Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis - The Lancet

Evidence before this study

Antidepressants are routinely used worldwide for the treatment of major depressive disorder, which is one of the most important global health challenges; however, in the scientific literature, there remains considerable debate about both their effectiveness as a group, and the potential differences in effectiveness and tolerability between individual drugs. With the marketing of new antidepressants and increasing numbers of trials published every year, an updated systematic review and network meta-analysis was required to synthesise the evidence in this important clinical area.

Added value of this study

This network meta-analysis represents a major update and extension of our previous study, which addressed 12 antidepressants with data for head-to-head comparisons only, and provides the best currently available evidence base to guide the choice about pharmacological treatment for adults with acute major depressive disorder. We now include a more comprehensive list of 21 antidepressants and placebo, consider three new clinical outcome measures and many potential effect modifiers, and use the most advanced statistical methodology for network meta-analysis to date.

Implications of all the available evidence

Our findings should inform clinical guidelines and assist the shared decision making process between patients, carers, and clinicians in routine practice on selecting the most appropriate antidepressant for adults with acute major depressive disorder. Future research should seek to extend network meta-analysis to combine aggregate and individual-patient data from trials in a so-called individual-patient data network meta-analysis. This analysis will allow the prediction of personalised clinical outcomes, such as early response or specific side-effects, and the estimate of comparative efficacy at multiple timepoints.

Introduction

Psychiatric disorders account for 22·8% of the global burden of diseases.1x 1 GBD 2013 DALYs and HALE Collaborators, Murray, CJ, Barber, RM et al. Global, regional, and national disability-adjusted life years (DALYs) for 306 diseases and injuries and healthy life expectancy (HALE) for 188 countries, 1990–2013: quantifying the epidemiological transition. Lancet . 2015 ; 386 : 2145–2191

Summary | Full Text | Full Text PDF | PubMed | Scopus (460) | Google Scholar See all References The leading cause of this disability is depression, which has substantially increased since 1990, largely driven by population growth and ageing.2x 2 GBD 2015 DALYs and HALE Collaborators. Global, regional, and national disability-adjusted life-years (DALYs) for 315 diseases and injuries and healthy life expectancy (HALE), 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet . 2016 ; 388 : 1603–1658

Summary | Full Text | Full Text PDF | PubMed | Scopus (231) | Google Scholar See all References With an estimated 350 million people affected globally, the economic burden of depressive disorders in the USA alone has been estimated to be more than US$210 billion, with approximately 45% attributable to direct costs, 5% to suicide-related costs, and 50% to workplace costs.3x 3 WHO. Depression: fact sheet. World Health Organisation , Geneva ; 2017 http://www.who.int/mediacentre/factsheets/fs369/en/. ()

Google Scholar See all References This trend poses a substantial challenge for health systems in both developed and developing countries, with the need to treat patients, optimise resources, and improve overall health care in mental health.

Grouped into various classes of drugs with slightly different mechanisms of action, antidepressants are widely used treatments for major depressive disorder, which are available worldwide. However, there is a long-lasting debate and concern about their efficacy and effectiveness, because short-term benefits are, on average, modest; and because long-term balance of benefits and harms is often understudied.4x 4 Ioannidis, JP. Effectiveness of antidepressants: an evidence myth constructed from a thousand randomized trials?. Philos Ethics Humanit Med . 2008 ; 3 : 14

Crossref | PubMed | Scopus (97) | Google Scholar See all References Therefore, innovation in psychopharmacology is of crucial importance, but the identification of new molecular targets is difficult, primarily because of the paucity of knowledge about how antidepressants work.5x 5 Harmer, CJ, Duman, RS, and Cowen, PJ. How do antidepressants work? New perspectives for refining future treatment approaches. Lancet Psychiatry . 2017 ; 4 : 409–418

Summary | Full Text | Full Text PDF | PubMed | Scopus (7) | Google Scholar See all References In routine practice, clinicians have a wide choice of individual drugs and they need good evidence to make the best choice for each individual patient. Network meta-analyses of existing datasets make it possible to estimate comparative efficacy, summarise and interpret the wider picture of the evidence base, and to understand the relative merits of the multiple interventions.6x 6 Higgins, JP and Welton, NJ. Network meta-analysis: a norm for comparative effectiveness?. Lancet . 2015 ; 386 : 628–630

Summary | Full Text | Full Text PDF | PubMed | Scopus (11) | Google Scholar See all References Therefore, in this study, we aimed to do a systematic review and network meta-analysis to inform clinical practice by comparing different antidepressants for the acute treatment of adults with unipolar major depressive disorder.

