Published on 16 September 2026
Ketamine versus midazolam in acute suicidality: what is left against an active comparator?
In brief
Ten randomised controlled trials, 649 participants by the authors’ count, all comparing ketamine not with an inert solution but with midazolam. Risk of bias is judged low in every domain for all ten trials. Yet no estimate covers the whole body of trials: the MADRS total score rests on three trials and 84 patients, the MADRS suicidal ideation item on four trials and 178 patients, the Beck Scale for Suicide Ideation on three trials and 118 patients. On the MADRS total, ketamine does better than midazolam by 6.23 points, with no detectable heterogeneity and moderate GRADE certainty, downgraded for imprecision. On the suicidal ideation item, the difference is 1.23 points on a scale that has six, with heterogeneity of 63%. On the Beck scale, the initial difference of 4.30 points does not survive the removal of a single trial. The only result graded high certainty is not an efficacy result: it is derealisation, 34.1% on ketamine against 6.5% on midazolam. Follow-up runs from 240 minutes to fourteen days according to the methods section, and up to about one month according to the discussion and the supplementary table of characteristics.
The context
An objection earlier trials could not answer
For about ten years, intravenous ketamine at 0.5 mg/kg has held a particular place in psychiatric emergency departments and short-stay inpatient units: a drug that acts within hours where antidepressants take weeks, in patients for whom the coming days are precisely what is feared. The objection fitted in one sentence, and every psychiatrist has voiced it or heard it: a patient who is sedated, anxiolysed, with consciousness altered for forty minutes, will report fewer suicidal thoughts that same evening, and that says nothing about an effect specific to the drug.
Earlier meta-analyses could not answer it, because they pooled heterogeneous comparators, saline and midazolam alike. A patient who receives ketamine knows very quickly that something has been given, and so does the rater, and blinding holds poorly. This systematic review changes the comparator: it retains only trials run against midazolam, a short-acting benzodiazepine that produces sedation, anxiolysis and altered consciousness without having an established antidepressant effect or sharing the presumed mechanism of ketamine. That methodological choice is what makes the work worth reading, and it is also what makes its results less spectacular than those the topic has accustomed us to.
The mechanism
What the comparator isolates, and what it does not
The reasoning behind the design is subtractive. If ketamine did nothing but sedate, midazolam should do as well, and the gap between the two arms should tend towards zero. A persistent gap in favour of ketamine therefore supports the hypothesis of an effect that cannot be reduced to the non-specific effects the comparator reproduces.
This reading has two limits that need stating from the outset. First, the review measures clinical scores, not mechanisms: no biological marker, no imaging, no direct index of the glutamatergic pathway is among the outcomes. What is tested is the specificity of the clinical effect, not its explanation. Second, midazolam controls for sedation and anxiolysis, but it does not reproduce the dissociative experience of ketamine, and the safety result reported below shows how common that experience is on ketamine. The review does not address this question head on for the trials it includes: it raises unblinding and expectancy effects in relation to the earlier literature, and presents the use of midazolam as the way to reduce them, but the blinding domain of the risk of bias tool is rated low risk for all ten trials without the residual risk linked to dissociative effects being discussed.
