Published on 15 September 2026
Esketamine in bipolar depression: 2126 patients, no excess manic switch, no proof of safety
In brief
Fear of a manic switch keeps esketamine away from many depressed patients with bipolar disorder. A retrospective cohort study on the international TriNetX network compared 2126 adults with bipolar depression after 1:1 propensity score matching: half on esketamine added to a mood stabiliser, half on the mood stabiliser alone. No excess of manic switch was detected over one year, and the observed rate is even significantly lower at 180 and 365 days. Suicide-related events are less frequent in the exposed group, with a hazard ratio of 0.439 in the first week. This is an observational comparison: none of these figures establishes a causal effect, and the bias that threatens the result most is the one matching does not correct, the selection of patients by the clinician. No labelling authorises esketamine in bipolar depression, so the prescription is off-label wherever you practise.
The context
Bipolar depression takes up most of the time spent ill, it often resists the first lines of treatment, and the clinician has few tolerated options. Intranasal esketamine changed the tempo of treatment in unipolar treatment-resistant depression, but extending it to bipolar disorder runs into an old objection: anything that lifts mood quickly is suspected of tipping it over.
The question asked in the consulting room is a concrete one. A depressed patient with bipolar disorder, on a mood stabiliser, after several failed lines, with suicidal ideation: does referring them for esketamine expose them to a switch? Until now the answer rested on series of a few dozen patients. This study shifts the order of magnitude.
The mechanism
Where the fear of switching comes from, and what the study did not measure
No biological measurement was taken here. The study compares rates of events coded in medical records, with no marker, no assay, no imaging. The table below recalls where the fear of switching comes from, and serves to separate what is established from what rests on reasoning by analogy.
| Element | Status | What this study says about it |
|---|---|---|
| Switch under a monoaminergic antidepressant | Established for tricyclics, debated for selective serotonin reuptake inhibitors | Nothing. Applying that fear to esketamine is a class extrapolation, not a finding |
| NMDA antagonism and glutamatergic signalling | Accepted mechanism of the rapid antidepressant effect of ketamine | No mechanistic marker measured, no pharmacological control arm |
| Effect of esketamine on its own | Unknown in this population | Cannot be isolated: every exposed patient was also receiving a mood stabiliser |
| Dose and duration of exposure | Decisive in practice | Not reported. Exposure is defined by the prescription, not by the number of sessions |
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Adults with bipolar depression, TriNetX global collaborative network. 2126 patients after matching, mean age 47.0 years, 63% female. Study conducted in May 2025, collection period not stated | |
| Intervention | |
| Esketamine added to a mood stabiliser | |
| Comparator | |
| Mood stabiliser alone, 1:1 propensity score matched. The figure of 1063 patients per arm follows from one-to-one matching, it is not spelled out in the abstract | |
| Outcomes | |
| Suicide-related events and manic switches, across the 1 to 7, 1 to 30, 1 to 90, 1 to 180 and 1 to 365 day intervals | |
| Design | |
| Retrospective cohort on a health data warehouse, survival analysis, model not specified in the abstract. PEB appraisal: CEBM level 3b, risk of bias judged with the ROBINS-I tool |
Quality control
| Criterion | Status |
|---|---|
| Sample size and power | Sound |
| Finding2126 matched patients, analyses run out to 365 days. The largest series we are aware of on this question | |
| Matching | Sound |
| Finding1:1 propensity score. The published abstract details neither the matching variables nor the residual standardised differences | |
| Choice of comparator | Sound |
| FindingMood stabiliser alone against mood stabiliser plus esketamine: the comparison isolates the addition, which is the right clinical question | |
| Case definition | Reservation |
| FindingDiagnoses and events drawn from record codes, with no structured interview. Classifying bipolar disorder and classifying a switch both depend on the quality of the coding | |
| Multiplicity | Problem |
| FindingTwo outcomes, five intervals, three subgroup dimensions. No correction for multiple testing is mentioned in the published abstract | |
| Sensitivity analyses | Reservation |
| FindingThe published abstract reports no sensitivity analysis. The robustness of the estimates therefore cannot be judged | |
| Funding and declared interests | Reservation |
| FindingThe record carries no funding statement, and that element could not be verified. The authors declare no competing interest. Hospital and university team: two of the five authors only belong to the department of psychiatry, the others practising in nutrition, in internal medicine and in hospital medicine | |
The findings
| Outcome | Published result |
|---|---|
