Skip to content

Bipolar depression

TL;DR — Bipolar depression is not safely treated as ordinary major depressive disorder: treatment must balance depressive benefit against activation, switch and longer-term polarity effects. The largest recent network meta-analysis found moderate-confidence benefit for olanzapine–fluoxetine, quetiapine, olanzapine, lurasidone, lumateperone, cariprazine and lamotrigine, but mean effects were small to moderate and adverse-effect profiles differed (Yildiz 2023, PMID 37595997). Direct trials give effect sizes around 0.5 for lurasidone, quetiapine and lumateperone, while lithium and paroxetine failed primary acute endpoints in large comparator trials (Loebel 2014, PMID 24170180; Thase 2006, PMID 17110817; Calabrese 2021, PMID 34551584; Young 2010, PMID 20122369; McElroy 2010, PMID 20122366). Lamotrigine presents a classic evidence tension: four of five individual trials missed their primary endpoint, yet patient-level pooling found a modest response advantage concentrated in more severe depression (Calabrese 2008, PMID 18271912; Geddes 2009, PMID 19118318). Evidence for light therapy, modafinil, pramipexole and ketamine is promising but narrower, smaller or partly indirect.

Why this phase needs separate treatment logic

The central trial problem is two-dimensional: reduce depression without causing hypomania, mania, mixed activation or cycle destabilization. Acute studies usually last six to eight weeks, whereas switch, relapse and functional recovery can evolve later. A 101-RCT network meta-analysis covering 20,081 participants used depressive symptoms and manic switch as co-primary outcomes, making this trade-off explicit (Yildiz 2023, PMID 37595997).

Diagnostic confidence matters because a depressive episode alone does not reveal polarity. The diagnosis and bipolar spectrum page covers history and mimics; this page assumes a bipolar-spectrum diagnosis and asks what changes depressive outcomes.

The treatment target is not simply absence of sadness. Residual sleep disruption, anhedonia, cognition and role impairment can persist after scale-defined response, while emerging activation can make a falling depression score falsely reassuring.

Comparative efficacy map

Treatment Direct or pooled depressive result Switch / tolerability signal Evidence boundary
Olanzapine–fluoxetine (OFC) Network SMD 0.41 (95% CI 0.19–0.64); pivotal remission 48.8% vs 24.5% placebo Treatment-emergent mania 6.4% vs 6.7% placebo; weight/metabolic burden Specific fixed combination, not proof for antidepressant monotherapy (Yildiz 2023, PMID 37595997; Tohen 2003, PMID 14609883)
Quetiapine 300 mg BOLDER II effect size 0.61; EMBOLDEN I MADRS −15.4 vs −11.8 placebo Sedation, somnolence, dry mouth; switch lower than paroxetine/placebo in EMBOLDEN II 600 mg did not show a major efficacy advantage (Thase 2006, PMID 17110817; Young 2010, PMID 20122369; McElroy 2010, PMID 20122366)
Lurasidone 20–60 mg MADRS −15.4 vs −10.7, ES 0.51; replication ES 0.33 Akathisia/nausea; minimal mean weight, lipid and glycemic change Higher dose did not outperform lower dose and missed significance in one replication (Loebel 2014, PMID 24170180; Kato 2020, PMID 32827348)
Lumateperone 42 mg MADRS LS mean difference −4.6; ES 0.56 Somnolence/nausea; low EPS and minimal mean metabolic change One six-week phase III program; longer-term comparative data needed (Calabrese 2021, PMID 34551584)
Cariprazine 1.5–3 mg Placebo-adjusted MADRS −2.5 (95% CI −4.6 to −0.4) and −3.0 (−5.1 to −0.9) Nausea, akathisia, dizziness, sedation; small mean metabolic changes CGI-S differences missed multiplicity-adjusted significance (Earley 2019, PMID 30845817)
Olanzapine MADRS −15.0 vs −11.9 placebo; remission 32.8% vs 24.5% Weight/metabolic burden Less effective than OFC in pivotal trial (Tohen 2003, PMID 14609883)
Lamotrigine IPD response RR 1.27 HRSD and 1.22 MADRS Generally well tolerated; slow titration constrains acute use Four of five primary trial endpoints negative; pooled benefit modest (Geddes 2009, PMID 19118318; Calabrese 2008, PMID 18271912)
Lithium EMBOLDEN I MADRS −13.6 vs −11.8, P=.123 Nausea in acute trial; monitoring burden Negative acute endpoint does not negate maintenance efficacy (Young 2010, PMID 20122369)
Paroxetine EMBOLDEN II MADRS −13.76 vs −12.60, P=.313 Mania/hypomania higher than quetiapine in this trial No monotherapy efficacy signal (McElroy 2010, PMID 20122366)

