Antidepressants for the treatment of adults with major depressive disorder in the maintenance phase: a systematic review and network meta-analysis

A systematic review and random-effects model network meta-analysis were conducted to compare the efficacy, acceptability, tolerability, and safety of antidepressants to treat adults with major depressive disorder (MDD) in the maintenance phase. This study searched the PubMed, Cochrane Library, and Embase databases and included only double-blind, randomized, placebo-controlled trials with an enrichment design: patients were stabilized on the antidepressant of interest during the open-label study and then randomized to receive the same antidepressant or placebo. The outcomes were the 6-month relapse rate (primary outcome, efficacy), all-cause discontinuation (acceptability), discontinuation due to adverse events (tolerability), and the incidence of individual adverse events. The risk ratio with a 95% credible interval was calculated. The meta-analysis comprised 34 studies (n = 9384, mean age = 43.80 years, and %females = 68.10%) on 20 antidepressants (agomelatine, amitriptyline, bupropion, citalopram, desvenlafaxine, duloxetine, escitalopram, fluoxetine, fluvoxamine, levomilnacipran, milnacipran, mirtazapine, nefazodone, paroxetine, reboxetine, sertraline, tianeptine, venlafaxine, vilazodone, and vortioxetine) and a placebo. In terms of the 6-month relapse rate, amitriptyline, citalopram, desvenlafaxine, duloxetine, fluoxetine, fluvoxamine, mirtazapine, nefazodone, paroxetine, reboxetine, sertraline, tianeptine, venlafaxine, and vortioxetine outperformed placebo. Compared to placebo, desvenlafaxine, paroxetine, sertraline, venlafaxine, and vortioxetine had lower all-cause discontinuation; however, sertraline had a higher discontinuation rate due to adverse events. Compared to placebo, venlafaxine was associated with a lower incidence of dizziness, while desvenlafaxine, sertraline, and vortioxetine were associated with a higher incidence of nausea/vomiting. In conclusion, desvenlafaxine, paroxetine, venlafaxine, and vortioxetine had reasonable efficacy, acceptability, and tolerability in the treatment of adults with stable MDD.


INTRODUCTION
Major depressive disorder (MDD) is a common mental illness [1], with a 12-month prevalence of 4.4% worldwide [2]. Individuals with MDD in the acute phase undergo pharmacotherapy (e.g., antidepressant therapy) [3] or non-pharmacotherapy (e.g., psychotherapy [4] and electroconvulsive therapy) [5]. Relapse/ recurrence rate of these patients is >85% within a decade of an index depressive episode and an average of ≥50% within 6 months of apparent clinical remission if the initially effective treatment is not continued [6]. Therefore, maintenance therapy is necessary to avoid relapse/recurrence [1]. Kato and colleagues recently conducted an important pairwise meta-analysis that included only double-blind, randomized placebo-controlled trials (DBRPCTs) with an enrichment design in which individuals with MDD were stabilized on the antidepressant of interest during the open-label study and then randomized to receive the same antidepressant or a placebo (40 studies, n = 8890) [7]. According to this meta-analysis, the antidepressant maintenance group had a significantly lower relapse rate than the antidepressant discontinuation group (odds ratio = 0.38, 95% confidence interval = 0.33-0.43, p < 0.00001). As the relapse rate remained unchanged in both the maintenance and discontinuation groups from 6 months to 1 year, Kato et al. concluded that antidepressant maintenance treatment for at least 6 months after remission is recommended to prevent relapse, with special attention to relapses and treatment failure during this 6-month period. Thanks to this excellent study, we conceived the new clinical question of which antidepressants were better in terms of efficacy, acceptability, tolerability, and safety for adult individuals with MDD as a maintenance treatment. A network meta-analysis on individuals with MDD in the acute phase demonstrated although some antidepressants (e.g., agomelatine, escitalopram, mirtazapine, paroxetine, and sertraline) have a relatively higher response rate and lower dropout rate than the others, fluvoxamine, reboxetine, and trazodone have been reported to have generally inferior efficacy and acceptability profiles compared with the other antidepressants [8]. This suggests that not all antidepressants have similar efficacies and acceptability in individuals with MDD in the acute phase. A network meta-analysis is a technique to compare three or more interventions simultaneously in a single analysis by combining both direct and indirect evidence across a network of studies [9]. A network meta-analysis also produces estimates of the relative effects between any pair of interventions in the network and usually yields more precise estimates than a single direct or indirect estimate, thereby allowing estimation of the ranking and hierarchy of interventions [9]. Results from a network meta-analysis cannot be obtained by a pairwise meta-analysis. Moreover, the previous pairwise metaanalyses for individuals with MDD in the maintenance phase did not evaluate the risk of individual adverse events of antidepressants [7,10,11]. To answer our clinical question, we conducted a systematic review and network meta-analysis on the 13 outcomes related to the efficacy, acceptability, tolerability, and safety of 20 antidepressants for the treatment of adults in the maintenance phase of MDD.