Methods

Search strategy and selection criteria

We did a systematic review and network meta-analysis. We searched the Cochrane Central Register of Controlled Trials, CINAHL, Embase, LILACS database, MEDLINE, MEDLINE In-Process, PsycINFO, AMED, the UK National Research Register, and PSYNDEX from the date of their inception to Jan 8, 2016, with no language restrictions. We used the search terms “depress*” OR “dysthymi*” OR “adjustment disorder*” OR “mood disorder*” OR “affective disorder” OR “affective symptoms” combined with a list of all included antidepressants.

We included double-blind, randomised controlled trials (RCTs) comparing antidepressants with placebo or another active antidepressant as oral monotherapy for the acute treatment of adults (≥18 years old and of both sexes) with a primary diagnosis of major depressive disorder according to standard operationalised diagnostic criteria (Feighner criteria, Research Diagnostic Criteria, DSM-III, DSM-III-R, DSM-IV, DSM-5, and ICD-10). We considered only double-blind trials because we included placebo in the network meta-analysis, and because this study design increases methodological rigour by minimising performance and ascertainment biases.7x 7 Hróbjartsson, A, Thomsen, AS, Emanuelsson, F et al. Observer bias in randomized clinical trials with measurement scale outcomes: a systematic review of trials with both blinded and nonblinded assessors. CMAJ . 2013 ; 185 : E201–E211

Crossref | PubMed | Scopus (154) | Google Scholar See all References Additionally, we included all second-generation antidepressants approved by the regulatory agencies in the USA, Europe, or Japan: agomelatine, bupropion, citalopram, desvenlafaxine, duloxetine, escitalopram, fluoxetine, fluvoxamine, levomilnacipran, milnacipran, mirtazapine, paroxetine, reboxetine, sertraline, venlafaxine, vilazodone, and vortioxetine. To inform clinical practice globally, we selected the two tricyclics (amitriptyline and clomipramine) included in the WHO Model List of Essential Medicines). We also included trazodone and nefazodone, because of their distinct effect and tolerability profiles. Additionally, we included trials that allowed rescue medications so long as they were equally provided among the randomised groups. We included data only for drugs within the therapeutic range (appendix pp 133, 134appendix pp 133, 134 ). Finally, we excluded quasi-randomised trials and trials that were incomplete or included 20% or more of participants with bipolar disorder, psychotic depression, or treatment-resistant depression; or patients with a serious concomitant medical illness.

The electronic database searches were supplemented with manual searches for published, unpublished, and ongoing RCTs in international trial registers, websites of drug approval agencies, and key scientific journals in the field.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References For example, we searched ClinicalTrials.gov using the search term “major depressive disorder” combined with a list of all included antidepressants. We contacted all the pharmaceutical companies marketing antidepressants and asked for supplemental unpublished information about both premarketing and post-marketing studies, with a specific focus on second-generation antidepressants. We also contacted study authors and drug manufacturers to supplement incomplete reports of the original papers or provide data for unpublished studies.

Six pairs of investigators (ACi, TAF, LZA, SL, HGR, YO, NT, YH, EHT, HI, KS, and AT) independently selected the studies, reviewed the main reports and supplementary materials, extracted the relevant information from the included trials, and assessed the risk of bias. Any discrepancies were resolved by consensus and arbitration by a panel of investigators within the review team (ACi, TAF, LZA, EHT, and JRG).

