Published on 15 September 2026
Bipolar depression: seven drugs beat placebo, and no ranking between them holds
In brief
Seven drugs do better than placebo on the depressive symptoms of bipolar disorder, with what the authors call good confidence in the evidence: olanzapine plus fluoxetine, quetiapine, olanzapine, lurasidone, lumateperone, cariprazine, lamotrigine. The paper published in 2026 is not a new network meta-analysis. It carries forward the one the same group published in 2023, built on 145 studies in qualitative synthesis and 101 in quantitative analysis, and adds a complementary systematic review covering April 2023 to November 2024. The 2026 abstract contains no effect size, no odds ratio, no confidence interval and no comparative tolerability figure. The magnitudes are published elsewhere: the abstract of the 2023 network meta-analysis gives standardised mean differences running from 0.41 (95% CI 0.19 to 0.64) for olanzapine plus fluoxetine to 0.16 (0.03 to 0.29) for lamotrigine. The effect is therefore small to moderate, and the lower bound for lamotrigine, 0.03, sits close to no effect. The fine ranking between the seven appears in neither abstract, and the tables of weight gain and sedation that circulate about this work come from the full text. One last point, decisive at the prescription pad, which does not come from the study but from a check run label by label for this analysis: of these seven drugs, one only, quetiapine, carries a marketing authorisation for the depressive episode of bipolar disorder in the European Union. The detail, drug by drug and source by source, appears further down.
The context
The work of Judd and colleagues on the natural history of bipolar disorder established that time spent with depressive symptoms far exceeds time spent on the manic or hypomanic side, in type I as in type II. Yet this is the phase for which the validated armamentarium is thinnest, and the one on which international guidelines diverge most, particularly on the place of antidepressants. In the consulting room the question is entirely concrete: faced with a patient with bipolar depression, what do you base the choice on when several drugs have been tested but rarely against one another?
The network meta-analysis answers part of that impasse. It combines direct comparisons, which are rare, with indirect comparisons reconstructed through placebo, which are many. What it produces is a ranking, not a demonstration of superiority between two treatments. The distinction changes everything at the moment of prescribing.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Adults in an acute depressive episode of bipolar disorder, type I, type II or not otherwise specified. The network meta-analysis carried forward covers 20,081 participants, 8,063 men (41.7%) and 11,263 women (58.3%), mean age 41.0 years. The split between type I, type II and not otherwise specified is published in neither abstract | |
| Intervention | |
| Pharmacological treatments for acute bipolar depression, with no restriction of class, as monotherapy, in combination or as add-on | |
| Comparator | |
| Placebo, and indirect comparisons between agents across a network of 68 medications plus placebo. Trials in patients with substance misuse, with unipolar depression or with schizophrenia are excluded | |
| Primary outcome | |
| Two co-primary outcomes: depressive symptoms, examined with standardised mean differences, and manic switch, examined with odds ratios. Four families of secondary outcomes: dropouts due to any reason, due to inefficacy, due to adverse events, and important side-effects. Neither functioning nor recurrence is among the outcomes retained | |
| Design | |
| A network meta-analysis carried forward (145 studies in qualitative synthesis, 101 in quantitative analysis, from inception to April 2023, trials lasting 2 to 16 weeks with masked outcome assessment) plus a complementary systematic review from April 2023 to November 2024 across nine databases, with narrative synthesis and risk of bias assessed with RoB 2, the revised Cochrane tool. PROSPERO registration CRD42020171726 · CEBM 1a |
A word on what this paper actually is. The authors write that they set out to revisit the results of a recent network meta-analysis and to update them. The 101 quantitative studies are therefore not a new corpus: they are those of the meta-analysis published in The Lancet Psychiatry in 2023, by four of the same five authors. Reading this text as an original synthesis would mean counting the same evidence twice.
