Switch to mania after acute antidepressant treatment for bipolar depression: a systematic review and network meta-analysis of randomised controlled trials

Summary

Background

The potential for antidepressants to induce a switch to mania remains a major concern in the treatment of bipolar depression, but the specific risk associated with different antidepressants remains unclear. This systematic review and network meta-analysis (NMA) assessed this risk by comparing individual antidepressants with each other and with a common placebo.

Methods

In this systematic review and network meta-analysis, we searched ClinicalTrials.gov, CENTRAL, PsycINFO, PubMed, Scopus, and Web of Science from database inception up to Feb 19, 2025, with no language restrictions, for randomised controlled trials (RCTs) assessing acute antidepressant treatment in bipolar depression. The primary outcome was the rate of switch to mania after antidepressant treatment. A frequentist NMA estimated risk ratios (RRs) and 95% confidence intervals. Sensitivity analyses were performed based on treatment regimen (monotherapy or add-on), baseline severity, switch to mania definition, study setting, psychiatric comorbidity, treatment duration, non-pharmacological combinations, industry sponsorship, and risk of bias. Certainty of evidence was assessed using the CINeMA framework. The protocol was preregistered on the Open Science Framework.

Findings

Of 2434 records screened, 13 RCTs (1362 patients; 818 [60.1%] female, 511 [37.5%] male, and 33 [2.4%] not disclosed) were included in the NMA. Although some evidence of increased risk of switching to mania was observed, no antidepressant was associated with a significantly higher risk of switch to mania compared to placebo. Venlafaxine showed the highest risk estimate among antidepressants, though not statistically significant RR (4.53 [95% CI 0.47–43.25]), and was the only compound with consistent signals of increased switch in individual studies. The evidence base was larger for add-on therapy, while fewer data were available for monotherapy. Sensitivity analyses confirmed the results. Heterogeneity was low. Overall confidence in the evidence was rated as low.

Interpretation

Antidepressants remain a treatment option for acute bipolar depression, particularly as add-on therapy. Their use should be individualised, considering patient-specific profiles and other potential risks, in line with a precision psychiatry approach. Further studies are needed to clarify long-term safety.

Funding

None.

Research in context.

Evidence before this study

Switching to mania after antidepressant treatment remains one of the most debated concerns in the pharmacological management of bipolar depression. Despite widespread off-label use, guidelines offer inconsistent recommendations due to a lack of clear comparative evidence among antidepressants. We searched ClinicalTrials.gov, CENTRAL, PsycINFO, PubMed, Scopus, and Web of Science up to January 12, 2024, using the terms “bipolar disorder,” “mania,” “hypomania,” “switch,” and “antidepressant∗,” to identify systematic reviews and meta-analyses. Previous work has suggested that antidepressants as a class may be associated with more switches than antipsychotics, but not more than placebo. No prior network meta-analysis has directly compared the risk of switching to mania across individual antidepressants in bipolar depression.

Added value of this study

This is the first systematic review and network meta-analysis focused exclusively on the risk of antidepressant-induced switch to mania in bipolar depression. By including only randomised controlled trials and following a pre-registered protocol with PRISMA-NMA guidance, RoB2 risk of bias assessment, and CINeMA grading of certainty, this study offers the most rigorous comparative safety evaluation to date. Although some evidence of increased risk of switching to mania was observed, no individual antidepressant showed a significantly increased switch risk compared to placebo. However, venlafaxine consistently showed the highest relative risk across individual studies, sensitivity and post-hoc analyses. These findings offer a more granular understanding of switch risk among antidepressants and highlight venlafaxine as a compound warranting special attention.

Implications of all the available evidence

Our findings support the cautious short-term use of antidepressants in bipolar depression, particularly as add-on therapy, in line with guideline recommendations. The absence of clear differences in switch risk suggests that treatment decisions should prioritize patient-specific factors—including bipolar subtype, treatment history, comorbidities, and prior switch episodes—over pharmacological class alone. While these results provide reassurance regarding acute safety, the evidence base remains limited by short trial durations and low confidence ratings. Longer-term studies and real-world data are needed to evaluate sustained safety and to guide the discontinuation of antidepressants after remission.

Introduction

Bipolar disorder (BD) is a chronic mental illness characterised by periods of mania and depression, with inter-critical periods of absent or subsyndromal mood symptoms.¹ The lifetime prevalence is estimated to be between 0.4% and 1.1%, making BD one of the leading causes of disability and social burden worldwide.²

Individuals with BD spend a significant proportion of their time experiencing mood symptoms, being euthymic only about 50% of the time³ Although mania is the hallmark feature distinguishing BD from recurrent major depression, depressive phases are three times more frequent than manic or hypomanic phases.³ Bipolar depression diminishes quality of life and functioning and also significantly increases mortality risk, including suicide.⁴ However, the treatment of bipolar depression remains a significant challenge in BD management, with only five pharmacological treatments having received FDA approval for bipolar depression: olanzapine/fluoxetine combination, quetiapine, lurasidone, cariprazine, and lumateperone.⁵ Despite the limited number of approved treatments, several guidelines recommend the use of other evidence-based off-label treatments, including antidepressants, mood stabilisers, and additional antipsychotics.⁶

