Published on 18 September 2026

Analysis · Depression · Psychopharmacology

Ketamine and suicidal ideation: does treatment resistance change the picture?

▬ Publication International Journal of Neuropsychopharmacology · 2026; 29(5): pyag021 · Liu KI et al. DOI 10.1093/ijnp/pyag021 PMID 42059646 Scientific 68 Editorial 69

Editorial collections A Study Under the Microscope

The essentials

One hundred and fifty-four patients with treatment-resistant depression, drawn from two randomized trials pooled after the fact, received a single infusion of ketamine against a comparator, and the authors asked one question item by item: does the degree of treatment resistance change the effect observed on each dimension of the depression scale used? Two items appear modulated by resistance, apparent sadness (interaction odds ratio 0.58, 95% confidence interval 0.41 to 0.83, p = 0.002) and inner tension (0.67, 0.48 to 0.95, p = 0.024). The suicidal ideation item shows no significant interaction (0.72, 0.50 to 1.04, p = 0.083). The authors call their work exploratory, and that is the right word: ten items tested in parallel for the treatment effect, then seven interaction terms, with no correction for multiple comparisons, on data pooling two different comparators across the two trials.

The context

For a patient who has already failed several lines of treatment, the pressing question is not average mood at six weeks but suicidal ideation in the days ahead. Ketamine occupies that place because it acts quickly. One legitimate clinical concern remains: the more a patient has resisted previous treatments, the less one expects from yet another one.

The study examined here tries to answer that concern by breaking the depression scale down instead of summing it. The idea is sound: a global score can mask that a treatment acts strongly on one dimension and not at all on another. It comes at a statistical cost, that of multiplying comparisons, and that cost is where the reading of this article turns.

The study at a glance

Population	154 adults aged 20 to 65 with treatment-resistant depression, defined as no response to at least two adequate treatments. Mean age 41.0 years, 72.1% women, mean illness duration 11.4 years, mean baseline depression score 35.9. Diagnosis: major depressive disorder in 89.0% of patients, bipolar I disorder in 3.2%, bipolar II disorder in 7.8%. Population exclusively of Han ethnicity, recruited in Taipei. 89 patients in the ketamine arm, 65 in the comparator arms.
Intervention	Single intravenous ketamine infusion over 40 minutes. Two dosages coexist in the pool: 66 patients received 0.5 mg/kg and 23 received 0.2 mg/kg. The first trial included both doses, the second only 0.5 mg/kg. Both dosages are pooled in the analysis.
Comparator	Saline in the first trial, midazolam 0.045 mg/kg in the second. These two comparators are not equivalent: one is inert, the other an active agent with sedative and anxiolytic effects. A pharmacological reminder, outside the scope of the trial itself.
Outcome measures	Total score and ten individual items of the Montgomery-Åsberg Depression Rating Scale, measured at baseline and on days 2, 3, 5, 7 and 15. Item improvement is defined as a decrease of at least 2 points from baseline. Degree of treatment resistance was quantified at baseline using the Maudsley Staging Method, which combines current episode duration, initial severity and the number of adequate treatment failures. Mean resistance score 9.4.
Design	Secondary analysis, conducted post hoc, on pooled data from two double-blind randomized trials. A third trial was considered then excluded, since its assessment schedule did not include a day-15 measurement; it contributes no patients. Linear mixed model for the total score, generalized estimating equations on logistic regression for the items, with interaction terms between treatment and degree of resistance fitted only on the items that reached significance. All models adjusted for age and sex.