The study at a glance
Population, intervention, comparator, outcomes
| Population | |
| Adults with acute suicidality or suicidal ideation, regardless of psychiatric diagnosis. The trials were run in psychiatric inpatient units, outpatient clinics and tertiary care hospitals, in North America, Europe and Asia, between 2013 and 2025. A stated total of 649 participants for ten trials. Adding up the sample sizes of the ten trials in supplementary table 1 gives 523 participants, 297 on ketamine and 226 on midazolam: the publication does not explain this gap of 126 patients. | |
| Intervention | |
| Ketamine at a subanaesthetic dose, most often 0.5 mg/kg as an intravenous infusion over about forty minutes. One trial gives racemic ketamine orally, another varies the dose from 0.1 to 1.0 mg/kg. Most trials give a single infusion, the others repeat administration over one to two weeks, up to six infusions in one trial. | |
| Comparator | |
| Midazolam, an active psychoactive comparator, given by the route matching that of ketamine, at an anxiolytic dose of 0.045 to 0.05 mg/kg intravenously in the trials that specify it. It is chosen because it reproduces the non-specific effects of sedation, anxiolysis and altered consciousness, while being neither an antidepressant nor an established treatment for suicidality. | |
| Outcomes | |
| Primary: suicidal ideation measured by the MADRS suicidal ideation item, scored 0 to 6, and by the Beck Scale for Suicide Ideation, described in the supplementary material as having 21 items scored 0 to 2. Secondary: depression severity measured by the MADRS total score, ten items scored 0 to 6. Safety: nausea and vomiting, emotional disturbances, derealisation, dizziness. | |
| Design | |
| Systematic review with meta-analysis of ten randomised controlled trials, searching PubMed, Embase and Cochrane Central without date restriction up to July 2025. Protocol registered with PROSPERO under number CRD420251110292. Reporting in line with PRISMA 2020 and the Cochrane Handbook. Risk of bias assessed with the Cochrane Risk of Bias Tool, in its five-domain version. Certainty of evidence by GRADE. Random-effects model, RevMan 5.4 software. |
Quality control
What the conduct of the review guarantees, and what it leaves open
| Item checked | Judgement |
|---|---|
| Prospective protocol registration | Verified |
| FindingPROSPERO number CRD420251110292, which makes it possible to compare the announced outcomes with the reported outcomes. | |
| Reporting standard | PRISMA 2020 |
| FindingThe review states that it follows the PRISMA 2020 checklist and the Cochrane Handbook for study selection and data reporting. The flow diagram is given in supplementary figure S1: 1349 records, 1020 after removal of 329 duplicates, 940 excluded at screening, 80 full texts read, 10 trials retained. | |
| Risk of bias of the included trials | Low in all domains |
| FindingThe tool used is the Cochrane Risk of Bias Tool in its five-domain version: sequence generation, allocation concealment, blinding of participants and outcome assessors, incomplete outcome data, selective reporting. It concludes to low risk for all ten trials. The plots are given in supplementary figure S2. The review nowhere mentions the RoB 2 tool and does not detail its judgements domain by domain in the body of the text. | |
| Study selection and data extraction | Independent, disagreements arbitrated |
| FindingTitles and abstracts screened by two independent investigators in Rayyan, disagreements settled by a third. Extraction carried out independently by four investigators, discrepancies resolved by discussion and then consensus, data cross-checked for consistency. | |
| Pooling model | Random effects |
| FindingA prudent choice, which does not assume a single effect common to all trials and widens the intervals as heterogeneity rises. | |
| Sensitivity analysis | Conditional, two results flip |
| FindingLeave-one-out removal of trials was planned in the statistical protocol only in case of heterogeneity above 50%. It was therefore not applied to the MADRS total score or to derealisation. Where it was carried out, removing Hochschild 2021 makes the Beck scale result lose its significance, and removing Tungpin 2023 does the same for dizziness. | |
| Participant count | Recount does not match |