MATERIALS AND METHODS
This study adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [12] ( Table S1) and was registered on the Open Science Framework (https://osf.io/xwezp). At least two authors double-checked the accuracy of the literature search, data transfer, and calculations.

Search strategy and inclusion criteria
A systematic literature review was conducted in accordance with the Population, Intervention, Comparison, Outcome strategy: the population comprised adults in the maintenance phase of MDD, the intervention was monotherapy with antidepressants, the comparator medication was a placebo, and the outcomes were described in the following section. The inclusion criteria were as follows: (1) DBRPCTs with a minimum duration of 12 weeks and (2) DBRPCTs with an enrichment design in which patients were stabilized on the antidepressant of interest during the open-label study and then randomized to receive the same antidepressant or a placebo. The following studies were excluded: (1) studies focusing on specific generations (e.g., children and/or adolescents or older individuals) because the efficacy and safety of antidepressants in children and older individuals differ from those in the general adult population [1]; (2) studies including individuals with a dual diagnosis of MDD and other disorders because these studies could lead to heterogeneity [1]; and (3) continuation studies in which individuals with acute symptoms were randomly assigned to treatment groups (i.e., the target population for a continuation study was individuals with MDD in the acute phase). In the present systematic review and meta-analysis, among adults with MDD who benefited symptomatically from antidepressant treatment (i.e., the target population for our systematic review and meta-analysis was individuals with MDD in the maintenance phase), the differences in relapse rates were compared between those who continued with the same antidepressant and those who discontinued the antidepressant. Information on the literature search is displayed in Fig. S1.

Data synthesis, outcome measures, and data extraction
The primary outcome was the 6-month relapse rate (efficacy), and the secondary outcome was all-cause discontinuation (acceptability). Other outcomes included discontinuation due to adverse events (tolerability) and the incidence of individual adverse events (safety). If at least five studies have data sufficient to perform a network meta-analysis for a specific safety outcome, a network meta-analysis was conducted for the safety outcome. In the International Classification of Diseases 11th Revision [13], recurrent depressive disorder is defined by a history of at least two depressive episodes with an interval of several months without substantial mood disturbance. In the present study, the term "relapse" is used for convenience rather than "recurrence" similar to the previous study [7], because few studies in this meta-analysis included cases in which worsening of symptoms during the study period was considered a recurrence. The definitions of relapse for each included study are presented in Table S2, and the data synthesis results are shown in Table S3. To avoid unit-of-analysis errors in studies involving two or more treatment arms of the same drug at different doses, data from the treatment arms were pooled for analysis [9]. The extracted data were analyzed based on intention-to-treat or modified intention-to-treat principles. If necessary data were missing from the studies, we searched for them in published systematic review articles; we also attempted to contact the original investigators in order to obtain previously unpublished data.

Meta-analysis methods
Both pairwise [14] and Bayesian network meta-analyses [15] were performed using the random-effects model [16]. Because all of the outcomes in our study were dichotomous, risk ratios (RRs) with 95% credible intervals (CrIs) were calculated as effect sizes. Network heterogeneity was assessed using τ² statistics. In pairwise metaanalyses, heterogeneity was assessed using I 2 statistics. A statistical evaluation of incoherence was not possible because there was no head-to-head study comparing different antidepressants. The treatments for each outcome were ranked using the surface under the curve cumulative ranking probabilities. The methodological quality of the included studies was evaluated using the Cochrane risk of bias tool for randomized trials (ROB2) (https:// www.riskofbias.info/welcome/rob-2-0-tool).
The assumption of transitivity was tested by extracting potential effect modifiers such as sample size, duration of study, and mean age and comparing their distribution across comparisons in the network. We determined whether the distribution differences were large enough to threaten the validity of the analysis by comparing the distribution of these possible effect modifiers across treatments included in the network meta-analysis using the Kruskal-Wallis test (continuous variables), the Pearson chi-squared test or the Fisher exact test (categorical variables) and by assessing their actual impact on the treatment effect through meta-regression analyses [17,18]. A metaregression analysis was performed to determine the relationship of potentially confounding factors (e.g., mean age, proportion of females, number of episodes, total number of participants, patient status, publication year, sponsorship, duration of preliminary phase, country, discontinuation methods, risk of bias, antidepressant class, dosage schedule, and antidepressant dose) to the magnitude of the effect on the primary outcome. Funnel plots were created to investigate potential publication bias. Finally, to assess the credibility of the findings of each network meta-analysis, the findings were incorporated into the Confidence in Network Meta-Analysis (CINeMA) application, which is an adaptation of the Grading of Recommendations Assessment, Development, and Evaluation approach [19][20][21].