The full protocol of this network meta-analysis has been published.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References

Outcomes

Our primary outcomes were efficacy (response rate measured by the total number of patients who had a reduction of ≥50% of the total score on a standardised observer-rating scale for depression) and acceptability (treatment discontinuation measured by the proportion of patients who withdrew for any reason).8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References All-cause discontinuation was used as a measure for the acceptability of treatments, because it encompasses efficacy and tolerability.9x 9 Cipriani, A, Furukawa, TA, Salanti, G et al. Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatment meta-analysis. Lancet . 2009 ; 373 : 746–758

Summary | Full Text | Full Text PDF | PubMed | Scopus (949) | Google Scholar See all References Secondary outcomes were endpoint depression score, remission rate, and the proportion of patients who dropped out early because of adverse events. When depressive symptoms had been measured with more than one standardised rating scale, we used a predefined hierarchy, based on psychometric properties and consistency of use across included trials.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References In the absence of information or supplemental data from the authors, response rate was calculated according to a validated imputation method.10x 10 Furukawa, TA, Cipriani, A, Barbui, C, Brambilla, P, and Watanabe, N. Imputing response rates from means and standard deviations in meta-analyses. Int Clin Psychopharmacol . 2005 ; 20 : 49–52

Crossref | PubMed | Scopus (105) | Google Scholar See all References We recorded the outcomes as close to 8 weeks as possible for all analyses.9x 9 Cipriani, A, Furukawa, TA, Salanti, G et al. Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatment meta-analysis. Lancet . 2009 ; 373 : 746–758

Summary | Full Text | Full Text PDF | PubMed | Scopus (949) | Google Scholar See all References If information at 8 weeks was not available, we used data ranging between 4 and 12 weeks (we gave preference to the timepoint closest to 8 weeks; if equidistant, we took the longer outcome). We checked trial protocols where available and compared published with unpublished data. We extracted data following a predefined hierarchy described in our protocol and gave priority to unpublished information in case of disagreement.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References

Data analysis

For studies published more than once (ie, duplicates), we included only the report with the most informative and complete data. Full details of the applied statistical approaches are provided in the protocol.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References We estimated summary odds ratios (ORs) for dichotomous outcomes and standardised mean differences (SMD, Cohen's d) for continuous outcomes using pairwise and network meta-analysis. In network meta-analysis, we used group-level data; the binomial likelihood was used for dichotomous outcomes and the normal likelihood for continuous outcomes. The study effect sizes were then synthesised using a random-effects network meta-analysis model. We accounted for the correlations induced by multi-group studies by using multivariate distributions. The variance in the random-effects distribution (heterogeneity variance) was considered to measure the extent of across-study and within-comparison variability on treatment effects. Additionally, in network meta-analysis, we assumed that the amount of heterogeneity was the same for all treatment comparisons. To assess the amount of heterogeneity, we compared the posterior distribution of the estimated heterogeneity variance with its predictive distribution.11x 11 Rhodes, KM, Turner, RM, and Higgins, JP. Predictive distributions were developed for the extent of heterogeneity in meta-analyses of continuous outcome data. J Clin Epidemiol . 2015 ; 68 : 52–60

Summary | Full Text | Full Text PDF | PubMed | Scopus (35) | Google Scholar See all References To rank the treatments for each outcome, we used the surface under the cumulative ranking curve (SUCRA) and the mean ranks.12x 12 Salanti, G, Ades, AE, and Ioannidis, JP. Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol . 2011 ; 64 : 163–171

Summary | Full Text | Full Text PDF | PubMed | Scopus (502) | Google Scholar See all References The transitivity assumption underlying network meta-analysis was evaluated by comparing the distribution of clinical and methodological variables that could act as effect modifiers across treatment comparisons.8x 8 Furukawa, TA, Salanti, G, Atkinson, LZ et al. Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis. BMJ Open . 2016 ; 6 : e010919

Crossref | PubMed | Scopus (13) | Google Scholar See all References We did a statistical evaluation of consistency (ie, the agreement between direct and indirect evidence) using the design-by-treatment test13x 13 Higgins, JP, Jackson, D, Barrett, JK, Lu, G, Ades, AE, and White, IR. Consistency and inconsistency in network meta-analysis: concepts and models for multi-arm studies. Res Synth Methods . 2012 ; 3 : 98–110

