Articles

Comparative efficacy and tolerability of pharmacological

interventions for acute bipolar depression in adults:
a systematic review and network meta-analysis
Ayşegül Yildiz*, Spyridon Siafis*, Dimitris Mavridis, Eduard Vieta†, Stefan Leucht†

Summary
Background Bipolar depression constitutes a major public health problem due to its substantial burden of disease. Lancet Psychiatry 2023;
Although pharmacological interventions are available, guidelines required updated evidence synthesis to improve 10: 693–705

their current recommendations. In order to inform evidence-based prescribing, we investigated the comparative See Comment page 655
efficacy and tolerability of pharmacological interventions for acute bipolar depression. *Contributed equally
†Contributed equally
Methods We conducted a systematic review and network meta-analysis. We searched for randomised controlled trials Department of Psychiatry,
comparing pharmacological interventions with each other or placebo in adults with acute bipolar depression (type I, Dokuz Eylül University, Izmir,
Turkey (A Yildiz MD); Section
type II, or not otherwise specified), excluding those with substance misuse, unipolar depression, or schizophrenia, in for Evidence-Based Medicine in
MEDLINE, Embase, PsycINFO, Google Scholar, Cochrane Library, Web of Knowledge, CINAHL, and LILACS from Psychiatry and Psychotherapy,
database inception up to April 13, 2023. Criteria for eligibility were a duration of 2–16 weeks with masked outcome Department of Psychiatry and
assessments, and we included combination, add-on design, and monotherapy studies. The co-primary outcomes Psychotherapy, School of
Medicine, Technical University
were depressive symptoms, examined with standardised mean differences (SMDs), and manic switch, examined with of Munich, Munich, Germany
odds ratios (ORs). We also investigated dropouts due to any reason, inefficacy, adverse events, and important side- (S Siafis MD, S Leucht MD);
effects as secondary outcomes. The confidence in the evidence was evaluated using Confidence-In-Network-Meta- Department of Primary
Analysis (CINeMA). The study was registered with PROSPERO, CRD42020171726. Education, University of
Ioannina, Ioannina, Greece
(D Mavridis PhD); Bipolar
Results We analysed data from 101 randomised controlled trials covering 20 081 participants, 8063 men (41·7%) and Disorders Program, Institute of
11 263 women (58·3%; sex not available in four studies), mean age 41·0 years (range of means 28·7–53·6 years), and Clinical Neuroscience, Hospital
68 medications and placebo. Ethnicity data were not available. With moderate confidence in the evidence, olanzapine Clinic, University of Barcelona,
IDIBAPS, CIBERSAM, Barcelona,
plus fluoxetine, quetiapine, olanzapine, lurasidone, lumateperone, cariprazine, and lamotrigine were more efficacious Catalonia, Spain
than placebo in reducing depressive symptoms, with SMDs ranging from 0·41 (95% CI 0·19–0·64) for olanzapine (E Vieta MD PhD)
plus fluoxetine to 0·16 (0·03–0·29) for lamotrigine. Several other drugs might also be efficacious, but the confidence Correspondence to:
in the evidence was very low to low. Antidepressants as a class seem to be efficacious, but had a higher risk for manic Dr Spyridon Siafis, Department
switch compared to antipsychotics. Medications differed in their side-effect profiles. of Psychiatry and Psychotherapy,
School of Medicine, Technical
University of Munich,
Interpretation This is, to our knowledge, the largest network meta-analysis of pharmacotherapy for bipolar depression Munich 81675, Germany
to date. Olanzapine plus fluoxetine, quetiapine, olanzapine, lurasidone, lumateperone, cariprazine, and lamotrigine spyridon.siafis@tum.de
were found to be more efficacious than placebo in adults with acute bipolar depression, with good confidence in the
evidence, and to differ in their side-effect profiles. These findings can inform evidence-based care and the development
of treatment guidelines internationally.

Funding None.

Copyright © 2023 Elsevier Ltd. All rights reserved.

Introduction pharmacological treatments differ in their efficacy and

The Global Burden of Diseases, Injuries, and Risk Factors
Study (GBD) recently confirmed that depressive disorders
are among the leading causes of global burden of disease
worldwide.1 Bipolar disorder (BD) is a subset of these
diseases, and is characterised by periods of mania and
depression. In the USA, the lifetime prevalence
of BD is 4·4%, and actuarial annual cost estimates are
between US$195 and $202 billion.2,3 The depressive
phase of the illness is especially challenging, not only
because suicidality is most common in this phase,4 but
also because this phase accounts for 72% of the time
experiencing illness in BD.5 However, evidence-based
tolerability, and available systematic reviews either
focused on antipsychotics alone,6 or did not apply meta-
analysis.7 Except for one network meta-analysis on
antipsychotics (sponsored by the manufacturer of
lurasidone),6 no studies that we know of employed any
assessment of the confidence in the evidence, which is
necessary to draw evidence-based conclusions.8,9 Given
clearly different pathophysiological backgrounds,
extrapolations from unipolar depression should be
avoided.4 Moreover, antidepressants carry risk for switch
to mania,10 but how individual agents differ in this regard,
and which other treatments may reduce this risk, needs

www.thelancet.com/psychiatry Vol 10 September 2023 693

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles
Research in context
Evidence before this study
The burden and cost of bipolar disorder (BD) is mostly
associated with its depressive pole, not only because a
significantly higher proportion of bipolar patients experience
periods of depression, but also because that pole of illness is
harder to treat and is associated with higher risk of functional
impairment and suicide. However, approved treatments differ
in their efficacy and tolerability. We searched PubMed for
network meta-analyses on the acute treatment of bipolar
depression in adults between database inception and
April 13, 2023 with no language restrictions. Using the search
terms “bipolar” AND “depression” AND “network-meta-
analysis” OR “multiple-treatment meta-analysis”, we found
several relevant systematic reviews and network meta-analyses.
Some involved only second-generation antipsychotics or
specific treatments, whereas others were restricted to only
monotherapy or adjunctive treatments, or also involved
patients with unipolar depression or schizoaffective disorder.
Added value of this study
To our knowledge, this is the largest and most comprehensive
network meta-analysis on pharmacological interventions for
acute bipolar depression. Specifically, we analysed data from
101 randomised trials encompassing 20 081 participants on
68 medications or dietary supplements, administered as
monotherapy, combination, or adjunctive treatments.Olanzapine
to be addressed in a quantitative way. Given these
limitations, treatment guidelines, which are nowadays
mainly based on systematic reviews, may need to be
updated as well.11–13 Therefore, we performed a systematic
review and network meta-analysis of pharmacological
interventions for acute treatment of bipolar depression,
and applied the GRADE approach, which currently is the
gold standard for evidence synthesis, with the aim to
evaluate the comparative efficacy and tolerability of
different pharmacological interventions.14,15
Methods
We followed the PRISMA-NMA16 reporting guidelines
See Online for appendix 1 (appendix 1 pp 3–5) and had a registered protocol
(appendix 1 pp 6–9, PROSPERO-ID: CRD42020171726).
Search strategy and selection criteria
We searched electronic databases (MEDLINE, Embase,
PsycINFO, Google Scholar, Cochrane Library, Web of
Knowledge, CINAHL, and LILACS) from inception until
April 13, 2023 (appendix 1 pp 12–16), previous reviews,6,8,9
and conference proceedings, with no language
restrictions. Additional data were requested from authors
or companies.
We included adult patients (≥18 years old) with an
acute major depressive episode of BD (type I, type II, and
unspecified-NOS) according to standard diagnostic
694
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
plus fluoxetine, quetiapine, olanzapine, lurasidone, lumateperone,
cariprazine, and lamotrigine were efficacious in reducing
depressive symptoms with small-to-medium effect sizes and
supported by good quality evidence. Some other interventions
might also be efficacious, but the quality of the evidence was low
or very low, mainly due to imprecision and small sample sizes.
Antidepressants and (with a smaller effect size) antiepileptics, as a
class, could be almost as efficacious as antipsychotics; however,
larger trials are required to confirm this finding.
Although none of the medications clearly increased the risk of
mania switch, antidepressants as a class had a higher risk
compared to antipsychotics, while quetiapine clearly decreased
the risk based on good quality evidence. Moreover,
antipsychotics were associated with multiple side-effects but
with differing profiles, while lamotrigine may have a more
benign profile but may also be less efficacious.
Implications of all the available evidence
This extension of the evidence for the pharmacological
treatment of acute bipolar depression in adults, encompassing
the efficacy, acceptability, and side-effect profiles of all
available interventions, should greatly contribute to
personalised clinical care, facilitate the development of
treatment guidelines internationally, and provide valuable
insights for future clinical research.
criteria (eg, DSM-III and ICD-10, or newer versions).
There was no restriction in terms of the presence of
psychotic features, comorbid anxiety symptoms or
disorders, or lack of response to previous treatments. We
excluded patients with ongoing substance misuse,
manic, mixed, and euthymic states, unipolar depression,
schizoaffective disorder, or schizophrenia.
Published or unpublished randomised controlled trials
with study duration of 2–16 weeks with masked outcome
assessments were eligible. Data from the first phase of
crossover studies were used to avoid carryover effects.
We included combination studies (co-administration of
two or more test agents), add-on designs (addition of a
test agent to ongoing mood stabilisers), and monotherapy
studies, but in a sensitivity analysis, we analysed add-on
trials separately.
Any licensed pharmacological intervention or dietary
supplement compared with each other, or a placebo, and
in any form of administration was eligible. We included
fixed-dose and flexible-dose studies, and combined
multiple-dose groups of the same medication with
appropriate methods.17 Low fixed-doses that could be
deemed as subtherapeutic were excluded, ie, lamotrigine
50 mg/day, cariprazine 0·75 mg/day, lumateperone
28 mg/day, and ramelteon 0·1 mg/day.
Study screening and data extraction were conducted by
AY and confirmed by the authors of the included studies,
www.thelancet.com/psychiatry Vol 10 September 2023
 Articles

