Abstract

Treatment-resistant depression is depression that has not responded well enough to adequate antidepressant treatment, so it is defined by treatment history rather than by symptoms. No definition has won agreement, staging models grade how resistant a case is, and some apparent cases reflect inadequate or untaken treatment or an unrecognised bipolar illness. Remission grows less likely with each further treatment step, and the condition is linked to poorer quality of life, lost work and strain on families. One study of older adults found broad cognitive deficits, and the treatment evidence covers added medicines, added psychotherapy, ketamine and esketamine, and brain stimulation, each with its own design limits. Psilocybin remains investigational, with early and mixed results.

Keywords: treatment-resistant depression, antidepressant response, augmentation, electroconvulsive therapy, ketamine

Treatment-resistant depression (TRD) is depression in which an antidepressant trial has produced an inadequate response (Fava, 2003). MeSH indexes it with refractory depression, therapy-resistant depression and treatment resistant depression as entry terms. In research on major depression, the working threshold is usually the failure of at least two antidepressants. An expert-group review found, however, that no definition has won consensus (McIntyre et al., 2023). This page covers depression that has not responded to adequate treatment; depressive mood in general is covered on the depression page, and the most common depressive diagnosis is covered on the major depressive disorder page. The condition matters because of its scale and cost. An industry-sponsored US cost model estimated that about 2.8 million of 8.9 million adults taking medication for major depressive disorder met its definition of TRD, roughly one in three. That definition was a proxy built from insurance claims, with no symptom data behind it. In the model, this group accounted for nearly half of the total modelled burden (Zhdanava et al., 2021).

The page begins with what treatment-resistant depression is and why its definition is still unsettled. It then describes the types the cited studies distinguish, from pseudo-resistance and unrecognised bipolarity to late-life and psychotic presentations. Next come staging and measurement, and then prevalence, course and burden, which is where the STAR*D trial of successive treatment steps is reported. The page then turns to cognitive features and other markers of non-response. The treatment sections cover switching, augmentation and psychotherapy, ketamine and esketamine, and brain stimulation, and emerging treatments, chiefly psilocybin, come after them. A section on living with treatment-resistant depression follows. It covers quality of life, work, caregivers and relationships, and what patients say about their treatment. Suicide risk comes next. A worked example then shows how to read a trial result: the risk difference, the number needed to treat and the confidence interval. The Discussion and Current Directions weigh what is settled and what remains open.

Key Takeaways

  • There is no agreed definition; one review counted 155 definitions across 150 studies. Research on major depression usually requires two failed antidepressants, and bipolar depression often only one. Staging models grade resistance by degree instead of yes or no.
  • The Maudsley staging method scores the treatments tried, how severe the episode is and how long it has lasted. Its developers judged its validity reasonable, but evidence that its scores predict outcome was only partial and came from one specialist setting.
  • Apparent resistance can come from treatment that was too short, too weak or not taken, or from a bipolar illness nobody has recognised. That is why the diagnosis is checked again first. The bipolarity evidence comes from a non-blind follow-up study and a literature review.
  • In the largest trial of successive treatment steps, remission became less likely at each further step. Relapse was more common in patients who needed more steps, in a follow-up that was not controlled. How many patients remitted overall is disputed.
  • People with treatment-resistant depression report poorer quality of life, more lost work and more hospital use than people whose depression responded. These survey data are self-reported and cannot show cause. Small interview studies describe strained relationships and how much preparation and support matter during treatment.
  • In one study of adults aged 60 and older, scores fell below norms for holding back automatic responses, processing speed and memory, but not for working memory or switching between rules. This gave the contested executive dysfunction hypothesis only partial support.
  • Approved atypical antipsychotics, thyroid hormones and lithium appeared better than placebo as add-on medicines in a large pooled analysis. Antipsychotic benefits were modest and may be inflated by trial design. Adding cognitive behavioural therapy to usual care roughly doubled response in one primary care trial in which everyone knew who received therapy.
  • Ketamine and esketamine act rapidly. For ketamine, its obvious effects may let patients guess their treatment and so inflate the measured benefit, and long-term safety data are scarce for intravenous ketamine. Electroconvulsive therapy beat simulated treatment in the short term, and the cognitive effects that were measured were concentrated in the first three days. Magnetic stimulation beat sham, vagus nerve stimulation evidence is observational, and deep brain stimulation did not beat sham in its blinded trial.

What Treatment-Resistant Depression Is

MeSH defines treatment-resistant depression as “failure to respond to two or more trials of antidepressant monotherapy or failure to respond to four or more trials of different antidepressant therapies” (MeSH, D061218). The descriptor, Depressive Disorder, Treatment-Resistant, was introduced in 2012, sits under depressive disorder, and lists refractory depression, therapy-resistant depression and treatment resistant depression as entry terms.

A Definition by Treatment History

Treatment-resistant depression (TRD) is defined by the course of treatment, not by a distinctive set of symptoms. A trial here means a course of one antidepressant given at an adequate dose for an adequate length of time. In Fava's narrative review, the label typically applies when at least one such trial produces an inadequate response (Fava, 2003). The diagnosis therefore does not describe a way of feeling that sets these patients apart from other depressed patients. It records what has already been tried and has not worked.

Research usage has settled on a numerical threshold. An expert-group narrative review gives the usual criterion in major depression as a minimum of two antidepressants that have failed (McIntyre et al., 2023). An expert-opinion review of ketamine and esketamine frames its population the same way, as patients with two prior treatment failures (McIntyre et al., 2021). The bar is lower for bipolar depression. A systematic review of definitions found one failed treatment to be the typical threshold there, against two in major depressive disorder (Gaynes et al., 2020). The term should also be kept apart from persistent depressive disorder. That is a separate construct, defined by how long the depression lasts rather than by how it has responded to treatment.

No Consensus Definition

The two-failure threshold is a convention, not an agreed standard. One response has been to propose difficult-to-treat depression, a broader concept for depression that continues to carry a significant burden despite the usual treatment efforts. Its authors judged it clinically useful. They also noted that it rests largely on expert consensus, and that it may lack the specificity and objectivity needed to define the populations of regulatory trials (McAllister-Williams et al., 2020).

A second response has been a definition built for research. A Delphi consensus is a structured method in which a panel of experts rates proposed statements until a stated level of agreement is reached. Using it, one guideline defined treatment-resistant depression as less than 25% improvement after two failed treatments, each lasting at least four weeks (Sforzini et al., 2022). It defined partially responsive depression as improvement of 25% to under 50%. Consensus was strong on the resistance definition and moderate on partial response. The authors acknowledged that the recommendations had not been objectively validated (Sforzini et al., 2022).

How fragmented the field is shows in a count of the definitions in use. A systematic review, paired with interviews of key informants (people chosen for their expert knowledge of the topic), found 155 distinct definitions of TRD across 150 studies from 27 countries (Brown et al., 2019). Its authors found no universal definition and marked variation among those in use. They also reported that the term itself is rarely used in clinical practice. Their conclusion was a call for a single definition shared by researchers and clinicians, not a proposal for one. The review's reach is limited in ways that bear on that conclusion. It covered only publications from 2016 and 2017 and only papers in English and French, and it drew on only six key informants, all in Calgary (Brown et al., 2019).

A second systematic review reached the same verdict. It examined both definitions and trials published through July 5, 2019 (Gaynes et al., 2020). Beyond the count of failed treatments, it found no consensus on what makes an antidepressant trial adequate, in either dose or duration. It also noted that measures of daily functioning and quality of life were rarely used as outcomes. Gaynes and colleagues concluded that the field needs a consensus definition together with a core package of outcomes. The expert group behind the 2023 review likewise treated the lack of an agreed definition as an open problem (McIntyre et al., 2023). The question remains unresolved: none of these reviews reports that a shared definition has been adopted.

Pseudo-Resistance and Re-Evaluation

Not every case that looks resistant is resistant. Pseudo-resistance is an apparent lack of response that does not reflect true treatment failure. Non-adherence, meaning not taking the medication as prescribed, is one of its recognised sources (McIntyre et al., 2023). Because the definition rests entirely on treatment history, an inadequate trial or an untaken drug can produce the label as readily as a genuinely refractory illness.

For this reason Fava regarded re-evaluating the diagnosis as an essential step when a patient fails to respond. His review also linked both medical and psychiatric comorbidity, the presence of other illnesses alongside the depression, to treatment resistance (Fava, 2003). One diagnostic question of this kind, unrecognised bipolarity, is taken up in the Types of Treatment-Resistant Depression section below.

Types of Treatment-Resistant Depression

MeSH lists no narrower descriptors under treatment-resistant depression, so this section does not reproduce an indexing tree. The types below are the boundary presentations and clinical subgroups that the cited studies themselves distinguish. Some are not resistance at all. They are conditions that imitate it. Others are subgroups in which resistance is more or less likely. The categories overlap, and one patient can belong to several.

Table 1

Boundary Presentations and Subgroups of Treatment-Resistant Depression

TypeStatusKey source
Pseudo-resistanceBoundary case: apparent, not true, resistanceMcIntyre et al., 2023; Fava, 2003
Unrecognised bipolarityBoundary case: naturalistic and review evidenceSharma et al., 2005; Correa et al., 2010
Treatment-resistant bipolar depressionBoundary case: rarely defined in the literatureBrown et al., 2019
Anxious distressSubgroup: poorer acute outcomes, secondary analysisFava et al., 2008
Late-life treatment-resistant depressionSubgroup, aged 60+: broad deficits against normsAinsworth et al., 2025
Psychotic featuresSubgroup: better ECT outcome, depression generallyvan Diermen et al., 2018

Pseudo-Resistance

Not every failure to improve is resistance. Pseudo-resistance is an apparent lack of response that does not reflect true treatment resistance. An expert-group narrative review traced it to causes such as treatment trials that were too short or too weak, or patients not taking the treatment as prescribed (non-adherence). The same review noted that the field still has no consensus definition of the condition (McIntyre et al., 2023). Fava's earlier review framed treatment-resistant depression as an inadequate response to an antidepressant trial. It held that re-evaluating the diagnosis is essential, which is why the first step with an apparently resistant patient is to check the diagnosis and whether the treatment given was adequate (Fava, 2003). Fava also linked both medical and psychiatric comorbidity, meaning other illnesses present at the same time, to treatment resistance. He noted too that the proposed methods for grading resistance into stages had not been tested empirically (Fava, 2003). Both sources are narrative reviews. They describe clinical reasoning; they do not measure how often pseudo-resistance occurs.

Cognitive Implications

The cited review addressed definitions and treatments, not cognition. Its forward-looking conclusion was that digital therapeutics could be a possible future option (McIntyre et al., 2023). No cited study has tested whether patients with pseudo-resistance differ cognitively from those with true resistance.

Unrecognised Bipolarity

A depression that fails one antidepressant after another may belong to the bipolar disorder spectrum without anyone having recognised it. Sharma and colleagues followed 61 clinic patients for a year or longer each. Their resistant depression had arisen after an earlier antidepressant response was lost, and it had been managed as unipolar. At intake, 35% had a bipolar diagnosis; by follow-up, 59% did, and 80% showed bipolarity of some kind. The authors judged that most supposedly unipolar resistant cases probably carry a bipolar tendency (Sharma et al., 2005). Diagnoses used the Structured Clinical Interview for DSM-IV. Data collection was not blind, however, which leaves room for expectation to shape the reassessment (Sharma et al., 2005).

A later literature review reached a more measured view. Correa and colleagues surveyed 196 articles published between January 1998 and January 2008. They concluded that the evidence gives at least moderate support to unrecognised bipolar disorder playing a part in depression that only appears resistant, with high rates of it found among resistant patients (Correa et al., 2010). The review was more cautious about the warning signs clinicians often assume. It found the evidence inconclusive on whether relapsing while on antidepressants points to the bipolar spectrum. Among patients first diagnosed with major depressive disorder, switching into mania or hypomania on these drugs was not common (Correa et al., 2010). The included studies had not been designed to test the hypothesis, and only English-language articles were searched (Correa et al., 2010).

Cognitive Implications

The review concerned diagnosis and course, including subtle soft signs of bipolarity, rather than cognitive performance (Correa et al., 2010). Neither cited study examined cognition. Whether unrecognised bipolarity carries a distinct cognitive profile within resistant depression is therefore not addressed here.

Treatment-Resistant Bipolar Depression

Resistance is defined differently once bipolar disorder is established. A systematic review of definitions and trials found that definitions for bipolar depression often require only one failed treatment, against two for major depressive disorder. It also found no consensus on what counts as an adequate dose or an adequate length of treatment (Gaynes et al., 2020). The definitional literature rarely studies resistant bipolar depression on its own. Across the patient populations of 150 studies, 1.3% were treatment-resistant bipolar depression. By comparison, 42.7% were unipolar depression, 30.7% major depressive disorder and 25.3% mixed samples (Brown et al., 2019). That review covered only publications from 2016 and 2017, in English and French (Brown et al., 2019). Patients with bipolar II disorder have been included in intervention research. One open-label long-term follow-up of subcallosal cingulate deep brain stimulation, in which both patients and clinicians knew the treatment given, enrolled 28 patients. Seven had bipolar II disorder, and all were followed for 4 to 8 years (Crowell et al., 2019). The section on Brain Stimulation reports its outcomes.

Cognitive Implications

The definitional review did not examine cognition. It noted that measures of daily functioning and quality of life were rarely used as outcomes (Gaynes et al., 2020). The cited evidence therefore leaves cognitive outcomes specific to resistant bipolar depression unaddressed.

Anxious Distress

Prominent anxiety marks a subgroup that does worse on standard treatment. The evidence comes from a secondary analysis of the first two levels of STARD, a large multisite trial of sequential antidepressant treatment. Anxious depression was defined as a score of 7 or higher on the anxiety/somatization factor of the 17-item Hamilton Depression Rating Scale. Of 2,876 adult outpatients starting citalopram, 53.2% met that threshold (Fava et al., 2008). These patients were less likely than nonanxious patients to reach remission*, the near-absence of symptoms. They also took longer to remit and carried a heavier burden of side effects and adverse events (Fava et al., 2008). The disadvantage persisted at Level 2, where 1,292 patients were either switched to a new drug or given a second one alongside the first (augmentation). Anxious patients fared worse under both strategies (Fava et al., 2008). The authors framed the result as an association with poorer acute outcomes. Because the analysis was secondary, it was not designed to test anxiety as a cause of non-response (Fava et al., 2008). The definition is also a rating-scale threshold, so the subgroup it picks out is not identical to a diagnostic specifier.

Cognitive Implications

The STAR*D analysis measured remission, response and side effects on clinical rating scales. It did not assess cognition (Fava et al., 2008). The cited study did not test whether anxiety-related cognitive processes explain the poorer outcomes.