| FindingThe stated total is 649 participants. Adding up the ten trials in supplementary table 1 gives 523, that is 297 on ketamine and 226 on midazolam. The same table puts trial size between 13 and 84 participants, whereas the publication writes that it ranges from 20 to 84. Neither discrepancy is explained. | |
| Size of the analyses | From 84 to 236 patients |
| FindingNo estimate covers the ten trials together. The GRADE table gives 178 participants and 4 trials for the MADRS suicidal ideation item, 118 and 3 for the Beck scale, 84 and 3 for the MADRS total score, 236 and 3 for derealisation. The results are to be read on these numbers, not on 649. | |
| Length of follow-up | Reported inconsistently |
| FindingThe study characteristics section places follow-up between 240 minutes and fourteen days after infusion. The discussion writes that outcomes were assessed at times ranging from 24 hours to about one month, and the limitations section gives follow-up of fourteen days or less for most trials. Supplementary table 2 gives an assessment time of one month for three trials. In every case, nothing informs beyond one month. | |
| Publication bias | Rated without a reported test |
| FindingThe GRADE table carries the entry none in the publication bias column for all seven outcomes, and the discussion states that “few concerns were raised regarding indirectness or publication bias”. No funnel plot and no statistical test are reported. With three to four trials per outcome, such a search would in any case have very little power to detect anything: the absence of concern here is not an absence of signal, it is an absence of analysis. | |
| Internal consistency of the publication | Two contradictions found |
| FindingThe abstract lists nausea among the adverse events that were “much more frequent in the ketamine group”, whereas the results section and the GRADE table show the reverse. The section on the Beck scale states that “The overall effect was not affected by sensitivity analysis”, whereas the p value goes from 0.02 to 0.08 and the confidence interval crosses zero in the GRADE table. | |
| Funding and competing interests | None declared |
| FindingThe funding section reads “The authors have nothing to report.” and the conflicts of interest section “The authors declare no conflicts of interest.” The data are said to be available from the corresponding author on reasonable request. | |
The findings
One clear outcome with few patients, one fragile outcome, one safety signal
| Outcome | Reported value |
|---|---|
| MADRS total score, 3 trials, 84 patients | Mean difference −6.23 (95% CI −10.37 to −2.08), p = 0.003, I² = 0%. Moderate GRADE certainty, downgraded for imprecision. |
| ReadingThe widest difference in magnitude, on the thinnest base: 84 patients spread over three trials, the least populated outcome in the whole review. Zero heterogeneity does not mean the trials agree perfectly, it means that with three trials no heterogeneity can be detected, which is a low-power situation. The authors’ own footnote acknowledges it without hedging: a very wide confidence interval, and too few patients to produce a precise estimate. | |
| MADRS suicidal ideation item, 4 trials, 178 patients | Mean difference −1.23 (95% CI −2.14 to −0.32), p = 0.008, I² = 63%. After removal of Hochschild 2021: −0.86 (95% CI −1.42 to −0.29), p = 0.003, I² = 27%. Moderate GRADE certainty, downgraded for inconsistency. |
| ReadingSignificant before and after sensitivity analysis, but the magnitude stays modest on a scale of only six points, and it is the estimate after removal, −0.86 points, that the GRADE table retains. The initial heterogeneity of 63% is mostly due to the Hochschild 2021 trial, whose removal brings it down to 27%. | |
| Beck Scale for Suicide Ideation, 3 trials, 118 patients | Mean difference −4.30 (95% CI −8.01 to −0.59), p = 0.02, I² = 64%. After removal of Hochschild 2021: −2.44, p = 0.08, I² = 0%, with an interval that crosses zero in the GRADE table, from −5.15 to +0.28. Moderate GRADE certainty, downgraded for inconsistency. |
| ReadingRemoving a single trial out of three pushes the p value above the conventional threshold. No conclusion of efficacy of ketamine on this outcome can therefore be drawn, and no conclusion of no effect either: a result that does not survive a sensitivity analysis is an undetermined result, not a negative one. The publication nevertheless writes that “The overall effect was not affected by sensitivity analysis”, which its own figures contradict. The body of the text also prints the interval as −5.15 to −0.28, which is incompatible with a p value of 0.08; supplementary table S3 prints −5.15 to +0.28. | |
