Published on 24 September 2026
Esketamine in treatment-resistant depression: do adverse effects depend on the dose?
Frontiers in Pharmacology · 2026; 17: 1792570 · Qu et al.
DOI 10.3389/fphar.2026.1792570
PMID 42292802
Scientific 72
Editorial 79
The essentials
A systematic review with meta-analysis of nine randomized trials, 1,449 adult patients with treatment-resistant depression, stratified the adverse effects of esketamine by dose received: high dose at 56 mg or more by nasal spray, or 0.40 mg/kg intravenously; low dose at 28 mg or less, or 0.20 mg/kg. The authors report nine of eleven adverse event categories significantly more frequent under esketamine; their own results table counts only eight, since somnolence does not reach the threshold. The clearest signal is dissociation, with a relative risk of 10.65 in the high-dose subgroup versus 3.27 in the low-dose subgroup. Nausea reaches a relative risk of 3.72 at high dose and does not reach significance at low dose, which may reflect insufficient power and does not demonstrate an absence of effect. Discontinuation for adverse events is roughly doubled, relative risk 2.22, while all-cause discontinuation does not differ from control. Clinical response is increased, relative risk 1.94, but with 81% heterogeneity and a prediction interval that crosses the null. Two limitations should govern how this is read: the comparison between the high-dose and low-dose subgroups is not a randomized comparison, and these data concern only short-term tolerability.
Context
In consultation, the question is no longer whether to use esketamine in treatment-resistant depression, but how. Two dose steps coexist in practice by nasal spray, 56 mg and 84 mg, and the prescriber has to tell the patient what to expect during the first administrations, decide on the intensity of monitoring, and anticipate reasons for discontinuation. Earlier syntheses mostly pooled adverse effects across all doses, which produces an average frequency of little use at the bedside. The authors position their contribution against this body of work: an analysis built on dose thresholds defined in advance, rather than on pooled categories. They make no claim of priority anywhere, and none is added here.
The study at a glance
| Element | Content |
|---|---|
| Population | Adults aged 18 and older with treatment-resistant depression, without major associated medical illness. |
| DetailResistance is defined as failure of at least two antidepressants given at adequate dose and duration during the current episode. Nine randomized controlled trials, 1,449 patients in total, 30 to 344 patients per trial. Search across six databases, including three Chinese databases, from inception to March 2025. | |
| Intervention | Esketamine, nasal spray 28 to 84 mg, or intravenous route 0.20 to 0.40 mg/kg. |
| DetailDose subgroups are defined a priori: high dose at 56 mg or more, or 0.40 mg/kg; low dose at 28 mg or less, or 0.20 mg/kg. Per the characteristics table, six trials use the nasal route and three the intravenous route. | |
| Comparator | Placebo, background antidepressant, or another medication, depending on the trial. |
| DetailThe publication describes three trials with inert placebo, five trials in which the control group receives a background antidepressant, and one trial in which the comparator is another medication, propofol, given alongside electroconvulsive therapy. The comparator is therefore not homogeneous across trials, which weighs on the interpretation of the pooled estimates. | |
| Outcomes | Eleven categories of adverse events, treatment discontinuation, clinical response or remission. |
| DetailThe eleven categories are nausea, paresthesia, dissociative symptoms, fatigue, somnolence, sleep disturbance, headache, dizziness, dysgeusia, vertigo, and elevated blood pressure. Serious adverse events are analyzed separately. Response is defined as a reduction of at least 50% in the depression score. | |
| Design | Systematic review with meta-analysis of randomized trials, level 1a on the Oxford CEBM scale. |
| DetailRelative risks with 95% confidence intervals, fixed-effects model, random-effects model when I² exceeds 50%, 95% prediction intervals, sensitivity analysis by successive exclusion of one trial, subgroups prespecified by dose, route, and region. Registered protocol, PRISMA 2020, Cochrane risk-of-bias tool in its 2011 version, RevMan 5.4.1. | |
Quality control
| Domain | Verdict |
|---|---|
| Protocol and reporting | Registered, grading missing |
| FindingThe protocol is registered with PROSPERO under number CRD420251024830 and conduct follows PRISMA 2020: this is the work’s formal strength. However, the methods section announces a GRADE assessment of evidence quality for each outcome and the creation of an evidence profile: no GRADE profile and no criterion-by-criterion grading appears in the published article or its supplementary material. | |
| Risk of bias in included trials | Low, except two trials |
| FindingThe 2011 Cochrane tool is applied to the nine trials. On the first six domains, all trials are judged low risk, except Ochs-Ross 2020, rated high risk for missing data. The “other bias” domain is rated high risk for Singh 2016 and unclear for five trials. The authors nonetheless conclude that all included trials were of high quality and that these domains did not affect the results, without substantiating this judgment. | |
