Intranasal Esketamine for Treatment-Resistant Depression: A Practical Clinical Guide to Patient Selection, Dosing, Combination Strategies and Adverse-Event Management

1. Introduction

Treatment-resistant depression (TRD), usually defined as failure to respond to at least two adequately dosed and delivered antidepressant trials in the current episode, affects roughly a third of patients with major depressive disorder and accounts for a disproportionate share of chronicity, disability and cost. Intranasal esketamine, the S-enantiomer of ketamine, received European and United States approval in 2019 as adjunctive treatment for TRD. An indication for depressive symptoms in adults with major depressive disorder and acute suicidal ideation or behaviour followed in 2020, and a United States monotherapy indication in January 2025 [1,2,3]. Esketamine was the first antidepressant acting primarily outside the monoamine system to be widely approved.

The efficacy data have been reviewed many times and remain contested. In the ESCAPE-TRD trial, esketamine was superior to extended-release quetiapine, with remission at week 8 in 27.1% versus 17.6% of patients [4]. On the other hand, an independent individual-patient-data meta-analysis of seven randomized placebo-controlled trials found a mean MADRS advantage of only 2.94 points (95% CI −5.39 to −0.48) in the initiation trials; no value in that interval reached the 6.5-point difference that the sponsor had itself pre-specified as clinically meaningful [5]. A PRISMA meta-analysis of 87 articles reported pooled effect sizes of 0.15–0.23 at weeks 2–4, similar to those of atypical antipsychotic augmentation, and no significant effect on suicidality at any timepoint [6].

We do not reopen that debate here. Our starting point is that most of the decisions a prescriber faces fall outside the evidence on which the licence was granted. Almost none of the patients in the registration trials met the usual definition of advanced resistance [5], and the programme as a whole excluded bipolar disorder, psychosis, obsessive–compulsive disorder, four personality disorders, moderate-to-severe substance use disorder and recent suicidal intent (Section 4.1). Patients treated in practice are different. In a comparison of 418 esketamine-treated patients with 830,047 other patients with TRD, those receiving esketamine had more severe depression and more suicidal ideation; they had more often received electroconvulsive therapy (ECT) or repetitive transcranial magnetic stimulation (rTMS); they more often had comorbid anxiety, post-traumatic stress disorder (PTSD), substance use disorder or chronic pain; and their Charlson comorbidity scores were higher [7].

Clinicians therefore have to decide whom to prioritise, what dose to aim for, what to do with the benzodiazepine the patient has taken for a decade, how to manage dissociation at 40 minutes, whether a history of alcohol dependence is a barrier, and when to stop. The label says little about any of this. Di Vincenzo and colleagues have already examined several common misconceptions about esketamine [8]. Here we take these decisions one at a time, summarise what the current evidence supports for each, and say whether a recommendation rests on the label, on expert consensus or only on clinical experience.

2. Esketamine at a Glance

Table 1 summarises the drug’s general characteristics, pharmacokinetics, approved indications and standard dosing schedule, together with the main advantages and limitations discussed in later sections.

3. Mechanism of Action and Its Clinical Corollaries

Prescribers do not need the full molecular account, but a few mechanistic facts change what we tell patients and what we do in the clinic. Table 2 lists the receptor pharmacology and its clinical corollaries, and the subsections below discuss the points we find most useful in practice.

3.1. Low Affinity and the Importance of Peak Concentration

In the most comprehensive enantiomer-resolved screen available, a 98-target panel produced no hits at 100 nM and only two at 10 µM, and target deconvolution across more than 16,000 human gene products found no high-affinity binding site for either enantiomer [9]. Esketamine is therefore a low-affinity drug whose effect depends on a brief peak in concentration, and the route and rate of delivery matter at least as much as the total dose in milligrams. Esketamine is also neither a reuptake inhibitor nor a dopaminergic agent: its affinities at the monoamine transporters lie two orders of magnitude above clinical exposure, and an earlier report of high-affinity D2 binding was not replicated [10].

3.2. From NMDA Receptor Blockade to Synaptic Plasticity

NMDA receptor blockade starts the process, but the antidepressant effect appears to come from the glutamate surge that follows and from AMPA-driven plasticity. Blocking NMDA receptors on GABAergic interneurons disinhibits pyramidal neurons. This has been shown directly in hippocampal slices at clinically relevant concentrations [11] and in awake mice, in which ketamine suppressed somatostatin interneuron activity and enlarged dendritic calcium transients [12]. Deleting GluN2B selectively from GABAergic interneurons, but not from principal neurons, abolishes the antidepressant effect [13]. Downstream, the cascade runs through AMPA receptor activation, eEF2 kinase de-suppression and rapid BDNF translation [14], and through mTORC1-dependent synaptogenesis [15]. The comparison with memantine is useful. Memantine matches ketamine’s potency at the NMDA receptor but fails to block receptors at rest under physiological magnesium concentrations, and it has no antidepressant effect [16]. This account has limits. The preference for interneuron NMDA receptors is inferred from circuit-level studies and has not been measured as a property of the molecule. And when mTORC1 inhibition was tested in patients, rapamycin prolonged the antidepressant response instead of abolishing it, so mTORC1 should not be presented to patients as the mechanism [17].

3.3. Induction and Maintenance of Response

Longitudinal two-photon imaging showed that prefrontal spine formation was required to maintain ketamine’s behavioural effect but not to induce it [18]. In clinical terms, early response can occur before structural changes are consolidated, and relapse after discontinuation is better explained by the loss of these changes than by the loss of receptor occupancy. This is also the main argument for tapering by spacing out sessions instead of stopping abruptly.

3.4. Hydroxynorketamine and Stereochemistry

Findings on hydroxynorketamine are often cited in discussions of esketamine, but most of them do not apply to it. The well-known metabolite results concern (2R,6R)-hydroxynorketamine, which derives from (R)-ketamine [19]. Ring hydroxylation preserves the C2 configuration, so a patient given esketamine produces (2S,6S)-hydroxynorketamine. This stereoisomer is more potent at NMDA receptors [20] and was less active behaviourally in the original work. Later target-deconvolution studies also found that (2R,6R)-hydroxynorketamine binds none of the known or proposed ketamine targets [21]. The same applies to σ1 receptor affinity, which favours arketamine; the enantiomer preference there is the reverse of that at the NMDA receptor (Table 2). Mechanistic claims taken from studies of racemic or (R)-ketamine should therefore be checked before they are applied to esketamine.

4. Ideal and Real-World Candidates

4.1. The Gap Between the Trial Population and the Clinic

The patients in the registration trials were quite unlike those now receiving esketamine. In the individual-patient-data meta-analysis, resistance could be staged in 1013 patients: 334 had failed one antidepressant class, 648 had failed two, 29 had failed three and 2 had failed four or more [5]. Only one of the seven studies recorded prior ECT. Bipolar disorder, current or prior psychotic disorder, obsessive–compulsive disorder, borderline, antisocial, histrionic and narcissistic personality disorder, moderate-to-severe substance use disorder and recent suicidal intent were exclusion criteria across the programme. ESCAPE-TRD also required patients to show “signs of minimal clinical improvement” on their current antidepressant, which excluded both complete non-responders and good responders [4].

Real-world cohorts include many of these patients, and their outcomes are not worse. The Italian REAL-ESK consortium reported response and remission rates of 64.2% and 40.6% at three months in 116 patients and found no difference in efficacy between patients with and without psychiatric comorbidity [22]. A meta-analysis of nine real-world studies found a pooled Hedges’ g of −1.98, and effectiveness was not influenced by the presence of anxiety, personality or substance use disorders [23]. In the French ELLIPSE cohort of 200 patients, the MADRS score fell from 31.9 to 12.4 at 12 months, with response and remission reaching a plateau at month 2 and maintained thereafter [24]. These figures come from uncontrolled single-arm studies and cannot be set against drug–placebo differences. They support a narrower claim: in most datasets that have examined it, comorbidity has not predicted non-response (Section 5.1 describes an exception).