| Derealisation, 3 trials, 236 patients | 44 cases out of 129 on ketamine (34.1%) against 7 out of 107 on midazolam (6.5%). Odds ratio 11.71 (95% CI 4.62 to 29.64), p = 0.00001, I² = 0%. Absolute difference of 385 more cases per 1000 (95% CI 179 to 609). High GRADE certainty. |
| ReadingThe only result in the review graded high certainty, and it concerns tolerability, not efficacy. Heterogeneity is zero and the sample is the largest in the whole review. This is the most solid piece of information the work gives the clinician, and it is better expressed in frequencies than as an odds ratio: about one patient in three on ketamine, against one in fifteen on midazolam. | |
| Emotional disturbances, 3 trials, 184 patients | 16 cases out of 93 on ketamine (17.2%) against 0 out of 91 on midazolam. Odds ratio 12.30 (95% CI 2.20 to 68.74), p = 0.004, I² = 0%. Moderate GRADE certainty, downgraded for imprecision. |
| ReadingThe complete absence of events in the midazolam arm makes the odds ratio very unstable, as shown by an interval running from 2.20 to 68.74. The direction is hard to dispute, the size cannot be read. The absolute effects column of the GRADE table in fact shows zero per 1000, an artefact of the zero in the event count. | |
| Dizziness, 3 trials, 176 patients | Before sensitivity analysis: odds ratio 3.23 (95% CI 1.20 to 8.68), p = 0.02, I² = 57%. After removal of Tungpin 2023: 1.97 (95% CI 0.95 to 4.08), p = 0.07. Retained rates: 34 cases out of 99 on ketamine (34.3%) against 15 out of 77 on midazolam (19.5%). Moderate GRADE certainty, downgraded for inconsistency. |
| ReadingThe estimate the GRADE table retains does not reach the conventional threshold and its interval crosses one. The publication nevertheless concludes that dizziness is increased on ketamine, which this estimate does not support. The observed direction does point that way; the claim goes beyond the precision available. | |
| Nausea and vomiting, 4 trials, 208 patients | Before sensitivity analysis: odds ratio 0.78 (95% CI 0.23 to 2.63), p = 0.69, I² = 59%. After removal of Murrough 2013: 0.39 (95% CI 0.17 to 0.87), p = 0.02, I² = 0%. Retained rates: 11 cases out of 105 on ketamine (10.5%) against 23 out of 103 on midazolam (22.3%), that is 122 fewer cases per 1000. Moderate GRADE certainty, downgraded for inconsistency. |
| ReadingThe direction is the opposite of what the abstract of the publication states, since it lists nausea among the adverse events much more frequent on ketamine. The results and the GRADE table give a lower frequency on ketamine. The contradiction is internal to the publication, and it is the abstract that is at fault. A reader who stopped at the abstract would take away the opposite of what the data show. | |
Critical appraisal
Domain by domain
| Domain | Judgement |
|---|---|
| Choice of comparator | High methodological standard |
| FindingRestricting inclusion to trials against midazolam is the strong point of the work. The symmetrical consequence has to be drawn: midazolam is an active control, not a placebo. A difference in favour of ketamine does not mean that midazolam is inert, and the review does not claim to measure that. | |
| Internal validity of the included trials | Low risk of bias |
| FindingNo problematic domain in any of the ten trials, according to the five-domain Cochrane Risk of Bias Tool. This is the foundation for the moderate to high certainty assigned to the various outcomes. The judgement is, however, given as a block, with no detail by trial and by domain in the body of the text. | |
| Numerical base of the estimates | Three to four trials per outcome |
| FindingThis is the most underestimated limitation of the work. The announced body of ten trials is found in no estimate: each outcome rests on three or four trials and on 84 to 236 patients. The discussion concedes as much, writing that “The evidence base for ketamine versus midazolam remains modest, and several pooled estimates necessarily include small numbers of studies after sensitivity analyses”. It also points out that one of the included trials, a midazolam-controlled inpatient pilot trial, “was terminated early and therefore contributes limited precision”: it is the smallest in the corpus, 13 patients in the supplementary table. | |
| Consistency across trials | Varies by outcome |