SS, and EV. AY and EV contacted authors and companies modifiers—ie, age and sex (post-hoc), baseline severity of
for each study to confirm and request data. AY evaluated depressive and mania symptoms, presence of psychotic
the risk of bias for the primary outcomes using the risk or rapid cycling features, comorbid anxiety, previous
of bias tool 2 (RoB 2),18 and this was confirmed by BA, SS, treatment failure, background mood stabilisers, type of
and SL. Further details of these procedures are reported BD, route of administration, number of sites, study
in appendix 1 pp 6–11. duration, and publication year. Moreover, we examined
the impact of these factors on the effect sizes of placebo-
Data analysis controlled comparisons with multivariable meta-
The primary efficacy outcome was change in depressive regressions.
symptoms as measured by validated scales (appendix We assumed a common heterogeneity variance (τ²),
p 7). Secondary efficacy outcomes were the number of and we quantified heterogeneity as low, moderate, and
participants who responded to treament and those who high by comparing τ² with its empirical distributions.22,23
achieved remission; any study definition was eligible. We downrated the confidence in the estimates due to
Efficacy outcomes were recorded as close to 8 weeks as indirectness when the comparisons were based on
possible. studies in which participants had a low baseline severity,
The primary safety outcome was the number of or had experienced treatment failure, or trial design as
patients with study-defined manic/hypomanic switch. add-on to mood stabilisers (appendix 1 p 197).
Secondary safety outcomes were the number of patients Incoherence was evaluated with global (design-by-
with extrapyramidal side-effects (EPS), akathisia, treatment-interaction test) and local (separating
QTc prolongation, weight gain (≥7% increase from the
total baseline body weight), and sedation. We also
30 467 records identified through 161 records identified from references
investigated overall drug acceptability along with the database searching of reviews and included studies
number of dropouts due to inefficacy or adverse events. 4952 Embase
We post-hoc explored change in mania symptoms, 7360 Google Scholar
3731 Cochrane Library
QTc interval (ms), and weight (kg) as continuous 4989 Web of Science
outcomes. Safety outcomes were recorded at study 2357 OVID (MEDLINE)
1845 PubMed (MEDLINE)
endpoint. Details on data extraction and imputation 1894 CINAHL
methods can be found in the appendix 1 pp 6–11. 2014 PsycINFO
We calculated random-effects, pairwise meta-analyses 450 ISRCTN registry
459 ICTRP registry
(PMA) and network meta-analyses in a frequentist 257 ClinicalTrials.gov registry
framework. A fixed-effects Mantel-Haenszel model was 92 GSK registry
67 LILACS
used for dichotomous outcomes with rare events
(appendix 1 pp 10, 126–131).19 We evaluated confidence in
the evidence for the primary outcomes using
CINeMA (Confidence-In-Network-Meta-Analysis),14 and 15 675 records after duplicates removed
conclusions were drawn using a minimally contextualised
framework of the GRADE approach.15 Following this
15 675 records screened
framework, we classified interventions into efficacious or
not convincingly different to placebo by taking into
consideration both the effect sizes and the confidence in 15 090 records excluded (based on title and abstract)
the evidence as evaluated with CINeMA. We examined 4280 non-bipolar depression studies
2471 non-pharmacological intervention
each individual medication separately, but we also 1232 reviews
conducted a post-hoc analysis of the following drug 7107 for other reasons
classes: antipsychotics, antidepressants, antipsychotics
plus antidepressants, antiepileptics, and lithium 585 full-text records assessed for eligibility
(appendix 1 p 11).
The effect sizes were mean differences (weight change
in kg, QTc interval in ms) and standardized mean 440 records excluded (based on full text)
377 excluded per inclusion and exclusion criteria
differences (SMDs/Hedges’ g) for continuous outcomes, 61 supplementary records
and odds ratios (ORs) for dichotomous outcomes, 2 full text could not be accessed
accompanied with their 95% CIs. ORs were also
converted to absolute risk in percent in order to ease
145 studies included in qualitative synthesis (in 140 records)
interpretability (appendix 1 p 10).20 P-scores were used to
rank treatments.21
The transitivity assumption of network meta-analysis 101 studies included in meta-analysis (in 99 records)
was assessed by examining the distribution of clinical
and methodological variables that might act as effect Figure 1: Study selection

www.thelancet.com/psychiatry Vol 10 September 2023 695

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles

indirectfrom direct evidence approach) methods.24 Role of the funding source

We performed only PMA when there was
substantial incoherence (significant design-by-
treatment-interaction test at alpha 0·1 and/or
≥20% incoherent closed loops).
We examined the robustness of the primary outcomes
with predefined sensitivity analyses by using a fixed-
effects model (or a random-effects model if originally a
fixed-effects was used), excluding studies based on the
above-mentioned potential effect-modifiers, excluding
studies with imputed values, excluding studies from less
developed countries (post-hoc),25 completers-only data,
and studies with an overall high risk of bias according to
the applied RoB 2.18 Small-study effects and potential
publication bias were investigated by visual inspection of
contour-enhanced, comparison-adjusted funnel plots,
and an Egger’s test.26–28
We used R version 4.0.3,29 and its packages meta-
version 5.1-1,30 and netmeta-version 2.1-0.31 Alpha was set
at two-sided 0·05, except for incoherence and funnel plot
analyses (alpha 0·1).
There was no funding source for this study.
Results
We identified 30 467 references, assessed 585 full-text
articles after title and abstract screening and included
101 studies with 20 081 participants (included studies are
shown in figure 1 and appendix 1 pp 17–34, 80–99). The
participants had a median age of 41·3 years
(IQR 38·9–43·8), 58·8% (IQR 53·3–64·2) were female;
41·2% (IQR 35·8–46·7) were male; 34 trials (33·7%)
included only patients with bipolar I, six (5·9%) only
patients with bipolar II, and 58 (57·4%) included both or
mixed groups (three studies did not state bipolar type).
We did not extract data about ethnicity.
The mean baseline severity was 48·0% (IQR 42·5–50·6)
of the maximum scale score (appendix 1 p 112); 21 studies
included participants who had experienced prior treatment
failure. Study authors provided additional data for
60 studies. There were some concerns about overall risk of
bias in about 10% of the studies (appendix 1 pp 100–109).