| Heterogeneity | High on three outcomes |
| FindingI² at 76% for somnolence, 66% for dysgeusia, and 81% for the efficacy outcome. On these three outcomes, the pooled estimate poorly summarizes trials that do not tell the same story, and the 95% prediction interval crosses the null. | |
| Subgroups | Prespecified |
| FindingThe dose, route, and region subgroups are announced in advance, which rules out opportunistic post hoc splitting. This does not, however, make the comparison between subgroups randomized. | |
| Sensitivity analysis | Conducted and informative |
| FindingSuccessive exclusion of one trial brings I² down from 76% to 42% for somnolence by removing Singh 2016, from 66% to 32% for dysgeusia by removing Daly 2019, a flexible-dose trial, and from 81% to 55% for efficacy by removing Ochs-Ross 2020, the only trial conducted in older adults. Heterogeneity is therefore largely attributable to three identifiable trials. | |
| Route of administration | The publication contradicts itself |
| FindingThe characteristics table lists three trials by intravenous route, Singh 2016, Han 2022, and Zeng 2022; the limitations section mentions only one, Singh 2016. Whichever count is used, the intravenous estimates rest on small samples and very wide intervals, up to 1.14 to 64.23 for elevated blood pressure. | |
| Supplementary material | Incomplete and inconsistent |
| FindingThe methods announce the full search strategy for all six databases searched: the supplementary material provides only the PubMed strategy, duplicated, and that string contains a MeSH descriptor that does not exist in the form cited, which would make it inoperative as written. Another supplementary table indicates that Singh 2016 was excluded from the analysis, although this trial is among the nine included trials and appears in the forest plots. | |
| Funding and conflicts of interest | Institutional, none declared |
| FindingFunding is institutional: Sichuan Medical Association, number 2025RF27, and Medical Association of Jintang County, number 2024001. The authors declare no commercial or financial relationship. The publication, however, reports neither the sponsor nor the conflicts of interest of the trials it includes: this point can therefore be neither confirmed nor ruled out on the basis of this publication. | |
| Data horizon | Short-term tolerability |
| FindingFollow-up durations range from 4 hours to 24 weeks, with most trials covering the acute phase, typically 4 weeks. Long-term data are not reported, which allows no conclusion about what happens afterward. | |
| Publication venue | General pharmacology journal |
| FindingFrontiers in Pharmacology is not a psychiatry journal. Its peer review is open: the editor and the two reviewers are named in the article. The published text contains several internal discrepancies between its abstract, its tables, and its discussion, noted further below. This is one element of appraisal among others, not a reason for rejection. | |
Results
Relative risk of dissociation in the high-dose subgroup, at 56 mg or more or 0.40 mg/kg, 95% confidence interval 6.20 to 18.29, versus 3.27, interval 1.54 to 6.93, in the low-dose subgroup. Both subgroups are compared with their own respective control, not with each other.
| Result | Value |
|---|---|
| Adverse events increased | 9 of 11 categories per the text, 8 per the table |
| ReadingThe abstract and the discussion report nine of eleven categories. The results table gives only eight below the 0.05 threshold: somnolence appears there at 2.00, interval 0.93 to 4.28, p = 0.07. The other two categories without a significant difference are sleep disturbance and dysgeusia. | |
| Most increased effects, all doses | Dissociation 8.17; vertigo 6.23; paresthesia 5.51 |
| ReadingNext come dizziness at 3.69, nausea at 3.23, elevated blood pressure at 2.88, fatigue at 2.09, and headache at 1.65. The 95% prediction interval for dissociation, 2.54 to 18.42, remains above the null: it is the only outcome whose signal holds across every context considered. | |
| Dissociation, high dose | RR = 10.65 (6.20 to 18.29) |
| ReadingThis is the strongest signal in the synthesis, and the one that should shape what is told to the patient. | |
| Dissociation, low dose | RR = 3.27 (1.54 to 6.93) |
| ReadingThe risk remains markedly increased at low dose. The direction of the gradient is consistent; its exact magnitude is less so, for lack of a direct comparison. | |
| Nausea, high dose | RR = 3.72 (2.60 to 5.32) |
| ReadingA significant result in the high-dose subgroup. | |
| Nausea, low dose | RR = 1.69 (0.81 to 3.55), p = 0.16 |
| ReadingNot significant does not mean absent. The point estimate points toward an increase, the interval is wide, and a subgroup drawn from nine trials lacks the power to settle the question. | |
| Headache by dose | 2.88 at low dose versus 1.41 at high dose |
| ReadingNot every adverse effect follows the same gradient. For headache and for dizziness, the relative risk is higher in the low-dose subgroup; for elevated blood pressure it is comparable in both, 2.27 and 2.41. The dose gradient holds for dissociation, paresthesia, and nausea, not for the tolerability profile as a whole. | |