Quality control

Analysis status	Post hoc Secondary analysis of already-collected data, which the authors themselves present as exploratory and hypothesis-generating. The originating trials are registered, but nothing indicates that the item-by-item analysis plan was filed before the data were accessed. This does not make the results false, it removes their confirmatory value.
Pooled comparators	Heterogeneous, adjusted Pooling an inert placebo arm with a midazolam arm amounts to comparing ketamine against two different references, and the estimated effect depends on the weight of each trial in the pool. The authors anticipated the objection: a sensitivity analysis adding the originating trial as a covariate preserves all seven item-level effects, with odds ratios from 2.02 to 6.02, and preserves both interactions, apparent sadness at 0.58 and p = 0.001, inner tension at 0.67 and p = 0.014. This adjustment narrows the objection without removing it, since it does not recreate a common comparator.
Correction for multiple testing	Absent, acknowledged Ten items tested in parallel for the treatment effect, then seven interaction terms on the retained items. The authors state explicitly that no correction was applied, on the grounds of strong intercorrelation among the scale’s items. The argument is understandable, it does not remove the risk of a false positive. Across seven comparisons, a Bonferroni threshold would sit at 0.0071: apparent sadness, at p = 0.002, crosses it; inner tension, at p = 0.024, does not.
Precision of estimates	Intervals reported All odds ratios are published with their 95% confidence interval. Several are very wide, such as reduced appetite, 5.88, 1.88 to 18.42, and the effect on inner tension touches the boundary of no effect, 2.01, 1.00 to 4.05. The one to watch is the suicidal ideation interaction: at 0.50 to 1.04, it is compatible with a halving of the effect as much as with no attenuation at all.
Pooled dosages	Two regimens A factor of two and a half separates 0.5 and 0.2 mg/kg. The authors state that the sample size of this secondary analysis did not allow an interaction analysis stratified by dose, even though a dose-dependent effect was reported in this same population by one of the pooled trials. The limitation is acknowledged, not resolved.
Funding and conflicts of interest	Public and institutional Funded by Taipei Veterans General Hospital, the Yen Tjing Ling Medical Foundation, the Kun-Po Soo Medical Foundation, Taiwan’s ministry for science and technology, joint programs of Taiwanese university hospitals, and Cheng Hsin General Hospital. The authors state that the funding sources played no role in the study and that they have no conflicts of interest or financial relationships to declare.

The results

0.083

The significance level of the interaction between treatment and degree of resistance on the suicidal ideation item, for an odds ratio of 0.72 whose confidence interval runs from 0.50 to 1.04. This value means that the study did not detect a modulation, not that it showed its absence.

Total depression score	5.27-point difference, p less than 0.001. Linear mixed model adjusted for age and sex, confidence interval 7.98 to 2.55 points in favor of ketamine. The treatment-by-time interaction, 0.18, 0.05 to 0.31, p = 0.006, indicates that the gap narrows over the fifteen days. An expected result, consistent with the trials this data comes from. It is not the contribution of this work.
Individual items improved	Seven items out of ten. Odds ratios for improvement under ketamine: apparent sadness 3.46, 1.62 to 7.41; inner tension 2.01, 1.00 to 4.05; reduced appetite 5.88, 1.88 to 18.42; concentration difficulties 2.91, 1.14 to 7.40; lassitude 2.68, 1.10 to 6.55; pessimistic thoughts 3.14, 1.51 to 6.52; suicidal thoughts 3.54, 1.63 to 7.69. Three items show no demonstrated effect: reported sadness, reduced sleep, inability to feel.
Direct effect on suicidal ideation	Odds ratio 3.54, p = 0.001. This is the clearest effect in the study, and it should not be lost behind the discussion of the interaction. In the model that includes degree of resistance, it reaches 4.47, 1.92 to 10.38, p less than 0.001. This is an average effect across the whole cohort, not an individual guarantee.
Resistance-by-treatment interaction, apparent sadness	0.58, 0.41 to 0.83, p = 0.002. This is the most solid result of the study, and the only one that would cross a threshold corrected for multiplicity. It suggests that the more marked the resistance, the less improvement occurs on this dimension. The authors note that this item is scored on the examiner’s observation, which exposes it to unblinding.
Resistance-by-treatment interaction, inner tension	0.67, 0.48 to 0.95, p = 0.024. A more fragile result: the upper bound of the interval approaches 1, and the corrected threshold would not be crossed. The objection about pooling comparators, since midazolam acts precisely on this dimension, is weaker than it appears, as the interaction remains significant after adjustment for originating trial.
Resistance-by-treatment interaction, suicidal ideation	0.72, 0.50 to 1.04, p = 0.083. The most tempting point to seize on, and the one that must be handled with the greatest caution. The point estimate points in the same direction as the two significant interactions, and its interval does not exclude a halving of the effect. A non-significant interaction does not demonstrate that the effect is independent of resistance, it indicates that the available data cannot settle the question in either direction.