Quality control
| Criterion | Status |
|---|---|
| Protocol registration | Sound |
| FindingPROSPERO CRD42020171726, cited in the published abstract | |
| Search strategy | Sound |
| FindingNine sources searched for the update: MEDLINE, EMBASE, PsycINFO, CINAHL, LILACS, Cochrane, Web of Science, OVID and Google Scholar | |
| Risk of bias of the studies | Sound |
| FindingRoB 2, the revised Cochrane risk of bias tool, applied to the studies identified by the complementary review | |
| Nature of the work | Reservation |
| FindingA meta-analysis already published in 2023 carried forward, not a new analysis. The risk is editorial rather than methodological: that of counting the same evidence twice | |
| Quantitative data available | Reservation |
| FindingThe 2026 abstract gives no effect size, no odds ratio, no confidence interval and no tolerability figure: nothing in it is verifiable beyond the list of the seven drugs. The magnitudes are, on the other hand, published in the abstract of the 2023 meta-analysis, from 0.41 (95% CI 0.19 to 0.64) to 0.16 (0.03 to 0.29) | |
| Yield of the complementary review | Insufficient |
| FindingThe 2026 abstract counts the trials the complementary review identified and says how many of them have usable published results, but it reports no finding from them: no estimate, no revised ranking, nothing on what those trials change in the list of seven | |
| Tool for confidence in the evidence | Reservation |
| FindingThe tool is identified: the abstract of the meta-analysis carried forward names CINeMA. What is not detailed is its application domain by domain | |
| Funding | Sound |
| FindingUnited States National Institute of Mental Health, intramural programme, grant ZIA MH002927 according to the record. No industry funding identified | |
| Competing interests | Reservation |
| FindingThe record does carry a declaration. Two of the five authors declare no competing interest; the three others report industry ties, from research work carried out with several companies before joining a public institute, for the first author, to consulting, advisory or lecture honoraria from long lists of companies, for the two senior authors. The point deserves the reader’s attention in a field with a strong industry presence | |
The findings
| Result | What the published abstract says |
|---|---|
| Efficacy against placebo | Seven drugs superior to placebo with good confidence |
| PEB readingDemonstrated no magnitude in the 2026 abstract, but the standardised mean differences appear in the abstract of the 2023 meta-analysis: from 0.41 (95% CI 0.19 to 0.64) for olanzapine plus fluoxetine to 0.16 (0.03 to 0.29) for lamotrigine. The effect is small to moderate, and the lower bound for lamotrigine, 0.03, sits close to no effect | |
| Other drugs | Several others might also be efficacious, but confidence in the evidence is very low to low |
| PEB readingUncertainty, not inefficacy low confidence reflects missing data, it demonstrates no absence of effect | |
| Ranking among the seven | No numerical order in the 2026 abstract |
| PEB readingNot verifiable here the bounds published in 2023, 0.41 and 0.16, frame the set without establishing a rank drug by drug, and no ranking statistic appears in either abstract. The fine hierarchies in circulation come from the full text and rest largely on indirect comparisons | |
| Comparative tolerability | No figure on weight, on sedation or on manic switch in the 2026 abstract |
| PEB readingNot verifiable here the 2023 source meta-analysis concludes that the drugs differ in their side-effect profiles, without this abstract giving the measure of it | |
| Complementary review, 2023 to 2024 | The abstract reports the yield of the complementary review, 24 trials identified, seven with published results suitable for meta-analysis and 17 either ongoing or completed with no available results, but no effect size from those seven |
| PEB readingNot concluded so it cannot be said whether the literature published since April 2023 confirms, widens or modifies the list of seven | |
The size of the corpus is established and sourceable: the abstract of the 2023 meta-analysis reports 101 randomised controlled trials and 20,081 participants. Other figures attributed to this work circulate with no verifiable support. A weight gain of 3.2 kg on olanzapine plus fluoxetine against 0.2 kg on lurasidone, sedation multiplied by 6.9 on quetiapine, and above all the claim that quetiapine would be the only drug to reduce manic switch. None of these three values appears in the published abstracts. PEB does not carry them. The last is in addition the weakest as reasoning: reducing switch below placebo would presuppose an active protective effect, a claim that requires solid safety data and a published confidence interval, not a table row.
What can be retained without error is more modest. The list of seven is the 2023 list, and the 2026 paper carries it forward unchanged. But the abstract reports no finding from the complementary review covering the following nineteen months: it counts the trials, it does not say what they show. So it cannot be written that the list withstood that additional literature, nor that it comes out of it modified. The point stays open, and the full text will be needed to settle it.