Late-Life Treatment-Resistant Depression

In older adults, resistant depression comes with measurable cognitive deficits. The evidence is a secondary analysis of the OPTIMUM randomized trial, run at four sites in the US and one in Canada. It tested 369 participants aged 60 and older with the NIH Toolbox-Cognition Battery (Ainsworth et al., 2025). Against population norms, the group scored lower on inhibitory control (holding back an automatic response), processing speed, memory and the fluid cognition composite. It did not score lower on working memory or on set shifting (switching flexibly between rules). Premorbid IQ, an estimate of ability before the illness taken from oral reading, was high (Ainsworth et al., 2025). Older age and physical comorbidity were each associated with slower processing speed (Ainsworth et al., 2025). Impaired set shifting predicted less improvement, whereas impaired inhibitory control did not predict response (Ainsworth et al., 2025). The section on Cognitive Features, Bias and Non-Response sets this study beside the meta-analytic evidence.

Cognitive Implications

The OPTIMUM result gives only partial support to the executive-function account of late-life non-response, which holds that problems with higher-level control of thinking predict a poor outcome. One executive measure predicted outcome and another did not. The authors read the poorer outcomes linked to impaired set shifting as a reason to identify alternative treatment strategies (Ainsworth et al., 2025). A meta-analysis pooling eight studies of outpatients aged 50 and older with late-life depression found a similarly selective picture. Only the planning and organization domain separated responders from nonresponders. The effect was large but imprecise (g = 0.91, a standardized difference between the groups, with a 95% CI of 0.32-1.58, a wide plausible range). The other five domains, cognitive flexibility and response inhibition among them, did not reach significance (Pimontel et al., 2016). The planning estimate rested on a single study, and the treatment designs differed. The authors therefore suggested only that planning-focused therapies may be worth testing as an addition to antidepressants (Pimontel et al., 2016). That meta-analysis concerned antidepressant response in late-life depression generally, not resistant depression. Its null result for cognitive flexibility also sits uneasily with the set-shifting finding in OPTIMUM. Neither study settles which executive process matters.

Psychotic Features

Psychotic features point the other way from most subgroups: they predict a better response to electroconvulsive therapy (ECT). A meta-analysis of 34 studies, with 3,276 patients treated with ECT for depression, found that psychotic features predicted higher rates of both remission and response, as did older age (van Diermen et al., 2018). The patients were people with depression treated with ECT, not specifically people with resistant depression (van Diermen et al., 2018). The comparison with ketamine has a gap here. The randomized trial testing whether ketamine is noninferior to ECT, that is, not meaningfully worse, enrolled only patients without psychosis. Its finding that ketamine was noninferior therefore does not extend to psychotic depression (Anand et al., 2023). The section on Brain Stimulation reports that trial in full.

Cognitive Implications

The meta-analysis examined clinical predictors of ECT outcome, not cognition. Besides psychotic features and age, symptom severity predicted response with a small effect but did not predict remission. Melancholic features did not predict outcome, though the melancholia data were few and inconsistently defined (van Diermen et al., 2018). The cited evidence does not address the cognitive profile of psychotic resistant depression.

Staging and Measurement of Treatment Resistance

Every claim about treatment-resistant depression rests on two measurements. The first is how resistant a given patient is. The second is whether a treatment produced response (a clear fall in symptom scores) or remission (symptoms reduced to a minimal level). Staging models address the first question, and symptom rating scales address the second.

Staging Models

Staging treats resistance as a matter of degree. A staging method does not sort patients into resistant and non-resistant groups. It places them along a graded scale of treatment resistance (Fava, 2003). When Fava's narrative review set out this idea, the evidence had not yet caught up with it. The review judged the available staging methods untested. It argued that they needed empirical evaluation before anyone could rely on them. It also held that a fresh diagnostic assessment is essential before a patient's depression is labelled resistant at all (Fava, 2003).

Several staging instruments now appear in the research literature. One systematic review examined 150 studies and added interviews with six key informants. Staging models were referenced in 38.1% of the definitions it collected, most often the Thase and Rush Staging Model (Brown et al., 2019; Thase & Rush, 1995). Its two validation studies used the Antidepressant Treatment Response Questionnaire and an extension of the Maudsley Staging Model (Brown et al., 2019). The review sampled only publications from 2016 and 2017, in English and French. It therefore shows which tools were in use over a short window, not how widely each has been adopted.

The Maudsley Staging Method

The Maudsley staging method is multidimensional and scored in points. It rates resistance on three factors: the treatments a patient has received, the severity of the episode, and the episode's duration. The factor scores add up to a single stage (Fekadu et al., 2009). Severity is scored from 1 to 5. Treatment failures sit on a five-point subscale, and the duration of the current episode on a three-point subscale. The combined total therefore runs from 3 to 15 (Fekadu et al., 2009).

The developers tested the method empirically. They judged its face validity (whether it looks as if it measures resistance) and its predictive validity (whether its scores forecast later outcomes) to be reasonable. They hedged both claims. The predictive validity they reported was only partial. They also presented the grouping of totals into resistance categories as something that could possibly be specified, not as a settled set of cut-offs (Fekadu et al., 2009). The method thus fills part of the gap Fava identified, with evidence from a single specialist setting.

How a Points-Based Staging Score Adds Up

Staging grades treatment resistance by degree. Move each slider to see how three subscales add to a single score.

313

Total on the three modelled subscales: 5 (range 3 to 13).

Note. Subscale ranges follow Fekadu et al. (2009); the labels for each level are illustrative. The published method's total runs from 3 to 15, and this demonstration models only the three subscales shown. It is not a clinical tool.

Rating Response and Remission

Fava's review noted that clinician-rated or self-rated instruments may help in assessing treatment resistance systematically (Fava, 2003). The trials cited in this article used different instruments, so remission does not mean exactly the same thing in each.

STAR*D moved 3,671 adult outpatients with nonpsychotic major depressive disorder through successive treatment steps. A patient counted as remitted on scoring 5 or below on the QIDS-SR16, a self-report symptom questionnaire. The Hamilton Rating Scale for Depression, a clinician-rated scale, was also given (Rush et al., 2006a).

A trial comparing esketamine with quetiapine in 676 patients set remission at a Montgomery-Åsberg Depression Rating Scale (MADRS) score of 10 or less at week 8. The trial was open-label, meaning patients and clinicians knew which treatment was given, and its raters were single-blind (Reif et al., 2023). An open-label trial comparing ketamine with electroconvulsive therapy in 403 patients assessed response with the QIDS-SR. It measured memory separately, with the Hopkins Verbal Learning Test-Revised (Anand et al., 2023).

Symptom scales dominate this literature. A systematic review of definitions and trials found no consensus on what counts as an adequate dose or duration for a failed treatment. It found functional and quality-of-life measures rare. It called for a consensus definition with a core package of outcomes (Gaynes et al., 2020).

Prevalence, Course and Burden

How often treatment-resistant depression (TRD) occurs depends on how it is defined. No definition has won consensus. An expert group's narrative review found no agreed definition, although failure on a minimum of two antidepressants keeps recurring as a working threshold (McIntyre et al., 2023). Each figure below therefore carries its authors' chosen definition. The largest estimates come from models, not from finding cases by their symptoms.

How Common It Is

The most detailed recent US estimate comes from a cost model, not a patient survey. The model started from 8.9 million adults taking medication for major depressive disorder. It estimated that 2.8 million of them (30.9%) met its TRD definition (Zhdanava et al., 2021). That definition was a proxy: it was built from insurance claims, with no symptom data behind it. The model also combined claims studies from four payers with the Kantar National Health and Wellness Survey, and these sources used different methods (Zhdanava et al., 2021). The share varied by payer, at 44.2% for Medicaid against 21.4% for Medicare (Zhdanava et al., 2021). The study was industry-sponsored, and the sponsor helped shape its design. Its totals are best read as one modelled estimate among other possible ones (Zhdanava et al., 2021).

An earlier narrative review defined TRD as an inadequate response to an adequate antidepressant trial. It linked both medical and psychiatric comorbidity, meaning other illnesses present alongside the depression, to resistance (Fava, 2003). The same review stressed that apparent resistance calls for a fresh look at the diagnosis before it is accepted as real (Fava, 2003).

Remission Across Successive Treatment Steps

STARD was a large US trial that offered depressed outpatients up to four treatment steps in turn, moving on when a step did not work. It enrolled 3,671 adult outpatients with nonpsychotic major depressive disorder. Remission, meaning symptoms low enough to count as well, was defined as a score of 5 or less on the QIDS-SR16, a self-report symptom questionnaire (Rush et al., 2006a). Remission rates fell at each step: 36.8% at the first, 30.6% at the second, and 13.7% and 13.0% at the third and fourth (Rush et al., 2006a). The trial reported a cumulative remission rate, the share of patients who remitted at any step, of 67% (Rush et al., 2006a). A 12-month follow-up then tracked relapse*, a return of depression after improvement. Relapse was more common among patients who had needed more steps. Patients who entered follow-up in remission relapsed less often than those who had not remitted, a contrast reported for steps 1 to 3 (Rush et al., 2006a). The follow-up was naturalistic rather than controlled, so these patterns describe the cohort and do not isolate the effect of any treatment (Rush et al., 2006a).

A 2023 reanalysis, part of the Restoring Invisible and Abandoned Trials initiative, scored remission on the Hamilton Rating Scale for Depression (HRSD), the outcome the original protocol named. On that scale, cumulative remission was 35.0%, about half the reported 67% (Pigott et al., 2023). Of 4,041 adults enrolled, the reanalysis included 3,110. It left out 99 patients who were already in remission on the HRSD at the start and 125 who were in remission on entering their next step; the original report had counted both groups (Pigott et al., 2023). Adding the QIDS-SR to the HRSD raised the cumulative figure to 41.3%. Remission by step fell from 25.5% at step 1 to 21.3%, 13.2% and 10.4% at steps 2 to 4, each lower than the originally reported rate (Pigott et al., 2023). Dropout was higher at step 3 than at step 1. Response, a smaller degree of improvement than remission, was 40.5% at the first step. The first-step remission and response rates both sat below the 48.4% and 65.2% averaged across comparator trials of 7,030 patients (Pigott et al., 2023). The reanalysis notes its own limits. It used descriptive statistics only, 1,330 patients had no exit HRSD, and the protocol was silent on four points the authors had to decide themselves (Pigott et al., 2023). The two cumulative figures rest on different scales and different rules about which patients counted. STAR*D was also open-label, meaning patients and clinicians knew which treatment was given, and it had no control group. That design limits the original reading and the reanalysis alike, since neither can show how much of the remission the treatments caused (Pigott et al., 2023).

Cumulative Remission Depends on Who Stays

Choose a set of step-wise remission rates, then set how many patients who have not remitted go on to the next step.

Step 136.8%Step 230.6%Step 313.7%Step 413.0%

Cumulative remission across four steps: 67.1%.

Note. Reported step rates from Rush et al. (2006a); reanalysed rates from Pigott et al. (2023), whose cumulative figure was 35.0%. The cumulative value here is computed by assuming that each step's remission rate applies to everyone who reaches it. Original schematic.

Burden

The burden figures come from the same cost model. They are modelled estimates, not observed spending. The model put the total yearly burden of medication-treated major depressive disorder at $92.7 billion. TRD accounted for $43.8 billion of it (47.2%), well above its 30.9% share of patients (Zhdanava et al., 2021). TRD's share was largest for health care costs, at 56.6% ($25.8 billion), and smaller for lost productivity, at 32.2% ($9.3 billion) (Zhdanava et al., 2021). The model also assigned TRD a disproportionate share of the unemployment burden (Zhdanava et al., 2021). Some costs fell outside it. The model extended data from insured patients to the uninsured, and it left out losses in quality of life and the burden on caregivers (Zhdanava et al., 2021). Its effects on daily life, work and families are described in the Living with Treatment-Resistant Depression section.

Cognitive Features, Bias and Non-Response

Cognitive measures enter the study of treatment resistance in two ways. They describe what patients who have not recovered can and cannot do, and they are tested as possible markers of who will fail the next treatment. Most of this evidence comes from older adults. None of it shows that a cognitive profile causes resistance.

Executive Function and Processing Speed

One cognitive profile of treatment-resistant late-life depression comes from the OPTIMUM study. This was a multicentre randomized trial run at four sites in the United States and one in Canada. Ainsworth and colleagues reanalysed its cognitive testing, done with the NIH Toolbox-Cognition Battery, in 369 participants aged 60 and older (Ainsworth et al., 2025). They compared the scores with normative data, that is, with the typical scores of people of similar age. Participants scored below norms on inhibitory control (holding back an automatic response), processing speed, memory and a combined fluid cognition score (Ainsworth et al., 2025). Working memory and set shifting (switching between tasks or rules) were at norms. Estimated premorbid IQ, the ability level before illness as indexed by oral reading, was high (Ainsworth et al., 2025). Within the sample, older people and those with more physical illness tended to have slower processing speed (Ainsworth et al., 2025).

The analysis tested the executive dysfunction hypothesis. This is the proposal that impaired executive function, the set of control processes that plan and regulate behaviour, underlies poor antidepressant response in later life. The idea traces to an earlier randomized maintenance trial in 58 elderly patients whose depression had remitted (Alexopoulos et al., 2000). In that trial, abnormal initiation and perseveration predicted relapse and recurrence over up to two years of follow-up, while memory impairment did not. Its authors said the result needed confirmation (Alexopoulos et al., 2000). The hypothesis remains contested, and the OPTIMUM data supported it only in part. Impaired set shifting went with less improvement. Inhibitory control, although below norms, did not predict response (Ainsworth et al., 2025). The authors took the poorer outcomes linked to impaired set shifting as a reason to look for alternative treatment strategies (Ainsworth et al., 2025).

A second line of evidence is a meta-analysis, a study that pools the results of earlier studies. It combined eight studies of outpatients aged 50 and older with late-life depression, who were not selected for treatment resistance, using a Bayesian hierarchical model (Pimontel et al., 2016). Its 25 executive measures were sorted into six domains. Only planning and organization separated the groups: non-responders did worse than responders (Pimontel et al., 2016). The effect size, Hedges' g, expresses a group difference in standard-deviation units. Here it was large (g = 0.91), but its 95% confidence interval, the range of values compatible with the data, ran widely from 0.32 to 1.58. The other five domains did not differ reliably (Pimontel et al., 2016). The planning estimate rested on a single study. Many study-by-domain cells were empty, treatment designs varied, and there were too few studies to test moderators (Pimontel et al., 2016). The authors suggested, as a possibility rather than a finding, that therapies aimed at planning might add to the effect of antidepressants (Pimontel et al., 2016).

Attention, Memory and Interpretation Biases

None of the studies cited in this article tested attentional, memory or interpretation biases in samples defined by treatment resistance. Whether such biases differ between resistant and responsive patients is therefore untested here. The biases found in depression in general are described on the depression page, and they are not assumed to carry over.