The safe and appropriate use of antidepressants for the acute treatment of bipolar depression is a topic of ongoing debate. Concerns arise from evidence indicating that antidepressants may induce manic and hypomanic episodes.⁷ The emergence of a manic or hypomanic episode during a depressive episode is often referred to as a “switch to mania”, and may result from the natural course of the illness, substance use, pharmacological treatment, or other causes. Terms such as “antidepressant-induced” or “treatment-emergent mania” are often used specifically to describe switches triggered by antidepressant use. Despite these nuances in terminology, the clinical relevance of manic switch remains a central concern when considering antidepressants for bipolar depression. The largest network meta-analysis (NMA) of acute pharmacotherapy for bipolar depression to date found that antidepressants as a class led to more switches to mania than other classes, such as antipsychotics.⁸ However, no differences were observed when the antidepressant class was compared to placebo.⁸ Similar results were reported when individual antidepressants (i.e., each single drug) were analysed separately alongside other individual medications from other pharmacological classes: no increased risk of switching was observed for any of the antidepressants studied in randomised controlled trials (RCTs) compared to placebo.⁸ However, in that analysis, antidepressants were either grouped together as a pharmacological class or evaluated within a large network including other treatments such as antipsychotics and mood stabilisers. This broader network structure may have limited the resolution of treatment-specific effects within the antidepressant class, reducing the ability to detect potential differences between individual agents.

No NMA has yet compared the rates of switch to mania in patients with bipolar depression exclusively within the class of antidepressants. In addition, no NMA has specifically examined differences among distinct antidepressant classes. We hypothesised that such a focused network including only antidepressants allows for direct and indirect comparisons among individual agents within the same class, potentially improving both precision and clinical applicability, thus offering valuable insights to guide clinical decision-making in the acute management of bipolar depression. To address this, we conducted a systematic review and NMA of antidepressant interventions for the acute treatment of bipolar depression to evaluate the risk of switching to mania across different antidepressant treatments.

Methods

Study design and ethics

We followed the PRISMA-NMA reporting guidelines.⁹ The protocol was preregistered on the Open Science Framework (OSF) platform (at https://osf.io/vw4zq/?view_only=863eeddfbe3a450a88f7fe5b9225c464; amendements are reported in Appendix 1 p 7). This study is based on data extracted from previously published and publicly available trials. Therefore, ethical approval and informed consent were not required.

Search strategy and selection criteria

We searched electronic databases (ClinicalTrials.gov, CENTRAL, PsycINFO, PubMed, Scopus, and Web of Science) from database inception until Feb 19, 2025 (Appendix 2 pp 8–12) and previous reviews,^8,10,11 with no language restrictions.

We considered published and unpublished RCTs including: Population, participants of any age in any setting (both inpatients and outpatients), with a diagnosis of BD and a current depressive episode according to any version of the Research Diagnostic Criteria (RDC), the Diagnostic and Statistical Manual of Mental Disorders (DSM), or the International Classification of Diseases (ICD); Intervention, acute antidepressant treatment classified under the Anatomical Therapeutic Chemical (ATC: N06A) system, administered either as monotherapy (i.e., without concurrent mood stabiliser, such as lithium, anticonvulsants, or antipsychotics) and add-on therapy (i.e., added to an existing mood stabiliser regimen). Combination regimens with other pharmacological classes (defined as the simultaneous initiation of antidepressants and other psychotropic agents, such as antipsychotics or mood stabilisers) were excluded to isolate the specific contribution of antidepressant treatment to the risk of switching to mania, and to avoid confounding from concurrent pharmacological interventions; Comparison, placebo or another antidepressant agent; Outcome, switch to manic or hyponamic episode following antidepressant treatment during an acute depressive episode, as reported by the original study authors. This definition encompasses what is often referred to as antidepressant-induced mania or treatment-emergent affective switch, and we consistently use the term switch to mania throughout the manuscript to reflect this phenomenon. No additional operational criteria were applied beyond those provided in the original trials, in line with previous meta-analyses⁸; and Timeframe, studies evaluating treatment during the acute phase of the current depressive episode. Trials conducted during continuation or maintenance phases were excluded.

When we identified studies that met our eligibility criteria, the following data on study-level characteristics were systematically extracted: first author, year of publication, study design, country and geographical region, clinical setting, sponsor, trial duration, washout period, specific interventions and comparators, classification of interventions according to the Neuroscience-based Nomenclature-3 (NbN-3) framework¹² and mechanisms of action,¹³ treatment regimen (monotherapy or add-on therapy), medication doses, permitted concomitant medications, sample size for each arm, diagnostic criteria, and structured diagnostic interviews used. For each study, we extracted aggregate data on participant characteristics, including: age, sex, baseline symptom severity, duration of illness, duration of the current depressive episode, age at onset of BD, number of previous depressive and manic episodes, psychiatric comorbidities, number of patients diagnosed with BD-I and BD-II, and characteristics of prior mood stabiliser treatment. Regarding outcomes, we extracted the definitions and measurement tools used to assess switch to mania, the type of statistical analysis (intention-to-treat or per-protocol), and the reported switch rates.

Study screening, data extraction, and risk of bias assessment using the Risk of Bias tool 2 (RoB2) were conducted in duplicate by ELS and SP and confirmed by VO.