Critical appraisal
| Domain | Judgement |
|---|---|
| Question and protocol | Sound |
| FindingA clear clinical question, prior registration, a declared method | |
| Completeness of the search | Sound |
| FindingNine sources searched, an explicit time window | |
| Transitivity of the network | Reservation |
| FindingThe frame is partly set: 68 medications plus placebo, monotherapy, combination and add-on trials, exclusion of substance misuse, of unipolar depression and of schizophrenia. The fine comparability of populations across trials is not documented for all that. Yet a network meta-analysis holds only if the patients could have been randomised indifferently to any arm | |
| Coherence of the network | Reservation |
| FindingNo inconsistency test result in the abstract. The agreement between direct and indirect evidence is therefore not verifiable at this level of reading | |
| Precision of the estimates | Reservation |
| FindingNo confidence interval in the 2026 abstract. Those of the 2023 meta-analysis are published and frame the effect, from 0.41 (95% CI 0.19 to 0.64) to 0.16 (0.03 to 0.29). The magnitude is small to moderate, and the lower bound for lamotrigine, 0.03, sits close to no effect | |
| Publication bias | Reservation |
| FindingThe abstract reports no test for publication bias. It does state that part of the trials identified by the update are still ongoing or completed with no available results, which is precisely the mechanism by which negative trials stay in the drawer | |
| Outcomes measured | Reservation |
| FindingThe two co-primary outcomes, depressive symptoms and manic switch, are relevant but remain short term, on trials of two to sixteen weeks. Neither functioning, nor recurrence, nor the metabolic trajectory at medium term is among the outcomes retained | |
| External validity and regulatory status | Reservation |
| FindingThe ranking is international, prescribing is not. Of the seven drugs, one only, quetiapine, carries an authorisation for this indication in the European Union. Four are authorised for another indication: olanzapine, lamotrigine, lurasidone and cariprazine. Two have no European authorisation at all, olanzapine plus fluoxetine and lumateperone. Statuses checked on 10 August 2026 against the summaries of product characteristics and the European public assessment reports, and the indication wording re-read in English-language labelling on 15 September 2026 | |
| Independence | Reservation |
| FindingIntramural public funding, which is a real strength. The declaration of competing interests, which the record carries, reports industry ties for three of the five authors and none for the two others, a point that deserves the reader’s attention in a field with a strong industry presence | |
Three faulty readings should be set aside at once. Superior to placebo does not mean superior to the others: with 68 treatments for 101 trials, the network rests mainly on comparisons anchored to placebo, not on direct confrontations between drugs. Very low confidence attached to a drug does not disqualify it, it says that nobody has tested it properly. And a tolerability profile measured over trials of two to sixteen weeks, in selected patients, tells you nothing about what a weight or a blood glucose becomes after three years, which is the real horizon of bipolar disorder. A fourth reading is just as wrong and more common: none of this transfers to unipolar major depression, which the included trials excluded by design.
Level of evidence
PEB appraisal: moderate to good confidence on the CINeMA scale that an effect exists for the seven drugs named, low confidence on any ranking between them, none at all on the tolerability figures until they are read in the full text. The source is in fact discordant with itself on this point: the Results section of the meta-analysis carried forward states moderate confidence, its Interpretation section good confidence, and the 2026 paper carries the second wording forward. The gap is not trivial when the grading serves as the main argument. The methodological base remains first rate, prior registration included. What holds the scientific score at 82, three points below the full Premium threshold, is not the unavailability of the data, since the magnitudes are accessible in the 2023 abstract. It is the nature of the work: a corpus already published and carried forward, a hierarchy built mainly on indirect comparisons anchored to placebo, and a short observation horizon, two to sixteen weeks, far below the real duration of the illness.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Seven drugs that beat placebo, fine, but it is the same meta-analysis as in 2023 and the abstract says nothing about what the update found. The fine ranking between the seven, I distrust it: these are indirect comparisons, on trials of two to sixteen weeks. And the effect is modest, 0.41 at best, 0.16 for lamotrigine. In practice I still choose on the side-effect profile and on what I am allowed to prescribe, and no network will tell me that. ”
What this means in practice: keep the list as a basis for discussion, refuse the ranking as a prescribing guide, and bring the decision back down to where it is really made, the patient’s profile and the regulatory frame that applies where you practise.
What you can do with this on Monday morning. The prescription does not get rewritten, but four points become immediately usable in the consulting room.
