort. The evidence is insufficient to draw conclusions about differences in serious adverse events, such as suicidal ideas and behavior.

Our confidence in findings from the comparisons of remaining treatment options was low or insufficient, indicating that these bodies of evidence had major or unacceptable deficiencies. Nevertheless, for most comparisons the overall findings did not show statistically significant differences in benefits but indicated a lower risk of adverse events for nonpharmacological treatment options. Notable exceptions are omega-3-fatty acids, which appear to have lower effectiveness than SGAs; the combination of SGAs with acupuncture, which appears to have higher response rates than SGA monotherapy; and the combination of SGAs with interpersonal psychotherapy, which appears to have better effectiveness than SGA monotherapy. Our confidence in these findings, however, is low, and results have to be interpreted cautiously. In addition, for many comparisons that are limited to single trials, determining whether similar treatment effects between SGAs and other interventions are based on similar effectiveness or high placebo response rates is impossible. Furthermore, we emphasize that detecting no statistically significant difference does not necessarily mean the treatments are equivalent.

The available data offer no conclusions on how selection of treatment strategies might differ based on a patient’s severity of depression. Overall, data do not indicate differences in comparative effectiveness between SGAs and nonpharmacological interventions for patients with severe MDD. This important question concerning MDD severity, although raised by a few systematic reviews,^17-19 remains without a clear answer.

Beyond the two articles identified comparing switching and augmentation strategies employing a limited number of medication options or CT, the absence of relevant comparative data about which treatment options are most effective for those needing second-step treatment (about 70% of patients with MDD)^20,21 was striking.

Our findings are consistent with several prior systematic reviews and meta-analyses that compared SGAs with nonpharmacological interventions. Most of these reviews, however, included populations that were not eligible for our review, such as patients with minor depression, bipolar disorder, or dysthymia.

Our results are partially consistent with the recommendations of both the American Psychiatric Association²² and the Department of Veterans Affairs/Department of Defense.²³ These consider both pharmacotherapy and psychotherapy to be appropriate individual first-step treatments for patients with mild to moderate MDD, and state that the combination of pharmacotherapy and psychotherapy may be necessary in cases of moderate to severe depression.

In terms of clinical decisionmaking, the information in this review can be helpful to physicians because they can provide a summary of the available evidence base indicating the advantages and disadvantages of these options, and patients can identify which intervention they would prefer. Some options, such as medication and St. John’s wort, would require physician supervision and monitoring, given potential side effects and drug interactions. Moreover, patients who would like to maintain or start an exercise regimen in addition to undergoing SGA therapy can be encouraged to do so. The enhanced potential for increasing physical well-being and expanding social interactions may be an added incentive to encourage an exercise regimen.

The scope of this review was limited to trials that enrolled adult patients with MDD. We did not attempt to review literature on interventions for children with MDD or for patients with subthreshold depression (depressive symptoms not severe enough to meet diagnostic criteria for a major depressive episode), dysthymia, psychotic depression, or perinatal depression. The included trials covered populations with mild, moderate, and severe MDD; the majority of participants were women. Most trial populations, however, excluded patients with medical comorbidities or suicidal ideas and behaviors; few trials included elderly patients. We did not find evidence to confirm or refute whether treatments are more or less efficacious for various subgroups (i.e., patients characterized by sex, race, or ethnicity, or individuals with coexisting psychiatric conditions).

With few exceptions, interventions in included trials were in line with clinical practice. Except for some CAM trials in which patients received SGA dosages at the lower end of the recommended range, prescribing patterns and doses in the SGA arms of our evidence base were consistent with clinical practice. Some newer SGAs, such as desvenlafaxine, levomilnacipran, vilazodone, or vortioxetine, have never been compared with psychological or CAM treatments or exercise. Nevertheless, reliable evidence indicates that the comparative effectiveness of SGAs is similar.²⁴ Consequently, we believe that our findings are applicable across the class of SGAs.