| Suicide, 1 to 7 days | HR 0.439 (95% CI 0.256 to 0.753) |
| PEB readingStrong association wide interval, event count over seven days probably small | |
| Suicide, 1 to 30 days | HR 0.485 (95% CI 0.324 to 0.726) |
| PEB readingClear association the interval comfortably excludes 1 | |
| Suicide, 1 to 90 days | HR 0.641 (95% CI 0.456 to 0.901) |
| PEB readingAssociation maintained steady attenuation | |
| Suicide, 1 to 180 days | Not reported in the published abstract |
| PEB readingGap in the series the run of intervals is incomplete for this outcome | |
| Suicide, 1 to 365 days | HR 0.754 (95% CI 0.577 to 0.985) |
| PEB readingSignal at the edge of the threshold upper bound at 0.985, not to be presented as an established lasting benefit | |
| Manic switch, short intervals | Risk not increased, no significant result reported at 7, 30 and 90 days |
| PEB readingNo significant difference this is not proof that there is no effect, only the absence of a detectable signal | |
| Manic switch, 1 to 180 days | HR 0.643 (95% CI 0.442 to 0.935) |
| PEB readingReduction observed a glutamatergic drug that would protect against switching at six months calls for more than a database | |
| Manic switch, 1 to 365 days | HR 0.673 (95% CI 0.477 to 0.950) |
| PEB readingReduction observed same reservation, upper bound close to 1 | |
| Subgroups | Reduction in suicide risk consistent across age, sex and race. Reduction in switching in women at the longer intervals, not found in men |
| PEB readingNot to be used a subgroup contrast with no interaction test reported, in a design already loaded with comparisons | |
The most interesting point is not the size of the hazard ratios, it is their direction. What was expected was an excess of switches in the exposed group, or at best parity. The opposite is observed at the longer intervals. A reduction in switching under esketamine has no established pharmacological rationale, which points towards an explanation by selection rather than towards an effect of the drug.
Critical appraisal
The authors conclude that esketamine used alongside mood stabilisers is “a safe and potentially effective treatment for bipolar depression”, with “sustained anti-suicidal benefits”. PEB does not endorse that wording. A matched observational design does not measure safety in use: it compares two groups assembled by the prescriber’s decision, and propensity score matching rebalances only the coded variables. No systematic collection of adverse events is described, which rules out the word safe. The lasting character of the benefit rests on estimates whose upper confidence bound brushes 1 at 365 days. These results justify putting the question to a randomised trial, they do not settle it.
| ROBINS-I domain | Judgement |
|---|---|
| D1, confounding | Serious |
| FindingPropensity score matching on a large record network, which balances the measured variables, but residual confounding by indication that cannot be measured. The bias lies above all in the implicit contraindication: a clinician does not offer esketamine to a patient they judge unstable or close to switching. That judgement exists in no coded variable, so no propensity score can balance it. It is exactly the bias that would produce the result observed on switching | |
| D2, selection of participants | Reservation |
| FindingThe index date is not described in the abstract. A prescription of esketamine marks a moment of crisis, a prescription of a mood stabiliser marks any moment in the trajectory: the two cohorts are not necessarily observed at the same point in the course of illness | |
| D3, classification of the intervention | Reservation |
| FindingExposure defined by the prescription. No dose, no frequency, no duration, no adherence. No dose-response relationship can be examined | |
| D4, deviations from intended intervention | Reservation |
| FindingEsketamine is given under direct supervision, with closely spaced visits. The exposed group receives a care package, not just a molecule. The study cannot separate the two | |
| D5, missing data | Reservation |
| FindingCare delivered outside the network escapes the count, and deaths by suicide are poorly captured by a record warehouse. The composite outcome measures documented events, not mortality | |
| D6, measurement of outcomes | Reservation |
| FindingA composite outcome whose composition is not detailed in the abstract, which speaks only of suicide-related events. The simplest assumption is that documented ideation weighs more heavily in it than attempts, but that is only an assumption. A patient seen more often has their symptoms coded more often, which works against the direction of the observed results rather than in their favour | |
That last remark deserves to be followed through, because it cuts both ways. Most of the biases that are easy to name here work against the published results: patients referred for esketamine are generally more severe, more resistant, more suicidal, and they are seen more often, so they are better coded. An excess of events in the exposed group would be expected, not a shortfall. That is what makes the negative conclusion, the absence of excess risk of switching, more credible than the positive one.