| FindingHeterogeneity goes from 0% on the MADRS total score and on derealisation to 63% on the suicidal ideation item and 64% on the Beck scale. In other words, the trials agree on the overall reduction in depressive symptoms and diverge on suicidal ideation itself, which is nonetheless the reason for asking the question. It should be added that an I² of 0% computed on three trials does not demonstrate agreement, it records that agreement cannot be faulted. | |
| Robustness to sensitivity analyses | Two outcomes do not hold |
| FindingThe dependence of the Beck scale result on keeping a single trial is the most important limitation for the clinician, because it concerns the scale specifically built to measure suicidal ideation. Dizziness meets the same fate after removal of Tungpin 2023. | |
| Internal consistency of the report | Abstract wrong on nausea |
| FindingThe abstract contradicts the results on the direction of the effect for nausea, and the section on the Beck scale contradicts its own figures. The participant count cannot be rebuilt from the supplementary table. These flaws do not call the estimates themselves into question, but they require reading the body of the text and the GRADE table rather than the abstract. | |
| External validity | Hospital and outpatient clinics, short follow-up |
| FindingA population recruited in psychiatric inpatient units, outpatient clinics and tertiary centres, administration overwhelmingly intravenous and monitored, follow-up extending at most to about one month. Nothing transfers to office-based practice, and nothing informs on the medium term. The presence of one oral trial and one variable-dose trial also introduces clinical heterogeneity into a corpus described as intravenous at a fixed dose. | |
| Fit between claim and evidence | Wording somewhat broad |
| FindingThe GRADE grading is presented honestly, outcome by outcome, and the discussion specifies that the review is intended as “a rigorous, comparator-specific synthesis rather than a claim of a novel efficacy signal”. The statement that ketamine is “much more effective” than midazolam, used in the abstract and the conclusion, nevertheless needs to be brought back to what the figures carry: 1.23 points out of six for suicidal ideation, reduced to 0.86 after sensitivity analysis, and a Beck scale result that does not withstand the removal of one trial. | |
| Independence | No declared interests |
| FindingNo funding and no competing interests declared, a public protocol, peer review history made available online by the publisher. On the standing of the journal, the assessment that follows is the editors’ own and does not commit the source: it is a sound general journal, without being a top-tier journal in the field. | |
Two inference risks remain open. The first concerns publication bias: the GRADE table rates it as absent for all seven outcomes, but no funnel plot and no statistical test are reported, and with three to four trials per outcome such a search would in any case lack power. So this is not an absence of signal, it is an absence of analysis. The second concerns the nature of the outcomes: suicidal ideation scales are intermediate outcomes, not suicidal events. A reduction in score, however clear, does not translate mechanically into a reduction in suicidal acts, and none of the figures presented here concerns the latter.
Level of evidence
Where confidence is high, and where it is not
The level of evidence attached to the design is the highest on the Centre for Evidence-Based Medicine scale, level 1a, since this is a meta-analysis of randomised trials. This classification is the editors’ own; the publication does not use it. It is not the same as the certainty of each result, and that is the whole point of GRADE applied outcome by outcome rather than as a block.
Certainty is high on a single point, the occurrence of derealisation on ketamine, across three trials and 236 patients. It is moderate for the MADRS total score, downgraded for imprecision, and moderate for the MADRS suicidal ideation item as for the Beck scale, both downgraded for inconsistency. These downgrades do not mean the same thing: in one case the signal is consistent but the sample is very small, 84 patients, in the other the trials do not agree. On the Beck scale, the review does not provide a result that can be relied on, since it does not survive the removal of one trial. The four safety outcomes are rated moderate, with the exception of derealisation.