Statistical analysis

A frequentist NMA based on random-effects models was conducted. This approach combines both direct evidence (obtained from conventional pairwise meta-analyses of head-to-head trials) and indirect evidence (obtained through pairwise comparisons across the network via a common comparator) to estimate relative treatment effects. The risk ratio (RR) was calculated along with 95% confidence intervals (95% CI). Statistical heterogeneity in the networks was assessed using tau² for between-study variance and I² for global heterogeneity. Transitivity was evaluated based on study design, diagnosis, and the comparison of potential effect modifiers (e.g., mean age, sex, and baseline symptom severity) across studies. In case of identification of outliers in these comparisons, sensitivity analyses were conducted by removing them from the main analysis. Inconsistency was evaluated locally using the node-splitting analysis and globally across the entire network using the design-by-treatment Q statistic. P-scores were calculated to estimate the relative ranking of treatments in terms of safety (i.e., lower risk of switch). They are based on the point estimates and standard errors from the NMA and range from 0 (worst) to 1 (best), representing the mean certainty that a treatment is safer than the others. Publication bias was evaluated by visual inspection of the contour-enhanced funnel plot (i.e., for the comparison of any pharmacological intervention versus placebo) and comparison-adjusted funnel plot (i.e., assuming the direction of bias to the more recent interventions of the comparison), and Egger's test. Sensitivity analyses were conducted based on treatment regimen (antidepressants used exclusively as add-on therapy, exclusively as monotherapy, or both within the same sample), baseline symptom severity, definition of switch to mania (if based on standardised criteria and validated rating scales, or on clinical observation only), study setting, psychiatric comorbidity, study duration, inclusion of non-pharmacological treatments, industry sponsorship, and exclusion of studies with high overall risk of bias. Post-hoc analyses were conducted to group antidepressants according to the NbN-3¹² and the most commonly used pharmacological classification (which considers mechanisms of action).¹³ Additionally, a post-hoc pairwise meta-analysis was conducted to explore the risk of switching to mania using only studies that directly compared antidepressants and placebo.

Certainty of evidence was assessed for primary outcomes using the Confidence-In-Network-Meta-Analysis (CINeMA) framework.

All statistical analyses were performed using R version 4.4.1, and its packages netmeta and metafor.

Role of the funding source

There was no funding source for this study.

Results

A total of 2434 records were identified from different sources After title and abstract screening, 209 full-text articles were assessed, 13 RCTs met the inclusion criteria and were included in the systematic review and NMA (Fig. 1).14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26 The characteristics of the included studies are provided in Table 1 and in Appendix 3 pp 13–15. A list of excluded studies, with the respective reasons for exclusion, is provided in Appendix 3 pp 16–30.

Fig. 1.

Study selection.

Table 1.

Charateristics of included studies in systematic review.

Author, year (country)	Intervention (regimen)	N.	Mean ± SD age; % of females	Baseline depression severity (category)	N. of switches to mania (%)	Definition of switch to mania	Duration (study setting)	Diagnostic criteria (psychiatric comorbidity)	Industry sponsorship	RoB
Amsterdam et al., 2005 (USA)	Fluoxetine (mono and add-on therapy)	8	41 ± 13; 37.5	HDRS-17 = 25 ± 5.9 (severe)	4 (50)	DSM IV criteria and YMRS ≥12	8 weeks (outpatients)	DSM-IV	Yes	Some concerns
	Placebo (mono and add-on therapy)	9	39 ± 11; 22.23	HDRS-17 = 25 ± 5.2 (severe)	3 (33.34)
Baumhackl et al., 1989 (Austria)	Moclobemide (mono and add-on therapy)	18	53.4 ± 17.8; 74.6a	HDRS-17 = 25 ± 5.7 (severe)a	1 (5.56)	clinical observation	4 weeks (mixed)	DSM-III (not allowed)	No	High
	Imipramine (mono and add-on therapy)	15	55.6 ± 16.7; 74.5a	HDRS-17 = 24.3 ± 5.9 (severe)a	0 (0)
Benedetti et al., 1996 (Italy)	Amineptine (monotherapy)	11	50 ± 6.5; 72.72	MADRS = 31.64 ± 4.78 (moderate)	1 (9.09)	clinical observation	12 days	DSM-III-R (allowed)	No	Some concerns
	Placebo (monotherapy)	11	52.82 ± 11.96; 81.81	MADRS = 30.09 ± 8.43 (moderate)	3 (27.27)
Cohn et al., 1989 (USA)	Fluoxetine (mono and add-on therapy)	30	37 ± 8.75; 66.66	HDRS-21 = 27.7 ± 4.93 (severe)	0 (0)	clinical observation	6 weeks (outpatients)	DSM-III-R (allowed)	No	Some concerns
	Imipramine (mono and add-on therapy)	30	40 ± 10.75; 63.33	HDRS-21 = 26 ± 4.93 (severe)	2 (6.67)
	Placebo (mono and add-on therapy)	29	42 ± 12.75; 68.96	HDRS-21 = 27.2 ± 4.93 (severe)	1 (3.45)
Himmelhoch et al., 1991 (USA)	Tranylcypromine (monotherapy)	28	40.2 ± 11.6; 57.14	HDRS-17 = 22.1 ± 4.1 (severe)	3 (10.71)	RMS >5 and RDC criteria	6 weeks (outpatients)	DSM-III and RDC (not allowed)	No	Low
	Imipramine (monotherapy)	28	37.5 ± 10.8; 64.29	HDRS-17 = 23.2 ± 5 (severe)	5 (17.86)
McElroy et al., 2010 (Multicenter)	Paroxetine (monotherapy)	118	39.3 ± 11.45; 66.3	HDRS-17 = 24.1 ± 4.41 (severe)	13 (11.02)	YMRS ≥16	8 weeks	DSM-IV (not allowed)	Yes	Low
	Placebo (monotherapy)	121	38.7 ± 11.45; 66.9	HDRS-17 = 24.2 ± 4.41 (severe)	11 (9.09)
Nemeroff et al., 2001 (USA)	Paroxetine (add-on therapy)	35	42.5 ± 7.09; 54.3	HDRS-21 = 20.38 ± 3.91 (moderate)	0 (0)	clinical observation	10 weeks (outpatients)	DSM-III-R (not allowed)	Yes	Low
	Imipramine (add-on therapy)	39	41.9 ± 8.01; 59	HDRS-21 = 20.71 ± 3.9 (moderate)	3 (7.69)
	Placebo (add-on therapy)	43	40.4 ± 6.86; 53.5	HDRS-21 = 21.57 ± 3.87 (moderate)	1 (2.32)
Pilhatsch et al., 2010 (Germany)	Paroxetine (add-on therapy)	18	NA; 61	HDRS-21 = 24.4 ± 6.08 (severe)	1 (5.56)	DOTES	6 weeks	DSM-III-R	Yes	High
	Amitriptyline (add-on therapy)	22	NA; 50	HDRS-21 = 26 ± 5.46 (severe)	0 (0)
Post et al., 2006 (Multicenter)	Venlafaxine (add-on therapy)	65	40.6 ± 12.1; 50.8	IDS = 34.08 ± 10.2 (moderate)	10 (15.38)	YMRS ≥13	10 weeks (outpatients)	DSM-IV (not allowed)	Yes	Low
	Bupropion (add-on therapy)	51	43.4 ± 14.2; 44.8	IDS = 35.8 ± 9.7 (moderate)	2 (3.92)
	Sertraline (add-on therapy)	58	41 ± 11.8; 54.9	IDS = 30.6 ± 10.6 (moderate)	4 (6.89)
Sachs et al., 1994 (USA)	Desipramine (add-on therapy)	7	32.29 ± 14.58; 85.7	NA	2 (28.57)	clinical observation	8 weeks (outpatients)	DSM-III-R (allowed)	Yes	Some concerns
	Bupropion (add-on therapy)	8	41.62 ± 13.91; 75	NA	1 (12.5)
Silverstone et al., 2001 (Multicenter)	Moclobemide (add-on therapy)	81	44.4 ± 5.44; 70	HDRS-17 = 23 ± 3.8 (severe)	5 (6.17)	YMRS >10	8 weeks (mixed)	DSM-III-R (not allowed)	No	Low
	Imipramine (add-on therapy)	75	40.7 ± 5.66; 72	HDRS-17 = 22.5 ± 3.6 (severe)	7 (9.34)
Vieta et al., 2002 (Spain)	Paroxetine (add-on therapy)	30	47.1 ± 15.2; 63	HDRS-17 = 20.7 ± 3 (moderate)	1 (3.34)	DSM IV criteria and YMRS >11	6 weeks (mixed)	DSM-IV (not allowed)	No	Low
	Venlafaxine (add-on therapy)	30	45.5 ± 13.7; 70	HDRS-17 = 21.2 ± 3.2 (moderate)	4 (13.34)
Yatham et al., 2016 (Multicenter)	Agomelatine (add-on therapy)	172	45.6 ± 13.3; 61	HDRS-17 = 24.9 ± 3.8 (severe)	7 (4.07)	DSM-IV criteria	8 weeks	DSM-IV-TR (allowed)	Yes	Low
	Placebo (add-on therapy)	172	44.7 ± 12; 61	HDRS-17 = 25.1 ± 3.7 (severe)	6 (3.49)