Predictors of Non-Response

A history of childhood maltreatment is one of the better-documented risk markers. Nanni and colleagues pooled two sets of studies, reporting results as odds ratios (Nanni et al., 2012). An odds ratio compares the odds of an outcome in one group with the odds in another, so a value of 2 means the odds are twice as high. Across 16 epidemiological studies with 23,544 participants, people who had been maltreated were more likely to have recurrent or persistent depression (odds ratio 2.27; 95% confidence interval 1.80 to 2.87). Across 10 clinical trials with 3,098 participants, maltreatment also raised the odds of non-response or non-remission (odds ratio 1.43; 95% confidence interval 1.11 to 1.83) (Nanni et al., 2012). Neither result changed materially with publication bias, study quality, the outcome measure, sample age or the use of prevalence rather than incidence samples (Nanni et al., 2012).

Anxious depression, described under Types of Treatment-Resistant Depression, is a second marker. Fava and colleagues ran a secondary analysis of 2,876 STAR*D outpatients treated first with citalopram (Fava et al., 2008). Patients were classed as anxious when they scored 7 or higher on the anxiety/somatization factor of the 17-item Hamilton Depression Rating Scale; 53.2% met that mark. Compared with nonanxious patients, they reached remission (near-absence of symptoms) less often and more slowly, and they reported more side effects (Fava et al., 2008). The gap persisted among the 1,292 patients who moved on to a second step of switching or augmentation (Fava et al., 2008).

These markers are associations seen in cohorts and trial samples. They are not proven causes of resistance, and a maltreatment history or an anxious presentation does not by itself decide a person's outcome. A recent expert review concluded that few factors predict non-response consistently across treatment types (McIntyre et al., 2023).

Treatment Evidence: Switching, Augmentation and Psychotherapy

When a first antidepressant fails, clinicians have three broad moves. They can replace it, add something to it, or combine it with a psychological treatment. Rapid-acting glutamatergic agents, which act on the brain's main excitatory signalling chemical, now form a fourth option. The evidence for each differs in quantity and in quality, and the subsections below take them in turn.

Switching Antidepressants

Changing to a different drug is the simplest next step in psychopharmacology. The largest test of it found no clear winner. The second level of the STARD trial enrolled 727 adult outpatients with nonpsychotic major depression whose selective serotonin reuptake inhibitor (SSRI) had failed (Rush et al., 2006b). They were randomly assigned to bupropion-SR (239 patients), sertraline (238) or venlafaxine-XR (250) for up to 14 weeks, at 18 primary care and 23 psychiatric settings. Remission* means that symptoms fall to a minimal level. On the clinician-rated 17-item Hamilton scale (HRSD-17), remission rates for the three drugs were 21.3%, 17.6% and 24.8%. On the self-rated QIDS-SR-16 they were 25.5%, 26.6% and 25.0% (Rush et al., 2006b). None of these differences was significant, and the groups did not differ in tolerability or adverse events (Rush et al., 2006b). The authors concluded that any of the three is a reasonable second step. The sobering qualifier is that only about one patient in four remitted, whichever drug was chosen (Rush et al., 2006b).

Augmentation

Augmentation means adding a second agent, usually one that is not itself a standard antidepressant, to a drug the patient keeps taking. The broadest comparison is a network meta-analysis, a method that pools trials and also compares treatments indirectly through their shared comparisons with placebo. It covered 65 randomised trials in 12,415 adults and ranked 19 augmentation agents for treatment-resistant depression, over periods of up to 24 weeks (Nuñez et al., 2022). Ten agents beat placebo for response, a substantial drop in symptoms, and six did so for the stricter outcome of remission. The approved atypical antipsychotics, thyroid hormones and lithium came out as the strongest options, and modafinil also beat placebo (Nuñez et al., 2022). Timing mattered for two of them. Thyroid hormone given for three weeks or less was no better than placebo. Lithium over the same short window still was. Ziprasidone, mirtazapine and cariprazine had more all-cause discontinuation than placebo, meaning more patients stopped them for any reason (Nuñez et al., 2022).

The authors were explicit about the limits of the ranking. Trials defined treatment resistance in different ways and ran for different lengths of time. The older studies had weaker designs, and few trials informed any single comparison. Only 24% of studies were at low risk of bias, against 63% at moderate and 13% at high risk. The authors called for head-to-head trials before any agent is declared best (Nuñez et al., 2022).

Lithium has the longest record. An earlier meta-analysis included 9 of 11 placebo-controlled, double-blind studies, 234 patients in all (Bauer & Döpfmer, 1999). Its main estimate came from the three studies (110 patients) with adequate dosing: at least 800 mg/day, or a serum level of 0.5 mEq/L, for at least 2 weeks. In those core studies, the odds of response with lithium were 3.31 times the odds with placebo, an odds ratio of 3.31. Its 95% confidence interval ran from 1.46 to 7.53, so the true effect could be modest or large. The absolute difference in response rates was 27%. The number needed to treat was 3.7, meaning that about four patients need lithium for one more to respond than would on placebo (Bauer & Döpfmer, 1999). The authors judged lithium a first-choice option after antidepressant failure. Only three studies met their dosing criteria, however (Bauer & Döpfmer, 1999).

Adjunctive antipsychotics have a contested reading. A meta-analysis pooled 14 short-term placebo-controlled trials in treatment-resistant depression, each lasting 4 to 12 weeks (Spielmans et al., 2013). Adding an antipsychotic lowered observer-rated depression, but only modestly. The standardised effect size, Hedges' g, ranged from 0.26 to 0.48, a small-to-moderate benefit. Odds ratios for remission were 2.01 for aripiprazole, 1.79 for quetiapine, 2.37 for risperidone and 1.42 for the olanzapine-fluoxetine combination. The number needed to treat was nine for each of the first three drugs and 19 for olanzapine-fluoxetine (Spielmans et al., 2013). For response, olanzapine-fluoxetine did not differ significantly from placebo (odds ratio 1.30). Quality of life barely improved with any drug except risperidone (Spielmans et al., 2013). The authors urged caution. They argued that shortcomings in trial design and data reporting may inflate the apparent benefit. Several analyses were post hoc, chosen after the data were in, and adverse events may be undercounted (Spielmans et al., 2013).

A Cochrane review of second-generation antipsychotics covered 28 trials with 8487 participants across five drugs (Komossa et al., 2010). It did not pool the trials into one estimate. Its scope also went beyond resistant depression, to major depression and dysthymia in general. Adding aripiprazole, quetiapine or risperidone beat placebo, and quetiapine also beat placebo on its own. The MADRS (Montgomery-Åsberg Depression Rating Scale) is a clinician-rated measure of depression severity. With aripiprazole added to an antidepressant, MADRS scores fell by a mean of 3.04 points more than with placebo. Olanzapine augmentation helped on some outcomes but not others (Komossa et al., 2010). The symptom gains came at the cost of more side effects. The reviewers called the data very limited and mostly short-term. They also noted that the risperidone samples were small and that the drugs' manufacturers funded most of the trials (Komossa et al., 2010).

Psychotherapy Added to Medication

Psychological treatment has its strongest evidence in resistant depression as an addition to medication, not a replacement for it. The CoBalT trial randomised 469 adults aged 18-75 years from 73 UK general practices (Wiles et al., 2013). All were still depressed despite antidepressant treatment, with a Beck Depression Inventory score of 14 or more. Of these, 234 received cognitive behavioral therapy on top of usual care, and 235 continued usual care alone. At six months, 46% of the CBT group had responded, against 22% of the usual-care group. Response roughly doubled. The odds ratio was 3.26 (p < 0.001), so chance is a very unlikely explanation. Follow-up reached 90% of participants at six months and 84% at twelve (Wiles et al., 2013). The design could not mask the treatment, however. Both participants and assessors knew who had received therapy (Wiles et al., 2013).

From Two Percentages to a Number Needed to Treat

Pick a comparison. The number needed to treat is how many patients would need the first option, rather than the second, for one more patient to reach the outcome.

CBT plus usual care46.0%Usual care22.0%
Outcomeresponse
Difference24.0 percentage points
Number needed to treat5

Wiles et al. (2013). No masking was possible.

Note. Percentages are those reported by each study; the number needed to treat is computed here as one divided by the difference, rounded up. Studies differ in design, outcome and follow-up, so the numbers are not comparable across rows.

Whether psychotherapy alone can treat resistant depression is a separate and open question. An expert review found it not established as a stand-alone treatment for this group (McIntyre et al., 2023).

Ketamine and Esketamine

The first controlled test was tiny. Seven depressed patients took part in a randomised, double-blind crossover trial (Berman et al., 2000). In a crossover design, each person receives both treatments on different occasions. Here there were two test days, with 0.5 mg/kg ketamine on one and saline on the other. Over the next 72 hours, scores on the 25-item Hamilton scale fell by 14 points (SD 10) after ketamine and by 0 points (SD 12) after saline. The authors suggested only that drugs acting on the NMDA receptor, a glutamate receptor, may help depression (Berman et al., 2000).

A 2006 crossover trial then tested the same dose in 18 people with treatment-resistant major depression, with the test days a week apart (Zarate et al., 2006). Ketamine beat placebo within 110 minutes. The effect size d expresses the group difference in standard-deviation units. It was 1.46 at 24 hours and had fallen to 0.68 by one week. Of the 17 people treated with ketamine, 71% responded and 29% remitted the day after the single intravenous dose, and 35% kept their response for at least a week (Zarate et al., 2006).

As off-label use spread, meaning prescription outside a drug's licensed indication, a consensus statement reviewed the case for ketamine in treatment-resistant mood disorders. It accepted the rapid antidepressant effect. It also noted that the trials were small, longer-term efficacy data were lacking, and safety data were limited (Sanacora et al., 2017).

The intranasal enantiomer, esketamine, was then tested in licensing trials. (An enantiomer is one of the two mirror-image forms of a molecule.) In a phase 3, double-blind trial with an active comparator, 227 adults at 39 outpatient centres took an antidepressant plus either esketamine nasal spray (56 or 84 mg twice weekly) or a placebo spray (Popova et al., 2019). By day 28 the esketamine arm had improved more on the MADRS. The model-adjusted gap, a least-squares mean difference of -4.0 points, favoured esketamine. Adverse events were more common in that arm, though. Seven percent of the esketamine arm stopped for an adverse event, against 0.9% of the comparator arm (Popova et al., 2019).

A later phase 3b trial compared esketamine with quetiapine in 676 patients (Reif et al., 2023). At week 8, 27.1% of the esketamine group had remitted, defined as a MADRS score of 10 or less, against 17.6% of the quetiapine group, a significant difference. At week 32, more of the esketamine group were in remission without relapse (21.7% vs 14.1%) (Reif et al., 2023). The trial was open-label, so patients knew which drug they had, and only the raters were blind to it. It also counted dropouts as unfavourable outcomes (Reif et al., 2023).

The comparison between the two forms of ketamine rests on pooled data. A meta-analysis of 24 randomised trials in 1877 depressed adults split the results by formulation (Bahji et al., 2021). It reported risk ratios, the rate of an outcome in the treated group divided by the rate in the control group. Intravenous racemic ketamine, an equal mix of both mirror-image forms, had larger risk ratios than intranasal esketamine for response (3.01 vs 1.38) and for remission (3.70 vs 1.47). The two did not differ significantly in change in depression score or in suicidality. Whether trials enrolled treatment-resistant or other patients did not significantly change the outcomes either (Bahji et al., 2021). The authors read intravenous ketamine as apparently more efficacious. Yet heterogeneity was high, meaning results varied widely from trial to trial. Correcting for publication bias, the tendency of positive results to reach print more often, also made the effects smaller (Bahji et al., 2021). The approval evidence for esketamine has also been criticised. One commentary set out seven concerns about the efficacy data and about the FDA's decision to approve it for treatment-resistant depression (Turner, 2019).

Expert and guideline bodies have tempered their endorsement with safety warnings. An expert review noted the FDA approval of esketamine in March 2019. It found the rapid efficacy of both drugs well supported, while safety questions remain open (McIntyre et al., 2021). Its authors pointed out that the obvious acute effects can unblind participants, so expectancy may inflate the measured benefit. Dissociation, a sense of detachment from oneself or one's surroundings, was reported in 72% of intravenous studies against 36% of non-intravenous ones. It was not necessary for an antidepressant response. Blood pressure rose with dose, and esketamine did not separate from placebo in patients aged 65 or older. Whether long-term ketamine exposure causes neurotoxicity remains unknown (McIntyre et al., 2021). The Canadian CANMAT task force reviewed the literature to January 31, 2020. It found Level 1 evidence, its highest grade, for a single intravenous dose of racemic ketamine and recommended it as a third-line option. It rated the evidence for repeated infusions as limited and for non-intravenous forms as weak. It also flagged misuse and diversion, dissociative symptoms and hypertension as risks (Swainson et al., 2021).

How ketamine relieves depression is still debated, and the field holds several hypotheses at once. One review groups them into two broad families (Zanos & Gould, 2018). The first centres on blocking NMDA receptors, including those on GABAergic interneurons, the inhibitory cells that damp local activity. The second centres on downstream AMPA receptor activation and the plasticity signalling that follows, involving the growth factor BDNF and the mTOR pathway. Still other accounts attribute part of the effect to the metabolite (2R,6R)-HNK rather than to ketamine itself (Zanos & Gould, 2018). The authors argue that these mechanisms need not exclude one another and may work together. They also note that ketamine's abuse potential and dissociative effects limit its use (Zanos & Gould, 2018). A broader review presents ketamine as the first antidepressant to work rapidly, an action that came to light in people whose symptoms had resisted treatment (Krystal et al., 2019). Neither review settles which mechanism carries the clinical effect, and the question remains open.

Brain Stimulation: ECT, rTMS, VNS and Deep Brain Stimulation

When several medication trials have failed, the next options usually act on the brain directly. They work through electrical current, magnetic pulses or implanted electrodes. The methods differ widely in how invasive they are, how long they have been studied and how strong their controlled evidence is. The four subsections below take them roughly in that order of evidence, from the most extensively tested to the least settled.

Electroconvulsive Therapy

Electroconvulsive therapy (ECT) induces a brief, controlled seizure. It has the deepest evidence base of any treatment in this section. One systematic review and meta-analysis gathered randomised trials and observational studies, and its short-term efficacy analysis drew on the randomised trials (UK ECT Review Group, 2003). A meta-analysis pools the results of many studies into a single estimate. Here the pooled result was reported as a standardised effect size, which expresses a difference in standard-deviation units so that studies using different scales can be combined. Across six trials with 256 patients, real ECT was more effective than simulated ECT, with a standardised effect size of -0.91 (UK ECT Review Group, 2003). ECT also came out ahead of drug therapy in 18 trials with 1144 participants (standardised effect size -0.80), and higher doses did better than lower ones (UK ECT Review Group, 2003). The reviewers worded their verdict cautiously. ECT probably beats medication, they concluded, and their pooled evidence covers only short-term efficacy (UK ECT Review Group, 2003).