Crossref | PubMed | Google Scholar See all References and by separating direct evidence from indirect evidence.14x 14 Dias, S, Welton, NJ, Caldwell, DM, and Ades, AE. Checking consistency in mixed treatment comparison meta-analysis. Stat Med . 2010 ; 29 : 932–944

Crossref | PubMed | Scopus (378) | Google Scholar See all References

We assessed the studies' risk of bias in accordance to the Cochrane Handbook for Systematic Reviews of Interventions. Additionally, we assessed the certainty of evidence contributing to network estimates of the main outcomes with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework.15x 15 Salanti, G, Del Giovane, C, Chaimani, A, Caldwell, DM, and Higgins, JP. Evaluating the quality of evidence from a network meta-analysis. PLoS One . 2014 ; 9 : e99682

Crossref | PubMed | Scopus (106) | Google Scholar See all References

We evaluated whether treatment effects for the two primary outcomes were robust in subgroup analyses and network meta-regression using study year, sponsorship, depressive severity at baseline, dosing schedule, study precision (ie, small study effect), and novelty effect.16x 16 Salanti, G, Dias, S, Welton, NJ et al. Evaluating novel agent effects in multiple-treatments meta-regression. Stat Med . 2010 ; 29 : 2369–2383

PubMed | Google Scholar See all References The appendix (pp 133–36)appendix (pp 133–36) summarises the definition of covariates. The sensitivity of our conclusions was evaluated by analysing the dataset with the following restrictions: studies with reported response rate, studies using accepted doses in all groups, studies with unpublished data, multi-centre studies, and head-to-head studies. We used comparison-adjusted funnel plots to investigate whether results in imprecise trials differ from those in more precise trials.17x 17 Chaimani, A, Higgins, JP, Mavridis, D, Spyridonos, P, and Salanti, G. Graphical tools for network meta-analysis in STATA. PLoS One . 2013 ; 8 : e76654

Crossref | PubMed | Scopus (307) | Google Scholar See all References

We fitted all models in OpenBUGS (version 3.2.2)18x 18 Lunn, D, Spiegelhalter, D, Thomas, A et al. The BUGS project: evolution, critique and future directions. Stat Med . 2009 ; 28 : 3049–3067

Crossref | PubMed | Scopus (904) | Google Scholar See all References using the binomial likelihood for dichotomous outcomes, uninformative prior distributions for the treatment effects, and a minimally informative prior distribution for the common heterogeneity SD. We assumed uninformative priors—ie, N(0,1000)—for all meta-regression coefficients. Convergence of models was ensured by visual inspection of three chains and after considering the Brooks–Gelman–Rubin diagnostic. The codes of analyses, statistical details of the meta-analysis, and meta-regression models are presented in the appendix (pp 182, 183)appendix (pp 182, 183) . Statistical evaluation of inconsistency and production of network graphs and result figures were done using the network and network graphs packages in Stata (version 14.2).19x 19 Chaimani, A and Salanti, G. Visualizing assumptions and results in network meta-analysis: the network graphs package. Stata J . 2015 ; 15 : 905–950

Google Scholar See all References Network meta-analyses of the primary outcomes were duplicated using the netmeta 0.9-6 package in R (version 3.4.0).20x 20 Schwarzer, G. Network meta-analysis. in: G Schwarzer, JR Carpenter, G Rücker (Eds.) Meta-analysis with R . Springer , Berlin ; 2015 : 187–216

Crossref | Google Scholar See all References The appendix (p 289)appendix (p 289) lists the changes to the original protocol. The study was done from March 12, 2012, to June 4, 2016, and data analysis was done from June 5, 2016, to Sept 18, 2017.

This study is registered with PROSPERO, number CRD42012002291.

Data sharing

With the publication of this Article, the full dataset will be freely available online in Mendeley Data, a secure online repository for research data, which allows archiving of any file type and assigns a permanent and unique digital object identifier (DOI) so that the files can be easily referenced (DOI:10.17632/83rthbp8ys.2).

Role of the funding source

The funder of this study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or in the decision to submit for publication. ACi, TAF, GS, ACh, LZA, and YO had full access to all the data, and ACi was responsible for the decision to submit for publication.