One bias does run in the direction of the result, and it is the only one that really counts: channelling by caution. The patients offered esketamine are the ones the prescriber judges stable on the manic side. No matching on diagnostic codes reproduces that clinical assessment. Until a randomised trial neutralises it, the observed drop in switching at 180 and 365 days should be read as a plausible artefact of selection, not as a property of the drug.
Level of evidence
PEB appraisal: reasonable confidence on the absence of a switch signal, low confidence on everything else. What the study supports well is that no excess of manic switch appears over one year in depressed patients with bipolar disorder who receive esketamine in combination with a mood stabiliser, within a supervised care setting. That already counts for a great deal, because the opposite doubt was blocking the prescription. What the study does not support: that esketamine reduces switching, that it reduces suicide, that it is safe without a mood stabiliser, that it acts differently in women. Those four statements are hypotheses, and three of them are not even particularly plausible.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ It reassures me about the thing that frightened me, and it does not convince me about the rest. Two thousand patients with no excess of switching, I will take that, it is the figure I was missing. But if you tell me esketamine protects against switching, I answer that it was the patients we chose to treat who were not going to switch. And I am not touching the mood stabiliser. ”
What this means in practice: these data are not enough to rule out the risk of switching, they show only that no excess was detected in selected patients kept on a mood stabiliser. Vigilance still applies at every session, and it applies in full as soon as the mood stabiliser is missing.
What you can do with this on Monday morning. The course of action does not change, but five things become usable straight away.
- Answer precisely the patient who asks whether esketamine might make them switch: in the largest series we are aware of, 2126 matched patients and analyses run out to 365 days, no excess of switching was detected in patients who kept their mood stabiliser. Then say that this is not a randomised trial, and why that changes the reach of the sentence.
- Set the regulatory frame before the clinical discussion. Bipolar depression appears in no approved indication for esketamine. Where the European wording applies, the authorisation covers the treatment-resistant major depressive episode of moderate to severe intensity, in combination with an SSRI or an SNRI, and the rapid reduction of depressive symptoms in a psychiatric emergency, in an episode of moderate to severe intensity and co-administered with an oral antidepressant. Status and wording are decided jurisdiction by jurisdiction, and the labelling in force where you practise is the one that applies: prescribing here is off-label, which brings whatever obligations your own jurisdiction attaches to that, informing the patient, marking it on the prescription and recording the reasoning in the notes.
- Do not lighten the mood stabiliser during a course of esketamine. Every patient in the cohort was on one. The study says nothing about esketamine alone in a patient with bipolar disorder, and can say nothing about it.
- Keep watching for a switch at every session, with a scale to hand, despite the absence of a statistical signal. An average cohort says nothing about the patient in front of you, and the summary of product characteristics explicitly advises caution in the presence or history of mania or bipolar disorder.
- Do not forget what the study was not looking at: blood pressure before and about forty minutes after each administration, monitoring until the patient is clinically stable, and the contraindication in patients for whom a rise in blood pressure or intracranial pressure is a serious risk, in particular aneurysmal vascular disease (intracranial, thoracic or abdominal aorta, peripheral arteries), a history of intracerebral haemorrhage, or a recent cardiovascular event, within six weeks.
One question the study leaves wide open: in a patient with bipolar I disorder and a history of switching under antidepressants, should the threshold of caution be different? The abstract reports neither the split between types I and II nor any history of switching. That is a point of vigilance, not an additional contraindication.
Frequently asked questions
Can esketamine be prescribed to a patient with bipolar disorder?
