Adverse effects of 21 antidepressants on sleep during acute-phase treatment in major depressive disorder: a systemic review and dose-effect network meta-analysis

Abstract Study Objectives Sleep-related adverse effects during acute treatment with antidepressants undermine adherence and impede remission. We aimed to address subtypes of sleep-related adverse effects and depict the relationship between dose and sleep-related adverse events. Methods We searched PubMed, Embase, Cochrane Central Register of Controlled Trials, and Web of Science for double-blind randomized controlled trials of depression published before April 30th, 2023. Eligible studies reporting sleep-related adverse effects during short-term monotherapy were included. The odds ratios (ORs) for sleep-related adverse effects were addressed with network meta-analysis. A Bayesian approach was used to depict the dose-effect relationship. Heterogeneity among studies was assessed using the τ2 and I2 statistics. Sensitivity analyses were performed without studies featuring high risk of bias. Results Studies with 64 696 patients were examined from 216 trials. Compared to placebo, 13 antidepressants showed higher ORs for somnolence, of which fluvoxamine (OR = 6.32; 95% CI: 3.56 to 11.21) ranked the top. Eleven had higher risks for insomnia, reboxetine ranked the top (OR = 3.47; 95% CI: 2.77 to 4.36). The dose-effect relationships curves between somnolence or insomnia and dose included linear shape, inverted U-shape, and other shapes. There was no significant heterogeneity among individual studies. The quality of evidence for results in network meta-analyses was rated as very low to moderate by Grading of Recommendations Assessment, Development, and Evaluation. Conclusions Most antidepressants had higher risks for insomnia or somnolence than placebo. The diverse relationship curves between somnolence or insomnia and dose of antidepressants can guide clinicians to adjust the doses. These findings suggest clinicians pay more attention to sleep-related adverse effects during acute treatment with antidepressants.


Introduction
Major depressive disorder (MDD) is a common mood disorder, affecting about 20% of people worldwide, which is predicted to rank as the leading cause of global impact of psychiatric disease by 2030 [1].Antidepressants are usually prescribed for MDD patients and are recommended as a first-line treatment for moderate and severe depression [2].At present, more than 30 antidepressants are generally used in treatment of MDD, including selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors, and others.Although most of these are effective, it is necessary for clinicians to balance their efficacy and acceptability [3].There are many adverse effects commonly associated with antidepressants, including disordered sleep, sexual dysfunction, and gastrointestinal side effects, and these may result in discontinuation during the acute-phase treatment [2].Reviews and meta-analyses on sexual dysfunction [4] and gastrointestinal side effects [5] were reported recently.Sleep-related adverse effects during short-term treatment with antidepressants not only undermine patient adherence but are also associated with an impediment to achieve remission, greater functional impairment, and higher risk of recurrence [6].But meta-analysis on the sleep-related adverse effects is still scarce and the prevalence of treatment-emergent sleep disturbance in patients with MDD taking antidepressants is ambiguous.
Some systematic reviews have reported the influence of antidepressants on sleep architecture and physiology [7,8], providing evidence that different antidepressants with different action mechanisms and pharmacokinetics may have different adverse effects on sleep [9].Insomnia and somnolence have a significant influence on patients who require alertness in their work, which may include driving or operating heavy machinery, resulting in severe adverse events.A post-marketing adverse drug reaction study [10] assessed and ranked the odds ratios (ORs) of somnolence among 30 antidepressants under a wide array of clinical indications circumstances not limited to MDD.Some meta-analyses [11,12] qualified and compared the rates of insomnia and somnolence associated with second-generation antidepressants during acute-phase treatment of MDD.Other sleep disorders, including nightmares, sleep terrors, restless leg syndrome, sleep paralysis, sleep-related hallucinations, and sleepwalking can also be found and affect clinicians' choice of antidepressants [13].Thus, it is important to clarify the association between antidepressants and adverse effects on sleep [7,9].A comprehensive comparative analysis that ranks the odds of adverse effects of common antidepressants on sleep during short-term treatment for MDD is an unmet clinical need.Network meta-analyses of datasets from high-quality double-blinded randomized controlled trials (RCTs) make it possible to measure the rates and risks of sleep-related adverse events and provide essential evidence for clinicians to conduct optimal treatment [3].
Dose-effect relationships of antidepressants have been reported in several studies [14][15][16], which show that efficacy is not always dose-dependent.The debate on whether higher doses are more efficacious or not is still ongoing, and the same happens with dose and sleep-related adverse effects.
For instance, different dosages of trazodone have different effects on sleep architecture [8].Low doses of mirtazapine are often prescribed off-label for insomnia clinically.And its sedation effect might be attenuated at higher doses, probably due to increased serotonin and norepinephrine release [17].It is essential for clinicians to identify the relationship between dosage and sleep-related adverse effects to make better use of antidepressants, but few studies focus on this aspect.
Our study was specifically designed to address the different subtypes of sleep-related adverse effects during acute treatment with antidepressants for patients living with MDD and depict the relationship between dose and these adverse events.To this end, we conducted a systematic review and dose-effect network meta-analysis.