- Put the list of seven forward as a frame for discussion with the patient, separating clearly what belongs to efficacy demonstrated against placebo from what belongs to a ranking built on indirect comparisons.
- Establish the regulatory status of each drug before you talk about it. The table below sets out, drug by drug, the wording of the indication as it appears in the summary of product characteristics or in the European public assessment report, consulted on 10 August 2026. Slow titration of lamotrigine remains mandatory because of the risk of severe cutaneous reactions.
- Write in the notes what justifies a prescription outside the authorised indication when you decide on one, and explain it to the patient before starting it, not at the first renewal.
- Make weight and metabolic monitoring an explicit criterion of choice from the first prescription, since that is precisely where these drugs part company, rather than a catch-up at six months.
| Drug | Status in the depressive episode of bipolar disorder | Wording of the authorised indication, source |
|---|---|---|
| Quetiapine | Indicated | “For the treatment of major depressive episodes associated with bipolar disorder”, section 4.1 of the summary of product characteristics of a prolonged-release quetiapine medicine. |
| Olanzapine | Not indicated | Bipolar indication limited to the moderate to severe manic episode and, in patients whose manic episode has responded to olanzapine, to the prevention of recurrence, European public assessment report. The depressive episode does not appear in it. |
| Lamotrigine | Prevention only | “Prevention of depressive episodes in patients with bipolar I disorder who experience predominantly depressive episodes”, section 4.1 of the summary of product characteristics, which states that the medicine is “not indicated for the acute treatment of manic or depressive episodes”. |
| Lurasidone | Not indicated | European indication limited to the treatment of schizophrenia in adults, European public assessment report. |
| Cariprazine | Not indicated | European indication limited to “the treatment of schizophrenia in adult patients”, European public assessment report. |
| Lumateperone | No European authorisation | No medicine containing this substance in the register of the European Medicines Agency. The known authorisations are American. |
| Olanzapine plus fluoxetine | No European authorisation | The fixed combination is not authorised in the European Union. The European public assessment report for olanzapine contains no combination with fluoxetine. |
One green cell out of seven. That is the exact sense of the headline: of the seven drugs this work places ahead of placebo, one only, quetiapine, carries an authorisation for this precise indication in the European Union. Four do hold an authorisation, but for another indication. Two do not exist in the European circuit at all. This regulatory reading does not come from the study analysed, which does not deal with marketing authorisations: it was established by opening each summary of product characteristics and each European public assessment report, listed in the bibliography. Authorisation status is decided jurisdiction by jurisdiction, so check the labelling in force where you practise before you rely on any line of this table.
Two useful points. In a patient already taking valproate, starting lamotrigine calls for a reduced dose and an even slower titration, the interaction raising lamotrigine concentrations markedly. And valproate carries a contraindication that no prescription in bipolar disorder can leave aside: because of an established teratogenic and neurodevelopmental risk, it must not be used in girls, in female adolescents, in women of childbearing potential or in pregnancy, except where the conditions of the pregnancy prevention programme set out in the labelling in force where you practise are met. Valproate is not among the drugs named in the published abstract of this work, so PEB takes no position here on its place in bipolar depression.
Frequently asked questions
Is this a new network meta-analysis?
No. The authors revisit a network meta-analysis published in 2023 in The Lancet Psychiatry, whose 101 quantitative studies they carry forward, and they add a complementary systematic review covering April 2023 to November 2024. What is new is the update, not the network.
Can the seven drugs be ranked against one another?
Not from the published abstracts. No numerical value in the 2026 abstract, the magnitudes appearing in the abstract of the 2023 meta-analysis: from 0.41 (95% CI 0.19 to 0.64) for olanzapine plus fluoxetine to 0.16 (0.03 to 0.29) for lamotrigine. Those two bounds frame the set, they establish no rank drug by drug. Above all they indicate a small to moderate effect, whose lower bound, 0.03 for lamotrigine, sits close to no effect. The weight or sedation hierarchies attributed to this work come from the full text and have not been verified here. Even verified, they would rest on mainly indirect comparisons.
Does this work validate antidepressants in bipolar depression?