| Discontinuation for adverse events | RR = 2.22 (1.11 to 4.45), p = 0.025 |
| ReadingThis is the outcome closest to the patient’s experience, since it reflects tolerability judged unacceptable by the patient or the physician. The interval remains wide: doubling is the order of magnitude, not a precise value. | |
| All-cause discontinuation | RR = 1.15 (0.85 to 1.56), p = 0.36 |
| ReadingAll-cause discontinuation does not differ from control. The excess risk therefore concerns the tolerability reason for stopping, not the total number of trial exits. The text of the publication also mentions an overall p of 0.001 for treatment discontinuation, a value that none of its tables reproduces. | |
| Clinical response | RR = 1.94 (1.21 to 3.10) |
| ReadingBenefit is not the main subject of the synthesis. It is affected by 81% heterogeneity and a 95% prediction interval of 0.51 to 4.30, which crosses the null: how much benefit to expect in a given context remains highly uncertain. The publication labels this outcome response in its abstract and remission in its discussion, where it also cites a relative risk of 1.35 that appears in none of its tables. | |
| Serious adverse events | No significant difference, p = 0.14 |
| ReadingNo count, relative risk, or interval is reported for this outcome. With nine trials and a rare event, the absence of a significant difference does not demonstrate the absence of excess risk. | |
| Regional subgroup | Differences in both directions |
| ReadingNausea, somnolence, and headache are more frequent in international trials; elevated blood pressure is more frequent in Chinese trials, 6.67 versus 2.67. For efficacy, the interaction test between regions is not significant, p = 0.208. The comparison concerns sets of trials that also differ in their populations, comparators, and methods of collecting adverse events: it is an exploratory observation. The text of the publication also swaps the values for the two regions for dizziness relative to its own table. | |
Critical appraisal
| Domain | Judgment |
|---|---|
| Nature of the dose comparison | Indirect comparison |
| FindingThe two dose subgroups do not come from a randomization pitting them against each other. They pool different trials, with different populations, comparators, and modes of data collection. The observed gradient is consistent, but the 3.3-fold ratio between the two dissociation relative risks should not be handled as a measured dose effect. | |
| Power of the subgroups | Small subgroups |
| FindingThe low-dose subgroup is necessarily smaller than the whole. The non-significant results that come out of it, chief among them nausea, may reflect insufficient power and do not demonstrate an absence of effect. | |
| Heterogeneity | Three outcomes involved |
| FindingI² at 76% for somnolence, 66% for dysgeusia, 81% for efficacy. On these outcomes, the pooled value should be considered uninformative. The sensitivity analysis identifies a responsible trial in each of the three cases, which is to the work’s credit. | |
| Comparison of administration routes | Exploratory |
| FindingThree trials by intravenous route per the characteristics table, only one per the limitations section. The corresponding estimates rest on very small samples and intervals that can range from 1.14 to 64.23. No conclusion about the comparative tolerability of the nasal and intravenous routes can be drawn from this work, in either direction. | |
| Internal consistency of the publication | Several discrepancies |
| FindingThe number of significant categories, the adverse-event table count of 24 of 807 against 246 in the corresponding forest plot, the 81% heterogeneity attributed at times to response and at times to remission, a relative risk of 1.35 cited without support, the swapped regional values for dizziness, an overall p of 0.001 for discontinuation absent from the tables, and the contradictory status of Singh 2016 in the supplementary material: the publication contradicts itself on at least seven points. Each is minor on its own; their accumulation calls for relying only on values confirmed by the forest plots. | |
| Conflicts of interest of included trials | Not reported |
| FindingThe synthesis’s authors declare no commercial or financial relationship, and its funding is institutional. The publication, however, reports neither the sponsor nor the conflicts of interest of the trials it pools. This silence allows neither confirming nor ruling out an influence on the collection and classification of adverse events: the point remains open and would need checking trial by trial against the original publications. | |
| Time horizon | Short term only |
| FindingLong-term data are not reported. Absence of data should not be written as absence of signal: the long-term question remains open, particularly for cognitive, urinary, and misuse-related effects. The authors acknowledge this explicitly among their limitations. | |
| Dosing guidance | Unsupported recommendation |
| FindingThe authors recommend starting at 56 mg to establish tolerability before moving to 84 mg. This recommendation does not follow from their analysis: the synthesis compares two dose classes on tolerability, without comparing titration strategies or weighing tolerability against efficacy dose by dose. The 56 mg threshold here separates two classes defined in advance; it is not derived from a result. | |