Critical appraisal

Study design	Exploratory A secondary analysis of randomized data retains the protection of randomization for the main comparison, but not for subgroup analyses or interactions, which are observational in nature. The authors note that the two groups were not matched on baseline characteristics, since the pooling was done after the fact.
Power to detect an interaction	Not calculated No power calculation appears in the publication. Detecting an interaction ordinarily requires a sample far larger than one sufficient to detect a main effect, and yet two interactions reached the threshold in these 154 patients. The useful conclusion is therefore not that the study lacked power, but the one given by the interval: at 0.50 to 1.04 for the suicidal ideation item, the modulation is neither demonstrated nor ruled out.
Choice of comparator	Not homogeneous Midazolam has been chosen in several ketamine trials precisely because it reproduces part of the immediate subjective effects. That methodological quality becomes a problem once it is merged with an inert placebo. The authors further acknowledge that neither trial formally assessed the integrity of blinding and that functional unblinding may have occurred in either direction.
External validity	Homogeneous population Recruitment in Taipei, at two institutions according to the methods section, which the authors describe as a single-center limitation. Population exclusively of Han ethnicity, aged 20 to 65, including about 11% of patients with bipolar disorder. Generalization to other populations is not established, which holds for pharmacokinetics as much as for symptom expression.
Adequacy of conclusions	Cautious The exploratory nature is stated in the title, and the conclusion calls for prospective confirmation. One nuance, however: the abstract states that the antisuicidal effect is “robust and not significantly modulated,” a phrasing that slides faster toward absence of modulation than a negative test allows. The risk of overinterpretation will mainly come from citing the result without its caveats.
Transparency	Adequate, data on request Funding detailed, absence of conflicts of interest declared, supplementary material available, author contributions specified. The publication cites three registration numbers with the Japanese UMIN registry, UMIN000016985, UMIN000023581 and UMIN000033916, without specifying which corresponds to which trial. The data are not public and are available only on request from the corresponding author, the publication citing Taiwanese regulations.

Level of evidence

Scientific

Editorial

Confidence is adequate regarding ketamine’s overall antidepressant effect in this population, but that point is not new and does not come from this analysis. It is also adequate regarding the approach: breaking a scale down rather than summing it is a sound way to frame a clinical question.

It is weak on each specific conclusion. Two significant interactions among seven terms tested, themselves drawn from ten item-by-item comparisons without correction, on data pooling two different comparators and two different dosages, amount to a lead, not a result. And the most tempting claim, that of an antisuicidal effect that would withstand treatment resistance, rests on a negative test whose interval leaves open a halving of the effect. What is suggested is that ketamine’s effect is not uniform across symptom dimensions. What is demonstrated in this cohort is a clear direct effect on the suicidal ideation item, odds ratio 3.54. What is not established is the clinical meaning of the difference between dimensions. What cannot be claimed is that treatment resistance leaves the antisuicidal effect intact.

The colleague test

What an experienced colleague would say if shown this study in two minutes, between two consultations.

“This is a post hoc analysis on 154 patients, with two comparators mixed together, ten items tested and then seven interactions, and no correction at all. The message people will take from it, that the antisuicidal effect holds despite treatment resistance, rests on an odds ratio of 0.72 whose interval goes down to 0.50. In other words, a halving of the effect has not been ruled out. I see a hypothesis worth testing, not an argument to put to a patient.”

Translation for practice: nothing in this work changes an indication. It sheds light on a way of framing the question, separating the dimensions of depression instead of following a global score, and that approach deserves to be taken up in a trial designed to answer it.