Results

28 552 citations were identified by the search and 680 potentially eligible articles were retrieved in full text (figure 1figure 1 ). We included 421 trials from the database search, 86 unpublished studies from trial registries and pharmaceutical company websites, and 15 from personal communication or hand-searching other review articles. Overall, 522 double-blind, parallel, RCTs (comprising 116 477 patients) done between 1979 and 2016, and comparing 21 antidepressants or placebo were included in the analysis (appendix pp 6–64appendix pp 6–64 ). The appendix (pp 65–114)appendix (pp 65–114) summarises the characteristics of included studies. The mean study sample size was 224 participants (SD 186). In total, 87 052 participants were randomly assigned to an active drug and 29 425 were randomly assigned to placebo. The mean age was 44 years (SD 9) for both men and women; 38 404 (62·3%) of 61 681 of the sample population were women. The median duration of the acute treatment was 8 weeks (IQR 6–8). 243 (47%) of 522 studies randomly assigned participants to three or more groups, and 304 (58%) of 522 were placebo-controlled trials. 391 (83%) of 472 were multi-centre studies and 335 (77%) of 437 studies recruited outpatients only. 252 (48%) of 522 trials recruited patients from North America, 37 (7%) from Asia, and 140 (27%) from Europe (59 [11%] trials were cross-continental and the remaining 34 [7%] were either from other regions or did not specify). The great majority of patients had moderate-to-severe major depressive disorder, with a mean reported baseline severity score on the Hamilton Depression Rating Scale 17-item of 25·7 (SD 3·97) among 464 (89%) of 522 studies. Response rate was imputed in 20 608 (17·7%) of 116 447 cases. Rescue medications (typically benzodiazepines or other sedative hypnotics) were allowed in 187 (36%) of 522 studies. 409 (78%) of 522 studies were funded by pharmaceutical companies. We retrieved unpublished information for 274 (52%) of the included trials. Consistent with the study protocol, the primary analysis was based on the 474 studies (comprising 106 966 patients) that used drugs within the licensed dose range (ie, the dosage approved by the regulatory agencies in the USA and Europe; appendix pp 133, 134appendix pp 133, 134 ).

Figure 1

Study selection process

RCTs=randomised controlled trials. *Industry websites, contact with authors, and trial registries. The total number of unpublished records is the total number of results for each drug and on each unpublished database source. †522 RCTs corresponded to 814 treatment groups.

Figure 2Figure 2 shows the network of eligible comparisons for efficacy and acceptability. All antidepressant drugs, except milnacipran, had at least one placebo-controlled trial. Only levomilnacipran was not directly compared with at least another active drug in any of the networks. The appendix (pp 139–44)appendix (pp 139–44) provides detailed results of pairwise meta-analyses. Figure 3Figure 3 shows the network meta-analysis' results for the primary outcomes. In terms of efficacy (432 RCTs, comprising 102 443 patients), all antidepressants were more effective than placebo, with ORs ranging between 2·13 (95% credible interval [CrI] 1·89–2·41) for amitriptyline and 1·37 (1·16–1·63) for reboxetine. In terms of acceptability (422 RCTs, comprising 99 787 patients), agomelatine (OR 0·84, 95% CrI 0·72–0·97) and fluoxetine (0·88, 0·80–0·96) were associated with fewer dropouts than placebo; by contrast, clomipramine was worse than placebo (1·30, 1·01–1·68).

Figure 2

Network meta-analysis of eligible comparisons for efficacy (A) and acceptability (B)

Width of the lines is proportional to the number of trials comparing every pair of treatments. Size of every circle is proportional to the number of randomly assigned participants (ie, sample size).

Figure 3

Forest plots of network meta-analysis of all trials for efficacy (A) and acceptability (B)

Antidepressants were compared with placebo, which was the reference compound. OR=odds ratio. CrI=credible interval.

The relative efficacy of antidepressants compared with placebo is also shown for remission (appendix pp 152, 153appendix pp 152, 153 ). The random-effects summary SMD for all antidepressants was 0·30 (95% CrI 0·26–0·34; p