The network estimate should be read as a map, not a rigid league table. Moderate confidence supported seven interventions, with SMDs from 0.41 for OFC to 0.16 for lamotrigine; several additional drugs had only low or very-low-confidence evidence (Yildiz 2023, PMID 37595997). Small mean differences can still contain meaningful responders, but they also make adverse-effect and patient-preference differences proportionally more important.

Direct trial anchors

Quetiapine

BOLDER II randomized 509 people with bipolar I or II depression. Both 300 and 600 mg separated from placebo from week 1 through week 8; effect sizes were 0.61 and 0.54, with no major dose advantage and common dry mouth, sedation, somnolence, dizziness and constipation (Thase 2006, PMID 17110817).

EMBOLDEN I provides an active-comparator caution. Among 802 participants, week-8 MADRS change was −15.4 with quetiapine 300 mg, −16.1 with 600 mg, −13.6 with lithium and −11.8 with placebo; both quetiapine doses met P<.001, whereas lithium gave P=.123 (Young 2010, PMID 20122369). EMBOLDEN II similarly found both quetiapine doses effective, while paroxetine 20 mg did not separate from placebo (P=.313) (McElroy 2010, PMID 20122366).

Lurasidone

The pivotal monotherapy trial randomized 505 participants for six weeks. Both dose ranges reduced MADRS by 15.4 points versus 10.7 on placebo, effect size 0.51; adverse-event discontinuation was 6.6%, 5.9% and 6.5%, respectively (Loebel 2014, PMID 24170180).

The multinational replication was informative about dose. The 20–60 mg group achieved MADRS −13.6 versus −10.6, adjusted P=.007 and ES 0.33, while 80–120 mg gave −12.6, adjusted P=.057 and ES 0.22 (Kato 2020, PMID 32827348). A pooled post-hoc analysis found effect sizes 0.41–0.47 in non-rapid cycling but 0.02–0.04 in 85 rapid-cycling participants, where placebo improved strongly; this small subgroup should generate, not settle, a moderator hypothesis (Kato 2023, PMID 37245552).

Lumateperone and cariprazine

Lumateperone 42 mg improved MADRS by 4.6 points more than placebo, effect size 0.56, in 377 participants; benefit appeared in bipolar I and II subgroups, with little mean metabolic or EPS change over six weeks (Calabrese 2021, PMID 34551584).

Cariprazine produced smaller placebo-adjusted mean differences, −2.5 and −3.0 MADRS points, but both confidence intervals excluded zero. CGI-S did not survive multiplicity adjustment, reminding readers that statistical success can depend on endpoint hierarchy (Earley 2019, PMID 30845817).

Olanzapine–fluoxetine

In 833 adults with bipolar I depression, week-8 MADRS change was −11.9 placebo, −15.0 olanzapine and −18.5 OFC; remission was 24.5%, 32.8% and 48.8%, respectively (Tohen 2003, PMID 14609883). A treatment-evaluation synthesis converted the placebo comparison to NNT 4 (95% CI 3–8) for response and 5 (3–8) for remission, but also NNH 6 (4–10) for at least 7% weight gain (Citrome 2011, PMID 22035291).