4.2. Predictors of Response

Clinicians would like a selection rule, but the two largest systematic reviews of predictors, published within a year of each other, agree mainly that most candidate variables do not predict outcome. A review of 122 studies, 12,674 patients and 77 predictor variables concluded that “most variables had more reports revealing no association with antidepressant outcomes than a positive or negative predictive ability”; only early response and family history of substance use disorder emerged as promising positive signals, with anxiety during administration and childhood maltreatment as negative ones [25]. A parallel review of 44 studies found the predictive value of demographic and clinical variables “limited with either no significant relationship or inconsistent results”, and reached partly contradictory conclusions, notably on childhood trauma, which one review classes as negative and the other as positive [26].

The clearest positive signal comes from the REAL-ESK machine-learning sub-study, in which random-forest classifiers achieved 66–69% accuracy and identified severe anhedonia, anxious distress, mixed features and bipolarity as positive predictors, with benzodiazepine use and greater baseline severity predicting delayed response [27]. Accuracies in this range are modest and the model has not been externally validated, so the findings are hypothesis-generating. The anti-anhedonic signal is supported by a second REAL-ESK analysis: in 253 REAL-ESK patients the anti-anhedonic effect was statistically dissociable from overall depressive-symptom improvement, with anhedonia-subscale response of 38% in unipolar and 51.9% in bipolar TRD at three months [28].

Some findings run against intuition. Degree of treatment resistance did not moderate efficacy in the trials (p = 0.80), but the range studied was narrow; the result shows that the effect is not concentrated in more resistant patients and does not show that the drug works equally well at stage IV [5]. Across the wider range seen in practice, milder resistance on the Maudsley Staging Method predicted a better outcome (OR 3.162, p = 0.041) [29]. Access is also skewed in ways the evidence does not justify. Substance use disorder, Medicaid insurance, higher comorbidity burden and older age all predict lower initiation, and initiation falls by 92.8% at 50 miles or more from a certified centre compared with under 10 miles [30].

4.3. Patients with Suicidal Ideation

Suicidal ideation is common among patients treated with esketamine in practice [7], and the controlled data on this outcome are more favourable than the primary results of the phase 3 programme suggest. In the phase 2 trial in 68 patients at imminent risk of suicide, esketamine 84 mg added to standard care reduced the MADRS suicidal-thoughts item more than placebo at 4 hours (effect size 0.67, p = 0.002), and on post hoc analysis suicide risk had resolved at 24 hours in 40.0% of patients versus 6.5% [31]. In the pooled ASPIRE trials (451 patients), resolution of suicidality on the Clinical Global Impression–Severity of Suicidality–Revised (CGI-SS-R) at 4 hours was more frequent with esketamine (33.2% vs 20.0%) [32]. Meta-analyses of randomized trials point the same way, with an odds ratio of 2.04 (95% CI 1.37–3.05) for resolution of suicidality at 2–4 hours [33] and effect sizes of 0.26 at 4–6 hours and 0.30 at 24 hours for intranasal esketamine [34]. Repeated intravenous esketamine also reduced suicidal ideation more than midazolam in a randomized trial in adolescents [35].

Real-world cohorts show reductions that persist over months. In 80 Italian outpatients, the ideation score of the Columbia-Suicide Severity Rating Scale (C-SSRS) fell from 1.56 to 0.78 after one week and to 0.12 at six months, and 68.4% had a suicidal-ideation response at one week [36]. In the 90-patient cohort described in Section 5.1, suicidal ideation, suicidal behaviour and self-harm decreased from the first month in patients with and without borderline personality disorder [37], and in ECT non-responders the MADRS suicidal-thoughts item fell from 3.5 to 2.1 [38]. In a six-month naturalistic comparison of 62 patients at high suicide risk, ideation fell faster and further with esketamine than with other pharmacological strategies, with a number needed to treat of 1.35 to prevent one case of high suicide risk [39]. In a matched cohort of 2126 patients with bipolar depression, esketamine was associated with fewer suicide-related outcomes at 30 days (hazard ratio 0.485) and at one year (hazard ratio 0.754) [40].

These findings should be read alongside the negative ones. At 24 hours, the CGI-SS-R did not differ from placebo in either phase 3 trial [41,42]. A large meta-analysis found no significant effect on suicidality at any timepoint [6], and another judged the evidence for esketamine insufficient [43]. In a cohort of 261 patients, the improvement in suicidal ideation disappeared after adjustment for the other depressive symptoms, which suggests that it follows the antidepressant response [44], and in US claims data esketamine was not associated with fewer suicide attempts or episodes of self-harm [45]. Both labels state that an effect in preventing suicide or in reducing suicidal ideation or behaviour has not been demonstrated, and the US label adds that esketamine does not remove the need for hospitalisation when it is clinically warranted [2,3]. We therefore regard suicidal ideation as no reason to withhold esketamine, since it may improve within hours of the first dose, but the drug should be given together with safety planning and crisis care and not in place of them.

Table 3. Practical patient-selection framework for intranasal esketamine in treatment-resistant depression.

Patient feature	Basis	Practical implication
PRIORITISE: consistent signals across datasets
Prominent anhedonia	Positive predictor in machine-learning analysis; anti-anhedonic effect statistically dissociable from overall improvement in 253 patients	The most consistent positive signal so far. Consider measuring anhedonia separately at baseline and follow-up
Anxious distress or comorbid anxiety disorder	Positive predictor in two independent cohorts (OR 3.149 for anxiety comorbidity); measurable HAM-A reduction; unaffected effectiveness in pooled real-world data	Trait or disorder-level anxiety is favourable. Distinguish this from acute procedural anxiety during dosing, which is a negative predictor
Suicidal ideation	Greater reduction than placebo at 4 h in randomized trials and meta-analyses; reductions maintained over months in real-world cohorts	A reasonable choice. Measure with the C-SSRS at each session and combine with safety planning and crisis care (Section 4.3)
Mixed features or bipolar spectrum	Positive predictor in machine-learning analysis; equivalent response with greater anxiolysis in bipolar TRD	Off-label in bipolar depression. Maintain mood-stabiliser cover and monitor for switch
Comorbid PTSD or significant trauma history	Two randomized trials of intravenous ketamine in PTSD itself; PTSD patients respond as well despite worse baseline scores	Do not regard trauma history as a barrier. Prepare the patient for the possibility of re-experiencing during sessions
Comorbid substance use disorder, stable or in remission	No effectiveness difference; pharmacovigilance reporting odds ratios not elevated for esketamine on any misuse parameter	Supervised administration already limits misuse (Section 9)
Failure of ≥ 2 antidepressants without extreme resistance	The population the trials actually studied (97% stage I–II)	This is where the licensing evidence applies most directly
APPROACH WITH CAUTION: treat with an adapted protocol and explicit consent about uncertainty
Comorbid borderline personality disorder	No randomized evidence. One prospective 90-patient cohort (reported in three analyses) encouraging; one retrospective mixed ketamine/esketamine cohort reports near-null response; case reports of impulsive suicide attempt and self-harm escalation, clustering around interruption	Treat where options are exhausted. Baseline C-SSRS, self-harm inventory and impulsivity measures; concurrent structured psychotherapy; plan the taper explicitly; consider inpatient induction if recent attempts, frequent self-harm or severe baseline dissociation
Bipolar TRD	Off-label; manic switch 5.6% vs 2.0% in unipolar (p = 1.000, underpowered)	Mood-stabiliser cover; structured hypomania monitoring at each visit
PTSD with active re-experiencing	In the only case series, re-experiencing during sessions resolved with continued treatment in 72.7% but caused cessation in 27.3%	Pre-session psychoeducation that flashbacks may occur and usually abate; trauma-informed environment; agree a stopping rule in advance
Age ≥ 75 years	Very limited data; the dedicated geriatric trial (TRANSFORM-3) failed its primary endpoint and showed no signal above 75 [46]	Falls-risk assessment, extended observation, slower titration. Extended induction beyond eight sessions may help
Cardiovascular disease or imperfectly controlled hypertension	Transient systolic rise of about 9 mmHg at 40 min [5]; serum esketamine concentration correlates with the magnitude of the rise	Optimise blood pressure before induction. Consider 56 mg and closer monitoring if on a CYP-inhibiting antidepressant
Metabolic syndrome	Associated with reduced initial but not sustained response in a 150-patient ketamine cohort [47]	Persist beyond induction before judging
Traumatic brain injury or severe obstructive sleep apnoea	Depression did not improve in these subgroups in a US veterans cohort (n = 38 and n = 9 respectively) [48]	Screen for and treat sleep apnoea before starting; set expectations in brain injury
High-dose chronic benzodiazepine therapy	Dose-dependent attenuation of the delayed effect; delayed response in machine-learning analysis	Reduce the standing dose where safe before induction. Do not withhold treatment on this basis alone
Very low body mass index	Fixed-dose product means higher mg/kg exposure; no data	Clinician judgement with enhanced haemodynamic and dissociation monitoring; document the absence of evidence
Substantial travel distance from a certified centre	Initiation falls by 92.8% at ≥ 50 miles; distance predicts interruption	Arrange transport and scheduling before induction
AVOID, OR REQUIRE A DOCUMENTED SPECIFIC JUSTIFICATION
Active psychotic features, or current or prior psychotic disorder	Excluded from every trial; no usable data in either direction	Both labels advise caution rather than prohibition, but the clinician acts without evidence and should document the reasoning
Lifetime ketamine use disorder	Excluded from the registration trials; no study has examined it	The substance-use history of most concern
Active, untreated moderate-to-severe substance or alcohol use disorder	Alcohol use disorder predicts treatment interruption, and the protocol requires reliable attendance; no misuse signal has been shown	Stabilise the substance use disorder first, then reconsider
Esketamine used in place of crisis care for imminent suicide risk	Neither label claims an anti-suicidal effect; the US label states that esketamine does not remove the need for hospitalization	Use alongside hospitalization, safety planning and close follow-up when these are needed
Esketamine as a substitute where ECT is indicated	An adequate course of ECT in the current episode was an exclusion criterion, so the comparison was never made in that direction	Catatonia, food refusal, psychotic depression and imminent risk remain indications for ECT