Level of evidence
The design sits at level 1a on the Oxford CEBM scale: systematic review with meta-analysis of randomized trials, registered protocol, conduct reported under PRISMA 2020. The GRADE grading announced in the methods was not published, which deprives the reader of the authors’ own confidence judgment, criterion by criterion. What follows is therefore the review’s own appraisal. Confidence is high on one point: under esketamine, adverse effects are more frequent than under control, and dissociation is the most consistent of them, its prediction interval staying above the null. It is moderate on the dose gradient, whose direction is consistent for dissociation, paresthesia, and nausea, but whose magnitude rests on a comparison between subgroups not protected by randomization, and which does not hold for either headache or blood pressure. It is low on three points: the estimates affected by high heterogeneity, somnolence, dysgeusia, and efficacy; the comparative tolerability of the two administration routes, which the available material does not allow addressing; and everything concerning the long term, absent from the work. The publication venue, a general pharmacology journal, and the number of internal discrepancies found between the abstract, the tables, and the discussion are two further elements of appraisal, neither disqualifying on its own.
The colleague test
What an experienced colleague might say about this study in two minutes, between two consultations.
“That dissociation rises with the dose, honestly, tells me nothing new. What interests me is the number: ten times more at high dose, three times more at low dose. Except these aren’t two randomized arms pitted against each other, they’re two bundles of trials looked at from a distance, so I’m not taking the ratio between them at face value. And nausea being non-significant at low dose, with nine trials, mostly means we didn’t have enough to see it. What I take away for tomorrow morning isn’t a dose, it’s what I tell the patient before the first administration, and the fact that discontinuations for poor tolerability are roughly doubled while the total number of dropouts doesn’t budge.”
Translated for practice: this work changes how you inform and monitor, not how you dose. It gives orders of magnitude for informed consent, and it is a reminder that the most likely reason for stopping is tolerability, not lack of effect.
What you can do with this
- What this tells you. Dissociation is the adverse effect whose relative frequency varies most with the dose received, by a factor of roughly 10 versus control in the high-dose subgroup against roughly 3 in the low-dose subgroup. The direction of the gradient is solid, its exact magnitude less so, and this gradient does not hold for every adverse effect: headache and dizziness are, if anything, more frequent at low dose, and elevated blood pressure is comparable at both dose steps.
- What you can tell the patient. That adverse effects are frequent and expected, that dissociation is the most characteristic of them, and that it is plausible it is more marked at higher dose. That an effect not found to be significant at low dose is not an effect ruled out: nausea is the example here.
- What to monitor. Discontinuation for adverse effects is roughly doubled, relative risk 2.22, while the total number of discontinuations does not differ from control. Plan explicitly, from initiation, what will trigger a dose reduction or discontinuation, and document it in the chart. Monitor blood pressure regardless of dose step, since its elevation does not depend on dose in this analysis.
- What you cannot conclude. This synthesis does not provide a starting dose. The authors recommend starting at 56 mg before moving to 84 mg, but their analysis compares no titration schedule and does not weigh tolerability against efficacy dose by dose. Nor does it allow comparing the tolerability of the nasal and intravenous routes, say anything about the long term, or rule out an excess risk of serious adverse effects, for lack of reported figures on that outcome.
- Regulatory context to check. Authorization, marketing, and collective reimbursement decisions are three distinct matters, and administrative or regulatory conditions on how a treatment may be prescribed, dispensed, and monitored are set independently of what any single meta-analysis shows. In France, for instance, nasal esketamine is subject to specific prescribing, dispensing, and monitoring requirements, and intravenous use falls under a different framework. Check the rules in force in your own jurisdiction at the time of prescribing.
Frequently asked questions
Should this be read as recommending the lowest starting dose?
Not on the basis of this work. The authors do recommend favoring 56 mg before escalating to 84 mg, but their analysis does not support this recommendation: it compares two dose classes on tolerability, the high class starting at 56 mg, without evaluating any initiation or titration strategy and without weighing tolerability against efficacy dose by dose. The choice of starting dose depends on the product’s conditions of use and on the clinical situation, not on this meta-analysis.
Is nausea absent at low dose?