OFC outcome Benefit / harm estimate Meaning
Response NNT 4 (95% CI 3–8) About one additional responder per four treated vs placebo
Remission NNT 5 (3–8) About one additional remitter per five treated
Weight gain adverse event NNH 7 (5–16) Harm frequency is of similar order to benefit
At least 7% weight gain NNH 6 (4–10) Metabolic monitoring is not optional

The lamotrigine paradox

Lamotrigine’s acute evidence changes depending on the level of analysis. Four of five randomized trials did not meet their primary endpoint, and only one showed separation on several secondary measures (Calabrese 2008, PMID 18271912). Yet individual-patient-data pooling of all five trials found HRSD response RR 1.27 (95% CI 1.09–1.47) and MADRS response RR 1.22 (1.06–1.41) (Geddes 2009, PMID 19118318).

Severity appeared to modify the pooled result: baseline HRSD above 24 gave RR 1.47 (1.16–1.87), while HRSD 24 or lower gave RR 1.07 (0.90–1.27) (Geddes 2009, PMID 19118318). Because the interaction P value was .04 and retrospective subgrouping can overstate modifiers, severity-guided selection remains plausible rather than validated.

Lamotrigine also requires titration to reduce serious rash risk, making it poorly suited to situations requiring immediate symptom relief. Its strongest role is often continuation and depressive-relapse prevention, discussed in maintenance.

Smaller and adjunctive strategies

Strategy Quantitative signal Why it is not a default first-line conclusion
Adjunctive modafinil Response 44% vs 23%; remission 39% vs 18%; mania/hypomania 6 vs 5 participants n=85, six weeks, inadequately responsive population (Frye 2007, PMID 17671288)
Adjunctive pramipexole Response 60% vs 9%, P=.02 n=21 bipolar II proof-of-concept; one hypomanic participant on drug (Zarate 2004, PMID 15219473)
Adjunctive bright-light therapy Depression SMD 0.43 (95% CI 0.04–0.82); response OR 2.32 (1.12–4.81) Seven trials, n=259; heterogeneous timing/dose and short follow-up (Lam 2020, PMID 31826657)
Ketamine / esketamine Mixed-depression meta-analysis: response RR 2.14, remission RR 1.64, severity d −0.63 Evidence combined unipolar and bipolar depression; not a bipolar-specific pooled estimate (Bahji 2022, PMID 35231204)

The ketamine estimate is deliberately labeled indirect. A mixed diagnostic meta-analysis cannot establish bipolar-specific switch risk, durability or comparative benefit even when bipolar participants were eligible (Bahji 2022, PMID 35231204).

Mixed features and rapid cycling

Depression with mixed features changes the risk-benefit calculation. The updated systematic review found short-term evidence for selected atypical antipsychotics, but most data were post-hoc and antidepressant monotherapy had not been adequately evaluated (Xiao 2025, PMID 40808264). The mania and mixed states page details this evidence boundary.

For lurasidone, rapid-cycling subgroup estimates were near zero despite ordinary-range effects in non-rapid-cycling participants; only 85 participants were rapid cycling and placebo response was high (Kato 2023, PMID 37245552). This is insufficient to declare lurasidone ineffective in rapid cycling, but sufficient to avoid assuming uniform effect.

From symptom response to recovery

Outcome What it contributes Common blind spot
MADRS/HDRS change Sensitive measure of depressive symptoms A few-point difference may not restore function
Response Relative symptom improvement Baseline-dependent; residual symptoms can remain substantial
Remission Low endpoint symptom burden Definitions and handling of missing data vary
Affective switch Immediate polarity safety Short trials miss later cycle acceleration
Function / quality of life Patient-important recovery Less consistently reported than symptom scales
Weight/metabolic change Treatment burden Six-to-eight-week means understate cumulative risk

Lurasidone’s trial program included functional measures: path analysis in 485 participants found an indirect functional effect through depression improvement of −0.23 and a direct standardized effect of −0.17, P=.04 (Rajagopalan 2016, PMID 26928122). Such analyses are useful but remain secondary and model-dependent.

Evidence limitations

  • Acute trials are usually six to eight weeks; they cannot settle maintenance benefit, late switching or cycle acceleration.
  • Enriched, low-comorbidity samples may not represent routine bipolar depression.
  • Mean-score differences and response rates can tell different stories; both should be reported.
  • Drug class labels conceal dose, formulation and combination specificity; OFC evidence cannot be generalized to fluoxetine alone.
  • Network meta-analysis depends on cross-trial comparability and yields rankings more precise-looking than the underlying data may warrant (Yildiz 2023, PMID 37595997).
  • Bipolar II and rapid-cycling subgroups are commonly underpowered.