A Depressive symptoms
Esamisulpride
Esketamine Dexamphetamine
Fluoxetine Creatine
Idazoxan Ubidecarenone
Imipramine Citalopram
Infliximab Celecoxib
Inositol Carnitine + Lipoic-acid
Lamotrigine Cariprazine
Levatiracetam Carbamazepine + FEWP

Lithium Carbamazepine

Lumateperone Bupropion

Lurasidone Aspirin
Armodafinil
Memantine
Aripiprazole + Citalopram
Minocycline
Aripiprazole
Minocycline + Aspirin
Agomelatine
Minocycline + Celecoxib
Zonisamide
Moclobemide
Ziprasidone
Modafinil
Vitamin D
N-acetylcysteine
Venlafaxine
Naltrexone
Valproate + Memantine
Nutrients cocktail
Valproate + Cytidine
Olanzapine
Valproate
Olanzapine + Fluoxetine
Topiramate
Omega-3 (EPA) T4
Paroxetine
T3
Pioglitazone
Sertraline + Li
Placebo
Pramipexole Sertraline
Pregnenolone Selegiline
SAM
Quetiapine
Risperidone + Paroxetin
Ramelteon
Risperidone
Riluzole

(Figure 2 continues on next page)

696 www.thelancet.com/psychiatry Vol 10 September 2023

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles

B Mania switch
Esamisulpride Desipramine Creatine
Fluoxetine Citalopram
Imipramine Celecoxib
Infliximab Cariprazine
Lamotrigine Carbamazepine + FEWP

Levatiracetam Carbamazepine

Lithium Bupropion

Lumateperone Aspirin

Lurasidone Armodafinil

Aripiprazole
Memantine

Agomelatine
Minocycline

Minocycline + Aspirin
Zonisamide

Minocycline + Celecoxib
Ziprasidone

Moclobemide
Venlafaxine

Modafinil
Valproate + Cytidine

N-acetylcysteine
Valproate

Nutrients cocktail T4
Olanzapine
Sertraline+Li

Olanzapine + Fluoxetine Sertraline

Paroxetine SAM
Placebo Risperidone + Paroxetin
Pramipexole Risperidone
Riluzole
Quetiapine Ramelteon

Figure 2: Network plots for the co-primary outcomes

(A) Depressive symptoms. (B) Mania switch. The nodes of the networks represent the different treatments, and their size is proportional to the number of
participants receiving the treatment in the randomised controlled trials. The edges between nodes represent the presence of direct comparisons between
two treatments and their width is proportional to the number of randomised controlled trials that compared the two treatments. FEWP=Free and Easy Wanderer
Plus. Li=lithium. SAM=S-adenosyl-L-methione. T3=triiodothyronine. T4=thyroxine. EPA=eicosapentaenoic acid.

68 medications and placebo were investigated as add-
on treatments to mood stabilisers in 59 studies, and as
monotherapy or combination treatments without
ongoing mood stabilisers in 42 studies (figure 2).
96 RCTs involving 63 interventions with 18 669
(about 93%) participants contributed to the network on
the co-primary outcome, change in depressive symptoms
(figure 3).
Figure 3 presents the effect sizes of the treatments on
depressive symptoms compared with placebo. In this and
all other figures, treatments were grouped according to
the GRADE framework. This means that medications
which were statistically significantly more efficacious than
placebo and for which there was good confidence in the
evidence are presented on the top (category I), drugs with
significant effects but without good evidence in the middle
(category II), and non-efficacious drugs at the bottom
(category III). Within these three categories, drugs
were ordered according to their P-score rankings.
Seven treatments fell in category I: olanzapine
plus fluoxetine (SMD 0·41, 95% CI 0·19–0·64),
quetiapine (0·35, 0·23–0·47), olanzapine (0·35,
0·17–0·54), lurasidone (0·29, 0·14–0·45), lumateperone
(0·27, 0·09–0·45), cariprazine (0·23, 0·06–0·39), and
lamotrigine (0·16, 0·03–0·29; figure 3). Olanzapine plus
fluoxetine and quetiapine were more efficacious in
comparison with lamotrigine (appendix 2). Drugs See Online for appendix 2
from esketamine (1·07, 0·51–1·63) to the omega−3
fatty acid eicosapentaenoic acid (0·50, 0·01–0·98) in
figure 3 fell in the second category—ie, they had
statistically significantly different effects from placebo, but
with low or very low confidence in the evidence, mainly
due to lack of precision (see in particular small sample
sizes, and note that in mental health approximately

www.thelancet.com/psychiatry Vol 10 September 2023 697

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles
Figure 3: Forest plot for
depressive symptoms
This forest plot represents the
findings from the network
meta-analysis using placebo as
reference. The treatments
were ranked and grouped
according to the minimally
contextualised framework of
the GRADE approach that
takes into consideration both
the treatment effects, the
ranking (ie, P-scores) and the
confidence in the evidence
(ie, CINeMA). The colour key
represents the confidence in
the evidence: blue for
moderate, pink for low, and
red for very low confidence.
SMD=standardised mean
difference. k=number of
studies. n=number of
participants receiving the
treatment. FEWP=Free and
Easy Wanderer Plus.
Li=lithium. SAM=S-adenosyl-
L-methione.
T3=triiodothyronine.
698
SMD (95% CI)
Efficacious in Olanzapine + Fluoxetine (k=3, n=294) I 0·41 (0·19 to 0·64)
reducing Quetiapine (k=7, n=2152) I 0·35 (0·23 to 0·47)
depressive Olanzapine (k=4, n=732) I 0·35 (0·17 to 0·54)
symptoms Lurasidone (k=4, n=1029) I 0·29 (0·14 to 0·45)
(moderate Lumateperone (k=3, n=546) I 0·27 (0·09 to 0·45)
evidence; Cariprazine (k=4, n=997) I 0·23 (0·06 to 0·39)
category I)
Lamotrigine (k=11, n=948) I 0·16 (0·03 to 0·29)
Esketamine (k=1, n=66) I 1·07 (0·51 to 1·63)
Efficacious in Pramipexole (k=2, n=22) I 1·03 (0·37 to 1·69)
reducing Ubidecarenone (k=1, n=36) I 0·99 (0·43 to 1·54)
depressive Fluoxetine (k=2, n=38) I 0·75 (0·30 to 1·20)
symptoms Valproate + Cytidine (k=1, n=18) I 0·87 (0·08 to 1·67)
(very low or Valproate + Memantine (k=1, n=81) I 0·61 (0·07 to 1·14)
low evidence;
Valproate (k=6, n=162) I 0·51 (0·15 to 0·87)
category II)
Venlafaxine (k=3, n=157) I 0·47 (0·11 to 0·83)
Omega-3 (EPA) (k=2, n=56) I 0·50 (0·01 to 0·98)
Idazoxan (k=1, n=7) I 0·82 (−0·34 to 1·99)
Carbamazepine + FEWP (k=1, n=46) I 0·55 (−0·01 to 1·11)
Aripiprazole + Citalopram (k=1, n=10) I 0·70 (−0·36 to 1·76)
Sertraline (k=2, n=103) I 0·42 (0·00 to 0·83)
Modafinil (k=1, n=41) I 0·43 (−0·06 to 0·93)
Imipramine (k=3, n=141) I 0·32 (−0·03 to 0·67)
Esamisulpride (k=1, n=224) I 0·30 (−0·03 to 0·63)
Dexamphetamine (k=1, n=11) I 0·33 (−0·50 to 1·16)
Memantine (k=1, n=14) I 0·31 (−0·46 to 1·08)
Naltrexone (k=1, n=17) I 0·27 (−0·47 to 1·01)
Carbamazepine (k=1, n=47) I 0·24 (−0·32 to 0·79)
Creatine (k=1, n=16) I 0·25 (−0·56 to 1·06)
Selegiline (k=1, n=23) I 0·24 (−0·64 to 1·13)
Sertraline + Li (k=1, n=48) I 0·21 (−0·28 to 0·71)
Paroxetine (k=4, n=189) 0·18 (−0·06 to 0·42)
Lithium (k=4, n=298) 0·18 (−0·06 to 0·42)
Topiramate (k=1, n=18) 0·17 (−0·70 to 1·04)
Bupropion (k=3, n=78) 0·17 (−0·35 to 0·69)
T4 (k=1, n=31) 0·13 (−0·42 to 0·68)
Armodafinil (k=4, n=807) 0·13 (−0·03 to 0·29)
No clear Aripiprazole (k=1, n=162) 0·12 (−0·20 to 0·44)
difference N-acetylcysteine (k=2, n=99) 0·10 (−0·22 to 0·43)
from placebo Ziprasidone (k=4, n=674) 0·11 (−0·07 to 0·29)
in changing Citalopram (k=3, n=71) 0·08 (−0·32 to 0·48)
depressive Aspirin (k=1, n=19)
0·04 (−0·55 to 0·63)
symptoms
Inositol (k=2, n=21) 0·02 (−0·63 to 0·68)
(very low or
low evidence; Minocycline + Aspirin (k=1, n=30) 0·03 (−0·49 to 0·56)
category III) Pioglitazone (k=2, n=39) 0·03 (−0·44 to 0·50)
Risperidone + Paroxetin (k=1, n=10) −0·04 (−0·98 to 0·90)
Agomelatine (k=1, n=168) 0·02 (−0·30 to 0·34)
Moclobemide (k=1, n=81) −0·04 (−0·57 to 0·49)
Placebo Reference
Carnitine + Lipoic acid (k=1, n=20) −0·11 (−0·77 to 0·56)
Ramelteon (k=2, n=347) −0·04 (−0·28 to 0·20)
Risperidone (k=1, n=10) I −0·20 (−1·15 to 0·74)
Celecoxib (k=1, n=66) −0·13 (−0·53 to 0·27)
Nutrients−cocktail (k=1, n=61) −0·13 (−0·53 to 0·28)
Infliximab (k=1, n=28) I −0·19 (−0·76 to 0·38)
SAM (k=1, n=14) I −0·25 (−1·07 to 0·56)
Minocycline (k=2, n=84) I −0·18 (−0·53 to 0·17)
Levatiracetam (k=1, n=17) I −0·32 (−1·06 to 0·41)
Zonisamide (k=1, n=50) I −0·25 (−0·72 to 0·22)
Riluzole (k=2, n=55) I −0·27 (−0·70 to 0·16)
Minocycline + Celecoxib (k=1, n=68) I −0·32 (−0·73 to 0·08)
Vitamin D (k=1, n=13) I −0·49 (−1·32 to 0·35)
Pregnenolone (k=1, n=38) I −0·38 (−0·90 to 0·14)
T3 (k=1, n=5) I −0·80 (−2·06 to 0·46)
–2 –1 0 1 2
SMD (95% CI)
Favours placebo Favours medication
www.thelancet.com/psychiatry Vol 10 September 2023
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.