C-SSRS, Columbia-Suicide Severity Rating Scale; CYP, cytochrome P450; ECT, electroconvulsive therapy; HAM-A, Hamilton Anxiety Rating Scale; OR, odds ratio; PTSD, post-traumatic stress disorder; TRD, treatment-resistant depression.

Table 3. Practical patient-selection framework for intranasal esketamine in treatment-resistant depression.

Patient feature	Basis	Practical implication
PRIORITISE: consistent signals across datasets
Prominent anhedonia	Positive predictor in machine-learning analysis; anti-anhedonic effect statistically dissociable from overall improvement in 253 patients	The most consistent positive signal so far. Consider measuring anhedonia separately at baseline and follow-up
Anxious distress or comorbid anxiety disorder	Positive predictor in two independent cohorts (OR 3.149 for anxiety comorbidity); measurable HAM-A reduction; unaffected effectiveness in pooled real-world data	Trait or disorder-level anxiety is favourable. Distinguish this from acute procedural anxiety during dosing, which is a negative predictor
Suicidal ideation	Greater reduction than placebo at 4 h in randomized trials and meta-analyses; reductions maintained over months in real-world cohorts	A reasonable choice. Measure with the C-SSRS at each session and combine with safety planning and crisis care (Section 4.3)
Mixed features or bipolar spectrum	Positive predictor in machine-learning analysis; equivalent response with greater anxiolysis in bipolar TRD	Off-label in bipolar depression. Maintain mood-stabiliser cover and monitor for switch
Comorbid PTSD or significant trauma history	Two randomized trials of intravenous ketamine in PTSD itself; PTSD patients respond as well despite worse baseline scores	Do not regard trauma history as a barrier. Prepare the patient for the possibility of re-experiencing during sessions
Comorbid substance use disorder, stable or in remission	No effectiveness difference; pharmacovigilance reporting odds ratios not elevated for esketamine on any misuse parameter	Supervised administration already limits misuse (Section 9)
Failure of ≥ 2 antidepressants without extreme resistance	The population the trials actually studied (97% stage I–II)	This is where the licensing evidence applies most directly
APPROACH WITH CAUTION: treat with an adapted protocol and explicit consent about uncertainty
Comorbid borderline personality disorder	No randomized evidence. One prospective 90-patient cohort (reported in three analyses) encouraging; one retrospective mixed ketamine/esketamine cohort reports near-null response; case reports of impulsive suicide attempt and self-harm escalation, clustering around interruption	Treat where options are exhausted. Baseline C-SSRS, self-harm inventory and impulsivity measures; concurrent structured psychotherapy; plan the taper explicitly; consider inpatient induction if recent attempts, frequent self-harm or severe baseline dissociation
Bipolar TRD	Off-label; manic switch 5.6% vs 2.0% in unipolar (p = 1.000, underpowered)	Mood-stabiliser cover; structured hypomania monitoring at each visit
PTSD with active re-experiencing	In the only case series, re-experiencing during sessions resolved with continued treatment in 72.7% but caused cessation in 27.3%	Pre-session psychoeducation that flashbacks may occur and usually abate; trauma-informed environment; agree a stopping rule in advance
Age ≥ 75 years	Very limited data; the dedicated geriatric trial (TRANSFORM-3) failed its primary endpoint and showed no signal above 75 [46]	Falls-risk assessment, extended observation, slower titration. Extended induction beyond eight sessions may help
Cardiovascular disease or imperfectly controlled hypertension	Transient systolic rise of about 9 mmHg at 40 min [5]; serum esketamine concentration correlates with the magnitude of the rise	Optimise blood pressure before induction. Consider 56 mg and closer monitoring if on a CYP-inhibiting antidepressant
Metabolic syndrome	Associated with reduced initial but not sustained response in a 150-patient ketamine cohort [47]	Persist beyond induction before judging
Traumatic brain injury or severe obstructive sleep apnoea	Depression did not improve in these subgroups in a US veterans cohort (n = 38 and n = 9 respectively) [48]	Screen for and treat sleep apnoea before starting; set expectations in brain injury
High-dose chronic benzodiazepine therapy	Dose-dependent attenuation of the delayed effect; delayed response in machine-learning analysis	Reduce the standing dose where safe before induction. Do not withhold treatment on this basis alone
Very low body mass index	Fixed-dose product means higher mg/kg exposure; no data	Clinician judgement with enhanced haemodynamic and dissociation monitoring; document the absence of evidence
Substantial travel distance from a certified centre	Initiation falls by 92.8% at ≥ 50 miles; distance predicts interruption	Arrange transport and scheduling before induction
AVOID, OR REQUIRE A DOCUMENTED SPECIFIC JUSTIFICATION
Active psychotic features, or current or prior psychotic disorder	Excluded from every trial; no usable data in either direction	Both labels advise caution rather than prohibition, but the clinician acts without evidence and should document the reasoning
Lifetime ketamine use disorder	Excluded from the registration trials; no study has examined it	The substance-use history of most concern
Active, untreated moderate-to-severe substance or alcohol use disorder	Alcohol use disorder predicts treatment interruption, and the protocol requires reliable attendance; no misuse signal has been shown	Stabilise the substance use disorder first, then reconsider
Esketamine used in place of crisis care for imminent suicide risk	Neither label claims an anti-suicidal effect; the US label states that esketamine does not remove the need for hospitalization	Use alongside hospitalization, safety planning and close follow-up when these are needed
Esketamine as a substitute where ECT is indicated	An adequate course of ECT in the current episode was an exclusion criterion, so the comparison was never made in that direction	Catatonia, food refusal, psychotic depression and imminent risk remain indications for ECT

C-SSRS, Columbia-Suicide Severity Rating Scale; CYP, cytochrome P450; ECT, electroconvulsive therapy; HAM-A, Hamilton Anxiety Rating Scale; OR, odds ratio; PTSD, post-traumatic stress disorder; TRD, treatment-resistant depression.