Mixed features and incremental benefit

In 200 bipolar-depression participants with mixed features, lumateperone 42 mg produced a placebo-adjusted MADRS difference of −5.7 points (95% CI −8.29 to −3.05; effect size −0.64) at day 43. In the combined major-depressive/bipolar mixed-features population (n=388 randomized), somnolence occurred in 12.5% versus 1.6% with placebo, dizziness in 12.0% versus 2.1% and nausea in 9.9% versus 1.6%; no mania/hypomania treatment-emergent events were reported (Durgam 2025, PMID 39946099). This is stronger than merely labeling the trial “positive,” but the six-week window cannot establish maintenance safety or prevention of polarity switch.

Adjunctive lumateperone illustrates a smaller incremental effect. Among 529 participants inadequately responsive to lithium or valproate, 42 mg/day produced a placebo-adjusted MADRS difference of −2.4 points (P=.02), whereas 28 mg produced −1.7 points (P=.10); somnolence, dizziness and nausea occurred in 11.3%, 10.7% and 8.5% at 42 mg (Suppes 2023, PMID 36779257). Statistical efficacy therefore does not answer whether the mean increment is clinically meaningful for a given patient.

Ketamine versus ECT: speed and durability

A six-trial meta-analysis (n=340, predominantly severe unipolar or bipolar depression) favored ECT over ketamine for acute depressive severity (SMD −0.69, 95% CI −0.89 to −0.48), without detected differences in cognition/memory or serious adverse events (Rhee 2022, PMID 36260324). A later five-RCT synthesis (n=664) found ketamine faster at 24 hours (SMD −0.49, −0.96 to −0.03) but no end-of-treatment response difference (RR 0.96, 0.70–1.31) or remission difference (RR 0.91, 0.45–1.82), with high heterogeneity (I²=75% for response and 90% for remission at end of treatment, and 99% for the pooled symptom difference) (Shi 2025, PMID 39549887); that paper carries a published expression of concern (J Affect Disord. 2025;378:350–351), so its estimates should be treated as provisional. Different time points and mixed diagnostic samples explain part of the apparent contradiction.

Longitudinal observational evidence associated ECT with lower all-cause mortality (HR 0.70, 95% CI 0.61–0.81) across 17 studies and 1,182,501 people, but suicide reduction appeared only at three months (HR 0.53, 0.39–0.72), not at 1, 6 or 12 months (Rhee 2025, PMID 41407487). Confounding by indication and care intensity prevents interpreting this as randomized mortality protection.

Open questions

  • Which baseline features predict response to lurasidone, quetiapine, lumateperone, cariprazine or OFC rather than merely average group benefit (Yildiz 2023, PMID 37595997)?
  • Is lamotrigine’s larger effect in severe depression reproducible in a prospectively stratified trial (Geddes 2009, PMID 19118318)?
  • Can adjunctive light therapy’s response OR 2.32 be reproduced with standardized timing, dose and objective circadian measures (Lam 2020, PMID 31826657)?
  • What is ketamine’s bipolar-specific durability and switch risk, separated from unipolar depression (Bahji 2022, PMID 35231204)?
  • Which treatments improve work, relationships and cognitive function after symptom remission rather than only rating scales (Rajagopalan 2016, PMID 26928122)?