A clinical review, built around a single case, called ECT the most effective acute treatment for severe depression (Kellner et al., 2012). The same review acknowledged that ECT remains contested, that mistaken beliefs about it persist and that many clinicians do not know how modern ECT is given (Kellner et al., 2012). A later review reached a similar verdict. It described ECT as effective, especially in resistant depression, and treated the seizure as required for the therapeutic effect (Espinoza & Kellner, 2022). It also acknowledged that the mechanism is unknown, that memory effects are usually transient and that stigma still restricts use (Espinoza & Kellner, 2022).

Some patients respond better than others. A meta-analysis of 34 studies covered 3276 depressed patients receiving ECT (van Diermen et al., 2018). Psychotic features predicted both remission and response, with odds ratios of 1.47 and 1.69 for the two outcomes (van Diermen et al., 2018). An odds ratio above 1 means the outcome was more likely in the group with the feature. Response means a substantial fall in symptoms, and remission means symptoms fall to a minimal level. Older patients also did better than younger ones on both measures (van Diermen et al., 2018). Greater severity predicted response only weakly and did not significantly predict remission. Melancholic features showed no reliable link to outcome, although the authors judged the melancholia data inconclusive because few studies used consistent definitions (van Diermen et al., 2018). The psychotic subtype is described under Types of Treatment-Resistant Depression.

ECT is now being compared with newer rapid-acting treatments. One trial randomly assigned 403 patients with treatment-resistant major depression and no psychosis, at five sites, to either ketamine or ECT (Anand et al., 2023). It was open-label, meaning that patients and clinicians knew which treatment was given. It was also a noninferiority trial, designed to test whether ketamine was not meaningfully worse than ECT rather than whether it was better. On the QIDS-SR, a self-report depression scale, 55.4% of the ketamine group responded against 41.2% of the ECT group (Anand et al., 2023). The gap was 14.2 percentage points, and even the low end of its 95% confidence interval (3.9 points) lay well above the preset margin of -10 points. That result established ketamine as noninferior (Anand et al., 2023). Patients given ECT showed a fall in memory recall that eased gradually over follow-up. Quality of life improved by a comparable amount in the two arms (Anand et al., 2023). Because the trial was open-label and excluded psychotic depression, it does not show that ketamine can replace ECT in psychotic depression.

Keeping a remission after ECT is a separate problem. A six-month multisite randomised trial enrolled 201 patients with unipolar depression who had remitted with ECT (Kellner et al., 2006). They received either continuation ECT (10 treatments) or continuation medication. Relapse occurred in 37.1% of the continuation-ECT group and 31.6% of the medication group, a difference that was not significant (Kellner et al., 2006). Remission was sustained in about 46% of each arm, and time to relapse did not differ significantly either (Kellner et al., 2006). Both strategies looked better than a historical placebo comparison, but the trial had no placebo arm of its own. More than half the sample relapsed or dropped out. The investigators therefore concluded that both approaches have limited efficacy and that better strategies are needed (Kellner et al., 2006).

Why ECT works is still debated. A narrative review weighed whether the benefit comes from the seizure or from the electric field that produces it (Deng et al., 2024). The question is hard to answer because in conventional ECT the two always occur together. Some findings favour the seizure. Low-dose right unilateral ECT, which still produces seizures, works less well than high-dose right unilateral ECT. Patients with stronger postictal suppression, a dampening of brain electrical activity right after the seizure, responded more often (74% against 55%) (Deng et al., 2024). Evidence on the field points mainly to side effects. Hippocampal field strength had a direct effect on impairment of phonemic fluency, the ability to produce words beginning with a given sound, but was not directly related to antidepressant outcome (Deng et al., 2024). Magnetic seizure therapy induces a seizure with a much smaller field. It has shown efficacy with minimal cognitive impairment, which the reviewers read as support for the seizure being key to efficacy (Deng et al., 2024). The review treated the anticonvulsant, neurogenesis and amnesia hypotheses as unproven. It noted that a confirmatory trial of magnetic seizure therapy is still underway and left the seizure-versus-field question open until pending trials report (Deng et al., 2024).

Cognitive Effects of ECT

A meta-analysis of 84 studies covering 2981 depressed patients examined standardised cognitive tests (Semkovska & McLoughlin, 2010). The tests covered processing speed, working memory, anterograde memory (the learning of new material) and executive function. In the first three days after a course, 72% of the cognitive variables were worse than before treatment (Semkovska & McLoughlin, 2010). Between 4 and 15 days, performance was back near pretreatment levels. Beyond 15 days, 57% of the variables showed improvement over baseline (Semkovska & McLoughlin, 2010). After the first three days, neither electrode placement nor stimulus waveform significantly predicted cognitive outcome (Semkovska & McLoughlin, 2010). The analysis has one important limit. It could not assess retrograde amnesia, the loss of memory for events before treatment, because no standardised tests of it were available (Semkovska & McLoughlin, 2010).

Repetitive Transcranial Magnetic Stimulation

Transcranial magnetic stimulation uses a magnetic coil held against the scalp to induce current in the cortex beneath it. Its repetitive form (rTMS) usually targets the left prefrontal cortex. Earlier trials gave mixed results, partly because their sham conditions were poorly masked. A sham condition imitates the treatment without delivering it, so that expectation alone can be measured. A multisite trial addressed the masking problem with a sham coil carrying a metal insert (George et al., 2010). Among 199 drug-free outpatients with major depressive disorder, 14.1% remitted with active 10 Hz left prefrontal stimulation, against 5.1% with sham (George et al., 2010). The number needed to treat was 12, meaning that about 12 patients had to receive active stimulation for one additional remission. Adverse effects did not differ between arms, and most remitters had low antidepressant treatment resistance (George et al., 2010).

A meta-analysis focused on treatment-resistant depression itself. It pooled 18 sham-controlled trials in adults who had failed at least two antidepressants (Gaynes et al., 2014). Scores on the Hamilton depression scale, a clinician-rated measure of symptom severity, fell by 4.53 points more with rTMS than with sham (Gaynes et al., 2014). A relative risk gives how many times as likely an outcome is in one group as in the other. Response was more than three times as likely with rTMS (relative risk 3.38), and remission about five times as likely (relative risk 5.07); raw remission rates were 30% against 6% (Gaynes et al., 2014). The authors rated the evidence for severity and response as high and for remission as moderate. They called rTMS a reasonable option after two failed medications (Gaynes et al., 2014). Two features of the data temper that conclusion. Effect sizes fell with publication year, by about 0.45 Hamilton points per year. No included trial followed patients for more than a week, so it is unclear whether the benefit lasts (Gaynes et al., 2014).

A network meta-analysis compared different rTMS methods across 81 studies and 4233 patients (Brunoni et al., 2017). This kind of analysis ranks several treatments at once, including pairs that were never tested head to head. Five methods produced higher response rates than sham: priming low-frequency, bilateral, high-frequency, theta-burst and low-frequency stimulation, with odds ratios from 2.37 to 4.66 (Brunoni et al., 2017). Accelerated, synchronised and deep rTMS did not separate from sham. Differences among the effective methods were small. The estimates were imprecise and several methods rested on few trials, so the authors called for larger randomised trials (Brunoni et al., 2017).

Theta-burst stimulation delivers short, rapid bursts of pulses and shortens each session. A randomised multicentre trial of 414 adults with treatment-resistant depression tested its intermittent form (iTBS) (Blumberger et al., 2018). A 3-minute iTBS session was compared with 37.5 minutes of 10 Hz rTMS over the left dorsolateral prefrontal cortex. Scores on the 17-item Hamilton scale fell to 13.4 in both arms, from baselines of 23.6 and 23.5 (Blumberger et al., 2018). That outcome met the noninferiority margin of 2.25 points. Dropout and side effects were similar, although participants rated iTBS as slightly more painful (Blumberger et al., 2018). Because each device could treat many more patients, the authors saw a capacity gain. Two limits qualify the result. The stimulation was given open-label, although the investigators and the outcome assessors did not know group assignments. The trial's main analysis also counted participants per protocol, that is, only those who completed treatment as planned (Blumberger et al., 2018).

Accelerated, individually targeted protocols go further still. In the open-label SAINT series, 22 participants with treatment-resistant depression received 50 iTBS sessions over five days, ten a day at 50-minute intervals (Cole et al., 2020). Each participant's target was a left dorsolateral prefrontal site chosen by resting-state functional connectivity with the subgenual anterior cingulate. Functional connectivity is the degree to which activity in two brain regions rises and falls together. Of the 21 who completed treatment, 19 (90.5%) met the remission criterion on the Montgomery-Åsberg Depression Rating Scale (MADRS), a clinician-rated severity scale (Cole et al., 2020). By intention to treat, which counts everyone who started, the rate was 86.4%. The regimen was well tolerated. With no sham arm and a small sample, the authors stated that controlled confirmation was needed (Cole et al., 2020).

An earlier double-blind randomised trial had already found the accelerated protocol superior to sham. In a planned interim analysis of 29 participants with treatment-resistant depression, MADRS scores fell by 52.5% with active stimulation (N=14) and by 11.1% with sham (N=15) (Cole et al., 2022). Whether the benefit lasted was left to further trials (Cole et al., 2022). A second double-blind sham-controlled trial of the protocol, now called Stanford neuromodulation therapy (SNT), randomised 48 adults to test whether that result would replicate (Kratter et al., 2026). Remission reached 50.0% with active treatment against 20.8% with sham, and response 54.2% against 25.0%; both differences were statistically significant (Kratter et al., 2026). The trial also recorded the electroencephalogram (EEG), the brain's electrical activity measured at the scalp. Frontal beta power fell significantly with active treatment but not with sham, and reductions in left anterior cingulate beta tracked clinical improvement (Kratter et al., 2026). The authors offered beta suppression as a possible mechanism and baseline beta power as a candidate biomarker. They presented the trial as a replication of the efficacy that an earlier randomized controlled trial had shown (Kratter et al., 2026). The striking open-label rates from SAINT therefore have controlled support. The controlled effects, however, are smaller and come from two small trials.

Vagus Nerve Stimulation

Vagus nerve stimulation (VNS) is a stimulation treatment whose evidence in depression is mainly observational. A five-year registry at 61 U.S. sites followed 795 patients with treatment-resistant depression (Aaronson et al., 2017). The patients received either adjunctive VNS, added to their other care, or treatment as usual. Cumulative response was 67.6% with VNS against 40.9% with treatment as usual. First-time remission was 43.3% against 25.7% (Aaronson et al., 2017). The VNS advantage held both in patients who had previously responded to ECT and in those who had not. The VNS group also started out more severely ill, with baseline MADRS scores of 33.1 against 29.3 (Aaronson et al., 2017). Assignment was not random and the registry was open-label. These figures therefore show an association with added benefit and do not establish that VNS caused it (Aaronson et al., 2017).

Deep Brain Stimulation

Deep brain stimulation (DBS) is a surgical stimulation treatment; earlier surgical approaches to psychiatric illness are covered under psychosurgery. The first series in depression stimulated the subgenual cingulate (Brodmann area 25) in six patients with refractory illness (Mayberg et al., 2005). Four of the six reached remission. Positron emission tomography, an imaging method that can track regional blood flow, showed flow falling in the stimulated region as the depression lifted (Mayberg et al., 2005). The authors proposed that electrically interrupting abnormal activity within limbic-cortical circuits might reverse treatment-resistant depression (Mayberg et al., 2005).

The sham-controlled test did not confirm the effect. A multisite trial randomised 90 patients with treatment-resistant depression to active stimulation of subcallosal cingulate white matter (60) or sham (30) (Holtzheimer et al., 2017). In the blinded phase, response did not differ significantly between the arms. The trial was stopped after a futility analysis, an interim check of whether a positive result was still likely, on half its planned sample (Holtzheimer et al., 2017). It was sponsored by the device maker, Abbott. In uncontrolled open-label follow-up, response was 40% at 12 months, 51% at 18 months and 48% at 24 months, with remission between 17% and 25% (Holtzheimer et al., 2017). A separate cohort of 28 patients given subcallosal cingulate deep brain stimulation was followed openly for four to eight years. Most kept an antidepressant response, but just 21% showed continuous response from the first year onward (Crowell et al., 2019).

The evidence on DBS is therefore split. Open-label series and follow-up describe lasting benefit for many implanted patients (Mayberg et al., 2005; Crowell et al., 2019). The only sham-controlled trial, however, found no difference from sham in its blinded phase (Holtzheimer et al., 2017). DBS for treatment-resistant depression thus remains unproven under blinded conditions, and the open-label gains come from designs without a control condition.

Emerging Treatments

Psilocybin is an investigational treatment for treatment-resistant depression. Its evidence base is early and mixed. The first test in this population was a feasibility study with an open-label design and no control group (Carhart-Harris et al., 2016). Open-label means that patients and investigators both knew what was being given. Twelve patients with unipolar treatment-resistant depression received a 10 mg dose and a 25 mg dose, 7 days apart (Carhart-Harris et al., 2016). Symptoms were rated on the 16-item Quick Inventory of Depressive Symptoms (QIDS). One week on, scores sat 11·8 points below baseline (p=0·002), and at 3 months they were still 9·2 points lower (p=0·003) (Carhart-Harris et al., 2016). The matching effect sizes were Hedges' g=3·1 and g=2 (Carhart-Harris et al., 2016). Hedges' g expresses a change in units of the scores' spread, so a larger value means a larger change. The investigators also assessed anxiety, anhedonia and psychedelic effects. Those effects were rated as more intense at the high dose (0·75) than at the low dose (0·51) (Carhart-Harris et al., 2016). The authors read the results as preliminary support only and called for rigorous trials (Carhart-Harris et al., 2016). Without a comparison condition, an uncontrolled study cannot show how much of the change the drug caused.

A larger test followed. It was a phase 2 randomized controlled trial, double-blind (Goodwin et al., 2022). The 233 adults with treatment-resistant depression received 25 mg or 10 mg of psilocybin, or a 1 mg control dose, with psychological support as part of the protocol (Goodwin et al., 2022). The outcome was the Montgomery-Åsberg Depression Rating Scale (MADRS), a depression scale on which lower scores mean fewer symptoms. By week 3, scores had fallen from baseline by 12.0 points at 25 mg, 7.9 points at 10 mg and 5.4 points at 1 mg (Goodwin et al., 2022). Only the highest dose beat the control. The 25 mg group improved 6.6 points more than the 1 mg group (95% CI, -10.2 to -2.9; P<0.001) (Goodwin et al., 2022). The 10 mg group's difference of 2.5 points was not significant (P = 0.18; its confidence interval [-6.2, 1.2] includes zero) (Goodwin et al., 2022). Its confidence interval, the range of plausible true differences, includes zero. The early gain at 25 mg did not become a sustained response at 12 weeks (Goodwin et al., 2022). Adverse events were common, and suicidal ideation was reported across dose groups (Goodwin et al., 2022). The trial did not compare psilocybin with existing treatments. Its authors concluded that larger and longer trials are needed (Goodwin et al., 2022). Taken together, the two studies give psilocybin an early and mixed standing. A single large dose beat a minimal dose in the short term, a moderate dose did not, and lasting benefit has not been shown.