1000 participants are needed to make meta-analyses
stable).32 All other drugs were not clearly different from
placebo (category III; figure 3).
96 studies involving 64 interventions with
19 871 (about 99%) participants reported data for the
co-primary outcome mania switch, but 25 studies among
them could not be analysed due to the reporting of zero
events in all arms (figure 4). Quetiapine was the only
drug which clearly reduced the switching risk compared
to placebo (risk 1·0% vs 2·0%, odds ratio [OR] 0·49,
95% CI 0·33–0·75), and the confidence in the evidence
was moderate. All other drugs were not clearly different
from placebo, with confidence in the evidence being low
or very low, although some point estimates were
rather low, suggesting decreased risk, such as for
lithium (1·2%, 0·57, 0·19–1·73) and olanzapine
(1·3%, 0·63, 0·36–1·08), or high, suggesting increased
risk, such as for imipramine (4·4%, 2·22, 0·64–7·67;
figure 4). The findings of a post-hoc analysis using rating
scales measuring symptoms of mania were generally in
agreement with the main findings (appendix 1 pp 149–150).
Given the large number of medications, we focused the
description of the secondary outcomes on the category of
medications that were found to be efficacious in reducing
depressive symptoms with moderate confidence
(appendix 1 pp 200–293 and 348–349): cariprazine,
olanzapine (alone or combined with fluoxetine),
quetiapine, lurasidone, lumateperone, and lamotrigine.
More results are presented in figure 5 and appendix 1
pp 144–168.
The primary outcome (depressive symptoms) results
were in general corroborated by the secondary efficacy
outcomes response to treatment and remission
(appendix 1 pp 145–148). For category I drugs, response
rates ranged from 56·7% (OR 2·47, 95% CI 1·72–3·55)
for olanzapine plus fluoxetine to 42·4% for
lumateperone (1·39, 1·05–1·83), compared with
34·7% for placebo. Patterns for remission rates were
similar, except that the 95% CI for lumateperone did
not exclude the null (1·12, 0·84–1·50).
All-cause discontinuation was analysed with PMA due
to incoherence (appendix 1 pp 132, 151–152). It was less
frequent with olanzapine plus fluoxetine (11·7%,
OR 0·35, 95% CI 0·22–0·57) and olanzapine (20·4%,
0·68, 0·54–0·87), compared with placebo (27·3%). It was
also less frequent with olanzapine plus fluoxetine
compared with olanzapine alone. There was no clear
difference between placebo and quetiapine (26·5%),
lurasidone (27·1%), lamotrigine (27·5%), cariprazine
(29·2%), and lumateperone (33·8%) as all 95% CIs did
not exclude the null. The pattern for dropouts due
to inefficacy was mostly analogous to the afore­
mentioned efficacy outcomes. Conversely, the risk of
discontinuation due to adverse events was higher for
lumateperone (12·8%, OR 2·71, 95% CI 1·29–5·71) and
quetiapine (9·8%, 1·99, 1·51–2·63) compared with
placebo (5·1%). Both lumateperone and quetiapine had
www.thelancet.com/psychiatry Vol 10 September 2023
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
Articles
higher risk of discontinuation compared with
olanzapine (6·0%), lamotrigine (5·1%) and olanzapine
plus fluoxetine (4·5%). The risk of discontinuation with
quetiapine (9·8%) was also higher compared with
lurasidone (6·0%; appendix 2).
We analysed side-effects commonly associated with the
use of psychotropic drugs (figure 5).
Weight gain greater than or equal to 7% from the total
baseline body weight to endpoint occurred with most of
these medications, with a substantial risk of 30·3% with
olanzapine plus fluoxetine (OR 40·65, 95% CI
13·43–123·02) and 28·4% with olanzapine (37·20,
13·89–99·63) falling to 5·4% with lurasidone (5·39,
1·62–18·00), 3·5% with quetiapine (3·39, 2·14–5·35),
and 3·1% with cariprazine (2·98, 1·28–6·94) as
compared to 1·1% risk with placebo (appendix 1
pp 157–158). Exceptions were lamotrigine (0·4%, 0·35,
0·08–1·49) and lumateperone (0·8%, 0·79, 0·21–2·98).
Post-hoc analysis of mean change in weight revealed
mean differences (MDs) of 3·19 kg (95% CI 2·57–3·81)
with olanzapine plus fluoxetine, 2·87 kg (2·56–3·19)
with olanzapine, and 1·25 kg (1·01–1·49) with quetiapine
as compared to placebo (appendix 1 pp 159–160). Weight
gain with cariprazine was small (MD 0·67 kg,
95% CI 0·41–0·93), while with lurasidone, lumateperone,
and lamotrigine it was not clearly distinguishable from
placebo (appendix 1 pp 159–160).
Due to substantial incoherence (appendix 1
pp 132, 161–162), sedation was analysed with PMA. The
following medications had an increased risk ranging
from 24·2% for quetiapine (OR 6·89, 95% CI 5·58–8·50)
and 18·4% for lumateperone (4·89, 2·23–10·69)
to 11·4% for olanzapine (2·79, 2·05–3·78), 8·4% for
lurasidone (1·98, 1·31–3·01), 7·9% for olanzapine
plus fluoxetine (1·86, 1·02–3·40), and 7·7% for
cariprazine (1·80, 1·09–2·97) compared with placebo
(4·4%) (appendix 1 pp 161–162).
EPS and akathisia were mainly recorded in studies of
antipsychotics. The risk of EPS was higher, yet still low,
compared with placebo (0·4%), ranging from 1·1% for
quetiapine (OR 2·38, 95% CI 1·56–3·64) to 0·8%
for lurasidone (1·90, 1·26–2·86), and the 95% CIs for
olanzapine plus fluoxetine and lumateperone were wide
and did not exclude the null effect (appendix 1 pp 163 –164).
An exeption was olanzapine, which did not differ from
placebo (0·4%, 0·78, 0·36–1·70) and had a lower risk of
EPS than the other medications (appendix 2).
Similarly, the risk of akathisia compared with
placebo (0·2%) was higher for all of these medications,
ranging from 0·9% for cariprazine (OR 4·00,
95% CI 2·20–7·28) to 0·4% for quetiapine (1·75,
1·04–2·95); the 95% CIs for olanzapine and
lumateperone were wide and did not exclude the null
effect (appendix 1 pp 165–166). Cariprazine had higher
risk for akathisia compared to quetiapine (0·44,
0·20–0·97, appendix 2). There were no data for
olanzapine plus fluoxetine.
699
 Articles