References

  1. Yildiz A, et al. Comparative efficacy and tolerability of pharmacological interventions for acute bipolar depression in adults: a systematic review and network meta-analysis. Lancet Psychiatry. 2023. PMID 37595997
  2. Loebel A, et al. Lurasidone monotherapy in the treatment of bipolar I depression: a randomized, double-blind, placebo-controlled study. American Journal of Psychiatry. 2014. PMID 24170180
  3. Kato T, et al. Double-blind, placebo-controlled study of lurasidone monotherapy for the treatment of bipolar I depression. Psychiatry and Clinical Neurosciences. 2020. PMID 32827348
  4. Thase ME, et al. Efficacy of quetiapine monotherapy in bipolar I and II depression: the BOLDER II study. Journal of Clinical Psychopharmacology. 2006. PMID 17110817
  5. Young AH, et al. A double-blind, placebo-controlled study of quetiapine and lithium monotherapy in bipolar depression (EMBOLDEN I). Journal of Clinical Psychiatry. 2010. PMID 20122369
  6. McElroy SL, et al. A double-blind, placebo-controlled study of quetiapine and paroxetine in bipolar depression (EMBOLDEN II). Journal of Clinical Psychiatry. 2010. PMID 20122366
  7. Earley W, et al. Cariprazine treatment of bipolar depression: a randomized double-blind placebo-controlled phase 3 study. American Journal of Psychiatry. 2019. PMID 30845817
  8. Calabrese JR, et al. Efficacy and safety of lumateperone for major depressive episodes associated with bipolar I or bipolar II disorder. American Journal of Psychiatry. 2021. PMID 34551584
  9. Tohen M, et al. Efficacy of olanzapine and olanzapine-fluoxetine combination in the treatment of bipolar I depression. Archives of General Psychiatry. 2003. PMID 14609883
  10. Citrome L. Olanzapine-fluoxetine combination for the treatment of bipolar depression. Expert Opinion on Pharmacotherapy. 2011. PMID 22035291
  11. Geddes JR, et al. Lamotrigine for treatment of bipolar depression: individual patient data from five randomised trials. British Journal of Psychiatry. 2009. PMID 19118318
  12. Calabrese JR, et al. Lamotrigine in the acute treatment of bipolar depression: results of five double-blind, placebo-controlled trials. Bipolar Disorders. 2008. PMID 18271912
  13. Zarate CA Jr, et al. Pramipexole for bipolar II depression: a placebo-controlled proof of concept study. Biological Psychiatry. 2004. PMID 15219473
  14. Frye MA, et al. A placebo-controlled evaluation of adjunctive modafinil in bipolar depression. American Journal of Psychiatry. 2007. PMID 17671288
  15. Lam RW, et al. Light therapy for patients with bipolar depression: systematic review and meta-analysis of randomized controlled trials. Canadian Journal of Psychiatry. 2020. PMID 31826657
  16. Bahji A, et al. Efficacy and safety of racemic ketamine and esketamine for depression: a systematic review and meta-analysis. Expert Opinion on Drug Safety. 2022. PMID 35231204
  17. Kato M, et al. Efficacy and safety of lurasidone in bipolar I depression with and without rapid cycling. Journal of Affective Disorders. 2023. PMID 37245552
  18. Xiao N, et al. Efficacy of pharmacological interventions in depression with mixed features: a systematic review. Bipolar Disorders. 2025. PMID 40808264
  19. Rajagopalan K, et al. Direct and indirect effects of lurasidone monotherapy on functional improvement among patients with bipolar depression. International Journal of Bipolar Disorders. 2016. PMID 26928122
  20. Durgam S, et al. Lumateperone for major depressive disorder or bipolar depression with mixed features: a randomized placebo-controlled trial. J Clin Psychopharmacol. 2025;45:67–75. PMID 39946099
  21. Suppes T, et al. Adjunctive lumateperone in the treatment of bipolar depression: results from a randomized placebo-controlled clinical trial. Bipolar Disord. 2023;25:478–488. PMID 36779257
  22. Rhee TG, et al. Efficacy and safety of ketamine versus electroconvulsive therapy among patients with major depressive episode: a systematic review and meta-analysis. JAMA Psychiatry. 2022;79:1162–1172. PMID 36260324
  23. Shi ZM, et al. Intravenous ketamine versus electroconvulsive therapy for major depressive disorder or bipolar depression: a meta-analysis of randomized controlled trials. J Affect Disord. 2025;371:45–53. PMID 39549887
  24. Rhee TG, et al. Longitudinal associations of electroconvulsive therapy with all-cause mortality and suicide deaths in severe unipolar or bipolar depression: a systematic review and meta-analysis. BMJ Ment Health. 2025;28:e302083. PMID 41407487