Methods
We reported this study according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [18].The protocol was registered on PROSPERO, the International prospective register of systematic reviews (CRD42022339567).

Data source and search strategy
We systematically searched multiple databases, including PubMed, Embase, the Cochrane Central Register of Controlled Trials (CENTRAL), and Web of Science, for articles published before April 30th, 2023.We also searched clinicaltrials.govfor unpublished trials.The reference lists of previous systematic reviews were screened to supplement study inclusion.The search terms included "depress," "dysthymia," and names of included antidepressants, among others.The detailed search strategy is described in Supplementary Table S1.

Study selection and data extraction
Double-blinded RCTs involving adults (≥18 years) with MDD and that compared antidepressants with placebo, compared different antidepressants, or compared different doses of antidepressants were eligible.Studies reporting sleep-related adverse effects that occurred during the trial period were included.For the screened studies, we went through their method section and protocol, if applicable, to confirm all the investigated outcomes.If any sleep-related adverse effect was mentioned in the method section or protocol but not reported in the result, we consider zero events occurred for this outcome.According to the previous large-scale reviews [3,4] and clinical practice, we included 21 antidepressants in the analysis of odds risks of sleep-related adverse effects and dose-effect network meta-analysis (see Table 1 and Table 2).Toludesvenlafaxine, a new triple reuptake inhibitor with a good effect profile [19,20], was only included in the analysis of the risk of somnolence because of the limited study data, which were a revision from original protocol.Only monotherapy was considered.We excluded RCTs of women with postpartum depression, of patients with post-stroke depression, of participants that consisted of more than 20% of bipolar or psychotic depression, and of participants with resistant depression and concomitant medical illness.
We extracted the name and the dosage per day of the antidepressants, as well as the feature of the study populations, including sample size, baseline severity, percentage of females, and mean age.Basic information was also extracted, such as year of publication, first author, and study region.We summarized information on study design and result reporting to assess the risk of bias within each individual study.
The main outcomes that we focused on were somnolence and insomnia.Both self-reported and clinically confirmed treatment-emergent somnolence and insomnia were extracted.In addition, we also extracted the numbers of participants that experienced other subtypes of sleep-related adverse effects, The risk of bias in each included study was assessed using the Cochrane risk of bias tool (RoB2.0)[21], which assesses the following domains: randomization, deviation from the intended interventions, missing outcome data, outcome measures, selection of the reported results, and overall bias.The quality of evidence for results in network meta-analyses was rated through the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) process from five aspects, including risk of bias, imprecision, inconsistency, indirectness, and publication bias [22].
Two groups of reviewers selected the studies, extracted information, and conducted the quality assessment independently.Every group consisted of three reviews, who were master or PhD candidates (PL, JWZ, YMT and XFL, MZ, and FQL).We did not use any automation tool in this process.Data were double-checked across two groups who worked independently.Discrepancies were resolved by discussion or consultation with senior investigators (FS and XZL), both were associate professors and specialists.