Legend: DSM, Diagnostic and Statistical Manual of Mental Disorders; DOTES, Dosage Record and Treatment Emergent Symptom Scale; HDRS, Hamilton Depression Rating Scale; IDS, Inventory of Depressive Symptomatology; MADRS, Montgomery-Åsberg Depression Rating Scale; RDC, Research Diagnostic Criteria; RMS, Raskin Mania Scale; RoB, Risk of Bias; YMRS, Young Mania Rating Scale.

^aThis study included both patients with unipolar major depression and BD. While switch to mania refers to patients with BD, these data refer to the entire sample evaluated in the trial, including patients with unipolar major depression.

All studies included in the NMA were parallel-group RCTs and used diagnostic criteria from various versions of the DSM for the diagnosis of bipolar depression. Mean trial duration was 6.9 weeks (range 1.7–10 weeks). Three studies used antidepressant as monotherapy,^16,18,19 seven studies used antidepressants as add-on therapies,20, 21, 22, 23, 24, 25, 26 and three studies included both patients receiving antidepressants as monotherapy and as add-on therapy, reporting outcomes for the combined population without stratification by treatment regimen.^14,15,17 All studies included adult patients, except one that did not provide specific age ranges or other information regarding age.²¹ Six studies included outpatients,^{14,17,18,20,22,26} three included both inpatients and outpatients,^15,23,24 and the remaining four did not specify the study setting. All studies used oral medications. Eleven studies adopted the Hamilton Rating Scale for Depression (HRSD) to assess depression severity,^14,15,17, 18, 19, 20, 21, 22, 23, 24, 25 one study used the Montgomery-Asberg Depression Rating Scale (MADRS),¹⁶ and one study used the Inventory for Depressive Symptomatology (IDS).²⁶ Switch to mania was defined using standardised criteria or rating scales in eight studies: two used both DSM criteria and the Young Mania Rating Scale (YMRS),^14,24 one used both RDC criteria and the Raskin Mania Scale (RMS),¹⁸ one used DSM criteria alone,²⁵ three used the YMRS alone,^19,23,26 one used the Dosage Record and Treatment Emergent Symptom Scale (DOTES).²¹ The remaining five studies defined switch to mania based on clinical observation. All RCTs performed an intention to treat (ITT) analysis for switch to mania.

A total of 1362 people with bipolar depression were included in the NMA, of which 818 (60.1%) were female, 511 (37.5%) were male, and 33 (2.4%) were not disclosed. Of the total sample, 688 (50.5%) participants had BD-I, 183 (13.4%) had BD-II, and 491 (36.1%) did not have their BD type disclosed. Furthermore, 930 (68.3%) patients were taking antidepressants as add-on therapy, while 432 (31.7%) were on antidepressant monotherapy.