5. Psychiatric Comorbidity, Including Personality Disorders

5.1. Borderline Personality Disorder

Borderline personality disorder (BPD) is common among patients with TRD, and the evidence on esketamine in this group is thin. BPD was an exclusion criterion in TRANSFORM-2, ESCAPE-TRD and the ASPIRE suicidality trials; the ASPIRE trials also excluded patients “not meeting full DSM-5 criteria for borderline personality disorder but exhibiting recurrent suicidal gestures, threats, or self-mutilating behaviors.” There is therefore no randomized evidence in this group. The excluded group is large: BPD is diagnosed in 10–30% of patients with major depressive disorder, and depression occurs in 71–83% of people with BPD. The combination is associated with more severe symptoms, delayed remission and a poorer response to antidepressants, psychotherapy and ECT [49].

What exists is encouraging but uncontrolled. The first systematic review in this population (9 studies, 281 patients) concluded that patients with depression and comorbid BPD or traits “were equally likely to respond to ketamine/esketamine as those with depression but without”, while noting case reports of suicidal ideation and self-harm following treatment cessation and “indications of a higher risk of acute dissociation” [50]. The largest controlled comparison, using intravenous ketamine in 100 patients matched for BPD status, found large improvements in both groups with no significant between-group difference, and improvement in borderline symptoms themselves [51].

The best esketamine-specific evidence comes from a single prospective multicentre Italian cohort of 90 consecutive TRD outpatients (45 with comorbid BPD and 45 with no personality disorder) followed for six months and reported in three separate analyses, two of which include authors of the present review. Depressive symptoms fell substantially (ηp² = 0.81) with a significant time × BPD interaction indicating faster early improvement in the BPD group; remission at six months was 53.3% overall with no difference by BPD status (48.9% vs 57.8%, p = 0.398), and response rates were higher in the BPD group from month one onward. Anxiety and impulsivity decreased, there was no deterioration in cognitive screening over six months, and there were no serious adverse events and no dropouts [52]. In the same cohort, suicidal ideation, suicidal behaviour and deliberate self-harm all decreased significantly from month one, with self-harm episode frequency in the BPD group falling from a mean of 30.8 at baseline to 2.4 at six months, and no increase in suicidality in any patient [37]. A subsequent analysis of the same cohort indicated that suicidal ideation covaried with depressive severity rather than following an independent trajectory [53].

A larger retrospective dataset points the other way. In a US cohort of 200 patients receiving either intravenous ketamine or intranasal esketamine, comorbidity profile was a strong moderator: pooled across both modalities, patients without comorbidity improved by 22.0 MADRS points and those with BPD by only 3.7, leading the authors to conclude that comorbid conditions, “especially BPD, attenuate the benefits of both ketamine and esketamine” [54]. That study also found intravenous ketamine superior to esketamine overall by roughly 5 MADRS points, so its comorbidity signal is not esketamine-specific. Individual harms are also documented: a young woman with recurrent depression and BPD showed initial rapid improvement on intravenous esketamine followed by disinhibition, increased emotional responsivity and an impulsive suicide attempt after the fifth infusion [55].

In our view, esketamine can be offered to patients with BPD when other options have been exhausted, provided the protocol is adapted. Suicidality, self-harm and impulsivity should be measured with structured instruments at baseline and at each review; structured psychotherapy should run alongside treatment; the taper and any interruption in supply should be planned explicitly, since this is when the reported harms cluster; and the threshold for inpatient induction should be lower in patients with recent suicide attempts, frequent self-harm or severe dissociative symptoms at baseline.

5.2. Anxiety

Anxious distress is a positive predictor of response [27], comorbid anxiety disorder predicted a better outcome with subcutaneous esketamine (OR 3.149, p = 0.028) [29], and the anxiolytic effect is measurable and large, with Cohen’s d of 0.840 at one month rising to 1.419 at three months in older adults [56]. What matters clinically is the difference between trait or disorder-level anxiety, which appears favourable, and anxiety during the dosing session, which predicts a poorer response [25]. Preparation, setting and continuity of staff therefore matter, and a patient with generalised anxiety should not be assumed to tolerate sessions badly.

5.3. Post-Traumatic Stress Disorder and Trauma History

Two randomized trials support intravenous ketamine in PTSD itself, the later repeated-dose trial showing a CAPS-5 difference of 11.88 points (d = 1.13) and 67% versus 20% response [57,58]; there is no esketamine randomized trial in PTSD and none in comorbid TRD with PTSD. For everyday practice, the more useful data concern re-experiencing of trauma during sessions. In a retrospective series of 22 patients across 11 French departments, all of whom experienced flashbacks during esketamine administration, re-experiencing resolved as sessions progressed in 16 of 22 (72.7%), while 6 of 22 (27.3%) stopped treatment because it persisted; among those continuing, depression response was 45.5% and PTSD improvement 45.5% [59]. Because patients entered the series only if they had had a flashback, it cannot estimate how often this occurs, only what happens when it does. Protocols integrating esketamine with structured traumatic-memory work have been proposed, and their authors note that the literature supports mechanistic plausibility and feasibility more than efficacy [60].

5.4. Bipolar Depression

All use in bipolar depression is off-label. In the REAL-ESK comparison of 35 patients with bipolar and 35 with unipolar TRD, response and remission rates did not differ, esketamine showed greater anxiolytic action in the bipolar group, and depressive improvement was not associated with treatment-emergent affective switch [61]. The largest secondary analysis quantified switch rates as 5.6% in bipolar versus 2.0% in unipolar TRD [28]: numerically higher, but the groups (54 and 199 patients, three and four events) are too small to exclude a real difference of that size, and the published p-value of 1.000 is difficult to reconcile with those denominators. A five-year observational cohort recorded no treatment-emergent hypomania or mania over 260 weeks [62]. Mood-stabiliser cover and structured hypomania monitoring are reasonable precautions, although neither has been tested.

5.5. Comorbidities Without Data

Comorbid obsessive–compulsive disorder, eating disorders and attention-deficit/hyperactivity disorder have not been studied with esketamine in TRD; current obsessive–compulsive disorder was an explicit exclusion. For autism spectrum disorder the evidence amounts to a three-patient case series, in which two of three achieved remission at six months with quality-of-life improvement and no major adverse effects [63]. Clinicians treating these patients are working without evidence and should document this. Because the product is dosed at fixed 28, 56 or 84 mg rather than by weight, very low body mass implies a higher mg/kg exposure: a theoretical concern in severe eating disorders for which no empirical data exist.

6. Dose Selection and Titration

Many clinicians regard 84 mg as the most effective dose. The randomized data do not support this view.

6.1. The Randomized Dose Data

TRANSFORM-1 is the only fixed-dose phase 3 trial and was designed to compare the two doses with placebo. Pre-specified hierarchical testing required the 84 mg arm to be significant before 56 mg could be formally tested. The 84 mg arm did not separate from placebo: least-squares mean difference versus placebo −3.2 (95% CI −6.88 to 0.45, p = 0.088). The 56 mg arm, which could not be formally tested, showed a numerically larger difference of −4.1 (95% CI −7.67 to −0.49, nominal p = 0.027). Response favored 56 mg (54.1% vs 53.1%) while remission favored 84 mg (36.0% vs 38.8%), a flat and non-monotonic pattern [64].

The main evidence for a dose–response relationship is the phase 2 dose-ranging study, which found an ascending relationship (28 mg −4.2; 56 mg −6.3; 84 mg −9.0; p < 0.001 for trend), but with only 11–12 patients per active arm, and it was not replicated in phase 3 [65]. The monotherapy trial separated both doses from placebo with observed effect sizes of 0.48 (56 mg) and 0.63 (84 mg) at day 28, but 56 mg was numerically better at 24 hours, the confidence intervals overlap substantially, and no formal statistical comparison of the two doses was performed [1]. A meta-analysis of dose escalation concluded that 56–84 mg was superior to 28 mg but did not separate 84 from 56 mg [66].

Adverse effects, by contrast, clearly increase with dose. In a meta-analysis of nine randomized trials, higher doses (≥ 56 mg) versus lower (≤ 28 mg) produced risk ratios of 3.72 versus 1.69 for nausea and 10.65 versus 3.27 for dissociation, with a 2.22-fold increase in discontinuation due to adverse events [67].