Statistical analyses
We conducted network meta-analyses using the frequentist method, where different doses of the same antidepressant were treated as a single treatment.A random effects model was used to calculate pooled ORs and their 95% confidence intervals (95% CI).We calculated the surface under the cumulative ranking curve (SUCRA) to rank different treatments.Heterogeneity among studies was assessed using the τ 2 and I 2 statistics.We performed a meta-regression using the Bayesian method instead of subgroup analysis to explore the potential interaction between the study characteristics and antidepressants, according to study region, mean age of the study population (≥50/<50 years), percentage of females, and baseline severity (depression rating score).This alternative analysis method was a modification of the original protocol which had been amended in PROSPERO.In the Bayesian models, the number of the chain was three and the number of total interactions per chain was 50 000.Global inconsistency was evaluated using a generalized Q test, while local inconsistencies were detected using a node split approach [23].Funnel plots were drawn to check for publication bias.The transitivity assumption was assessed by comparing the baseline characteristics of populations with different treatment comparisons, and box plots were used to display the similarity of the baseline characteristics.Sensitivity analysis was carried out by excluding studies that were assessed to feature high risk of bias.
Using a Bayesian approach, we implemented model-based network meta-analyses to explore the dose-effect relationship between dose and the risks for sleep-related adverse effects [24].
We estimated the maximum effect (ORmax) and its 95% credible interval (95%CrI) using an Emax model [25].Only fixed-dose arms were included in the dose-effect modeling.
We performed the statistical analyses with R 4.0.1, and mainly used the netmeta, gemtc, and model-based network meta-analysis dose packages.p < .05 was considered to indicate significance in all statistical tests.

Results
In all, 38 929 records were identified from the databases, and 104 studies were retrieved from the reference lists of published reviews.After screening the titles and abstracts, the full texts of 999 studies were reviewed.Following this, 216 studies were included in our systematic review, of which 163 were included in the meta-analyses for somnolence and 166 were included in the meta-analyses for insomnia (Figure 1).Other subtypes of sleep-related adverse effects were reported sparsely and could not be synthesized quantitively.

Study characteristics
In total, 64 696 participants were enrolled in the included studies, of whom 63.2% were female, with a mean age of 45.5 years.Of the included studies, 127 (58.8%) studies involved a group of placebo controls, while 45 (20.8%) included a group treated with paroxetine, which is the most common active drug treatment among all 22 antidepressants.In addition, 85 (39.4%) studies were carried out in North America, 58 (26.9%) were conducted in Europe, 20 (9.7%) were from Asia, and 32 (14.8%) were carried out across multiple continents (detailed information on characteristics of individual studies can be viewed in Supplementary Table S2).

Risk of bias and quality of evidence
Most studies (n = 158) were assessed to be of some concern in terms of overall bias.To be more specific, more than 80% of the included studies probably introduced risk of bias during the randomization process and outcome measurement.Because we only included double-blinded RCTs, there was a low risk of deviation from the intended interventions in the majority of the studies.Notably, 51 (23.6%) studies were at high risk for missing outcome data (Figure 2) (detailed information on risks for each study can be viewed in Supplementary Table >S3).