As noted previously, detecting no statistically significant difference does not necessarily mean that the treatments are equivalent. The studies involved were designed to test whether an outcome for one intervention was different from the outcome for another rather than to test equivalence, which would generally require a much larger sample size. This point is especially relevant for findings with a low SOE. While confidence intervals were relatively narrow and risk ratios were often close to 1 (findings consistent with equivalent outcomes), a conclusion of equivalence cannot be made. Further, while moderate-strength evidence at a group level did not detect a difference between SGAs and CBT, how best to tailor this information to an individual patient is still not clear. Indeed, other potentially relevant indicators (e.g., depressive severity, comorbid psychiatric illness) may favor one over another, but the current evidence base (as indicated in the KQ 1b and 2b findings) is quite limited.

Finally, many trials, particularly for CAM interventions, were conducted outside the United States. Whether and how differences in ethnic or cultural backgrounds and health systems affect the applicability of results to U.S. populations remain uninvestigated and unanswered.

Across all comparisons of interventions, major research gaps pertain to information about patient-centered outcomes, such as functional capacity and quality of life, and the comparative risk of harms. Lack of information about harms can lead to a biased knowledge base and the potential for decisions that cause more harm than good.

We found no eligible studies that compared SGAs with behavior therapy or behavior modification, humanistic therapies, yoga, or mindfulness interventions. Given the wide use of these types of psychotherapies in clinical practice, further research into their comparative effectiveness with SGAs in treating MDD patients is desirable. For many psychotherapies and all CAM therapies that have been evaluated against an SGA, the data were insufficient because trials did not report important outcomes, most notably quality of life and functional capacity. Future studies should assess remission, response to treatment, quality of life, functional capacity, suicidal ideas and behaviors, and adverse events using standardized measures to allow for more direct comparisons across studies using the same or similar SGAs and psychological interventions. These same deficiencies in the literature extend to the comparative effectiveness of SGAs and both psychological and CAM interventions for treating MDD as a function of depression severity. For CAM interventions, we found that most studies did not include the full range of SGA doses for comparison, and many studies made comparisons with only the very lowest SGA doses. To truly compare any CAM intervention for MDD treatment, future studies will need to incorporate SGA dosing strategies that use the entire SGA dosage range. Finally, a major gap in the evidence is the lack of studies addressing different treatment options for patients who have not achieved remission with first-step therapy. No second-step therapy data at all exist that compare SGA with CAM or exercise treatments. This void in the evidence base is a major one that will perplex and confound clinicians, patients, policymakers, and guideline developers alike.

Overall, the available evidence indicates that SGAs and CBT do not differ significantly in symptomatic relief as first-step treatments for adult outpatients with mild to severe MDD. The evidence is insufficient to draw conclusions about the comparative risk of serious adverse events, such as suicidal ideas and behaviors. Given comparable benefits among treatment options, the choice of the initial treatment of MDD should consider results of previous treatments, patient preferences, and feasibility (e.g., costs, likely adherence, and availability) following a discussion of the advantages and disadvantages of each treatment option, including risks of particular adverse effects and potential drug interactions.

Differences with respect to adverse events, personal engagement, and costs may be taken into consideration for the choice of a first-step treatment. Such shared and informed decisionmaking might enhance treatment adherence and improve treatment outcomes for patients with MDD, especially because treatment continuity is one of the main challenges in treating such patients. For second-step therapies, although evidence is limited, no clear benefit emerges to suggest that either switching to a particular SGA or CT, or augmenting with a particular medication or CT, is preferable. Available data suggest that switching to another SGA, switching to CT, or augmenting with a particular medication or CT are all reasonable options. The more important decision appears to be simply to try a different evidence-based approach.

In the manuscript summarizing the findings of this report for journal submission, we employed a different statistical approach for random effects meta-analyses than in the AHRQ report. We followed journal policy and used restricted maximum likelihood models instead of DerSimonian and Laird methods. As a consequence, point estimates and the width of some confidence intervals for some effect estimates are slightly different between the AHRQ report and journal manuscript. Differences are minor and do not change conclusions.