Digital therapeutics are more speculative still. An expert-group narrative review named them as a potential future option for treatment-resistant depression (McIntyre et al., 2023). None of the evidence drawn on in this article reports a trial of a digital therapeutic in treatment-resistant depression. Their standing is that of a proposal, not a tested treatment.

Living with Treatment-Resistant Depression

Statistics on response rates say little about what treatment-resistant depression (TRD) does to a working week, a household or a person's sense of self. Two kinds of evidence speak to that. Surveys and reviews compare people with TRD against people whose depression responded, and they measure quality of life, work and use of services. Qualitative interview studies ask patients and the people close to them what the condition and its treatments were like. The second kind describes experiences in a few people's own terms; it cannot say how common any of those experiences are.

Quality of Life, Work and Daily Functioning

People with TRD report poorer health and more lost work than people whose depression is not resistant. The largest single source is a retrospective, cross-sectional analysis of the National Health and Wellness Survey, which drew 52,060 respondents from five European countries (Jaffe et al., 2019). Of its 3308 respondents with major depressive disorder, 622 (18.8%) met the study's TRD definition and 2686 did not. Relative to the general population, adjusted SF-12v2 scores were lower in both depressed groups: the mental component by 18.1 with TRD and 12.1 without it, and the physical component by 5.4 and 2.5, so health-related quality of life was lowest with resistance. After adjustment, TRD carried more absenteeism (rate ratio 1.53), more hospitalisation (odds ratio 1.79) and more emergency-room visits (odds ratio 1.53), as well as more presenteeism, meaning reduced productivity while still at work. Against the general population, the odds of visits to health care providers were 15.9 for TRD and 5.4 for non-resistant depression. Visits to general practitioners did not differ significantly between the two depressed groups, and nor did sex, smoking or alcohol use. The authors read the findings as a large unmet need in Europe. The limits are substantial: every measure was self-reported, a cross-sectional design cannot show that resistance caused the burden, TRD subtypes could not be examined, and Janssen-Cilag funded the study.

The same gradient appears when resistance is treated as a matter of degree. A systematic review of PubMed and Medline, covering 12 economic studies and 19 studies of health-related quality of life, found that direct and indirect costs rose and quality of life fell at each higher level of treatment resistance (Johnston et al., 2019). The review's own caveats limit how far the pattern can be read. TRD was defined differently from study to study, which made comparisons between studies hard, and the review could not say what caused the burden, nor whether a longer or more severe illness, rather than resistance as such, explained it. Its authors concluded that better treatments for TRD are needed.

A review of US data points the same way. Drawing on 62 English-language articles published from 1996 to 2013, with 59,462 patients in total, it found quality of life in TRD lower by .41 than in people whose depression had remitted and lower by .26 than in people who had responded (Mrazek et al., 2014). The same review set TRD against treatment-responsive depression and found an excess of $5,481 in health care costs and $4,048 in lost productivity. Because it was confined to US sources, the figures describe one health care system.

Caregivers and Relationships

The burden reaches the household. Thirty-five people caring for someone with severe TRD that needed in-patient treatment were compared with 23 controls in a case-control study, and the carers reported high burden and high psychiatric caseness (Rane et al., 2012). The carers also had lower cortisol after waking than the controls, on a stress-hormone measure taken from saliva, though the design cannot show that caring caused the difference (Rane et al., 2012). An opinion review on the concept of TRD gives a caregiver caseness figure of 34% against 4% in the comparison it reports, and it treats caregiver burden as part of the overall weight of the condition (Demyttenaere & Van Duppen, 2019). As an opinion piece, it argues a position from selected literature rather than pooling studies systematically.

Successful treatment can strain relationships too. In a prospective qualitative study of deep brain stimulation (DBS) targeting the bed nucleus of the stria terminalis, six patients and five caregivers gave 21 semi-structured interviews, analysed thematically, covering the period up to nine months after stimulation began (Thomson et al., 2023). Two patients described a self that felt restored, three spoke of modest or short-lived benefit with glimpses of who they had been before, and two reported no benefit. Benefit went with a restored sense of self and with accepting the device. Relationships, however, did not simply improve. Accounts included irritability and anger (three participants), and four patients reported memory difficulties; the authors related the difficulty of moving from chronic illness to wellness to the burden of normality, a concept from earlier neurological DBS research. The sample was too small for robust quantitative analysis, five of its six patients were women, all had extreme resistance (Stage V), and only two had sustained meaningful benefit at follow-up. The authors concluded that response tends to come gradually and that patients' own narratives should be collected.

Experiences of Treatment

Interviews about newer treatments find that how a session is set up matters to patients as much as the drug or the device. Seventeen people with TRD, 11 of them women, described a six-week, twice-weekly course of oral esketamine in in-depth interviews analysed by interpretative phenomenological analysis (Breeksema et al., 2022). The central theme was letting go of control: sessions in which patients let go were experienced better than sessions in which they resisted, and preparation, support and the setting all helped them let go. Patients found a private room more comfortable than a shared one. The authors suggested that better preparation, setting and support might improve the treatment. The sample was small and gave no basis for statistical generalisation, only people who completed treatment were interviewed, and the esketamine was oral and off-label at two sites.

A similar study interviewed 11 people with TRD (8 women and 3 men, aged 24 to 66) who had taken part in a trial of psilocybin at 1, 10 or 25 mg, again using interpretative phenomenological analysis (Breeksema et al., 2024). Three themes organised their accounts: building trust and managing expectations, navigating the experience, and the need for a more comprehensive treatment. Music served as a guide during sessions. Three of the 11 were responders at week 12, and dose had no obvious relation either to outcome or to whether the experience felt good or bad. Participants wanted more trust-building and preparation, more than one dosing session and ongoing therapy afterwards. The group was heterogeneous and still blinded. Interviews took place by video during Covid-19, 3 to 15 months after dosing, at two sites in one country, and sampling bias was possible, so the trial context limits how far the accounts generalise.

For DBS, how patients' expectations compared with what they experienced was itself a subject of study. Interviewing the same six patients and five caregivers in Australia before surgery and nine months after stimulation began, Thomson et al. (2021) compared what participants had expected with what they experienced; expectations varied, and some had expected no benefit at all. By nine months after stimulation began, two patients reported meaningful improvement, three reported little or no benefit, one reported worse depression, and two had attempted suicide. The participants' outcomes rose and fell rather than holding steady. Participants were more aware of the risks of surgery than of the risks of stimulation, and a media story had raised high hopes that later turned to disappointment. The authors called for informed consent in DBS trials to be revisited as a trial goes on, not settled once. With six patients in one trial, the study shows what can happen, not how often.

What the Label Means

The term itself is part of the experience. An opinion review argues that TRD is not a single entity: the label covers very different illnesses, and inconsistent concepts of resistance get in the way of understanding it (Demyttenaere & Van Duppen, 2019). The same review discusses stigma alongside the burden of the condition, which it describes as immense and under-recognised. Because it is an opinion review, it offers an argued position rather than measured rates, and its point is a warning: the concept is heterogeneous, and confusing concepts can hamper understanding.

Suicide Risk and Suicidal Ideation

Thoughts of suicide are a central clinical concern when depression does not respond to treatment. On this outcome, the strongest evidence comes from studies of ketamine. One individual-participant meta-analysis pooled the raw data of each patient rather than each study's summary. It drew on 167 patients who had suicidal ideation at the start, from 10 of 11 eligible studies, and found that ideation fell rapidly after a single dose of ketamine (Wilkinson et al., 2018). The comparison drugs were saline or midazolam. Against them, ketamine reduced ideation within 1 day, and the benefit lasted for up to 1 week (Wilkinson et al., 2018). The effect sizes ran from Cohen's d = 0.48 to 0.85 (Wilkinson et al., 2018). The drop in ideation stayed significant after the analysis allowed for change in depressive symptoms. The authors therefore judged the effect to be partly independent of mood (Wilkinson et al., 2018). They also gave three cautions: ketamine has not been shown to lower suicide risk itself, one of the 11 studies was missing from the pooled data, and long-term safety still needs research (Wilkinson et al., 2018).

Esketamine, the intranasal form, has not matched this result. An expert narrative review found that esketamine did not differ significantly from placebo on suicidal ideation. The reviewers added that no evidence yet shows it reduces deaths by suicide (McIntyre et al., 2021). A meta-analysis of 24 trials set racemic ketamine against esketamine. Across the trials, the overall suicidality effect was a standardized mean difference (SMD, a group difference expressed in standard-deviation units) of −0.3867, and the two drugs did not differ significantly on this outcome (Bahji et al., 2021). The same authors noted that the benefit for suicidality faded over time. By 4 weeks it was no longer significant (p = 0.2741) (Bahji et al., 2021). Beyond drug treatment, a registry that compared adjunctive vagus nerve stimulation with treatment as usual also took suicidality outcomes within its scope (Aaronson et al., 2017).

None of the studies cited here estimates how much treatment resistance itself raises the risk of suicide. The size of that excess risk is therefore an open question in this article's evidence base. The effects on ideation are rapid but short-lived, which also leaves unsettled how acute treatments fit into longer-term suicide prevention.

Help is available for anyone in crisis or thinking about suicide: in the United States, calling or texting 988 reaches the 988 Suicide & Crisis Lifeline; elsewhere, the International Association for Suicide Prevention lists crisis centres by country.

Worked Example

The question a reader most often brings to a trial is how much difference the treatment makes. Two verified trials answer it in a form that is easy to read. The arithmetic below is computed here from the numbers the trials published; only the inputs, and the rTMS trial's own figure of 12, come from the reports.

The first trial added cognitive behavioural therapy (CBT) to usual care for 469 adults recruited through 73 UK general practices (Wiles et al., 2013). With the added CBT, 46% of patients responded; with usual care alone, 22% did (Wiles et al., 2013). The risk difference is the gap between those two proportions: 46 minus 22, or 24 percentage points. The number needed to treat (NNT) turns that gap into people. It is 1 divided by 0.24, about 4.2, and it is rounded up to 5 by convention. So about five patients would need the added CBT for one more of them to respond than with usual care alone.

The same trial reported an odds ratio of 3.26 (Wiles et al., 2013). That figure looks much larger than a 24-point gain, but it measures something else. Odds set the responders against the non-responders in each group, rather than against the whole group. When response becomes common, odds climb faster than proportions do, so a ratio of odds outruns the plain difference. The risk difference and the NNT are the easier figures to read. The trial's main limitation was that masking was not possible: participants and assessors both knew who had received CBT (Wiles et al., 2013).

The second trial compared left prefrontal repetitive transcranial magnetic stimulation (rTMS) with sham stimulation in 199 drug-free outpatients with major depressive disorder (George et al., 2010). A sham is a dummy procedure that mimics the real one without delivering the active treatment; here it used a sham coil with a metal insert (George et al., 2010). Remission, meaning that symptoms had largely resolved, came to 14.1% with active rTMS and 5.1% with sham (George et al., 2010). The difference is 9.0 percentage points. Dividing 1 by 0.090 gives about 11.1, which rounds up to 12. That matches the NNT of 12 that the trial itself reported (George et al., 2010).

Figure 1

Response or Remission With and Without the Added Treatment in Two Trials

Paired bars. CBT added to usual care: 46 percent responded, against 22 percent with usual care alone. Active rTMS: 14.1 percent remitted, against 5.1 percent with sham stimulation. 0% 25% 50% 46 22 14.1 5.1 CBT trial (response) rTMS trial (remission) Added treatment Comparison
Note. CBT trial: CBT plus usual care against usual care alone, Wiles et al. (2013). rTMS trial: active against sham stimulation, George et al. (2010). Original schematic.

A single difference is an estimate, and a trial also reports how uncertain it is. In an open-label trial of ketamine against electroconvulsive therapy (ECT) in 403 patients without psychosis, response ran 55.4% with ketamine and 41.2% with ECT, a difference of 14.2 percentage points (Anand et al., 2023). Its 95% confidence interval ran from 3.9 to 24.2 points (Anand et al., 2023). In plain terms, the true difference could plausibly lie anywhere in that range, from a small advantage for ketamine to a large one. Because the trial was open-label, patients and clinicians knew which treatment was given.

The calculator demonstration in the Treatment Evidence section runs the same arithmetic on any pair of response rates.

Discussion

Some conclusions hold across the evidence on this page. Treatment-resistant depression still has no agreed definition. Even so, two failed treatments recur as the working threshold in major depressive disorder, and one failure is often used in bipolar depression (Gaynes et al., 2020). Remission grows less likely at each further treatment step (Rush et al., 2006a); the dispute over STAR*D's overall total is set out under Prevalence, Course and Burden. Electroconvulsive therapy beat simulated treatment and probably beats medication over the short term (UK ECT Review Group, 2003). A clinical review called it the most effective acute treatment for severe depression (Kellner et al., 2012). Ketamine and esketamine both act rapidly in resistant depression (McIntyre et al., 2021). The treatment and brain stimulation sections give the trials behind each point.

Three questions remain contested. The first is how large and how lasting the ketamine and esketamine effects are. Their obvious acute effects can reveal to patients which treatment they received, so expectancy may inflate the benefit. No head-to-head trial has compared the two drugs, and long-term safety data are scarce for intravenous ketamine, though they were collected systematically for esketamine (McIntyre et al., 2021). The second is whether deep brain stimulation works beyond open-label follow-up, in which everyone knows the treatment. The only sham-controlled trial found no difference from sham in its blinded phase. The open-label gains therefore remain unconfirmed under blinded conditions (Holtzheimer et al., 2017). The third is why electroconvulsive therapy works. Some findings favour the seizure as the source of benefit and link the electric field mainly to side effects, but the reviewers found support for both sides, and the question stays open until pending trials report (Deng et al., 2024).

Each dispute points to the kind of study that would settle it. Accelerated stimulation has begun to pass sham-controlled replication, and deep brain stimulation has not. Head-to-head trials are still missing for the augmentation agents and for intravenous ketamine against esketamine. Longer follow-up is needed wherever benefit has been measured only over days or weeks. Outcomes also need widening. Trials rarely measure daily functioning or quality of life (Gaynes et al., 2020), although the Living with Treatment-Resistant Depression section shows how much of the burden lies there. Under all of this sits the definitional problem. Systematic reviews have called for a definition shared by researchers and clinicians (Brown et al., 2019). They have also called for a consensus definition paired with a core set of outcomes (Gaynes et al., 2020). Until one exists, response rates from different trials rest on different thresholds for resistance.

Current Directions

Three open questions now drive research on treatment-resistant depression. The first is whether faster, more precisely aimed brain stimulation holds up under blinded testing. The second is how newer drugs compare with treatments clinicians already use. The third is what the term itself should mean.