6.2. Dosing in Practice

Most patients nonetheless reach 84 mg: 64.4% of 1148 patients in the long-term extension were on 84 mg at final maintenance dose [68], and in the French routine-practice cohort 80.7% of under-65s escalated to 84 mg at a median of 8.5 days [69]. The real-world datasets we identified that tested dose against outcome found no relationship. Expert consensus favors maximization: the European Delphi panel of 30 psychiatrists reached ≥ 80% agreement that dose and frequency maximization to 84 mg weekly should be pursued to enhance acute-phase outcomes [70].

We recommend titrating to the maximum tolerated dose because consensus supports it and most patients reach it. Patients should not be told that 84 mg works better than 56 mg, and the dose should not be pushed through intolerable adverse effects in the hope of rescuing a non-responder, a strategy that has never been tested prospectively. In patients aged 65 or over, the European label requires a 28 mg starting dose with 28 mg increments. The dedicated geriatric trial failed its primary endpoint, with benefit concentrated in those aged 65–74 and no signal above 75 [46]. The lower starting dose is a tolerability precaution and does not cap the dose, and in routine practice 72.7% of patients over 65 are still escalated to 84 mg [3,69]. In one inpatient cohort, the group with the highest proportion on 84 mg (92.1%) had the lowest response rate (28.8%) [71], most likely because the dose is raised in patients who are not responding.

7. Concomitant Medications and Combination Strategies

Both product labels address only four concomitant-medication issues: additive sedation, blood pressure, nasal co-medication timing and, briefly, cytochrome P450 interactions [2,3]. Neither discusses whether any concomitant drug alters efficacy, although clinical data now exist for benzodiazepines, opioid antagonists, lithium and lamotrigine. Table 4 summarises how to handle the rest of the regimen.

7.1. Which Oral Antidepressant

The European indication requires a concomitant SSRI or SNRI; the FDA now also licenses monotherapy, so in the United States an oral antidepressant is no longer required. The only large comparative dataset, a propensity-matched analysis of 55,480 patients, found esketamine plus an SNRI associated with lower all-cause mortality (5.3% vs 9.1%), hospitalisation and depressive relapse than esketamine plus an SSRI, at the cost of a small absolute increase in suicide attempts (0.5% vs 0.3%) [72]. We would treat this as hypothesis-generating and would not switch a stable, responding patient on this basis. Matching was on age and sex only, which leaves the main confounder untouched: clinicians choose SNRIs and SSRIs for different patients. The reported 9.1% all-cause mortality in a cohort of mean age 46 is also difficult to reconcile with the same study’s very low hospitalisation rate, which raises a question about ascertainment across network sites. Vortioxetine is the one off-label partner with esketamine-specific outcome data, showing comparable efficacy with a numerically larger reduction in emotional blunting and fewer treatment-emergent adverse events in 10 patients per arm [73].

One pharmacokinetic finding has direct practical consequences. In 53 patients on 56 mg esketamine, serum concentrations at 20 minutes and total exposure were significantly higher in patients co-prescribed paroxetine, fluoxetine, duloxetine or venlafaxine than in those on sertraline, citalopram, escitalopram or vortioxetine. The difference did not affect clinical outcome, but the rise in systolic blood pressure correlated positively with serum esketamine, leading the authors to suggest a lower esketamine dose in patients with cardiovascular comorbidity on those four agents [74].

7.2. Benzodiazepines

Benzodiazepines are the concomitant drugs with the most evidence. The mechanistic prediction is that, because the first step is the removal of GABAergic inhibition, a GABA-A positive allosteric modulator should blunt the downstream cascade that carries the sustained effect while leaving the acute effect largely intact; intracortical muscimol has this effect in animals [75]. The human data broadly fit this prediction.

The effect appears to be dose-dependent, so studies that code benzodiazepine use as present or absent may miss it. The three strongest human datasets described below used intravenous racemic ketamine, not intranasal esketamine, so their thresholds apply to esketamine only by inference. In the only randomized dataset that separated the two approaches, benzodiazepine use as a binary predictor had no impact, whereas benzodiazepine dose as a continuous predictor did (p = 0.018 for depression severity at day 1) [76]. Observational data show the same pattern. Any use was equally distributed between responders and non-responders, but the diazepam-equivalent dose differed (7.7 ± 4.5 mg in responders vs 32.1 ± 24.9 mg in non-responders, p = 0.007); a receiver-operating-characteristic cut-point above 8 mg separated the groups with 80% sensitivity and 84.6% specificity, and the adjusted odds ratio for non-response was 9.7 (95% CI 1.2–69.7). Ketamine and norketamine plasma levels did not differ between groups, which points to a pharmacodynamic rather than a pharmacokinetic interaction [77]. In a small repeated-dose study, response rates were identical, but in benzodiazepine users response came 5 days later and relapse 16 days sooner [78]. Benzodiazepine use predicted delayed response in the machine-learning analysis [27] and influenced latent trajectory-class membership in two esketamine cohorts [79].

Set against this, the pooled analysis of the two ASPIRE trials, in which 68.5% of 451 patients used a concomitant benzodiazepine, found the 24-hour antidepressant effect fully intact in benzodiazepine users [80]. The result is reassuring, but it covers a narrower situation than is often assumed: benzodiazepines were prohibited within 8 hours before and 4 hours after the first dose and within 8 hours of the 24-hour efficacy assessment, so the trial tested background prescription with a protected peri-dose window against an acute endpoint, in patients in suicidal crisis. It says nothing about weeks 2–12 of ambulatory maintenance. The same analysis did find sedation more frequent with concomitant benzodiazepine use, while dissociation was unaffected.

Confounding may explain part of the signal. Benzodiazepine users had more severe baseline anxiety in the randomized trial, and the largest esketamine cohort found a main effect without a medication-by-time interaction: users scored about 4.5 MADRS points higher throughout, including at baseline [81]. This fits a more severe phenotype in benzodiazepine users as well as it fits a pharmacodynamic interaction. No study has tested whether tapering benzodiazepines before esketamine improves the response, and in our view this is the most important open question about concomitant medication. A taper is at least feasible around ketamine treatment. In 22 chronic benzodiazepine users with TRD given a planned 10–25% weekly taper, timed so that the last dose fell within a day or two of the first ketamine session, 20 (91%) had stopped entirely by week 4 and 14 (64%) remained abstinent at a mean of 12 months; depression, anxiety and suicidality improved during withdrawal [82].

7.3. Lithium, Antipsychotics and Lamotrigine

No human dataset shows that lithium augments or impairs the response. The one randomized trial designed to test lithium continuation after ketamine was negative on its primary endpoint (p = 0.91, 95% CI −7.87 to 8.76) [83], as was the largest esketamine cohort (β = 1.70, p = 0.343) [81], and a systematic review found no reported interaction [84]. A quarter of European patients receiving esketamine in routine practice take lithium, with no reported safety signal [69]. Lithium should be continued for its own indications, but there is no reason to add it to potentiate esketamine or to stop it beforehand.

Antipsychotics are the most common augmenting agents in practice, taken by 36.3% of patients at esketamine initiation [69], yet no study has reported antidepressant outcomes for esketamine combined with an atypical antipsychotic. A five-year cohort found antipsychotic side-effect burden negligible on esketamine, with median scores of zero from week 13 [62]. The concern that an antipsychotic will blunt dissociation and so reduce efficacy is not supported, because dissociation intensity does not predict antidepressant benefit (Section 8.1). Quetiapine is a special case: in a patient already taking extended-release quetiapine, the head-to-head trial gives randomized grounds for tapering it once esketamine is established [4].

With lamotrigine, the mechanistic and clinical data disagree. The concern that it attenuates the glutamate surge is supported by psychotomimetic and neuroimaging endpoints in healthy volunteers, and a systematic review concluded that lamotrigine “probably” reduces ketamine’s treatment outcome [84]. However, both studies that measured depression outcomes in patients were negative: 347 patients at a single centre with full covariate adjustment [85], and 178 patients in a second cohort (β = 1.04, p = 0.777) [81]. Lamotrigine appears to attenuate the psychotomimetic effects without attenuating the antidepressant effect, which is consistent with the separation between dissociation and response (Section 8.1). Lamotrigine should be continued.