Efficacious in %, OR (95%CI)
reducing the
risk of mania Quetiapine (k=7, n=2268) I 1·0%, 0·49 (0·33–0·75)
switch
(moderate Carbamazepine (k=1, n=49) I 0·3%, 0·15 (0·02–1·55)
evidence) Sertraline + Li (k=1, n=48) I 0·7%, 0·34 (0·07–1·65)
Bupropion (k=2, n=60) I 0·9%, 0·42 (0·09–2·02)
Valproate + Cytidine (k=1, n=19) I 0·6%, 0·31 (0·01–9·67)
Infliximab (k=1, n=30) I 0·7%, 0·34 (0·01–8·80)
Celecoxib (k=1, n=66) I 1·1%, 0·55 (0·16–1·97)
Sertraline (k=2, n=103) I 1·1%, 0·55 (0·14–2·11)
Lithium (k=4, n=304) I 1·2%, 0·57 (0·19–1·73)
Olanzapine (k=4, n=756) I 1·3%, 0·63 (0·36–1·08)
Risperidone (k=1, n=11) I 0·8%, 0·38 (0·01–11·17)
Risperidone + Paroxetin (k=1, n=11) I 0·8%, 0·38 (0·01–11·17)
Carbamazepine+FEWP (k=1, n=50) I 1·3%, 0·64 (0·13–3·10)
Memantine (k=1, n=14) I 1·4%, 0·67 (0·09–4·73)
Moclobemide (k=1, n=81) 1·6%, 0·79 (0·10–6·31)
Cariprazine (k=4, n=1022) 1·9%, 0·92 (0·48–1·78)
Ziprasidone (k=3, n=702) 1·9%, 0·95 (0·31–2·95)
Olanzapine + Fluoxetine (k=3, n=299) 2·0%, 0·99 (0·5–1·96)
Pramipexole (k=2, n=23) 2·0%, 0·99 (0·15–6·46)
Placebo I
2·0%, reference
Esamisulpride (k=1, n=227) I 2·0%, 1·00 (0·25–4·09)
Armodafinil (k=4, n=826) I 2·1%, 1·01 (0·45–2·24)
No clear Minocycline (k=2, n=86) I 2·1%, 1·03 (0·36–3·00)
difference
from placebo in N-acetylcysteine (k=1, n=59) I 2·1%, 1·04 (0·31–3·42)
the risk of Citalopram (k=2, n=70) I 2·1%, 1·04 (0·20–5·46)
mania switch Valproate (k=3, n=57) I 2·1%, 1·05 (0·37–2·99)
(very low or
Ramelteon (k=2, n=431) I 2·3%, 1·12 (0·20–6·13)
low evidence)
Venlafaxine (k=3, n=160) I 2·2%, 1·08 (0·28–4·18)
Aspirin (k=1, n=20) I 2·6%, 1·27 (0·04–41·15)
Lumateperone (k=3, n=550) I 2·4%, 1·16 (0·40–3·33)
Agomelatine (k=1, n=172) I 2·4%, 1·17 (0·39–3·57)
Fluoxetine (k=2, n=39) I 2·5%, 1·23 (0·28–5·36)
Lurasidone (k=4, n=1051) I 2·4%, 1·21 (0·55–2·65)
Lamotrigine (k=8, n=615) I 2·4%, 1·19 (0·71–1·98)
Desipramine (k=1, n=10) I 2·9%, 1·43 (0·08–26·97)
Paroxetine (k=4, n=199) I 2·5%, 1·25 (0·62–2·51)
Modafinil (k=1, n=41) I 2·7%, 1·34 (0·37–4·77)
Levatiracetam (k=1, n=19) I 3·5%, 1·76 (0·15–21·47)
Nutrients cocktail (k=1, n=61) I 3·2%, 1·59 (0·53–4·77)
Minocycline + Aspirin (k=1, n=32) I 4·8%, 2·44 (0·18–33·69)
Zonisamide (k=1, n=53) I 5·8%, 2·94 (0·12–73·93)
T4 (k=1, n=32) I 6·1%, 3·10 (0·12–79·05)
Aripiprazole (k=2, n=373) I 3·8%, 1·88 (0·69–5·14)
Minocycline + Celecoxib (k=1, n=68) I 4·0%, 2·02 (0·76–5·39)
Imipramine (k=3, n=146) I 4·4%, 2·22 (0·64–7·67)
Riluzole (k=1, n=9) I 8·7%, 4·60 (0·16–128·55)
SAM (k=1, n=25) I 9·4%, 5·00 (0·23–110·12)
Creatine (k=1, n=18) I 11·0%, 5·97 (0·27–133·87)

0·01 0·10 1·00 10·00 100·00

OR (95%CI)

Favours medication Favours placebo

Figure 4: Forest plot for mania switch

This forest plot represents the findings from the network meta-analysis using placebo as reference. The treatments were ranked and grouped according to the
minimally contextualised framework of the GRADE approach that takes into consideration both the treatment effects, the ranking (ie, P-scores) and the confidence in
the evidence (ie, CINeMA). The colour key represents the confidence in the evidence: blue for moderate, pink for low, and red for very low confidence. OR=odds ratio.
k=number of studies. n=number of participants receiving the treatment. %=risk of mania switch in patients receiving the treatment. FEWP=Free and Easy Wanderer
Plus. Li=lithium. SAM=S-adenosyl-L-methione. T4=thyroxine.

700 www.thelancet.com/psychiatry Vol 10 September 2023

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles

1 Weight gain 2 Sedation 3 EPS 4 Akathisia 5 QTc prolongation

Olanzapine + Fluoxetine
Efficacious in
Quetiapine
reducing
Olanzapine depressive
Lurasidone symptoms
Lumateperone (moderate
evidence;
Cariprazine category I)
Lamotrigine
Esketamine Efficacious in
Pramipexole reducing
Fluoxetine depressive
symptoms
Valproate + Cytidine
(very low or
Valproate low evidence;
Venlafaxine category II)
Carbamazepine + FEWP
Imipramine
Esamisulpride
Dexamphetamine
Memantine
Carbamazepine
Creatine
Paroxetine
Lithium
T4
No clear
Armodafinil difference
Aripiprazole from placebo
N-acetylcysteine in changing
depressive
Ziprasidone symptoms
Citalopram (very low or
Pioglitazone low evidence;
category III)
Risperidone + Paroxetin
Agomelatine
Moclobemide
Ramelteon
Risperidone
Nutrients-cocktail
Levatiracetam
Riluzole
Vitamin D
T3
0

0
10

00

0· 0

10

0
00

1·0

0
·0

·0

1·0

·0

1·0

·0

0
3
01

1
01

1
·0

·0
00

0·

00

0·
0·
1·0

·0

·0

1·0

·0

·0

3·

0·

0·
10

30

10

10
0·

0·
0·

0·

00

00
0·

0·
10

00

10

00

10

10
0·
10

10

10

10
10

10

OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)

Favours Favours Favours Favours Favours Favours Favours Favours Favours Favours
medication placebo medication placebo medication placebo medication placebo medication placebo

Figure 5: Forest plots for side-effects

Panel of forest plots that present the important side-effects, ie, weight gain, sedation, EPS, akathisia, and QTc prolongation. The forest plots represent the findings from the network meta-analysis
using placebo as reference, except for sedation for which only pairwise meta-analyses were conducted due to incoherence. The treatments were ranked and grouped according to findings on the
primary outcome of depressive symptoms, ie, efficacious medications with moderate evidence (category I), efficacious medications with poor evidence (category II), and medications not clearly
different to placebo (category III). EPS=extrapyramidal symptoms. OR=odds ratio. FEWP=Free and Easy Wanderer Plus. T3=triiodothyronine. T4=thyroxine.

Data for QTc prolongation were mainly available for
antipsychotics. No drug was associated with a clearly
increased number of participants with QTc prolongation
compared with placebo (0·1%), ranging from 0·4% for
lurasidone (OR 2·72, 95% CI 0·11–67·25) to 0·0% for
lumateperone (0·33, 0·03–3·22, appendix 1 p 167). This
finding was corroborated by post-hoc analysis of mean
change of QTc interval. The MDs compared with placebo
ranged from 2·32 ms (95% CI –0·07 to 4·71) for
olanzapine to –1·55 ms (–3·94 to 0·84) for quetiapine
(appendix 1 p 168). Olanzapine increased the QTc interval
more than quetiapine in this analysis (appendix 2). There
were no data for olanzapine plus fluoxetine and
lamotrigine.