Study characteristics
The literature search and selection strategy are depicted in Fig. S1. The initial search retrieved 148 articles, 50 of which were excluded as duplicates, 95 were excluded based on a review of the abstract and/or title, and three were included in our study [22][23][24]. In addition, 31 studies were retrieved  by manually searching the reference lists of previous review article [7]. There were no additional studies found in the clinical trial registers. Finally, the present review included a total of 34 DBRPCTs comprising 9384 patients with MDD (mean age = 43.80 years and %females = 68.10%). The characteristics of the 34 DBRPCTs included are summarized in Table S4. The average length of the study was 40.94 ± 16.27 weeks. Adults in the maintenance group were administered agomelatine (K = 2), amitriptyline (K = 1), bupropion (K = 1), citalopram (K = 3), desvenlafaxine (K = 2), duloxetine (K = 2), escitalopram (K = 1), fluoxetine (K = 4), fluvoxamine (K = 1), levomilnacipran (K = 2), milnacipran (K = 1), mirtazapine (K = 1), nefazodone (K = 1), paroxetine (K = 2), reboxetine (K = 1), sertraline (K = 2), tianeptine (K = 1), venlafaxine (K = 3), vilazodone (K = 1), and vortioxetine (K = 2). In 32 studies, participants in the acute study were required to have a scale-derived minimum of symptoms at baseline. However, one study lacked such a criterion, while another lacked detailed information on the criterion. Although 20 of the studies included only outpatients, six included both inpatients and outpatients, and the remaining eight did not report the status. All studies employed operationalized criteria such as those found in Diagnostic and Statistical Manual of Mental Disorders [56]. For the placebo group, the drug was discontinued abruptly (7 studies) and gradually (12 studies), and the remaining 15 studies did not report the detailed method of drug discontinuation. In addition, 31 studies were sponsored by the industry. The distribution of potential effect modifiers was similar across the comparisons in the network (Table S5). In at least one domain of the ROB2 tool, no studies were determined to be at high risk of bias (Table S6).

Network meta-analysis results
The network meta-analysis results are shown in Appendices S1-S13.
Tolerability and safety outcomes Compared to placebo, sertraline was associated with a higher rate of discontinuation due to adverse events ( Fig. 2 and Appendix S3). Compared to placebo, although desvenlafaxine, sertraline, and vortioxetine were associated with a higher incidence of nausea/ vomiting ( Fig. 2 and Appendix S4), venlafaxine was associated with a lower incidence of dizziness (Appendix S5). Compared to placebo, any antidepressants were not associated with an increased incidence of headache, somnolence, insomnia, dry mouth, constipation, sweating, weight gain, or sexual dysfunction (Appendices S6-13).
Heterogeneity, inconsistency, and network meta-analysis results graded using the CINeMA application Global heterogeneity was rated as moderate for all outcomes, except for constipation and sexual dysfunction, for which global heterogeneity was rated as high (Appendices S1-13). A considerable local heterogeneity was observed for the majority of outcomes in specific comparisons. Statistical evaluation of incoherence was impossible due to the absence of a head-tohead study comparing various antidepressants. Between network meta-analysis and pairwise meta-analysis, results showed differences in the following in comparison to placebo: agomelatine and levomilnacipran for the 6-month relapse rates, tianeptine for allcause discontinuation rate, desvenlafaxine and mirtazapine for discontinuation rates due to adverse events, sertraline for nausea/ vomiting, desvenlafaxine for dizziness, duloxetine for dry mouth, citalopram for constipation, and sertraline for sexual dysfunction. The within-study bias was rated as "some concerns" for all comparisons. Because funnel plots with fewer than 10 studies were not meaningful [9], all comparisons for publication bias were rated as "suspected," and any inconsistency could not be evaluated. Furthermore, the comparison was downgraded one level if it was based only on indirect evidence. Therefore, the confidence in the evidence for all comparisons other than vortioxetine versus placebo (low) in terms of the primary outcome was rated as "very low (Appendix S1)."