7.4. Opioid Antagonists and Partial Agonists

Opioid receptor antagonism is seldom mentioned in clinical guidance, although several studies, including two randomized crossover trials, have examined it. In a crossover trial stopped at interim analysis, naltrexone 50 mg markedly attenuated the antidepressant effect of ketamine but left dissociation intact [86]. An independent randomized crossover study found that naltrexone attenuated both the anterior cingulate glutamatergic response and the day-1 reduction in depression [87]. The literature is not unanimous. A 2026 systematic review of 16 studies (12 clinical, n = 790) found that naltrexone attenuated ketamine’s effects in three, had no such effect in four and gave mixed results in one, and the only esketamine-specific test, an analysis of a functional µ-opioid receptor polymorphism in two phase 3 trials, was negative [88,89]. Both positive human trials used intravenous racemic ketamine. Even so, the selectivity of the effect matters clinically: a patient taking naltrexone or buprenorphine may feel the drug working and still not improve. This is better discussed before a course starts than after three weeks without improvement.

8. Preventing and Managing Adverse Effects

Most adverse effects of esketamine are predictable and can be managed without additional drugs. The large majority occur on dosing days and resolve the same day [90]. Discontinuation for adverse events is uncommon, at 4.6% versus 1.4% on placebo in the short term, and the commonest single cause is anxiety (1.2%), ahead of depression (0.9%) and of blood-pressure elevation and dizziness (0.6% each) [2]. Tolerability also improves with time: in a long-term cohort the proportion reporting at least one side effect fell from 71.8% at six months to 42% at twelve [91]. Table 5 gives a management protocol for each effect.

8.1. Dissociation

Dissociative and perceptual changes are the most visible effect of a session. Frequencies and time course are given in Table 5, and severity decreases over consecutive doses [92]. The main clinical question is whether dissociation predicts benefit, and the largest analyses indicate that it does not. In an analysis of phase 3 trial data, response rates at day 2 and day 28 did not differ between patients above and below a peak dissociation threshold, correlations were close to zero, and dissociation did not mediate symptom change or time to relapse [92]. A pooled analysis of 576 patients found an association of −0.04 MADRS points per dissociation point, which is statistically detectable but of no clinical relevance [93]. Earlier single-infusion work did report a positive correlation [94]. Dissociation may accompany response, but it is not required for it. The dose should not be titrated to a dissociative endpoint, and patients who do not dissociate should not be told that their dose is too low. A qualitative study adds the patient’s perspective. In 36 patients, the main experiential themes were psychological distance from suffering (83.3%), suspension of time (58.3%) and altered perception of body and space (55.6%), and most described the experience as neutral or meaningful rather than distressing. This supports structured psychoeducation, support during the session, and integration afterward [95].

8.2. Blood Pressure

Table 5 gives the size and time course of the pressor effect. What the labels do not give is a number at which to act. They also differ on duration: the US label puts it at about 4 hours, and the European label at 1–2 hours [2,3]. Neither label sets a hard numerical stop. Both treat a pre-dose pressure above 140/90 mmHg (above 150/90 mmHg at age 65 or over in the European label) as a threshold for a documented risk–benefit decision about delaying treatment, and both prohibit administration, in qualitative terms, when a rise in blood pressure or intracranial pressure poses a serious risk. Each clinic needs its own numerical rule. The best-sourced figures are the withholding and discontinuation criteria of the long-term phase 3 safety study, which should be described as trial-protocol thresholds and not as label requirements [90]. The US label also points out that a substantial rise can follow any dose, however well the earlier ones went, so monitoring should not be relaxed with experience [2].

8.3. Session Logistics that Prevent Problems

Three label instructions prevent many common problems. Patients should not eat for at least 2 hours or drink for at least 30 minutes before administration. The stated reason is nausea, not aspiration, so longer fasts are unnecessary. Any nasal corticosteroid or decongestant should be given at least 1 hour beforehand; under that condition, pharmacokinetic studies show no effect on esketamine exposure, and symptomatic rhinitis needs no dose adjustment [96]. Patients must not drive or operate machinery until the next day, after a restful sleep [2,3]. Outside the label, Martiadis and colleagues have published a phase-based operational framework with a minimum session protocol, documentation covering six domains, and escalation triggers that separate expected transient effects from events needing urgent review. They also note that the post-dose period is generally unsuitable for formal psychotherapy, which belongs in separate sessions after full recovery [97].

CNS, central nervous system. Frequencies are drawn from the US prescribing information and European summary of product characteristics unless otherwise indicated in the text; real-world figures are cited in the corresponding sections.

9. Esketamine in TRD with Comorbid Substance Use Disorder

A history of moderate-to-severe substance or alcohol use disorder was an exclusion criterion across the registration programme, and clinicians have reasonably been cautious. The evidence gathered since then is more reassuring for this comorbidity than for most of the other excluded groups.

On effectiveness, substance use disorder did not influence outcome in the pooled real-world meta-analysis [23], and the first REAL-ESK cohort found no difference in efficacy by comorbidity status [22]. The dedicated sub-analysis is small but directly relevant: in 26 patients with TRD and comorbid substance use disorder, esketamine produced a significant reduction in symptoms, with no dropouts and no recorded case of esketamine abuse or misuse under supervised administration [98].

On misuse liability, a structured narrative review covering clinical trials, real-world cohorts and high-risk subgroups found no documented cases of abuse, addiction or withdrawal over cumulative follow-up approaching 4.5 years, a single published case report of craving since market introduction, and no confirmed instance of illicit acquisition or diversion of the medicinal product [99]. Pharmacovigilance data point the same way. In a World Health Organization database replication study, reporting odds ratios for ketamine were elevated for most misuse parameters (substance use disorder 170.44, substance dependence 12.48, drug use disorder 11.54), whereas esketamine was not associated with an increased reporting odds ratio for any misuse parameter; its ratios were in fact reduced for substance abuse (0.41), drug dependence (0.083) and drug abuse (0.052) [100]. As the authors point out, reporting odds ratios do not establish causation when elevated or benefit when reduced. Still, the two compounds were compared in the same database and the same analysis, and the difference is large. Supervised administration is a plausible explanation, although these data cannot confirm it.

Not all findings are reassuring. A disproportionality analysis of 5132 esketamine reports flagged substance abuse among rare signals, based on 12 cases [101], and a critical meta-analysis still lists abuse potential among unresolved concerns [6]. Practice is also more restrictive than the evidence warrants: substance use disorder predicts a lower likelihood of esketamine initiation, and alcohol use disorder predicts more frequent treatment interruption [30].

A history of substance use disorder calls for assessment and monitoring, which is how both labels frame it, and we do not consider it a reason to exclude a patient [2,3]. The exception is lifetime ketamine use disorder, which the registration trials excluded and no study has examined. Active, untreated moderate-to-severe substance or alcohol use disorder is mainly a question of timing: the twice-weekly supervised protocol requires reliable attendance, so the disorder should be stabilised first.

10. Duration of Treatment and Management of Non-Response

10.1. Continuing Treatment After Response

The randomized-withdrawal trial remains the main source of evidence. Among patients who had completed 16 weeks of esketamine and achieved stable remission, relapse occurred in 26.7% of those continuing versus 45.3% of those switched to placebo spray (hazard ratio 0.49, 95% CI 0.29–0.84; number needed to treat 6). Among stable responders without remission, relapse occurred in 25.8% versus 57.6% (hazard ratio 0.30, 95% CI 0.16–0.55; number needed to treat 4) [102]. The benefit of continuing was larger in partial responders than in full remitters, and patients with residual symptoms have more to lose from stopping.

Long-term data are reassuring. In the open-label extension of 1148 patients with a mean exposure of 42.9 months and 3777 cumulative patient-years, remission rose from 35.6% at the end of induction to 49.6% at maintenance endpoint, cognition remained stable, there was no treatment-related interstitial or ulcerative cystitis, and discontinuation was 5.3% for lack of efficacy and 6.4% for adverse events [68]. Cognitive safety is independently supported by a pooled analysis of four phase 3 studies [103].