Results of network meta-analyses
In all, 163 studies were included to construct a network for comparing the associations between different antidepressants and somnolence (Figure 3A).The prevalence of somnolence for placebo was approximately 4.5%.moderate, due to withinstudy bias somnolence for every antidepressant was calculated based on the OR of relative effect and the prevalence for placebo, such as agomelatine 6.1% and amitriptyline 17.3% (Table 1).Taking the placebo as a control group, fluvoxamine (OR = 6.32; 95% CI: 3.56 to 11.21), trazodone (OR = 4.64; 95% CI: 3.17 Bupropion (OR = 0.50; 95% CI: 0.30 to 0.82) had a lower risk for somnolence compared to placebo (Figure 4A).The 166 studies that compared different antidepressants with respect to insomnia outcomes were included in a network (Figure 3B).The prevalence of insomnia for placebo was approximately 5.4%.The prevalence of insomnia for every antidepressant was calculated based on the OR of relative effect and the prevalence for placebo, such as agomelatine 5.3% and amitriptyline 3.5% (Table 2).Compared to placebo, reboxetine (OR = 3.47; 95% CI: 2.77 to 4.36), vilazodone (OR = 2.99; 95% CI: 1.78 to 5.03), desvenlafaxine (OR = 2.12; 95% CI: 1.50 to 2.99), duloxetine (OR = 1.96; 95% CI: 1.60 to 2.42), bupropion (OR = 1.83; 95% CI: 1.42 to 2.36), venlafaxine (OR = 1.77; 95% CI:  .37;95% CI: 1.08 to 1.75) had a higher risk for insomnia.However, amitriptyline (OR = 0.63; 95% CI: 0.42 to 0.92) was associated with a lower risk for insomnia than that of placebo (Figure 4B).Heterogeneity among individual studies was not significant for somnolence or insomnia (τ 2 = 0.098, I 2 = 29.1% [95% CI: 13.2%~41.1%]and τ 2 = 0.013, I 2 = 5.7% [95% CI: 0.0%~22.5%],respectively).In the meta-regression, there was no significant interaction between baseline characteristics (mean age, percentage of females, and baseline severity) and antidepressants in terms of the risk for both main outcomes, except for gender, and baseline severity had an effect on toludesvenlafaxine (Supplementary Tables S4-S9).However, a significant global inconsistency was detected in the networks of both outcomes.The local inconsistency was explored using the node split approach (Supplementary Table S10).No publication bias was detected from the funnel plot (Supplementary Figures S3 and S4).Baseline characteristics, including the percentage of females, mean age, and baseline severity, were similar across different designs (Supplementary Figure S5).When studies with high risk of bias were excluded from the network, the results were consistent with the main analyses (Supplementary Figures S6 and S7).
Information on other subtypes of sleep-related adverse effects was presented in Supplementary Table S11.Sleep disorders, abnormal dreams, and yawning were the most commonly reported secondary outcomes, although the rates were very low.
The quality of evidence for results in network meta-analyses was rated as very low to moderate.The results of GRADE assessment are displayed in Table 1 and Table 2.

Dose-effect relationship of each drug between dose and sleep-related adverse effects
Dose-effect relationships of antidepressants between dose and sleep-related adverse events were presented in Figure 5.As for fluoxetine, milnacipran, nefazodone, and sertraline, the risks of somnolence increased linearly along with the dose increasing within conventional therapeutic doses.When it comes to amitriptyline, desvenlafaxine, duloxetine, escitalopram, paroxetine, toludesvenlafaxine, trazodone, and venlafaxine the risks of somnolence gradually increased from low doses to moderate doses and then showed a decreasing trend through the high doses.In    Meanwhile, the risks of insomnia stayed flat from low doses to moderate doses and then increased through high doses of bupropion, citalopram, escitalopram, paroxetine, sertraline, and vortioxetine.With respect to desvenlafaxine, duloxetine, fluoxetine, venlafaxine, and vilazodone, the risks of insomnia increased gently and then presented a flat trend.Some antidepressants only appeared in one dose in the constructed model, and due to the limited sample size, some of the 95% CI were wide both in somnolence and insomnia.
Maximum effects were detected in the relationship between somnolence and doses of 21 antidepressants, 13 antidepressants had significantly higher risks than that placebo, not including milnacipran due to the wide 95% CI (OR = 1.68; 95% CI: 0.91 to 12.91).Regarding to maximum effects of insomnia, 11 antidepressants were estimated to have significantly higher risks of insomnia compared to placebo, which was consistent with the results of network meta-analysis.Bupropion (OR = 0.49; 95% CI: 0.28 to 0.83) had lower risks for somnolence even at maximum effects, whereas amitriptyline (OR = 0.63; 95% CI: 0.31 to 0.96) still had lower risks for insomnia compared to placebo (Table 3).The dose-effect models were fitted with the estimated effect size of the specific dose of antidepressants.