DISCUSSION
To the best of our knowledge, this is the first systematic review and network meta-analysis to investigate which antidepressant has the best balance of efficacy and acceptability for the treatment of adult individuals with MDD in the maintenance phase. Although desvenlafaxine, paroxetine, sertraline, venlafaxine, and vortioxetine had the best balance, sertraline was not well tolerated due to its association with nausea/vomiting. Therefore, desvenlafaxine, paroxetine, venlafaxine, and vortioxetine may be beneficial to individuals with MDD in the maintenance phase. However, desvenlafaxine and vortioxetine were associated with a risk of nausea/vomiting in adults with MDD in the maintenance phase as well as in the acute phase [57]. The efficacy, acceptability, tolerability, and safety of the treatment of MDD in the maintenance phase should be carefully considered as treatments prescribed for an acute depressive episode are typically continued into maintenance. Results of a network meta-analysis of adults with acute MDD also revealed that desvenlafaxine, paroxetine, venlafaxine, and vortioxetine had good efficacy and acceptability [8].
In contrast, the findings of the present network meta-analysis suggest that agomelatine, bupropion, escitalopram, levomilnacipran, milnacipran, and vilazodone did not outperform the placebo in terms of 6-month relapse rate. The original DBRPCTs reported that although vilazodone did not differ from placebo in terms of relapse rate at the study-endpoint [23], escitalopram and bupropion were superior to placebo [43,55]. Two DBRPCTs on agomelatine had inconsistent results [32,33]. One DBRPCT reported that levomilnacipran outperformed placebo in terms of relapse rate at the study-endpoint [22], while another DBRPCT did not report the statistical result of the outcome [49]; one trial investigating milnacipran also did not report the statistical results [47]. Our pairwise meta-analysis showed that agomelatine and levomilnacipran outperformed the placebo (Appendix S1). Due to the small number of individuals in these antidepressant trials, the 95% CrIs for the primary outcome in the network meta-analysis might be wider. As a result, our network meta-analysis might not Fig. 1 Forest plots for the 6-month relapse and all-cause discontinuation rates. A 6-month relapse rate and B all-cause discontinuation rate.
Medications were compared to a placebo. Colors indicate the presence or absence of a statistically significant difference, with blue indicating that the drug was superior to the placebo and black indicating that the drug was comparable to the placebo. 95% CrI 95% credible interval, AGO agomelatine, AMI amitriptyline, BUP bupropion, CIT citalopram, DES desvenlafaxine, DUL duloxetine, ESC escitalopram, FLUO fluoxetine, FLUV fluvoxamine, LEV levomilnacipran, MIL milnacipran, MIR mirtazapine, NEF nefazodone, PAR paroxetine, REB reboxetine, RR risk ratio, SER sertraline, TIA tianeptine, VEN venlafaxine, VIL vilazodone, VOR vortioxetine. Fig. 2 Forest plots for discontinuation rate due to adverse events and nausea/vomiting. A Discontinuation rate due to adverse events and B nausea/vomiting. Medications were compared with a placebo. Colors indicate the presence or absence of a statistically significant difference, with red indicating that the drug was inferior to the placebo and black indicating that the drug was comparable to the placebo. 95% CrI 95% credible interval, AGO agomelatine, BUP bupropion, CIT citalopram, DES desvenlafaxine, DUL duloxetine, ESC escitalopram, FLUO fluoxetine, FLUV fluvoxamine, LEV levomilnacipran, MIR mirtazapine, NEF nefazodone, PAR paroxetine, REB reboxetine, RR risk ratio, SER sertraline, VEN venlafaxine, VIL vilazodone, VOR vortioxetine. be able to detect the significant differences between these antidepressants and placebo.
A previous meta-regression analysis based on a pairwise metaanalysis showed that the effect size of the relapse rates was greater for tricyclics, selective serotonin reuptake inhibitors, and other newer agents, in that order, compared with the placebo [7]. However, our study did not demonstrate this trend (Appendix S1). Through a network meta-analysis, the relative effects can be estimated using any pair of interventions in the network simultaneously as well as the ranking and hierarchy of the interventions based on effectiveness [9,58]. Thus, when comparing the efficacy of individual antidepressants, a network meta-analysis is likely to yield more robust results than a pairwise meta-analysis.
There are some limitations to this study. First, the number of participants and DBRPCTs for some antidepressants, especially for tricyclic antidepressants, is small. The results of the present metaanalysis for some antidepressants were based on only one study. Second, important clinical issues regarding treatment decisionmaking in routine clinical practice (e.g., monotherapy or combination of antidepressants with nonpharmacological treatments) were not covered. A Finnish nationwide cohort study of individuals with severe MDD requiring hospitalization (mean follow-up time, 7.9 ± 5.3 years) found that lithium treatment was associated with the lowest risk of hospital readmission in patients with severe unipolar depression compared with other pharmacological treatments such as antidepressant and antipsychotics [59,60]. Sim and colleagues also reported that psychotherapy may have long-term benefits, particularly for patients with at least three previous major depressive episodes [10]. However, because there were no DBRPCT with an enrichment design for those treatments, our study did not evaluate these treatments for individuals with MDD. Third, due to a lack of available data, our study did not include some important antidepressant side effects such as agitation.
In conclusion, antidepressants such as desvenlafaxine, paroxetine, venlafaxine, and vortioxetine had balanced efficacy, acceptability, and tolerability in the treatment of adults with MDD in the maintenance phase. However, desvenlafaxine and vortioxetine had a risk of nausea/vomiting in adults with MDD in both the maintenance and acute phases.

DATA AVAILABILITY
The current study data were reported in articles cited in this paper.