On duration, the European label recommends at least 6 months after depressive symptoms improve, with periodic re-evaluation; the US label specifies no minimum or maximum [2,3]. The 6-month figure derives from a small expert consensus rather than from trial data. The European Delphi panel frames continuation as 6–9 months and maintenance as 12 months or longer, with the decision driven by degree of worsening on tapering, relapse risk and consequences, chronicity and residual symptoms [70].

10.2. Stepping Down the Dosing Frequency

One maintenance dataset examined what happens when dosing frequency is reduced. When sessions were spaced from weekly to fortnightly, 19% of patients improved further, 49% maintained benefit and 32% worsened; among those who lost remission and returned to weekly dosing, 47% improved and only 10% worsened further [104]. Reducing frequency can be treated as a reversible trial with a planned review. A second full induction after relapse is also effective: in 96 patients who had relapsed and received a second induction, response was 71.9–73.4% and remission 60.9–62.5%, with no discontinuations for adverse events during that induction [105].

The European label states that discontinuation does not require tapering because the withdrawal risk is low [3], whereas expert consensus favours gradual reduction with regular monitoring [70]. The two positions are hard to reconcile, partly for a methodological reason: withdrawal symptoms and depressive relapse are symptomatically indistinguishable, so trials looking for “withdrawal” may have classified it as relapse. Reducing frequency instead of dose, with scheduled reviews, is a reasonable compromise.

10.3. Time to Response and Premature Discontinuation

In routine practice, many patients who will eventually respond have not yet done so at the end of induction. Response rises steeply between month 1 and month 3, more than doubling in the Italian consortium (28.4% to 64.2%) [22] and rising by about half in the French cohort (40.2% to 56.2%), where median time to response was 5.7 weeks (95% CI 4.1–8.4) [69]. Both figures are calculated among patients still in treatment, so attrition inflates them to an unknown degree, and the French ELLIPSE cohort found response and remission plateauing at month 2 rather than continuing to climb [24]. A cautious reading is that a meaningful minority of patients respond only after induction has ended. In randomized data, 54.9% of patients who had not responded by day 2 nonetheless responded by day 28 (odds ratio 1.61, 95% CI 1.09–2.40) [106]. The European Delphi panel accordingly defines the acute phase as lasting 4–12 weeks and reached consensus that even modest or subjective reduction in core symptoms justifies continuation, particularly with long disease course or multiple prior failures [70].

Response can come even later. In a long-term real-world cohort, 4 of 15 patients who were non-responders at six months had improved substantially by twelve [91]. This rests on four patients in an uncontrolled retrospective cohort in which only 19 reached twelve months, and it cannot by itself justify indefinite continuation in a non-responder. The predictors of six-month improvement in the same cohort, a shorter current episode and less psychiatric comorbidity, are more useful for setting expectations. In an inpatient cohort, extending induction beyond eight sessions appeared especially helpful for older patients and those with psychiatric comorbidity [71].

Yet treatment is often stopped much earlier. In the French cohort, 79.6% of patients discontinued within 12 months, at a median of 13.1 weeks, and 52.0% of them stopped because of unsatisfactory effect, at a median of 5.4 weeks, slightly before the median time to response of 5.7 weeks [69]. The European label, by contrast, recommends continuing for at least six months after improvement.

10.4. After Esketamine Failure

No prospective randomized trial has examined what to do after esketamine fails. Two indirect findings are relevant. First, an individual-participant-data comparison of two trials, after restriction and propensity matching, found rTMS and esketamine both superior to a new antidepressant, with rTMS numerically but not significantly better than esketamine (−2.46, 95% CI −5.82 to 0.89) [107]. The analysis is exploratory and non-randomized, but it supports rTMS as a reasonable next step. Second, esketamine can work in patients in whom ECT has failed: in 96 inpatients, history of ECT non-response had no significant interaction with treatment outcome over time, and response and remission did not differ significantly from those without a failed ECT course [38]. Prior ECT failure should therefore not exclude a patient from esketamine.

11. Who Should Not Receive Esketamine

There are few absolute contraindications, and the two agencies differ on some of them. Both labels prohibit use in hypersensitivity to esketamine or ketamine, in aneurysmal vascular disease (intracranial, thoracic or abdominal aortic, or peripheral) and in patients with a history of intracerebral haemorrhage. The European label adds a recent cardiovascular event (within 6 weeks), including myocardial infarction, as an absolute contraindication. The US label does not list it and instead asks prescribers to assess cardiovascular and cerebrovascular conditions carefully before prescribing [2,3]. The US label names arteriovenous malformation explicitly; the European text covers it through its qualitative clause. European prescribers are bound by the stricter rule, and we would apply it elsewhere too, in line with international expert opinion [108].

Beyond these, the main rule in both labels is qualitative: do not administer if a rise in blood pressure or intracranial pressure poses a serious risk. Severe hepatic impairment (Child–Pugh C) is “not recommended” in both. Pregnancy and breastfeeding are not recommended, with the European label adding an explicit contraception expectation; the two agencies differ on whether esketamine is known to be present in human milk, with the US label stating that it is and the European stating that this is unknown [2,3].

Contraindications That are Commonly Over-Applied

Some conditions are treated as contraindications even though the labels do not list them as such, and this keeps patients from a treatment that could help them. A history of substance use disorder is a monitoring requirement in both labels (Section 9). Controlled or treated hypertension is not a contraindication either; both labels ask for blood pressure to be optimised before starting. Rhinitis, nasal congestion or the use of a nasal steroid or decongestant is not a reason to cancel a session (Section 8.3). Established benzodiazepine therapy calls for sedation monitoring. A history of psychosis is listed as a caution in both labels, and bipolar disorder as a caution in the European label. Dissociation is a separate matter: it does not mean the dose is too high, and its absence does not mean the dose is too low. Dosing should follow clinical response and tolerability [92,97].

Table 6. Contraindications, cautions and commonly over-applied exclusions, with the divergences between the US prescribing information and the European summary of product characteristics.

Condition	Regulatory position	Practical note
ABSOLUTE CONTRAINDICATIONS
Hypersensitivity to esketamine, ketamine or excipients	Contraindicated (both agencies)	No exceptions
Aneurysmal vascular disease (intracranial, thoracic or abdominal aortic, peripheral)	Contraindicated (both agencies)	Consider the possibility of undiagnosed aortic disease in older patients with relevant risk factors
History of intracerebral haemorrhage	Contraindicated (both agencies)	No time limit is specified by either agency
Arteriovenous malformation	Explicitly contraindicated by the FDA; covered by the EMA’s qualitative clause	Treat as absolute in both jurisdictions
Recent (within 6 weeks) cardiovascular event, including myocardial infarction	Contraindicated by the EMA; not listed as a contraindication by the FDA, which asks for careful assessment before prescribing	The most clinically relevant difference between the labels. Expert opinion supports the stricter European position, and we recommend adopting it
Any condition in which a rise in blood pressure or intracranial pressure poses a serious risk	Qualitative absolute contraindication (both agencies)	A catch-all clause with no numerical threshold; document the clinical judgement
CAUTIONS REQUIRING DOCUMENTED RISK–BENEFIT ASSESSMENT
Other cardiovascular and cerebrovascular disease	Careful pre-prescribing assessment	Optimise blood pressure first; consider 56 mg and closer monitoring if on a CYP-inhibiting antidepressant
Current or prior psychosis	Caution (both agencies)	No usable data in either direction; document the reasoning
Mania or bipolar disorder	Caution (EMA)	Off-label. Maintain mood-stabiliser cover and monitor for switch at each visit
History of drug abuse or dependence	Assessment and monitoring required (both agencies)	Not an exclusion (Section 9)
Severe hepatic impairment (Child–Pugh C)	Not recommended (both agencies)	In moderate impairment, exposure is higher and the observation period should be extended
Pregnancy	Not recommended (both agencies); the EMA adds an explicit contraception expectation	If pregnancy occurs during treatment, discontinue and counsel. Balance against the risks of untreated depression in pregnancy
Breastfeeding	Not recommended (both agencies)	The FDA states that esketamine is present in human milk; the EMA states that this is unknown
Age ≥ 65 years	Not a contraindication; the EMA mandates a 28 mg starting dose	Higher exposure, greater falls risk and a higher rate of marked blood-pressure elevation. Data for age ≥ 75 are very limited
Age < 18 years	Safety and efficacy not established (both agencies)	The class suicidality warning carries its largest risk difference in this age group
COMMONLY OVER-APPLIED: NOT CONTRAINDICATIONS
Treated or controlled hypertension	Not a contraindication	The concern is uncontrolled hypertension; both labels ask for blood pressure to be optimised before starting
Rhinitis, nasal congestion, nasal steroid or decongestant use	Not a contraindication	No dose adjustment needed in symptomatic rhinitis; give nasal medication ≥ 1 h beforehand. Not a reason to cancel a session
Established benzodiazepine therapy	Not a contraindication	Monitor sedation and review the standing daily dose (Section 7.2)
Borderline personality disorder	Not a label exclusion	Excluded from the trials but not by the labels. Can be treated with an adapted protocol (Section 5.1)
Concurrent MAOI	Not a contraindication; blood-pressure monitoring required	No hypertensive crisis has been reported, but published experience is very limited
Prior ECT failure	Not a contraindication	Excluded from the trials; in real-world data, ECT non-responders improved as much as other patients