www.thelancet.com/psychiatry Vol 10 September 2023 701

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
 Articles

There was some heterogeneity for depressive
symptoms (τ²=0·015, low-to-moderate), and some
incoherence in the co-primary outcomes according to the
number of incoherent loops (18·9%, 16·7%) but not
according to the design-by-treatment test (p=0·30, p=0·15,
respectively; appendix 1 pp 132–143). We therefore
downgraded the evidence in CINeMA (appendix 1
pp 197–199), and the results did not materially change in
sensitivity analyses excluding the incoherent loops
(appendix 1 pp 169–182). In terms of secondary outcomes,
there was low heterogeneity and no important concerns
of incoherence, except for dropouts due to any reason
and sedation, which were analysed with PMA
(appendix 1 pp 132–143).
There might be confounding because of the sparseness
of the network and given that the multivariable meta-
regression analysis of placebo-controlled trials identified
potential moderators of larger effect sizes for depressive
symptoms (lower mean age, higher percentage of female
participants, monotherapy treatment, prior treatment
failure and lower placebo response), and for manic
switch (higher mean age) (appendix 1 pp 183–184).
Nevertheless, the results were generally robust in the
predefined sensitivity analyses (appendix 1 pp 169–182).
Moreover, there were no material differences when we
analysed separately medications given as monotherapy
A Depressive symptoms
Antipsychotic plus antidepressant (k=5, n=314) I
Antipsychotics (k=29, n=6526) I
Antidepressants (k=13, n=698) I 0·28 (0·12 to 0·43)
Antiepileptics (k=19, n=1149) I
Lithium (k=4, n=298) I
Placebo
–0·25 0 0·25 0·50
Placebo better Medication better
B Mania switch
Antipsychotics (k=29, n=6960) I
Antipsychotic plus antidepressant (k=4, n=310)
I
Placebo
Lithium (k=4, n=304)
I
or adjunctive to mood stabilisers (appendix 1 pp 181), and
studies including more or fewer patients with bipolar II
than the median of the studies (27%, appendix 1 p 182).
Distribution of bipolar types across medications revealed
that antipsychotics were primarily investigated in
patients with bipolar I, while antidepressants were
studied more frequently in patients with bipolar II
(appendix 1 p 117).
The post-hoc analyses on drug classes revealed that
antidepressants in combination with antipsychotics
(SMD 0·40, 95% CI 0·17 to 0·63), antipsychotics alone
(0·28, 0·21 to 0·35), antidepressants alone (0·28,
0·12 to 0·43), and antiepileptics as a group (0·16,
0·04 to 0·28) were significantly more efficacious than
placebo, while the confidence interval for lithium
included no effect (0·09, –0·13 to 0·32). Antipsychotics
reduced the risk of switching to mania compared to
placebo (OR 0·74, 95% CI 0·58 to 0·95); antipsychotics
plus antidepressants (0·98, 0·52 to 1·87), lithium (1·05,
0·48 to 2·29), and antiepileptics (1·12, 0·73 to 1·72) were
neutral. Antidepressants were not clearly different from
placebo (1·39, 0·86 to 2·23) but were associated with
more switches than antipsychotics (figure 6, appendix 1
pp 353–356).
There was no indication of small-study effects in the
funnel plot analyses, yet there were unpublished data
that we could not obtain for a few comparisons and the
evidence was downrated accordingly (appendix 1
SMD (95% CI) pp 185–196).
0·40 (0·17 to 0·63)
0·28 (0·21 to 0·35)
Discussion
The present review, based on 101 studies with
0·16 (0·04 to 0·28) 68 pharmaceutical or supplementary compounds and
0·09 (–0·13 to 0·32) 20 081 participants, is the largest network meta-analysis
Reference in bipolar depression. Previous reviews based on
0·75
17 studies of ten mood stabilisers and antipsychotics,7
18 studies of six antipsychotics,6 or 46 studies of
22 psychotropic compounds8 either did not examine all
agents or applied less rigorous approaches. In particular,
none of these studies evaluated the confidence in
OR (95% CI)
evidence. For this reason, we have reached substantially
0·74 (0·58 to 0·95) different conclusions. Applying the GRADE approach,
0·98 (0·52 to 1·87)
only seven drugs, olanzapine plus fluoxetine, olanzapine,
Reference
quetiapine, lurasidone, lumateperone, cariprazine, and
1·05 (0·48 to 2·29)
lamotrigine can be clearly recommended.

Antiepileptics (k=13, n=775) I 1·12 (0·73 to 1·72)

We cannot confirm the superiority of lurasidone
Antidepressants (k=10, n=627) I 1·39 (0·86 to 2·23)
compared to cariprazine, as reported by
0 1 2 3 Kadakia and colleagues in 2021.6 Three lurasidone
Medication better Placebo better RCTs33–35 were not included in this previous network
meta-analysis by lurasidone’s manufacturer.6
Figure 6: Network meta-analysis by classifying medications into their drug classes
Cai and colleagues did not address lumateperone,
7
(A) Depressive symptoms. (B) Mania switch. The forest plots present the findings on the following drug classes
using placebo as reference: antidepressants (ie, agomelatine, bupropion, citalopram, esketamine, fluoxetine, olanzapine monotherapy, and anti­depressants in their
imipramine, moclobemide, imipramine, paroxetine, venlafaxine, and sertraline), antiepileptics (ie, lamotrigine, 2023 review. Bahji and colleagues, in two studies from
levetiracetam, topiramate, valproate, zonisamide, and carbamazepine), antipsychotics (ie, aripiprazole, cariprazine, 20208 and 2021,9 found the following treatments to be
esamisulpride, lumateperone, lurasidone, olanzapine, quetiapine, risperidone, and ziprasidone), lithium,
more efficacious than placebo: divalproex, olanzapine
antipsychotic plus antidepressant (ie, aripiprazole plus citalopram, olanzapine plus fluoxetine, risperidone plus
paroxetine). Other drugs were not considered as they were more heterogenous. The treatments were ranked plus fluoxetine, olanzapine, quetiapine, cariprazine, and
according to their P-scores. SMD=standardised mean difference. OR=odds ratio. lamotrigine as monotherapy; and intravenous ketamine,

702 www.thelancet.com/psychiatry Vol 10 September 2023

Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.