Accelerated stimulation guided by brain connectivity has moved from an uncontrolled signal to a sham-controlled test. A sham condition mimics the procedure without delivering active stimulation, so it controls for expectation. The first report of the SAINT protocol was an open-label trial, meaning that participants and staff knew the treatment, and it had no sham arm. Its 22 participants received 50 sessions of intermittent theta-burst stimulation, a rapid patterned form of magnetic stimulation, given as 10 daily sessions over 5 days. Each session was aimed by resting-state fcMRI, a scan of how brain regions co-activate at rest, at the left DLPFC-sgACC circuit. Of the 21 completers, 19 (90.5%) reached remission. Remission here meant a Montgomery-Åsberg Depression Rating Scale score below 11. The authors judged the protocol safe and well tolerated. With so small a sample and no control condition, however, they said the result needed sham-controlled confirmation (Cole et al., 2020).

That test has now been run. A double-blind trial randomized 48 adults with treatment-resistant depression to active or sham Stanford neuromodulation therapy, 24 per arm. Half the active group (50.0%) remitted, against 20.8% with sham. Response came to 54.2% against 25.0%. Both gaps were statistically significant (p = 0.035 and p = 0.039). The trial also recorded electroencephalography, which measures electrical activity at the scalp. Frontal beta power, one band of that activity, fell significantly under active stimulation but not under sham. Patients whose beta power fell more in the left anterior cingulate cortex improved more. The authors proposed beta suppression as a possible mechanism, but they treated the marker as a candidate only. They presented the trial as replicating the efficacy that an earlier randomized controlled trial had shown (Kratter et al., 2026). The section on brain stimulation sets these protocols beside ECT, rTMS, VNS and deep brain stimulation.

Drug trials have begun to test new agents against active treatments rather than placebo. In a phase 3b trial, 676 patients were randomized in equal numbers to esketamine (336) or quetiapine (340). Remission meant a Montgomery-Åsberg score of 10 or less. At week 8, 27.1% of the esketamine group had remitted, against 17.6% of the quetiapine group (P = 0.003). By week 32, 21.7% of the esketamine group and 14.1% of the quetiapine group had remitted without relapse. Two design features limit how far this can be read. The trial was open-label, with only the raters blinded, and every dropout was counted as an unfavorable outcome (Reif et al., 2023).

A second trial compared ketamine with ECT itself. It used a noninferiority design, which asks whether a new treatment is not meaningfully worse than an established one, rather than whether it is better. The trial was open-label, ran at five sites and randomized 403 patients without psychosis. Response came to 55.4% with ketamine and 41.2% with ECT. The difference was 14.2 percentage points (95% CI, 3.9 to 24.2). Even the low end of that confidence interval, the range of plausible true differences, sat above the prespecified margin of -10 percentage points. Ketamine was therefore shown to be noninferior. ECT brought a decline in memory recall that recovered gradually. Quality of life improved to a similar degree in both arms (Anand et al., 2023).

Psilocybin remains at the dose-finding stage. A phase 2 double-blind trial in 233 adults compared doses of 25 mg, 10 mg and a 1 mg control, each given with psychological support. At week 3, Montgomery-Åsberg scores fell by 12.0 points with 25 mg and by 5.4 points with 1 mg. The gap of 6.6 points was statistically significant (P < 0.001). The 10 mg dose did not differ significantly from control (P = 0.18). The safety and durability picture was less clear. Adverse events affected 179 of the 233 participants (77%), and suicidal ideation appeared in every dose group. By 12 weeks there was no sustained response, and no existing treatment served as a comparator. The authors called for larger and longer trials (Goodwin et al., 2022). The section on emerging treatments places each of these agents in its wider evidence base.

The least technical question is still open. A systematic review of studies published in 2016 and 2017 found 155 definitions of treatment-resistant depression across 150 studies, and no universal one. Its six key-informant interviews also suggested that clinicians rarely use the term. The review covered only English and French publications from those two years, so its count describes the field at one moment (Brown et al., 2019). A second review of definitions and trials reached the same verdict. Investigators had not agreed on what counts as an adequate dose or an adequate length of treatment. They also seldom measured daily functioning or quality of life. The authors asked for a consensus definition together with a core outcome package, a shared set of measures every trial would report (Gaynes et al., 2020). Until such a definition exists, response rates from different trials rest on different thresholds for resistance. That is why the staging and measurement section matters for every comparison made here.

Common Misconceptions

Treatment-resistant depression means that no treatment can help.
The label records that earlier treatments failed, not that every treatment will. Electroconvulsive therapy beat simulated treatment and probably beats drug therapy in the short term (UK ECT Review Group, 2003). In primary care patients still depressed despite antidepressants, adding cognitive behavioural therapy to usual care led to response in 46% against 22% at six months. That trial could not hide from participants or assessors who received therapy (Wiles et al., 2013). The belief arises because resistance sounds like a fixed property of the illness. The definition describes a treatment history.
Anyone who has not improved on antidepressants has true treatment resistance.
Some apparent failures are pseudo-resistance, where treatment was too short, too weak or not taken as prescribed (McIntyre et al., 2023). Others may be an unrecognised bipolar illness. In a non-blind follow-up of 61 patients treated for resistant unipolar depression, bipolar diagnoses rose from 35% at intake to 59% (Sharma et al., 2005). Because the diagnosis rests entirely on what has been tried, an inadequate or untaken treatment can produce it as readily as a truly refractory illness.
Treatment resistance affects only mood.
In a survey across five European countries, people with treatment-resistant depression reported lower health-related quality of life than people whose depression was not resistant. They also reported more missed work, more hospital stays and more emergency visits. The data were self-reported and cross-sectional, so they cannot show that resistance caused the burden (Jaffe et al., 2019). A systematic review found that costs rose and quality of life fell at each higher level of resistance, though it could not say what caused the burden (Johnston et al., 2019). An opinion review counts caregiver burden as part of the condition's weight (Demyttenaere & Van Duppen, 2019). The belief persists because trials score symptoms and rarely measure daily life.
Electroconvulsive therapy permanently erases memory.
A meta-analysis of 84 studies found that 72% of cognitive measures were worse in the first three days after a course. Performance was back near pretreatment levels between 4 and 15 days. After 15 days, 57% of the measures were better than at baseline (Semkovska & McLoughlin, 2010). The analysis could not test retrograde amnesia, the loss of memory for events before treatment, because no standardised tests of it were available. The correction therefore applies to the functions it measured.

Glossary

Augmentation.
Adding a second agent, usually one that is not itself a standard antidepressant, to an antidepressant the patient keeps taking. Lithium, thyroid hormones and atypical antipsychotics are among the agents tested in resistant depression.
Confidence interval.
The range of values for a true effect that are compatible with a study's data. A wide interval means the estimate is imprecise.
Deep brain stimulation.
A surgical treatment that stimulates a deep brain target through implanted electrodes, in most of the depression studies cited here the subcallosal cingulate region. Benefit has been reported in open-label series but not confirmed against sham in the only blinded trial.
Dissociation.
A sense of detachment from oneself or one's surroundings, a common acute effect of ketamine that was not necessary for its antidepressant response.
Electroconvulsive therapy.
A treatment that uses electric current to induce a brief, controlled seizure. It is given as a course and is sometimes continued afterwards to prevent relapse.
Esketamine.
One of the two mirror-image forms of ketamine, given as a nasal spray alongside an oral antidepressant and approved by the FDA in March 2019. One meta-analysis found smaller response and remission effects for it than for intravenous ketamine.
Ketamine.
A drug that blocks the NMDA receptor, a receptor for the signalling chemical glutamate. A single intravenous dose of 0.5 mg/kg reduced depressive symptoms within hours in the early trials. How it does so is still debated.
Maudsley Staging Method.
A points-based scale that grades treatment resistance from the treatments received, the severity of the episode and its duration. Totals run from 3 to 15.
Meta-analysis.
A study that pools the results of earlier studies into a single estimate. A network meta-analysis also compares treatments that were never tested against each other directly.
Noninferiority trial.
A randomised comparison designed to show that a treatment is not worse than an established one by more than a margin fixed in advance, as in the trial of ketamine against electroconvulsive therapy.
Number needed to treat.
How many patients must receive a treatment instead of its comparison for one more of them to benefit. It is 1 divided by the risk difference, rounded up.
Odds ratio.
The odds of an outcome in one group divided by the odds in another, so a value of 2 means the odds are twice as high. When an outcome is common, it looks larger than the plain difference in rates.
Open-label.
A study design in which patients and clinicians know which treatment is being given, so expectation can shape the results.
Pseudo-resistance.
An apparent lack of response that does not reflect true treatment resistance, for instance because a treatment trial was too short or too weak, or the medication was not taken.
Remission.
A fall in symptoms to a minimal level, judged by a cut-off on a rating scale, such as a MADRS score of 10 or less in one esketamine trial. The cut-off differs from trial to trial.
Repetitive transcranial magnetic stimulation.
A noninvasive treatment in which repeated magnetic pulses from a coil held against the scalp induce current in the cortex beneath it. In depression it usually targets the left prefrontal cortex.
Response.
A clear fall in symptom scores that may stop short of remission. Each trial sets its own threshold on its rating scale.
Risk difference.
The gap between the proportion of patients with an outcome in one group and in another, given in percentage points.
Sham.
A dummy procedure that imitates a treatment without delivering it, so that the effect of expectation can be measured.
Staging model.
A method that places patients on a graded scale of treatment resistance instead of sorting them into resistant and non-resistant groups. The Thase and Rush model and the Maudsley method are examples.
Theta-burst stimulation.
A patterned form of rTMS given in short, rapid bursts. In one trial, a 3-minute session of intermittent theta-burst stimulation was noninferior to 37.5 minutes of 10 Hz rTMS.
Vagus nerve stimulation.
Stimulation of the vagus nerve added to usual treatment. Its evidence in depression comes mainly from a nonrandomised, open-label registry.

Key Researchers

Andre R. Brunoni

Psychiatrist at the University of Texas Southwestern Medical Center, previously at the University of São Paulo, who led a network meta-analysis of transcranial magnetic stimulation treatments for depression. ORCID - Google Scholar - Faculty Page

Maurizio Fava

Psychiatrist at Massachusetts General Hospital and Harvard Medical School who reviewed how treatment-resistant depression is diagnosed, defined and staged. Faculty Page

Charles H. Kellner

Psychiatrist at the Icahn School of Medicine at Mount Sinai who led a randomised trial of continuation electroconvulsive therapy for relapse prevention. ORCID - Faculty Page

Helen S. Mayberg

Neurologist at the Icahn School of Medicine at Mount Sinai whose 2005 case series was the first to apply deep brain stimulation to the subcallosal cingulate in depression that had not responded to other treatments. ORCID - Wikipedia - Faculty Page

A. John Rush

Psychiatrist at Duke-NUS Medical School who led the STAR*D reports of remission across successive treatment steps. ORCID - Wikipedia - Google Scholar - Faculty Page

Carlos A. Zarate

Chief of the Experimental Therapeutics and Pathophysiology Branch at the National Institute of Mental Health who led the 2006 randomised trial of ketamine in treatment-resistant major depression. Faculty Page

Frequently Asked Questions

What is treatment-resistant depression?

Treatment-resistant depression is depression that has not responded adequately to antidepressant treatment, so it is identified by what has been tried rather than by distinct symptoms. In research on major depression it usually means at least two failed antidepressants, although an expert review found that no definition has won consensus (McIntyre et al., 2023).

How common is treatment-resistant depression?

A US cost model estimated that about 2.8 million of 8.9 million adults taking medication for major depressive disorder met its definition, roughly one in three. The definition was a proxy built from insurance claims with no symptom data, and the study was industry-sponsored (Zhdanava et al., 2021).

Why might depression only appear to be treatment-resistant?

Because the label rests on treatment history, an inadequate trial or an untaken medicine can produce it. Fava's review therefore held that re-evaluating the diagnosis is essential, and it linked other medical and psychiatric illnesses to resistance (Fava, 2003). Unrecognised bipolar illness, a further possibility, is examined in the Types section.

How does treatment-resistant depression affect daily life?

In a survey across five European countries, people with treatment-resistant depression reported lower quality of life, more missed work and more hospital stays than people whose depression was not resistant. The data were self-reported and cross-sectional, so they cannot show that resistance caused the burden (Jaffe et al., 2019). Interview studies with patients and caregivers are described in the Living with Treatment-Resistant Depression section.

Does treatment-resistant depression affect thinking and memory?

In a study of 369 adults aged 60 and older, scores fell below norms for inhibitory control, processing speed and memory, but not for working memory or set shifting. Impaired set shifting went with less improvement, while impaired inhibitory control did not predict response (Ainsworth et al., 2025). The short-lived cognitive effects of electroconvulsive therapy are covered in the Brain Stimulation section.

Which medication strategies have evidence after an antidepressant fails?

A network meta-analysis of 65 randomised trials found that approved atypical antipsychotics, thyroid hormones and lithium appeared better than placebo as add-on agents. Its authors noted that trials defined resistance in different ways and that head-to-head studies are still needed (Nuñez et al., 2022). Switching to a different antidepressant is covered in the Treatment Evidence section.

Does psychotherapy help in treatment-resistant depression?

In a UK primary care trial of 469 adults still depressed despite antidepressants, 46% of those given cognitive behavioural therapy on top of usual care responded at six months, against 22% with usual care alone. Neither participants nor assessors could be kept unaware of who received therapy (Wiles et al., 2013).

What do ketamine, esketamine and the stimulation treatments offer?

An expert review found rapid efficacy for ketamine and esketamine in treatment-resistant depression but judged their safety questions unresolved. It noted that patients may guess their treatment from its obvious effects, which can inflate the measured benefit of ketamine, and that long-term safety data are scarce for intravenous ketamine (McIntyre et al., 2021). Electroconvulsive therapy, magnetic stimulation, vagus nerve stimulation and deep brain stimulation are compared in the Brain Stimulation section.

How This Article Was Prepared

This is a narrative reference article, not a systematic review. Its sources were identified through a survey of the 15 most-cited works on the subject, the most-cited research of four of its key researchers and six logged searches of OpenAlex, which matched 810 records, of which the 100 most cited were screened by title, from 20 November 2026 to 21 November 2026.

- Considered and excluded: 18 candidate works, each with a recorded reason (for example, superseded by a later review, a single case report, or outside the article's scope). - Included: 66 sources, each confirmed against the publisher's or indexer's own record: 65 by DOI and one by ISBN. 21 read in full text, 43 from their abstracts and two cited without a retrievable abstract. Abstracts came from PubMed, Crossref and OpenAlex. - Identifiers: 65 references carry a PubMed ID and 23 a PubMed Central ID, each confirmed by the National Library of Medicine against the cited DOI. - Retractions: 65 sources were checked for retractions and editorial notices; none was retracted. - Publication dates: two from the 1990s, 12 from the 2000s, 29 from the 2010s and 23 from 2020 or later. - Checked against sources: each of the 296 cited sentences was matched to a passage in its source (295 supported by that passage, one resting on a source without a retrievable abstract), and an independent reviewer checked the article against those sources over three passes before publication. - Limits: a narrative review reflects choices about which studies to include, so it can over-represent some lines of work more than a systematic review would. To limit this, each of the field's 15 most-cited works was considered; every logged search and screening decision is recorded; and the reviewer searched each cited source for contrary evidence before accepting a sentence.