CYP, cytochrome P450; ECT, electroconvulsive therapy; EMA, European Medicines Agency; FDA, US Food and Drug Administration; MAOI, monoamine oxidase inhibitor. Label positions are drawn from the current US prescribing information and European summary of product characteristics; prescribers should verify against the version current in their jurisdiction at the time of prescribing.

Table 6. Contraindications, cautions and commonly over-applied exclusions, with the divergences between the US prescribing information and the European summary of product characteristics.

Condition	Regulatory position	Practical note
ABSOLUTE CONTRAINDICATIONS
Hypersensitivity to esketamine, ketamine or excipients	Contraindicated (both agencies)	No exceptions
Aneurysmal vascular disease (intracranial, thoracic or abdominal aortic, peripheral)	Contraindicated (both agencies)	Consider the possibility of undiagnosed aortic disease in older patients with relevant risk factors
History of intracerebral haemorrhage	Contraindicated (both agencies)	No time limit is specified by either agency
Arteriovenous malformation	Explicitly contraindicated by the FDA; covered by the EMA’s qualitative clause	Treat as absolute in both jurisdictions
Recent (within 6 weeks) cardiovascular event, including myocardial infarction	Contraindicated by the EMA; not listed as a contraindication by the FDA, which asks for careful assessment before prescribing	The most clinically relevant difference between the labels. Expert opinion supports the stricter European position, and we recommend adopting it
Any condition in which a rise in blood pressure or intracranial pressure poses a serious risk	Qualitative absolute contraindication (both agencies)	A catch-all clause with no numerical threshold; document the clinical judgement
CAUTIONS REQUIRING DOCUMENTED RISK–BENEFIT ASSESSMENT
Other cardiovascular and cerebrovascular disease	Careful pre-prescribing assessment	Optimise blood pressure first; consider 56 mg and closer monitoring if on a CYP-inhibiting antidepressant
Current or prior psychosis	Caution (both agencies)	No usable data in either direction; document the reasoning
Mania or bipolar disorder	Caution (EMA)	Off-label. Maintain mood-stabiliser cover and monitor for switch at each visit
History of drug abuse or dependence	Assessment and monitoring required (both agencies)	Not an exclusion (Section 9)
Severe hepatic impairment (Child–Pugh C)	Not recommended (both agencies)	In moderate impairment, exposure is higher and the observation period should be extended
Pregnancy	Not recommended (both agencies); the EMA adds an explicit contraception expectation	If pregnancy occurs during treatment, discontinue and counsel. Balance against the risks of untreated depression in pregnancy
Breastfeeding	Not recommended (both agencies)	The FDA states that esketamine is present in human milk; the EMA states that this is unknown
Age ≥ 65 years	Not a contraindication; the EMA mandates a 28 mg starting dose	Higher exposure, greater falls risk and a higher rate of marked blood-pressure elevation. Data for age ≥ 75 are very limited
Age < 18 years	Safety and efficacy not established (both agencies)	The class suicidality warning carries its largest risk difference in this age group
COMMONLY OVER-APPLIED: NOT CONTRAINDICATIONS
Treated or controlled hypertension	Not a contraindication	The concern is uncontrolled hypertension; both labels ask for blood pressure to be optimised before starting
Rhinitis, nasal congestion, nasal steroid or decongestant use	Not a contraindication	No dose adjustment needed in symptomatic rhinitis; give nasal medication ≥ 1 h beforehand. Not a reason to cancel a session
Established benzodiazepine therapy	Not a contraindication	Monitor sedation and review the standing daily dose (Section 7.2)
Borderline personality disorder	Not a label exclusion	Excluded from the trials but not by the labels. Can be treated with an adapted protocol (Section 5.1)
Concurrent MAOI	Not a contraindication; blood-pressure monitoring required	No hypertensive crisis has been reported, but published experience is very limited
Prior ECT failure	Not a contraindication	Excluded from the trials; in real-world data, ECT non-responders improved as much as other patients

CYP, cytochrome P450; ECT, electroconvulsive therapy; EMA, European Medicines Agency; FDA, US Food and Drug Administration; MAOI, monoamine oxidase inhibitor. Label positions are drawn from the current US prescribing information and European summary of product characteristics; prescribers should verify against the version current in their jurisdiction at the time of prescribing.

12. Conclusions

Esketamine is a useful addition to the treatment of TRD. Its average effect over placebo is smaller than its reputation suggests, and how much it helps in practice depends on decisions that the licensing trials did not address.

Patient selection should rest on clinical phenotype more than on the degree of resistance. Prominent anhedonia, anxious distress and mixed features are the most consistent positive signals. Resistance stage did not moderate the effect within the narrow range sampled in the trials, and greater resistance predicted a poorer response across the wider range seen in practice. Comorbidity, including BPD and substance use disorder, has not predicted non-response in most real-world datasets; the main exception is a retrospective cohort that combined intravenous ketamine and intranasal esketamine. Suicidal ideation is not a reason to withhold the drug; it often improves within hours, although esketamine does not replace crisis care.

We suggest titrating to the maximum tolerated dose without telling patients that a higher dose works better. The only fixed-dose phase 3 trial pointed the other way, no trial has formally compared 56 with 84 mg, adverse events increase with dose, and escalation has never been tested prospectively as a way to rescue non-response. Among concomitant drugs, the standing benzodiazepine dose matters more than the timing of doses on treatment days. Naltrexone should be discussed before starting, and lithium, lamotrigine and antipsychotics can be continued.

Dissociation is a tolerability problem to manage and should not be used as a therapeutic target; in the largest analyses it neither predicts nor mediates benefit. Anxiety during the session, more than dissociation, hypertension or nausea, is the commonest reason patients stop, so preparation, setting and continuity of staff are worth the effort.

The end of induction is too early to judge response. In routine practice the median time to response is close to six weeks and response rates keep rising into the third month, yet half of the patients who stop do so for lack of effect at about five weeks. Once response is achieved, continuing treatment prevents relapse, and the benefit is larger in partial responders than in remitters.

Several conditions often treated as contraindications are not listed as such in either label: a history of substance use disorder, treated hypertension, rhinitis, established benzodiazepine therapy, comorbid BPD and prior ECT failure. Recent cardiovascular events, on the other hand, are handled differently by the two agencies, and the stricter European position is the safer one to follow.

Many of the most practical questions remain untested. No trial has examined whether tapering benzodiazepines improves the esketamine response, whether dose escalation rescues non-responders, what to do after esketamine fails, or how to treat the comorbid presentations that now make up much of the treated population. Until such trials are done, practice will have to rely on real-world cohorts and expert consensus, with the uncertainty stated openly.