Not within the approved indication. Where the European wording applies, that indication covers the treatment-resistant major depressive episode of moderate to severe intensity, in combination with an SSRI or an SNRI, and the rapid reduction of depressive symptoms in a psychiatric emergency, in an episode of moderate to severe intensity and co-administered with an oral antidepressant. Bipolar depression is an off-label use, with the obligations your own jurisdiction attaches to it: informing the patient, marking it explicitly on the prescription, recording the reasoning in the notes. Authorisation is decided jurisdiction by jurisdiction, and the labelling in force where you practise is the one that applies. The summary of product characteristics also asks for caution in patients with a presence or history of mania or bipolar disorder.
Does this study prove that esketamine protects against manic switch?
No. It is an observational comparison, and a hazard ratio is not an effect. The reduction observed at 180 and 365 days has a simpler explanation: clinicians do not offer esketamine to patients they judge unstable. A propensity score matches on what is coded, not on clinical judgement.
Does the absence of a significant difference at seven days mean the short-term risk of switching is nil?
No. An absence of significant difference is not proof that there is no effect. Over seven days, coded switches are rare, and the number of events governs the ability to detect a difference. The abstract gives neither the number of events nor any power calculation for this outcome.
Does a hazard ratio of 0.44 justify offering esketamine in a suicidal emergency in bipolar disorder?
It is not enough. This is a composite outcome whose composition is not detailed in the abstract, which speaks only of suicide-related events. One can assume that documented ideation weighs more heavily in it than attempts, but that is only an assumption. In any case, the study does not show a fall in deaths by suicide, which this type of record database captures poorly. A strong early signal in a matched cohort is a reason to study the question, not a protocol.
Should anything be concluded from the result observed in women?
No, not at this stage. A contrast between subgroups counts only if the interaction is tested and reported, which does not appear in the abstract. With two outcomes, five intervals and three subgroup dimensions with no correction for multiple testing, an isolated result of this kind is what chance produces.
Annotated bibliography
Liu TH, Wu JY, Huang PY, Cheng WL, Lai CC. (2025). Risk of manic switch and suicidal outcomes in bipolar depression treated with esketamine: A one-year retrospective cohort study of 2126 patients. European Neuropsychopharmacology, 99, 3-12. DOI 10.1016/j.euroneuro.2025.07.006 · PMID 40774133. Source study analysed here. Retrospective TriNetX cohort, 1:1 propensity score matching, 2126 patients, analyses run out to 365 days. Hospital and university team.
Martinotti G, Dell’Osso B, Di Lorenzo G, Maina G, Bertolino A, Clerici M, et al.; REAL-ESK Study Group. (2023). Treating bipolar depression with esketamine: Safety and effectiveness data from a naturalistic multicentric study on esketamine in bipolar versus unipolar treatment-resistant depression. Bipolar Disorders, 25(3), 233-244. DOI 10.1111/bdi.13296 · PMID 36636839. Italian naturalistic study comparing 35 patients with treatment-resistant bipolar depression to 35 with treatment-resistant unipolar depression, over three months. Depressive improvement in both groups, with no association with an emergent affective switch. Very small sample, no unexposed group, short follow-up: this is the state of the literature that Liu et al. widen, and it shows the gap in scale.
Rodolico A, Cutrufelli P, Di Francesco A, Aguglia A, Catania G, Concerto C, et al. (2024). Efficacy and safety of ketamine and esketamine for unipolar and bipolar depression: an overview of systematic reviews with meta-analysis. Frontiers in Psychiatry, 15, 1325399. DOI 10.3389/fpsyt.2024.1325399 · PMID 38362031. Overview of reviews, 26 systematic reviews and 44 randomised trials, 3316 subjects. It concludes to an apparent efficacy and tolerability, but 23 of the 26 included reviews are of critically low quality on AMSTAR-2, the other three of low quality, and almost none of this work rates the certainty of the evidence. Useful for placing the general fragility of the field, including on its randomised side.
European Medicines Agency. Esketamine nasal spray, product information and summary of product characteristics. Product information, European Medicines Agency, English version read on 15 September 2026. Regulatory source for the indications, contraindications and conditions of administration cited here. Bipolar depression does not figure among its indications. This wording is that of one jurisdiction: authorisation is decided jurisdiction by jurisdiction, and the labelling in force where the reader practises is the one that applies.