What you can do with this

Keep the approach rather than the result: tracking suicidal ideation and mood separately in a patient treated urgently is clinically more informative than a global score, and it requires no additional tool.
Do not reassure a patient or a family by asserting that the effect on suicidal thoughts is independent of how many treatment lines have already failed. The published interval does not exclude a substantial attenuation.
In a highly resistant patient, keep in mind that some dimensions may respond less well than others. That is grounds for closer follow-up, not for giving up.
Remember that the measurement covers fifteen days after a single infusion. Nothing here speaks to the durability of the effect or to the value of repeated infusions.
Before discussing ketamine or esketamine with a patient, verify the exact regulatory status, approved form, and prescribing and monitoring framework wherever you practice. These rules vary from one country to another and continue to evolve.
For a trainee, this article is a good illustration of the difference between a main effect and an interaction, and of what it costs to pool two trials that did not share the same comparator.
The recommended approach to a suicidal crisis follows standard risk-assessment and safety-planning practice, independent of what this trial does or does not show about ketamine.

Can it be said that ketamine keeps its antisuicidal effect in patients with severe treatment resistance?

No, not on the basis of this work. The test designed to detect a difference did not reach significance, but the point estimate, 0.72, points in the same direction as the significant interactions, and its interval extends down to 0.50. The correct statement is that this study does not show a modulation, and that it does not rule one out either, including a substantial one.

Why is mixing the comparators a problem?

Because the quantity measured is a difference between ketamine and its reference. If the reference changes from one trial to the other, the difference does not carry the same meaning in both, and the pooled estimate depends on the weight of each trial. The authors addressed this with a sensitivity analysis adding the originating trial as a covariate: all results are preserved. This statistical adjustment remains less solid than a single comparator, but it makes the objection less decisive than it appears.

What changes if multiple testing is corrected for?

The threshold required for each comparison becomes stricter. Across the seven interaction terms, a Bonferroni correction would lower the threshold to 0.0071: apparent sadness, at p = 0.002, would remain significant; inner tension, at p = 0.024, would not. The authors justify the absence of correction by the strong intercorrelation among the scale’s items, an argument that supports a less severe correction than Bonferroni, not its abandonment.

Does a post hoc analysis have any value?

Yes, that of generating hypotheses. It loses the protection that preregistration provides against choosing analyses after the fact. The usual rule is to cite it as the starting point for future work, never as a decisive argument.

Are these results transferable to intranasal esketamine?

Nothing in this study allows that claim. Molecule, route of administration, treatment schedule and population all differ. The question deserves to be asked separately.

Annotated bibliography

Source study. Liu KI, Lin WC, Li CT, Bai YM, Su TP, Chen MH. Distinct therapeutic profiles of ketamine in treatment-resistant depression: an exploratory analysis. International Journal of Neuropsychopharmacology. 2026; 29(5): pyag021. DOI 10.1093/ijnp/pyag021. Open-access publication dated 30 April 2026. Taiwanese institutional funding, with no role for the funder; the authors declare no conflicts of interest.

Originating trials. Su TP, Chen MH, Li CT, et al. Dose-related effects of adjunctive ketamine in Taiwanese patients with treatment-resistant depression. Neuropsychopharmacology. 2017; 42: 2482-2492. DOI 10.1038/npp.2017.94. Trial against saline, including both dosages. Su TP, Li CT, Lin WC, et al. A randomized, double-blind, midazolam-controlled trial of low-dose ketamine infusion in patients with treatment-resistant depression and prominent suicidal ideation. International Journal of Neuropsychopharmacology. 2023; 26: 331-339. DOI 10.1093/ijnp/pyad014. Trial against midazolam. The publication cites three registration numbers, UMIN000016985 registered on 30 June 2016, UMIN000023581 registered on 31 December 2018, and UMIN000033916 registered on 30 November 2021, without indicating which corresponds to which trial.

Excluded trial. Chen MH, Cheng CM, Gueorguieva R, et al. Maintenance of antidepressant and antisuicidal effects by D-cycloserine among patients with treatment-resistant depression who responded to low-dose ketamine infusion. Neuropsychopharmacology. 2019; 44: 2112-2118. DOI 10.1038/s41386-019-0480-y. Considered then excluded from the pooling, since its assessment schedule did not include a day-15 measurement; it contributes no patients to the analysis.

Editorial collections

A Study Under the Microscope