What is demonstrated, sticking to what is measured: derealisation is far more frequent on ketamine than on midazolam, about one patient in three against one in fifteen. What is suggested, on a narrow numerical base and with heterogeneity that calls for caution: an advantage of ketamine on depressive symptoms measured by the MADRS total, and an advantage of modest size on suicidal ideation measured by the dedicated item. What belongs to interpretation rather than measurement: the idea that this advantage reflects a specific anti-suicidal effect rather than a consequence of overall mood improvement. The review does not separate these two pathways, and its own text acknowledges that “decreases in suicidal thoughts were observed in parallel with reductions in depression severity”.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ In a short-stay unit, this fits with what we already do with ketamine at 0.5 mg/kg in acute suicidality, and above all it answers the objection we hear every time, that it is just sedation. Against midazolam, something is left. That said, when you look closely, the MADRS total figure rests on three trials and 84 patients, and I would not tell a trainee that it works on the Beck scale: it does not hold up in the sensitivity analysis. The information on derealisation, on the other hand, I always give beforehand, not afterwards. ”
What this means in practice: the argument that it is merely a sedative effect no longer holds to the same degree, but the size of the advantage on suicidal ideation itself remains modest, the measured benefit bears first on overall depressive symptoms, and it rests on far smaller samples than the figure of 649 participants suggests. The only high-certainty statement that can be repeated as it stands to a patient concerns the frequency of derealisation.
What you can do with this
To take into the consulting room or on call
An answer to an objection, not a change of protocol. Faced with the colleague who explains that ketamine does nothing but sedate, these ten trials against midazolam provide part of an answer that meta-analyses with mixed comparators could not give. The difference persists when the comparator reproduces sedation, anxiolysis and altered consciousness.
- Inform about derealisation, before administration. It is the only high-certainty result in the review, across 236 patients and with no heterogeneity between trials: about 34% on ketamine against 6.5% on midazolam. A patient who knows that an experience of derealisation is common and transient lives through it differently from one who discovers it. This information is part of informed consent, in the same way as haemodynamic monitoring.
- Look at the sample size of the outcome, not that of the corpus. The 649 participants announced carry no estimate. The figure of 6.23 points on the MADRS total rests on three trials and 84 patients; adding up the sample sizes of the ten trials in the supplementary table gives 523, not 649.
- Do not turn non-significance into proof of absence. The Beck scale result does not survive the sensitivity analysis: that means undetermined, not ineffective. The symmetry applies to the comparator: midazolam is an active control, and nothing here allows one to say that it has no effect of its own in these patients.
- Know what the headline figure measures. The 6.23 points concern the MADRS total, that is, depression, not suicide. On the suicidal ideation item, the difference is 1.23 points out of six, reduced to 0.86 after sensitivity analysis. Ideation scales remain intermediate outcomes: none of the results presented concerns suicidal behaviour.
- Do not rely on the abstract for adverse events. The abstract announces more frequent nausea on ketamine, the results and the GRADE table give the reverse, 10.5% against 22.3%. For dizziness, the estimate retained after sensitivity analysis does not reach the conventional threshold, contrary to what the text asserts.
- Limit extrapolation in time and place. Follow-up extends at most to about one month, with most trials stopping at fourteen days; the population is recruited in hospital and in outpatient clinics, and administration is almost always intravenous and monitored. Nothing in this corpus informs on relapse prevention or on office-based outpatient practice.
- Keep the two products and their frameworks apart. Racemic ketamine and intranasal esketamine do not fall under the same regulatory framework, and the status of each has to be checked for the country where one practises. Authorisation, marketing and reimbursement are three separate questions. In France, taken here as an example, the intravenous use of racemic ketamine in acute suicidality, where it occurs, takes place in hospital and outside the marketing authorisation for this indication; it does not transfer to office-based practice, and off-label use there is not freely permitted: it requires, in particular, the absence of an appropriate authorised alternative. As at 1 September 2026, injectable racemic ketamine holds a French authorisation only in anaesthetic indications, so its use in psychiatry in France is entirely off-label. In the European Union, intranasal esketamine has held a marketing authorisation since 18 December 2019 in treatment-resistant major depressive disorder and for the rapid reduction of symptoms in a psychiatric emergency, with the prescribing decision resting with a psychiatrist, self-administration under direct supervision and monitoring until the patient is clinically stable.
Frequently asked questions
Why use midazolam rather than placebo as the comparator in ketamine trials?