Discussion
Our study investigated 216 double-blinded RCTs including 64 696 depressed patients to identify the ORs of insomnia and somnolence induced by different antidepressants.We found that fluvoxamine, trazodone, mirtazapine, amitriptyline, duloxetine, escitalopram, paroxetine, nefazodone, sertraline, fluoxetine, venlafaxine, desvenlafaxine, milnacipran, and agomelatine had higher ORs for somnolence and bupropion had a lower OR than placebo.For insomnia, reboxetine, vilazodone, desvenlafaxine, duloxetine, bupropion, venlafaxine, sertraline, citalopram, fluoxetine, paroxetine, and escitalopram had higher ORs and amitriptyline had a lower OR than placebo.But the risks of somnolence and insomnia were not always increased linearly along with the dose increase for 21 antidepressants.
With the exception of fluvoxamine, selective serotonin reuptake inhibitors and serotonin and norepinephrine reuptake inhibitors had higher risks for somnolence and insomnia than placebo.This finding is consistent with previous studies [2,9] that reported that patients taking an SSRI or SNRI had similar likelihood of presenting with somnolence or insomnia.If patients complained of lethargy after taking an SSRI or SNRI in the morning, it was appropriate to administer it closer to bedtime.However, for fluvoxamine which had significantly higher risks for hypersomnia but not for insomnia than placebo, patients would benefit more by taking it at night.This might be related to the well-characterized ability of fluvoxamine to increase nocturnal serum levels of melatonin by 2-to 3-fold [26], with probable mechanism of inhibiting hepatic metabolism of melatonin by cytochrome P450 enzymes.
Agomelatine, mirtazapine, and trazodone are usually used to improve insomnia in depressed patients by changing their polysomnographic sleep architecture [8,16].Our results are consistent with the findings of previous studies.But these three antidepressants have different properties due to their different mechanisms.Agomelatine was given top acceptability in a previous network meta-analysis [3].As a melatonin receptor agonist (MT1 and MT2), agomelatine can increase total sleep time, improve sleep efficiency [16], and restore circadian rhythm.Compared to mirtazapine, agomelatine has a lower frequency of oversedation or tiredness within 90 days of treatment [27].Mirtazapine is a noradrenaline and specific serotonergic antidepressant that shows antagonism against the alpha-2 autoreceptor and heteroreceptors and strong antagonism against the 5-HT2, 5-HT3, and H1 receptors.In one study, mirtazapine was the most frequently associated with akathisia and restless leg syndrome, which can lead to difficulty falling asleep [28].Mirtazapine might show optimal acceptability at 30 mg instead of 45 mg [29].In a previous study, compared to the control group, somnolence, and dizziness occurred with greater frequency in the trazodone group [8], and a particularly complex action of antagonist on H1 histamine receptor, alpha 1, and alpha 2 adrenergic receptors of trazodone resulted in these unwanted side effects [30].Whether trazodone should be first-line therapy for insomnia is still under discussion [31].

Figure 1 .
Figure 1.The flow diagram for study selection.

Figure 2 .
Figure 2. The result of risk of bias assessment.

Figure 3 .
Figure 3.The network of evidence for main outcomes: (A) Somnolence; (B) Insomnia.The size of the nodes represents the sample size of each treatment group, and the width of the lines connecting different nodes is proportional to the number of RCTs comparing every pair of treatments.

Figure 4 .
Figure 4. Forest plot displaying the result of network meta-analysis for main outcomes.(A) Somnolence; (B) Insomnia.

Figure 5 .
Figure 5. Dose-effect relationship of each drug between dose and sleep-related adverse events.(A) somnolence; (B) insomnia; Y-axis represents the absolute risk for somnolence or insomnia of different doses of antidepressants.X-axis represents doses of antidepressants, 0 represents placebo.

Table 1 .
The estimated prevalence of Continued Zhou et al. | 5

Table 2 .
Summary of Finding for Insomnia

Table 3 .
The Odds Ratios of Somnolence and Insomnia of Antidepressants Compared to Placebo in Dose-Effect Models