They are not among the seven drugs named with good confidence. The abstract states that several other treatments might be efficacious with very low to low confidence, without naming them. That wording is not a demonstration of inefficacy, it is a statement of insufficient data. The meta-analysis carried forward does go a little further: it concludes that antidepressants as a class seem to be efficacious, but had a higher risk for manic switch compared to antipsychotics. The caveat is substantial, a class result does not hold drug by drug, and antidepressants are a heterogeneous set. On manic switch itself, the context reference is the network meta-analysis by Oliva and colleagues (2025), cited in the bibliography.
What is actually licensed for this indication?
Only quetiapine carries a marketing authorisation in the depressive episode of bipolar disorder in the European Union, the exact wording of its section 4.1 being “for the treatment of major depressive episodes associated with bipolar disorder”. Lamotrigine is indicated for the prevention of depressive episodes in bipolar I disorder in patients who experience predominantly depressive episodes, and its labelling explicitly rules out treatment of the acute episode. Olanzapine is authorised in the manic episode and in the prevention of recurrence, not in the depressive episode. Lurasidone and cariprazine hold a European authorisation in schizophrenia only. Olanzapine plus fluoxetine and lumateperone have no European authorisation. The detailed statuses, with their sources, appear in the regulatory table above, and they are decided jurisdiction by jurisdiction: the labelling in force where you practise is what counts. Any other prescription in this indication is therefore use outside the authorised indication, which is neither forbidden nor trivial: it engages the prescriber’s responsibility and must be argued in the notes.
Does public funding settle the question of independence?
No, these are two different questions. The funding here is institutional and not industrial, which is a favourable point. It does not remove the need to read the declaration of competing interests, which the record carries: two of the five authors declare none, the three others report industry ties, from research work carried out with several companies to consulting, advisory or lecture honoraria from long lists of companies. The point deserves the reader’s attention in a field with a strong industry presence.
Annotated bibliography
Yalin N, Yildiz A, Siafis S, Vieta E, Leucht S (2026). Treatment of bipolar depression: results from a comprehensive network meta-analysis and updated systematic review. European Neuropsychopharmacology, 108, 112818. DOI 10.1016/j.euroneuro.2026.112818 · PMID 41855620. Source study analysed here. PROSPERO registration CRD42020171726. Funded by the United States National Institute of Mental Health, intramural programme, grant ZIA MH002927 according to the record. Main limitation: a published abstract with no quantitative data of its own and silent on what the complementary review found.
Yildiz A, Siafis S, Mavridis D, Vieta E, Leucht S (2023). Comparative efficacy and tolerability of pharmacological interventions for acute bipolar depression in adults: a systematic review and network meta-analysis. The Lancet Psychiatry, 10(9), 693-705. DOI 10.1016/S2215-0366(23)00199-2 · PMID 37595997. The network meta-analysis whose 101 quantitative studies the 2026 paper carries forward, that is 20,081 participants and 68 medications plus placebo. This is where the effect sizes are, standardised mean differences from 0.41 (95% CI 0.19 to 0.64) to 0.16 (0.03 to 0.29), together with the comparative tolerability profiles absent from the 2026 abstract. Limitation: the structure of the network rests largely on indirect comparisons through placebo.
Oliva V, De Prisco M, La Spina E, Paolucci S, Fico G, Anmella G, et al. (2025). Switch to mania after acute antidepressant treatment for bipolar depression: a systematic review and network meta-analysis of randomised controlled trials. eClinicalMedicine, 87, 103413. DOI 10.1016/j.eclinm.2025.103413 · PMID 40823496. Context reference on manic switch under antidepressant, a question the abstract of the article analysed does not address. Transparency note: only its metadata have been verified, none of its numerical results is carried here.
Regulatory sources for the table of indications, all consulted on 10 August 2026. The indication wording quoted here is that of English-language summaries of product characteristics read on 15 September 2026; those documents are national, and their wording can differ from one country to another. Summary of product characteristics, section 4.1, of a prolonged-release quetiapine medicine. Summary of product characteristics, section 4.1, of a lamotrigine medicine. European Medicines Agency, European public assessment reports for olanzapine, lurasidone and cariprazine. Register of the European Medicines Agency, search on the active substance lumateperone, no result. These documents do not come from the study analysed: they were opened one by one to establish the status of each drug, the source study not dealing with marketing authorisations. Authorisation wording differs from one jurisdiction to another, and the labelling in force where the reader practises is the one that applies.