Because an inert solution reproduces nothing of the subjective experience of ketamine, which weakens blinding and leaves fully open the possibility that the observed effect is a non-specific sedation effect. Midazolam produces sedation, anxiolysis and altered consciousness without being an antidepressant or sharing the presumed mechanism of ketamine: the difference that remains between the two arms is easier to interpret. It is a higher methodological standard than that of meta-analyses that mix comparators, and it also explains why the effect sizes reported here are more sober.
Why 84 patients when the meta-analysis reports 649 participants?
Because 649 is the size of the corpus, not that of the analyses. The GRADE table of the publication gives, for each outcome, the number of participants and trials actually pooled: 84 patients and three trials for the MADRS total score, 178 and four for the suicidal ideation item, 118 and three for the Beck scale, 236 and three for derealisation. A single corpus can feed analyses of very unequal size, depending on the outcomes each trial reported. Note in passing that adding up the sample sizes of the ten trials, as given in the supplementary table of characteristics, yields 523, not 649.
Is the Beck Scale for Suicide Ideation result negative?
No. The initial difference of 4.30 points was significant at the conventional threshold, and it stops being so when one trial out of the three in the analysis is removed. A result that depends on keeping a single study is an unstable result, which can be used neither to assert efficacy nor to assert a lack of efficacy. The right way to put it in consultation is that this outcome is not settled by this review.
How can a statistically significant result have only moderate GRADE certainty?
These are two different questions. The p value and the confidence interval describe the precision of the estimate within the available data; GRADE certainty describes how much confidence can be placed in that estimate given the risk of bias, consistency across trials, precision, the directness of the outcomes and the risk of publication bias. Here the MADRS total score is downgraded for imprecision, the suicidal ideation item and the Beck scale for inconsistency: two distinct reasons, two distinct readings.
Should the risk of derealisation lead to ketamine being withheld?
That is not what the review says, since it reports a frequency, not a severity or a duration. What the result calls for is information beforehand: derealisation occurs in about one patient in three on ketamine against one in fifteen on midazolam, with perfect consistency across the three trials concerned and the only high certainty rating in the work. How to proceed with a given patient is a matter of clinical assessment and of the unit’s protocols.
Do these results apply to intranasal esketamine?
Nothing in this corpus allows it. The ten trials concern ketamine given almost exclusively intravenously, and in its racemic form in the trials that specify the form. The two products differ in molecule, route of administration, pharmacokinetics and their respective regulatory frameworks. Transposing the figures from one to the other would be an extrapolation, not a reading.
Annotated bibliography
Source study. Shah A, Sawaira FNU, Misbahuddin, Mohmand MS, Khan S, Bacha Z, Iftikhar H, Jamal A, Khattak F, Khalid AA, Syed A, Kamil KA. Comparative Efficacy and Safety of Ketamine Versus Midazolam for Suicidality: A GRADE-Assessed Systematic Review and Meta-Analysis of Randomized Controlled Trials. Brain and Behavior 2026;16(2):e71255. DOI 10.1002/brb3.71255. PMID 41656677. Received 27 August 2025, revised 13 January 2026, accepted 24 January 2026; a correction dated 20 August 2026 added the peer review history. Protocol registered with PROSPERO under number CRD420251110292; reporting in line with PRISMA 2020 and the Cochrane Handbook; risk of bias assessed with the five-domain Cochrane Risk of Bias Tool; certainty of evidence assessed with GRADE. Funding: the authors state that they have nothing to report. Competing interests: the authors declare no conflicts of interest. The document provides a pooling restricted to trials run against an active comparator, which the authors present as “more conservative and arguably more clinically relevant” than syntheses with mixed comparators, while stating explicitly that it does not claim a novel efficacy signal, and a certainty grading that varies from one outcome to another rather than being applied uniformly to all results. Supplementary material consulted: scale definitions, PRISMA flow diagram, risk of bias plots, tables of trial and patient characteristics, GRADE summary of findings table, forest plots before and after sensitivity analysis.