No. The result reported at low dose is non-significant, with a point estimate pointing toward an increase, relative risk 1.69, interval 0.81 to 3.55. With nine trials in total and an even smaller low-dose subgroup, the lack of significance may reflect insufficient power. Absence of evidence is not evidence of absence.
Is the intravenous route better or worse tolerated than the nasal route?
This work cannot answer that. The publication itself is inconsistent about how many trials are concerned: its characteristics table lists three by intravenous route, its limitations section mentions only one. Either way, the samples are too small and the intervals too wide, up to 1.14 to 64.23 for elevated blood pressure, for the comparison between the two routes to be interpretable in either direction.
What should be made of the difference between international and Chinese trials?
That the differences run in both directions. Nausea, somnolence, and headache are reported more often in international trials; elevated blood pressure is more frequent in Chinese trials, relative risk 6.67 versus 2.67. On efficacy, the interaction test between regions is not significant, p = 0.208. This stratification compares sets of trials that also differ in their populations, comparators, and methods of collecting adverse events. It is an exploratory observation that does not transpose to individual patients.
Do these results say anything about long-term tolerability?
No. Follow-up durations range from 4 hours to 24 weeks and most trials cover the acute phase; long-term data are not reported in this work. That does not mean no long-term signal exists, only that the question is not addressed here. The authors acknowledge this among their limitations.
Annotated bibliography
Source study. Qu Y, Li S, Tian L, Tian X, Wu Y. Dose-dependent adverse events of esketamine in treatment-resistant depression: a systematic review and meta-analysis of randomized controlled trials. Frontiers in Pharmacology, 2026; 17: 1792570. DOI 10.3389/fphar.2026.1792570 PMID 42292802
What it contributes: a stratification of adverse effects by dose class, drawn from nine trials and 1,449 patients, on subgroups announced in advance, with prediction intervals and sensitivity analysis. Its limitations: comparison between subgroups not randomized, high heterogeneity on three outcomes, very small samples by intravenous route, horizon limited to the short term, GRADE grading announced but not published, and several internal discrepancies between the abstract, the tables, and the discussion.
Most structuring included trials. Fedgchin M et al. Efficacy and safety of fixed-dose esketamine nasal spray combined with a new oral antidepressant in treatment-resistant depression (TRANSFORM-1). Int J Neuropsychopharmacol, 2019; 22: 616 to 630. 10.1093/ijnp/pyz039
What it contributes: the largest single contingent in the synthesis, 344 patients, and a fixed-dose parallel-group design, which makes it a pillar of the dose stratification. Its limitation: the control group receives a newly initiated antidepressant, which raises the comparison incidence of adverse effects.
Ochs-Ross R et al. Efficacy and safety of esketamine nasal spray plus an oral antidepressant in elderly patients with treatment-resistant depression (TRANSFORM-3). Am J Geriatr Psychiatry, 2020; 28: 121 to 141. 10.1016/j.jagp.2019.10.008
What it contributes: the only trial conducted in older adults, and the trial whose exclusion brings the heterogeneity of the efficacy outcome down from 81% to 55%. Its limitation: it is also the only trial rated at high risk of bias for missing data in the authors’ assessment.
Singh JB et al. Intravenous esketamine in adult treatment-resistant depression: a double-blind, double-randomization, placebo-controlled study. Biol Psychiatry, 2016; 80: 424 to 431. 10.1016/j.biopsych.2015.10.018
What it contributes: one of the synthesis’s intravenous-route trials, and the identified source of heterogeneity for somnolence. Its limitations: 30 patients, a double sequential randomization design, a high risk-of-bias judgment on the “other bias” domain, and a contradictory status in the supplementary material, which lists it as excluded although it appears in the analysis.
To put the tolerability signal in context. Taillefer de Laportalière T et al. Reporting of harms in clinical trials of esketamine in depression: a systematic review. Psychol Med, 2023; 53: 4305 to 4315. 10.1017/S0033291723001058
What it contributes: an assessment of how adverse effects are themselves reported in esketamine trials, a question that precedes any meta-analysis of tolerability. Its limitation: it addresses the quality of harms reporting, not the frequency of events.
What was consulted. References verified on August 12, 2026 against the published version and its supplementary material, four files in total: the PubMed search strategy, provided in two identical copies, the table of reasons for discontinuation, and the list of candidate trials with reasons for exclusion. The DOI, volume, article number, and PMID are taken from the published version itself. Context references are transcribed from the source publication’s bibliography and have not been independently reverified against the original articles. The article underwent independent double reading.
Editorial collections
Tags
Verified on August 12, 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on September 23, 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.