Overall, the risk of bias was rated as high in two studies,^15,21 with some concerns in four studies,^14,16,17,22 and low in the remaining eight. A summary of the risk of bias assessment is available in Appendix 3 pp 31. The evaluation of transitivity assumption is reported in Appendix 3 pp 32–34. The results of the main analysis are documented in Appendix 3 pp 35 and 36.

Fig. 2 shows the network of antidepressants for switch to mania (k = 13, n = 1362). The network considered 12 different antidepressants (agomelatine, amineptine, amitriptyline, bupropion, desipramine, fluoxetine, imipramine, moclobemide, paroxetine, sertraline, tranylcypromine, venlafaxine), and placebo.

Fig. 2.

Networks of eligible comparisons for the network meta-analyses assessing the risk of switch to mania after antidepressant treatment in patients with bipolar depression. Legend: Each node represents a specific antidepressant or placebo. The size of each node is proportional to the total number of participants randomised to that treatment across studies. Edges between nodes represent direct head-to-head comparisons from included randomised controlled trials. The thickness of each edge is proportional to the number of studies informing that comparison.

Compared to placebo, RRs ranged from 4.53 to 0.31. RRs above 1, listed in descending order, were observed for venlafaxine, imipramine, desipramine, moclobemide, sertraline, tranylcypromine, fluoxetine, agomelatine, bupropion, and paroxetine. Amineptine and amitriptyline had RRs below 1 (Fig. 3). Consistent P-scores are reported in Table 2. In head-to-head comparisons, no antidepressant demonstrated a significantly higher switch to mania rate than another (Table 3).

Fig. 3.

Forest plot for risk of switch to mania after antidepressant treatment in patients with bipolar depression using placebo as reference. Legend: The plot displays point estimates (dots) and 95% confidence intervals (horizontal lines) for the relative risk of switch to mania for each treatment versus placebo. A risk ratio (RR) above 1 suggests a higher risk compared to placebo; below 1 suggests a lower risk. The vertical line at RR = 1 indicates no difference in risk. Note: RRs reported in this forest plot may differ in direction from those in Table 3 (league table), as they were inverted here to consistently represent the comparison “antidepressant versus placebo” and facilitate interpretation.

Table 2.

P-score rankings of antidepressants for the risk of switch to mania in the treatment of bipolar depression.

Antidepressant	P-score
Amineptine	0.8386
Amitriptyline	0.7909
Placebo	0.6364
Bupropion	0.5769
Paroxetine	0.5742
Agomelatine	0.5557
Fluoxetine	0.5239
Tranylcypromine	0.4525
Sertraline	0.411
Moclobemide	0.3667
Desipramine	0.3585
Imipramine	0.2498
Venlafaxine	0.1649

Legend: P-scores range from 0 (worst) to 1 (best) and reflect the probability that each treatment ranks highest in terms of safety (i.e., lowest risk of switch to mania).

Table 3.

Switch to mania after antidepressant treatment in patients with bipolar depression.

Amineptine

−

0.33 [0.04; 2.73]

−

−

−

−

−

−

−

−

−

−

1.07 [0.02; 50.55]

Amitriptyline

−

−

0.27 [0.01; 6.34]

−

−

−

−

−

−

−

−

0.33 [0.04; 2.73]

0.31 [0.01; 7.82]

Placebo

−

0.88 [0.42; 1.84]

0.86 [0.29; 2.50]

0.80 [0.27; 2.36]

−

−

−

−

0.49 [0.13; 1.87]

−

0.29 [0.01; 8.81]

0.27 [0.00; 15.77]

0.87 [0.06; 12.82]

Bupropion

−

−

−

−

0.57 [0.11; 2.98]

−

0.44 [0.05; 3.85]

−

0.25 [0.06; 1.11]

0.29 [0.03; 2.74]

0.27 [0.01; 6.34]

0.88 [0.42; 1.85]

1.02 [0.08; 13.61]

Paroxetine

−

−

−

−

−

−

0.16 [0.01; 2.97]

0.25 [0.03; 2.11]

0.29 [0.03; 3.02]

0.27 [0.01; 7.97]

0.86 [0.29; 2.50]

0.99 [0.05; 17.97]

0.97 [0.26; 3.56]

Agomelatine

−

−

−

−

−

−

−

0.26 [0.02; 2.78]

0.24 [0.01; 7.30]

0.78 [0.27; 2.31]

0.91 [0.05; 16.48]

0.89 [0.24; 3.27]

0.92 [0.20; 4.18]

Fluoxetine

−

−

−

−

0.20 [0.01; 4.00]

−

0.20 [0.01; 3.25]

0.19 [0.00; 7.48]

0.60 [0.10; 3.74]

0.70 [0.03; 17.64]

0.68 [0.10; 4.71]

0.70 [0.09; 5.83]

0.77 [0.10; 6.00]

Tranylcypromine

−

−

−

0.60 [0.16; 2.27]

−

0.16 [0.01; 4.35]

0.15 [0.00; 7.97]

0.49 [0.04; 6.07]

0.57 [0.11; 2.98]

0.56 [0.05; 6.15]

0.57 [0.04; 8.81]

0.63 [0.04; 9.64]

0.82 [0.04; 17.75]

Sertraline

−

−

−

0.45 [0.15; 1.35]

0.16 [0.01; 2.24]

0.15 [0.00; 5.32]

0.47 [0.09; 2.39]

0.55 [0.02; 12.38]

0.54 [0.09; 3.05]

0.55 [0.08; 3.85]

0.60 [0.09; 3.94]

0.78 [0.14; 4.24]

0.96 [0.05; 18.63]

Moclobemide

−

0.77 [0.27; 2.17]

.