See our Editorial Policy for the full method.

References

Aaronson, S. T., Sears, P., Ruvuna, F., Bunker, M., Conway, C. R., Dougherty, D. D., Reimherr, F. W., Schwartz, T. L., & Zajecka, J. M. (2017). A 5-year observational study of patients with treatment-resistant depression treated with vagus nerve stimulation or treatment as usual: Comparison of response, remission, and suicidality. American Journal of Psychiatry, 174(7), 640-648. https://doi.org/10.1176/appi.ajp.2017.16010034 PMID: 28359201

Ainsworth, N. J., Oughli, H., Lavretsky, H., Blumberger, D. M., Brown, P. J., Butters, M. A., Karp, J. F., Lenard, E., Lenze, E. J., McAndrews, M. P., Miller, J. P., Pollock, B. G., Reynolds, C. F., & Mulsant, B. H. (2025). The cognitive profile of older adults with treatment-resistant depression: An analysis of the OPTIMUM randomized controlled trial. The American Journal of Geriatric Psychiatry, 33(4), 361-371. https://doi.org/10.1016/j.jagp.2024.09.018 PMID: 39438235

Alexopoulos, G. S., Meyers, B. S., Young, R. C., Kalayam, B., Kakuma, T., Gabrielle, M., Sirey, J. A., & Hull, J. (2000). Executive dysfunction and long-term outcomes of geriatric depression. Archives of General Psychiatry, 57(3), 285-290. https://doi.org/10.1001/archpsyc.57.3.285 PMID: 10711915

Anand, A., Mathew, S. J., Sanacora, G., Murrough, J. W., Goes, F. S., Altinay, M., Aloysi, A. S., Asghar-Ali, A. A., Barnett, B. S., Chang, L. C., Collins, K. A., Costi, S., Iqbal, S., Jha, M. K., Krishnan, K., Malone, D. A., Nikayin, S., Nissen, S. E., Ostroff, R. B., Reti, I. M., Wilkinson, S. T., Wolski, K., & Hu, B. (2023). Ketamine versus ECT for nonpsychotic treatment-resistant major depression. New England Journal of Medicine, 388(25), 2315-2325. https://doi.org/10.1056/NEJMoa2302399 PMID: 37224232

Bahji, A., Vazquez, G. H., & Zarate, C. A., Jr. (2021). Comparative efficacy of racemic ketamine and esketamine for depression: A systematic review and meta-analysis. Journal of Affective Disorders, 278, 542-555. https://doi.org/10.1016/j.jad.2020.09.071 PMID: 33022440 · PMCID: PMC7704936

Bauer, M., & Döpfmer, S. (1999). Lithium augmentation in treatment-resistant depression: Meta-analysis of placebo-controlled studies. Journal of Clinical Psychopharmacology, 19(5), 427-434. https://doi.org/10.1097/00004714-199910000-00006 PMID: 10505584

Berman, R. M., Cappiello, A., Anand, A., Oren, D. A., Heninger, G. R., Charney, D. S., & Krystal, J. H. (2000). Antidepressant effects of ketamine in depressed patients. Biological Psychiatry, 47(4), 351-354. https://doi.org/10.1016/S0006-3223(99)00230-9 PMID: 10686270

Blumberger, D. M., Vila-Rodriguez, F., Thorpe, K. E., Feffer, K., Noda, Y., Giacobbe, P., Knyahnytska, Y., Kennedy, S. H., Lam, R. W., Daskalakis, Z. J., & Downar, J. (2018). Effectiveness of theta burst versus high-frequency repetitive transcranial magnetic stimulation in patients with depression (THREE-D): A randomised non-inferiority trial. The Lancet, 391(10131), 1683-1692. https://doi.org/10.1016/S0140-6736(18)30295-2 PMID: 29726344

Breeksema, J. J., Niemeijer, A., Krediet, E., Karsten, T., Kamphuis, J., Vermetten, E., van den Brink, W., & Schoevers, R. (2024). Patient perspectives and experiences with psilocybin treatment for treatment-resistant depression: A qualitative study. Scientific Reports, 14(1), 2929. https://doi.org/10.1038/s41598-024-53188-9 PMID: 38316896 · PMCID: PMC10844281

Breeksema, J. J., Niemeijer, A., Kuin, B., Veraart, J., Kamphuis, J., Schimmel, N., van den Brink, W., Vermetten, E., & Schoevers, R. (2022). Holding on or letting go? Patient experiences of control, context, and care in oral esketamine treatment for treatment-resistant depression: A qualitative study. Frontiers in Psychiatry, 13, 948115. https://doi.org/10.3389/fpsyt.2022.948115 PMID: 36506427 · PMCID: PMC9732097

Brown, S., Rittenbach, K., Cheung, S., McKean, G., MacMaster, F. P., & Clement, F. (2019). Current and common definitions of treatment-resistant depression: Findings from a systematic review and qualitative interviews. Canadian Journal of Psychiatry, 64(6), 380-387. https://doi.org/10.1177/0706743719828965 PMID: 30763119 · PMCID: PMC6591751

Brunoni, A. R., Chaimani, A., Moffa, A. H., Razza, L. B., Gattaz, W. F., Daskalakis, Z. J., & Carvalho, A. F. (2017). Repetitive transcranial magnetic stimulation for the acute treatment of major depressive episodes: A systematic review with network meta-analysis. JAMA Psychiatry, 74(2), 143-152. https://doi.org/10.1001/jamapsychiatry.2016.3644 PMID: 28030740

Carhart-Harris, R. L., Bolstridge, M., Rucker, J., Day, C. M. J., Erritzoe, D., Kaelen, M., Bloomfield, M., Rickard, J. A., Forbes, B., Feilding, A., Taylor, D., Pilling, S., Curran, H. V., & Nutt, D. J. (2016). Psilocybin with psychological support for treatment-resistant depression: An open-label feasibility study. The Lancet Psychiatry, 3(7), 619-627. https://doi.org/10.1016/S2215-0366(16)30065-7 PMID: 27210031

Cole, E. J., Phillips, A. L., Bentzley, B. S., Stimpson, K. H., Nejad, R., Barmak, F., Veerapal, C., Khan, N., Cherian, K., Felber, E., Brown, R., Choi, E., King, S., Pankow, H., Bishop, J. H., Azeez, A., Coetzee, J., Rapier, R., Odenwald, N., Carreon, D., Hawkins, J., Chang, M., Keller, J., Raj, K., DeBattista, C., Jo, B., Espil, F. M., Schatzberg, A. F., Sudheimer, K. D., & Williams, N. R. (2022). Stanford Neuromodulation Therapy (SNT): A double-blind randomized controlled trial. American Journal of Psychiatry, 179(2), 132-141. https://doi.org/10.1176/appi.ajp.2021.20101429 PMID: 34711062

Cole, E. J., Stimpson, K. H., Bentzley, B. S., Gulser, M., Cherian, K., Tischler, C., Nejad, R., Pankow, H., Choi, E., Aaron, H., Espil, F. M., Pannu, J., Xiao, X., Duvio, D., Solvason, H. B., Hawkins, J., Guerra, A., Jo, B., Raj, K. S., … Williams, N. R. (2020). Stanford Accelerated Intelligent Neuromodulation Therapy for treatment-resistant depression. The American Journal of Psychiatry, 177(8), 716–726. https://doi.org/10.1176/appi.ajp.2019.19070720 PMID: 32252538

Correa, R., Akiskal, H., Gilmer, W., Nierenberg, A. A., Trivedi, M., & Zisook, S. (2010). Is unrecognized bipolar disorder a frequent contributor to apparent treatment resistant depression? Journal of Affective Disorders, 127(1-3), 10-18. https://doi.org/10.1016/j.jad.2010.06.036 PMID: 20655113

Crowell, A. L., Riva-Posse, P., Holtzheimer, P. E., Garlow, S. J., Kelley, M. E., Gross, R. E., Denison, L., Quinn, S., & Mayberg, H. S. (2019). Long-term outcomes of subcallosal cingulate deep brain stimulation for treatment-resistant depression. American Journal of Psychiatry, 176(11), 949–956. https://doi.org/10.1176/appi.ajp.2019.18121427 PMID: 31581800

Demyttenaere, K., & Van Duppen, Z. (2019). The impact of (the concept of) treatment-resistant depression: An opinion review. International Journal of Neuropsychopharmacology, 22(2), 85-92. https://doi.org/10.1093/ijnp/pyy052 PMID: 29961822 · PMCID: PMC6368367

Deng, Z.-D., Robins, P. L., Regenold, W., Rohde, P., Dannhauer, M., & Lisanby, S. H. (2024). How electroconvulsive therapy works in the treatment of depression: Is it the seizure, the electricity, or both? Neuropsychopharmacology, 49(1), 150-162. https://doi.org/10.1038/s41386-023-01677-2 PMID: 37488281 · PMCID: PMC10700353

Espinoza, R. T., & Kellner, C. H. (2022). Electroconvulsive therapy. New England Journal of Medicine, 386(7), 667-672. https://doi.org/10.1056/NEJMra2034954 PMID: 35172057

Fava, M. (2003). Diagnosis and definition of treatment-resistant depression. Biological Psychiatry, 53(8), 649-659. https://doi.org/10.1016/s0006-3223(03)00231-2 PMID: 12706951

Fava, M., Rush, A. J., Alpert, J. E., Balasubramani, G. K., Wisniewski, S. R., Carmin, C. N., Biggs, M. M., Zisook, S., Leuchter, A., Howland, R., Warden, D., & Trivedi, M. H. (2008). Difference in treatment outcome in outpatients with anxious versus nonanxious depression: A STAR*D report. American Journal of Psychiatry, 165(3), 342-351. https://doi.org/10.1176/appi.ajp.2007.06111868 PMID: 18172020

Fekadu, A., Wooderson, S., Donaldson, C., Markopoulou, K., Masterson, B., Poon, L., & Cleare, A. J. (2009). A multidimensional tool to quantify treatment resistance in depression: The Maudsley staging method. Journal of Clinical Psychiatry, 70(2), 177-184. https://doi.org/10.4088/jcp.08m04309 PMID: 19192471

Gaynes, B. N., Lloyd, S. W., Lux, L., Gartlehner, G., Hansen, R. A., Brode, S., Jonas, D. E., Swinson Evans, T., Viswanathan, M., & Lohr, K. N. (2014). Repetitive transcranial magnetic stimulation for treatment-resistant depression: A systematic review and meta-analysis. The Journal of Clinical Psychiatry, 75(5), 477-489. https://doi.org/10.4088/jcp.13r08815 PMID: 24922485

Gaynes, B. N., Lux, L., Gartlehner, G., Asher, G., Forman-Hoffman, V., Green, J., Boland, E., Weber, R. P., Randolph, C., Bann, C., Coker-Schwimmer, E., Viswanathan, M., & Lohr, K. N. (2020). Defining treatment-resistant depression. Depression and Anxiety, 37(2), 134-145. https://doi.org/10.1002/da.22968 PMID: 31638723

George, M. S., Lisanby, S. H., Avery, D., McDonald, W. M., Durkalski, V., Pavlicova, M., Anderson, B., Nahas, Z., Bulow, P., Zarkowski, P., Holtzheimer, P. E., Schwartz, T., & Sackeim, H. A. (2010). Daily left prefrontal transcranial magnetic stimulation therapy for major depressive disorder: A sham-controlled randomized trial. Archives of General Psychiatry, 67(5), 507–516. https://doi.org/10.1001/archgenpsychiatry.2010.46 PMID: 20439832

Goodwin, G. M., Aaronson, S. T., Alvarez, O., Arden, P. C., Baker, A., Bennett, J. C., Bird, C., Blom, R. E., Brennan, C., Brusch, D., et al. (2022). Single-dose psilocybin for a treatment-resistant episode of major depression. New England Journal of Medicine, 387(18), 1637-1648. https://doi.org/10.1056/NEJMoa2206443 PMID: 36322843

Holtzheimer, P. E., Husain, M. M., Lisanby, S. H., Taylor, S. F., Whitworth, L. A., McClintock, S., Slavin, K. V., Berman, J., McKhann, G. M., Patil, P. G., Rittberg, B. R., Abosch, A., Pandurangi, A. K., Holloway, K. L., Lam, R. W., Honey, C. R., Neimat, J. S., Henderson, J. M., DeBattista, C., … Mayberg, H. S. (2017). Subcallosal cingulate deep brain stimulation for treatment-resistant depression: A multisite, randomised, sham-controlled trial. The Lancet Psychiatry, 4(11), 839–849. https://doi.org/10.1016/S2215-0366(17)30371-1 PMID: 28988904

Jaffe, D. H., Rive, B., & Denee, T. R. (2019). The humanistic and economic burden of treatment-resistant depression in Europe: A cross-sectional study. BMC Psychiatry, 19(1), 247. https://doi.org/10.1186/s12888-019-2222-4 PMID: 31391065 · PMCID: PMC6686569

Johnston, K. M., Powell, L. C., Anderson, I. M., Szabo, S., & Cline, S. (2019). The burden of treatment-resistant depression: A systematic review of the economic and quality of life literature. Journal of Affective Disorders, 242, 195-210. https://doi.org/10.1016/j.jad.2018.06.045 PMID: 30195173

Kellner, C. H., Greenberg, R. M., Murrough, J. W., Bryson, E. O., Briggs, M. C., & Pasculli, R. M. (2012). ECT in treatment-resistant depression. American Journal of Psychiatry, 169(12), 1238-1244. https://doi.org/10.1176/appi.ajp.2012.12050648 PMID: 23212054

Kellner, C. H., Knapp, R. G., Petrides, G., Rummans, T. A., Husain, M. M., Rasmussen, K., Mueller, M., Bernstein, H. J., O'Connor, K., Smith, G., Biggs, M., Bailine, S. H., Malur, C., Yim, E., McClintock, S., Sampson, S., & Fink, M. (2006). Continuation electroconvulsive therapy vs pharmacotherapy for relapse prevention in major depression: A multisite study from the Consortium for Research in Electroconvulsive Therapy (CORE). Archives of General Psychiatry, 63(12), 1337-1344. https://doi.org/10.1001/archpsyc.63.12.1337 PMID: 17146008 · PMCID: PMC3708140