ubidecarenone, pramipexole, fluoxetine, and lamotrigine
as augmentation strategies. If we had followed their
approach of using statistical significance as the only
criterion, we would have recommended these drugs, and
several other ones, as well (figure 3). But to judge the
confidence in the evidence within a GRADE approach is
an essential component of recommendations.14,15 An
obvious problem is the small numbers of participants in
the category II drugs, which ranged between 18 and 162
(figure 3). For example, fluoxetine which was especially
recommended by Bahji and colleagues’ 2021 study9 had a
sample size of only 38 participants. It has been shown
that the results of meta-analyses in mental health only
become stable once approximately 1000 participants are
available.32 For these reasons, our results do also not
agree with all recommendations of the guideline by
Yatham and colleagues.11
All category I drugs except lamotrigine were
antipsychotics, which are problematic due to their side
effects. These can make them unacceptable for patients,
as documented by higher adverse-event-related dropouts
recorded for some antipsychotics relative to placebo
(appendix 1 p 156). An umbrella review including bipolar
depression also pointed out that most antipsychotics
have an unfavourable risk profile.36
Weight gain is extremely important, as atypical
symptoms including hyperphagia occur as frequently as
in 90% of patients with BD.37 Moreover, BD is associated
with a 10–20 years shorter life span, and cardiovascular
disease is the most common cause of premature
mortality.4,36 Olanzapine (and olanzapine plus fluoxetine)
were associated with the highest number of participants
experiencing weight gain greater than or equal to 7%
increase from their total baseline body weight, followed
by lurasidone, quetiapine, and cariprazine, which were
approximately equal, while lumateperone and
lamotrigine were weight neutral. We post-hoc examined
mean weight gain, and found that olanzapine caused an
average 3·19 kg within 8 weeks, olanzapine plus
fluoxetine 2·87 kg, quetiapine 1·25 kg, cariprazine 0·67 kg,
and lurasidone, lumateperone, and lamotrigine did not
produce significantly more weight gain than placebo.
The mean weight gain results are in line with those in
schizophrenia.36,38
There were no clear differences in patients with
QTc prolongation except olanzapine being worse than
quetiapine. The mean increases in msec compared to
placebo were very small, in line with a recent review.36
Meta-analyses of trials of patients with schizophrenia
found that antipsychotics (except sertindole, amisulpride
and, arguably, ziprasidone) were relatively benign.38
Nevertheless, patients with BD often receive
polypharmacy, and QTc prolongation can be
cumulative.4,5,11–13,36,37
All antipsychotics were sedating, and quetiapine and
lumateperone were the most sedating. Although this
side-effect can be transient, it can be a reason for
www.thelancet.com/psychiatry Vol 10 September 2023
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
Articles
non-adherence and may aggravate hypersomnia which is
common in BD.37
Almost all antipsychotics, with the possible exception
of olanzapine and lumateperone, produced more EPS
and akathisia than placebo. The percentages were small
(estimates between 1·1% and 0·4% for the drugs in
category I), but it should be noted that EPS are dose
related.39 Moreover, the occurrence of EPS in the short
term is associated with tardive dyskinesia in the long
term.40
Finally, lamotrigine was the only non-antipsychotic in
category I. Apart from rare, but potentially life-
threatening Stevens-Johnson syndrome,11,37 it has a good
tolerability profile, but only a small mean efficacy effect
size (0·16).
There was not a single antidepressant which can
be recommended without reservation for bipolar
depression. However, absence of evidence is not
evidence of absence.41 A post-hoc analysis revealed that,
as a class, antidepressants (and with a smaller effect size
antiepileptics) may be as efficacious as antipsychotics
(figure 6). The evidence on antidepressants is mainly
constituted by small randomised controlled trials, which
can exaggerate treatment effects.42 This result should
therefore be tested in larger studies. Moreover,
antidepressants led to more manic switches than
antipsychotics (appendix 1 pp 355–356). This effect was
not significant compared to placebo, but the short
duration of the trials (median eight weeks), and the fact
that antidepressants were usually added to mood
stabilisers, must be considered. Antidepressants were
more frequently studied in patients with bipolar II and
antipsychotics in patients with bipolar I (appendix 1 p 117).
Antidepressants might be preferably tested in bipolar II
populations, where a potential switch to hypomania
could have fewer detrimental effects, at least in the short
term, compared with a switch to mania in patients with
bipolar I. Regardless of these considerations, clinicians
should be aware that concurrent administration of a
mood stabiliser or antipsychotic is a prerequisite if
antidepressants are used in bipolar depression.10,11,37
In addition to lamotrigine, valproate could be an option
for some patients, even though it fell only in category II.
By contrast, neither the present network meta-analysis
nor a recent systematic review on use of lithium in BD
based on 78 studies found lithium efficacious for acute
bipolar depression.43 Thus, lamotrigine and valproate
might be preferred yet, given the low confidence in the
evidence with the latter, this judgment might be changed
by future evidence. Supportively, a recent network meta-
analysis on maintenance treatment of BD reported
lamotrigine and valproate in combination with each
other and other drugs to reduce recurrence or relapse of
depressive episodes.44
This network meta-analysis had certain limitations.
Given the bipolar nature of the disorder, protection against
switch to mania by mood stabilisers is necessary; thus, we
703
 Articles
included add-on trials. The term mood stabilisers is ill
defined. We used it for the background medications in
add-on trials, following the definitions of the original
authors. This, as well as inclusion of trials on patients who
had experienced previous treatment failure, might raise
concerns about the transitivity assumption as revealed by
the incoherence indicated by 18·9% and 16·7% of the
loops for the primary efficacy and safety outcomes,
respectively. However, when studies of incoherent loops
were excluded from the network, no indication of
incoherence was left, and results remained largely
unchanged (appendix 1 pp 169–180). In addition, the
network was sparse and most randomised controlled trials
were placebo controlled. Hence, there was very little flow
of indirect evidence and treatment effect estimates were
mainly informed by direct evidence. For this reason, we
downrated the evidence and explored potential reasons
with meta-regression and sensitivity analyses. We could
not substantively address differences in terms of patients
with bipolar I and bipolar II, because only 6% of the
studies included only patients with bipolar II, most
populations were mixed. The sensitivity analysis based on
the median percentage of patients with bipolar II had
limited statistical power. Future individual patient data
meta-analyses are needed to clarify this issue, as well as
the effect of other participant-related characteristics such
as age, sex, ethnicity, and baseline symptom severity.
When treatment starts to work, it can be challenging
to differentiate between improvement of depressive
symptoms and switch to hypomania. Similarly, anti­
depressants can also lead to mixed states, but this was
usually not recorded. Finally, as a general limitation, meta-
analyses are usually better at answering global questions
than specific ones, for example, we did not address which
drug should be used if the first failed.
In conclusion, olanzapine plus fluoxetine, quetiapine,
olanzapine, lurasidone, lumateperone, cariprazine, and
lamotrigine are efficacious for bipolar depression with
sufficient confidence in the evidence, and without
increased risk for switch to mania. Antipsychotics have
multiple side-effects, but these differ between individual
drugs and not all patients will develop them. We present
the side effects in figure 5 and appendix 1 (pp 144–168),
which might support individualised treatment.
Lamotrigine is a more tolerated choice, but its small
efficacy effect size must be considered. Antidepressants
were efficacious as a group, but larger trials on individual
drugs are needed, and protection from manic switch is a
prerequisite.
Contributors
AY, SS, DM, EV, and SL contributed to the study. AY with inputs from
SS, SL, and EV wrote the study protocol. Study screening and data
extraction were conducted by AY and confirmed by the authors of
included studies, SS, and EV. AY and EV contacted authors and
companies for each study to confirm and request data. AY evaluated the
risk of bias for the primary outcomes using the RoB 2, and this was also
confirmed by BA, SS, and SL. SS, with inputs from DM and SL, did all
the statistical analysis, and prepared Article figures and appendices.
704
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
SL supervised the conduct of the review. AY wrote the first draft of the
report with input from SS and SL. All authors had full access to all the
data in the study and had final responsibility for the decision to submit
for publication. AY and SS directly accessed and verified the underlying
data reported in the manuscript.
Declaration of interests
SS, DM declare no competing interests. AY has received honoraria for
lectures from Genveon. EV has received grants and served as consultant,
advisor, or CME speaker for the following entities: AB-Biotics, AbbVie,
Adamed, Angelini, Biogen, Boehringer-Ingelheim, Celon Pharma,
Compass, Dainippon Sumitomo Pharma, Ethypharm, Ferrer, Gedeon
Richter, GH Research, Glaxo-Smith Kline, Janssen, Lundbeck,
Medincell, Merck, Novartis, Orion Corporation, Organon, Otsuka,
Roche, Rovi, Sage, Sanofi-Aventis, Sunovion, Takeda, and Viatris, outside
the submitted work. In the past 3 years, SL has received honoraria as
advisor and/or for lectures and/or for educational material from
Alkermes, Angelini, Apsen, Eisai, Gedeon Richter, Janssen, Karuna,
Kynexis, Lundbeck, Medichem, Medscape, Merck Sharp and Dohme,
Mitshubishi, Neurotorium, NovoNordisk, Otsuka, Recordati, Roche,
Rovi, Sanofi Aventis, and TEVA.
Data sharing
Please contact the corresponding author if you would like to see any data
that are not included in the Article or the appendices.
Acknowledgments
We would like to acknowledge Burç Aydin* and Keming Geo for their
initial contributions to the present work. We thank Binnur Kavlak for
her assistance on the search update. We also thank Abdullah Murat Mete
for his enduring technical support. We present our gratitude to the
fabulous field authors Gary Sachs, Bruce Cohen, Suzann Babb,
Beth Murphy, Nassir Ghaimi, Sujung Yoon, Perry Renshaw,
Paul Markovitz, Ishrat Husain, Suresh Durgam, Arnim Quante,
Jay Amsterdam, Angelos Halaris, Trisha Suppes, Beny Lafer,
Lakshmi Yatham, Michael Berk, Alyna Turner, Olivia Dean,
Wendy Marsh, Heinz Grunze, Sylwia Janowska, Maciej Wieczorek,
Allan Young, Rasmus Licht, Raphael Braga, Roy Chengappa,
Charles Nemeroff, Michael Thase, John Hodsoll, Brian Brennan,
Lori Davis, Salih Demircioglu, Mehmet Cüneyt Yazicioglu, Robert Post,
Zubin Bhagwagar, Mark Frye, Michael Bauer, Roger McIntyre,
Carlos Zarate, and Zhang-Jin Zhang who did their best to check
extracted data and provide us all missing information they could reach
under their tough schedule.
*Burç Aydin died in July, 2023.
References
1 Vos T, Lim SS, Abbafati C, et al. Global burden of 369 diseases and
injuries in 204 countries and territories, 1990–2019: a systematic
analysis for the Global Burden of Disease Study 2019. Lancet 2020;
396: 1204–22.
2 Bessonova L, Ogden K, Doane MJ, O’Sullivan AK, Tohen M.
The economic burden of bipolar disorder in the United States:
a systematic literature review. Clinicoecon Outcomes Res 2020;
12: 481–97.
3 Cloutier M, Greene M, Guerin A, Touya M, Wu E. The economic
burden of bipolar I disorder in the United States in 2015.
J Affect Disord 2018; 226: 45–51.
4 McIntyre RS, Berk M, Brietzke E, et al. Bipolar disorders. Lancet
2020; 396: 1841–56.
5 Forte A, Baldessarini RJ, Tondo L, Vázquez GH, Pompili M,
Girardi P. Long-term morbidity in bipolar-I, bipolar-II, and
unipolar major depressive disorders. J Affect Disord 2015;
178: 71–78.
6 Kadakia A, Dembek C, Heller V, et al. Efficacy and tolerability of
atypical antipsychotics for acute bipolar depression: a network
meta-analysis. BMC Psychiatry 2021; 21: 249.
7 Cai L, Chen G, Yang H, Bai Y. Efficacy and safety profiles of mood
stabilizers and antipsychotics for bipolar depression: a systematic
review. Int Clin Psychopharmacol 2023; 38: 249–60.
8 Bahji A, Ermacora D, Stephenson C, Hawken ER, Vazquez G.
Comparative efficacy and tolerability of pharmacological treatments
for the treatment of acute bipolar depression: a systematic review
and network meta-analysis. J Affect Disord 2020; 269: 154–84.
www.thelancet.com/psychiatry Vol 10 September 2023