0.13 [0.00; 7.26]

0.12 [0.00; 11.90]

0.38 [0.01; 12.10]

0.44 [0.05; 3.85]

0.43 [0.01; 12.65]

0.44 [0.01; 16.59]

0.48 [0.01; 18.13]

0.63 [0.01; 30.86]

0.77 [0.05; 11.84]

0.80 [0.02; 36.05]

Desipramine

−

−

0.12 [0.01; 1.39]

0.11 [0.00; 3.50]

0.36 [0.10; 1.26]

0.42 [0.02; 7.94]

0.41 [0.10; 1.66]

0.42 [0.08; 2.18]

0.46 [0.10; 2.20]

0.60 [0.16; 2.27]

0.74 [0.05; 11.83]

0.77 [0.27; 2.17]

0.96 [0.02; 37.17]

Imipramine

−

0.07 [0.00; 1.61]

0.07 [0.00; 3.05]

0.22 [0.02; 2.11]

0.25 [0.06; 1.11]

0.25 [0.03; 2.11]

0.26 [0.02; 3.13]

0.28 [0.02; 3.42]

0.37 [0.02; 6.49]

0.45 [0.15; 1.35]

0.47 [0.03; 7.33]

0.58 [0.04; 8.06]

0.61 [0.05; 7.79]

Venlafaxine

Treatments are reported according to P-Score. Results from the pairwise meta-analysis are presented in the upper right half and results from the network meta-analysis in the left lower half. Comparisons between treatments should be read from left to right and the estimate is in the cell in common between the column-defining treatment and the row-defining treatment. In the left lower half, a risk ratio (RR) lower than 1 indicates a lower risk of switch to mania for the column-defining treatmentcompared to the row-defining treatment. In the upper right half, an RR lower than 1 indicates a lower risk of switch to mania for the row-defining treatment compared to the column-defining treatment.

Global heterogeneity was low, although some uncertainty is reflected in the 95% CI of the I² estimate (0% [0.0%–79.2%]). Between-study variance was null (tau² = 0). No signs of global inconsistency were detected, as indicated by a non-significant Q statistic (p = 0.74), nor were any signs of local inconsistency observed (node-splitting analysis reported in Appendix 3 p 37). No publication bias was detected (Appendix 3 p 38).

None of the sensitivity analyses altered the main findings (Appendix 3 pp 39–80).

The post-hoc analyses (both k = 13, n = 1362) grouped antidepressants according to their mechanisms of action and by NbN-3 (Appendix 3 pp 81–84). Compared to placebo, RRs ranged from 2.93 to 0.69 in the mechanism of action grouping. Serotonin–norepinephrine reuptake inhibitors (SNRIs) were associated with the highest RR (2.93 [95% CI 0.97–8.82]). In the NbN-3 grouping, RRs compared to placebo ranged from 3.09 to 0.33. Serotonin, noradrenaline – reuptake inhibitors were associated with the highest RR (3.09 [0.97–9.86]).

The post-hoc pairwise meta-analysis assessing the risk of switching to mania between antidepressants and placebo included six studies (the only RCTs with a direct comparison to placebo) investigating agomelatine, amineptine, fluoxetine, imipramine, and paroxetine. The pooled RR was 1.17 [0.40–3.40], with no evidence of heterogeneity (Q = 1.70, p = 0.89; I² = 0.0%). None of the studies showed a significant increase in the risk of switching to mania with antidepressants compared to placebo (Appendix 3 p 85).

The CINeMA assessment is detailed in Appendix 4 pp 86–91. Of the 78 treatment comparisons evaluated, 77 were rated as having low confidence and 1 as very low confidence.

Discussion

This study is the first systematic review and NMA examining the risk of switching to mania associated with acute antidepressant treatment for bipolar depression, focusing specifically on antidepressants. While some evidence of increased risk of switching to mania was observed across several agents—suggesting a possible, albeit non-significant, class effect—no individual antidepressants demonstrated a significantly higher rate of switch to mania compared to placebo and to each other in the acute treatment of bipolar depression. Similarly, when grouping antidepressants based on their mechanisms of action and on NbN-3, no increased risk of switching to mania was found for any of the considered groups.

These findings are consistent with previous research, including the largest NMA on acute pharmacotherapy for bipolar depression, which found a higher risk of switching to mania when antidepressants were analysed as a class, but did not identify any individual antidepressant associated with a significantly increased risk compared to placebo or other agents.⁸ However, by focusing specifically on antidepressants, our analysis provides more detailed insights into their safety profile, which is of high clinical relevance, and allowed for a more detailed exploration of clinical moderators. The consistency of our results with those of broader NMAs reinforces the robustness of the evidence and supports the reliability of our focused, clinically oriented findings.

While we did not detect statistically significant differences, some individual agents showed larger effect estimates that warrant further investigation. In particular, venlafaxine emerged as the antidepressant with the highest relative risk of switch to mania. Notably, whether grouped by classical pharmacological classification (as a SNRI) or by the NbN-3 system (as a serotonin–noradrenaline reuptake inhibitor), venlafaxine consistently showed the highest associated risk. It was studied as add-on therapy in two trials—one comparing it to paroxetine²⁴ and another evaluating it alongside bupropion and sertraline.²⁶ Both trials reported a significantly increased risk of mania in participants treated with venlafaxine compared to other antidepressants. This elevated risk may be explained by venlafaxine's pharmacological profile: by increasing noradrenaline levels, it may enhance arousal and energy, potentially destabilising mood and triggering manic episodes in vulnerable individuals.²⁷ A similar mechanism may apply to other drug classes, such as tricyclic antidepressants, which also increase noradrenaline levels, although with more complex and less selective pharmacological profiles. Notably, the second highest (though still non-significant) risk of switch was observed for imipramine, a tricyclic antidepressant.