Komossa, K., Depping, A. M., Gaudchau, A., Kissling, W., & Leucht, S. (2010). Second-generation antipsychotics for major depressive disorder and dysthymia. Cochrane Database of Systematic Reviews, (12), CD008121. https://doi.org/10.1002/14651858.cd008121.pub2 PMID: 21154393 · PMCID: PMC11994262

Kratter, I. H., Austelle, C. W., Lissemore, J. I., Wada, M., Geoly, A., Chaiken, A., Kaloiani, I., Johnson, N., Wan, S., Kozyr, L., Makarewycz, E., Wong, B., Sridhar, M., Espil, F. M., Bassano, N., Kim, B., Ehrie, J., Maron-Katz, A., Tischler, C., ... Williams, N. R. (2026). Stanford neuromodulation therapy for treatment-resistant depression: A randomized controlled trial confirming efficacy, and an EEG study providing insight into mechanism of action and a potentially predictive biomarker of efficacy. World Psychiatry, 25(1), 105-116. https://doi.org/10.1002/wps.70032 PMID: 41536095 · PMCID: PMC12805067

Krystal, J. H., Abdallah, C. G., Sanacora, G., Charney, D. S., & Duman, R. S. (2019). Ketamine: A paradigm shift for depression research and treatment. Neuron, 101(5), 774-778. https://doi.org/10.1016/j.neuron.2019.02.005 PMID: 30844397 · PMCID: PMC6560624

Mayberg, H. S., Lozano, A. M., Voon, V., McNeely, H. E., Seminowicz, D., Hamani, C., Schwalb, J. M., & Kennedy, S. H. (2005). Deep brain stimulation for treatment-resistant depression. Neuron, 45(5), 651–660. https://doi.org/10.1016/j.neuron.2005.02.014 PMID: 15748841

McAllister-Williams, R. H., Arango, C., Blier, P., Demyttenaere, K., Falkai, P., Gorwood, P., Hopwood, M., Javed, A., Kasper, S., Malhi, G. S., Soares, J. C., Vieta, E., Young, A. H., Papadopoulos, A., & Rush, A. J. (2020). The identification, assessment and management of difficult-to-treat depression: An international consensus statement. Journal of Affective Disorders, 267, 264-282. https://doi.org/10.1016/j.jad.2020.02.023 PMID: 32217227

McIntyre, R. S., Alsuwaidan, M., Baune, B. T., Berk, M., Demyttenaere, K., Goldberg, J. F., Gorwood, P., Ho, R., Kasper, S., Kennedy, S. H., Ly-Uson, J., Mansur, R. B., McAllister-Williams, R. H., Murrough, J. W., Nemeroff, C. B., Nierenberg, A. A., Rosenblat, J. D., Sanacora, G., Schatzberg, A. F., Shelton, R., Stahl, S. M., Trivedi, M. H., Vieta, E., Vinberg, M., Williams, N., Young, A. H., & Maj, M. (2023). Treatment-resistant depression: Definition, prevalence, detection, management, and investigational interventions. World Psychiatry, 22(3), 394-412. https://doi.org/10.1002/wps.21120 PMID: 37713549 · PMCID: PMC10503923

McIntyre, R. S., Rosenblat, J. D., Nemeroff, C. B., Sanacora, G., Murrough, J. W., Berk, M., Brietzke, E., Dodd, S., Gorwood, P., Ho, R., Iosifescu, D. V., Lopez Jaramillo, C., Kasper, S., Kratiuk, K., Lee, J. G., Lee, Y., Lui, L. M. W., Mansur, R. B., Papakostas, G. I., Subramaniapillai, M., Thase, M., Vieta, E., Young, A. H., Zarate, C. A., Jr., & Stahl, S. (2021). Synthesizing the evidence for ketamine and esketamine in treatment-resistant depression: An international expert opinion on the available evidence and implementation. American Journal of Psychiatry, 178(5), 383-399. https://doi.org/10.1176/appi.ajp.2020.20081251 PMID: 33726522 · PMCID: PMC9635017

Mrazek, D. A., Hornberger, J. C., Altar, C. A., & Degtiar, I. (2014). A review of the clinical, economic, and societal burden of treatment-resistant depression: 1996–2013. Psychiatric Services, 65(8), 977-987. https://doi.org/10.1176/appi.ps.201300059 PMID: 24789696

Nanni, V., Uher, R., & Danese, A. (2012). Childhood maltreatment predicts unfavorable course of illness and treatment outcome in depression: A meta-analysis. American Journal of Psychiatry, 169(2), 141-151. https://doi.org/10.1176/appi.ajp.2011.11020335 PMID: 22420036

Nuñez, N. A., Joseph, B., Pahwa, M., Kumar, R., Gardea Resendez, M., Prokop, L. J., Veldic, M., Seshadri, A., Biernacka, J. M., Frye, M. A., Wang, Z., & Singh, B. (2022). Augmentation strategies for treatment resistant major depression: A systematic review and network meta-analysis. Journal of Affective Disorders, 302, 385-400. https://doi.org/10.1016/j.jad.2021.12.134 PMID: 34986373 · PMCID: PMC9328668

Pigott, H. E., Kim, T., Xu, C., Kirsch, I., & Amsterdam, J. (2023). What are the treatment remission, response and extent of improvement rates after up to four trials of antidepressant therapies in real-world depressed patients? A reanalysis of the STAR*D study's patient-level data with fidelity to the original research protocol. BMJ Open, 13(7), e063095. https://doi.org/10.1136/bmjopen-2022-063095 PMID: 37491091 · PMCID: PMC10373710

Pimontel, M. A., Rindskopf, D., Rutherford, B. R., Brown, P. J., Roose, S. P., & Sneed, J. R. (2016). A meta-analysis of executive dysfunction and antidepressant treatment response in late-life depression. American Journal of Geriatric Psychiatry, 24(1), 31-41. https://doi.org/10.1016/j.jagp.2015.05.010 PMID: 26282222 · PMCID: PMC4928373

Popova, V., Daly, E. J., Trivedi, M., Cooper, K., Lane, R., Lim, P., Mazzucco, C., Hough, D., Thase, M. E., Shelton, R. C., Molero, P., Vieta, E., Bajbouj, M., Manji, H., Drevets, W. C., & Singh, J. B. (2019). Efficacy and safety of flexibly dosed esketamine nasal spray combined with a newly initiated oral antidepressant in treatment-resistant depression: A randomized double-blind active-controlled study. American Journal of Psychiatry, 176(6), 428-438. https://doi.org/10.1176/appi.ajp.2019.19020172 PMID: 31109201

Rane, L. J., Fekadu, A., Papadopoulos, A. S., Wooderson, S. C., Poon, L., Markopoulou, K., & Cleare, A. J. (2012). Psychological and physiological effects of caring for patients with treatment-resistant depression. Psychological Medicine, 42(9), 1825-1833. https://doi.org/10.1017/S0033291711003035 PMID: 22251699

Reif, A., Bitter, I., Buyze, J., Cebulla, K., Frey, R., Fu, D.-J., Ito, T., Kambarov, Y., Llorca, P.-M., Oliveira-Maia, A. J., Messer, T., Mulhern-Haughey, S., Rive, B., von Holt, C., Young, A. H., Godinov, Y., & ESCAPE-TRD Investigators. (2023). Esketamine nasal spray versus quetiapine for treatment-resistant depression. New England Journal of Medicine, 389(14), 1298-1309. https://doi.org/10.1056/NEJMoa2304145 PMID: 37792613

Rush, A. J., Trivedi, M. H., Wisniewski, S. R., Nierenberg, A. A., Stewart, J. W., Warden, D., Niederehe, G., Thase, M. E., Lavori, P. W., Lebowitz, B. D., McGrath, P. J., Rosenbaum, J. F., Sackeim, H. A., Kupfer, D. J., Luther, J., & Fava, M. (2006). Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: A STAR*D report. American Journal of Psychiatry, 163(11), 1905-1917. https://doi.org/10.1176/ajp.2006.163.11.1905 PMID: 17074942

Rush, A. J., Trivedi, M. H., Wisniewski, S. R., Stewart, J. W., Nierenberg, A. A., Thase, M. E., Ritz, L., Biggs, M. M., Warden, D., Luther, J. F., Shores-Wilson, K., Niederehe, G., Fava, M., & STAR*D Study Team. (2006). Bupropion-SR, sertraline, or venlafaxine-XR after failure of SSRIs for depression. New England Journal of Medicine, 354(12), 1231-1242. https://doi.org/10.1056/NEJMoa052963 PMID: 16554525

Sanacora, G., Frye, M. A., McDonald, W., Mathew, S. J., Turner, M. S., Schatzberg, A. F., Summergrad, P., & Nemeroff, C. B. (2017). A consensus statement on the use of ketamine in the treatment of mood disorders. JAMA Psychiatry, 74(4), 399-405. https://doi.org/10.1001/jamapsychiatry.2017.0080 PMID: 28249076

Semkovska, M., & McLoughlin, D. M. (2010). Objective cognitive performance associated with electroconvulsive therapy for depression: A systematic review and meta-analysis. Biological Psychiatry, 68(6), 568-577. https://doi.org/10.1016/j.biopsych.2010.06.009 PMID: 20673880

Sforzini, L., Worrell, C., Kose, M., Anderson, I. M., Aouizerate, B., Arolt, V., et al. (2022). A Delphi-method-based consensus guideline for definition of treatment-resistant depression for clinical trials. Molecular Psychiatry, 27(3), 1286-1299. https://doi.org/10.1038/s41380-021-01381-x PMID: 34907394 · PMCID: PMC9095475

Sharma, V., Khan, M., & Smith, A. (2005). A closer look at treatment resistant depression: Is it due to a bipolar diathesis? Journal of Affective Disorders, 84(2-3), 251-257. https://doi.org/10.1016/j.jad.2004.01.015 PMID: 15708423

Spielmans, G. I., Berman, M. I., Linardatos, E., Rosenlicht, N. Z., Perry, A., & Tsai, A. C. (2013). Adjunctive atypical antipsychotic treatment for major depressive disorder: A meta-analysis of depression, quality of life, and safety outcomes. PLoS Medicine, 10(3), e1001403. https://doi.org/10.1371/journal.pmed.1001403 PMID: 23554581 · PMCID: PMC3595214

Swainson, J., McGirr, A., Blier, P., Brietzke, E., Richard-Devantoy, S., Ravindran, N., Blier, J., Beaulieu, S., Frey, B. N., Kennedy, S. H., McIntyre, R. S., Milev, R. V., Parikh, S. V., Schaffer, A., Taylor, V. H., Tourjman, V., van Ameringen, M., Yatham, L. N., Ravindran, A. V., & Lam, R. W. (2021). The Canadian Network for Mood and Anxiety Treatments (CANMAT) Task Force recommendations for the use of racemic ketamine in adults with major depressive disorder. The Canadian Journal of Psychiatry, 66(2), 113-125. https://doi.org/10.1177/0706743720970860 PMID: 33174760 · PMCID: PMC7918868

Thase, M. E., & Rush, A. J. (1995). Treatment-resistant depression. In F. E. Bloom & D. J. Kupfer (Eds.), Psychopharmacology: The fourth generation of progress. Raven Press.

Thomson, C. J., Segrave, R. A., Fitzgerald, P. B., Richardson, K. E., Racine, E., & Carter, A. (2021). "Nothing to lose, absolutely everything to gain": Patient and caregiver expectations and subjective outcomes of deep brain stimulation for treatment-resistant depression. Frontiers in Human Neuroscience, 15, 755276. https://doi.org/10.3389/fnhum.2021.755276 PMID: 34658822 · PMCID: PMC8511461

Thomson, C. J., Segrave, R. A., Fitzgerald, P. B., Richardson, K. E., Racine, E., & Carter, A. (2023). Personal and relational changes following deep brain stimulation for treatment-resistant depression: A prospective qualitative study with patients and caregivers. PLOS ONE, 18(4), e0284160. https://doi.org/10.1371/journal.pone.0284160 PMID: 37023074 · PMCID: PMC10079124

Turner, E. H. (2019). Esketamine for treatment-resistant depression: Seven concerns about efficacy and FDA approval. The Lancet Psychiatry, 6(12), 977-979. https://doi.org/10.1016/S2215-0366(19)30394-3 PMID: 31680014

UK ECT Review Group. (2003). Efficacy and safety of electroconvulsive therapy in depressive disorders: A systematic review and meta-analysis. The Lancet, 361(9360), 799-808. https://doi.org/10.1016/S0140-6736(03)12705-5 PMID: 12642045

van Diermen, L., van den Ameele, S., Kamperman, A. M., Sabbe, B. C. G., Vermeulen, T., Schrijvers, D., & Birkenhager, T. K. (2018). Prediction of electroconvulsive therapy response and remission in major depression: Meta-analysis. The British Journal of Psychiatry, 212(2), 71-80. https://doi.org/10.1192/bjp.2017.28 PMID: 29436330

Wiles, N., Thomas, L., Abel, A., Ridgway, N., Turner, N., Campbell, J., Garland, A., Hollinghurst, S., Jerrom, B., Kessler, D., Kuyken, W., Morrison, J., Turner, K., Williams, C., Peters, T., & Lewis, G. (2013). Cognitive behavioural therapy as an adjunct to pharmacotherapy for primary care based patients with treatment resistant depression: Results of the CoBalT randomised controlled trial. The Lancet, 381(9864), 375-384. https://doi.org/10.1016/S0140-6736(12)61552-9 PMID: 23219570

Wilkinson, S. T., Ballard, E. D., Bloch, M. H., Mathew, S. J., Murrough, J. W., Feder, A., Sos, P., Wang, G., Zarate, C. A., Jr., & Sanacora, G. (2018). The effect of a single dose of intravenous ketamine on suicidal ideation: A systematic review and individual participant data meta-analysis. American Journal of Psychiatry, 175(2), 150-158. https://doi.org/10.1176/appi.ajp.2017.17040472 PMID: 28969441 · PMCID: PMC5794524

Zanos, P., & Gould, T. D. (2018). Mechanisms of ketamine action as an antidepressant. Molecular Psychiatry, 23(4), 801-811. https://doi.org/10.1038/mp.2017.255 PMID: 29532791 · PMCID: PMC5999402

Zarate, C. A., Jr., Singh, J. B., Carlson, P. J., Brutsche, N. E., Ameli, R., Luckenbaugh, D. A., Charney, D. S., & Manji, H. K. (2006). A randomized trial of an N-methyl-D-aspartate antagonist in treatment-resistant major depression. Archives of General Psychiatry, 63(8), 856-864. https://doi.org/10.1001/archpsyc.63.8.856 PMID: 16894061

Zhdanava, M., Pilon, D., Ghelerter, I., Chow, W., Joshi, K., Lefebvre, P., & Sheehan, J. J. (2021). The prevalence and national burden of treatment-resistant depression and major depressive disorder in the United States. The Journal of Clinical Psychiatry, 82(2), 20m13699. https://doi.org/10.4088/jcp.20m13699 PMID: 33989464