9 Bahji A, Ermacora D, Stephenson C, Hawken ER, Vazquez G.
Comparative efficacy and tolerability of adjunctive
pharmacotherapies for acute bipolar depression: a systematic review
and network meta-analysis. Can J Psychiatry 2021; 66: 274–88.
10 Fornaro M, Anastasia A, Novello S, et al. Incidence, prevalence and
clinical correlates of antidepressant-emergent mania in bipolar
depression: a systematic review and meta-analysis. Bipolar Disord
2018; 20: 195–227.
11 Yatham LN, Kennedy SH, Parikh SV, et al. Canadian Network for
Mood and Anxiety Treatments (CANMAT) and International Society
for Bipolar Disorders (ISBD) 2018 guidelines for the management
of patients with bipolar disorder. Bipolar Disord 2018; 20: 97–170.
12 Grunze H, Vieta E, Goodwin GM, et al. The World Federation of
Societies of Biological Psychiatry (WFSBP) guidelines for the
biological treatment of bipolar disorders: update 2010 on the
treatment of acute bipolar depression. World J Biol Psychiatry 2010;
11: 81–109.
13 Baldessarini RJ, Vázquez GH, Tondo L. Bipolar depression: a major
unsolved challenge. Int J Bipolar Disord 2020; 8: 1.
14 Nikolakopoulou A, Higgins JPT, Papakonstantinou T, et al.
CINeMA: an approach for assessing confidence in the results of a
network meta-analysis. PLoS Med 2020; 17: e1003082.
15 Brignardello-Petersen R, Florez ID, Izcovich A, et al. GRADE
approach to drawing conclusions from a network meta-analysis
using a minimally contextualised framework. BMJ 2020;
371: m3900.
16 Hutton B, Salanti G, Caldwell DM, et al. The PRISMA extension
statement for reporting of systematic reviews incorporating network
meta-analyses of health care interventions: checklist and
explanations. Ann Intern Med 2015; 162: 777–84.
17 Higgins JPT, Thomas J, Chandler J, et al, eds. Cochrane handbook
for systematic reviews of interventions. New Jersey:
John Wiley & Sons, 2019.
18 Sterne JAC, Savović J, Page MJ, et al. RoB 2: a revised tool for
assessing risk of bias in randomised trials. BMJ 2019; 366: l4898.
19 Efthimiou O, Rücker G, Schwarzer G, Higgins JPT, Egger M,
Salanti G. Network meta-analysis of rare events using the
Mantel-Haenszel method. Stat Med 2019; 38: 2992–3012.
20 Leucht S, Siafis S, Engel RR, et al. How efficacious are antipsychotic
drugs for schizophrenia? An interpretation based on 13 effect size
indices. Schizophr Bull 2022; 48: 27–36.
21 Rücker G, Schwarzer G. Ranking treatments in frequentist network
meta-analysis works without resampling methods.
BMC Med Res Methodol 2015; 15: 58.
22 Rhodes KM, Turner RM, Higgins JP. Predictive distributions were
developed for the extent of heterogeneity in meta-analyses of
continuous outcome data. J Clin Epidemiol 2015; 68: 52–60.
23 Turner RM, Davey J, Clarke MJ, Thompson SG, Higgins JP.
Predicting the extent of heterogeneity in meta-analysis, using
empirical data from the Cochrane Database of Systematic Reviews.
Int J Epidemiol 2012; 41: 818–27.
24 Efthimiou O, Debray TP, van Valkenhoef G, et al. GetReal in
network meta-analysis: a review of the methodology.
Res Synth Methods 2016; 7: 236–63.
25 Panagiotou OA, Contopoulos-Ioannidis DG, Ioannidis JPA.
Comparative effect sizes in randomised trials from less developed
and more developed countries: meta-epidemiological assessment.
BMJ 2013; 346: f707.
26 Peters JL, Sutton AJ, Jones DR, Abrams KR, Rushton L.
Contour-enhanced meta-analysis funnel plots help distinguish
publication bias from other causes of asymmetry. J Clin Epidemiol
2008; 61: 991–96.
www.thelancet.com/psychiatry Vol 10 September 2023
Downloaded for Anonymous User (n/a) at University of Monterrey from ClinicalKey.com by Elsevier on August 21,
2023. For personal use only. No other uses without permission. Copyright ©2023. Elsevier Inc. All rights reserved.
Articles
27 Chaimani A, Higgins JP, Mavridis D, Spyridonos P, Salanti G.
Graphical tools for network meta-analysis in STATA. PLoS One
2013; 8: e76654.
28 Egger M, Davey Smith G, Schneider M, Minder C. Bias in
meta-analysis detected by a simple, graphical test. BMJ 1997;
315: 629–34.
29 R Core Team: R: a language and environment for statistical
computing. Vienna: Scientific Research, 2018.
30 Balduzzi S, Rücker G, Schwarzer G. How to perform a meta-
analysis with R: a practical tutorial. Evid Based Ment Health 2019;
22: 153–60.
31 Balduzzi S, Rücker G, Nikolakopoulou A, et al. netmeta:
an R package for network meta-analysis using frequentist methods.
J Stat Soft 2023; 106:1–40.
32 Trikalinos TA, Churchill R, Ferri M, et al. Effect sizes in cumulative
meta-analyses of mental health randomized trials evolved over time.
J Clin Epidemiol 2004; 57: 1124–30.
33 Suppes T, Kroger H, Pikalov A, Loebel A. Lurasidone adjunctive
with lithium or valproate for bipolar depression: a placebo-
controlled trial utilizing prospective and retrospective enrolment
cohorts. J Psychiatr Res 2016; 78: 86–93.
34 Loebel A, Cucchiaro J, Silva R, et al. Lurasidone as adjunctive
therapy with lithium or valproate for the treatment of bipolar I
depression: a randomized, double-blind, placebo-controlled study.
Am J Psychiatry 2014; 171: 169–77.
35 Kato T, Ishigooka J, Miyajima M, et al. Double-blind, placebo-
controlled study of lurasidone monotherapy for the treatment of
bipolar I depression. Psychiatry Clin Neurosci 2020; 74: 635–44.
36 Croatto G, Vancampfort D, Miola A, et al. The impact of
pharmacological and non-pharmacological interventions on
physical health outcomes in people with mood disorders across the
lifespan: an umbrella review of the evidence from randomised
controlled trials. Mol Psychiatry 2023; 28: 369–90.
37 Grande I, Berk M, Birmaher B, Vieta E. Bipolar disorder. Lancet
2016; 387: 1561–72.
38 Huhn M, Nikolakopoulou A, Schneider-Thoma J, et al. Comparative
efficacy and tolerability of 32 oral antipsychotics for the acute
treatment of adults with multi-episode schizophrenia: a systematic
review and network meta-analysis. Lancet 2019; 394: 939–51.
39 Siafis S, Wu H, Wang D, et al. Antipsychotic dose, dopamine D2
receptor occupancy and extrapyramidal side-effects: a systematic
review and dose response meta-analysis. Mol Psychiatry (in press).
40 Tenback DE, van Harten PN, Slooff CJ, van Os J. Evidence that early
extrapyramidal symptoms predict later tardive dyskinesia:
a prospective analysis of 10,000 patients in the European
Schizophrenia Outpatient Health Outcomes (SOHO) study.
Am J Psychiatry 2006; 163: 1438–40.
41 Altman DG, Bland JM. Absence of evidence is not evidence of
absence. BMJ 1995; 311: 485.
42 Dechartres A, Trinquart L, Boutron I, Ravaud P. Influence of trial
sample size on treatment effect estimates: meta-epidemiological
study. BMJ 2013; 346: f2304.
43 Fountoulakis KN, Tohen M, Zarate CA Jr. Lithium treatment of
bipolar disorder in adults: a systematic review of randomized trials
and meta-analyses. Eur Neuropsychopharmacol 2022; 54: 100–15.
44 Kishi T, Ikuta T, Matsuda Y, et al. Mood stabilizers and/or
antipsychotics for bipolar disorder in the maintenance phase:
a systematic review and network meta-analysis of randomized
controlled trials. Mol Psychiatry 2021; 26: 4146–57.
705