Expert consensus and clinical guidelines recommend the use of antidepressants as a second-line option in bipolar depression, specifically as add-on therapy alongside mood stabilisers or second-generation antipsychotics.^6,28 Our findings support this recommendation. Although sensitivity analyses did not show differences from the main analysis, the monotherapy subgroup included fewer studies, offering less robust evidence compared to the larger and more consistent data set for add-on therapy.

Importantly, our study focuses on the acute treatment of bipolar depression, which raises further considerations. One question is whether the duration of the included RCTs is sufficient to detect switches to mania, as delayed or rare events may be underestimated in short-term trials.²⁹ According to the International Society for Bipolar Disorders (ISBD) nomenclature task force, a switch to mania after antidepressant treatment must occur at least two weeks after treatment initiation to be considered treatment-emergent.³⁰ In line with this, we conducted a sensitivity analysis excluding the study with a shorter duration,¹⁶ which did not alter the main findings. While this confirms that our data meet the minimum threshold to detect early switches, the longest studies included in our analysis lasted 10 weeks.^20,26 Thus, our results provide reassurance regarding the risk of switch to mania within the acute treatment window, but the long-term risk remains uncertain. Further long-term RCTs or observational studies are needed to evaluate the safety of continued antidepressant use beyond the acute phase.³¹ A recent target trial emulation using Danish nationwide health registers found no significant association between antidepressant use and the risk of mania over a 1-year follow-up period, even when stratifying by concurrent mood stabiliser use.³² This result support the notion that even the long-term risk of switch to mania may be lower than often assumed, although residual confounding and selection bias cannot be ruled out. Similarly, a recent long-term RCT conducted in patients with remitted bipolar depression found no benefit of maintaining antidepressant treatment beyond the short term for preventing relapse into any mood episode.³³ Taken together, these findings support a cautious but potentially safe use of antidepressants as short-term add-on therapy in bipolar depression, with timely discontinuation following remission to minimise the risk of long-term mood destabilisation—particularly in light of the limited evidence supporting their continued benefit.

Although concerns regarding the use of antidepressants remain—particularly given the observed signals of increased switch risk with certain agents such as venlafaxine and imipramine—the absence of significant differences across treatments suggests that the risk of switching to mania may be more strongly influenced by patient-related factors than by pharmacological properties alone. This reinforces the need to understand for whom these treatments may be appropriate and how they should be used,³⁴ in line with the principles of precision psychiatry.³⁵ Several clinical factors have been identified as increasing the risk of antidepressant-induced mania. Younger age and early onset of illness have consistently been associated with a higher likelihood of switching, particularly in patients with a history of rapid cycling or severe manic symptoms.^7,36 Comorbidities such as anxiety disorders, substance or alcohol use, and a history of suicidal behavior further elevate this risk.^7,36 Moreover, patients with BD-I are at a higher risk for switching to (hypo)mania compared to those with BD-II.^7,28 Additional risk factors include a prior history of antidepressant-induced switch, low response to previous antidepressant trials, and a history of rapid cycling.^34,36,37

To explore these potential confounding factors, we conducted sensitivity analyses based on baseline depression severity, treatment setting, and the presence of psychiatric comorbidities. These analyses did not yield results that differed from the main analysis. In this context, identifying reliable biomarkers for predicting switches to mania is a key research priority. Promising, though still preliminary, results have emerged from genetic studies, such as genome-wide association studies (GWAS), which have identified potential genetic markers linked to an increased risk of antidepressant-induced switch to mania.³⁸ Recent findings also suggest that polygenic risk scores (PRSs) for bipolar disorder, depression, and schizophrenia are associated with a tendency toward specific episode polarities in bipolar disorder, with higher PRSs for bipolar disorder and schizophrenia linked to manic episodes.³⁹

A major strength of this network meta-analysis lies in its exclusive focus on a single outcome of critical clinical relevance—switch to mania—which is often cited as a key concern when prescribing antidepressants for bipolar depression. By isolating this outcome and conducting a rigorous comparison across a range of antidepressant agents, our study provides valuable evidence to inform clinical decision-making. However, several limitations should be acknowledged. Although RCTs are considered the gold standard for evaluating treatment efficacy and safety, they often do not reflect real-world clinical settings due to strict inclusion criteria.⁴⁰ Participants in RCTs frequently present with less severe illness—typically without suicide risk, psychotic features, or significant psychiatric and physical comorbidities—factors that may influence the risk of switching to mania.⁴¹ Nevertheless, we performed multiple sensitivity analyses accounting for multiple confounding factors, none of which altered the main findings. Still, given the heterogeneity and complexity of bipolar disorder, residual confounding cannot be entirely ruled out, and may have affected both our results and those of previous studies. In particular, due to incomplete reporting in the original trials, we were unable to perform stratified analyses by bipolar subtype, which may limit the specificity of our findings. Despite these limitations, RCTs remain the only design capable of establishing causal inferences regarding the relationship between antidepressant exposure and switch to mania—unlike cross-sectional or case–control studies, which are not designed to address causality.⁴² Another important limitation relates to the heterogeneity in the definition of switch to mania across studies, ranging from standardised diagnostic criteria or rating scales to clinical observation.³⁵ Although our sensitivity analysis confirmed the robustness of results across definitions, subtle inconsistencies may have gone undetected. This variability also hindered the consistent assessment of mania versus hypomania and related phenomena such as mixed states, which were not distinguished or systematically reported across trials. To enhance comparability across trials, future research should employ standardised tools to define and measure switch to mania, hypomania, and mixed states. The overall confidence in the evidence was rated as low, limiting the strength of our conclusions. In some cases, small sample sizes may have reduced statistical power, affecting the precision of risk estimates and the width of confidence intervals.⁴³ Moreover, two included studies were assessed as having a high risk of bias,^15,21 but a sensitivity analysis excluding these studies yielded results consistent with the main analysis, reinforcing the robustness of our findings. Additionally, this NMA did not assess the safety of novel treatments—such as ketamine, esketamine, and psilocybin—as it focused specifically on antidepressants classified under the ATC system (N06A). To date, three RCTs44, 45, 46 —one for each of these agents—have evaluated their use in bipolar depression, but none reported any switches to mania in either the intervention or control arms. This absence of events precluded their inclusion in our network, as NMA relies on event data to estimate relative treatment effects. While preliminary findings suggest a favorable safety profile, larger studies are needed to better assess the risk of switch to mania and other potential adverse outcomes associated with these emerging treatments. Similarly, combination regimens involving antidepressants and other pharmacological classes—such as fluoxetine plus olanzapine or sertraline plus lithium—were excluded from this analysis to isolate the specific contribution of antidepressants to switch risk. However, the four RCTs that evaluated these combinations reported no significant differences in switch rates compared to control conditions, suggesting a favorable safety profile.47, 48, 49, 50 Finally, the results of our post-hoc analyses, including classifications based on mechanisms of action and the NbN-3 system, should be interpreted with caution as they were not based on predefined hypotheses.⁵¹ Future studies should explicitly investigate these pharmacological frameworks to confirm and refine our exploratory findings.

In conclusion, this systematic review and NMA has important implications for clinical practice, particularly in light of the limited number of FDA-approved treatments for the acute phase of bipolar depression. While no individual antidepressant was associated with a statistically significant increase in the risk of switch to mania, some evidence of increased risk was observed across several treatments, suggesting that a potential class-level effect cannot be entirely excluded. These findings support the cautious use of antidepressants as short-term, add-on therapy, particularly in combination with a mood stabiliser or a second-generation antipsychotic. Special attention may be warranted for specific agents, such as venlafaxine, which showed consistent signals of higher risk in individual studies.

Informed prescribing decisions should not be based solely on a diagnosis of BD, but should incorporate the specific agent, treatment regimen, patient history of manic switches, prior treatment response, and clinical profile. This individualised approach, consistent with precision psychiatry, may optimise the safe use of antidepressants while minimising the risk of mood destabilisation.

Contributors

VO, with input from JR and EV, conceptualised and planned the study. VO managed and coordinated the research activity planning and execution. VO, ELS, and SP screened the literature and extracted the data. VO and MDP accessed and verified the underlying data, and conducted the statistical analysis, under the supervision of JR. All authors interpreted the results. VO prepared the first draft of the manuscript, with important contributions from MDP and GF. All authors critically reviewed and commented on the manuscript. JR and EV had oversight and leadership responsibility for the research activity planning and execution. All authors had full access to all data in the study and had final responsibility for the decision to submit for publication. All authors read and approved the final version of the manuscript.

Data sharing statement

Requests to see any data that are not included in the Article or the appendix should be directed to the corresponding author.

Declaration of interests

VO, MDP, ELS, SP, GF, DHM, MF, and AY have no conflicts to declare. GA has received CME-related honoraria, or consulting fees from Abartis Pharma, Adamed, Angelini, Casen Recordati, Johnson & Johnson, Lundbeck, Lundbeck/Otsuka, Rovi, and Viatris, with no financial or other relationship relevant to the subject of this article. AM has received honoraria/has been a consultant for Idorsia and Angelini. MS has received honoraria/has been a consultant for AbbVie, Angelini, Bausch Health, Boehringer Ingelheim, Lundbeck, Otsuka, Teva. MP wishes to disclose that in the last three years, he has received honoraria for lectures and/or educational activities, or institutional support for clinical trial activities, from Angelini Pharma, Janssen, Lundbeck, Merck Sharp and Dohme (MSD), Otsuka, Rovi, Pfizer Inc, Fidia, Viatris, Recordati, Newron, Neopharmed Gentili, and Teva; all disclosures are unrelated to the present work. MS received honoraria/has been a consultant for Angelini, AbbVie, Boehringer Ingelheim, Lundbeck, Otsuka. In the last three years SL has received honoraria for advising/consulting and/or for lectures and/or for educational material from Angelini, Apsen, Boehringer Ingelheim, Janssen, Karuna, Kynexis, Lundbeck, Medscape, Otsuka, Neurotorium, NovoNordisk, Orionpharma, Roche, Rovi, TEVA. EV has received grants, consulting fees, advisory roles, and/or honoraria for lectures or educational activities from the following entities, outside the submitted work: AB-Biotics, AbbVie, Abbott, Adamed, Angelini, Beckley-Psytech, Biogen, Biohaven, Boehringer-Ingelheim, Celon Pharma, Compass, Dainippon Sumitomo Pharma, Ferrer, Gedeon Richter, GH Research, GlaxoSmithKline, Idorsia, Janssen, Lundbeck, Merck, Mitsubishi, Newron, Novartis, Orion Corporation, Organon, Otsuka, Roche, Rovi, Sage, Sanofi-Aventis, Sunovion, Takeda, Tanabe Pharma, Teva, Viatris, and HMNC. JR has received CME-related honoraria from Adamed, outside the submitted work.